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 "qemu/osdep.h"
26 #include "block/block_int.h"
27 #include "qemu-common.h"
31 typedef struct Qcow2CachedTable {
39 Qcow2CachedTable *entries;
40 struct Qcow2Cache *depends;
43 bool depends_on_flush;
46 uint64_t cache_clean_lru_counter;
49 static inline void *qcow2_cache_get_table_addr(Qcow2Cache *c, int table)
51 return (uint8_t *) c->table_array + (size_t) table * c->table_size;
54 static inline int qcow2_cache_get_table_idx(Qcow2Cache *c, void *table)
56 ptrdiff_t table_offset = (uint8_t *) table - (uint8_t *) c->table_array;
57 int idx = table_offset / c->table_size;
58 assert(idx >= 0 && idx < c->size && table_offset % c->table_size == 0);
62 static inline const char *qcow2_cache_get_name(BDRVQcow2State *s, Qcow2Cache *c)
64 if (c == s->refcount_block_cache) {
65 return "refcount block";
66 } else if (c == s->l2_table_cache) {
69 /* Do not abort, because this is not critical */
74 static void qcow2_cache_table_release(Qcow2Cache *c, int i, int num_tables)
76 /* Using MADV_DONTNEED to discard memory is a Linux-specific feature */
78 void *t = qcow2_cache_get_table_addr(c, i);
79 int align = getpagesize();
80 size_t mem_size = (size_t) c->table_size * num_tables;
81 size_t offset = QEMU_ALIGN_UP((uintptr_t) t, align) - (uintptr_t) t;
82 size_t length = QEMU_ALIGN_DOWN(mem_size - offset, align);
83 if (mem_size > offset && length > 0) {
84 madvise((uint8_t *) t + offset, length, MADV_DONTNEED);
89 static inline bool can_clean_entry(Qcow2Cache *c, int i)
91 Qcow2CachedTable *t = &c->entries[i];
92 return t->ref == 0 && !t->dirty && t->offset != 0 &&
93 t->lru_counter <= c->cache_clean_lru_counter;
96 void qcow2_cache_clean_unused(Qcow2Cache *c)
102 /* Skip the entries that we don't need to clean */
103 while (i < c->size && !can_clean_entry(c, i)) {
107 /* And count how many we can clean in a row */
108 while (i < c->size && can_clean_entry(c, i)) {
109 c->entries[i].offset = 0;
110 c->entries[i].lru_counter = 0;
116 qcow2_cache_table_release(c, i - to_clean, to_clean);
120 c->cache_clean_lru_counter = c->lru_counter;
123 Qcow2Cache *qcow2_cache_create(BlockDriverState *bs, int num_tables)
125 BDRVQcow2State *s = bs->opaque;
128 c = g_new0(Qcow2Cache, 1);
129 c->size = num_tables;
130 c->table_size = s->cluster_size;
131 c->entries = g_try_new0(Qcow2CachedTable, num_tables);
132 c->table_array = qemu_try_blockalign(bs->file->bs,
133 (size_t) num_tables * c->table_size);
135 if (!c->entries || !c->table_array) {
136 qemu_vfree(c->table_array);
145 int qcow2_cache_destroy(Qcow2Cache *c)
149 for (i = 0; i < c->size; i++) {
150 assert(c->entries[i].ref == 0);
153 qemu_vfree(c->table_array);
160 static int qcow2_cache_flush_dependency(BlockDriverState *bs, Qcow2Cache *c)
164 ret = qcow2_cache_flush(bs, c->depends);
170 c->depends_on_flush = false;
175 static int qcow2_cache_entry_flush(BlockDriverState *bs, Qcow2Cache *c, int i)
177 BDRVQcow2State *s = bs->opaque;
180 if (!c->entries[i].dirty || !c->entries[i].offset) {
184 trace_qcow2_cache_entry_flush(qemu_coroutine_self(),
185 c == s->l2_table_cache, i);
188 ret = qcow2_cache_flush_dependency(bs, c);
189 } else if (c->depends_on_flush) {
190 ret = bdrv_flush(bs->file->bs);
192 c->depends_on_flush = false;
200 if (c == s->refcount_block_cache) {
201 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_REFCOUNT_BLOCK,
202 c->entries[i].offset, c->table_size);
203 } else if (c == s->l2_table_cache) {
204 ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L2,
205 c->entries[i].offset, c->table_size);
207 ret = qcow2_pre_write_overlap_check(bs, 0,
208 c->entries[i].offset, c->table_size);
215 if (c == s->refcount_block_cache) {
216 BLKDBG_EVENT(bs->file, BLKDBG_REFBLOCK_UPDATE_PART);
217 } else if (c == s->l2_table_cache) {
218 BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE);
221 ret = bdrv_pwrite(bs->file, c->entries[i].offset,
222 qcow2_cache_get_table_addr(c, i), c->table_size);
227 c->entries[i].dirty = false;
232 int qcow2_cache_write(BlockDriverState *bs, Qcow2Cache *c)
234 BDRVQcow2State *s = bs->opaque;
239 trace_qcow2_cache_flush(qemu_coroutine_self(), c == s->l2_table_cache);
241 for (i = 0; i < c->size; i++) {
242 ret = qcow2_cache_entry_flush(bs, c, i);
243 if (ret < 0 && result != -ENOSPC) {
251 int qcow2_cache_flush(BlockDriverState *bs, Qcow2Cache *c)
253 int result = qcow2_cache_write(bs, c);
256 int ret = bdrv_flush(bs->file->bs);
265 int qcow2_cache_set_dependency(BlockDriverState *bs, Qcow2Cache *c,
266 Qcow2Cache *dependency)
270 if (dependency->depends) {
271 ret = qcow2_cache_flush_dependency(bs, dependency);
277 if (c->depends && (c->depends != dependency)) {
278 ret = qcow2_cache_flush_dependency(bs, c);
284 c->depends = dependency;
288 void qcow2_cache_depends_on_flush(Qcow2Cache *c)
290 c->depends_on_flush = true;
293 int qcow2_cache_empty(BlockDriverState *bs, Qcow2Cache *c)
297 ret = qcow2_cache_flush(bs, c);
302 for (i = 0; i < c->size; i++) {
303 assert(c->entries[i].ref == 0);
304 c->entries[i].offset = 0;
305 c->entries[i].lru_counter = 0;
308 qcow2_cache_table_release(c, 0, c->size);
315 static int qcow2_cache_do_get(BlockDriverState *bs, Qcow2Cache *c,
316 uint64_t offset, void **table, bool read_from_disk)
318 BDRVQcow2State *s = bs->opaque;
322 uint64_t min_lru_counter = UINT64_MAX;
323 int min_lru_index = -1;
327 trace_qcow2_cache_get(qemu_coroutine_self(), c == s->l2_table_cache,
328 offset, read_from_disk);
330 if (!QEMU_IS_ALIGNED(offset, c->table_size)) {
331 qcow2_signal_corruption(bs, true, -1, -1, "Cannot get entry from %s "
332 "cache: Offset %#" PRIx64 " is unaligned",
333 qcow2_cache_get_name(s, c), offset);
337 /* Check if the table is already cached */
338 i = lookup_index = (offset / c->table_size * 4) % c->size;
340 const Qcow2CachedTable *t = &c->entries[i];
341 if (t->offset == offset) {
344 if (t->ref == 0 && t->lru_counter < min_lru_counter) {
345 min_lru_counter = t->lru_counter;
348 if (++i == c->size) {
351 } while (i != lookup_index);
353 if (min_lru_index == -1) {
354 /* This can't happen in current synchronous code, but leave the check
355 * here as a reminder for whoever starts using AIO with the cache */
359 /* Cache miss: write a table back and replace it */
361 trace_qcow2_cache_get_replace_entry(qemu_coroutine_self(),
362 c == s->l2_table_cache, i);
364 ret = qcow2_cache_entry_flush(bs, c, i);
369 trace_qcow2_cache_get_read(qemu_coroutine_self(),
370 c == s->l2_table_cache, i);
371 c->entries[i].offset = 0;
372 if (read_from_disk) {
373 if (c == s->l2_table_cache) {
374 BLKDBG_EVENT(bs->file, BLKDBG_L2_LOAD);
377 ret = bdrv_pread(bs->file, offset,
378 qcow2_cache_get_table_addr(c, i),
385 c->entries[i].offset = offset;
387 /* And return the right table */
390 *table = qcow2_cache_get_table_addr(c, i);
392 trace_qcow2_cache_get_done(qemu_coroutine_self(),
393 c == s->l2_table_cache, i);
398 int qcow2_cache_get(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
401 return qcow2_cache_do_get(bs, c, offset, table, true);
404 int qcow2_cache_get_empty(BlockDriverState *bs, Qcow2Cache *c, uint64_t offset,
407 return qcow2_cache_do_get(bs, c, offset, table, false);
410 void qcow2_cache_put(Qcow2Cache *c, void **table)
412 int i = qcow2_cache_get_table_idx(c, *table);
417 if (c->entries[i].ref == 0) {
418 c->entries[i].lru_counter = ++c->lru_counter;
421 assert(c->entries[i].ref >= 0);
424 void qcow2_cache_entry_mark_dirty(Qcow2Cache *c, void *table)
426 int i = qcow2_cache_get_table_idx(c, table);
427 assert(c->entries[i].offset != 0);
428 c->entries[i].dirty = true;
431 void *qcow2_cache_is_table_offset(Qcow2Cache *c, uint64_t offset)
435 for (i = 0; i < c->size; i++) {
436 if (c->entries[i].offset == offset) {
437 return qcow2_cache_get_table_addr(c, i);
443 void qcow2_cache_discard(Qcow2Cache *c, void *table)
445 int i = qcow2_cache_get_table_idx(c, table);
447 assert(c->entries[i].ref == 0);
449 c->entries[i].offset = 0;
450 c->entries[i].lru_counter = 0;
451 c->entries[i].dirty = false;
453 qcow2_cache_table_release(c, i, 1);