2
* QEMU Enhanced Disk Format Table I/O
4
* Copyright IBM, Corp. 2010
7
* Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
8
* Anthony Liguori <aliguori@us.ibm.com>
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.
16
#include "qemu_socket.h" /* for EINPROGRESS on Windows */
28
static void qed_read_table_cb(void *opaque, int ret)
30
QEDReadTableCB *read_table_cb = opaque;
31
QEDTable *table = read_table_cb->table;
32
int noffsets = read_table_cb->iov.iov_len / sizeof(uint64_t);
35
/* Handle I/O error */
40
/* Byteswap offsets */
41
for (i = 0; i < noffsets; i++) {
42
table->offsets[i] = le64_to_cpu(table->offsets[i]);
47
trace_qed_read_table_cb(read_table_cb->s, read_table_cb->table, ret);
48
gencb_complete(&read_table_cb->gencb, ret);
51
static void qed_read_table(BDRVQEDState *s, uint64_t offset, QEDTable *table,
52
BlockDriverCompletionFunc *cb, void *opaque)
54
QEDReadTableCB *read_table_cb = gencb_alloc(sizeof(*read_table_cb),
56
QEMUIOVector *qiov = &read_table_cb->qiov;
57
BlockDriverAIOCB *aiocb;
59
trace_qed_read_table(s, offset, table);
62
read_table_cb->table = table;
63
read_table_cb->iov.iov_base = table->offsets,
64
read_table_cb->iov.iov_len = s->header.cluster_size * s->header.table_size,
66
qemu_iovec_init_external(qiov, &read_table_cb->iov, 1);
67
aiocb = bdrv_aio_readv(s->bs->file, offset / BDRV_SECTOR_SIZE, qiov,
68
read_table_cb->iov.iov_len / BDRV_SECTOR_SIZE,
69
qed_read_table_cb, read_table_cb);
71
qed_read_table_cb(read_table_cb, -EIO);
80
bool flush; /* flush after write? */
86
static void qed_write_table_cb(void *opaque, int ret)
88
QEDWriteTableCB *write_table_cb = opaque;
90
trace_qed_write_table_cb(write_table_cb->s,
91
write_table_cb->orig_table,
92
write_table_cb->flush,
99
if (write_table_cb->flush) {
100
/* We still need to flush first */
101
write_table_cb->flush = false;
102
bdrv_aio_flush(write_table_cb->s->bs, qed_write_table_cb,
108
qemu_vfree(write_table_cb->table);
109
gencb_complete(&write_table_cb->gencb, ret);
114
* Write out an updated part or all of a table
117
* @offset: Offset of table in image file, in bytes
119
* @index: Index of first element
120
* @n: Number of elements
121
* @flush: Whether or not to sync to disk
122
* @cb: Completion function
123
* @opaque: Argument for completion function
125
static void qed_write_table(BDRVQEDState *s, uint64_t offset, QEDTable *table,
126
unsigned int index, unsigned int n, bool flush,
127
BlockDriverCompletionFunc *cb, void *opaque)
129
QEDWriteTableCB *write_table_cb;
130
BlockDriverAIOCB *aiocb;
131
unsigned int sector_mask = BDRV_SECTOR_SIZE / sizeof(uint64_t) - 1;
132
unsigned int start, end, i;
135
trace_qed_write_table(s, offset, table, index, n);
137
/* Calculate indices of the first and one after last elements */
138
start = index & ~sector_mask;
139
end = (index + n + sector_mask) & ~sector_mask;
141
len_bytes = (end - start) * sizeof(uint64_t);
143
write_table_cb = gencb_alloc(sizeof(*write_table_cb), cb, opaque);
144
write_table_cb->s = s;
145
write_table_cb->orig_table = table;
146
write_table_cb->flush = flush;
147
write_table_cb->table = qemu_blockalign(s->bs, len_bytes);
148
write_table_cb->iov.iov_base = write_table_cb->table->offsets;
149
write_table_cb->iov.iov_len = len_bytes;
150
qemu_iovec_init_external(&write_table_cb->qiov, &write_table_cb->iov, 1);
153
for (i = start; i < end; i++) {
154
uint64_t le_offset = cpu_to_le64(table->offsets[i]);
155
write_table_cb->table->offsets[i - start] = le_offset;
158
/* Adjust for offset into table */
159
offset += start * sizeof(uint64_t);
161
aiocb = bdrv_aio_writev(s->bs->file, offset / BDRV_SECTOR_SIZE,
162
&write_table_cb->qiov,
163
write_table_cb->iov.iov_len / BDRV_SECTOR_SIZE,
164
qed_write_table_cb, write_table_cb);
166
qed_write_table_cb(write_table_cb, -EIO);
171
* Propagate return value from async callback
173
static void qed_sync_cb(void *opaque, int ret)
175
*(int *)opaque = ret;
178
int qed_read_l1_table_sync(BDRVQEDState *s)
180
int ret = -EINPROGRESS;
182
async_context_push();
184
qed_read_table(s, s->header.l1_table_offset,
185
s->l1_table, qed_sync_cb, &ret);
186
while (ret == -EINPROGRESS) {
195
void qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n,
196
BlockDriverCompletionFunc *cb, void *opaque)
198
BLKDBG_EVENT(s->bs->file, BLKDBG_L1_UPDATE);
199
qed_write_table(s, s->header.l1_table_offset,
200
s->l1_table, index, n, false, cb, opaque);
203
int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
206
int ret = -EINPROGRESS;
208
async_context_push();
210
qed_write_l1_table(s, index, n, qed_sync_cb, &ret);
211
while (ret == -EINPROGRESS) {
227
static void qed_read_l2_table_cb(void *opaque, int ret)
229
QEDReadL2TableCB *read_l2_table_cb = opaque;
230
QEDRequest *request = read_l2_table_cb->request;
231
BDRVQEDState *s = read_l2_table_cb->s;
232
CachedL2Table *l2_table = request->l2_table;
235
/* can't trust loaded L2 table anymore */
236
qed_unref_l2_cache_entry(l2_table);
237
request->l2_table = NULL;
239
l2_table->offset = read_l2_table_cb->l2_offset;
241
qed_commit_l2_cache_entry(&s->l2_cache, l2_table);
243
/* This is guaranteed to succeed because we just committed the entry
246
request->l2_table = qed_find_l2_cache_entry(&s->l2_cache,
248
assert(request->l2_table != NULL);
251
gencb_complete(&read_l2_table_cb->gencb, ret);
254
void qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset,
255
BlockDriverCompletionFunc *cb, void *opaque)
257
QEDReadL2TableCB *read_l2_table_cb;
259
qed_unref_l2_cache_entry(request->l2_table);
261
/* Check for cached L2 entry */
262
request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset);
263
if (request->l2_table) {
268
request->l2_table = qed_alloc_l2_cache_entry(&s->l2_cache);
269
request->l2_table->table = qed_alloc_table(s);
271
read_l2_table_cb = gencb_alloc(sizeof(*read_l2_table_cb), cb, opaque);
272
read_l2_table_cb->s = s;
273
read_l2_table_cb->l2_offset = offset;
274
read_l2_table_cb->request = request;
276
BLKDBG_EVENT(s->bs->file, BLKDBG_L2_LOAD);
277
qed_read_table(s, offset, request->l2_table->table,
278
qed_read_l2_table_cb, read_l2_table_cb);
281
int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request, uint64_t offset)
283
int ret = -EINPROGRESS;
285
async_context_push();
287
qed_read_l2_table(s, request, offset, qed_sync_cb, &ret);
288
while (ret == -EINPROGRESS) {
296
void qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
297
unsigned int index, unsigned int n, bool flush,
298
BlockDriverCompletionFunc *cb, void *opaque)
300
BLKDBG_EVENT(s->bs->file, BLKDBG_L2_UPDATE);
301
qed_write_table(s, request->l2_table->offset,
302
request->l2_table->table, index, n, flush, cb, opaque);
305
int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
306
unsigned int index, unsigned int n, bool flush)
308
int ret = -EINPROGRESS;
310
async_context_push();
312
qed_write_l2_table(s, request, index, n, flush, qed_sync_cb, &ret);
313
while (ret == -EINPROGRESS) {