xref: /qemu/block/qcow2-cache.c (revision 93913dfd)
1 /*
2  * L2/refcount table cache for the QCOW2 format
3  *
4  * Copyright (c) 2010 Kevin Wolf <kwolf@redhat.com>
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 
25 #include "block_int.h"
26 #include "qemu-common.h"
27 #include "qcow2.h"
28 
29 typedef struct Qcow2CachedTable {
30     void*   table;
31     int64_t offset;
32     bool    dirty;
33     int     cache_hits;
34     int     ref;
35 } Qcow2CachedTable;
36 
37 struct Qcow2Cache {
38     Qcow2CachedTable*       entries;
39     struct Qcow2Cache*      depends;
40     int                     size;
41     bool                    depends_on_flush;
42     bool                    writethrough;
43 };
44 
45 Qcow2Cache *qcow2_cache_create(BlockDriverState *bs, int num_tables,
46     bool writethrough)
47 {
48     BDRVQcowState *s = bs->opaque;
49     Qcow2Cache *c;
50     int i;
51 
52     c = qemu_mallocz(sizeof(*c));
53     c->size = num_tables;
54     c->entries = qemu_mallocz(sizeof(*c->entries) * num_tables);
55     c->writethrough = writethrough;
56 
57     for (i = 0; i < c->size; i++) {
58         c->entries[i].table = qemu_blockalign(bs, s->cluster_size);
59     }
60 
61     return c;
62 }
63 
64 int qcow2_cache_destroy(BlockDriverState* bs, Qcow2Cache *c)
65 {
66     int i;
67 
68     for (i = 0; i < c->size; i++) {
69         assert(c->entries[i].ref == 0);
70         qemu_vfree(c->entries[i].table);
71     }
72 
73     qemu_free(c->entries);
74     qemu_free(c);
75 
76     return 0;
77 }
78 
79 static int qcow2_cache_flush_dependency(BlockDriverState *bs, Qcow2Cache *c)
80 {
81     int ret;
82 
83     ret = qcow2_cache_flush(bs, c->depends);
84     if (ret < 0) {
85         return ret;
86     }
87 
88     c->depends = NULL;
89     c->depends_on_flush = false;
90 
91     return 0;
92 }
93 
94 static int qcow2_cache_entry_flush(BlockDriverState *bs, Qcow2Cache *c, int i)
95 {
96     BDRVQcowState *s = bs->opaque;
97     int ret = 0;
98 
99     if (!c->entries[i].dirty || !c->entries[i].offset) {
100         return 0;
101     }
102 
103     if (c->depends) {
104         ret = qcow2_cache_flush_dependency(bs, c);
105     } else if (c->depends_on_flush) {
106         ret = bdrv_flush(bs->file);
107         if (ret >= 0) {
108             c->depends_on_flush = false;
109         }
110     }
111 
112     if (ret < 0) {
113         return ret;
114     }
115 
116     if (c == s->refcount_block_cache) {
117         BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_UPDATE_PART);
118     } else if (c == s->l2_table_cache) {
119         BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
120     }
121 
122     ret = bdrv_pwrite(bs->file, c->entries[i].offset, c->entries[i].table,
123         s->cluster_size);
124     if (ret < 0) {
125         return ret;
126     }
127 
128     c->entries[i].dirty = false;
129 
130     return 0;
131 }
132 
133 int qcow2_cache_flush(BlockDriverState *bs, Qcow2Cache *c)
134 {
135     int result = 0;
136     int ret;
137     int i;
138 
139     for (i = 0; i < c->size; i++) {
140         ret = qcow2_cache_entry_flush(bs, c, i);
141         if (ret < 0 && result != -ENOSPC) {
142             result = ret;
143         }
144     }
145 
146     if (result == 0) {
147         ret = bdrv_flush(bs->file);
148         if (ret < 0) {
149             result = ret;
150         }
151     }
152 
153     return result;
154 }
155 
156 int qcow2_cache_set_dependency(BlockDriverState *bs, Qcow2Cache *c,
157     Qcow2Cache *dependency)
158 {
159     int ret;
160 
161     if (dependency->depends) {
162         ret = qcow2_cache_flush_dependency(bs, dependency);
163         if (ret < 0) {
164             return ret;
165         }
166     }
167 
168     if (c->depends && (c->depends != dependency)) {
169         ret = qcow2_cache_flush_dependency(bs, c);
170         if (ret < 0) {
171             return ret;
172         }
173     }
174 
175     c->depends = dependency;
176     return 0;
177 }
178 
179 void qcow2_cache_depends_on_flush(Qcow2Cache *c)
180 {
181     c->depends_on_flush = true;
182 }
183 
184 static int qcow2_cache_find_entry_to_replace(Qcow2Cache *c)
185 {
186     int i;
187     int min_count = INT_MAX;
188     int min_index = -1;
189 
190 
191     for (i = 0; i < c->size; i++) {
192         if (c->entries[i].ref) {
193             continue;
194         }
195 
196         if (c->entries[i].cache_hits < min_count) {
197             min_index = i;
198             min_count = c->entries[i].cache_hits;
199         }
200 
201         /* Give newer hits priority */
202         /* TODO Check how to optimize the replacement strategy */
203         c->entries[i].cache_hits /= 2;
204     }
205 
206     if (min_index == -1) {
207         /* This can't happen in current synchronous code, but leave the check
208          * here as a reminder for whoever starts using AIO with the cache */
209         abort();
210     }
211     return min_index;
212 }
213 
214 static int qcow2_cache_do_get(BlockDriverState *bs, Qcow2Cache *c,
215     uint64_t offset, void **table, bool read_from_disk)
216 {
217     BDRVQcowState *s = bs->opaque;
218     int i;
219     int ret;
220 
221     /* Check if the table is already cached */
222     for (i = 0; i < c->size; i++) {
223         if (c->entries[i].offset == offset) {
224             goto found;
225         }
226     }
227 
228     /* If not, write a table back and replace it */
229     i = qcow2_cache_find_entry_to_replace(c);
230     if (i < 0) {
231         return i;
232     }
233 
234     ret = qcow2_cache_entry_flush(bs, c, i);
235     if (ret < 0) {
236         return ret;
237     }
238 
239     c->entries[i].offset = 0;
240     if (read_from_disk) {
241         if (c == s->l2_table_cache) {
242             BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
243         }
244 
245         ret = bdrv_pread(bs->file, offset, c->entries[i].table, s->cluster_size);
246         if (ret < 0) {
247             return ret;
248         }
249     }
250 
251     /* Give the table some hits for the start so that it won't be replaced
252      * immediately. The number 32 is completely arbitrary. */
253     c->entries[i].cache_hits = 32;
254     c->entries[i].offset = offset;
255 
256     /* And return the right table */
257 found:
258     c->entries[i].cache_hits++;
259     c->entries[i].ref++;
260     *table = c->entries[i].table;
261     return 0;
262 }
263 
264 int qcow2_cache_get(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
265     void **table)
266 {
267     return qcow2_cache_do_get(bs, c, offset, table, true);
268 }
269 
270 int qcow2_cache_get_empty(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
271     void **table)
272 {
273     return qcow2_cache_do_get(bs, c, offset, table, false);
274 }
275 
276 int qcow2_cache_put(BlockDriverState *bs, Qcow2Cache *c, void **table)
277 {
278     int i;
279 
280     for (i = 0; i < c->size; i++) {
281         if (c->entries[i].table == *table) {
282             goto found;
283         }
284     }
285     return -ENOENT;
286 
287 found:
288     c->entries[i].ref--;
289     *table = NULL;
290 
291     assert(c->entries[i].ref >= 0);
292 
293     if (c->writethrough) {
294         return qcow2_cache_entry_flush(bs, c, i);
295     } else {
296         return 0;
297     }
298 }
299 
300 void qcow2_cache_entry_mark_dirty(Qcow2Cache *c, void *table)
301 {
302     int i;
303 
304     for (i = 0; i < c->size; i++) {
305         if (c->entries[i].table == table) {
306             goto found;
307         }
308     }
309     abort();
310 
311 found:
312     c->entries[i].dirty = true;
313 }
314 
315 bool qcow2_cache_set_writethrough(BlockDriverState *bs, Qcow2Cache *c,
316     bool enable)
317 {
318     bool old = c->writethrough;
319 
320     if (!old && enable) {
321         qcow2_cache_flush(bs, c);
322     }
323 
324     c->writethrough = enable;
325     return old;
326 }
327