projects/linux·blob v3.7 · refs

620 lines · 16.4 KiB

1/*
2 * linux/fs/mbcache.c
3 * (C) 2001-2002 Andreas Gruenbacher, <[email protected]>
4 */
5
6/*
7 * Filesystem Meta Information Block Cache (mbcache)
8 *
9 * The mbcache caches blocks of block devices that need to be located
10 * by their device/block number, as well as by other criteria (such
11 * as the block's contents).
12 *
13 * There can only be one cache entry in a cache per device and block number.
14 * Additional indexes need not be unique in this sense. The number of
15 * additional indexes (=other criteria) can be hardwired at compile time
16 * or specified at cache create time.
17 *
18 * Each cache entry is of fixed size. An entry may be `valid' or `invalid'
19 * in the cache. A valid entry is in the main hash tables of the cache,
20 * and may also be in the lru list. An invalid entry is not in any hashes
21 * or lists.
22 *
23 * A valid cache entry is only in the lru list if no handles refer to it.
24 * Invalid cache entries will be freed when the last handle to the cache
25 * entry is released. Entries that cannot be freed immediately are put
26 * back on the lru list.
27 */
28
29#include <linux/kernel.h>
30#include <linux/module.h>
31
32#include <linux/hash.h>
33#include <linux/fs.h>
34#include <linux/mm.h>
35#include <linux/slab.h>
36#include <linux/sched.h>
37#include <linux/init.h>
38#include <linux/mbcache.h>
39
40
41#ifdef MB_CACHE_DEBUG
42# define mb_debug(f...) do { \
43 printk(KERN_DEBUG f); \
44 printk("\n"); \
45 } while (0)
46#define mb_assert(c) do { if (!(c)) \
47 printk(KERN_ERR "assertion " #c " failed\n"); \
48 } while(0)
49#else
50# define mb_debug(f...) do { } while(0)
51# define mb_assert(c) do { } while(0)
52#endif
53#define mb_error(f...) do { \
54 printk(KERN_ERR f); \
55 printk("\n"); \
56 } while(0)
57
58#define MB_CACHE_WRITER ((unsigned short)~0U >> 1)
59
60static DECLARE_WAIT_QUEUE_HEAD(mb_cache_queue);
61
62MODULE_AUTHOR("Andreas Gruenbacher <[email protected]>");
63MODULE_DESCRIPTION("Meta block cache (for extended attributes)");
64MODULE_LICENSE("GPL");
65
66EXPORT_SYMBOL(mb_cache_create);
67EXPORT_SYMBOL(mb_cache_shrink);
68EXPORT_SYMBOL(mb_cache_destroy);
69EXPORT_SYMBOL(mb_cache_entry_alloc);
70EXPORT_SYMBOL(mb_cache_entry_insert);
71EXPORT_SYMBOL(mb_cache_entry_release);
72EXPORT_SYMBOL(mb_cache_entry_free);
73EXPORT_SYMBOL(mb_cache_entry_get);
74#if !defined(MB_CACHE_INDEXES_COUNT) || (MB_CACHE_INDEXES_COUNT > 0)
75EXPORT_SYMBOL(mb_cache_entry_find_first);
76EXPORT_SYMBOL(mb_cache_entry_find_next);
77#endif
78
79/*
80 * Global data: list of all mbcache's, lru list, and a spinlock for
81 * accessing cache data structures on SMP machines. The lru list is
82 * global across all mbcaches.
83 */
84
85static LIST_HEAD(mb_cache_list);
86static LIST_HEAD(mb_cache_lru_list);
87static DEFINE_SPINLOCK(mb_cache_spinlock);
88
89/*
90 * What the mbcache registers as to get shrunk dynamically.
91 */
92
93static int mb_cache_shrink_fn(struct shrinker *shrink,
94 struct shrink_control *sc);
95
96static struct shrinker mb_cache_shrinker = {
97 .shrink = mb_cache_shrink_fn,
98 .seeks = DEFAULT_SEEKS,
99};
100
101static inline int
102__mb_cache_entry_is_hashed(struct mb_cache_entry *ce)
103{
104 return !list_empty(&ce->e_block_list);
105}
106
107
108static void
109__mb_cache_entry_unhash(struct mb_cache_entry *ce)
110{
111 if (__mb_cache_entry_is_hashed(ce)) {
112 list_del_init(&ce->e_block_list);
113 list_del(&ce->e_index.o_list);
114 }
115}
116
117
118static void
119__mb_cache_entry_forget(struct mb_cache_entry *ce, gfp_t gfp_mask)
120{
121 struct mb_cache *cache = ce->e_cache;
122
123 mb_assert(!(ce->e_used || ce->e_queued));
124 kmem_cache_free(cache->c_entry_cache, ce);
125 atomic_dec(&cache->c_entry_count);
126}
127
128
129static void
130__mb_cache_entry_release_unlock(struct mb_cache_entry *ce)
131 __releases(mb_cache_spinlock)
132{
133 /* Wake up all processes queuing for this cache entry. */
134 if (ce->e_queued)
135 wake_up_all(&mb_cache_queue);
136 if (ce->e_used >= MB_CACHE_WRITER)
137 ce->e_used -= MB_CACHE_WRITER;
138 ce->e_used--;
139 if (!(ce->e_used || ce->e_queued)) {
140 if (!__mb_cache_entry_is_hashed(ce))
141 goto forget;
142 mb_assert(list_empty(&ce->e_lru_list));
143 list_add_tail(&ce->e_lru_list, &mb_cache_lru_list);
144 }
145 spin_unlock(&mb_cache_spinlock);
146 return;
147forget:
148 spin_unlock(&mb_cache_spinlock);
149 __mb_cache_entry_forget(ce, GFP_KERNEL);
150}
151
152
153/*
154 * mb_cache_shrink_fn() memory pressure callback
155 *
156 * This function is called by the kernel memory management when memory
157 * gets low.
158 *
159 * @shrink: (ignored)
160 * @sc: shrink_control passed from reclaim
161 *
162 * Returns the number of objects which are present in the cache.
163 */
164static int
165mb_cache_shrink_fn(struct shrinker *shrink, struct shrink_control *sc)
166{
167 LIST_HEAD(free_list);
168 struct mb_cache *cache;
169 struct mb_cache_entry *entry, *tmp;
170 int count = 0;
171 int nr_to_scan = sc->nr_to_scan;
172 gfp_t gfp_mask = sc->gfp_mask;
173
174 mb_debug("trying to free %d entries", nr_to_scan);
175 spin_lock(&mb_cache_spinlock);
176 while (nr_to_scan-- && !list_empty(&mb_cache_lru_list)) {
177 struct mb_cache_entry *ce =
178 list_entry(mb_cache_lru_list.next,
179 struct mb_cache_entry, e_lru_list);
180 list_move_tail(&ce->e_lru_list, &free_list);
181 __mb_cache_entry_unhash(ce);
182 }
183 list_for_each_entry(cache, &mb_cache_list, c_cache_list) {
184 mb_debug("cache %s (%d)", cache->c_name,
185 atomic_read(&cache->c_entry_count));
186 count += atomic_read(&cache->c_entry_count);
187 }
188 spin_unlock(&mb_cache_spinlock);
189 list_for_each_entry_safe(entry, tmp, &free_list, e_lru_list) {
190 __mb_cache_entry_forget(entry, gfp_mask);
191 }
192 return (count / 100) * sysctl_vfs_cache_pressure;
193}
194
195
196/*
197 * mb_cache_create() create a new cache
198 *
199 * All entries in one cache are equal size. Cache entries may be from
200 * multiple devices. If this is the first mbcache created, registers
201 * the cache with kernel memory management. Returns NULL if no more
202 * memory was available.
203 *
204 * @name: name of the cache (informal)
205 * @bucket_bits: log2(number of hash buckets)
206 */
207struct mb_cache *
208mb_cache_create(const char *name, int bucket_bits)
209{
210 int n, bucket_count = 1 << bucket_bits;
211 struct mb_cache *cache = NULL;
212
213 cache = kmalloc(sizeof(struct mb_cache), GFP_KERNEL);
214 if (!cache)
215 return NULL;
216 cache->c_name = name;
217 atomic_set(&cache->c_entry_count, 0);
218 cache->c_bucket_bits = bucket_bits;
219 cache->c_block_hash = kmalloc(bucket_count * sizeof(struct list_head),
220 GFP_KERNEL);
221 if (!cache->c_block_hash)
222 goto fail;
223 for (n=0; n<bucket_count; n++)
224 INIT_LIST_HEAD(&cache->c_block_hash[n]);
225 cache->c_index_hash = kmalloc(bucket_count * sizeof(struct list_head),
226 GFP_KERNEL);
227 if (!cache->c_index_hash)
228 goto fail;
229 for (n=0; n<bucket_count; n++)
230 INIT_LIST_HEAD(&cache->c_index_hash[n]);
231 cache->c_entry_cache = kmem_cache_create(name,
232 sizeof(struct mb_cache_entry), 0,
233 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD, NULL);
234 if (!cache->c_entry_cache)
235 goto fail2;
236
237 /*
238 * Set an upper limit on the number of cache entries so that the hash
239 * chains won't grow too long.
240 */
241 cache->c_max_entries = bucket_count << 4;
242
243 spin_lock(&mb_cache_spinlock);
244 list_add(&cache->c_cache_list, &mb_cache_list);
245 spin_unlock(&mb_cache_spinlock);
246 return cache;
247
248fail2:
249 kfree(cache->c_index_hash);
250
251fail:
252 kfree(cache->c_block_hash);
253 kfree(cache);
254 return NULL;
255}
256
257
258/*
259 * mb_cache_shrink()
260 *
261 * Removes all cache entries of a device from the cache. All cache entries
262 * currently in use cannot be freed, and thus remain in the cache. All others
263 * are freed.
264 *
265 * @bdev: which device's cache entries to shrink
266 */
267void
268mb_cache_shrink(struct block_device *bdev)
269{
270 LIST_HEAD(free_list);
271 struct list_head *l, *ltmp;
272
273 spin_lock(&mb_cache_spinlock);
274 list_for_each_safe(l, ltmp, &mb_cache_lru_list) {
275 struct mb_cache_entry *ce =
276 list_entry(l, struct mb_cache_entry, e_lru_list);
277 if (ce->e_bdev == bdev) {
278 list_move_tail(&ce->e_lru_list, &free_list);
279 __mb_cache_entry_unhash(ce);
280 }
281 }
282 spin_unlock(&mb_cache_spinlock);
283 list_for_each_safe(l, ltmp, &free_list) {
284 __mb_cache_entry_forget(list_entry(l, struct mb_cache_entry,
285 e_lru_list), GFP_KERNEL);
286 }
287}
288
289
290/*
291 * mb_cache_destroy()
292 *
293 * Shrinks the cache to its minimum possible size (hopefully 0 entries),
294 * and then destroys it. If this was the last mbcache, un-registers the
295 * mbcache from kernel memory management.
296 */
297void
298mb_cache_destroy(struct mb_cache *cache)
299{
300 LIST_HEAD(free_list);
301 struct list_head *l, *ltmp;
302
303 spin_lock(&mb_cache_spinlock);
304 list_for_each_safe(l, ltmp, &mb_cache_lru_list) {
305 struct mb_cache_entry *ce =
306 list_entry(l, struct mb_cache_entry, e_lru_list);
307 if (ce->e_cache == cache) {
308 list_move_tail(&ce->e_lru_list, &free_list);
309 __mb_cache_entry_unhash(ce);
310 }
311 }
312 list_del(&cache->c_cache_list);
313 spin_unlock(&mb_cache_spinlock);
314
315 list_for_each_safe(l, ltmp, &free_list) {
316 __mb_cache_entry_forget(list_entry(l, struct mb_cache_entry,
317 e_lru_list), GFP_KERNEL);
318 }
319
320 if (atomic_read(&cache->c_entry_count) > 0) {
321 mb_error("cache %s: %d orphaned entries",
322 cache->c_name,
323 atomic_read(&cache->c_entry_count));
324 }
325
326 kmem_cache_destroy(cache->c_entry_cache);
327
328 kfree(cache->c_index_hash);
329 kfree(cache->c_block_hash);
330 kfree(cache);
331}
332
333/*
334 * mb_cache_entry_alloc()
335 *
336 * Allocates a new cache entry. The new entry will not be valid initially,
337 * and thus cannot be looked up yet. It should be filled with data, and
338 * then inserted into the cache using mb_cache_entry_insert(). Returns NULL
339 * if no more memory was available.
340 */
341struct mb_cache_entry *
342mb_cache_entry_alloc(struct mb_cache *cache, gfp_t gfp_flags)
343{
344 struct mb_cache_entry *ce = NULL;
345
346 if (atomic_read(&cache->c_entry_count) >= cache->c_max_entries) {
347 spin_lock(&mb_cache_spinlock);
348 if (!list_empty(&mb_cache_lru_list)) {
349 ce = list_entry(mb_cache_lru_list.next,
350 struct mb_cache_entry, e_lru_list);
351 list_del_init(&ce->e_lru_list);
352 __mb_cache_entry_unhash(ce);
353 }
354 spin_unlock(&mb_cache_spinlock);
355 }
356 if (!ce) {
357 ce = kmem_cache_alloc(cache->c_entry_cache, gfp_flags);
358 if (!ce)
359 return NULL;
360 atomic_inc(&cache->c_entry_count);
361 INIT_LIST_HEAD(&ce->e_lru_list);
362 INIT_LIST_HEAD(&ce->e_block_list);
363 ce->e_cache = cache;
364 ce->e_queued = 0;
365 }
366 ce->e_used = 1 + MB_CACHE_WRITER;
367 return ce;
368}
369
370
371/*
372 * mb_cache_entry_insert()
373 *
374 * Inserts an entry that was allocated using mb_cache_entry_alloc() into
375 * the cache. After this, the cache entry can be looked up, but is not yet
376 * in the lru list as the caller still holds a handle to it. Returns 0 on
377 * success, or -EBUSY if a cache entry for that device + inode exists
378 * already (this may happen after a failed lookup, but when another process
379 * has inserted the same cache entry in the meantime).
380 *
381 * @bdev: device the cache entry belongs to
382 * @block: block number
383 * @key: lookup key
384 */
385int
386mb_cache_entry_insert(struct mb_cache_entry *ce, struct block_device *bdev,
387 sector_t block, unsigned int key)
388{
389 struct mb_cache *cache = ce->e_cache;
390 unsigned int bucket;
391 struct list_head *l;
392 int error = -EBUSY;
393
394 bucket = hash_long((unsigned long)bdev + (block & 0xffffffff),
395 cache->c_bucket_bits);
396 spin_lock(&mb_cache_spinlock);
397 list_for_each_prev(l, &cache->c_block_hash[bucket]) {
398 struct mb_cache_entry *ce =
399 list_entry(l, struct mb_cache_entry, e_block_list);
400 if (ce->e_bdev == bdev && ce->e_block == block)
401 goto out;
402 }
403 __mb_cache_entry_unhash(ce);
404 ce->e_bdev = bdev;
405 ce->e_block = block;
406 list_add(&ce->e_block_list, &cache->c_block_hash[bucket]);
407 ce->e_index.o_key = key;
408 bucket = hash_long(key, cache->c_bucket_bits);
409 list_add(&ce->e_index.o_list, &cache->c_index_hash[bucket]);
410 error = 0;
411out:
412 spin_unlock(&mb_cache_spinlock);
413 return error;
414}
415
416
417/*
418 * mb_cache_entry_release()
419 *
420 * Release a handle to a cache entry. When the last handle to a cache entry
421 * is released it is either freed (if it is invalid) or otherwise inserted
422 * in to the lru list.
423 */
424void
425mb_cache_entry_release(struct mb_cache_entry *ce)
426{
427 spin_lock(&mb_cache_spinlock);
428 __mb_cache_entry_release_unlock(ce);
429}
430
431
432/*
433 * mb_cache_entry_free()
434 *
435 * This is equivalent to the sequence mb_cache_entry_takeout() --
436 * mb_cache_entry_release().
437 */
438void
439mb_cache_entry_free(struct mb_cache_entry *ce)
440{
441 spin_lock(&mb_cache_spinlock);
442 mb_assert(list_empty(&ce->e_lru_list));
443 __mb_cache_entry_unhash(ce);
444 __mb_cache_entry_release_unlock(ce);
445}
446
447
448/*
449 * mb_cache_entry_get()
450 *
451 * Get a cache entry by device / block number. (There can only be one entry
452 * in the cache per device and block.) Returns NULL if no such cache entry
453 * exists. The returned cache entry is locked for exclusive access ("single
454 * writer").
455 */
456struct mb_cache_entry *
457mb_cache_entry_get(struct mb_cache *cache, struct block_device *bdev,
458 sector_t block)
459{
460 unsigned int bucket;
461 struct list_head *l;
462 struct mb_cache_entry *ce;
463
464 bucket = hash_long((unsigned long)bdev + (block & 0xffffffff),
465 cache->c_bucket_bits);
466 spin_lock(&mb_cache_spinlock);
467 list_for_each(l, &cache->c_block_hash[bucket]) {
468 ce = list_entry(l, struct mb_cache_entry, e_block_list);
469 if (ce->e_bdev == bdev && ce->e_block == block) {
470 DEFINE_WAIT(wait);
471
472 if (!list_empty(&ce->e_lru_list))
473 list_del_init(&ce->e_lru_list);
474
475 while (ce->e_used > 0) {
476 ce->e_queued++;
477 prepare_to_wait(&mb_cache_queue, &wait,
478 TASK_UNINTERRUPTIBLE);
479 spin_unlock(&mb_cache_spinlock);
480 schedule();
481 spin_lock(&mb_cache_spinlock);
482 ce->e_queued--;
483 }
484 finish_wait(&mb_cache_queue, &wait);
485 ce->e_used += 1 + MB_CACHE_WRITER;
486
487 if (!__mb_cache_entry_is_hashed(ce)) {
488 __mb_cache_entry_release_unlock(ce);
489 return NULL;
490 }
491 goto cleanup;
492 }
493 }
494 ce = NULL;
495
496cleanup:
497 spin_unlock(&mb_cache_spinlock);
498 return ce;
499}
500
501#if !defined(MB_CACHE_INDEXES_COUNT) || (MB_CACHE_INDEXES_COUNT > 0)
502
503static struct mb_cache_entry *
504__mb_cache_entry_find(struct list_head *l, struct list_head *head,
505 struct block_device *bdev, unsigned int key)
506{
507 while (l != head) {
508 struct mb_cache_entry *ce =
509 list_entry(l, struct mb_cache_entry, e_index.o_list);
510 if (ce->e_bdev == bdev && ce->e_index.o_key == key) {
511 DEFINE_WAIT(wait);
512
513 if (!list_empty(&ce->e_lru_list))
514 list_del_init(&ce->e_lru_list);
515
516 /* Incrementing before holding the lock gives readers
517 priority over writers. */
518 ce->e_used++;
519 while (ce->e_used >= MB_CACHE_WRITER) {
520 ce->e_queued++;
521 prepare_to_wait(&mb_cache_queue, &wait,
522 TASK_UNINTERRUPTIBLE);
523 spin_unlock(&mb_cache_spinlock);
524 schedule();
525 spin_lock(&mb_cache_spinlock);
526 ce->e_queued--;
527 }
528 finish_wait(&mb_cache_queue, &wait);
529
530 if (!__mb_cache_entry_is_hashed(ce)) {
531 __mb_cache_entry_release_unlock(ce);
532 spin_lock(&mb_cache_spinlock);
533 return ERR_PTR(-EAGAIN);
534 }
535 return ce;
536 }
537 l = l->next;
538 }
539 return NULL;
540}
541
542
543/*
544 * mb_cache_entry_find_first()
545 *
546 * Find the first cache entry on a given device with a certain key in
547 * an additional index. Additional matches can be found with
548 * mb_cache_entry_find_next(). Returns NULL if no match was found. The
549 * returned cache entry is locked for shared access ("multiple readers").
550 *
551 * @cache: the cache to search
552 * @bdev: the device the cache entry should belong to
553 * @key: the key in the index
554 */
555struct mb_cache_entry *
556mb_cache_entry_find_first(struct mb_cache *cache, struct block_device *bdev,
557 unsigned int key)
558{
559 unsigned int bucket = hash_long(key, cache->c_bucket_bits);
560 struct list_head *l;
561 struct mb_cache_entry *ce;
562
563 spin_lock(&mb_cache_spinlock);
564 l = cache->c_index_hash[bucket].next;
565 ce = __mb_cache_entry_find(l, &cache->c_index_hash[bucket], bdev, key);
566 spin_unlock(&mb_cache_spinlock);
567 return ce;
568}
569
570
571/*
572 * mb_cache_entry_find_next()
573 *
574 * Find the next cache entry on a given device with a certain key in an
575 * additional index. Returns NULL if no match could be found. The previous
576 * entry is atomatically released, so that mb_cache_entry_find_next() can
577 * be called like this:
578 *
579 * entry = mb_cache_entry_find_first();
580 * while (entry) {
581 * ...
582 * entry = mb_cache_entry_find_next(entry, ...);
583 * }
584 *
585 * @prev: The previous match
586 * @bdev: the device the cache entry should belong to
587 * @key: the key in the index
588 */
589struct mb_cache_entry *
590mb_cache_entry_find_next(struct mb_cache_entry *prev,
591 struct block_device *bdev, unsigned int key)
592{
593 struct mb_cache *cache = prev->e_cache;
594 unsigned int bucket = hash_long(key, cache->c_bucket_bits);
595 struct list_head *l;
596 struct mb_cache_entry *ce;
597
598 spin_lock(&mb_cache_spinlock);
599 l = prev->e_index.o_list.next;
600 ce = __mb_cache_entry_find(l, &cache->c_index_hash[bucket], bdev, key);
601 __mb_cache_entry_release_unlock(prev);
602 return ce;
603}
604
605#endif /* !defined(MB_CACHE_INDEXES_COUNT) || (MB_CACHE_INDEXES_COUNT > 0) */
606
607static int __init init_mbcache(void)
608{
609 register_shrinker(&mb_cache_shrinker);
610 return 0;
611}
612
613static void __exit exit_mbcache(void)
614{
615 unregister_shrinker(&mb_cache_shrinker);
616}
617
618module_init(init_mbcache)
619module_exit(exit_mbcache)
620