root / block / qed-table.c @ 0e326109
History | View | Annotate | Download (8.9 kB)
1 |
/*
|
---|---|
2 |
* QEMU Enhanced Disk Format Table I/O
|
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 "trace.h" |
16 |
#include "qemu_socket.h" /* for EINPROGRESS on Windows */ |
17 |
#include "qed.h" |
18 |
|
19 |
typedef struct { |
20 |
GenericCB gencb; |
21 |
BDRVQEDState *s; |
22 |
QEDTable *table; |
23 |
|
24 |
struct iovec iov;
|
25 |
QEMUIOVector qiov; |
26 |
} QEDReadTableCB; |
27 |
|
28 |
static void qed_read_table_cb(void *opaque, int ret) |
29 |
{ |
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); |
33 |
int i;
|
34 |
|
35 |
/* Handle I/O error */
|
36 |
if (ret) {
|
37 |
goto out;
|
38 |
} |
39 |
|
40 |
/* Byteswap offsets */
|
41 |
for (i = 0; i < noffsets; i++) { |
42 |
table->offsets[i] = le64_to_cpu(table->offsets[i]); |
43 |
} |
44 |
|
45 |
out:
|
46 |
/* Completion */
|
47 |
trace_qed_read_table_cb(read_table_cb->s, read_table_cb->table, ret); |
48 |
gencb_complete(&read_table_cb->gencb, ret); |
49 |
} |
50 |
|
51 |
static void qed_read_table(BDRVQEDState *s, uint64_t offset, QEDTable *table, |
52 |
BlockDriverCompletionFunc *cb, void *opaque)
|
53 |
{ |
54 |
QEDReadTableCB *read_table_cb = gencb_alloc(sizeof(*read_table_cb),
|
55 |
cb, opaque); |
56 |
QEMUIOVector *qiov = &read_table_cb->qiov; |
57 |
BlockDriverAIOCB *aiocb; |
58 |
|
59 |
trace_qed_read_table(s, offset, table); |
60 |
|
61 |
read_table_cb->s = s; |
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, |
65 |
|
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); |
70 |
if (!aiocb) {
|
71 |
qed_read_table_cb(read_table_cb, -EIO); |
72 |
} |
73 |
} |
74 |
|
75 |
typedef struct { |
76 |
GenericCB gencb; |
77 |
BDRVQEDState *s; |
78 |
QEDTable *orig_table; |
79 |
QEDTable *table; |
80 |
bool flush; /* flush after write? */ |
81 |
|
82 |
struct iovec iov;
|
83 |
QEMUIOVector qiov; |
84 |
} QEDWriteTableCB; |
85 |
|
86 |
static void qed_write_table_cb(void *opaque, int ret) |
87 |
{ |
88 |
QEDWriteTableCB *write_table_cb = opaque; |
89 |
|
90 |
trace_qed_write_table_cb(write_table_cb->s, |
91 |
write_table_cb->orig_table, |
92 |
write_table_cb->flush, |
93 |
ret); |
94 |
|
95 |
if (ret) {
|
96 |
goto out;
|
97 |
} |
98 |
|
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, |
103 |
write_table_cb); |
104 |
return;
|
105 |
} |
106 |
|
107 |
out:
|
108 |
qemu_vfree(write_table_cb->table); |
109 |
gencb_complete(&write_table_cb->gencb, ret); |
110 |
return;
|
111 |
} |
112 |
|
113 |
/**
|
114 |
* Write out an updated part or all of a table
|
115 |
*
|
116 |
* @s: QED state
|
117 |
* @offset: Offset of table in image file, in bytes
|
118 |
* @table: Table
|
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
|
124 |
*/
|
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)
|
128 |
{ |
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; |
133 |
size_t len_bytes; |
134 |
|
135 |
trace_qed_write_table(s, offset, table, index, n); |
136 |
|
137 |
/* Calculate indices of the first and one after last elements */
|
138 |
start = index & ~sector_mask; |
139 |
end = (index + n + sector_mask) & ~sector_mask; |
140 |
|
141 |
len_bytes = (end - start) * sizeof(uint64_t);
|
142 |
|
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);
|
151 |
|
152 |
/* Byteswap table */
|
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; |
156 |
} |
157 |
|
158 |
/* Adjust for offset into table */
|
159 |
offset += start * sizeof(uint64_t);
|
160 |
|
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); |
165 |
if (!aiocb) {
|
166 |
qed_write_table_cb(write_table_cb, -EIO); |
167 |
} |
168 |
} |
169 |
|
170 |
/**
|
171 |
* Propagate return value from async callback
|
172 |
*/
|
173 |
static void qed_sync_cb(void *opaque, int ret) |
174 |
{ |
175 |
*(int *)opaque = ret;
|
176 |
} |
177 |
|
178 |
int qed_read_l1_table_sync(BDRVQEDState *s)
|
179 |
{ |
180 |
int ret = -EINPROGRESS;
|
181 |
|
182 |
async_context_push(); |
183 |
|
184 |
qed_read_table(s, s->header.l1_table_offset, |
185 |
s->l1_table, qed_sync_cb, &ret); |
186 |
while (ret == -EINPROGRESS) {
|
187 |
qemu_aio_wait(); |
188 |
} |
189 |
|
190 |
async_context_pop(); |
191 |
|
192 |
return ret;
|
193 |
} |
194 |
|
195 |
void qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n, |
196 |
BlockDriverCompletionFunc *cb, void *opaque)
|
197 |
{ |
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);
|
201 |
} |
202 |
|
203 |
int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index, |
204 |
unsigned int n) |
205 |
{ |
206 |
int ret = -EINPROGRESS;
|
207 |
|
208 |
async_context_push(); |
209 |
|
210 |
qed_write_l1_table(s, index, n, qed_sync_cb, &ret); |
211 |
while (ret == -EINPROGRESS) {
|
212 |
qemu_aio_wait(); |
213 |
} |
214 |
|
215 |
async_context_pop(); |
216 |
|
217 |
return ret;
|
218 |
} |
219 |
|
220 |
typedef struct { |
221 |
GenericCB gencb; |
222 |
BDRVQEDState *s; |
223 |
uint64_t l2_offset; |
224 |
QEDRequest *request; |
225 |
} QEDReadL2TableCB; |
226 |
|
227 |
static void qed_read_l2_table_cb(void *opaque, int ret) |
228 |
{ |
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; |
233 |
|
234 |
if (ret) {
|
235 |
/* can't trust loaded L2 table anymore */
|
236 |
qed_unref_l2_cache_entry(l2_table); |
237 |
request->l2_table = NULL;
|
238 |
} else {
|
239 |
l2_table->offset = read_l2_table_cb->l2_offset; |
240 |
|
241 |
qed_commit_l2_cache_entry(&s->l2_cache, l2_table); |
242 |
|
243 |
/* This is guaranteed to succeed because we just committed the entry
|
244 |
* to the cache.
|
245 |
*/
|
246 |
request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, |
247 |
l2_table->offset); |
248 |
assert(request->l2_table != NULL);
|
249 |
} |
250 |
|
251 |
gencb_complete(&read_l2_table_cb->gencb, ret); |
252 |
} |
253 |
|
254 |
void qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset,
|
255 |
BlockDriverCompletionFunc *cb, void *opaque)
|
256 |
{ |
257 |
QEDReadL2TableCB *read_l2_table_cb; |
258 |
|
259 |
qed_unref_l2_cache_entry(request->l2_table); |
260 |
|
261 |
/* Check for cached L2 entry */
|
262 |
request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset); |
263 |
if (request->l2_table) {
|
264 |
cb(opaque, 0);
|
265 |
return;
|
266 |
} |
267 |
|
268 |
request->l2_table = qed_alloc_l2_cache_entry(&s->l2_cache); |
269 |
request->l2_table->table = qed_alloc_table(s); |
270 |
|
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; |
275 |
|
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); |
279 |
} |
280 |
|
281 |
int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request, uint64_t offset)
|
282 |
{ |
283 |
int ret = -EINPROGRESS;
|
284 |
|
285 |
async_context_push(); |
286 |
|
287 |
qed_read_l2_table(s, request, offset, qed_sync_cb, &ret); |
288 |
while (ret == -EINPROGRESS) {
|
289 |
qemu_aio_wait(); |
290 |
} |
291 |
|
292 |
async_context_pop(); |
293 |
return ret;
|
294 |
} |
295 |
|
296 |
void qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
|
297 |
unsigned int index, unsigned int n, bool flush, |
298 |
BlockDriverCompletionFunc *cb, void *opaque)
|
299 |
{ |
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); |
303 |
} |
304 |
|
305 |
int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
|
306 |
unsigned int index, unsigned int n, bool flush) |
307 |
{ |
308 |
int ret = -EINPROGRESS;
|
309 |
|
310 |
async_context_push(); |
311 |
|
312 |
qed_write_l2_table(s, request, index, n, flush, qed_sync_cb, &ret); |
313 |
while (ret == -EINPROGRESS) {
|
314 |
qemu_aio_wait(); |
315 |
} |
316 |
|
317 |
async_context_pop(); |
318 |
return ret;
|
319 |
} |