root / block / qed.c @ 34b5d2c6
History | View | Annotate | Download (46.7 kB)
| 1 |
/*
|
|---|---|
| 2 |
* QEMU Enhanced Disk Format
|
| 3 |
*
|
| 4 |
* Copyright IBM, Corp. 2010
|
| 5 |
*
|
| 6 |
* Authors:
|
| 7 |
* Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
|
| 8 |
* Anthony Liguori <aliguori@us.ibm.com>
|
| 9 |
*
|
| 10 |
* This work is licensed under the terms of the GNU LGPL, version 2 or later.
|
| 11 |
* See the COPYING.LIB file in the top-level directory.
|
| 12 |
*
|
| 13 |
*/
|
| 14 |
|
| 15 |
#include "qemu/timer.h" |
| 16 |
#include "trace.h" |
| 17 |
#include "qed.h" |
| 18 |
#include "qapi/qmp/qerror.h" |
| 19 |
#include "migration/migration.h" |
| 20 |
|
| 21 |
static void qed_aio_cancel(BlockDriverAIOCB *blockacb) |
| 22 |
{
|
| 23 |
QEDAIOCB *acb = (QEDAIOCB *)blockacb; |
| 24 |
bool finished = false; |
| 25 |
|
| 26 |
/* Wait for the request to finish */
|
| 27 |
acb->finished = &finished; |
| 28 |
while (!finished) {
|
| 29 |
qemu_aio_wait(); |
| 30 |
} |
| 31 |
} |
| 32 |
|
| 33 |
static const AIOCBInfo qed_aiocb_info = { |
| 34 |
.aiocb_size = sizeof(QEDAIOCB),
|
| 35 |
.cancel = qed_aio_cancel, |
| 36 |
}; |
| 37 |
|
| 38 |
static int bdrv_qed_probe(const uint8_t *buf, int buf_size, |
| 39 |
const char *filename) |
| 40 |
{
|
| 41 |
const QEDHeader *header = (const QEDHeader *)buf; |
| 42 |
|
| 43 |
if (buf_size < sizeof(*header)) { |
| 44 |
return 0; |
| 45 |
} |
| 46 |
if (le32_to_cpu(header->magic) != QED_MAGIC) {
|
| 47 |
return 0; |
| 48 |
} |
| 49 |
return 100; |
| 50 |
} |
| 51 |
|
| 52 |
/**
|
| 53 |
* Check whether an image format is raw
|
| 54 |
*
|
| 55 |
* @fmt: Backing file format, may be NULL
|
| 56 |
*/
|
| 57 |
static bool qed_fmt_is_raw(const char *fmt) |
| 58 |
{
|
| 59 |
return fmt && strcmp(fmt, "raw") == 0; |
| 60 |
} |
| 61 |
|
| 62 |
static void qed_header_le_to_cpu(const QEDHeader *le, QEDHeader *cpu) |
| 63 |
{
|
| 64 |
cpu->magic = le32_to_cpu(le->magic); |
| 65 |
cpu->cluster_size = le32_to_cpu(le->cluster_size); |
| 66 |
cpu->table_size = le32_to_cpu(le->table_size); |
| 67 |
cpu->header_size = le32_to_cpu(le->header_size); |
| 68 |
cpu->features = le64_to_cpu(le->features); |
| 69 |
cpu->compat_features = le64_to_cpu(le->compat_features); |
| 70 |
cpu->autoclear_features = le64_to_cpu(le->autoclear_features); |
| 71 |
cpu->l1_table_offset = le64_to_cpu(le->l1_table_offset); |
| 72 |
cpu->image_size = le64_to_cpu(le->image_size); |
| 73 |
cpu->backing_filename_offset = le32_to_cpu(le->backing_filename_offset); |
| 74 |
cpu->backing_filename_size = le32_to_cpu(le->backing_filename_size); |
| 75 |
} |
| 76 |
|
| 77 |
static void qed_header_cpu_to_le(const QEDHeader *cpu, QEDHeader *le) |
| 78 |
{
|
| 79 |
le->magic = cpu_to_le32(cpu->magic); |
| 80 |
le->cluster_size = cpu_to_le32(cpu->cluster_size); |
| 81 |
le->table_size = cpu_to_le32(cpu->table_size); |
| 82 |
le->header_size = cpu_to_le32(cpu->header_size); |
| 83 |
le->features = cpu_to_le64(cpu->features); |
| 84 |
le->compat_features = cpu_to_le64(cpu->compat_features); |
| 85 |
le->autoclear_features = cpu_to_le64(cpu->autoclear_features); |
| 86 |
le->l1_table_offset = cpu_to_le64(cpu->l1_table_offset); |
| 87 |
le->image_size = cpu_to_le64(cpu->image_size); |
| 88 |
le->backing_filename_offset = cpu_to_le32(cpu->backing_filename_offset); |
| 89 |
le->backing_filename_size = cpu_to_le32(cpu->backing_filename_size); |
| 90 |
} |
| 91 |
|
| 92 |
int qed_write_header_sync(BDRVQEDState *s)
|
| 93 |
{
|
| 94 |
QEDHeader le; |
| 95 |
int ret;
|
| 96 |
|
| 97 |
qed_header_cpu_to_le(&s->header, &le); |
| 98 |
ret = bdrv_pwrite(s->bs->file, 0, &le, sizeof(le)); |
| 99 |
if (ret != sizeof(le)) { |
| 100 |
return ret;
|
| 101 |
} |
| 102 |
return 0; |
| 103 |
} |
| 104 |
|
| 105 |
typedef struct { |
| 106 |
GenericCB gencb; |
| 107 |
BDRVQEDState *s; |
| 108 |
struct iovec iov;
|
| 109 |
QEMUIOVector qiov; |
| 110 |
int nsectors;
|
| 111 |
uint8_t *buf; |
| 112 |
} QEDWriteHeaderCB; |
| 113 |
|
| 114 |
static void qed_write_header_cb(void *opaque, int ret) |
| 115 |
{
|
| 116 |
QEDWriteHeaderCB *write_header_cb = opaque; |
| 117 |
|
| 118 |
qemu_vfree(write_header_cb->buf); |
| 119 |
gencb_complete(write_header_cb, ret); |
| 120 |
} |
| 121 |
|
| 122 |
static void qed_write_header_read_cb(void *opaque, int ret) |
| 123 |
{
|
| 124 |
QEDWriteHeaderCB *write_header_cb = opaque; |
| 125 |
BDRVQEDState *s = write_header_cb->s; |
| 126 |
|
| 127 |
if (ret) {
|
| 128 |
qed_write_header_cb(write_header_cb, ret); |
| 129 |
return;
|
| 130 |
} |
| 131 |
|
| 132 |
/* Update header */
|
| 133 |
qed_header_cpu_to_le(&s->header, (QEDHeader *)write_header_cb->buf); |
| 134 |
|
| 135 |
bdrv_aio_writev(s->bs->file, 0, &write_header_cb->qiov,
|
| 136 |
write_header_cb->nsectors, qed_write_header_cb, |
| 137 |
write_header_cb); |
| 138 |
} |
| 139 |
|
| 140 |
/**
|
| 141 |
* Update header in-place (does not rewrite backing filename or other strings)
|
| 142 |
*
|
| 143 |
* This function only updates known header fields in-place and does not affect
|
| 144 |
* extra data after the QED header.
|
| 145 |
*/
|
| 146 |
static void qed_write_header(BDRVQEDState *s, BlockDriverCompletionFunc cb, |
| 147 |
void *opaque)
|
| 148 |
{
|
| 149 |
/* We must write full sectors for O_DIRECT but cannot necessarily generate
|
| 150 |
* the data following the header if an unrecognized compat feature is
|
| 151 |
* active. Therefore, first read the sectors containing the header, update
|
| 152 |
* them, and write back.
|
| 153 |
*/
|
| 154 |
|
| 155 |
int nsectors = (sizeof(QEDHeader) + BDRV_SECTOR_SIZE - 1) / |
| 156 |
BDRV_SECTOR_SIZE; |
| 157 |
size_t len = nsectors * BDRV_SECTOR_SIZE; |
| 158 |
QEDWriteHeaderCB *write_header_cb = gencb_alloc(sizeof(*write_header_cb),
|
| 159 |
cb, opaque); |
| 160 |
|
| 161 |
write_header_cb->s = s; |
| 162 |
write_header_cb->nsectors = nsectors; |
| 163 |
write_header_cb->buf = qemu_blockalign(s->bs, len); |
| 164 |
write_header_cb->iov.iov_base = write_header_cb->buf; |
| 165 |
write_header_cb->iov.iov_len = len; |
| 166 |
qemu_iovec_init_external(&write_header_cb->qiov, &write_header_cb->iov, 1);
|
| 167 |
|
| 168 |
bdrv_aio_readv(s->bs->file, 0, &write_header_cb->qiov, nsectors,
|
| 169 |
qed_write_header_read_cb, write_header_cb); |
| 170 |
} |
| 171 |
|
| 172 |
static uint64_t qed_max_image_size(uint32_t cluster_size, uint32_t table_size)
|
| 173 |
{
|
| 174 |
uint64_t table_entries; |
| 175 |
uint64_t l2_size; |
| 176 |
|
| 177 |
table_entries = (table_size * cluster_size) / sizeof(uint64_t);
|
| 178 |
l2_size = table_entries * cluster_size; |
| 179 |
|
| 180 |
return l2_size * table_entries;
|
| 181 |
} |
| 182 |
|
| 183 |
static bool qed_is_cluster_size_valid(uint32_t cluster_size) |
| 184 |
{
|
| 185 |
if (cluster_size < QED_MIN_CLUSTER_SIZE ||
|
| 186 |
cluster_size > QED_MAX_CLUSTER_SIZE) {
|
| 187 |
return false; |
| 188 |
} |
| 189 |
if (cluster_size & (cluster_size - 1)) { |
| 190 |
return false; /* not power of 2 */ |
| 191 |
} |
| 192 |
return true; |
| 193 |
} |
| 194 |
|
| 195 |
static bool qed_is_table_size_valid(uint32_t table_size) |
| 196 |
{
|
| 197 |
if (table_size < QED_MIN_TABLE_SIZE ||
|
| 198 |
table_size > QED_MAX_TABLE_SIZE) {
|
| 199 |
return false; |
| 200 |
} |
| 201 |
if (table_size & (table_size - 1)) { |
| 202 |
return false; /* not power of 2 */ |
| 203 |
} |
| 204 |
return true; |
| 205 |
} |
| 206 |
|
| 207 |
static bool qed_is_image_size_valid(uint64_t image_size, uint32_t cluster_size, |
| 208 |
uint32_t table_size) |
| 209 |
{
|
| 210 |
if (image_size % BDRV_SECTOR_SIZE != 0) { |
| 211 |
return false; /* not multiple of sector size */ |
| 212 |
} |
| 213 |
if (image_size > qed_max_image_size(cluster_size, table_size)) {
|
| 214 |
return false; /* image is too large */ |
| 215 |
} |
| 216 |
return true; |
| 217 |
} |
| 218 |
|
| 219 |
/**
|
| 220 |
* Read a string of known length from the image file
|
| 221 |
*
|
| 222 |
* @file: Image file
|
| 223 |
* @offset: File offset to start of string, in bytes
|
| 224 |
* @n: String length in bytes
|
| 225 |
* @buf: Destination buffer
|
| 226 |
* @buflen: Destination buffer length in bytes
|
| 227 |
* @ret: 0 on success, -errno on failure
|
| 228 |
*
|
| 229 |
* The string is NUL-terminated.
|
| 230 |
*/
|
| 231 |
static int qed_read_string(BlockDriverState *file, uint64_t offset, size_t n, |
| 232 |
char *buf, size_t buflen)
|
| 233 |
{
|
| 234 |
int ret;
|
| 235 |
if (n >= buflen) {
|
| 236 |
return -EINVAL;
|
| 237 |
} |
| 238 |
ret = bdrv_pread(file, offset, buf, n); |
| 239 |
if (ret < 0) { |
| 240 |
return ret;
|
| 241 |
} |
| 242 |
buf[n] = '\0';
|
| 243 |
return 0; |
| 244 |
} |
| 245 |
|
| 246 |
/**
|
| 247 |
* Allocate new clusters
|
| 248 |
*
|
| 249 |
* @s: QED state
|
| 250 |
* @n: Number of contiguous clusters to allocate
|
| 251 |
* @ret: Offset of first allocated cluster
|
| 252 |
*
|
| 253 |
* This function only produces the offset where the new clusters should be
|
| 254 |
* written. It updates BDRVQEDState but does not make any changes to the image
|
| 255 |
* file.
|
| 256 |
*/
|
| 257 |
static uint64_t qed_alloc_clusters(BDRVQEDState *s, unsigned int n) |
| 258 |
{
|
| 259 |
uint64_t offset = s->file_size; |
| 260 |
s->file_size += n * s->header.cluster_size; |
| 261 |
return offset;
|
| 262 |
} |
| 263 |
|
| 264 |
QEDTable *qed_alloc_table(BDRVQEDState *s) |
| 265 |
{
|
| 266 |
/* Honor O_DIRECT memory alignment requirements */
|
| 267 |
return qemu_blockalign(s->bs,
|
| 268 |
s->header.cluster_size * s->header.table_size); |
| 269 |
} |
| 270 |
|
| 271 |
/**
|
| 272 |
* Allocate a new zeroed L2 table
|
| 273 |
*/
|
| 274 |
static CachedL2Table *qed_new_l2_table(BDRVQEDState *s)
|
| 275 |
{
|
| 276 |
CachedL2Table *l2_table = qed_alloc_l2_cache_entry(&s->l2_cache); |
| 277 |
|
| 278 |
l2_table->table = qed_alloc_table(s); |
| 279 |
l2_table->offset = qed_alloc_clusters(s, s->header.table_size); |
| 280 |
|
| 281 |
memset(l2_table->table->offsets, 0,
|
| 282 |
s->header.cluster_size * s->header.table_size); |
| 283 |
return l2_table;
|
| 284 |
} |
| 285 |
|
| 286 |
static void qed_aio_next_io(void *opaque, int ret); |
| 287 |
|
| 288 |
static void qed_plug_allocating_write_reqs(BDRVQEDState *s) |
| 289 |
{
|
| 290 |
assert(!s->allocating_write_reqs_plugged); |
| 291 |
|
| 292 |
s->allocating_write_reqs_plugged = true;
|
| 293 |
} |
| 294 |
|
| 295 |
static void qed_unplug_allocating_write_reqs(BDRVQEDState *s) |
| 296 |
{
|
| 297 |
QEDAIOCB *acb; |
| 298 |
|
| 299 |
assert(s->allocating_write_reqs_plugged); |
| 300 |
|
| 301 |
s->allocating_write_reqs_plugged = false;
|
| 302 |
|
| 303 |
acb = QSIMPLEQ_FIRST(&s->allocating_write_reqs); |
| 304 |
if (acb) {
|
| 305 |
qed_aio_next_io(acb, 0);
|
| 306 |
} |
| 307 |
} |
| 308 |
|
| 309 |
static void qed_finish_clear_need_check(void *opaque, int ret) |
| 310 |
{
|
| 311 |
/* Do nothing */
|
| 312 |
} |
| 313 |
|
| 314 |
static void qed_flush_after_clear_need_check(void *opaque, int ret) |
| 315 |
{
|
| 316 |
BDRVQEDState *s = opaque; |
| 317 |
|
| 318 |
bdrv_aio_flush(s->bs, qed_finish_clear_need_check, s); |
| 319 |
|
| 320 |
/* No need to wait until flush completes */
|
| 321 |
qed_unplug_allocating_write_reqs(s); |
| 322 |
} |
| 323 |
|
| 324 |
static void qed_clear_need_check(void *opaque, int ret) |
| 325 |
{
|
| 326 |
BDRVQEDState *s = opaque; |
| 327 |
|
| 328 |
if (ret) {
|
| 329 |
qed_unplug_allocating_write_reqs(s); |
| 330 |
return;
|
| 331 |
} |
| 332 |
|
| 333 |
s->header.features &= ~QED_F_NEED_CHECK; |
| 334 |
qed_write_header(s, qed_flush_after_clear_need_check, s); |
| 335 |
} |
| 336 |
|
| 337 |
static void qed_need_check_timer_cb(void *opaque) |
| 338 |
{
|
| 339 |
BDRVQEDState *s = opaque; |
| 340 |
|
| 341 |
/* The timer should only fire when allocating writes have drained */
|
| 342 |
assert(!QSIMPLEQ_FIRST(&s->allocating_write_reqs)); |
| 343 |
|
| 344 |
trace_qed_need_check_timer_cb(s); |
| 345 |
|
| 346 |
qed_plug_allocating_write_reqs(s); |
| 347 |
|
| 348 |
/* Ensure writes are on disk before clearing flag */
|
| 349 |
bdrv_aio_flush(s->bs, qed_clear_need_check, s); |
| 350 |
} |
| 351 |
|
| 352 |
static void qed_start_need_check_timer(BDRVQEDState *s) |
| 353 |
{
|
| 354 |
trace_qed_start_need_check_timer(s); |
| 355 |
|
| 356 |
/* Use QEMU_CLOCK_VIRTUAL so we don't alter the image file while suspended for
|
| 357 |
* migration.
|
| 358 |
*/
|
| 359 |
timer_mod(s->need_check_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + |
| 360 |
get_ticks_per_sec() * QED_NEED_CHECK_TIMEOUT); |
| 361 |
} |
| 362 |
|
| 363 |
/* It's okay to call this multiple times or when no timer is started */
|
| 364 |
static void qed_cancel_need_check_timer(BDRVQEDState *s) |
| 365 |
{
|
| 366 |
trace_qed_cancel_need_check_timer(s); |
| 367 |
timer_del(s->need_check_timer); |
| 368 |
} |
| 369 |
|
| 370 |
static void bdrv_qed_rebind(BlockDriverState *bs) |
| 371 |
{
|
| 372 |
BDRVQEDState *s = bs->opaque; |
| 373 |
s->bs = bs; |
| 374 |
} |
| 375 |
|
| 376 |
static int bdrv_qed_open(BlockDriverState *bs, QDict *options, int flags, |
| 377 |
Error **errp) |
| 378 |
{
|
| 379 |
BDRVQEDState *s = bs->opaque; |
| 380 |
QEDHeader le_header; |
| 381 |
int64_t file_size; |
| 382 |
int ret;
|
| 383 |
|
| 384 |
s->bs = bs; |
| 385 |
QSIMPLEQ_INIT(&s->allocating_write_reqs); |
| 386 |
|
| 387 |
ret = bdrv_pread(bs->file, 0, &le_header, sizeof(le_header)); |
| 388 |
if (ret < 0) { |
| 389 |
return ret;
|
| 390 |
} |
| 391 |
qed_header_le_to_cpu(&le_header, &s->header); |
| 392 |
|
| 393 |
if (s->header.magic != QED_MAGIC) {
|
| 394 |
return -EMEDIUMTYPE;
|
| 395 |
} |
| 396 |
if (s->header.features & ~QED_FEATURE_MASK) {
|
| 397 |
/* image uses unsupported feature bits */
|
| 398 |
char buf[64]; |
| 399 |
snprintf(buf, sizeof(buf), "%" PRIx64, |
| 400 |
s->header.features & ~QED_FEATURE_MASK); |
| 401 |
qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, |
| 402 |
bs->device_name, "QED", buf);
|
| 403 |
return -ENOTSUP;
|
| 404 |
} |
| 405 |
if (!qed_is_cluster_size_valid(s->header.cluster_size)) {
|
| 406 |
return -EINVAL;
|
| 407 |
} |
| 408 |
|
| 409 |
/* Round down file size to the last cluster */
|
| 410 |
file_size = bdrv_getlength(bs->file); |
| 411 |
if (file_size < 0) { |
| 412 |
return file_size;
|
| 413 |
} |
| 414 |
s->file_size = qed_start_of_cluster(s, file_size); |
| 415 |
|
| 416 |
if (!qed_is_table_size_valid(s->header.table_size)) {
|
| 417 |
return -EINVAL;
|
| 418 |
} |
| 419 |
if (!qed_is_image_size_valid(s->header.image_size,
|
| 420 |
s->header.cluster_size, |
| 421 |
s->header.table_size)) {
|
| 422 |
return -EINVAL;
|
| 423 |
} |
| 424 |
if (!qed_check_table_offset(s, s->header.l1_table_offset)) {
|
| 425 |
return -EINVAL;
|
| 426 |
} |
| 427 |
|
| 428 |
s->table_nelems = (s->header.cluster_size * s->header.table_size) / |
| 429 |
sizeof(uint64_t);
|
| 430 |
s->l2_shift = ffs(s->header.cluster_size) - 1;
|
| 431 |
s->l2_mask = s->table_nelems - 1;
|
| 432 |
s->l1_shift = s->l2_shift + ffs(s->table_nelems) - 1;
|
| 433 |
|
| 434 |
if ((s->header.features & QED_F_BACKING_FILE)) {
|
| 435 |
if ((uint64_t)s->header.backing_filename_offset +
|
| 436 |
s->header.backing_filename_size > |
| 437 |
s->header.cluster_size * s->header.header_size) {
|
| 438 |
return -EINVAL;
|
| 439 |
} |
| 440 |
|
| 441 |
ret = qed_read_string(bs->file, s->header.backing_filename_offset, |
| 442 |
s->header.backing_filename_size, bs->backing_file, |
| 443 |
sizeof(bs->backing_file));
|
| 444 |
if (ret < 0) { |
| 445 |
return ret;
|
| 446 |
} |
| 447 |
|
| 448 |
if (s->header.features & QED_F_BACKING_FORMAT_NO_PROBE) {
|
| 449 |
pstrcpy(bs->backing_format, sizeof(bs->backing_format), "raw"); |
| 450 |
} |
| 451 |
} |
| 452 |
|
| 453 |
/* Reset unknown autoclear feature bits. This is a backwards
|
| 454 |
* compatibility mechanism that allows images to be opened by older
|
| 455 |
* programs, which "knock out" unknown feature bits. When an image is
|
| 456 |
* opened by a newer program again it can detect that the autoclear
|
| 457 |
* feature is no longer valid.
|
| 458 |
*/
|
| 459 |
if ((s->header.autoclear_features & ~QED_AUTOCLEAR_FEATURE_MASK) != 0 && |
| 460 |
!bdrv_is_read_only(bs->file) && !(flags & BDRV_O_INCOMING)) {
|
| 461 |
s->header.autoclear_features &= QED_AUTOCLEAR_FEATURE_MASK; |
| 462 |
|
| 463 |
ret = qed_write_header_sync(s); |
| 464 |
if (ret) {
|
| 465 |
return ret;
|
| 466 |
} |
| 467 |
|
| 468 |
/* From here on only known autoclear feature bits are valid */
|
| 469 |
bdrv_flush(bs->file); |
| 470 |
} |
| 471 |
|
| 472 |
s->l1_table = qed_alloc_table(s); |
| 473 |
qed_init_l2_cache(&s->l2_cache); |
| 474 |
|
| 475 |
ret = qed_read_l1_table_sync(s); |
| 476 |
if (ret) {
|
| 477 |
goto out;
|
| 478 |
} |
| 479 |
|
| 480 |
/* If image was not closed cleanly, check consistency */
|
| 481 |
if (!(flags & BDRV_O_CHECK) && (s->header.features & QED_F_NEED_CHECK)) {
|
| 482 |
/* Read-only images cannot be fixed. There is no risk of corruption
|
| 483 |
* since write operations are not possible. Therefore, allow
|
| 484 |
* potentially inconsistent images to be opened read-only. This can
|
| 485 |
* aid data recovery from an otherwise inconsistent image.
|
| 486 |
*/
|
| 487 |
if (!bdrv_is_read_only(bs->file) &&
|
| 488 |
!(flags & BDRV_O_INCOMING)) {
|
| 489 |
BdrvCheckResult result = {0};
|
| 490 |
|
| 491 |
ret = qed_check(s, &result, true);
|
| 492 |
if (ret) {
|
| 493 |
goto out;
|
| 494 |
} |
| 495 |
} |
| 496 |
} |
| 497 |
|
| 498 |
s->need_check_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, |
| 499 |
qed_need_check_timer_cb, s); |
| 500 |
|
| 501 |
out:
|
| 502 |
if (ret) {
|
| 503 |
qed_free_l2_cache(&s->l2_cache); |
| 504 |
qemu_vfree(s->l1_table); |
| 505 |
} |
| 506 |
return ret;
|
| 507 |
} |
| 508 |
|
| 509 |
/* We have nothing to do for QED reopen, stubs just return
|
| 510 |
* success */
|
| 511 |
static int bdrv_qed_reopen_prepare(BDRVReopenState *state, |
| 512 |
BlockReopenQueue *queue, Error **errp) |
| 513 |
{
|
| 514 |
return 0; |
| 515 |
} |
| 516 |
|
| 517 |
static void bdrv_qed_close(BlockDriverState *bs) |
| 518 |
{
|
| 519 |
BDRVQEDState *s = bs->opaque; |
| 520 |
|
| 521 |
qed_cancel_need_check_timer(s); |
| 522 |
timer_free(s->need_check_timer); |
| 523 |
|
| 524 |
/* Ensure writes reach stable storage */
|
| 525 |
bdrv_flush(bs->file); |
| 526 |
|
| 527 |
/* Clean shutdown, no check required on next open */
|
| 528 |
if (s->header.features & QED_F_NEED_CHECK) {
|
| 529 |
s->header.features &= ~QED_F_NEED_CHECK; |
| 530 |
qed_write_header_sync(s); |
| 531 |
} |
| 532 |
|
| 533 |
qed_free_l2_cache(&s->l2_cache); |
| 534 |
qemu_vfree(s->l1_table); |
| 535 |
} |
| 536 |
|
| 537 |
static int qed_create(const char *filename, uint32_t cluster_size, |
| 538 |
uint64_t image_size, uint32_t table_size, |
| 539 |
const char *backing_file, const char *backing_fmt) |
| 540 |
{
|
| 541 |
QEDHeader header = {
|
| 542 |
.magic = QED_MAGIC, |
| 543 |
.cluster_size = cluster_size, |
| 544 |
.table_size = table_size, |
| 545 |
.header_size = 1,
|
| 546 |
.features = 0,
|
| 547 |
.compat_features = 0,
|
| 548 |
.l1_table_offset = cluster_size, |
| 549 |
.image_size = image_size, |
| 550 |
}; |
| 551 |
QEDHeader le_header; |
| 552 |
uint8_t *l1_table = NULL;
|
| 553 |
size_t l1_size = header.cluster_size * header.table_size; |
| 554 |
Error *local_err = NULL;
|
| 555 |
int ret = 0; |
| 556 |
BlockDriverState *bs = NULL;
|
| 557 |
|
| 558 |
ret = bdrv_create_file(filename, NULL);
|
| 559 |
if (ret < 0) { |
| 560 |
return ret;
|
| 561 |
} |
| 562 |
|
| 563 |
ret = bdrv_file_open(&bs, filename, NULL, BDRV_O_RDWR | BDRV_O_CACHE_WB,
|
| 564 |
&local_err); |
| 565 |
if (ret < 0) { |
| 566 |
qerror_report_err(local_err); |
| 567 |
error_free(local_err); |
| 568 |
return ret;
|
| 569 |
} |
| 570 |
|
| 571 |
/* File must start empty and grow, check truncate is supported */
|
| 572 |
ret = bdrv_truncate(bs, 0);
|
| 573 |
if (ret < 0) { |
| 574 |
goto out;
|
| 575 |
} |
| 576 |
|
| 577 |
if (backing_file) {
|
| 578 |
header.features |= QED_F_BACKING_FILE; |
| 579 |
header.backing_filename_offset = sizeof(le_header);
|
| 580 |
header.backing_filename_size = strlen(backing_file); |
| 581 |
|
| 582 |
if (qed_fmt_is_raw(backing_fmt)) {
|
| 583 |
header.features |= QED_F_BACKING_FORMAT_NO_PROBE; |
| 584 |
} |
| 585 |
} |
| 586 |
|
| 587 |
qed_header_cpu_to_le(&header, &le_header); |
| 588 |
ret = bdrv_pwrite(bs, 0, &le_header, sizeof(le_header)); |
| 589 |
if (ret < 0) { |
| 590 |
goto out;
|
| 591 |
} |
| 592 |
ret = bdrv_pwrite(bs, sizeof(le_header), backing_file,
|
| 593 |
header.backing_filename_size); |
| 594 |
if (ret < 0) { |
| 595 |
goto out;
|
| 596 |
} |
| 597 |
|
| 598 |
l1_table = g_malloc0(l1_size); |
| 599 |
ret = bdrv_pwrite(bs, header.l1_table_offset, l1_table, l1_size); |
| 600 |
if (ret < 0) { |
| 601 |
goto out;
|
| 602 |
} |
| 603 |
|
| 604 |
ret = 0; /* success */ |
| 605 |
out:
|
| 606 |
g_free(l1_table); |
| 607 |
bdrv_unref(bs); |
| 608 |
return ret;
|
| 609 |
} |
| 610 |
|
| 611 |
static int bdrv_qed_create(const char *filename, QEMUOptionParameter *options, |
| 612 |
Error **errp) |
| 613 |
{
|
| 614 |
uint64_t image_size = 0;
|
| 615 |
uint32_t cluster_size = QED_DEFAULT_CLUSTER_SIZE; |
| 616 |
uint32_t table_size = QED_DEFAULT_TABLE_SIZE; |
| 617 |
const char *backing_file = NULL; |
| 618 |
const char *backing_fmt = NULL; |
| 619 |
|
| 620 |
while (options && options->name) {
|
| 621 |
if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
|
| 622 |
image_size = options->value.n; |
| 623 |
} else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { |
| 624 |
backing_file = options->value.s; |
| 625 |
} else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) { |
| 626 |
backing_fmt = options->value.s; |
| 627 |
} else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) { |
| 628 |
if (options->value.n) {
|
| 629 |
cluster_size = options->value.n; |
| 630 |
} |
| 631 |
} else if (!strcmp(options->name, BLOCK_OPT_TABLE_SIZE)) { |
| 632 |
if (options->value.n) {
|
| 633 |
table_size = options->value.n; |
| 634 |
} |
| 635 |
} |
| 636 |
options++; |
| 637 |
} |
| 638 |
|
| 639 |
if (!qed_is_cluster_size_valid(cluster_size)) {
|
| 640 |
fprintf(stderr, "QED cluster size must be within range [%u, %u] and power of 2\n",
|
| 641 |
QED_MIN_CLUSTER_SIZE, QED_MAX_CLUSTER_SIZE); |
| 642 |
return -EINVAL;
|
| 643 |
} |
| 644 |
if (!qed_is_table_size_valid(table_size)) {
|
| 645 |
fprintf(stderr, "QED table size must be within range [%u, %u] and power of 2\n",
|
| 646 |
QED_MIN_TABLE_SIZE, QED_MAX_TABLE_SIZE); |
| 647 |
return -EINVAL;
|
| 648 |
} |
| 649 |
if (!qed_is_image_size_valid(image_size, cluster_size, table_size)) {
|
| 650 |
fprintf(stderr, "QED image size must be a non-zero multiple of "
|
| 651 |
"cluster size and less than %" PRIu64 " bytes\n", |
| 652 |
qed_max_image_size(cluster_size, table_size)); |
| 653 |
return -EINVAL;
|
| 654 |
} |
| 655 |
|
| 656 |
return qed_create(filename, cluster_size, image_size, table_size,
|
| 657 |
backing_file, backing_fmt); |
| 658 |
} |
| 659 |
|
| 660 |
typedef struct { |
| 661 |
BlockDriverState *bs; |
| 662 |
Coroutine *co; |
| 663 |
uint64_t pos; |
| 664 |
int64_t status; |
| 665 |
int *pnum;
|
| 666 |
} QEDIsAllocatedCB; |
| 667 |
|
| 668 |
static void qed_is_allocated_cb(void *opaque, int ret, uint64_t offset, size_t len) |
| 669 |
{
|
| 670 |
QEDIsAllocatedCB *cb = opaque; |
| 671 |
BDRVQEDState *s = cb->bs->opaque; |
| 672 |
*cb->pnum = len / BDRV_SECTOR_SIZE; |
| 673 |
switch (ret) {
|
| 674 |
case QED_CLUSTER_FOUND:
|
| 675 |
offset |= qed_offset_into_cluster(s, cb->pos); |
| 676 |
cb->status = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset; |
| 677 |
break;
|
| 678 |
case QED_CLUSTER_ZERO:
|
| 679 |
cb->status = BDRV_BLOCK_ZERO; |
| 680 |
break;
|
| 681 |
case QED_CLUSTER_L2:
|
| 682 |
case QED_CLUSTER_L1:
|
| 683 |
cb->status = 0;
|
| 684 |
break;
|
| 685 |
default:
|
| 686 |
assert(ret < 0);
|
| 687 |
cb->status = ret; |
| 688 |
break;
|
| 689 |
} |
| 690 |
|
| 691 |
if (cb->co) {
|
| 692 |
qemu_coroutine_enter(cb->co, NULL);
|
| 693 |
} |
| 694 |
} |
| 695 |
|
| 696 |
static int64_t coroutine_fn bdrv_qed_co_get_block_status(BlockDriverState *bs,
|
| 697 |
int64_t sector_num, |
| 698 |
int nb_sectors, int *pnum) |
| 699 |
{
|
| 700 |
BDRVQEDState *s = bs->opaque; |
| 701 |
size_t len = (size_t)nb_sectors * BDRV_SECTOR_SIZE; |
| 702 |
QEDIsAllocatedCB cb = {
|
| 703 |
.bs = bs, |
| 704 |
.pos = (uint64_t)sector_num * BDRV_SECTOR_SIZE, |
| 705 |
.status = BDRV_BLOCK_OFFSET_MASK, |
| 706 |
.pnum = pnum, |
| 707 |
}; |
| 708 |
QEDRequest request = { .l2_table = NULL };
|
| 709 |
|
| 710 |
qed_find_cluster(s, &request, cb.pos, len, qed_is_allocated_cb, &cb); |
| 711 |
|
| 712 |
/* Now sleep if the callback wasn't invoked immediately */
|
| 713 |
while (cb.status == BDRV_BLOCK_OFFSET_MASK) {
|
| 714 |
cb.co = qemu_coroutine_self(); |
| 715 |
qemu_coroutine_yield(); |
| 716 |
} |
| 717 |
|
| 718 |
qed_unref_l2_cache_entry(request.l2_table); |
| 719 |
|
| 720 |
return cb.status;
|
| 721 |
} |
| 722 |
|
| 723 |
static int bdrv_qed_make_empty(BlockDriverState *bs) |
| 724 |
{
|
| 725 |
return -ENOTSUP;
|
| 726 |
} |
| 727 |
|
| 728 |
static BDRVQEDState *acb_to_s(QEDAIOCB *acb)
|
| 729 |
{
|
| 730 |
return acb->common.bs->opaque;
|
| 731 |
} |
| 732 |
|
| 733 |
/**
|
| 734 |
* Read from the backing file or zero-fill if no backing file
|
| 735 |
*
|
| 736 |
* @s: QED state
|
| 737 |
* @pos: Byte position in device
|
| 738 |
* @qiov: Destination I/O vector
|
| 739 |
* @cb: Completion function
|
| 740 |
* @opaque: User data for completion function
|
| 741 |
*
|
| 742 |
* This function reads qiov->size bytes starting at pos from the backing file.
|
| 743 |
* If there is no backing file then zeroes are read.
|
| 744 |
*/
|
| 745 |
static void qed_read_backing_file(BDRVQEDState *s, uint64_t pos, |
| 746 |
QEMUIOVector *qiov, |
| 747 |
BlockDriverCompletionFunc *cb, void *opaque)
|
| 748 |
{
|
| 749 |
uint64_t backing_length = 0;
|
| 750 |
size_t size; |
| 751 |
|
| 752 |
/* If there is a backing file, get its length. Treat the absence of a
|
| 753 |
* backing file like a zero length backing file.
|
| 754 |
*/
|
| 755 |
if (s->bs->backing_hd) {
|
| 756 |
int64_t l = bdrv_getlength(s->bs->backing_hd); |
| 757 |
if (l < 0) { |
| 758 |
cb(opaque, l); |
| 759 |
return;
|
| 760 |
} |
| 761 |
backing_length = l; |
| 762 |
} |
| 763 |
|
| 764 |
/* Zero all sectors if reading beyond the end of the backing file */
|
| 765 |
if (pos >= backing_length ||
|
| 766 |
pos + qiov->size > backing_length) {
|
| 767 |
qemu_iovec_memset(qiov, 0, 0, qiov->size); |
| 768 |
} |
| 769 |
|
| 770 |
/* Complete now if there are no backing file sectors to read */
|
| 771 |
if (pos >= backing_length) {
|
| 772 |
cb(opaque, 0);
|
| 773 |
return;
|
| 774 |
} |
| 775 |
|
| 776 |
/* If the read straddles the end of the backing file, shorten it */
|
| 777 |
size = MIN((uint64_t)backing_length - pos, qiov->size); |
| 778 |
|
| 779 |
BLKDBG_EVENT(s->bs->file, BLKDBG_READ_BACKING_AIO); |
| 780 |
bdrv_aio_readv(s->bs->backing_hd, pos / BDRV_SECTOR_SIZE, |
| 781 |
qiov, size / BDRV_SECTOR_SIZE, cb, opaque); |
| 782 |
} |
| 783 |
|
| 784 |
typedef struct { |
| 785 |
GenericCB gencb; |
| 786 |
BDRVQEDState *s; |
| 787 |
QEMUIOVector qiov; |
| 788 |
struct iovec iov;
|
| 789 |
uint64_t offset; |
| 790 |
} CopyFromBackingFileCB; |
| 791 |
|
| 792 |
static void qed_copy_from_backing_file_cb(void *opaque, int ret) |
| 793 |
{
|
| 794 |
CopyFromBackingFileCB *copy_cb = opaque; |
| 795 |
qemu_vfree(copy_cb->iov.iov_base); |
| 796 |
gencb_complete(©_cb->gencb, ret); |
| 797 |
} |
| 798 |
|
| 799 |
static void qed_copy_from_backing_file_write(void *opaque, int ret) |
| 800 |
{
|
| 801 |
CopyFromBackingFileCB *copy_cb = opaque; |
| 802 |
BDRVQEDState *s = copy_cb->s; |
| 803 |
|
| 804 |
if (ret) {
|
| 805 |
qed_copy_from_backing_file_cb(copy_cb, ret); |
| 806 |
return;
|
| 807 |
} |
| 808 |
|
| 809 |
BLKDBG_EVENT(s->bs->file, BLKDBG_COW_WRITE); |
| 810 |
bdrv_aio_writev(s->bs->file, copy_cb->offset / BDRV_SECTOR_SIZE, |
| 811 |
©_cb->qiov, copy_cb->qiov.size / BDRV_SECTOR_SIZE, |
| 812 |
qed_copy_from_backing_file_cb, copy_cb); |
| 813 |
} |
| 814 |
|
| 815 |
/**
|
| 816 |
* Copy data from backing file into the image
|
| 817 |
*
|
| 818 |
* @s: QED state
|
| 819 |
* @pos: Byte position in device
|
| 820 |
* @len: Number of bytes
|
| 821 |
* @offset: Byte offset in image file
|
| 822 |
* @cb: Completion function
|
| 823 |
* @opaque: User data for completion function
|
| 824 |
*/
|
| 825 |
static void qed_copy_from_backing_file(BDRVQEDState *s, uint64_t pos, |
| 826 |
uint64_t len, uint64_t offset, |
| 827 |
BlockDriverCompletionFunc *cb, |
| 828 |
void *opaque)
|
| 829 |
{
|
| 830 |
CopyFromBackingFileCB *copy_cb; |
| 831 |
|
| 832 |
/* Skip copy entirely if there is no work to do */
|
| 833 |
if (len == 0) { |
| 834 |
cb(opaque, 0);
|
| 835 |
return;
|
| 836 |
} |
| 837 |
|
| 838 |
copy_cb = gencb_alloc(sizeof(*copy_cb), cb, opaque);
|
| 839 |
copy_cb->s = s; |
| 840 |
copy_cb->offset = offset; |
| 841 |
copy_cb->iov.iov_base = qemu_blockalign(s->bs, len); |
| 842 |
copy_cb->iov.iov_len = len; |
| 843 |
qemu_iovec_init_external(©_cb->qiov, ©_cb->iov, 1);
|
| 844 |
|
| 845 |
qed_read_backing_file(s, pos, ©_cb->qiov, |
| 846 |
qed_copy_from_backing_file_write, copy_cb); |
| 847 |
} |
| 848 |
|
| 849 |
/**
|
| 850 |
* Link one or more contiguous clusters into a table
|
| 851 |
*
|
| 852 |
* @s: QED state
|
| 853 |
* @table: L2 table
|
| 854 |
* @index: First cluster index
|
| 855 |
* @n: Number of contiguous clusters
|
| 856 |
* @cluster: First cluster offset
|
| 857 |
*
|
| 858 |
* The cluster offset may be an allocated byte offset in the image file, the
|
| 859 |
* zero cluster marker, or the unallocated cluster marker.
|
| 860 |
*/
|
| 861 |
static void qed_update_l2_table(BDRVQEDState *s, QEDTable *table, int index, |
| 862 |
unsigned int n, uint64_t cluster) |
| 863 |
{
|
| 864 |
int i;
|
| 865 |
for (i = index; i < index + n; i++) {
|
| 866 |
table->offsets[i] = cluster; |
| 867 |
if (!qed_offset_is_unalloc_cluster(cluster) &&
|
| 868 |
!qed_offset_is_zero_cluster(cluster)) {
|
| 869 |
cluster += s->header.cluster_size; |
| 870 |
} |
| 871 |
} |
| 872 |
} |
| 873 |
|
| 874 |
static void qed_aio_complete_bh(void *opaque) |
| 875 |
{
|
| 876 |
QEDAIOCB *acb = opaque; |
| 877 |
BlockDriverCompletionFunc *cb = acb->common.cb; |
| 878 |
void *user_opaque = acb->common.opaque;
|
| 879 |
int ret = acb->bh_ret;
|
| 880 |
bool *finished = acb->finished;
|
| 881 |
|
| 882 |
qemu_bh_delete(acb->bh); |
| 883 |
qemu_aio_release(acb); |
| 884 |
|
| 885 |
/* Invoke callback */
|
| 886 |
cb(user_opaque, ret); |
| 887 |
|
| 888 |
/* Signal cancel completion */
|
| 889 |
if (finished) {
|
| 890 |
*finished = true;
|
| 891 |
} |
| 892 |
} |
| 893 |
|
| 894 |
static void qed_aio_complete(QEDAIOCB *acb, int ret) |
| 895 |
{
|
| 896 |
BDRVQEDState *s = acb_to_s(acb); |
| 897 |
|
| 898 |
trace_qed_aio_complete(s, acb, ret); |
| 899 |
|
| 900 |
/* Free resources */
|
| 901 |
qemu_iovec_destroy(&acb->cur_qiov); |
| 902 |
qed_unref_l2_cache_entry(acb->request.l2_table); |
| 903 |
|
| 904 |
/* Free the buffer we may have allocated for zero writes */
|
| 905 |
if (acb->flags & QED_AIOCB_ZERO) {
|
| 906 |
qemu_vfree(acb->qiov->iov[0].iov_base);
|
| 907 |
acb->qiov->iov[0].iov_base = NULL; |
| 908 |
} |
| 909 |
|
| 910 |
/* Arrange for a bh to invoke the completion function */
|
| 911 |
acb->bh_ret = ret; |
| 912 |
acb->bh = qemu_bh_new(qed_aio_complete_bh, acb); |
| 913 |
qemu_bh_schedule(acb->bh); |
| 914 |
|
| 915 |
/* Start next allocating write request waiting behind this one. Note that
|
| 916 |
* requests enqueue themselves when they first hit an unallocated cluster
|
| 917 |
* but they wait until the entire request is finished before waking up the
|
| 918 |
* next request in the queue. This ensures that we don't cycle through
|
| 919 |
* requests multiple times but rather finish one at a time completely.
|
| 920 |
*/
|
| 921 |
if (acb == QSIMPLEQ_FIRST(&s->allocating_write_reqs)) {
|
| 922 |
QSIMPLEQ_REMOVE_HEAD(&s->allocating_write_reqs, next); |
| 923 |
acb = QSIMPLEQ_FIRST(&s->allocating_write_reqs); |
| 924 |
if (acb) {
|
| 925 |
qed_aio_next_io(acb, 0);
|
| 926 |
} else if (s->header.features & QED_F_NEED_CHECK) { |
| 927 |
qed_start_need_check_timer(s); |
| 928 |
} |
| 929 |
} |
| 930 |
} |
| 931 |
|
| 932 |
/**
|
| 933 |
* Commit the current L2 table to the cache
|
| 934 |
*/
|
| 935 |
static void qed_commit_l2_update(void *opaque, int ret) |
| 936 |
{
|
| 937 |
QEDAIOCB *acb = opaque; |
| 938 |
BDRVQEDState *s = acb_to_s(acb); |
| 939 |
CachedL2Table *l2_table = acb->request.l2_table; |
| 940 |
uint64_t l2_offset = l2_table->offset; |
| 941 |
|
| 942 |
qed_commit_l2_cache_entry(&s->l2_cache, l2_table); |
| 943 |
|
| 944 |
/* This is guaranteed to succeed because we just committed the entry to the
|
| 945 |
* cache.
|
| 946 |
*/
|
| 947 |
acb->request.l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset); |
| 948 |
assert(acb->request.l2_table != NULL);
|
| 949 |
|
| 950 |
qed_aio_next_io(opaque, ret); |
| 951 |
} |
| 952 |
|
| 953 |
/**
|
| 954 |
* Update L1 table with new L2 table offset and write it out
|
| 955 |
*/
|
| 956 |
static void qed_aio_write_l1_update(void *opaque, int ret) |
| 957 |
{
|
| 958 |
QEDAIOCB *acb = opaque; |
| 959 |
BDRVQEDState *s = acb_to_s(acb); |
| 960 |
int index;
|
| 961 |
|
| 962 |
if (ret) {
|
| 963 |
qed_aio_complete(acb, ret); |
| 964 |
return;
|
| 965 |
} |
| 966 |
|
| 967 |
index = qed_l1_index(s, acb->cur_pos); |
| 968 |
s->l1_table->offsets[index] = acb->request.l2_table->offset; |
| 969 |
|
| 970 |
qed_write_l1_table(s, index, 1, qed_commit_l2_update, acb);
|
| 971 |
} |
| 972 |
|
| 973 |
/**
|
| 974 |
* Update L2 table with new cluster offsets and write them out
|
| 975 |
*/
|
| 976 |
static void qed_aio_write_l2_update(QEDAIOCB *acb, int ret, uint64_t offset) |
| 977 |
{
|
| 978 |
BDRVQEDState *s = acb_to_s(acb); |
| 979 |
bool need_alloc = acb->find_cluster_ret == QED_CLUSTER_L1;
|
| 980 |
int index;
|
| 981 |
|
| 982 |
if (ret) {
|
| 983 |
goto err;
|
| 984 |
} |
| 985 |
|
| 986 |
if (need_alloc) {
|
| 987 |
qed_unref_l2_cache_entry(acb->request.l2_table); |
| 988 |
acb->request.l2_table = qed_new_l2_table(s); |
| 989 |
} |
| 990 |
|
| 991 |
index = qed_l2_index(s, acb->cur_pos); |
| 992 |
qed_update_l2_table(s, acb->request.l2_table->table, index, acb->cur_nclusters, |
| 993 |
offset); |
| 994 |
|
| 995 |
if (need_alloc) {
|
| 996 |
/* Write out the whole new L2 table */
|
| 997 |
qed_write_l2_table(s, &acb->request, 0, s->table_nelems, true, |
| 998 |
qed_aio_write_l1_update, acb); |
| 999 |
} else {
|
| 1000 |
/* Write out only the updated part of the L2 table */
|
| 1001 |
qed_write_l2_table(s, &acb->request, index, acb->cur_nclusters, false,
|
| 1002 |
qed_aio_next_io, acb); |
| 1003 |
} |
| 1004 |
return;
|
| 1005 |
|
| 1006 |
err:
|
| 1007 |
qed_aio_complete(acb, ret); |
| 1008 |
} |
| 1009 |
|
| 1010 |
static void qed_aio_write_l2_update_cb(void *opaque, int ret) |
| 1011 |
{
|
| 1012 |
QEDAIOCB *acb = opaque; |
| 1013 |
qed_aio_write_l2_update(acb, ret, acb->cur_cluster); |
| 1014 |
} |
| 1015 |
|
| 1016 |
/**
|
| 1017 |
* Flush new data clusters before updating the L2 table
|
| 1018 |
*
|
| 1019 |
* This flush is necessary when a backing file is in use. A crash during an
|
| 1020 |
* allocating write could result in empty clusters in the image. If the write
|
| 1021 |
* only touched a subregion of the cluster, then backing image sectors have
|
| 1022 |
* been lost in the untouched region. The solution is to flush after writing a
|
| 1023 |
* new data cluster and before updating the L2 table.
|
| 1024 |
*/
|
| 1025 |
static void qed_aio_write_flush_before_l2_update(void *opaque, int ret) |
| 1026 |
{
|
| 1027 |
QEDAIOCB *acb = opaque; |
| 1028 |
BDRVQEDState *s = acb_to_s(acb); |
| 1029 |
|
| 1030 |
if (!bdrv_aio_flush(s->bs->file, qed_aio_write_l2_update_cb, opaque)) {
|
| 1031 |
qed_aio_complete(acb, -EIO); |
| 1032 |
} |
| 1033 |
} |
| 1034 |
|
| 1035 |
/**
|
| 1036 |
* Write data to the image file
|
| 1037 |
*/
|
| 1038 |
static void qed_aio_write_main(void *opaque, int ret) |
| 1039 |
{
|
| 1040 |
QEDAIOCB *acb = opaque; |
| 1041 |
BDRVQEDState *s = acb_to_s(acb); |
| 1042 |
uint64_t offset = acb->cur_cluster + |
| 1043 |
qed_offset_into_cluster(s, acb->cur_pos); |
| 1044 |
BlockDriverCompletionFunc *next_fn; |
| 1045 |
|
| 1046 |
trace_qed_aio_write_main(s, acb, ret, offset, acb->cur_qiov.size); |
| 1047 |
|
| 1048 |
if (ret) {
|
| 1049 |
qed_aio_complete(acb, ret); |
| 1050 |
return;
|
| 1051 |
} |
| 1052 |
|
| 1053 |
if (acb->find_cluster_ret == QED_CLUSTER_FOUND) {
|
| 1054 |
next_fn = qed_aio_next_io; |
| 1055 |
} else {
|
| 1056 |
if (s->bs->backing_hd) {
|
| 1057 |
next_fn = qed_aio_write_flush_before_l2_update; |
| 1058 |
} else {
|
| 1059 |
next_fn = qed_aio_write_l2_update_cb; |
| 1060 |
} |
| 1061 |
} |
| 1062 |
|
| 1063 |
BLKDBG_EVENT(s->bs->file, BLKDBG_WRITE_AIO); |
| 1064 |
bdrv_aio_writev(s->bs->file, offset / BDRV_SECTOR_SIZE, |
| 1065 |
&acb->cur_qiov, acb->cur_qiov.size / BDRV_SECTOR_SIZE, |
| 1066 |
next_fn, acb); |
| 1067 |
} |
| 1068 |
|
| 1069 |
/**
|
| 1070 |
* Populate back untouched region of new data cluster
|
| 1071 |
*/
|
| 1072 |
static void qed_aio_write_postfill(void *opaque, int ret) |
| 1073 |
{
|
| 1074 |
QEDAIOCB *acb = opaque; |
| 1075 |
BDRVQEDState *s = acb_to_s(acb); |
| 1076 |
uint64_t start = acb->cur_pos + acb->cur_qiov.size; |
| 1077 |
uint64_t len = |
| 1078 |
qed_start_of_cluster(s, start + s->header.cluster_size - 1) - start;
|
| 1079 |
uint64_t offset = acb->cur_cluster + |
| 1080 |
qed_offset_into_cluster(s, acb->cur_pos) + |
| 1081 |
acb->cur_qiov.size; |
| 1082 |
|
| 1083 |
if (ret) {
|
| 1084 |
qed_aio_complete(acb, ret); |
| 1085 |
return;
|
| 1086 |
} |
| 1087 |
|
| 1088 |
trace_qed_aio_write_postfill(s, acb, start, len, offset); |
| 1089 |
qed_copy_from_backing_file(s, start, len, offset, |
| 1090 |
qed_aio_write_main, acb); |
| 1091 |
} |
| 1092 |
|
| 1093 |
/**
|
| 1094 |
* Populate front untouched region of new data cluster
|
| 1095 |
*/
|
| 1096 |
static void qed_aio_write_prefill(void *opaque, int ret) |
| 1097 |
{
|
| 1098 |
QEDAIOCB *acb = opaque; |
| 1099 |
BDRVQEDState *s = acb_to_s(acb); |
| 1100 |
uint64_t start = qed_start_of_cluster(s, acb->cur_pos); |
| 1101 |
uint64_t len = qed_offset_into_cluster(s, acb->cur_pos); |
| 1102 |
|
| 1103 |
trace_qed_aio_write_prefill(s, acb, start, len, acb->cur_cluster); |
| 1104 |
qed_copy_from_backing_file(s, start, len, acb->cur_cluster, |
| 1105 |
qed_aio_write_postfill, acb); |
| 1106 |
} |
| 1107 |
|
| 1108 |
/**
|
| 1109 |
* Check if the QED_F_NEED_CHECK bit should be set during allocating write
|
| 1110 |
*/
|
| 1111 |
static bool qed_should_set_need_check(BDRVQEDState *s) |
| 1112 |
{
|
| 1113 |
/* The flush before L2 update path ensures consistency */
|
| 1114 |
if (s->bs->backing_hd) {
|
| 1115 |
return false; |
| 1116 |
} |
| 1117 |
|
| 1118 |
return !(s->header.features & QED_F_NEED_CHECK);
|
| 1119 |
} |
| 1120 |
|
| 1121 |
static void qed_aio_write_zero_cluster(void *opaque, int ret) |
| 1122 |
{
|
| 1123 |
QEDAIOCB *acb = opaque; |
| 1124 |
|
| 1125 |
if (ret) {
|
| 1126 |
qed_aio_complete(acb, ret); |
| 1127 |
return;
|
| 1128 |
} |
| 1129 |
|
| 1130 |
qed_aio_write_l2_update(acb, 0, 1); |
| 1131 |
} |
| 1132 |
|
| 1133 |
/**
|
| 1134 |
* Write new data cluster
|
| 1135 |
*
|
| 1136 |
* @acb: Write request
|
| 1137 |
* @len: Length in bytes
|
| 1138 |
*
|
| 1139 |
* This path is taken when writing to previously unallocated clusters.
|
| 1140 |
*/
|
| 1141 |
static void qed_aio_write_alloc(QEDAIOCB *acb, size_t len) |
| 1142 |
{
|
| 1143 |
BDRVQEDState *s = acb_to_s(acb); |
| 1144 |
BlockDriverCompletionFunc *cb; |
| 1145 |
|
| 1146 |
/* Cancel timer when the first allocating request comes in */
|
| 1147 |
if (QSIMPLEQ_EMPTY(&s->allocating_write_reqs)) {
|
| 1148 |
qed_cancel_need_check_timer(s); |
| 1149 |
} |
| 1150 |
|
| 1151 |
/* Freeze this request if another allocating write is in progress */
|
| 1152 |
if (acb != QSIMPLEQ_FIRST(&s->allocating_write_reqs)) {
|
| 1153 |
QSIMPLEQ_INSERT_TAIL(&s->allocating_write_reqs, acb, next); |
| 1154 |
} |
| 1155 |
if (acb != QSIMPLEQ_FIRST(&s->allocating_write_reqs) ||
|
| 1156 |
s->allocating_write_reqs_plugged) {
|
| 1157 |
return; /* wait for existing request to finish */ |
| 1158 |
} |
| 1159 |
|
| 1160 |
acb->cur_nclusters = qed_bytes_to_clusters(s, |
| 1161 |
qed_offset_into_cluster(s, acb->cur_pos) + len); |
| 1162 |
qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); |
| 1163 |
|
| 1164 |
if (acb->flags & QED_AIOCB_ZERO) {
|
| 1165 |
/* Skip ahead if the clusters are already zero */
|
| 1166 |
if (acb->find_cluster_ret == QED_CLUSTER_ZERO) {
|
| 1167 |
qed_aio_next_io(acb, 0);
|
| 1168 |
return;
|
| 1169 |
} |
| 1170 |
|
| 1171 |
cb = qed_aio_write_zero_cluster; |
| 1172 |
} else {
|
| 1173 |
cb = qed_aio_write_prefill; |
| 1174 |
acb->cur_cluster = qed_alloc_clusters(s, acb->cur_nclusters); |
| 1175 |
} |
| 1176 |
|
| 1177 |
if (qed_should_set_need_check(s)) {
|
| 1178 |
s->header.features |= QED_F_NEED_CHECK; |
| 1179 |
qed_write_header(s, cb, acb); |
| 1180 |
} else {
|
| 1181 |
cb(acb, 0);
|
| 1182 |
} |
| 1183 |
} |
| 1184 |
|
| 1185 |
/**
|
| 1186 |
* Write data cluster in place
|
| 1187 |
*
|
| 1188 |
* @acb: Write request
|
| 1189 |
* @offset: Cluster offset in bytes
|
| 1190 |
* @len: Length in bytes
|
| 1191 |
*
|
| 1192 |
* This path is taken when writing to already allocated clusters.
|
| 1193 |
*/
|
| 1194 |
static void qed_aio_write_inplace(QEDAIOCB *acb, uint64_t offset, size_t len) |
| 1195 |
{
|
| 1196 |
/* Allocate buffer for zero writes */
|
| 1197 |
if (acb->flags & QED_AIOCB_ZERO) {
|
| 1198 |
struct iovec *iov = acb->qiov->iov;
|
| 1199 |
|
| 1200 |
if (!iov->iov_base) {
|
| 1201 |
iov->iov_base = qemu_blockalign(acb->common.bs, iov->iov_len); |
| 1202 |
memset(iov->iov_base, 0, iov->iov_len);
|
| 1203 |
} |
| 1204 |
} |
| 1205 |
|
| 1206 |
/* Calculate the I/O vector */
|
| 1207 |
acb->cur_cluster = offset; |
| 1208 |
qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); |
| 1209 |
|
| 1210 |
/* Do the actual write */
|
| 1211 |
qed_aio_write_main(acb, 0);
|
| 1212 |
} |
| 1213 |
|
| 1214 |
/**
|
| 1215 |
* Write data cluster
|
| 1216 |
*
|
| 1217 |
* @opaque: Write request
|
| 1218 |
* @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2, QED_CLUSTER_L1,
|
| 1219 |
* or -errno
|
| 1220 |
* @offset: Cluster offset in bytes
|
| 1221 |
* @len: Length in bytes
|
| 1222 |
*
|
| 1223 |
* Callback from qed_find_cluster().
|
| 1224 |
*/
|
| 1225 |
static void qed_aio_write_data(void *opaque, int ret, |
| 1226 |
uint64_t offset, size_t len) |
| 1227 |
{
|
| 1228 |
QEDAIOCB *acb = opaque; |
| 1229 |
|
| 1230 |
trace_qed_aio_write_data(acb_to_s(acb), acb, ret, offset, len); |
| 1231 |
|
| 1232 |
acb->find_cluster_ret = ret; |
| 1233 |
|
| 1234 |
switch (ret) {
|
| 1235 |
case QED_CLUSTER_FOUND:
|
| 1236 |
qed_aio_write_inplace(acb, offset, len); |
| 1237 |
break;
|
| 1238 |
|
| 1239 |
case QED_CLUSTER_L2:
|
| 1240 |
case QED_CLUSTER_L1:
|
| 1241 |
case QED_CLUSTER_ZERO:
|
| 1242 |
qed_aio_write_alloc(acb, len); |
| 1243 |
break;
|
| 1244 |
|
| 1245 |
default:
|
| 1246 |
qed_aio_complete(acb, ret); |
| 1247 |
break;
|
| 1248 |
} |
| 1249 |
} |
| 1250 |
|
| 1251 |
/**
|
| 1252 |
* Read data cluster
|
| 1253 |
*
|
| 1254 |
* @opaque: Read request
|
| 1255 |
* @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2, QED_CLUSTER_L1,
|
| 1256 |
* or -errno
|
| 1257 |
* @offset: Cluster offset in bytes
|
| 1258 |
* @len: Length in bytes
|
| 1259 |
*
|
| 1260 |
* Callback from qed_find_cluster().
|
| 1261 |
*/
|
| 1262 |
static void qed_aio_read_data(void *opaque, int ret, |
| 1263 |
uint64_t offset, size_t len) |
| 1264 |
{
|
| 1265 |
QEDAIOCB *acb = opaque; |
| 1266 |
BDRVQEDState *s = acb_to_s(acb); |
| 1267 |
BlockDriverState *bs = acb->common.bs; |
| 1268 |
|
| 1269 |
/* Adjust offset into cluster */
|
| 1270 |
offset += qed_offset_into_cluster(s, acb->cur_pos); |
| 1271 |
|
| 1272 |
trace_qed_aio_read_data(s, acb, ret, offset, len); |
| 1273 |
|
| 1274 |
if (ret < 0) { |
| 1275 |
goto err;
|
| 1276 |
} |
| 1277 |
|
| 1278 |
qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); |
| 1279 |
|
| 1280 |
/* Handle zero cluster and backing file reads */
|
| 1281 |
if (ret == QED_CLUSTER_ZERO) {
|
| 1282 |
qemu_iovec_memset(&acb->cur_qiov, 0, 0, acb->cur_qiov.size); |
| 1283 |
qed_aio_next_io(acb, 0);
|
| 1284 |
return;
|
| 1285 |
} else if (ret != QED_CLUSTER_FOUND) { |
| 1286 |
qed_read_backing_file(s, acb->cur_pos, &acb->cur_qiov, |
| 1287 |
qed_aio_next_io, acb); |
| 1288 |
return;
|
| 1289 |
} |
| 1290 |
|
| 1291 |
BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); |
| 1292 |
bdrv_aio_readv(bs->file, offset / BDRV_SECTOR_SIZE, |
| 1293 |
&acb->cur_qiov, acb->cur_qiov.size / BDRV_SECTOR_SIZE, |
| 1294 |
qed_aio_next_io, acb); |
| 1295 |
return;
|
| 1296 |
|
| 1297 |
err:
|
| 1298 |
qed_aio_complete(acb, ret); |
| 1299 |
} |
| 1300 |
|
| 1301 |
/**
|
| 1302 |
* Begin next I/O or complete the request
|
| 1303 |
*/
|
| 1304 |
static void qed_aio_next_io(void *opaque, int ret) |
| 1305 |
{
|
| 1306 |
QEDAIOCB *acb = opaque; |
| 1307 |
BDRVQEDState *s = acb_to_s(acb); |
| 1308 |
QEDFindClusterFunc *io_fn = (acb->flags & QED_AIOCB_WRITE) ? |
| 1309 |
qed_aio_write_data : qed_aio_read_data; |
| 1310 |
|
| 1311 |
trace_qed_aio_next_io(s, acb, ret, acb->cur_pos + acb->cur_qiov.size); |
| 1312 |
|
| 1313 |
/* Handle I/O error */
|
| 1314 |
if (ret) {
|
| 1315 |
qed_aio_complete(acb, ret); |
| 1316 |
return;
|
| 1317 |
} |
| 1318 |
|
| 1319 |
acb->qiov_offset += acb->cur_qiov.size; |
| 1320 |
acb->cur_pos += acb->cur_qiov.size; |
| 1321 |
qemu_iovec_reset(&acb->cur_qiov); |
| 1322 |
|
| 1323 |
/* Complete request */
|
| 1324 |
if (acb->cur_pos >= acb->end_pos) {
|
| 1325 |
qed_aio_complete(acb, 0);
|
| 1326 |
return;
|
| 1327 |
} |
| 1328 |
|
| 1329 |
/* Find next cluster and start I/O */
|
| 1330 |
qed_find_cluster(s, &acb->request, |
| 1331 |
acb->cur_pos, acb->end_pos - acb->cur_pos, |
| 1332 |
io_fn, acb); |
| 1333 |
} |
| 1334 |
|
| 1335 |
static BlockDriverAIOCB *qed_aio_setup(BlockDriverState *bs,
|
| 1336 |
int64_t sector_num, |
| 1337 |
QEMUIOVector *qiov, int nb_sectors,
|
| 1338 |
BlockDriverCompletionFunc *cb, |
| 1339 |
void *opaque, int flags) |
| 1340 |
{
|
| 1341 |
QEDAIOCB *acb = qemu_aio_get(&qed_aiocb_info, bs, cb, opaque); |
| 1342 |
|
| 1343 |
trace_qed_aio_setup(bs->opaque, acb, sector_num, nb_sectors, |
| 1344 |
opaque, flags); |
| 1345 |
|
| 1346 |
acb->flags = flags; |
| 1347 |
acb->finished = NULL;
|
| 1348 |
acb->qiov = qiov; |
| 1349 |
acb->qiov_offset = 0;
|
| 1350 |
acb->cur_pos = (uint64_t)sector_num * BDRV_SECTOR_SIZE; |
| 1351 |
acb->end_pos = acb->cur_pos + nb_sectors * BDRV_SECTOR_SIZE; |
| 1352 |
acb->request.l2_table = NULL;
|
| 1353 |
qemu_iovec_init(&acb->cur_qiov, qiov->niov); |
| 1354 |
|
| 1355 |
/* Start request */
|
| 1356 |
qed_aio_next_io(acb, 0);
|
| 1357 |
return &acb->common;
|
| 1358 |
} |
| 1359 |
|
| 1360 |
static BlockDriverAIOCB *bdrv_qed_aio_readv(BlockDriverState *bs,
|
| 1361 |
int64_t sector_num, |
| 1362 |
QEMUIOVector *qiov, int nb_sectors,
|
| 1363 |
BlockDriverCompletionFunc *cb, |
| 1364 |
void *opaque)
|
| 1365 |
{
|
| 1366 |
return qed_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0); |
| 1367 |
} |
| 1368 |
|
| 1369 |
static BlockDriverAIOCB *bdrv_qed_aio_writev(BlockDriverState *bs,
|
| 1370 |
int64_t sector_num, |
| 1371 |
QEMUIOVector *qiov, int nb_sectors,
|
| 1372 |
BlockDriverCompletionFunc *cb, |
| 1373 |
void *opaque)
|
| 1374 |
{
|
| 1375 |
return qed_aio_setup(bs, sector_num, qiov, nb_sectors, cb,
|
| 1376 |
opaque, QED_AIOCB_WRITE); |
| 1377 |
} |
| 1378 |
|
| 1379 |
typedef struct { |
| 1380 |
Coroutine *co; |
| 1381 |
int ret;
|
| 1382 |
bool done;
|
| 1383 |
} QEDWriteZeroesCB; |
| 1384 |
|
| 1385 |
static void coroutine_fn qed_co_write_zeroes_cb(void *opaque, int ret) |
| 1386 |
{
|
| 1387 |
QEDWriteZeroesCB *cb = opaque; |
| 1388 |
|
| 1389 |
cb->done = true;
|
| 1390 |
cb->ret = ret; |
| 1391 |
if (cb->co) {
|
| 1392 |
qemu_coroutine_enter(cb->co, NULL);
|
| 1393 |
} |
| 1394 |
} |
| 1395 |
|
| 1396 |
static int coroutine_fn bdrv_qed_co_write_zeroes(BlockDriverState *bs, |
| 1397 |
int64_t sector_num, |
| 1398 |
int nb_sectors)
|
| 1399 |
{
|
| 1400 |
BlockDriverAIOCB *blockacb; |
| 1401 |
BDRVQEDState *s = bs->opaque; |
| 1402 |
QEDWriteZeroesCB cb = { .done = false };
|
| 1403 |
QEMUIOVector qiov; |
| 1404 |
struct iovec iov;
|
| 1405 |
|
| 1406 |
/* Refuse if there are untouched backing file sectors */
|
| 1407 |
if (bs->backing_hd) {
|
| 1408 |
if (qed_offset_into_cluster(s, sector_num * BDRV_SECTOR_SIZE) != 0) { |
| 1409 |
return -ENOTSUP;
|
| 1410 |
} |
| 1411 |
if (qed_offset_into_cluster(s, nb_sectors * BDRV_SECTOR_SIZE) != 0) { |
| 1412 |
return -ENOTSUP;
|
| 1413 |
} |
| 1414 |
} |
| 1415 |
|
| 1416 |
/* Zero writes start without an I/O buffer. If a buffer becomes necessary
|
| 1417 |
* then it will be allocated during request processing.
|
| 1418 |
*/
|
| 1419 |
iov.iov_base = NULL,
|
| 1420 |
iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE, |
| 1421 |
|
| 1422 |
qemu_iovec_init_external(&qiov, &iov, 1);
|
| 1423 |
blockacb = qed_aio_setup(bs, sector_num, &qiov, nb_sectors, |
| 1424 |
qed_co_write_zeroes_cb, &cb, |
| 1425 |
QED_AIOCB_WRITE | QED_AIOCB_ZERO); |
| 1426 |
if (!blockacb) {
|
| 1427 |
return -EIO;
|
| 1428 |
} |
| 1429 |
if (!cb.done) {
|
| 1430 |
cb.co = qemu_coroutine_self(); |
| 1431 |
qemu_coroutine_yield(); |
| 1432 |
} |
| 1433 |
assert(cb.done); |
| 1434 |
return cb.ret;
|
| 1435 |
} |
| 1436 |
|
| 1437 |
static int bdrv_qed_truncate(BlockDriverState *bs, int64_t offset) |
| 1438 |
{
|
| 1439 |
BDRVQEDState *s = bs->opaque; |
| 1440 |
uint64_t old_image_size; |
| 1441 |
int ret;
|
| 1442 |
|
| 1443 |
if (!qed_is_image_size_valid(offset, s->header.cluster_size,
|
| 1444 |
s->header.table_size)) {
|
| 1445 |
return -EINVAL;
|
| 1446 |
} |
| 1447 |
|
| 1448 |
/* Shrinking is currently not supported */
|
| 1449 |
if ((uint64_t)offset < s->header.image_size) {
|
| 1450 |
return -ENOTSUP;
|
| 1451 |
} |
| 1452 |
|
| 1453 |
old_image_size = s->header.image_size; |
| 1454 |
s->header.image_size = offset; |
| 1455 |
ret = qed_write_header_sync(s); |
| 1456 |
if (ret < 0) { |
| 1457 |
s->header.image_size = old_image_size; |
| 1458 |
} |
| 1459 |
return ret;
|
| 1460 |
} |
| 1461 |
|
| 1462 |
static int64_t bdrv_qed_getlength(BlockDriverState *bs)
|
| 1463 |
{
|
| 1464 |
BDRVQEDState *s = bs->opaque; |
| 1465 |
return s->header.image_size;
|
| 1466 |
} |
| 1467 |
|
| 1468 |
static int bdrv_qed_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
| 1469 |
{
|
| 1470 |
BDRVQEDState *s = bs->opaque; |
| 1471 |
|
| 1472 |
memset(bdi, 0, sizeof(*bdi)); |
| 1473 |
bdi->cluster_size = s->header.cluster_size; |
| 1474 |
bdi->is_dirty = s->header.features & QED_F_NEED_CHECK; |
| 1475 |
return 0; |
| 1476 |
} |
| 1477 |
|
| 1478 |
static int bdrv_qed_change_backing_file(BlockDriverState *bs, |
| 1479 |
const char *backing_file, |
| 1480 |
const char *backing_fmt) |
| 1481 |
{
|
| 1482 |
BDRVQEDState *s = bs->opaque; |
| 1483 |
QEDHeader new_header, le_header; |
| 1484 |
void *buffer;
|
| 1485 |
size_t buffer_len, backing_file_len; |
| 1486 |
int ret;
|
| 1487 |
|
| 1488 |
/* Refuse to set backing filename if unknown compat feature bits are
|
| 1489 |
* active. If the image uses an unknown compat feature then we may not
|
| 1490 |
* know the layout of data following the header structure and cannot safely
|
| 1491 |
* add a new string.
|
| 1492 |
*/
|
| 1493 |
if (backing_file && (s->header.compat_features &
|
| 1494 |
~QED_COMPAT_FEATURE_MASK)) {
|
| 1495 |
return -ENOTSUP;
|
| 1496 |
} |
| 1497 |
|
| 1498 |
memcpy(&new_header, &s->header, sizeof(new_header));
|
| 1499 |
|
| 1500 |
new_header.features &= ~(QED_F_BACKING_FILE | |
| 1501 |
QED_F_BACKING_FORMAT_NO_PROBE); |
| 1502 |
|
| 1503 |
/* Adjust feature flags */
|
| 1504 |
if (backing_file) {
|
| 1505 |
new_header.features |= QED_F_BACKING_FILE; |
| 1506 |
|
| 1507 |
if (qed_fmt_is_raw(backing_fmt)) {
|
| 1508 |
new_header.features |= QED_F_BACKING_FORMAT_NO_PROBE; |
| 1509 |
} |
| 1510 |
} |
| 1511 |
|
| 1512 |
/* Calculate new header size */
|
| 1513 |
backing_file_len = 0;
|
| 1514 |
|
| 1515 |
if (backing_file) {
|
| 1516 |
backing_file_len = strlen(backing_file); |
| 1517 |
} |
| 1518 |
|
| 1519 |
buffer_len = sizeof(new_header);
|
| 1520 |
new_header.backing_filename_offset = buffer_len; |
| 1521 |
new_header.backing_filename_size = backing_file_len; |
| 1522 |
buffer_len += backing_file_len; |
| 1523 |
|
| 1524 |
/* Make sure we can rewrite header without failing */
|
| 1525 |
if (buffer_len > new_header.header_size * new_header.cluster_size) {
|
| 1526 |
return -ENOSPC;
|
| 1527 |
} |
| 1528 |
|
| 1529 |
/* Prepare new header */
|
| 1530 |
buffer = g_malloc(buffer_len); |
| 1531 |
|
| 1532 |
qed_header_cpu_to_le(&new_header, &le_header); |
| 1533 |
memcpy(buffer, &le_header, sizeof(le_header));
|
| 1534 |
buffer_len = sizeof(le_header);
|
| 1535 |
|
| 1536 |
if (backing_file) {
|
| 1537 |
memcpy(buffer + buffer_len, backing_file, backing_file_len); |
| 1538 |
buffer_len += backing_file_len; |
| 1539 |
} |
| 1540 |
|
| 1541 |
/* Write new header */
|
| 1542 |
ret = bdrv_pwrite_sync(bs->file, 0, buffer, buffer_len);
|
| 1543 |
g_free(buffer); |
| 1544 |
if (ret == 0) { |
| 1545 |
memcpy(&s->header, &new_header, sizeof(new_header));
|
| 1546 |
} |
| 1547 |
return ret;
|
| 1548 |
} |
| 1549 |
|
| 1550 |
static void bdrv_qed_invalidate_cache(BlockDriverState *bs) |
| 1551 |
{
|
| 1552 |
BDRVQEDState *s = bs->opaque; |
| 1553 |
|
| 1554 |
bdrv_qed_close(bs); |
| 1555 |
memset(s, 0, sizeof(BDRVQEDState)); |
| 1556 |
bdrv_qed_open(bs, NULL, bs->open_flags, NULL); |
| 1557 |
} |
| 1558 |
|
| 1559 |
static int bdrv_qed_check(BlockDriverState *bs, BdrvCheckResult *result, |
| 1560 |
BdrvCheckMode fix) |
| 1561 |
{
|
| 1562 |
BDRVQEDState *s = bs->opaque; |
| 1563 |
|
| 1564 |
return qed_check(s, result, !!fix);
|
| 1565 |
} |
| 1566 |
|
| 1567 |
static QEMUOptionParameter qed_create_options[] = {
|
| 1568 |
{
|
| 1569 |
.name = BLOCK_OPT_SIZE, |
| 1570 |
.type = OPT_SIZE, |
| 1571 |
.help = "Virtual disk size (in bytes)"
|
| 1572 |
}, {
|
| 1573 |
.name = BLOCK_OPT_BACKING_FILE, |
| 1574 |
.type = OPT_STRING, |
| 1575 |
.help = "File name of a base image"
|
| 1576 |
}, {
|
| 1577 |
.name = BLOCK_OPT_BACKING_FMT, |
| 1578 |
.type = OPT_STRING, |
| 1579 |
.help = "Image format of the base image"
|
| 1580 |
}, {
|
| 1581 |
.name = BLOCK_OPT_CLUSTER_SIZE, |
| 1582 |
.type = OPT_SIZE, |
| 1583 |
.help = "Cluster size (in bytes)",
|
| 1584 |
.value = { .n = QED_DEFAULT_CLUSTER_SIZE },
|
| 1585 |
}, {
|
| 1586 |
.name = BLOCK_OPT_TABLE_SIZE, |
| 1587 |
.type = OPT_SIZE, |
| 1588 |
.help = "L1/L2 table size (in clusters)"
|
| 1589 |
}, |
| 1590 |
{ /* end of list */ }
|
| 1591 |
}; |
| 1592 |
|
| 1593 |
static BlockDriver bdrv_qed = {
|
| 1594 |
.format_name = "qed",
|
| 1595 |
.instance_size = sizeof(BDRVQEDState),
|
| 1596 |
.create_options = qed_create_options, |
| 1597 |
|
| 1598 |
.bdrv_probe = bdrv_qed_probe, |
| 1599 |
.bdrv_rebind = bdrv_qed_rebind, |
| 1600 |
.bdrv_open = bdrv_qed_open, |
| 1601 |
.bdrv_close = bdrv_qed_close, |
| 1602 |
.bdrv_reopen_prepare = bdrv_qed_reopen_prepare, |
| 1603 |
.bdrv_create = bdrv_qed_create, |
| 1604 |
.bdrv_has_zero_init = bdrv_has_zero_init_1, |
| 1605 |
.bdrv_co_get_block_status = bdrv_qed_co_get_block_status, |
| 1606 |
.bdrv_make_empty = bdrv_qed_make_empty, |
| 1607 |
.bdrv_aio_readv = bdrv_qed_aio_readv, |
| 1608 |
.bdrv_aio_writev = bdrv_qed_aio_writev, |
| 1609 |
.bdrv_co_write_zeroes = bdrv_qed_co_write_zeroes, |
| 1610 |
.bdrv_truncate = bdrv_qed_truncate, |
| 1611 |
.bdrv_getlength = bdrv_qed_getlength, |
| 1612 |
.bdrv_get_info = bdrv_qed_get_info, |
| 1613 |
.bdrv_change_backing_file = bdrv_qed_change_backing_file, |
| 1614 |
.bdrv_invalidate_cache = bdrv_qed_invalidate_cache, |
| 1615 |
.bdrv_check = bdrv_qed_check, |
| 1616 |
}; |
| 1617 |
|
| 1618 |
static void bdrv_qed_init(void) |
| 1619 |
{
|
| 1620 |
bdrv_register(&bdrv_qed); |
| 1621 |
} |
| 1622 |
|
| 1623 |
block_init(bdrv_qed_init); |