2 * L2/refcount table cache for the QCOW2 format
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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
25 #include "block_int.h"
26 #include "qemu-common.h"
30 typedef struct Qcow2CachedTable {
39 Qcow2CachedTable* entries;
40 struct Qcow2Cache* depends;
42 bool depends_on_flush;
46 Qcow2Cache *qcow2_cache_create(BlockDriverState *bs, int num_tables,
49 BDRVQcowState *s = bs->opaque;
53 c = g_malloc0(sizeof(*c));
55 c->entries = g_malloc0(sizeof(*c->entries) * num_tables);
56 c->writethrough = writethrough;
58 for (i = 0; i < c->size; i++) {
59 c->entries[i].table = qemu_blockalign(bs, s->cluster_size);
65 int qcow2_cache_destroy(BlockDriverState* bs, Qcow2Cache *c)
69 for (i = 0; i < c->size; i++) {
70 assert(c->entries[i].ref == 0);
71 qemu_vfree(c->entries[i].table);
80 static int qcow2_cache_flush_dependency(BlockDriverState *bs, Qcow2Cache *c)
84 ret = qcow2_cache_flush(bs, c->depends);
90 c->depends_on_flush = false;
95 static int qcow2_cache_entry_flush(BlockDriverState *bs, Qcow2Cache *c, int i)
97 BDRVQcowState *s = bs->opaque;
100 if (!c->entries[i].dirty || !c->entries[i].offset) {
104 trace_qcow2_cache_entry_flush(qemu_coroutine_self(),
105 c == s->l2_table_cache, i);
108 ret = qcow2_cache_flush_dependency(bs, c);
109 } else if (c->depends_on_flush) {
110 ret = bdrv_flush(bs->file);
112 c->depends_on_flush = false;
120 if (c == s->refcount_block_cache) {
121 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_UPDATE_PART);
122 } else if (c == s->l2_table_cache) {
123 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
126 ret = bdrv_pwrite(bs->file, c->entries[i].offset, c->entries[i].table,
132 c->entries[i].dirty = false;
137 int qcow2_cache_flush(BlockDriverState *bs, Qcow2Cache *c)
139 BDRVQcowState *s = bs->opaque;
144 trace_qcow2_cache_flush(qemu_coroutine_self(), c == s->l2_table_cache);
146 for (i = 0; i < c->size; i++) {
147 ret = qcow2_cache_entry_flush(bs, c, i);
148 if (ret < 0 && result != -ENOSPC) {
154 ret = bdrv_flush(bs->file);
163 int qcow2_cache_set_dependency(BlockDriverState *bs, Qcow2Cache *c,
164 Qcow2Cache *dependency)
168 if (dependency->depends) {
169 ret = qcow2_cache_flush_dependency(bs, dependency);
175 if (c->depends && (c->depends != dependency)) {
176 ret = qcow2_cache_flush_dependency(bs, c);
182 c->depends = dependency;
186 void qcow2_cache_depends_on_flush(Qcow2Cache *c)
188 c->depends_on_flush = true;
191 static int qcow2_cache_find_entry_to_replace(Qcow2Cache *c)
194 int min_count = INT_MAX;
198 for (i = 0; i < c->size; i++) {
199 if (c->entries[i].ref) {
203 if (c->entries[i].cache_hits < min_count) {
205 min_count = c->entries[i].cache_hits;
208 /* Give newer hits priority */
209 /* TODO Check how to optimize the replacement strategy */
210 c->entries[i].cache_hits /= 2;
213 if (min_index == -1) {
214 /* This can't happen in current synchronous code, but leave the check
215 * here as a reminder for whoever starts using AIO with the cache */
221 static int qcow2_cache_do_get(BlockDriverState *bs, Qcow2Cache *c,
222 uint64_t offset, void **table, bool read_from_disk)
224 BDRVQcowState *s = bs->opaque;
228 trace_qcow2_cache_get(qemu_coroutine_self(), c == s->l2_table_cache,
229 offset, read_from_disk);
231 /* Check if the table is already cached */
232 for (i = 0; i < c->size; i++) {
233 if (c->entries[i].offset == offset) {
238 /* If not, write a table back and replace it */
239 i = qcow2_cache_find_entry_to_replace(c);
240 trace_qcow2_cache_get_replace_entry(qemu_coroutine_self(),
241 c == s->l2_table_cache, i);
246 ret = qcow2_cache_entry_flush(bs, c, i);
251 trace_qcow2_cache_get_read(qemu_coroutine_self(),
252 c == s->l2_table_cache, i);
253 c->entries[i].offset = 0;
254 if (read_from_disk) {
255 if (c == s->l2_table_cache) {
256 BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
259 ret = bdrv_pread(bs->file, offset, c->entries[i].table, s->cluster_size);
265 /* Give the table some hits for the start so that it won't be replaced
266 * immediately. The number 32 is completely arbitrary. */
267 c->entries[i].cache_hits = 32;
268 c->entries[i].offset = offset;
270 /* And return the right table */
272 c->entries[i].cache_hits++;
274 *table = c->entries[i].table;
276 trace_qcow2_cache_get_done(qemu_coroutine_self(),
277 c == s->l2_table_cache, i);
282 int qcow2_cache_get(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
285 return qcow2_cache_do_get(bs, c, offset, table, true);
288 int qcow2_cache_get_empty(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
291 return qcow2_cache_do_get(bs, c, offset, table, false);
294 int qcow2_cache_put(BlockDriverState *bs, Qcow2Cache *c, void **table)
298 for (i = 0; i < c->size; i++) {
299 if (c->entries[i].table == *table) {
309 assert(c->entries[i].ref >= 0);
311 if (c->writethrough) {
312 return qcow2_cache_entry_flush(bs, c, i);
318 void qcow2_cache_entry_mark_dirty(Qcow2Cache *c, void *table)
322 for (i = 0; i < c->size; i++) {
323 if (c->entries[i].table == table) {
330 c->entries[i].dirty = true;
333 bool qcow2_cache_set_writethrough(BlockDriverState *bs, Qcow2Cache *c,
336 bool old = c->writethrough;
338 if (!old && enable) {
339 qcow2_cache_flush(bs, c);
342 c->writethrough = enable;