xref: /qemu/block/qcow2-cache.c (revision ab9056ff)
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 "qcow2.h"
27 #include "trace.h"
28 
29 typedef struct Qcow2CachedTable {
30     int64_t  offset;
31     uint64_t lru_counter;
32     int      ref;
33     bool     dirty;
34 } Qcow2CachedTable;
35 
36 struct Qcow2Cache {
37     Qcow2CachedTable       *entries;
38     struct Qcow2Cache      *depends;
39     int                     size;
40     int                     table_size;
41     bool                    depends_on_flush;
42     void                   *table_array;
43     uint64_t                lru_counter;
44     uint64_t                cache_clean_lru_counter;
45 };
46 
47 static inline void *qcow2_cache_get_table_addr(Qcow2Cache *c, int table)
48 {
49     return (uint8_t *) c->table_array + (size_t) table * c->table_size;
50 }
51 
52 static inline int qcow2_cache_get_table_idx(Qcow2Cache *c, void *table)
53 {
54     ptrdiff_t table_offset = (uint8_t *) table - (uint8_t *) c->table_array;
55     int idx = table_offset / c->table_size;
56     assert(idx >= 0 && idx < c->size && table_offset % c->table_size == 0);
57     return idx;
58 }
59 
60 static inline const char *qcow2_cache_get_name(BDRVQcow2State *s, Qcow2Cache *c)
61 {
62     if (c == s->refcount_block_cache) {
63         return "refcount block";
64     } else if (c == s->l2_table_cache) {
65         return "L2 table";
66     } else {
67         /* Do not abort, because this is not critical */
68         return "unknown";
69     }
70 }
71 
72 static void qcow2_cache_table_release(Qcow2Cache *c, int i, int num_tables)
73 {
74 /* Using MADV_DONTNEED to discard memory is a Linux-specific feature */
75 #ifdef CONFIG_LINUX
76     void *t = qcow2_cache_get_table_addr(c, i);
77     int align = qemu_real_host_page_size;
78     size_t mem_size = (size_t) c->table_size * num_tables;
79     size_t offset = QEMU_ALIGN_UP((uintptr_t) t, align) - (uintptr_t) t;
80     size_t length = QEMU_ALIGN_DOWN(mem_size - offset, align);
81     if (mem_size > offset && length > 0) {
82         madvise((uint8_t *) t + offset, length, MADV_DONTNEED);
83     }
84 #endif
85 }
86 
87 static inline bool can_clean_entry(Qcow2Cache *c, int i)
88 {
89     Qcow2CachedTable *t = &c->entries[i];
90     return t->ref == 0 && !t->dirty && t->offset != 0 &&
91         t->lru_counter <= c->cache_clean_lru_counter;
92 }
93 
94 void qcow2_cache_clean_unused(Qcow2Cache *c)
95 {
96     int i = 0;
97     while (i < c->size) {
98         int to_clean = 0;
99 
100         /* Skip the entries that we don't need to clean */
101         while (i < c->size && !can_clean_entry(c, i)) {
102             i++;
103         }
104 
105         /* And count how many we can clean in a row */
106         while (i < c->size && can_clean_entry(c, i)) {
107             c->entries[i].offset = 0;
108             c->entries[i].lru_counter = 0;
109             i++;
110             to_clean++;
111         }
112 
113         if (to_clean > 0) {
114             qcow2_cache_table_release(c, i - to_clean, to_clean);
115         }
116     }
117 
118     c->cache_clean_lru_counter = c->lru_counter;
119 }
120 
121 Qcow2Cache *qcow2_cache_create(BlockDriverState *bs, int num_tables,
122                                unsigned table_size)
123 {
124     BDRVQcow2State *s = bs->opaque;
125     Qcow2Cache *c;
126 
127     assert(num_tables > 0);
128     assert(is_power_of_2(table_size));
129     assert(table_size >= (1 << MIN_CLUSTER_BITS));
130     assert(table_size <= s->cluster_size);
131 
132     c = g_new0(Qcow2Cache, 1);
133     c->size = num_tables;
134     c->table_size = table_size;
135     c->entries = g_try_new0(Qcow2CachedTable, num_tables);
136     c->table_array = qemu_try_blockalign(bs->file->bs,
137                                          (size_t) num_tables * c->table_size);
138 
139     if (!c->entries || !c->table_array) {
140         qemu_vfree(c->table_array);
141         g_free(c->entries);
142         g_free(c);
143         c = NULL;
144     }
145 
146     return c;
147 }
148 
149 int qcow2_cache_destroy(Qcow2Cache *c)
150 {
151     int i;
152 
153     for (i = 0; i < c->size; i++) {
154         assert(c->entries[i].ref == 0);
155     }
156 
157     qemu_vfree(c->table_array);
158     g_free(c->entries);
159     g_free(c);
160 
161     return 0;
162 }
163 
164 static int qcow2_cache_flush_dependency(BlockDriverState *bs, Qcow2Cache *c)
165 {
166     int ret;
167 
168     ret = qcow2_cache_flush(bs, c->depends);
169     if (ret < 0) {
170         return ret;
171     }
172 
173     c->depends = NULL;
174     c->depends_on_flush = false;
175 
176     return 0;
177 }
178 
179 static int qcow2_cache_entry_flush(BlockDriverState *bs, Qcow2Cache *c, int i)
180 {
181     BDRVQcow2State *s = bs->opaque;
182     int ret = 0;
183 
184     if (!c->entries[i].dirty || !c->entries[i].offset) {
185         return 0;
186     }
187 
188     trace_qcow2_cache_entry_flush(qemu_coroutine_self(),
189                                   c == s->l2_table_cache, i);
190 
191     if (c->depends) {
192         ret = qcow2_cache_flush_dependency(bs, c);
193     } else if (c->depends_on_flush) {
194         ret = bdrv_flush(bs->file->bs);
195         if (ret >= 0) {
196             c->depends_on_flush = false;
197         }
198     }
199 
200     if (ret < 0) {
201         return ret;
202     }
203 
204     if (c == s->refcount_block_cache) {
205         ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_REFCOUNT_BLOCK,
206                 c->entries[i].offset, c->table_size, false);
207     } else if (c == s->l2_table_cache) {
208         ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2,
209                 c->entries[i].offset, c->table_size, false);
210     } else {
211         ret = qcow2_pre_write_overlap_check(bs, 0,
212                 c->entries[i].offset, c->table_size, false);
213     }
214 
215     if (ret < 0) {
216         return ret;
217     }
218 
219     if (c == s->refcount_block_cache) {
220         BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_UPDATE_PART);
221     } else if (c == s->l2_table_cache) {
222         BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
223     }
224 
225     ret = bdrv_pwrite(bs->file, c->entries[i].offset,
226                       qcow2_cache_get_table_addr(c, i), c->table_size);
227     if (ret < 0) {
228         return ret;
229     }
230 
231     c->entries[i].dirty = false;
232 
233     return 0;
234 }
235 
236 int qcow2_cache_write(BlockDriverState *bs, Qcow2Cache *c)
237 {
238     BDRVQcow2State *s = bs->opaque;
239     int result = 0;
240     int ret;
241     int i;
242 
243     trace_qcow2_cache_flush(qemu_coroutine_self(), c == s->l2_table_cache);
244 
245     for (i = 0; i < c->size; i++) {
246         ret = qcow2_cache_entry_flush(bs, c, i);
247         if (ret < 0 && result != -ENOSPC) {
248             result = ret;
249         }
250     }
251 
252     return result;
253 }
254 
255 int qcow2_cache_flush(BlockDriverState *bs, Qcow2Cache *c)
256 {
257     int result = qcow2_cache_write(bs, c);
258 
259     if (result == 0) {
260         int ret = bdrv_flush(bs->file->bs);
261         if (ret < 0) {
262             result = ret;
263         }
264     }
265 
266     return result;
267 }
268 
269 int qcow2_cache_set_dependency(BlockDriverState *bs, Qcow2Cache *c,
270     Qcow2Cache *dependency)
271 {
272     int ret;
273 
274     if (dependency->depends) {
275         ret = qcow2_cache_flush_dependency(bs, dependency);
276         if (ret < 0) {
277             return ret;
278         }
279     }
280 
281     if (c->depends && (c->depends != dependency)) {
282         ret = qcow2_cache_flush_dependency(bs, c);
283         if (ret < 0) {
284             return ret;
285         }
286     }
287 
288     c->depends = dependency;
289     return 0;
290 }
291 
292 void qcow2_cache_depends_on_flush(Qcow2Cache *c)
293 {
294     c->depends_on_flush = true;
295 }
296 
297 int qcow2_cache_empty(BlockDriverState *bs, Qcow2Cache *c)
298 {
299     int ret, i;
300 
301     ret = qcow2_cache_flush(bs, c);
302     if (ret < 0) {
303         return ret;
304     }
305 
306     for (i = 0; i < c->size; i++) {
307         assert(c->entries[i].ref == 0);
308         c->entries[i].offset = 0;
309         c->entries[i].lru_counter = 0;
310     }
311 
312     qcow2_cache_table_release(c, 0, c->size);
313 
314     c->lru_counter = 0;
315 
316     return 0;
317 }
318 
319 static int qcow2_cache_do_get(BlockDriverState *bs, Qcow2Cache *c,
320     uint64_t offset, void **table, bool read_from_disk)
321 {
322     BDRVQcow2State *s = bs->opaque;
323     int i;
324     int ret;
325     int lookup_index;
326     uint64_t min_lru_counter = UINT64_MAX;
327     int min_lru_index = -1;
328 
329     assert(offset != 0);
330 
331     trace_qcow2_cache_get(qemu_coroutine_self(), c == s->l2_table_cache,
332                           offset, read_from_disk);
333 
334     if (!QEMU_IS_ALIGNED(offset, c->table_size)) {
335         qcow2_signal_corruption(bs, true, -1, -1, "Cannot get entry from %s "
336                                 "cache: Offset %#" PRIx64 " is unaligned",
337                                 qcow2_cache_get_name(s, c), offset);
338         return -EIO;
339     }
340 
341     /* Check if the table is already cached */
342     i = lookup_index = (offset / c->table_size * 4) % c->size;
343     do {
344         const Qcow2CachedTable *t = &c->entries[i];
345         if (t->offset == offset) {
346             goto found;
347         }
348         if (t->ref == 0 && t->lru_counter < min_lru_counter) {
349             min_lru_counter = t->lru_counter;
350             min_lru_index = i;
351         }
352         if (++i == c->size) {
353             i = 0;
354         }
355     } while (i != lookup_index);
356 
357     if (min_lru_index == -1) {
358         /* This can't happen in current synchronous code, but leave the check
359          * here as a reminder for whoever starts using AIO with the cache */
360         abort();
361     }
362 
363     /* Cache miss: write a table back and replace it */
364     i = min_lru_index;
365     trace_qcow2_cache_get_replace_entry(qemu_coroutine_self(),
366                                         c == s->l2_table_cache, i);
367 
368     ret = qcow2_cache_entry_flush(bs, c, i);
369     if (ret < 0) {
370         return ret;
371     }
372 
373     trace_qcow2_cache_get_read(qemu_coroutine_self(),
374                                c == s->l2_table_cache, i);
375     c->entries[i].offset = 0;
376     if (read_from_disk) {
377         if (c == s->l2_table_cache) {
378             BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
379         }
380 
381         ret = bdrv_pread(bs->file, offset,
382                          qcow2_cache_get_table_addr(c, i),
383                          c->table_size);
384         if (ret < 0) {
385             return ret;
386         }
387     }
388 
389     c->entries[i].offset = offset;
390 
391     /* And return the right table */
392 found:
393     c->entries[i].ref++;
394     *table = qcow2_cache_get_table_addr(c, i);
395 
396     trace_qcow2_cache_get_done(qemu_coroutine_self(),
397                                c == s->l2_table_cache, i);
398 
399     return 0;
400 }
401 
402 int qcow2_cache_get(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
403     void **table)
404 {
405     return qcow2_cache_do_get(bs, c, offset, table, true);
406 }
407 
408 int qcow2_cache_get_empty(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
409     void **table)
410 {
411     return qcow2_cache_do_get(bs, c, offset, table, false);
412 }
413 
414 void qcow2_cache_put(Qcow2Cache *c, void **table)
415 {
416     int i = qcow2_cache_get_table_idx(c, *table);
417 
418     c->entries[i].ref--;
419     *table = NULL;
420 
421     if (c->entries[i].ref == 0) {
422         c->entries[i].lru_counter = ++c->lru_counter;
423     }
424 
425     assert(c->entries[i].ref >= 0);
426 }
427 
428 void qcow2_cache_entry_mark_dirty(Qcow2Cache *c, void *table)
429 {
430     int i = qcow2_cache_get_table_idx(c, table);
431     assert(c->entries[i].offset != 0);
432     c->entries[i].dirty = true;
433 }
434 
435 void *qcow2_cache_is_table_offset(Qcow2Cache *c, uint64_t offset)
436 {
437     int i;
438 
439     for (i = 0; i < c->size; i++) {
440         if (c->entries[i].offset == offset) {
441             return qcow2_cache_get_table_addr(c, i);
442         }
443     }
444     return NULL;
445 }
446 
447 void qcow2_cache_discard(Qcow2Cache *c, void *table)
448 {
449     int i = qcow2_cache_get_table_idx(c, table);
450 
451     assert(c->entries[i].ref == 0);
452 
453     c->entries[i].offset = 0;
454     c->entries[i].lru_counter = 0;
455     c->entries[i].dirty = false;
456 
457     qcow2_cache_table_release(c, i, 1);
458 }
459