xref: /linux/fs/mbcache.c (revision 25aee3debe0464f6c680173041fa3de30ec9ff54)
1 /*
2  * linux/fs/mbcache.c
3  * (C) 2001-2002 Andreas Gruenbacher, <a.gruenbacher@computer.org>
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 
60 static DECLARE_WAIT_QUEUE_HEAD(mb_cache_queue);
61 
62 MODULE_AUTHOR("Andreas Gruenbacher <a.gruenbacher@computer.org>");
63 MODULE_DESCRIPTION("Meta block cache (for extended attributes)");
64 MODULE_LICENSE("GPL");
65 
66 EXPORT_SYMBOL(mb_cache_create);
67 EXPORT_SYMBOL(mb_cache_shrink);
68 EXPORT_SYMBOL(mb_cache_destroy);
69 EXPORT_SYMBOL(mb_cache_entry_alloc);
70 EXPORT_SYMBOL(mb_cache_entry_insert);
71 EXPORT_SYMBOL(mb_cache_entry_release);
72 EXPORT_SYMBOL(mb_cache_entry_free);
73 EXPORT_SYMBOL(mb_cache_entry_get);
74 #if !defined(MB_CACHE_INDEXES_COUNT) || (MB_CACHE_INDEXES_COUNT > 0)
75 EXPORT_SYMBOL(mb_cache_entry_find_first);
76 EXPORT_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 
85 static LIST_HEAD(mb_cache_list);
86 static LIST_HEAD(mb_cache_lru_list);
87 static DEFINE_SPINLOCK(mb_cache_spinlock);
88 
89 /*
90  * What the mbcache registers as to get shrunk dynamically.
91  */
92 
93 static int mb_cache_shrink_fn(struct shrinker *shrink,
94 			      struct shrink_control *sc);
95 
96 static struct shrinker mb_cache_shrinker = {
97 	.shrink = mb_cache_shrink_fn,
98 	.seeks = DEFAULT_SEEKS,
99 };
100 
101 static 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 
108 static 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 
118 static 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 
129 static 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;
147 forget:
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  */
164 static int
165 mb_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  */
207 struct mb_cache *
208 mb_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 
248 fail2:
249 	kfree(cache->c_index_hash);
250 
251 fail:
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  */
267 void
268 mb_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  */
297 void
298 mb_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  */
341 struct mb_cache_entry *
342 mb_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  */
385 int
386 mb_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;
411 out:
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  */
424 void
425 mb_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  */
438 void
439 mb_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  */
456 struct mb_cache_entry *
457 mb_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 
496 cleanup:
497 	spin_unlock(&mb_cache_spinlock);
498 	return ce;
499 }
500 
501 #if !defined(MB_CACHE_INDEXES_COUNT) || (MB_CACHE_INDEXES_COUNT > 0)
502 
503 static 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  */
555 struct mb_cache_entry *
556 mb_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  */
589 struct mb_cache_entry *
590 mb_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 
607 static int __init init_mbcache(void)
608 {
609 	register_shrinker(&mb_cache_shrinker);
610 	return 0;
611 }
612 
613 static void __exit exit_mbcache(void)
614 {
615 	unregister_shrinker(&mb_cache_shrinker);
616 }
617 
618 module_init(init_mbcache)
619 module_exit(exit_mbcache)
620 
621