xref: /openbmc/qemu/block/qcow2-cache.c (revision c39f95dc)
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 "qemu/osdep.h"
26 #include "block/block_int.h"
27 #include "qemu-common.h"
28 #include "qcow2.h"
29 #include "trace.h"
30 
31 typedef struct Qcow2CachedTable {
32     int64_t  offset;
33     uint64_t lru_counter;
34     int      ref;
35     bool     dirty;
36 } Qcow2CachedTable;
37 
38 struct Qcow2Cache {
39     Qcow2CachedTable       *entries;
40     struct Qcow2Cache      *depends;
41     int                     size;
42     bool                    depends_on_flush;
43     void                   *table_array;
44     uint64_t                lru_counter;
45     uint64_t                cache_clean_lru_counter;
46 };
47 
48 static inline void *qcow2_cache_get_table_addr(BlockDriverState *bs,
49                     Qcow2Cache *c, int table)
50 {
51     BDRVQcow2State *s = bs->opaque;
52     return (uint8_t *) c->table_array + (size_t) table * s->cluster_size;
53 }
54 
55 static inline int qcow2_cache_get_table_idx(BlockDriverState *bs,
56                   Qcow2Cache *c, void *table)
57 {
58     BDRVQcow2State *s = bs->opaque;
59     ptrdiff_t table_offset = (uint8_t *) table - (uint8_t *) c->table_array;
60     int idx = table_offset / s->cluster_size;
61     assert(idx >= 0 && idx < c->size && table_offset % s->cluster_size == 0);
62     return idx;
63 }
64 
65 static void qcow2_cache_table_release(BlockDriverState *bs, Qcow2Cache *c,
66                                       int i, int num_tables)
67 {
68 /* Using MADV_DONTNEED to discard memory is a Linux-specific feature */
69 #ifdef CONFIG_LINUX
70     BDRVQcow2State *s = bs->opaque;
71     void *t = qcow2_cache_get_table_addr(bs, c, i);
72     int align = getpagesize();
73     size_t mem_size = (size_t) s->cluster_size * num_tables;
74     size_t offset = QEMU_ALIGN_UP((uintptr_t) t, align) - (uintptr_t) t;
75     size_t length = QEMU_ALIGN_DOWN(mem_size - offset, align);
76     if (length > 0) {
77         madvise((uint8_t *) t + offset, length, MADV_DONTNEED);
78     }
79 #endif
80 }
81 
82 static inline bool can_clean_entry(Qcow2Cache *c, int i)
83 {
84     Qcow2CachedTable *t = &c->entries[i];
85     return t->ref == 0 && !t->dirty && t->offset != 0 &&
86         t->lru_counter <= c->cache_clean_lru_counter;
87 }
88 
89 void qcow2_cache_clean_unused(BlockDriverState *bs, Qcow2Cache *c)
90 {
91     int i = 0;
92     while (i < c->size) {
93         int to_clean = 0;
94 
95         /* Skip the entries that we don't need to clean */
96         while (i < c->size && !can_clean_entry(c, i)) {
97             i++;
98         }
99 
100         /* And count how many we can clean in a row */
101         while (i < c->size && can_clean_entry(c, i)) {
102             c->entries[i].offset = 0;
103             c->entries[i].lru_counter = 0;
104             i++;
105             to_clean++;
106         }
107 
108         if (to_clean > 0) {
109             qcow2_cache_table_release(bs, c, i - to_clean, to_clean);
110         }
111     }
112 
113     c->cache_clean_lru_counter = c->lru_counter;
114 }
115 
116 Qcow2Cache *qcow2_cache_create(BlockDriverState *bs, int num_tables)
117 {
118     BDRVQcow2State *s = bs->opaque;
119     Qcow2Cache *c;
120 
121     c = g_new0(Qcow2Cache, 1);
122     c->size = num_tables;
123     c->entries = g_try_new0(Qcow2CachedTable, num_tables);
124     c->table_array = qemu_try_blockalign(bs->file->bs,
125                                          (size_t) num_tables * s->cluster_size);
126 
127     if (!c->entries || !c->table_array) {
128         qemu_vfree(c->table_array);
129         g_free(c->entries);
130         g_free(c);
131         c = NULL;
132     }
133 
134     return c;
135 }
136 
137 int qcow2_cache_destroy(BlockDriverState *bs, Qcow2Cache *c)
138 {
139     int i;
140 
141     for (i = 0; i < c->size; i++) {
142         assert(c->entries[i].ref == 0);
143     }
144 
145     qemu_vfree(c->table_array);
146     g_free(c->entries);
147     g_free(c);
148 
149     return 0;
150 }
151 
152 static int qcow2_cache_flush_dependency(BlockDriverState *bs, Qcow2Cache *c)
153 {
154     int ret;
155 
156     ret = qcow2_cache_flush(bs, c->depends);
157     if (ret < 0) {
158         return ret;
159     }
160 
161     c->depends = NULL;
162     c->depends_on_flush = false;
163 
164     return 0;
165 }
166 
167 static int qcow2_cache_entry_flush(BlockDriverState *bs, Qcow2Cache *c, int i)
168 {
169     BDRVQcow2State *s = bs->opaque;
170     int ret = 0;
171 
172     if (!c->entries[i].dirty || !c->entries[i].offset) {
173         return 0;
174     }
175 
176     trace_qcow2_cache_entry_flush(qemu_coroutine_self(),
177                                   c == s->l2_table_cache, i);
178 
179     if (c->depends) {
180         ret = qcow2_cache_flush_dependency(bs, c);
181     } else if (c->depends_on_flush) {
182         ret = bdrv_flush(bs->file->bs);
183         if (ret >= 0) {
184             c->depends_on_flush = false;
185         }
186     }
187 
188     if (ret < 0) {
189         return ret;
190     }
191 
192     if (c == s->refcount_block_cache) {
193         ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_REFCOUNT_BLOCK,
194                 c->entries[i].offset, s->cluster_size);
195     } else if (c == s->l2_table_cache) {
196         ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2,
197                 c->entries[i].offset, s->cluster_size);
198     } else {
199         ret = qcow2_pre_write_overlap_check(bs, 0,
200                 c->entries[i].offset, s->cluster_size);
201     }
202 
203     if (ret < 0) {
204         return ret;
205     }
206 
207     if (c == s->refcount_block_cache) {
208         BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_UPDATE_PART);
209     } else if (c == s->l2_table_cache) {
210         BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
211     }
212 
213     ret = bdrv_pwrite(bs->file, c->entries[i].offset,
214                       qcow2_cache_get_table_addr(bs, c, i), s->cluster_size);
215     if (ret < 0) {
216         return ret;
217     }
218 
219     c->entries[i].dirty = false;
220 
221     return 0;
222 }
223 
224 int qcow2_cache_write(BlockDriverState *bs, Qcow2Cache *c)
225 {
226     BDRVQcow2State *s = bs->opaque;
227     int result = 0;
228     int ret;
229     int i;
230 
231     trace_qcow2_cache_flush(qemu_coroutine_self(), c == s->l2_table_cache);
232 
233     for (i = 0; i < c->size; i++) {
234         ret = qcow2_cache_entry_flush(bs, c, i);
235         if (ret < 0 && result != -ENOSPC) {
236             result = ret;
237         }
238     }
239 
240     return result;
241 }
242 
243 int qcow2_cache_flush(BlockDriverState *bs, Qcow2Cache *c)
244 {
245     int result = qcow2_cache_write(bs, c);
246 
247     if (result == 0) {
248         int ret = bdrv_flush(bs->file->bs);
249         if (ret < 0) {
250             result = ret;
251         }
252     }
253 
254     return result;
255 }
256 
257 int qcow2_cache_set_dependency(BlockDriverState *bs, Qcow2Cache *c,
258     Qcow2Cache *dependency)
259 {
260     int ret;
261 
262     if (dependency->depends) {
263         ret = qcow2_cache_flush_dependency(bs, dependency);
264         if (ret < 0) {
265             return ret;
266         }
267     }
268 
269     if (c->depends && (c->depends != dependency)) {
270         ret = qcow2_cache_flush_dependency(bs, c);
271         if (ret < 0) {
272             return ret;
273         }
274     }
275 
276     c->depends = dependency;
277     return 0;
278 }
279 
280 void qcow2_cache_depends_on_flush(Qcow2Cache *c)
281 {
282     c->depends_on_flush = true;
283 }
284 
285 int qcow2_cache_empty(BlockDriverState *bs, Qcow2Cache *c)
286 {
287     int ret, i;
288 
289     ret = qcow2_cache_flush(bs, c);
290     if (ret < 0) {
291         return ret;
292     }
293 
294     for (i = 0; i < c->size; i++) {
295         assert(c->entries[i].ref == 0);
296         c->entries[i].offset = 0;
297         c->entries[i].lru_counter = 0;
298     }
299 
300     qcow2_cache_table_release(bs, c, 0, c->size);
301 
302     c->lru_counter = 0;
303 
304     return 0;
305 }
306 
307 static int qcow2_cache_do_get(BlockDriverState *bs, Qcow2Cache *c,
308     uint64_t offset, void **table, bool read_from_disk)
309 {
310     BDRVQcow2State *s = bs->opaque;
311     int i;
312     int ret;
313     int lookup_index;
314     uint64_t min_lru_counter = UINT64_MAX;
315     int min_lru_index = -1;
316 
317     trace_qcow2_cache_get(qemu_coroutine_self(), c == s->l2_table_cache,
318                           offset, read_from_disk);
319 
320     /* Check if the table is already cached */
321     i = lookup_index = (offset / s->cluster_size * 4) % c->size;
322     do {
323         const Qcow2CachedTable *t = &c->entries[i];
324         if (t->offset == offset) {
325             goto found;
326         }
327         if (t->ref == 0 && t->lru_counter < min_lru_counter) {
328             min_lru_counter = t->lru_counter;
329             min_lru_index = i;
330         }
331         if (++i == c->size) {
332             i = 0;
333         }
334     } while (i != lookup_index);
335 
336     if (min_lru_index == -1) {
337         /* This can't happen in current synchronous code, but leave the check
338          * here as a reminder for whoever starts using AIO with the cache */
339         abort();
340     }
341 
342     /* Cache miss: write a table back and replace it */
343     i = min_lru_index;
344     trace_qcow2_cache_get_replace_entry(qemu_coroutine_self(),
345                                         c == s->l2_table_cache, i);
346 
347     ret = qcow2_cache_entry_flush(bs, c, i);
348     if (ret < 0) {
349         return ret;
350     }
351 
352     trace_qcow2_cache_get_read(qemu_coroutine_self(),
353                                c == s->l2_table_cache, i);
354     c->entries[i].offset = 0;
355     if (read_from_disk) {
356         if (c == s->l2_table_cache) {
357             BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
358         }
359 
360         ret = bdrv_pread(bs->file, offset,
361                          qcow2_cache_get_table_addr(bs, c, i),
362                          s->cluster_size);
363         if (ret < 0) {
364             return ret;
365         }
366     }
367 
368     c->entries[i].offset = offset;
369 
370     /* And return the right table */
371 found:
372     c->entries[i].ref++;
373     *table = qcow2_cache_get_table_addr(bs, c, i);
374 
375     trace_qcow2_cache_get_done(qemu_coroutine_self(),
376                                c == s->l2_table_cache, i);
377 
378     return 0;
379 }
380 
381 int qcow2_cache_get(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
382     void **table)
383 {
384     return qcow2_cache_do_get(bs, c, offset, table, true);
385 }
386 
387 int qcow2_cache_get_empty(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
388     void **table)
389 {
390     return qcow2_cache_do_get(bs, c, offset, table, false);
391 }
392 
393 void qcow2_cache_put(BlockDriverState *bs, Qcow2Cache *c, void **table)
394 {
395     int i = qcow2_cache_get_table_idx(bs, c, *table);
396 
397     c->entries[i].ref--;
398     *table = NULL;
399 
400     if (c->entries[i].ref == 0) {
401         c->entries[i].lru_counter = ++c->lru_counter;
402     }
403 
404     assert(c->entries[i].ref >= 0);
405 }
406 
407 void qcow2_cache_entry_mark_dirty(BlockDriverState *bs, Qcow2Cache *c,
408      void *table)
409 {
410     int i = qcow2_cache_get_table_idx(bs, c, table);
411     assert(c->entries[i].offset != 0);
412     c->entries[i].dirty = true;
413 }
414 
415 void *qcow2_cache_is_table_offset(BlockDriverState *bs, Qcow2Cache *c,
416                                   uint64_t offset)
417 {
418     int i;
419 
420     for (i = 0; i < c->size; i++) {
421         if (c->entries[i].offset == offset) {
422             return qcow2_cache_get_table_addr(bs, c, i);
423         }
424     }
425     return NULL;
426 }
427 
428 void qcow2_cache_discard(BlockDriverState *bs, Qcow2Cache *c, void *table)
429 {
430     int i = qcow2_cache_get_table_idx(bs, c, table);
431 
432     assert(c->entries[i].ref == 0);
433 
434     c->entries[i].offset = 0;
435     c->entries[i].lru_counter = 0;
436     c->entries[i].dirty = false;
437 
438     qcow2_cache_table_release(bs, c, i, 1);
439 }
440