root / block / qcow2.c @ 72daa72e
History | View | Annotate | Download (70.6 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 |
} QEMU_PACKED 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 |
Error **errp) |
| 84 |
{
|
| 85 |
BDRVQcowState *s = bs->opaque; |
| 86 |
QCowExtension ext; |
| 87 |
uint64_t offset; |
| 88 |
int ret;
|
| 89 |
|
| 90 |
#ifdef DEBUG_EXT
|
| 91 |
printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset, end_offset);
|
| 92 |
#endif
|
| 93 |
offset = start_offset; |
| 94 |
while (offset < end_offset) {
|
| 95 |
|
| 96 |
#ifdef DEBUG_EXT
|
| 97 |
/* Sanity check */
|
| 98 |
if (offset > s->cluster_size)
|
| 99 |
printf("qcow2_read_extension: suspicious offset %lu\n", offset);
|
| 100 |
|
| 101 |
printf("attempting to read extended header in offset %lu\n", offset);
|
| 102 |
#endif
|
| 103 |
|
| 104 |
ret = bdrv_pread(bs->file, offset, &ext, sizeof(ext));
|
| 105 |
if (ret < 0) { |
| 106 |
error_setg_errno(errp, -ret, "qcow2_read_extension: ERROR: "
|
| 107 |
"pread fail from offset %" PRIu64, offset);
|
| 108 |
return 1; |
| 109 |
} |
| 110 |
be32_to_cpus(&ext.magic); |
| 111 |
be32_to_cpus(&ext.len); |
| 112 |
offset += sizeof(ext);
|
| 113 |
#ifdef DEBUG_EXT
|
| 114 |
printf("ext.magic = 0x%x\n", ext.magic);
|
| 115 |
#endif
|
| 116 |
if (ext.len > end_offset - offset) {
|
| 117 |
error_setg(errp, "Header extension too large");
|
| 118 |
return -EINVAL;
|
| 119 |
} |
| 120 |
|
| 121 |
switch (ext.magic) {
|
| 122 |
case QCOW2_EXT_MAGIC_END:
|
| 123 |
return 0; |
| 124 |
|
| 125 |
case QCOW2_EXT_MAGIC_BACKING_FORMAT:
|
| 126 |
if (ext.len >= sizeof(bs->backing_format)) { |
| 127 |
error_setg(errp, "ERROR: ext_backing_format: len=%u too large"
|
| 128 |
" (>=%zu)", ext.len, sizeof(bs->backing_format)); |
| 129 |
return 2; |
| 130 |
} |
| 131 |
ret = bdrv_pread(bs->file, offset, bs->backing_format, ext.len); |
| 132 |
if (ret < 0) { |
| 133 |
error_setg_errno(errp, -ret, "ERROR: ext_backing_format: "
|
| 134 |
"Could not read format name");
|
| 135 |
return 3; |
| 136 |
} |
| 137 |
bs->backing_format[ext.len] = '\0';
|
| 138 |
#ifdef DEBUG_EXT
|
| 139 |
printf("Qcow2: Got format extension %s\n", bs->backing_format);
|
| 140 |
#endif
|
| 141 |
break;
|
| 142 |
|
| 143 |
case QCOW2_EXT_MAGIC_FEATURE_TABLE:
|
| 144 |
if (p_feature_table != NULL) { |
| 145 |
void* feature_table = g_malloc0(ext.len + 2 * sizeof(Qcow2Feature)); |
| 146 |
ret = bdrv_pread(bs->file, offset , feature_table, ext.len); |
| 147 |
if (ret < 0) { |
| 148 |
error_setg_errno(errp, -ret, "ERROR: ext_feature_table: "
|
| 149 |
"Could not read table");
|
| 150 |
return ret;
|
| 151 |
} |
| 152 |
|
| 153 |
*p_feature_table = feature_table; |
| 154 |
} |
| 155 |
break;
|
| 156 |
|
| 157 |
default:
|
| 158 |
/* unknown magic - save it in case we need to rewrite the header */
|
| 159 |
{
|
| 160 |
Qcow2UnknownHeaderExtension *uext; |
| 161 |
|
| 162 |
uext = g_malloc0(sizeof(*uext) + ext.len);
|
| 163 |
uext->magic = ext.magic; |
| 164 |
uext->len = ext.len; |
| 165 |
QLIST_INSERT_HEAD(&s->unknown_header_ext, uext, next); |
| 166 |
|
| 167 |
ret = bdrv_pread(bs->file, offset , uext->data, uext->len); |
| 168 |
if (ret < 0) { |
| 169 |
error_setg_errno(errp, -ret, "ERROR: unknown extension: "
|
| 170 |
"Could not read data");
|
| 171 |
return ret;
|
| 172 |
} |
| 173 |
} |
| 174 |
break;
|
| 175 |
} |
| 176 |
|
| 177 |
offset += ((ext.len + 7) & ~7); |
| 178 |
} |
| 179 |
|
| 180 |
return 0; |
| 181 |
} |
| 182 |
|
| 183 |
static void cleanup_unknown_header_ext(BlockDriverState *bs) |
| 184 |
{
|
| 185 |
BDRVQcowState *s = bs->opaque; |
| 186 |
Qcow2UnknownHeaderExtension *uext, *next; |
| 187 |
|
| 188 |
QLIST_FOREACH_SAFE(uext, &s->unknown_header_ext, next, next) {
|
| 189 |
QLIST_REMOVE(uext, next); |
| 190 |
g_free(uext); |
| 191 |
} |
| 192 |
} |
| 193 |
|
| 194 |
static void GCC_FMT_ATTR(3, 4) report_unsupported(BlockDriverState *bs, |
| 195 |
Error **errp, const char *fmt, ...) |
| 196 |
{
|
| 197 |
char msg[64]; |
| 198 |
va_list ap; |
| 199 |
|
| 200 |
va_start(ap, fmt); |
| 201 |
vsnprintf(msg, sizeof(msg), fmt, ap);
|
| 202 |
va_end(ap); |
| 203 |
|
| 204 |
error_set(errp, QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, bs->device_name, "qcow2",
|
| 205 |
msg); |
| 206 |
} |
| 207 |
|
| 208 |
static void report_unsupported_feature(BlockDriverState *bs, |
| 209 |
Error **errp, Qcow2Feature *table, uint64_t mask) |
| 210 |
{
|
| 211 |
while (table && table->name[0] != '\0') { |
| 212 |
if (table->type == QCOW2_FEAT_TYPE_INCOMPATIBLE) {
|
| 213 |
if (mask & (1 << table->bit)) { |
| 214 |
report_unsupported(bs, errp, "%.46s", table->name);
|
| 215 |
mask &= ~(1 << table->bit);
|
| 216 |
} |
| 217 |
} |
| 218 |
table++; |
| 219 |
} |
| 220 |
|
| 221 |
if (mask) {
|
| 222 |
report_unsupported(bs, errp, "Unknown incompatible feature: %" PRIx64,
|
| 223 |
mask); |
| 224 |
} |
| 225 |
} |
| 226 |
|
| 227 |
/*
|
| 228 |
* Sets the dirty bit and flushes afterwards if necessary.
|
| 229 |
*
|
| 230 |
* The incompatible_features bit is only set if the image file header was
|
| 231 |
* updated successfully. Therefore it is not required to check the return
|
| 232 |
* value of this function.
|
| 233 |
*/
|
| 234 |
int qcow2_mark_dirty(BlockDriverState *bs)
|
| 235 |
{
|
| 236 |
BDRVQcowState *s = bs->opaque; |
| 237 |
uint64_t val; |
| 238 |
int ret;
|
| 239 |
|
| 240 |
assert(s->qcow_version >= 3);
|
| 241 |
|
| 242 |
if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
|
| 243 |
return 0; /* already dirty */ |
| 244 |
} |
| 245 |
|
| 246 |
val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY); |
| 247 |
ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features), |
| 248 |
&val, sizeof(val));
|
| 249 |
if (ret < 0) { |
| 250 |
return ret;
|
| 251 |
} |
| 252 |
ret = bdrv_flush(bs->file); |
| 253 |
if (ret < 0) { |
| 254 |
return ret;
|
| 255 |
} |
| 256 |
|
| 257 |
/* Only treat image as dirty if the header was updated successfully */
|
| 258 |
s->incompatible_features |= QCOW2_INCOMPAT_DIRTY; |
| 259 |
return 0; |
| 260 |
} |
| 261 |
|
| 262 |
/*
|
| 263 |
* Clears the dirty bit and flushes before if necessary. Only call this
|
| 264 |
* function when there are no pending requests, it does not guard against
|
| 265 |
* concurrent requests dirtying the image.
|
| 266 |
*/
|
| 267 |
static int qcow2_mark_clean(BlockDriverState *bs) |
| 268 |
{
|
| 269 |
BDRVQcowState *s = bs->opaque; |
| 270 |
|
| 271 |
if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
|
| 272 |
int ret = bdrv_flush(bs);
|
| 273 |
if (ret < 0) { |
| 274 |
return ret;
|
| 275 |
} |
| 276 |
|
| 277 |
s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY; |
| 278 |
return qcow2_update_header(bs);
|
| 279 |
} |
| 280 |
return 0; |
| 281 |
} |
| 282 |
|
| 283 |
/*
|
| 284 |
* Marks the image as corrupt.
|
| 285 |
*/
|
| 286 |
int qcow2_mark_corrupt(BlockDriverState *bs)
|
| 287 |
{
|
| 288 |
BDRVQcowState *s = bs->opaque; |
| 289 |
|
| 290 |
s->incompatible_features |= QCOW2_INCOMPAT_CORRUPT; |
| 291 |
return qcow2_update_header(bs);
|
| 292 |
} |
| 293 |
|
| 294 |
/*
|
| 295 |
* Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
|
| 296 |
* before if necessary.
|
| 297 |
*/
|
| 298 |
int qcow2_mark_consistent(BlockDriverState *bs)
|
| 299 |
{
|
| 300 |
BDRVQcowState *s = bs->opaque; |
| 301 |
|
| 302 |
if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
|
| 303 |
int ret = bdrv_flush(bs);
|
| 304 |
if (ret < 0) { |
| 305 |
return ret;
|
| 306 |
} |
| 307 |
|
| 308 |
s->incompatible_features &= ~QCOW2_INCOMPAT_CORRUPT; |
| 309 |
return qcow2_update_header(bs);
|
| 310 |
} |
| 311 |
return 0; |
| 312 |
} |
| 313 |
|
| 314 |
static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result, |
| 315 |
BdrvCheckMode fix) |
| 316 |
{
|
| 317 |
int ret = qcow2_check_refcounts(bs, result, fix);
|
| 318 |
if (ret < 0) { |
| 319 |
return ret;
|
| 320 |
} |
| 321 |
|
| 322 |
if (fix && result->check_errors == 0 && result->corruptions == 0) { |
| 323 |
ret = qcow2_mark_clean(bs); |
| 324 |
if (ret < 0) { |
| 325 |
return ret;
|
| 326 |
} |
| 327 |
return qcow2_mark_consistent(bs);
|
| 328 |
} |
| 329 |
return ret;
|
| 330 |
} |
| 331 |
|
| 332 |
static QemuOptsList qcow2_runtime_opts = {
|
| 333 |
.name = "qcow2",
|
| 334 |
.head = QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts.head), |
| 335 |
.desc = {
|
| 336 |
{
|
| 337 |
.name = QCOW2_OPT_LAZY_REFCOUNTS, |
| 338 |
.type = QEMU_OPT_BOOL, |
| 339 |
.help = "Postpone refcount updates",
|
| 340 |
}, |
| 341 |
{
|
| 342 |
.name = QCOW2_OPT_DISCARD_REQUEST, |
| 343 |
.type = QEMU_OPT_BOOL, |
| 344 |
.help = "Pass guest discard requests to the layer below",
|
| 345 |
}, |
| 346 |
{
|
| 347 |
.name = QCOW2_OPT_DISCARD_SNAPSHOT, |
| 348 |
.type = QEMU_OPT_BOOL, |
| 349 |
.help = "Generate discard requests when snapshot related space "
|
| 350 |
"is freed",
|
| 351 |
}, |
| 352 |
{
|
| 353 |
.name = QCOW2_OPT_DISCARD_OTHER, |
| 354 |
.type = QEMU_OPT_BOOL, |
| 355 |
.help = "Generate discard requests when other clusters are freed",
|
| 356 |
}, |
| 357 |
{
|
| 358 |
.name = QCOW2_OPT_OVERLAP, |
| 359 |
.type = QEMU_OPT_STRING, |
| 360 |
.help = "Selects which overlap checks to perform from a range of "
|
| 361 |
"templates (none, constant, cached, all)",
|
| 362 |
}, |
| 363 |
{
|
| 364 |
.name = QCOW2_OPT_OVERLAP_MAIN_HEADER, |
| 365 |
.type = QEMU_OPT_BOOL, |
| 366 |
.help = "Check for unintended writes into the main qcow2 header",
|
| 367 |
}, |
| 368 |
{
|
| 369 |
.name = QCOW2_OPT_OVERLAP_ACTIVE_L1, |
| 370 |
.type = QEMU_OPT_BOOL, |
| 371 |
.help = "Check for unintended writes into the active L1 table",
|
| 372 |
}, |
| 373 |
{
|
| 374 |
.name = QCOW2_OPT_OVERLAP_ACTIVE_L2, |
| 375 |
.type = QEMU_OPT_BOOL, |
| 376 |
.help = "Check for unintended writes into an active L2 table",
|
| 377 |
}, |
| 378 |
{
|
| 379 |
.name = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, |
| 380 |
.type = QEMU_OPT_BOOL, |
| 381 |
.help = "Check for unintended writes into the refcount table",
|
| 382 |
}, |
| 383 |
{
|
| 384 |
.name = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, |
| 385 |
.type = QEMU_OPT_BOOL, |
| 386 |
.help = "Check for unintended writes into a refcount block",
|
| 387 |
}, |
| 388 |
{
|
| 389 |
.name = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, |
| 390 |
.type = QEMU_OPT_BOOL, |
| 391 |
.help = "Check for unintended writes into the snapshot table",
|
| 392 |
}, |
| 393 |
{
|
| 394 |
.name = QCOW2_OPT_OVERLAP_INACTIVE_L1, |
| 395 |
.type = QEMU_OPT_BOOL, |
| 396 |
.help = "Check for unintended writes into an inactive L1 table",
|
| 397 |
}, |
| 398 |
{
|
| 399 |
.name = QCOW2_OPT_OVERLAP_INACTIVE_L2, |
| 400 |
.type = QEMU_OPT_BOOL, |
| 401 |
.help = "Check for unintended writes into an inactive L2 table",
|
| 402 |
}, |
| 403 |
{ /* end of list */ }
|
| 404 |
}, |
| 405 |
}; |
| 406 |
|
| 407 |
static const char *overlap_bool_option_names[QCOW2_OL_MAX_BITNR] = { |
| 408 |
[QCOW2_OL_MAIN_HEADER_BITNR] = QCOW2_OPT_OVERLAP_MAIN_HEADER, |
| 409 |
[QCOW2_OL_ACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L1, |
| 410 |
[QCOW2_OL_ACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_ACTIVE_L2, |
| 411 |
[QCOW2_OL_REFCOUNT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE, |
| 412 |
[QCOW2_OL_REFCOUNT_BLOCK_BITNR] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK, |
| 413 |
[QCOW2_OL_SNAPSHOT_TABLE_BITNR] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE, |
| 414 |
[QCOW2_OL_INACTIVE_L1_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L1, |
| 415 |
[QCOW2_OL_INACTIVE_L2_BITNR] = QCOW2_OPT_OVERLAP_INACTIVE_L2, |
| 416 |
}; |
| 417 |
|
| 418 |
static int qcow2_open(BlockDriverState *bs, QDict *options, int flags, |
| 419 |
Error **errp) |
| 420 |
{
|
| 421 |
BDRVQcowState *s = bs->opaque; |
| 422 |
int len, i, ret = 0; |
| 423 |
QCowHeader header; |
| 424 |
QemuOpts *opts; |
| 425 |
Error *local_err = NULL;
|
| 426 |
uint64_t ext_end; |
| 427 |
uint64_t l1_vm_state_index; |
| 428 |
const char *opt_overlap_check; |
| 429 |
int overlap_check_template = 0; |
| 430 |
|
| 431 |
ret = bdrv_pread(bs->file, 0, &header, sizeof(header)); |
| 432 |
if (ret < 0) { |
| 433 |
error_setg_errno(errp, -ret, "Could not read qcow2 header");
|
| 434 |
goto fail;
|
| 435 |
} |
| 436 |
be32_to_cpus(&header.magic); |
| 437 |
be32_to_cpus(&header.version); |
| 438 |
be64_to_cpus(&header.backing_file_offset); |
| 439 |
be32_to_cpus(&header.backing_file_size); |
| 440 |
be64_to_cpus(&header.size); |
| 441 |
be32_to_cpus(&header.cluster_bits); |
| 442 |
be32_to_cpus(&header.crypt_method); |
| 443 |
be64_to_cpus(&header.l1_table_offset); |
| 444 |
be32_to_cpus(&header.l1_size); |
| 445 |
be64_to_cpus(&header.refcount_table_offset); |
| 446 |
be32_to_cpus(&header.refcount_table_clusters); |
| 447 |
be64_to_cpus(&header.snapshots_offset); |
| 448 |
be32_to_cpus(&header.nb_snapshots); |
| 449 |
|
| 450 |
if (header.magic != QCOW_MAGIC) {
|
| 451 |
error_setg(errp, "Image is not in qcow2 format");
|
| 452 |
ret = -EMEDIUMTYPE; |
| 453 |
goto fail;
|
| 454 |
} |
| 455 |
if (header.version < 2 || header.version > 3) { |
| 456 |
report_unsupported(bs, errp, "QCOW version %d", header.version);
|
| 457 |
ret = -ENOTSUP; |
| 458 |
goto fail;
|
| 459 |
} |
| 460 |
|
| 461 |
s->qcow_version = header.version; |
| 462 |
|
| 463 |
/* Initialise version 3 header fields */
|
| 464 |
if (header.version == 2) { |
| 465 |
header.incompatible_features = 0;
|
| 466 |
header.compatible_features = 0;
|
| 467 |
header.autoclear_features = 0;
|
| 468 |
header.refcount_order = 4;
|
| 469 |
header.header_length = 72;
|
| 470 |
} else {
|
| 471 |
be64_to_cpus(&header.incompatible_features); |
| 472 |
be64_to_cpus(&header.compatible_features); |
| 473 |
be64_to_cpus(&header.autoclear_features); |
| 474 |
be32_to_cpus(&header.refcount_order); |
| 475 |
be32_to_cpus(&header.header_length); |
| 476 |
} |
| 477 |
|
| 478 |
if (header.header_length > sizeof(header)) { |
| 479 |
s->unknown_header_fields_size = header.header_length - sizeof(header);
|
| 480 |
s->unknown_header_fields = g_malloc(s->unknown_header_fields_size); |
| 481 |
ret = bdrv_pread(bs->file, sizeof(header), s->unknown_header_fields,
|
| 482 |
s->unknown_header_fields_size); |
| 483 |
if (ret < 0) { |
| 484 |
error_setg_errno(errp, -ret, "Could not read unknown qcow2 header "
|
| 485 |
"fields");
|
| 486 |
goto fail;
|
| 487 |
} |
| 488 |
} |
| 489 |
|
| 490 |
if (header.backing_file_offset) {
|
| 491 |
ext_end = header.backing_file_offset; |
| 492 |
} else {
|
| 493 |
ext_end = 1 << header.cluster_bits;
|
| 494 |
} |
| 495 |
|
| 496 |
/* Handle feature bits */
|
| 497 |
s->incompatible_features = header.incompatible_features; |
| 498 |
s->compatible_features = header.compatible_features; |
| 499 |
s->autoclear_features = header.autoclear_features; |
| 500 |
|
| 501 |
if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
|
| 502 |
void *feature_table = NULL; |
| 503 |
qcow2_read_extensions(bs, header.header_length, ext_end, |
| 504 |
&feature_table, NULL);
|
| 505 |
report_unsupported_feature(bs, errp, feature_table, |
| 506 |
s->incompatible_features & |
| 507 |
~QCOW2_INCOMPAT_MASK); |
| 508 |
ret = -ENOTSUP; |
| 509 |
goto fail;
|
| 510 |
} |
| 511 |
|
| 512 |
if (s->incompatible_features & QCOW2_INCOMPAT_CORRUPT) {
|
| 513 |
/* Corrupt images may not be written to unless they are being repaired
|
| 514 |
*/
|
| 515 |
if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_CHECK)) {
|
| 516 |
error_setg(errp, "qcow2: Image is corrupt; cannot be opened "
|
| 517 |
"read/write");
|
| 518 |
ret = -EACCES; |
| 519 |
goto fail;
|
| 520 |
} |
| 521 |
} |
| 522 |
|
| 523 |
/* Check support for various header values */
|
| 524 |
if (header.refcount_order != 4) { |
| 525 |
report_unsupported(bs, errp, "%d bit reference counts",
|
| 526 |
1 << header.refcount_order);
|
| 527 |
ret = -ENOTSUP; |
| 528 |
goto fail;
|
| 529 |
} |
| 530 |
s->refcount_order = header.refcount_order; |
| 531 |
|
| 532 |
if (header.cluster_bits < MIN_CLUSTER_BITS ||
|
| 533 |
header.cluster_bits > MAX_CLUSTER_BITS) {
|
| 534 |
error_setg(errp, "Unsupported cluster size: 2^%i", header.cluster_bits);
|
| 535 |
ret = -EINVAL; |
| 536 |
goto fail;
|
| 537 |
} |
| 538 |
if (header.crypt_method > QCOW_CRYPT_AES) {
|
| 539 |
error_setg(errp, "Unsupported encryption method: %i",
|
| 540 |
header.crypt_method); |
| 541 |
ret = -EINVAL; |
| 542 |
goto fail;
|
| 543 |
} |
| 544 |
s->crypt_method_header = header.crypt_method; |
| 545 |
if (s->crypt_method_header) {
|
| 546 |
bs->encrypted = 1;
|
| 547 |
} |
| 548 |
s->cluster_bits = header.cluster_bits; |
| 549 |
s->cluster_size = 1 << s->cluster_bits;
|
| 550 |
s->cluster_sectors = 1 << (s->cluster_bits - 9); |
| 551 |
s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */ |
| 552 |
s->l2_size = 1 << s->l2_bits;
|
| 553 |
bs->total_sectors = header.size / 512;
|
| 554 |
s->csize_shift = (62 - (s->cluster_bits - 8)); |
| 555 |
s->csize_mask = (1 << (s->cluster_bits - 8)) - 1; |
| 556 |
s->cluster_offset_mask = (1LL << s->csize_shift) - 1; |
| 557 |
s->refcount_table_offset = header.refcount_table_offset; |
| 558 |
s->refcount_table_size = |
| 559 |
header.refcount_table_clusters << (s->cluster_bits - 3);
|
| 560 |
|
| 561 |
s->snapshots_offset = header.snapshots_offset; |
| 562 |
s->nb_snapshots = header.nb_snapshots; |
| 563 |
|
| 564 |
/* read the level 1 table */
|
| 565 |
s->l1_size = header.l1_size; |
| 566 |
|
| 567 |
l1_vm_state_index = size_to_l1(s, header.size); |
| 568 |
if (l1_vm_state_index > INT_MAX) {
|
| 569 |
error_setg(errp, "Image is too big");
|
| 570 |
ret = -EFBIG; |
| 571 |
goto fail;
|
| 572 |
} |
| 573 |
s->l1_vm_state_index = l1_vm_state_index; |
| 574 |
|
| 575 |
/* the L1 table must contain at least enough entries to put
|
| 576 |
header.size bytes */
|
| 577 |
if (s->l1_size < s->l1_vm_state_index) {
|
| 578 |
error_setg(errp, "L1 table is too small");
|
| 579 |
ret = -EINVAL; |
| 580 |
goto fail;
|
| 581 |
} |
| 582 |
s->l1_table_offset = header.l1_table_offset; |
| 583 |
if (s->l1_size > 0) { |
| 584 |
s->l1_table = g_malloc0( |
| 585 |
align_offset(s->l1_size * sizeof(uint64_t), 512)); |
| 586 |
ret = bdrv_pread(bs->file, s->l1_table_offset, s->l1_table, |
| 587 |
s->l1_size * sizeof(uint64_t));
|
| 588 |
if (ret < 0) { |
| 589 |
error_setg_errno(errp, -ret, "Could not read L1 table");
|
| 590 |
goto fail;
|
| 591 |
} |
| 592 |
for(i = 0;i < s->l1_size; i++) { |
| 593 |
be64_to_cpus(&s->l1_table[i]); |
| 594 |
} |
| 595 |
} |
| 596 |
|
| 597 |
/* alloc L2 table/refcount block cache */
|
| 598 |
s->l2_table_cache = qcow2_cache_create(bs, L2_CACHE_SIZE); |
| 599 |
s->refcount_block_cache = qcow2_cache_create(bs, REFCOUNT_CACHE_SIZE); |
| 600 |
|
| 601 |
s->cluster_cache = g_malloc(s->cluster_size); |
| 602 |
/* one more sector for decompressed data alignment */
|
| 603 |
s->cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size |
| 604 |
+ 512);
|
| 605 |
s->cluster_cache_offset = -1;
|
| 606 |
s->flags = flags; |
| 607 |
|
| 608 |
ret = qcow2_refcount_init(bs); |
| 609 |
if (ret != 0) { |
| 610 |
error_setg_errno(errp, -ret, "Could not initialize refcount handling");
|
| 611 |
goto fail;
|
| 612 |
} |
| 613 |
|
| 614 |
QLIST_INIT(&s->cluster_allocs); |
| 615 |
QTAILQ_INIT(&s->discards); |
| 616 |
|
| 617 |
/* read qcow2 extensions */
|
| 618 |
if (qcow2_read_extensions(bs, header.header_length, ext_end, NULL, |
| 619 |
&local_err)) {
|
| 620 |
error_propagate(errp, local_err); |
| 621 |
ret = -EINVAL; |
| 622 |
goto fail;
|
| 623 |
} |
| 624 |
|
| 625 |
/* read the backing file name */
|
| 626 |
if (header.backing_file_offset != 0) { |
| 627 |
len = header.backing_file_size; |
| 628 |
if (len > 1023) { |
| 629 |
len = 1023;
|
| 630 |
} |
| 631 |
ret = bdrv_pread(bs->file, header.backing_file_offset, |
| 632 |
bs->backing_file, len); |
| 633 |
if (ret < 0) { |
| 634 |
error_setg_errno(errp, -ret, "Could not read backing file name");
|
| 635 |
goto fail;
|
| 636 |
} |
| 637 |
bs->backing_file[len] = '\0';
|
| 638 |
} |
| 639 |
|
| 640 |
ret = qcow2_read_snapshots(bs); |
| 641 |
if (ret < 0) { |
| 642 |
error_setg_errno(errp, -ret, "Could not read snapshots");
|
| 643 |
goto fail;
|
| 644 |
} |
| 645 |
|
| 646 |
/* Clear unknown autoclear feature bits */
|
| 647 |
if (!bs->read_only && s->autoclear_features != 0) { |
| 648 |
s->autoclear_features = 0;
|
| 649 |
ret = qcow2_update_header(bs); |
| 650 |
if (ret < 0) { |
| 651 |
error_setg_errno(errp, -ret, "Could not update qcow2 header");
|
| 652 |
goto fail;
|
| 653 |
} |
| 654 |
} |
| 655 |
|
| 656 |
/* Initialise locks */
|
| 657 |
qemu_co_mutex_init(&s->lock); |
| 658 |
|
| 659 |
/* Repair image if dirty */
|
| 660 |
if (!(flags & BDRV_O_CHECK) && !bs->read_only &&
|
| 661 |
(s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
|
| 662 |
BdrvCheckResult result = {0};
|
| 663 |
|
| 664 |
ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS); |
| 665 |
if (ret < 0) { |
| 666 |
error_setg_errno(errp, -ret, "Could not repair dirty image");
|
| 667 |
goto fail;
|
| 668 |
} |
| 669 |
} |
| 670 |
|
| 671 |
/* Enable lazy_refcounts according to image and command line options */
|
| 672 |
opts = qemu_opts_create(&qcow2_runtime_opts, NULL, 0, &error_abort); |
| 673 |
qemu_opts_absorb_qdict(opts, options, &local_err); |
| 674 |
if (error_is_set(&local_err)) {
|
| 675 |
error_propagate(errp, local_err); |
| 676 |
ret = -EINVAL; |
| 677 |
goto fail;
|
| 678 |
} |
| 679 |
|
| 680 |
s->use_lazy_refcounts = qemu_opt_get_bool(opts, QCOW2_OPT_LAZY_REFCOUNTS, |
| 681 |
(s->compatible_features & QCOW2_COMPAT_LAZY_REFCOUNTS)); |
| 682 |
|
| 683 |
s->discard_passthrough[QCOW2_DISCARD_NEVER] = false;
|
| 684 |
s->discard_passthrough[QCOW2_DISCARD_ALWAYS] = true;
|
| 685 |
s->discard_passthrough[QCOW2_DISCARD_REQUEST] = |
| 686 |
qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_REQUEST, |
| 687 |
flags & BDRV_O_UNMAP); |
| 688 |
s->discard_passthrough[QCOW2_DISCARD_SNAPSHOT] = |
| 689 |
qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_SNAPSHOT, true);
|
| 690 |
s->discard_passthrough[QCOW2_DISCARD_OTHER] = |
| 691 |
qemu_opt_get_bool(opts, QCOW2_OPT_DISCARD_OTHER, false);
|
| 692 |
|
| 693 |
opt_overlap_check = qemu_opt_get(opts, "overlap-check") ?: "cached"; |
| 694 |
if (!strcmp(opt_overlap_check, "none")) { |
| 695 |
overlap_check_template = 0;
|
| 696 |
} else if (!strcmp(opt_overlap_check, "constant")) { |
| 697 |
overlap_check_template = QCOW2_OL_CONSTANT; |
| 698 |
} else if (!strcmp(opt_overlap_check, "cached")) { |
| 699 |
overlap_check_template = QCOW2_OL_CACHED; |
| 700 |
} else if (!strcmp(opt_overlap_check, "all")) { |
| 701 |
overlap_check_template = QCOW2_OL_ALL; |
| 702 |
} else {
|
| 703 |
error_setg(errp, "Unsupported value '%s' for qcow2 option "
|
| 704 |
"'overlap-check'. Allowed are either of the following: "
|
| 705 |
"none, constant, cached, all", opt_overlap_check);
|
| 706 |
qemu_opts_del(opts); |
| 707 |
ret = -EINVAL; |
| 708 |
goto fail;
|
| 709 |
} |
| 710 |
|
| 711 |
s->overlap_check = 0;
|
| 712 |
for (i = 0; i < QCOW2_OL_MAX_BITNR; i++) { |
| 713 |
/* overlap-check defines a template bitmask, but every flag may be
|
| 714 |
* overwritten through the associated boolean option */
|
| 715 |
s->overlap_check |= |
| 716 |
qemu_opt_get_bool(opts, overlap_bool_option_names[i], |
| 717 |
overlap_check_template & (1 << i)) << i;
|
| 718 |
} |
| 719 |
|
| 720 |
qemu_opts_del(opts); |
| 721 |
bs->bl.write_zeroes_alignment = s->cluster_sectors; |
| 722 |
|
| 723 |
if (s->use_lazy_refcounts && s->qcow_version < 3) { |
| 724 |
error_setg(errp, "Lazy refcounts require a qcow2 image with at least "
|
| 725 |
"qemu 1.1 compatibility level");
|
| 726 |
ret = -EINVAL; |
| 727 |
goto fail;
|
| 728 |
} |
| 729 |
|
| 730 |
#ifdef DEBUG_ALLOC
|
| 731 |
{
|
| 732 |
BdrvCheckResult result = {0};
|
| 733 |
qcow2_check_refcounts(bs, &result, 0);
|
| 734 |
} |
| 735 |
#endif
|
| 736 |
return ret;
|
| 737 |
|
| 738 |
fail:
|
| 739 |
g_free(s->unknown_header_fields); |
| 740 |
cleanup_unknown_header_ext(bs); |
| 741 |
qcow2_free_snapshots(bs); |
| 742 |
qcow2_refcount_close(bs); |
| 743 |
g_free(s->l1_table); |
| 744 |
/* else pre-write overlap checks in cache_destroy may crash */
|
| 745 |
s->l1_table = NULL;
|
| 746 |
if (s->l2_table_cache) {
|
| 747 |
qcow2_cache_destroy(bs, s->l2_table_cache); |
| 748 |
} |
| 749 |
g_free(s->cluster_cache); |
| 750 |
qemu_vfree(s->cluster_data); |
| 751 |
return ret;
|
| 752 |
} |
| 753 |
|
| 754 |
static int qcow2_set_key(BlockDriverState *bs, const char *key) |
| 755 |
{
|
| 756 |
BDRVQcowState *s = bs->opaque; |
| 757 |
uint8_t keybuf[16];
|
| 758 |
int len, i;
|
| 759 |
|
| 760 |
memset(keybuf, 0, 16); |
| 761 |
len = strlen(key); |
| 762 |
if (len > 16) |
| 763 |
len = 16;
|
| 764 |
/* XXX: we could compress the chars to 7 bits to increase
|
| 765 |
entropy */
|
| 766 |
for(i = 0;i < len;i++) { |
| 767 |
keybuf[i] = key[i]; |
| 768 |
} |
| 769 |
s->crypt_method = s->crypt_method_header; |
| 770 |
|
| 771 |
if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0) |
| 772 |
return -1; |
| 773 |
if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0) |
| 774 |
return -1; |
| 775 |
#if 0
|
| 776 |
/* test */
|
| 777 |
{
|
| 778 |
uint8_t in[16];
|
| 779 |
uint8_t out[16];
|
| 780 |
uint8_t tmp[16];
|
| 781 |
for(i=0;i<16;i++)
|
| 782 |
in[i] = i;
|
| 783 |
AES_encrypt(in, tmp, &s->aes_encrypt_key);
|
| 784 |
AES_decrypt(tmp, out, &s->aes_decrypt_key);
|
| 785 |
for(i = 0; i < 16; i++)
|
| 786 |
printf(" %02x", tmp[i]);
|
| 787 |
printf("\n");
|
| 788 |
for(i = 0; i < 16; i++)
|
| 789 |
printf(" %02x", out[i]);
|
| 790 |
printf("\n");
|
| 791 |
}
|
| 792 |
#endif
|
| 793 |
return 0; |
| 794 |
} |
| 795 |
|
| 796 |
/* We have nothing to do for QCOW2 reopen, stubs just return
|
| 797 |
* success */
|
| 798 |
static int qcow2_reopen_prepare(BDRVReopenState *state, |
| 799 |
BlockReopenQueue *queue, Error **errp) |
| 800 |
{
|
| 801 |
return 0; |
| 802 |
} |
| 803 |
|
| 804 |
static int64_t coroutine_fn qcow2_co_get_block_status(BlockDriverState *bs,
|
| 805 |
int64_t sector_num, int nb_sectors, int *pnum) |
| 806 |
{
|
| 807 |
BDRVQcowState *s = bs->opaque; |
| 808 |
uint64_t cluster_offset; |
| 809 |
int index_in_cluster, ret;
|
| 810 |
int64_t status = 0;
|
| 811 |
|
| 812 |
*pnum = nb_sectors; |
| 813 |
qemu_co_mutex_lock(&s->lock); |
| 814 |
ret = qcow2_get_cluster_offset(bs, sector_num << 9, pnum, &cluster_offset);
|
| 815 |
qemu_co_mutex_unlock(&s->lock); |
| 816 |
if (ret < 0) { |
| 817 |
return ret;
|
| 818 |
} |
| 819 |
|
| 820 |
if (cluster_offset != 0 && ret != QCOW2_CLUSTER_COMPRESSED && |
| 821 |
!s->crypt_method) {
|
| 822 |
index_in_cluster = sector_num & (s->cluster_sectors - 1);
|
| 823 |
cluster_offset |= (index_in_cluster << BDRV_SECTOR_BITS); |
| 824 |
status |= BDRV_BLOCK_OFFSET_VALID | cluster_offset; |
| 825 |
} |
| 826 |
if (ret == QCOW2_CLUSTER_ZERO) {
|
| 827 |
status |= BDRV_BLOCK_ZERO; |
| 828 |
} else if (ret != QCOW2_CLUSTER_UNALLOCATED) { |
| 829 |
status |= BDRV_BLOCK_DATA; |
| 830 |
} |
| 831 |
return status;
|
| 832 |
} |
| 833 |
|
| 834 |
/* handle reading after the end of the backing file */
|
| 835 |
int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
|
| 836 |
int64_t sector_num, int nb_sectors)
|
| 837 |
{
|
| 838 |
int n1;
|
| 839 |
if ((sector_num + nb_sectors) <= bs->total_sectors)
|
| 840 |
return nb_sectors;
|
| 841 |
if (sector_num >= bs->total_sectors)
|
| 842 |
n1 = 0;
|
| 843 |
else
|
| 844 |
n1 = bs->total_sectors - sector_num; |
| 845 |
|
| 846 |
qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1)); |
| 847 |
|
| 848 |
return n1;
|
| 849 |
} |
| 850 |
|
| 851 |
static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num, |
| 852 |
int remaining_sectors, QEMUIOVector *qiov)
|
| 853 |
{
|
| 854 |
BDRVQcowState *s = bs->opaque; |
| 855 |
int index_in_cluster, n1;
|
| 856 |
int ret;
|
| 857 |
int cur_nr_sectors; /* number of sectors in current iteration */ |
| 858 |
uint64_t cluster_offset = 0;
|
| 859 |
uint64_t bytes_done = 0;
|
| 860 |
QEMUIOVector hd_qiov; |
| 861 |
uint8_t *cluster_data = NULL;
|
| 862 |
|
| 863 |
qemu_iovec_init(&hd_qiov, qiov->niov); |
| 864 |
|
| 865 |
qemu_co_mutex_lock(&s->lock); |
| 866 |
|
| 867 |
while (remaining_sectors != 0) { |
| 868 |
|
| 869 |
/* prepare next request */
|
| 870 |
cur_nr_sectors = remaining_sectors; |
| 871 |
if (s->crypt_method) {
|
| 872 |
cur_nr_sectors = MIN(cur_nr_sectors, |
| 873 |
QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); |
| 874 |
} |
| 875 |
|
| 876 |
ret = qcow2_get_cluster_offset(bs, sector_num << 9,
|
| 877 |
&cur_nr_sectors, &cluster_offset); |
| 878 |
if (ret < 0) { |
| 879 |
goto fail;
|
| 880 |
} |
| 881 |
|
| 882 |
index_in_cluster = sector_num & (s->cluster_sectors - 1);
|
| 883 |
|
| 884 |
qemu_iovec_reset(&hd_qiov); |
| 885 |
qemu_iovec_concat(&hd_qiov, qiov, bytes_done, |
| 886 |
cur_nr_sectors * 512);
|
| 887 |
|
| 888 |
switch (ret) {
|
| 889 |
case QCOW2_CLUSTER_UNALLOCATED:
|
| 890 |
|
| 891 |
if (bs->backing_hd) {
|
| 892 |
/* read from the base image */
|
| 893 |
n1 = qcow2_backing_read1(bs->backing_hd, &hd_qiov, |
| 894 |
sector_num, cur_nr_sectors); |
| 895 |
if (n1 > 0) { |
| 896 |
BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO); |
| 897 |
qemu_co_mutex_unlock(&s->lock); |
| 898 |
ret = bdrv_co_readv(bs->backing_hd, sector_num, |
| 899 |
n1, &hd_qiov); |
| 900 |
qemu_co_mutex_lock(&s->lock); |
| 901 |
if (ret < 0) { |
| 902 |
goto fail;
|
| 903 |
} |
| 904 |
} |
| 905 |
} else {
|
| 906 |
/* Note: in this case, no need to wait */
|
| 907 |
qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors); |
| 908 |
} |
| 909 |
break;
|
| 910 |
|
| 911 |
case QCOW2_CLUSTER_ZERO:
|
| 912 |
qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors); |
| 913 |
break;
|
| 914 |
|
| 915 |
case QCOW2_CLUSTER_COMPRESSED:
|
| 916 |
/* add AIO support for compressed blocks ? */
|
| 917 |
ret = qcow2_decompress_cluster(bs, cluster_offset); |
| 918 |
if (ret < 0) { |
| 919 |
goto fail;
|
| 920 |
} |
| 921 |
|
| 922 |
qemu_iovec_from_buf(&hd_qiov, 0,
|
| 923 |
s->cluster_cache + index_in_cluster * 512,
|
| 924 |
512 * cur_nr_sectors);
|
| 925 |
break;
|
| 926 |
|
| 927 |
case QCOW2_CLUSTER_NORMAL:
|
| 928 |
if ((cluster_offset & 511) != 0) { |
| 929 |
ret = -EIO; |
| 930 |
goto fail;
|
| 931 |
} |
| 932 |
|
| 933 |
if (s->crypt_method) {
|
| 934 |
/*
|
| 935 |
* For encrypted images, read everything into a temporary
|
| 936 |
* contiguous buffer on which the AES functions can work.
|
| 937 |
*/
|
| 938 |
if (!cluster_data) {
|
| 939 |
cluster_data = |
| 940 |
qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); |
| 941 |
} |
| 942 |
|
| 943 |
assert(cur_nr_sectors <= |
| 944 |
QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors); |
| 945 |
qemu_iovec_reset(&hd_qiov); |
| 946 |
qemu_iovec_add(&hd_qiov, cluster_data, |
| 947 |
512 * cur_nr_sectors);
|
| 948 |
} |
| 949 |
|
| 950 |
BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); |
| 951 |
qemu_co_mutex_unlock(&s->lock); |
| 952 |
ret = bdrv_co_readv(bs->file, |
| 953 |
(cluster_offset >> 9) + index_in_cluster,
|
| 954 |
cur_nr_sectors, &hd_qiov); |
| 955 |
qemu_co_mutex_lock(&s->lock); |
| 956 |
if (ret < 0) { |
| 957 |
goto fail;
|
| 958 |
} |
| 959 |
if (s->crypt_method) {
|
| 960 |
qcow2_encrypt_sectors(s, sector_num, cluster_data, |
| 961 |
cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
|
| 962 |
qemu_iovec_from_buf(qiov, bytes_done, |
| 963 |
cluster_data, 512 * cur_nr_sectors);
|
| 964 |
} |
| 965 |
break;
|
| 966 |
|
| 967 |
default:
|
| 968 |
g_assert_not_reached(); |
| 969 |
ret = -EIO; |
| 970 |
goto fail;
|
| 971 |
} |
| 972 |
|
| 973 |
remaining_sectors -= cur_nr_sectors; |
| 974 |
sector_num += cur_nr_sectors; |
| 975 |
bytes_done += cur_nr_sectors * 512;
|
| 976 |
} |
| 977 |
ret = 0;
|
| 978 |
|
| 979 |
fail:
|
| 980 |
qemu_co_mutex_unlock(&s->lock); |
| 981 |
|
| 982 |
qemu_iovec_destroy(&hd_qiov); |
| 983 |
qemu_vfree(cluster_data); |
| 984 |
|
| 985 |
return ret;
|
| 986 |
} |
| 987 |
|
| 988 |
static coroutine_fn int qcow2_co_writev(BlockDriverState *bs, |
| 989 |
int64_t sector_num, |
| 990 |
int remaining_sectors,
|
| 991 |
QEMUIOVector *qiov) |
| 992 |
{
|
| 993 |
BDRVQcowState *s = bs->opaque; |
| 994 |
int index_in_cluster;
|
| 995 |
int n_end;
|
| 996 |
int ret;
|
| 997 |
int cur_nr_sectors; /* number of sectors in current iteration */ |
| 998 |
uint64_t cluster_offset; |
| 999 |
QEMUIOVector hd_qiov; |
| 1000 |
uint64_t bytes_done = 0;
|
| 1001 |
uint8_t *cluster_data = NULL;
|
| 1002 |
QCowL2Meta *l2meta = NULL;
|
| 1003 |
|
| 1004 |
trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num, |
| 1005 |
remaining_sectors); |
| 1006 |
|
| 1007 |
qemu_iovec_init(&hd_qiov, qiov->niov); |
| 1008 |
|
| 1009 |
s->cluster_cache_offset = -1; /* disable compressed cache */ |
| 1010 |
|
| 1011 |
qemu_co_mutex_lock(&s->lock); |
| 1012 |
|
| 1013 |
while (remaining_sectors != 0) { |
| 1014 |
|
| 1015 |
l2meta = NULL;
|
| 1016 |
|
| 1017 |
trace_qcow2_writev_start_part(qemu_coroutine_self()); |
| 1018 |
index_in_cluster = sector_num & (s->cluster_sectors - 1);
|
| 1019 |
n_end = index_in_cluster + remaining_sectors; |
| 1020 |
if (s->crypt_method &&
|
| 1021 |
n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) {
|
| 1022 |
n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors; |
| 1023 |
} |
| 1024 |
|
| 1025 |
ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
|
| 1026 |
index_in_cluster, n_end, &cur_nr_sectors, &cluster_offset, &l2meta); |
| 1027 |
if (ret < 0) { |
| 1028 |
goto fail;
|
| 1029 |
} |
| 1030 |
|
| 1031 |
assert((cluster_offset & 511) == 0); |
| 1032 |
|
| 1033 |
qemu_iovec_reset(&hd_qiov); |
| 1034 |
qemu_iovec_concat(&hd_qiov, qiov, bytes_done, |
| 1035 |
cur_nr_sectors * 512);
|
| 1036 |
|
| 1037 |
if (s->crypt_method) {
|
| 1038 |
if (!cluster_data) {
|
| 1039 |
cluster_data = qemu_blockalign(bs, QCOW_MAX_CRYPT_CLUSTERS * |
| 1040 |
s->cluster_size); |
| 1041 |
} |
| 1042 |
|
| 1043 |
assert(hd_qiov.size <= |
| 1044 |
QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size); |
| 1045 |
qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
|
| 1046 |
|
| 1047 |
qcow2_encrypt_sectors(s, sector_num, cluster_data, |
| 1048 |
cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
|
| 1049 |
|
| 1050 |
qemu_iovec_reset(&hd_qiov); |
| 1051 |
qemu_iovec_add(&hd_qiov, cluster_data, |
| 1052 |
cur_nr_sectors * 512);
|
| 1053 |
} |
| 1054 |
|
| 1055 |
ret = qcow2_pre_write_overlap_check(bs, 0,
|
| 1056 |
cluster_offset + index_in_cluster * BDRV_SECTOR_SIZE, |
| 1057 |
cur_nr_sectors * BDRV_SECTOR_SIZE); |
| 1058 |
if (ret < 0) { |
| 1059 |
goto fail;
|
| 1060 |
} |
| 1061 |
|
| 1062 |
qemu_co_mutex_unlock(&s->lock); |
| 1063 |
BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO); |
| 1064 |
trace_qcow2_writev_data(qemu_coroutine_self(), |
| 1065 |
(cluster_offset >> 9) + index_in_cluster);
|
| 1066 |
ret = bdrv_co_writev(bs->file, |
| 1067 |
(cluster_offset >> 9) + index_in_cluster,
|
| 1068 |
cur_nr_sectors, &hd_qiov); |
| 1069 |
qemu_co_mutex_lock(&s->lock); |
| 1070 |
if (ret < 0) { |
| 1071 |
goto fail;
|
| 1072 |
} |
| 1073 |
|
| 1074 |
while (l2meta != NULL) { |
| 1075 |
QCowL2Meta *next; |
| 1076 |
|
| 1077 |
ret = qcow2_alloc_cluster_link_l2(bs, l2meta); |
| 1078 |
if (ret < 0) { |
| 1079 |
goto fail;
|
| 1080 |
} |
| 1081 |
|
| 1082 |
/* Take the request off the list of running requests */
|
| 1083 |
if (l2meta->nb_clusters != 0) { |
| 1084 |
QLIST_REMOVE(l2meta, next_in_flight); |
| 1085 |
} |
| 1086 |
|
| 1087 |
qemu_co_queue_restart_all(&l2meta->dependent_requests); |
| 1088 |
|
| 1089 |
next = l2meta->next; |
| 1090 |
g_free(l2meta); |
| 1091 |
l2meta = next; |
| 1092 |
} |
| 1093 |
|
| 1094 |
remaining_sectors -= cur_nr_sectors; |
| 1095 |
sector_num += cur_nr_sectors; |
| 1096 |
bytes_done += cur_nr_sectors * 512;
|
| 1097 |
trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors); |
| 1098 |
} |
| 1099 |
ret = 0;
|
| 1100 |
|
| 1101 |
fail:
|
| 1102 |
qemu_co_mutex_unlock(&s->lock); |
| 1103 |
|
| 1104 |
while (l2meta != NULL) { |
| 1105 |
QCowL2Meta *next; |
| 1106 |
|
| 1107 |
if (l2meta->nb_clusters != 0) { |
| 1108 |
QLIST_REMOVE(l2meta, next_in_flight); |
| 1109 |
} |
| 1110 |
qemu_co_queue_restart_all(&l2meta->dependent_requests); |
| 1111 |
|
| 1112 |
next = l2meta->next; |
| 1113 |
g_free(l2meta); |
| 1114 |
l2meta = next; |
| 1115 |
} |
| 1116 |
|
| 1117 |
qemu_iovec_destroy(&hd_qiov); |
| 1118 |
qemu_vfree(cluster_data); |
| 1119 |
trace_qcow2_writev_done_req(qemu_coroutine_self(), ret); |
| 1120 |
|
| 1121 |
return ret;
|
| 1122 |
} |
| 1123 |
|
| 1124 |
static void qcow2_close(BlockDriverState *bs) |
| 1125 |
{
|
| 1126 |
BDRVQcowState *s = bs->opaque; |
| 1127 |
g_free(s->l1_table); |
| 1128 |
/* else pre-write overlap checks in cache_destroy may crash */
|
| 1129 |
s->l1_table = NULL;
|
| 1130 |
|
| 1131 |
qcow2_cache_flush(bs, s->l2_table_cache); |
| 1132 |
qcow2_cache_flush(bs, s->refcount_block_cache); |
| 1133 |
|
| 1134 |
qcow2_mark_clean(bs); |
| 1135 |
|
| 1136 |
qcow2_cache_destroy(bs, s->l2_table_cache); |
| 1137 |
qcow2_cache_destroy(bs, s->refcount_block_cache); |
| 1138 |
|
| 1139 |
g_free(s->unknown_header_fields); |
| 1140 |
cleanup_unknown_header_ext(bs); |
| 1141 |
|
| 1142 |
g_free(s->cluster_cache); |
| 1143 |
qemu_vfree(s->cluster_data); |
| 1144 |
qcow2_refcount_close(bs); |
| 1145 |
qcow2_free_snapshots(bs); |
| 1146 |
} |
| 1147 |
|
| 1148 |
static void qcow2_invalidate_cache(BlockDriverState *bs) |
| 1149 |
{
|
| 1150 |
BDRVQcowState *s = bs->opaque; |
| 1151 |
int flags = s->flags;
|
| 1152 |
AES_KEY aes_encrypt_key; |
| 1153 |
AES_KEY aes_decrypt_key; |
| 1154 |
uint32_t crypt_method = 0;
|
| 1155 |
QDict *options; |
| 1156 |
|
| 1157 |
/*
|
| 1158 |
* Backing files are read-only which makes all of their metadata immutable,
|
| 1159 |
* that means we don't have to worry about reopening them here.
|
| 1160 |
*/
|
| 1161 |
|
| 1162 |
if (s->crypt_method) {
|
| 1163 |
crypt_method = s->crypt_method; |
| 1164 |
memcpy(&aes_encrypt_key, &s->aes_encrypt_key, sizeof(aes_encrypt_key));
|
| 1165 |
memcpy(&aes_decrypt_key, &s->aes_decrypt_key, sizeof(aes_decrypt_key));
|
| 1166 |
} |
| 1167 |
|
| 1168 |
qcow2_close(bs); |
| 1169 |
|
| 1170 |
options = qdict_new(); |
| 1171 |
qdict_put(options, QCOW2_OPT_LAZY_REFCOUNTS, |
| 1172 |
qbool_from_int(s->use_lazy_refcounts)); |
| 1173 |
|
| 1174 |
memset(s, 0, sizeof(BDRVQcowState)); |
| 1175 |
qcow2_open(bs, options, flags, NULL);
|
| 1176 |
|
| 1177 |
QDECREF(options); |
| 1178 |
|
| 1179 |
if (crypt_method) {
|
| 1180 |
s->crypt_method = crypt_method; |
| 1181 |
memcpy(&s->aes_encrypt_key, &aes_encrypt_key, sizeof(aes_encrypt_key));
|
| 1182 |
memcpy(&s->aes_decrypt_key, &aes_decrypt_key, sizeof(aes_decrypt_key));
|
| 1183 |
} |
| 1184 |
} |
| 1185 |
|
| 1186 |
static size_t header_ext_add(char *buf, uint32_t magic, const void *s, |
| 1187 |
size_t len, size_t buflen) |
| 1188 |
{
|
| 1189 |
QCowExtension *ext_backing_fmt = (QCowExtension*) buf; |
| 1190 |
size_t ext_len = sizeof(QCowExtension) + ((len + 7) & ~7); |
| 1191 |
|
| 1192 |
if (buflen < ext_len) {
|
| 1193 |
return -ENOSPC;
|
| 1194 |
} |
| 1195 |
|
| 1196 |
*ext_backing_fmt = (QCowExtension) {
|
| 1197 |
.magic = cpu_to_be32(magic), |
| 1198 |
.len = cpu_to_be32(len), |
| 1199 |
}; |
| 1200 |
memcpy(buf + sizeof(QCowExtension), s, len);
|
| 1201 |
|
| 1202 |
return ext_len;
|
| 1203 |
} |
| 1204 |
|
| 1205 |
/*
|
| 1206 |
* Updates the qcow2 header, including the variable length parts of it, i.e.
|
| 1207 |
* the backing file name and all extensions. qcow2 was not designed to allow
|
| 1208 |
* such changes, so if we run out of space (we can only use the first cluster)
|
| 1209 |
* this function may fail.
|
| 1210 |
*
|
| 1211 |
* Returns 0 on success, -errno in error cases.
|
| 1212 |
*/
|
| 1213 |
int qcow2_update_header(BlockDriverState *bs)
|
| 1214 |
{
|
| 1215 |
BDRVQcowState *s = bs->opaque; |
| 1216 |
QCowHeader *header; |
| 1217 |
char *buf;
|
| 1218 |
size_t buflen = s->cluster_size; |
| 1219 |
int ret;
|
| 1220 |
uint64_t total_size; |
| 1221 |
uint32_t refcount_table_clusters; |
| 1222 |
size_t header_length; |
| 1223 |
Qcow2UnknownHeaderExtension *uext; |
| 1224 |
|
| 1225 |
buf = qemu_blockalign(bs, buflen); |
| 1226 |
|
| 1227 |
/* Header structure */
|
| 1228 |
header = (QCowHeader*) buf; |
| 1229 |
|
| 1230 |
if (buflen < sizeof(*header)) { |
| 1231 |
ret = -ENOSPC; |
| 1232 |
goto fail;
|
| 1233 |
} |
| 1234 |
|
| 1235 |
header_length = sizeof(*header) + s->unknown_header_fields_size;
|
| 1236 |
total_size = bs->total_sectors * BDRV_SECTOR_SIZE; |
| 1237 |
refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
|
| 1238 |
|
| 1239 |
*header = (QCowHeader) {
|
| 1240 |
/* Version 2 fields */
|
| 1241 |
.magic = cpu_to_be32(QCOW_MAGIC), |
| 1242 |
.version = cpu_to_be32(s->qcow_version), |
| 1243 |
.backing_file_offset = 0,
|
| 1244 |
.backing_file_size = 0,
|
| 1245 |
.cluster_bits = cpu_to_be32(s->cluster_bits), |
| 1246 |
.size = cpu_to_be64(total_size), |
| 1247 |
.crypt_method = cpu_to_be32(s->crypt_method_header), |
| 1248 |
.l1_size = cpu_to_be32(s->l1_size), |
| 1249 |
.l1_table_offset = cpu_to_be64(s->l1_table_offset), |
| 1250 |
.refcount_table_offset = cpu_to_be64(s->refcount_table_offset), |
| 1251 |
.refcount_table_clusters = cpu_to_be32(refcount_table_clusters), |
| 1252 |
.nb_snapshots = cpu_to_be32(s->nb_snapshots), |
| 1253 |
.snapshots_offset = cpu_to_be64(s->snapshots_offset), |
| 1254 |
|
| 1255 |
/* Version 3 fields */
|
| 1256 |
.incompatible_features = cpu_to_be64(s->incompatible_features), |
| 1257 |
.compatible_features = cpu_to_be64(s->compatible_features), |
| 1258 |
.autoclear_features = cpu_to_be64(s->autoclear_features), |
| 1259 |
.refcount_order = cpu_to_be32(s->refcount_order), |
| 1260 |
.header_length = cpu_to_be32(header_length), |
| 1261 |
}; |
| 1262 |
|
| 1263 |
/* For older versions, write a shorter header */
|
| 1264 |
switch (s->qcow_version) {
|
| 1265 |
case 2: |
| 1266 |
ret = offsetof(QCowHeader, incompatible_features); |
| 1267 |
break;
|
| 1268 |
case 3: |
| 1269 |
ret = sizeof(*header);
|
| 1270 |
break;
|
| 1271 |
default:
|
| 1272 |
ret = -EINVAL; |
| 1273 |
goto fail;
|
| 1274 |
} |
| 1275 |
|
| 1276 |
buf += ret; |
| 1277 |
buflen -= ret; |
| 1278 |
memset(buf, 0, buflen);
|
| 1279 |
|
| 1280 |
/* Preserve any unknown field in the header */
|
| 1281 |
if (s->unknown_header_fields_size) {
|
| 1282 |
if (buflen < s->unknown_header_fields_size) {
|
| 1283 |
ret = -ENOSPC; |
| 1284 |
goto fail;
|
| 1285 |
} |
| 1286 |
|
| 1287 |
memcpy(buf, s->unknown_header_fields, s->unknown_header_fields_size); |
| 1288 |
buf += s->unknown_header_fields_size; |
| 1289 |
buflen -= s->unknown_header_fields_size; |
| 1290 |
} |
| 1291 |
|
| 1292 |
/* Backing file format header extension */
|
| 1293 |
if (*bs->backing_format) {
|
| 1294 |
ret = header_ext_add(buf, QCOW2_EXT_MAGIC_BACKING_FORMAT, |
| 1295 |
bs->backing_format, strlen(bs->backing_format), |
| 1296 |
buflen); |
| 1297 |
if (ret < 0) { |
| 1298 |
goto fail;
|
| 1299 |
} |
| 1300 |
|
| 1301 |
buf += ret; |
| 1302 |
buflen -= ret; |
| 1303 |
} |
| 1304 |
|
| 1305 |
/* Feature table */
|
| 1306 |
Qcow2Feature features[] = {
|
| 1307 |
{
|
| 1308 |
.type = QCOW2_FEAT_TYPE_INCOMPATIBLE, |
| 1309 |
.bit = QCOW2_INCOMPAT_DIRTY_BITNR, |
| 1310 |
.name = "dirty bit",
|
| 1311 |
}, |
| 1312 |
{
|
| 1313 |
.type = QCOW2_FEAT_TYPE_INCOMPATIBLE, |
| 1314 |
.bit = QCOW2_INCOMPAT_CORRUPT_BITNR, |
| 1315 |
.name = "corrupt bit",
|
| 1316 |
}, |
| 1317 |
{
|
| 1318 |
.type = QCOW2_FEAT_TYPE_COMPATIBLE, |
| 1319 |
.bit = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR, |
| 1320 |
.name = "lazy refcounts",
|
| 1321 |
}, |
| 1322 |
}; |
| 1323 |
|
| 1324 |
ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE, |
| 1325 |
features, sizeof(features), buflen);
|
| 1326 |
if (ret < 0) { |
| 1327 |
goto fail;
|
| 1328 |
} |
| 1329 |
buf += ret; |
| 1330 |
buflen -= ret; |
| 1331 |
|
| 1332 |
/* Keep unknown header extensions */
|
| 1333 |
QLIST_FOREACH(uext, &s->unknown_header_ext, next) {
|
| 1334 |
ret = header_ext_add(buf, uext->magic, uext->data, uext->len, buflen); |
| 1335 |
if (ret < 0) { |
| 1336 |
goto fail;
|
| 1337 |
} |
| 1338 |
|
| 1339 |
buf += ret; |
| 1340 |
buflen -= ret; |
| 1341 |
} |
| 1342 |
|
| 1343 |
/* End of header extensions */
|
| 1344 |
ret = header_ext_add(buf, QCOW2_EXT_MAGIC_END, NULL, 0, buflen); |
| 1345 |
if (ret < 0) { |
| 1346 |
goto fail;
|
| 1347 |
} |
| 1348 |
|
| 1349 |
buf += ret; |
| 1350 |
buflen -= ret; |
| 1351 |
|
| 1352 |
/* Backing file name */
|
| 1353 |
if (*bs->backing_file) {
|
| 1354 |
size_t backing_file_len = strlen(bs->backing_file); |
| 1355 |
|
| 1356 |
if (buflen < backing_file_len) {
|
| 1357 |
ret = -ENOSPC; |
| 1358 |
goto fail;
|
| 1359 |
} |
| 1360 |
|
| 1361 |
/* Using strncpy is ok here, since buf is not NUL-terminated. */
|
| 1362 |
strncpy(buf, bs->backing_file, buflen); |
| 1363 |
|
| 1364 |
header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
|
| 1365 |
header->backing_file_size = cpu_to_be32(backing_file_len); |
| 1366 |
} |
| 1367 |
|
| 1368 |
/* Write the new header */
|
| 1369 |
ret = bdrv_pwrite(bs->file, 0, header, s->cluster_size);
|
| 1370 |
if (ret < 0) { |
| 1371 |
goto fail;
|
| 1372 |
} |
| 1373 |
|
| 1374 |
ret = 0;
|
| 1375 |
fail:
|
| 1376 |
qemu_vfree(header); |
| 1377 |
return ret;
|
| 1378 |
} |
| 1379 |
|
| 1380 |
static int qcow2_change_backing_file(BlockDriverState *bs, |
| 1381 |
const char *backing_file, const char *backing_fmt) |
| 1382 |
{
|
| 1383 |
pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); |
| 1384 |
pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); |
| 1385 |
|
| 1386 |
return qcow2_update_header(bs);
|
| 1387 |
} |
| 1388 |
|
| 1389 |
static int preallocate(BlockDriverState *bs) |
| 1390 |
{
|
| 1391 |
uint64_t nb_sectors; |
| 1392 |
uint64_t offset; |
| 1393 |
uint64_t host_offset = 0;
|
| 1394 |
int num;
|
| 1395 |
int ret;
|
| 1396 |
QCowL2Meta *meta; |
| 1397 |
|
| 1398 |
nb_sectors = bdrv_getlength(bs) >> 9;
|
| 1399 |
offset = 0;
|
| 1400 |
|
| 1401 |
while (nb_sectors) {
|
| 1402 |
num = MIN(nb_sectors, INT_MAX >> 9);
|
| 1403 |
ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num,
|
| 1404 |
&host_offset, &meta); |
| 1405 |
if (ret < 0) { |
| 1406 |
return ret;
|
| 1407 |
} |
| 1408 |
|
| 1409 |
ret = qcow2_alloc_cluster_link_l2(bs, meta); |
| 1410 |
if (ret < 0) { |
| 1411 |
qcow2_free_any_clusters(bs, meta->alloc_offset, meta->nb_clusters, |
| 1412 |
QCOW2_DISCARD_NEVER); |
| 1413 |
return ret;
|
| 1414 |
} |
| 1415 |
|
| 1416 |
/* There are no dependent requests, but we need to remove our request
|
| 1417 |
* from the list of in-flight requests */
|
| 1418 |
if (meta != NULL) { |
| 1419 |
QLIST_REMOVE(meta, next_in_flight); |
| 1420 |
} |
| 1421 |
|
| 1422 |
/* TODO Preallocate data if requested */
|
| 1423 |
|
| 1424 |
nb_sectors -= num; |
| 1425 |
offset += num << 9;
|
| 1426 |
} |
| 1427 |
|
| 1428 |
/*
|
| 1429 |
* It is expected that the image file is large enough to actually contain
|
| 1430 |
* all of the allocated clusters (otherwise we get failing reads after
|
| 1431 |
* EOF). Extend the image to the last allocated sector.
|
| 1432 |
*/
|
| 1433 |
if (host_offset != 0) { |
| 1434 |
uint8_t buf[512];
|
| 1435 |
memset(buf, 0, 512); |
| 1436 |
ret = bdrv_write(bs->file, (host_offset >> 9) + num - 1, buf, 1); |
| 1437 |
if (ret < 0) { |
| 1438 |
return ret;
|
| 1439 |
} |
| 1440 |
} |
| 1441 |
|
| 1442 |
return 0; |
| 1443 |
} |
| 1444 |
|
| 1445 |
static int qcow2_create2(const char *filename, int64_t total_size, |
| 1446 |
const char *backing_file, const char *backing_format, |
| 1447 |
int flags, size_t cluster_size, int prealloc, |
| 1448 |
QEMUOptionParameter *options, int version,
|
| 1449 |
Error **errp) |
| 1450 |
{
|
| 1451 |
/* Calculate cluster_bits */
|
| 1452 |
int cluster_bits;
|
| 1453 |
cluster_bits = ffs(cluster_size) - 1;
|
| 1454 |
if (cluster_bits < MIN_CLUSTER_BITS || cluster_bits > MAX_CLUSTER_BITS ||
|
| 1455 |
(1 << cluster_bits) != cluster_size)
|
| 1456 |
{
|
| 1457 |
error_setg(errp, "Cluster size must be a power of two between %d and "
|
| 1458 |
"%dk", 1 << MIN_CLUSTER_BITS, 1 << (MAX_CLUSTER_BITS - 10)); |
| 1459 |
return -EINVAL;
|
| 1460 |
} |
| 1461 |
|
| 1462 |
/*
|
| 1463 |
* Open the image file and write a minimal qcow2 header.
|
| 1464 |
*
|
| 1465 |
* We keep things simple and start with a zero-sized image. We also
|
| 1466 |
* do without refcount blocks or a L1 table for now. We'll fix the
|
| 1467 |
* inconsistency later.
|
| 1468 |
*
|
| 1469 |
* We do need a refcount table because growing the refcount table means
|
| 1470 |
* allocating two new refcount blocks - the seconds of which would be at
|
| 1471 |
* 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
|
| 1472 |
* size for any qcow2 image.
|
| 1473 |
*/
|
| 1474 |
BlockDriverState* bs; |
| 1475 |
QCowHeader *header; |
| 1476 |
uint8_t* refcount_table; |
| 1477 |
Error *local_err = NULL;
|
| 1478 |
int ret;
|
| 1479 |
|
| 1480 |
ret = bdrv_create_file(filename, options, &local_err); |
| 1481 |
if (ret < 0) { |
| 1482 |
error_propagate(errp, local_err); |
| 1483 |
return ret;
|
| 1484 |
} |
| 1485 |
|
| 1486 |
ret = bdrv_file_open(&bs, filename, NULL, NULL, BDRV_O_RDWR, &local_err); |
| 1487 |
if (ret < 0) { |
| 1488 |
error_propagate(errp, local_err); |
| 1489 |
return ret;
|
| 1490 |
} |
| 1491 |
|
| 1492 |
/* Write the header */
|
| 1493 |
QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS) < sizeof(*header)); |
| 1494 |
header = g_malloc0(cluster_size); |
| 1495 |
*header = (QCowHeader) {
|
| 1496 |
.magic = cpu_to_be32(QCOW_MAGIC), |
| 1497 |
.version = cpu_to_be32(version), |
| 1498 |
.cluster_bits = cpu_to_be32(cluster_bits), |
| 1499 |
.size = cpu_to_be64(0),
|
| 1500 |
.l1_table_offset = cpu_to_be64(0),
|
| 1501 |
.l1_size = cpu_to_be32(0),
|
| 1502 |
.refcount_table_offset = cpu_to_be64(cluster_size), |
| 1503 |
.refcount_table_clusters = cpu_to_be32(1),
|
| 1504 |
.refcount_order = cpu_to_be32(3 + REFCOUNT_SHIFT),
|
| 1505 |
.header_length = cpu_to_be32(sizeof(*header)),
|
| 1506 |
}; |
| 1507 |
|
| 1508 |
if (flags & BLOCK_FLAG_ENCRYPT) {
|
| 1509 |
header->crypt_method = cpu_to_be32(QCOW_CRYPT_AES); |
| 1510 |
} else {
|
| 1511 |
header->crypt_method = cpu_to_be32(QCOW_CRYPT_NONE); |
| 1512 |
} |
| 1513 |
|
| 1514 |
if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
|
| 1515 |
header->compatible_features |= |
| 1516 |
cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS); |
| 1517 |
} |
| 1518 |
|
| 1519 |
ret = bdrv_pwrite(bs, 0, header, cluster_size);
|
| 1520 |
g_free(header); |
| 1521 |
if (ret < 0) { |
| 1522 |
error_setg_errno(errp, -ret, "Could not write qcow2 header");
|
| 1523 |
goto out;
|
| 1524 |
} |
| 1525 |
|
| 1526 |
/* Write an empty refcount table */
|
| 1527 |
refcount_table = g_malloc0(cluster_size); |
| 1528 |
ret = bdrv_pwrite(bs, cluster_size, refcount_table, cluster_size); |
| 1529 |
g_free(refcount_table); |
| 1530 |
|
| 1531 |
if (ret < 0) { |
| 1532 |
error_setg_errno(errp, -ret, "Could not write refcount table");
|
| 1533 |
goto out;
|
| 1534 |
} |
| 1535 |
|
| 1536 |
bdrv_close(bs); |
| 1537 |
|
| 1538 |
/*
|
| 1539 |
* And now open the image and make it consistent first (i.e. increase the
|
| 1540 |
* refcount of the cluster that is occupied by the header and the refcount
|
| 1541 |
* table)
|
| 1542 |
*/
|
| 1543 |
BlockDriver* drv = bdrv_find_format("qcow2");
|
| 1544 |
assert(drv != NULL);
|
| 1545 |
ret = bdrv_open(bs, filename, NULL,
|
| 1546 |
BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_FLUSH, drv, &local_err); |
| 1547 |
if (ret < 0) { |
| 1548 |
error_propagate(errp, local_err); |
| 1549 |
goto out;
|
| 1550 |
} |
| 1551 |
|
| 1552 |
ret = qcow2_alloc_clusters(bs, 2 * cluster_size);
|
| 1553 |
if (ret < 0) { |
| 1554 |
error_setg_errno(errp, -ret, "Could not allocate clusters for qcow2 "
|
| 1555 |
"header and refcount table");
|
| 1556 |
goto out;
|
| 1557 |
|
| 1558 |
} else if (ret != 0) { |
| 1559 |
error_report("Huh, first cluster in empty image is already in use?");
|
| 1560 |
abort(); |
| 1561 |
} |
| 1562 |
|
| 1563 |
/* Okay, now that we have a valid image, let's give it the right size */
|
| 1564 |
ret = bdrv_truncate(bs, total_size * BDRV_SECTOR_SIZE); |
| 1565 |
if (ret < 0) { |
| 1566 |
error_setg_errno(errp, -ret, "Could not resize image");
|
| 1567 |
goto out;
|
| 1568 |
} |
| 1569 |
|
| 1570 |
/* Want a backing file? There you go.*/
|
| 1571 |
if (backing_file) {
|
| 1572 |
ret = bdrv_change_backing_file(bs, backing_file, backing_format); |
| 1573 |
if (ret < 0) { |
| 1574 |
error_setg_errno(errp, -ret, "Could not assign backing file '%s' "
|
| 1575 |
"with format '%s'", backing_file, backing_format);
|
| 1576 |
goto out;
|
| 1577 |
} |
| 1578 |
} |
| 1579 |
|
| 1580 |
/* And if we're supposed to preallocate metadata, do that now */
|
| 1581 |
if (prealloc) {
|
| 1582 |
BDRVQcowState *s = bs->opaque; |
| 1583 |
qemu_co_mutex_lock(&s->lock); |
| 1584 |
ret = preallocate(bs); |
| 1585 |
qemu_co_mutex_unlock(&s->lock); |
| 1586 |
if (ret < 0) { |
| 1587 |
error_setg_errno(errp, -ret, "Could not preallocate metadata");
|
| 1588 |
goto out;
|
| 1589 |
} |
| 1590 |
} |
| 1591 |
|
| 1592 |
bdrv_close(bs); |
| 1593 |
|
| 1594 |
/* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning */
|
| 1595 |
ret = bdrv_open(bs, filename, NULL,
|
| 1596 |
BDRV_O_RDWR | BDRV_O_CACHE_WB | BDRV_O_NO_BACKING, |
| 1597 |
drv, &local_err); |
| 1598 |
if (error_is_set(&local_err)) {
|
| 1599 |
error_propagate(errp, local_err); |
| 1600 |
goto out;
|
| 1601 |
} |
| 1602 |
|
| 1603 |
ret = 0;
|
| 1604 |
out:
|
| 1605 |
bdrv_unref(bs); |
| 1606 |
return ret;
|
| 1607 |
} |
| 1608 |
|
| 1609 |
static int qcow2_create(const char *filename, QEMUOptionParameter *options, |
| 1610 |
Error **errp) |
| 1611 |
{
|
| 1612 |
const char *backing_file = NULL; |
| 1613 |
const char *backing_fmt = NULL; |
| 1614 |
uint64_t sectors = 0;
|
| 1615 |
int flags = 0; |
| 1616 |
size_t cluster_size = DEFAULT_CLUSTER_SIZE; |
| 1617 |
int prealloc = 0; |
| 1618 |
int version = 3; |
| 1619 |
Error *local_err = NULL;
|
| 1620 |
int ret;
|
| 1621 |
|
| 1622 |
/* Read out options */
|
| 1623 |
while (options && options->name) {
|
| 1624 |
if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
|
| 1625 |
sectors = options->value.n / 512;
|
| 1626 |
} else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { |
| 1627 |
backing_file = options->value.s; |
| 1628 |
} else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) { |
| 1629 |
backing_fmt = options->value.s; |
| 1630 |
} else if (!strcmp(options->name, BLOCK_OPT_ENCRYPT)) { |
| 1631 |
flags |= options->value.n ? BLOCK_FLAG_ENCRYPT : 0;
|
| 1632 |
} else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) { |
| 1633 |
if (options->value.n) {
|
| 1634 |
cluster_size = options->value.n; |
| 1635 |
} |
| 1636 |
} else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) { |
| 1637 |
if (!options->value.s || !strcmp(options->value.s, "off")) { |
| 1638 |
prealloc = 0;
|
| 1639 |
} else if (!strcmp(options->value.s, "metadata")) { |
| 1640 |
prealloc = 1;
|
| 1641 |
} else {
|
| 1642 |
error_setg(errp, "Invalid preallocation mode: '%s'",
|
| 1643 |
options->value.s); |
| 1644 |
return -EINVAL;
|
| 1645 |
} |
| 1646 |
} else if (!strcmp(options->name, BLOCK_OPT_COMPAT_LEVEL)) { |
| 1647 |
if (!options->value.s) {
|
| 1648 |
/* keep the default */
|
| 1649 |
} else if (!strcmp(options->value.s, "0.10")) { |
| 1650 |
version = 2;
|
| 1651 |
} else if (!strcmp(options->value.s, "1.1")) { |
| 1652 |
version = 3;
|
| 1653 |
} else {
|
| 1654 |
error_setg(errp, "Invalid compatibility level: '%s'",
|
| 1655 |
options->value.s); |
| 1656 |
return -EINVAL;
|
| 1657 |
} |
| 1658 |
} else if (!strcmp(options->name, BLOCK_OPT_LAZY_REFCOUNTS)) { |
| 1659 |
flags |= options->value.n ? BLOCK_FLAG_LAZY_REFCOUNTS : 0;
|
| 1660 |
} |
| 1661 |
options++; |
| 1662 |
} |
| 1663 |
|
| 1664 |
if (backing_file && prealloc) {
|
| 1665 |
error_setg(errp, "Backing file and preallocation cannot be used at "
|
| 1666 |
"the same time");
|
| 1667 |
return -EINVAL;
|
| 1668 |
} |
| 1669 |
|
| 1670 |
if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) { |
| 1671 |
error_setg(errp, "Lazy refcounts only supported with compatibility "
|
| 1672 |
"level 1.1 and above (use compat=1.1 or greater)");
|
| 1673 |
return -EINVAL;
|
| 1674 |
} |
| 1675 |
|
| 1676 |
ret = qcow2_create2(filename, sectors, backing_file, backing_fmt, flags, |
| 1677 |
cluster_size, prealloc, options, version, &local_err); |
| 1678 |
if (error_is_set(&local_err)) {
|
| 1679 |
error_propagate(errp, local_err); |
| 1680 |
} |
| 1681 |
return ret;
|
| 1682 |
} |
| 1683 |
|
| 1684 |
static int qcow2_make_empty(BlockDriverState *bs) |
| 1685 |
{
|
| 1686 |
#if 0
|
| 1687 |
/* XXX: not correct */
|
| 1688 |
BDRVQcowState *s = bs->opaque;
|
| 1689 |
uint32_t l1_length = s->l1_size * sizeof(uint64_t);
|
| 1690 |
int ret;
|
| 1691 |
|
| 1692 |
memset(s->l1_table, 0, l1_length);
|
| 1693 |
if (bdrv_pwrite(bs->file, s->l1_table_offset, s->l1_table, l1_length) < 0)
|
| 1694 |
return -1;
|
| 1695 |
ret = bdrv_truncate(bs->file, s->l1_table_offset + l1_length);
|
| 1696 |
if (ret < 0)
|
| 1697 |
return ret;
|
| 1698 |
|
| 1699 |
l2_cache_reset(bs);
|
| 1700 |
#endif
|
| 1701 |
return 0; |
| 1702 |
} |
| 1703 |
|
| 1704 |
static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs, |
| 1705 |
int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
|
| 1706 |
{
|
| 1707 |
int ret;
|
| 1708 |
BDRVQcowState *s = bs->opaque; |
| 1709 |
|
| 1710 |
/* Emulate misaligned zero writes */
|
| 1711 |
if (sector_num % s->cluster_sectors || nb_sectors % s->cluster_sectors) {
|
| 1712 |
return -ENOTSUP;
|
| 1713 |
} |
| 1714 |
|
| 1715 |
/* Whatever is left can use real zero clusters */
|
| 1716 |
qemu_co_mutex_lock(&s->lock); |
| 1717 |
ret = qcow2_zero_clusters(bs, sector_num << BDRV_SECTOR_BITS, |
| 1718 |
nb_sectors); |
| 1719 |
qemu_co_mutex_unlock(&s->lock); |
| 1720 |
|
| 1721 |
return ret;
|
| 1722 |
} |
| 1723 |
|
| 1724 |
static coroutine_fn int qcow2_co_discard(BlockDriverState *bs, |
| 1725 |
int64_t sector_num, int nb_sectors)
|
| 1726 |
{
|
| 1727 |
int ret;
|
| 1728 |
BDRVQcowState *s = bs->opaque; |
| 1729 |
|
| 1730 |
qemu_co_mutex_lock(&s->lock); |
| 1731 |
ret = qcow2_discard_clusters(bs, sector_num << BDRV_SECTOR_BITS, |
| 1732 |
nb_sectors, QCOW2_DISCARD_REQUEST); |
| 1733 |
qemu_co_mutex_unlock(&s->lock); |
| 1734 |
return ret;
|
| 1735 |
} |
| 1736 |
|
| 1737 |
static int qcow2_truncate(BlockDriverState *bs, int64_t offset) |
| 1738 |
{
|
| 1739 |
BDRVQcowState *s = bs->opaque; |
| 1740 |
int64_t new_l1_size; |
| 1741 |
int ret;
|
| 1742 |
|
| 1743 |
if (offset & 511) { |
| 1744 |
error_report("The new size must be a multiple of 512");
|
| 1745 |
return -EINVAL;
|
| 1746 |
} |
| 1747 |
|
| 1748 |
/* cannot proceed if image has snapshots */
|
| 1749 |
if (s->nb_snapshots) {
|
| 1750 |
error_report("Can't resize an image which has snapshots");
|
| 1751 |
return -ENOTSUP;
|
| 1752 |
} |
| 1753 |
|
| 1754 |
/* shrinking is currently not supported */
|
| 1755 |
if (offset < bs->total_sectors * 512) { |
| 1756 |
error_report("qcow2 doesn't support shrinking images yet");
|
| 1757 |
return -ENOTSUP;
|
| 1758 |
} |
| 1759 |
|
| 1760 |
new_l1_size = size_to_l1(s, offset); |
| 1761 |
ret = qcow2_grow_l1_table(bs, new_l1_size, true);
|
| 1762 |
if (ret < 0) { |
| 1763 |
return ret;
|
| 1764 |
} |
| 1765 |
|
| 1766 |
/* write updated header.size */
|
| 1767 |
offset = cpu_to_be64(offset); |
| 1768 |
ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, size), |
| 1769 |
&offset, sizeof(uint64_t));
|
| 1770 |
if (ret < 0) { |
| 1771 |
return ret;
|
| 1772 |
} |
| 1773 |
|
| 1774 |
s->l1_vm_state_index = new_l1_size; |
| 1775 |
return 0; |
| 1776 |
} |
| 1777 |
|
| 1778 |
/* XXX: put compressed sectors first, then all the cluster aligned
|
| 1779 |
tables to avoid losing bytes in alignment */
|
| 1780 |
static int qcow2_write_compressed(BlockDriverState *bs, int64_t sector_num, |
| 1781 |
const uint8_t *buf, int nb_sectors) |
| 1782 |
{
|
| 1783 |
BDRVQcowState *s = bs->opaque; |
| 1784 |
z_stream strm; |
| 1785 |
int ret, out_len;
|
| 1786 |
uint8_t *out_buf; |
| 1787 |
uint64_t cluster_offset; |
| 1788 |
|
| 1789 |
if (nb_sectors == 0) { |
| 1790 |
/* align end of file to a sector boundary to ease reading with
|
| 1791 |
sector based I/Os */
|
| 1792 |
cluster_offset = bdrv_getlength(bs->file); |
| 1793 |
cluster_offset = (cluster_offset + 511) & ~511; |
| 1794 |
bdrv_truncate(bs->file, cluster_offset); |
| 1795 |
return 0; |
| 1796 |
} |
| 1797 |
|
| 1798 |
if (nb_sectors != s->cluster_sectors) {
|
| 1799 |
ret = -EINVAL; |
| 1800 |
|
| 1801 |
/* Zero-pad last write if image size is not cluster aligned */
|
| 1802 |
if (sector_num + nb_sectors == bs->total_sectors &&
|
| 1803 |
nb_sectors < s->cluster_sectors) {
|
| 1804 |
uint8_t *pad_buf = qemu_blockalign(bs, s->cluster_size); |
| 1805 |
memset(pad_buf, 0, s->cluster_size);
|
| 1806 |
memcpy(pad_buf, buf, nb_sectors * BDRV_SECTOR_SIZE); |
| 1807 |
ret = qcow2_write_compressed(bs, sector_num, |
| 1808 |
pad_buf, s->cluster_sectors); |
| 1809 |
qemu_vfree(pad_buf); |
| 1810 |
} |
| 1811 |
return ret;
|
| 1812 |
} |
| 1813 |
|
| 1814 |
out_buf = g_malloc(s->cluster_size + (s->cluster_size / 1000) + 128); |
| 1815 |
|
| 1816 |
/* best compression, small window, no zlib header */
|
| 1817 |
memset(&strm, 0, sizeof(strm)); |
| 1818 |
ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION, |
| 1819 |
Z_DEFLATED, -12,
|
| 1820 |
9, Z_DEFAULT_STRATEGY);
|
| 1821 |
if (ret != 0) { |
| 1822 |
ret = -EINVAL; |
| 1823 |
goto fail;
|
| 1824 |
} |
| 1825 |
|
| 1826 |
strm.avail_in = s->cluster_size; |
| 1827 |
strm.next_in = (uint8_t *)buf; |
| 1828 |
strm.avail_out = s->cluster_size; |
| 1829 |
strm.next_out = out_buf; |
| 1830 |
|
| 1831 |
ret = deflate(&strm, Z_FINISH); |
| 1832 |
if (ret != Z_STREAM_END && ret != Z_OK) {
|
| 1833 |
deflateEnd(&strm); |
| 1834 |
ret = -EINVAL; |
| 1835 |
goto fail;
|
| 1836 |
} |
| 1837 |
out_len = strm.next_out - out_buf; |
| 1838 |
|
| 1839 |
deflateEnd(&strm); |
| 1840 |
|
| 1841 |
if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
|
| 1842 |
/* could not compress: write normal cluster */
|
| 1843 |
ret = bdrv_write(bs, sector_num, buf, s->cluster_sectors); |
| 1844 |
if (ret < 0) { |
| 1845 |
goto fail;
|
| 1846 |
} |
| 1847 |
} else {
|
| 1848 |
cluster_offset = qcow2_alloc_compressed_cluster_offset(bs, |
| 1849 |
sector_num << 9, out_len);
|
| 1850 |
if (!cluster_offset) {
|
| 1851 |
ret = -EIO; |
| 1852 |
goto fail;
|
| 1853 |
} |
| 1854 |
cluster_offset &= s->cluster_offset_mask; |
| 1855 |
|
| 1856 |
ret = qcow2_pre_write_overlap_check(bs, 0, cluster_offset, out_len);
|
| 1857 |
if (ret < 0) { |
| 1858 |
goto fail;
|
| 1859 |
} |
| 1860 |
|
| 1861 |
BLKDBG_EVENT(bs->file, BLKDBG_WRITE_COMPRESSED); |
| 1862 |
ret = bdrv_pwrite(bs->file, cluster_offset, out_buf, out_len); |
| 1863 |
if (ret < 0) { |
| 1864 |
goto fail;
|
| 1865 |
} |
| 1866 |
} |
| 1867 |
|
| 1868 |
ret = 0;
|
| 1869 |
fail:
|
| 1870 |
g_free(out_buf); |
| 1871 |
return ret;
|
| 1872 |
} |
| 1873 |
|
| 1874 |
static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs) |
| 1875 |
{
|
| 1876 |
BDRVQcowState *s = bs->opaque; |
| 1877 |
int ret;
|
| 1878 |
|
| 1879 |
qemu_co_mutex_lock(&s->lock); |
| 1880 |
ret = qcow2_cache_flush(bs, s->l2_table_cache); |
| 1881 |
if (ret < 0) { |
| 1882 |
qemu_co_mutex_unlock(&s->lock); |
| 1883 |
return ret;
|
| 1884 |
} |
| 1885 |
|
| 1886 |
if (qcow2_need_accurate_refcounts(s)) {
|
| 1887 |
ret = qcow2_cache_flush(bs, s->refcount_block_cache); |
| 1888 |
if (ret < 0) { |
| 1889 |
qemu_co_mutex_unlock(&s->lock); |
| 1890 |
return ret;
|
| 1891 |
} |
| 1892 |
} |
| 1893 |
qemu_co_mutex_unlock(&s->lock); |
| 1894 |
|
| 1895 |
return 0; |
| 1896 |
} |
| 1897 |
|
| 1898 |
static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
| 1899 |
{
|
| 1900 |
BDRVQcowState *s = bs->opaque; |
| 1901 |
bdi->unallocated_blocks_are_zero = true;
|
| 1902 |
bdi->can_write_zeroes_with_unmap = (s->qcow_version >= 3);
|
| 1903 |
bdi->cluster_size = s->cluster_size; |
| 1904 |
bdi->vm_state_offset = qcow2_vm_state_offset(s); |
| 1905 |
return 0; |
| 1906 |
} |
| 1907 |
|
| 1908 |
static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs)
|
| 1909 |
{
|
| 1910 |
BDRVQcowState *s = bs->opaque; |
| 1911 |
ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
|
| 1912 |
|
| 1913 |
*spec_info = (ImageInfoSpecific){
|
| 1914 |
.kind = IMAGE_INFO_SPECIFIC_KIND_QCOW2, |
| 1915 |
{
|
| 1916 |
.qcow2 = g_new(ImageInfoSpecificQCow2, 1),
|
| 1917 |
}, |
| 1918 |
}; |
| 1919 |
if (s->qcow_version == 2) { |
| 1920 |
*spec_info->qcow2 = (ImageInfoSpecificQCow2){
|
| 1921 |
.compat = g_strdup("0.10"),
|
| 1922 |
}; |
| 1923 |
} else if (s->qcow_version == 3) { |
| 1924 |
*spec_info->qcow2 = (ImageInfoSpecificQCow2){
|
| 1925 |
.compat = g_strdup("1.1"),
|
| 1926 |
.lazy_refcounts = s->compatible_features & |
| 1927 |
QCOW2_COMPAT_LAZY_REFCOUNTS, |
| 1928 |
.has_lazy_refcounts = true,
|
| 1929 |
}; |
| 1930 |
} |
| 1931 |
|
| 1932 |
return spec_info;
|
| 1933 |
} |
| 1934 |
|
| 1935 |
#if 0
|
| 1936 |
static void dump_refcounts(BlockDriverState *bs)
|
| 1937 |
{
|
| 1938 |
BDRVQcowState *s = bs->opaque;
|
| 1939 |
int64_t nb_clusters, k, k1, size;
|
| 1940 |
int refcount;
|
| 1941 |
|
| 1942 |
size = bdrv_getlength(bs->file);
|
| 1943 |
nb_clusters = size_to_clusters(s, size);
|
| 1944 |
for(k = 0; k < nb_clusters;) {
|
| 1945 |
k1 = k;
|
| 1946 |
refcount = get_refcount(bs, k);
|
| 1947 |
k++;
|
| 1948 |
while (k < nb_clusters && get_refcount(bs, k) == refcount)
|
| 1949 |
k++;
|
| 1950 |
printf("%" PRId64 ": refcount=%d nb=%" PRId64 "\n", k, refcount,
|
| 1951 |
k - k1);
|
| 1952 |
}
|
| 1953 |
}
|
| 1954 |
#endif
|
| 1955 |
|
| 1956 |
static int qcow2_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, |
| 1957 |
int64_t pos) |
| 1958 |
{
|
| 1959 |
BDRVQcowState *s = bs->opaque; |
| 1960 |
int64_t total_sectors = bs->total_sectors; |
| 1961 |
int growable = bs->growable;
|
| 1962 |
bool zero_beyond_eof = bs->zero_beyond_eof;
|
| 1963 |
int ret;
|
| 1964 |
|
| 1965 |
BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_SAVE); |
| 1966 |
bs->growable = 1;
|
| 1967 |
bs->zero_beyond_eof = false;
|
| 1968 |
ret = bdrv_pwritev(bs, qcow2_vm_state_offset(s) + pos, qiov); |
| 1969 |
bs->growable = growable; |
| 1970 |
bs->zero_beyond_eof = zero_beyond_eof; |
| 1971 |
|
| 1972 |
/* bdrv_co_do_writev will have increased the total_sectors value to include
|
| 1973 |
* the VM state - the VM state is however not an actual part of the block
|
| 1974 |
* device, therefore, we need to restore the old value. */
|
| 1975 |
bs->total_sectors = total_sectors; |
| 1976 |
|
| 1977 |
return ret;
|
| 1978 |
} |
| 1979 |
|
| 1980 |
static int qcow2_load_vmstate(BlockDriverState *bs, uint8_t *buf, |
| 1981 |
int64_t pos, int size)
|
| 1982 |
{
|
| 1983 |
BDRVQcowState *s = bs->opaque; |
| 1984 |
int growable = bs->growable;
|
| 1985 |
bool zero_beyond_eof = bs->zero_beyond_eof;
|
| 1986 |
int ret;
|
| 1987 |
|
| 1988 |
BLKDBG_EVENT(bs->file, BLKDBG_VMSTATE_LOAD); |
| 1989 |
bs->growable = 1;
|
| 1990 |
bs->zero_beyond_eof = false;
|
| 1991 |
ret = bdrv_pread(bs, qcow2_vm_state_offset(s) + pos, buf, size); |
| 1992 |
bs->growable = growable; |
| 1993 |
bs->zero_beyond_eof = zero_beyond_eof; |
| 1994 |
|
| 1995 |
return ret;
|
| 1996 |
} |
| 1997 |
|
| 1998 |
/*
|
| 1999 |
* Downgrades an image's version. To achieve this, any incompatible features
|
| 2000 |
* have to be removed.
|
| 2001 |
*/
|
| 2002 |
static int qcow2_downgrade(BlockDriverState *bs, int target_version) |
| 2003 |
{
|
| 2004 |
BDRVQcowState *s = bs->opaque; |
| 2005 |
int current_version = s->qcow_version;
|
| 2006 |
int ret;
|
| 2007 |
|
| 2008 |
if (target_version == current_version) {
|
| 2009 |
return 0; |
| 2010 |
} else if (target_version > current_version) { |
| 2011 |
return -EINVAL;
|
| 2012 |
} else if (target_version != 2) { |
| 2013 |
return -EINVAL;
|
| 2014 |
} |
| 2015 |
|
| 2016 |
if (s->refcount_order != 4) { |
| 2017 |
/* we would have to convert the image to a refcount_order == 4 image
|
| 2018 |
* here; however, since qemu (at the time of writing this) does not
|
| 2019 |
* support anything different than 4 anyway, there is no point in doing
|
| 2020 |
* so right now; however, we should error out (if qemu supports this in
|
| 2021 |
* the future and this code has not been adapted) */
|
| 2022 |
error_report("qcow2_downgrade: Image refcount orders other than 4 are "
|
| 2023 |
"currently not supported.");
|
| 2024 |
return -ENOTSUP;
|
| 2025 |
} |
| 2026 |
|
| 2027 |
/* clear incompatible features */
|
| 2028 |
if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
|
| 2029 |
ret = qcow2_mark_clean(bs); |
| 2030 |
if (ret < 0) { |
| 2031 |
return ret;
|
| 2032 |
} |
| 2033 |
} |
| 2034 |
|
| 2035 |
/* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
|
| 2036 |
* the first place; if that happens nonetheless, returning -ENOTSUP is the
|
| 2037 |
* best thing to do anyway */
|
| 2038 |
|
| 2039 |
if (s->incompatible_features) {
|
| 2040 |
return -ENOTSUP;
|
| 2041 |
} |
| 2042 |
|
| 2043 |
/* since we can ignore compatible features, we can set them to 0 as well */
|
| 2044 |
s->compatible_features = 0;
|
| 2045 |
/* if lazy refcounts have been used, they have already been fixed through
|
| 2046 |
* clearing the dirty flag */
|
| 2047 |
|
| 2048 |
/* clearing autoclear features is trivial */
|
| 2049 |
s->autoclear_features = 0;
|
| 2050 |
|
| 2051 |
ret = qcow2_expand_zero_clusters(bs); |
| 2052 |
if (ret < 0) { |
| 2053 |
return ret;
|
| 2054 |
} |
| 2055 |
|
| 2056 |
s->qcow_version = target_version; |
| 2057 |
ret = qcow2_update_header(bs); |
| 2058 |
if (ret < 0) { |
| 2059 |
s->qcow_version = current_version; |
| 2060 |
return ret;
|
| 2061 |
} |
| 2062 |
return 0; |
| 2063 |
} |
| 2064 |
|
| 2065 |
static int qcow2_amend_options(BlockDriverState *bs, |
| 2066 |
QEMUOptionParameter *options) |
| 2067 |
{
|
| 2068 |
BDRVQcowState *s = bs->opaque; |
| 2069 |
int old_version = s->qcow_version, new_version = old_version;
|
| 2070 |
uint64_t new_size = 0;
|
| 2071 |
const char *backing_file = NULL, *backing_format = NULL; |
| 2072 |
bool lazy_refcounts = s->use_lazy_refcounts;
|
| 2073 |
int ret;
|
| 2074 |
int i;
|
| 2075 |
|
| 2076 |
for (i = 0; options[i].name; i++) |
| 2077 |
{
|
| 2078 |
if (!options[i].assigned) {
|
| 2079 |
/* only change explicitly defined options */
|
| 2080 |
continue;
|
| 2081 |
} |
| 2082 |
|
| 2083 |
if (!strcmp(options[i].name, "compat")) { |
| 2084 |
if (!options[i].value.s) {
|
| 2085 |
/* preserve default */
|
| 2086 |
} else if (!strcmp(options[i].value.s, "0.10")) { |
| 2087 |
new_version = 2;
|
| 2088 |
} else if (!strcmp(options[i].value.s, "1.1")) { |
| 2089 |
new_version = 3;
|
| 2090 |
} else {
|
| 2091 |
fprintf(stderr, "Unknown compatibility level %s.\n",
|
| 2092 |
options[i].value.s); |
| 2093 |
return -EINVAL;
|
| 2094 |
} |
| 2095 |
} else if (!strcmp(options[i].name, "preallocation")) { |
| 2096 |
fprintf(stderr, "Cannot change preallocation mode.\n");
|
| 2097 |
return -ENOTSUP;
|
| 2098 |
} else if (!strcmp(options[i].name, "size")) { |
| 2099 |
new_size = options[i].value.n; |
| 2100 |
} else if (!strcmp(options[i].name, "backing_file")) { |
| 2101 |
backing_file = options[i].value.s; |
| 2102 |
} else if (!strcmp(options[i].name, "backing_fmt")) { |
| 2103 |
backing_format = options[i].value.s; |
| 2104 |
} else if (!strcmp(options[i].name, "encryption")) { |
| 2105 |
if ((options[i].value.n != !!s->crypt_method)) {
|
| 2106 |
fprintf(stderr, "Changing the encryption flag is not "
|
| 2107 |
"supported.\n");
|
| 2108 |
return -ENOTSUP;
|
| 2109 |
} |
| 2110 |
} else if (!strcmp(options[i].name, "cluster_size")) { |
| 2111 |
if (options[i].value.n != s->cluster_size) {
|
| 2112 |
fprintf(stderr, "Changing the cluster size is not "
|
| 2113 |
"supported.\n");
|
| 2114 |
return -ENOTSUP;
|
| 2115 |
} |
| 2116 |
} else if (!strcmp(options[i].name, "lazy_refcounts")) { |
| 2117 |
lazy_refcounts = options[i].value.n; |
| 2118 |
} else {
|
| 2119 |
/* if this assertion fails, this probably means a new option was
|
| 2120 |
* added without having it covered here */
|
| 2121 |
assert(false);
|
| 2122 |
} |
| 2123 |
} |
| 2124 |
|
| 2125 |
if (new_version != old_version) {
|
| 2126 |
if (new_version > old_version) {
|
| 2127 |
/* Upgrade */
|
| 2128 |
s->qcow_version = new_version; |
| 2129 |
ret = qcow2_update_header(bs); |
| 2130 |
if (ret < 0) { |
| 2131 |
s->qcow_version = old_version; |
| 2132 |
return ret;
|
| 2133 |
} |
| 2134 |
} else {
|
| 2135 |
ret = qcow2_downgrade(bs, new_version); |
| 2136 |
if (ret < 0) { |
| 2137 |
return ret;
|
| 2138 |
} |
| 2139 |
} |
| 2140 |
} |
| 2141 |
|
| 2142 |
if (backing_file || backing_format) {
|
| 2143 |
ret = qcow2_change_backing_file(bs, backing_file ?: bs->backing_file, |
| 2144 |
backing_format ?: bs->backing_format); |
| 2145 |
if (ret < 0) { |
| 2146 |
return ret;
|
| 2147 |
} |
| 2148 |
} |
| 2149 |
|
| 2150 |
if (s->use_lazy_refcounts != lazy_refcounts) {
|
| 2151 |
if (lazy_refcounts) {
|
| 2152 |
if (s->qcow_version < 3) { |
| 2153 |
fprintf(stderr, "Lazy refcounts only supported with compatibility "
|
| 2154 |
"level 1.1 and above (use compat=1.1 or greater)\n");
|
| 2155 |
return -EINVAL;
|
| 2156 |
} |
| 2157 |
s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; |
| 2158 |
ret = qcow2_update_header(bs); |
| 2159 |
if (ret < 0) { |
| 2160 |
s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; |
| 2161 |
return ret;
|
| 2162 |
} |
| 2163 |
s->use_lazy_refcounts = true;
|
| 2164 |
} else {
|
| 2165 |
/* make image clean first */
|
| 2166 |
ret = qcow2_mark_clean(bs); |
| 2167 |
if (ret < 0) { |
| 2168 |
return ret;
|
| 2169 |
} |
| 2170 |
/* now disallow lazy refcounts */
|
| 2171 |
s->compatible_features &= ~QCOW2_COMPAT_LAZY_REFCOUNTS; |
| 2172 |
ret = qcow2_update_header(bs); |
| 2173 |
if (ret < 0) { |
| 2174 |
s->compatible_features |= QCOW2_COMPAT_LAZY_REFCOUNTS; |
| 2175 |
return ret;
|
| 2176 |
} |
| 2177 |
s->use_lazy_refcounts = false;
|
| 2178 |
} |
| 2179 |
} |
| 2180 |
|
| 2181 |
if (new_size) {
|
| 2182 |
ret = bdrv_truncate(bs, new_size); |
| 2183 |
if (ret < 0) { |
| 2184 |
return ret;
|
| 2185 |
} |
| 2186 |
} |
| 2187 |
|
| 2188 |
return 0; |
| 2189 |
} |
| 2190 |
|
| 2191 |
static QEMUOptionParameter qcow2_create_options[] = {
|
| 2192 |
{
|
| 2193 |
.name = BLOCK_OPT_SIZE, |
| 2194 |
.type = OPT_SIZE, |
| 2195 |
.help = "Virtual disk size"
|
| 2196 |
}, |
| 2197 |
{
|
| 2198 |
.name = BLOCK_OPT_COMPAT_LEVEL, |
| 2199 |
.type = OPT_STRING, |
| 2200 |
.help = "Compatibility level (0.10 or 1.1)"
|
| 2201 |
}, |
| 2202 |
{
|
| 2203 |
.name = BLOCK_OPT_BACKING_FILE, |
| 2204 |
.type = OPT_STRING, |
| 2205 |
.help = "File name of a base image"
|
| 2206 |
}, |
| 2207 |
{
|
| 2208 |
.name = BLOCK_OPT_BACKING_FMT, |
| 2209 |
.type = OPT_STRING, |
| 2210 |
.help = "Image format of the base image"
|
| 2211 |
}, |
| 2212 |
{
|
| 2213 |
.name = BLOCK_OPT_ENCRYPT, |
| 2214 |
.type = OPT_FLAG, |
| 2215 |
.help = "Encrypt the image"
|
| 2216 |
}, |
| 2217 |
{
|
| 2218 |
.name = BLOCK_OPT_CLUSTER_SIZE, |
| 2219 |
.type = OPT_SIZE, |
| 2220 |
.help = "qcow2 cluster size",
|
| 2221 |
.value = { .n = DEFAULT_CLUSTER_SIZE },
|
| 2222 |
}, |
| 2223 |
{
|
| 2224 |
.name = BLOCK_OPT_PREALLOC, |
| 2225 |
.type = OPT_STRING, |
| 2226 |
.help = "Preallocation mode (allowed values: off, metadata)"
|
| 2227 |
}, |
| 2228 |
{
|
| 2229 |
.name = BLOCK_OPT_LAZY_REFCOUNTS, |
| 2230 |
.type = OPT_FLAG, |
| 2231 |
.help = "Postpone refcount updates",
|
| 2232 |
}, |
| 2233 |
{ NULL }
|
| 2234 |
}; |
| 2235 |
|
| 2236 |
static BlockDriver bdrv_qcow2 = {
|
| 2237 |
.format_name = "qcow2",
|
| 2238 |
.instance_size = sizeof(BDRVQcowState),
|
| 2239 |
.bdrv_probe = qcow2_probe, |
| 2240 |
.bdrv_open = qcow2_open, |
| 2241 |
.bdrv_close = qcow2_close, |
| 2242 |
.bdrv_reopen_prepare = qcow2_reopen_prepare, |
| 2243 |
.bdrv_create = qcow2_create, |
| 2244 |
.bdrv_has_zero_init = bdrv_has_zero_init_1, |
| 2245 |
.bdrv_co_get_block_status = qcow2_co_get_block_status, |
| 2246 |
.bdrv_set_key = qcow2_set_key, |
| 2247 |
.bdrv_make_empty = qcow2_make_empty, |
| 2248 |
|
| 2249 |
.bdrv_co_readv = qcow2_co_readv, |
| 2250 |
.bdrv_co_writev = qcow2_co_writev, |
| 2251 |
.bdrv_co_flush_to_os = qcow2_co_flush_to_os, |
| 2252 |
|
| 2253 |
.bdrv_co_write_zeroes = qcow2_co_write_zeroes, |
| 2254 |
.bdrv_co_discard = qcow2_co_discard, |
| 2255 |
.bdrv_truncate = qcow2_truncate, |
| 2256 |
.bdrv_write_compressed = qcow2_write_compressed, |
| 2257 |
|
| 2258 |
.bdrv_snapshot_create = qcow2_snapshot_create, |
| 2259 |
.bdrv_snapshot_goto = qcow2_snapshot_goto, |
| 2260 |
.bdrv_snapshot_delete = qcow2_snapshot_delete, |
| 2261 |
.bdrv_snapshot_list = qcow2_snapshot_list, |
| 2262 |
.bdrv_snapshot_load_tmp = qcow2_snapshot_load_tmp, |
| 2263 |
.bdrv_get_info = qcow2_get_info, |
| 2264 |
.bdrv_get_specific_info = qcow2_get_specific_info, |
| 2265 |
|
| 2266 |
.bdrv_save_vmstate = qcow2_save_vmstate, |
| 2267 |
.bdrv_load_vmstate = qcow2_load_vmstate, |
| 2268 |
|
| 2269 |
.bdrv_change_backing_file = qcow2_change_backing_file, |
| 2270 |
|
| 2271 |
.bdrv_invalidate_cache = qcow2_invalidate_cache, |
| 2272 |
|
| 2273 |
.create_options = qcow2_create_options, |
| 2274 |
.bdrv_check = qcow2_check, |
| 2275 |
.bdrv_amend_options = qcow2_amend_options, |
| 2276 |
}; |
| 2277 |
|
| 2278 |
static void bdrv_qcow2_init(void) |
| 2279 |
{
|
| 2280 |
bdrv_register(&bdrv_qcow2); |
| 2281 |
} |
| 2282 |
|
| 2283 |
block_init(bdrv_qcow2_init); |