xref: /openbmc/linux/fs/mbcache.c (revision 3821a065)
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 /*
30  * Lock descriptions and usage:
31  *
32  * Each hash chain of both the block and index hash tables now contains
33  * a built-in lock used to serialize accesses to the hash chain.
34  *
35  * Accesses to global data structures mb_cache_list and mb_cache_lru_list
36  * are serialized via the global spinlock mb_cache_spinlock.
37  *
38  * Each mb_cache_entry contains a spinlock, e_entry_lock, to serialize
39  * accesses to its local data, such as e_used and e_queued.
40  *
41  * Lock ordering:
42  *
43  * Each block hash chain's lock has the highest lock order, followed by an
44  * index hash chain's lock, mb_cache_bg_lock (used to implement mb_cache_entry's
45  * lock), and mb_cach_spinlock, with the lowest order.  While holding
46  * either a block or index hash chain lock, a thread can acquire an
47  * mc_cache_bg_lock, which in turn can also acquire mb_cache_spinlock.
48  *
49  * Synchronization:
50  *
51  * Since both mb_cache_entry_get and mb_cache_entry_find scan the block and
52  * index hash chian, it needs to lock the corresponding hash chain.  For each
53  * mb_cache_entry within the chain, it needs to lock the mb_cache_entry to
54  * prevent either any simultaneous release or free on the entry and also
55  * to serialize accesses to either the e_used or e_queued member of the entry.
56  *
57  * To avoid having a dangling reference to an already freed
58  * mb_cache_entry, an mb_cache_entry is only freed when it is not on a
59  * block hash chain and also no longer being referenced, both e_used,
60  * and e_queued are 0's.  When an mb_cache_entry is explicitly freed it is
61  * first removed from a block hash chain.
62  */
63 
64 #include <linux/kernel.h>
65 #include <linux/module.h>
66 
67 #include <linux/hash.h>
68 #include <linux/fs.h>
69 #include <linux/mm.h>
70 #include <linux/slab.h>
71 #include <linux/sched.h>
72 #include <linux/list_bl.h>
73 #include <linux/mbcache.h>
74 #include <linux/init.h>
75 #include <linux/blockgroup_lock.h>
76 #include <linux/log2.h>
77 
78 #ifdef MB_CACHE_DEBUG
79 # define mb_debug(f...) do { \
80 		printk(KERN_DEBUG f); \
81 		printk("\n"); \
82 	} while (0)
83 #define mb_assert(c) do { if (!(c)) \
84 		printk(KERN_ERR "assertion " #c " failed\n"); \
85 	} while(0)
86 #else
87 # define mb_debug(f...) do { } while(0)
88 # define mb_assert(c) do { } while(0)
89 #endif
90 #define mb_error(f...) do { \
91 		printk(KERN_ERR f); \
92 		printk("\n"); \
93 	} while(0)
94 
95 #define MB_CACHE_WRITER ((unsigned short)~0U >> 1)
96 
97 #define MB_CACHE_ENTRY_LOCK_BITS	ilog2(NR_BG_LOCKS)
98 #define	MB_CACHE_ENTRY_LOCK_INDEX(ce)			\
99 	(hash_long((unsigned long)ce, MB_CACHE_ENTRY_LOCK_BITS))
100 
101 static DECLARE_WAIT_QUEUE_HEAD(mb_cache_queue);
102 static struct blockgroup_lock *mb_cache_bg_lock;
103 static struct kmem_cache *mb_cache_kmem_cache;
104 
105 MODULE_AUTHOR("Andreas Gruenbacher <a.gruenbacher@computer.org>");
106 MODULE_DESCRIPTION("Meta block cache (for extended attributes)");
107 MODULE_LICENSE("GPL");
108 
109 EXPORT_SYMBOL(mb_cache_create);
110 EXPORT_SYMBOL(mb_cache_shrink);
111 EXPORT_SYMBOL(mb_cache_destroy);
112 EXPORT_SYMBOL(mb_cache_entry_alloc);
113 EXPORT_SYMBOL(mb_cache_entry_insert);
114 EXPORT_SYMBOL(mb_cache_entry_release);
115 EXPORT_SYMBOL(mb_cache_entry_free);
116 EXPORT_SYMBOL(mb_cache_entry_get);
117 #if !defined(MB_CACHE_INDEXES_COUNT) || (MB_CACHE_INDEXES_COUNT > 0)
118 EXPORT_SYMBOL(mb_cache_entry_find_first);
119 EXPORT_SYMBOL(mb_cache_entry_find_next);
120 #endif
121 
122 /*
123  * Global data: list of all mbcache's, lru list, and a spinlock for
124  * accessing cache data structures on SMP machines. The lru list is
125  * global across all mbcaches.
126  */
127 
128 static LIST_HEAD(mb_cache_list);
129 static LIST_HEAD(mb_cache_lru_list);
130 static DEFINE_SPINLOCK(mb_cache_spinlock);
131 
132 static inline void
133 __spin_lock_mb_cache_entry(struct mb_cache_entry *ce)
134 {
135 	spin_lock(bgl_lock_ptr(mb_cache_bg_lock,
136 		MB_CACHE_ENTRY_LOCK_INDEX(ce)));
137 }
138 
139 static inline void
140 __spin_unlock_mb_cache_entry(struct mb_cache_entry *ce)
141 {
142 	spin_unlock(bgl_lock_ptr(mb_cache_bg_lock,
143 		MB_CACHE_ENTRY_LOCK_INDEX(ce)));
144 }
145 
146 static inline int
147 __mb_cache_entry_is_block_hashed(struct mb_cache_entry *ce)
148 {
149 	return !hlist_bl_unhashed(&ce->e_block_list);
150 }
151 
152 
153 static inline void
154 __mb_cache_entry_unhash_block(struct mb_cache_entry *ce)
155 {
156 	if (__mb_cache_entry_is_block_hashed(ce))
157 		hlist_bl_del_init(&ce->e_block_list);
158 }
159 
160 static inline int
161 __mb_cache_entry_is_index_hashed(struct mb_cache_entry *ce)
162 {
163 	return !hlist_bl_unhashed(&ce->e_index.o_list);
164 }
165 
166 static inline void
167 __mb_cache_entry_unhash_index(struct mb_cache_entry *ce)
168 {
169 	if (__mb_cache_entry_is_index_hashed(ce))
170 		hlist_bl_del_init(&ce->e_index.o_list);
171 }
172 
173 /*
174  * __mb_cache_entry_unhash_unlock()
175  *
176  * This function is called to unhash both the block and index hash
177  * chain.
178  * It assumes both the block and index hash chain is locked upon entry.
179  * It also unlock both hash chains both exit
180  */
181 static inline void
182 __mb_cache_entry_unhash_unlock(struct mb_cache_entry *ce)
183 {
184 	__mb_cache_entry_unhash_index(ce);
185 	hlist_bl_unlock(ce->e_index_hash_p);
186 	__mb_cache_entry_unhash_block(ce);
187 	hlist_bl_unlock(ce->e_block_hash_p);
188 }
189 
190 static void
191 __mb_cache_entry_forget(struct mb_cache_entry *ce, gfp_t gfp_mask)
192 {
193 	struct mb_cache *cache = ce->e_cache;
194 
195 	mb_assert(!(ce->e_used || ce->e_queued || atomic_read(&ce->e_refcnt)));
196 	kmem_cache_free(cache->c_entry_cache, ce);
197 	atomic_dec(&cache->c_entry_count);
198 }
199 
200 static void
201 __mb_cache_entry_release(struct mb_cache_entry *ce)
202 {
203 	/* First lock the entry to serialize access to its local data. */
204 	__spin_lock_mb_cache_entry(ce);
205 	/* Wake up all processes queuing for this cache entry. */
206 	if (ce->e_queued)
207 		wake_up_all(&mb_cache_queue);
208 	if (ce->e_used >= MB_CACHE_WRITER)
209 		ce->e_used -= MB_CACHE_WRITER;
210 	/*
211 	 * Make sure that all cache entries on lru_list have
212 	 * both e_used and e_qued of 0s.
213 	 */
214 	ce->e_used--;
215 	if (!(ce->e_used || ce->e_queued || atomic_read(&ce->e_refcnt))) {
216 		if (!__mb_cache_entry_is_block_hashed(ce)) {
217 			__spin_unlock_mb_cache_entry(ce);
218 			goto forget;
219 		}
220 		/*
221 		 * Need access to lru list, first drop entry lock,
222 		 * then reacquire the lock in the proper order.
223 		 */
224 		spin_lock(&mb_cache_spinlock);
225 		if (list_empty(&ce->e_lru_list))
226 			list_add_tail(&ce->e_lru_list, &mb_cache_lru_list);
227 		spin_unlock(&mb_cache_spinlock);
228 	}
229 	__spin_unlock_mb_cache_entry(ce);
230 	return;
231 forget:
232 	mb_assert(list_empty(&ce->e_lru_list));
233 	__mb_cache_entry_forget(ce, GFP_KERNEL);
234 }
235 
236 /*
237  * mb_cache_shrink_scan()  memory pressure callback
238  *
239  * This function is called by the kernel memory management when memory
240  * gets low.
241  *
242  * @shrink: (ignored)
243  * @sc: shrink_control passed from reclaim
244  *
245  * Returns the number of objects freed.
246  */
247 static unsigned long
248 mb_cache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
249 {
250 	LIST_HEAD(free_list);
251 	struct mb_cache_entry *entry, *tmp;
252 	int nr_to_scan = sc->nr_to_scan;
253 	gfp_t gfp_mask = sc->gfp_mask;
254 	unsigned long freed = 0;
255 
256 	mb_debug("trying to free %d entries", nr_to_scan);
257 	spin_lock(&mb_cache_spinlock);
258 	while ((nr_to_scan-- > 0) && !list_empty(&mb_cache_lru_list)) {
259 		struct mb_cache_entry *ce =
260 			list_entry(mb_cache_lru_list.next,
261 				struct mb_cache_entry, e_lru_list);
262 		list_del_init(&ce->e_lru_list);
263 		if (ce->e_used || ce->e_queued || atomic_read(&ce->e_refcnt))
264 			continue;
265 		spin_unlock(&mb_cache_spinlock);
266 		/* Prevent any find or get operation on the entry */
267 		hlist_bl_lock(ce->e_block_hash_p);
268 		hlist_bl_lock(ce->e_index_hash_p);
269 		/* Ignore if it is touched by a find/get */
270 		if (ce->e_used || ce->e_queued || atomic_read(&ce->e_refcnt) ||
271 			!list_empty(&ce->e_lru_list)) {
272 			hlist_bl_unlock(ce->e_index_hash_p);
273 			hlist_bl_unlock(ce->e_block_hash_p);
274 			spin_lock(&mb_cache_spinlock);
275 			continue;
276 		}
277 		__mb_cache_entry_unhash_unlock(ce);
278 		list_add_tail(&ce->e_lru_list, &free_list);
279 		spin_lock(&mb_cache_spinlock);
280 	}
281 	spin_unlock(&mb_cache_spinlock);
282 
283 	list_for_each_entry_safe(entry, tmp, &free_list, e_lru_list) {
284 		__mb_cache_entry_forget(entry, gfp_mask);
285 		freed++;
286 	}
287 	return freed;
288 }
289 
290 static unsigned long
291 mb_cache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
292 {
293 	struct mb_cache *cache;
294 	unsigned long count = 0;
295 
296 	spin_lock(&mb_cache_spinlock);
297 	list_for_each_entry(cache, &mb_cache_list, c_cache_list) {
298 		mb_debug("cache %s (%d)", cache->c_name,
299 			  atomic_read(&cache->c_entry_count));
300 		count += atomic_read(&cache->c_entry_count);
301 	}
302 	spin_unlock(&mb_cache_spinlock);
303 
304 	return vfs_pressure_ratio(count);
305 }
306 
307 static struct shrinker mb_cache_shrinker = {
308 	.count_objects = mb_cache_shrink_count,
309 	.scan_objects = mb_cache_shrink_scan,
310 	.seeks = DEFAULT_SEEKS,
311 };
312 
313 /*
314  * mb_cache_create()  create a new cache
315  *
316  * All entries in one cache are equal size. Cache entries may be from
317  * multiple devices. If this is the first mbcache created, registers
318  * the cache with kernel memory management. Returns NULL if no more
319  * memory was available.
320  *
321  * @name: name of the cache (informal)
322  * @bucket_bits: log2(number of hash buckets)
323  */
324 struct mb_cache *
325 mb_cache_create(const char *name, int bucket_bits)
326 {
327 	int n, bucket_count = 1 << bucket_bits;
328 	struct mb_cache *cache = NULL;
329 
330 	if (!mb_cache_bg_lock) {
331 		mb_cache_bg_lock = kmalloc(sizeof(struct blockgroup_lock),
332 			GFP_KERNEL);
333 		if (!mb_cache_bg_lock)
334 			return NULL;
335 		bgl_lock_init(mb_cache_bg_lock);
336 	}
337 
338 	cache = kmalloc(sizeof(struct mb_cache), GFP_KERNEL);
339 	if (!cache)
340 		return NULL;
341 	cache->c_name = name;
342 	atomic_set(&cache->c_entry_count, 0);
343 	cache->c_bucket_bits = bucket_bits;
344 	cache->c_block_hash = kmalloc(bucket_count *
345 		sizeof(struct hlist_bl_head), GFP_KERNEL);
346 	if (!cache->c_block_hash)
347 		goto fail;
348 	for (n=0; n<bucket_count; n++)
349 		INIT_HLIST_BL_HEAD(&cache->c_block_hash[n]);
350 	cache->c_index_hash = kmalloc(bucket_count *
351 		sizeof(struct hlist_bl_head), GFP_KERNEL);
352 	if (!cache->c_index_hash)
353 		goto fail;
354 	for (n=0; n<bucket_count; n++)
355 		INIT_HLIST_BL_HEAD(&cache->c_index_hash[n]);
356 	if (!mb_cache_kmem_cache) {
357 		mb_cache_kmem_cache = kmem_cache_create(name,
358 			sizeof(struct mb_cache_entry), 0,
359 			SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD, NULL);
360 		if (!mb_cache_kmem_cache)
361 			goto fail2;
362 	}
363 	cache->c_entry_cache = mb_cache_kmem_cache;
364 
365 	/*
366 	 * Set an upper limit on the number of cache entries so that the hash
367 	 * chains won't grow too long.
368 	 */
369 	cache->c_max_entries = bucket_count << 4;
370 
371 	spin_lock(&mb_cache_spinlock);
372 	list_add(&cache->c_cache_list, &mb_cache_list);
373 	spin_unlock(&mb_cache_spinlock);
374 	return cache;
375 
376 fail2:
377 	kfree(cache->c_index_hash);
378 
379 fail:
380 	kfree(cache->c_block_hash);
381 	kfree(cache);
382 	return NULL;
383 }
384 
385 
386 /*
387  * mb_cache_shrink()
388  *
389  * Removes all cache entries of a device from the cache. All cache entries
390  * currently in use cannot be freed, and thus remain in the cache. All others
391  * are freed.
392  *
393  * @bdev: which device's cache entries to shrink
394  */
395 void
396 mb_cache_shrink(struct block_device *bdev)
397 {
398 	LIST_HEAD(free_list);
399 	struct list_head *l;
400 	struct mb_cache_entry *ce, *tmp;
401 
402 	l = &mb_cache_lru_list;
403 	spin_lock(&mb_cache_spinlock);
404 	while (!list_is_last(l, &mb_cache_lru_list)) {
405 		l = l->next;
406 		ce = list_entry(l, struct mb_cache_entry, e_lru_list);
407 		if (ce->e_bdev == bdev) {
408 			list_del_init(&ce->e_lru_list);
409 			if (ce->e_used || ce->e_queued ||
410 				atomic_read(&ce->e_refcnt))
411 				continue;
412 			spin_unlock(&mb_cache_spinlock);
413 			/*
414 			 * Prevent any find or get operation on the entry.
415 			 */
416 			hlist_bl_lock(ce->e_block_hash_p);
417 			hlist_bl_lock(ce->e_index_hash_p);
418 			/* Ignore if it is touched by a find/get */
419 			if (ce->e_used || ce->e_queued ||
420 				atomic_read(&ce->e_refcnt) ||
421 				!list_empty(&ce->e_lru_list)) {
422 				hlist_bl_unlock(ce->e_index_hash_p);
423 				hlist_bl_unlock(ce->e_block_hash_p);
424 				l = &mb_cache_lru_list;
425 				spin_lock(&mb_cache_spinlock);
426 				continue;
427 			}
428 			__mb_cache_entry_unhash_unlock(ce);
429 			mb_assert(!(ce->e_used || ce->e_queued ||
430 				atomic_read(&ce->e_refcnt)));
431 			list_add_tail(&ce->e_lru_list, &free_list);
432 			l = &mb_cache_lru_list;
433 			spin_lock(&mb_cache_spinlock);
434 		}
435 	}
436 	spin_unlock(&mb_cache_spinlock);
437 
438 	list_for_each_entry_safe(ce, tmp, &free_list, e_lru_list) {
439 		__mb_cache_entry_forget(ce, GFP_KERNEL);
440 	}
441 }
442 
443 
444 /*
445  * mb_cache_destroy()
446  *
447  * Shrinks the cache to its minimum possible size (hopefully 0 entries),
448  * and then destroys it. If this was the last mbcache, un-registers the
449  * mbcache from kernel memory management.
450  */
451 void
452 mb_cache_destroy(struct mb_cache *cache)
453 {
454 	LIST_HEAD(free_list);
455 	struct mb_cache_entry *ce, *tmp;
456 
457 	spin_lock(&mb_cache_spinlock);
458 	list_for_each_entry_safe(ce, tmp, &mb_cache_lru_list, e_lru_list) {
459 		if (ce->e_cache == cache)
460 			list_move_tail(&ce->e_lru_list, &free_list);
461 	}
462 	list_del(&cache->c_cache_list);
463 	spin_unlock(&mb_cache_spinlock);
464 
465 	list_for_each_entry_safe(ce, tmp, &free_list, e_lru_list) {
466 		list_del_init(&ce->e_lru_list);
467 		/*
468 		 * Prevent any find or get operation on the entry.
469 		 */
470 		hlist_bl_lock(ce->e_block_hash_p);
471 		hlist_bl_lock(ce->e_index_hash_p);
472 		mb_assert(!(ce->e_used || ce->e_queued ||
473 			atomic_read(&ce->e_refcnt)));
474 		__mb_cache_entry_unhash_unlock(ce);
475 		__mb_cache_entry_forget(ce, GFP_KERNEL);
476 	}
477 
478 	if (atomic_read(&cache->c_entry_count) > 0) {
479 		mb_error("cache %s: %d orphaned entries",
480 			  cache->c_name,
481 			  atomic_read(&cache->c_entry_count));
482 	}
483 
484 	if (list_empty(&mb_cache_list)) {
485 		kmem_cache_destroy(mb_cache_kmem_cache);
486 		mb_cache_kmem_cache = NULL;
487 	}
488 	kfree(cache->c_index_hash);
489 	kfree(cache->c_block_hash);
490 	kfree(cache);
491 }
492 
493 /*
494  * mb_cache_entry_alloc()
495  *
496  * Allocates a new cache entry. The new entry will not be valid initially,
497  * and thus cannot be looked up yet. It should be filled with data, and
498  * then inserted into the cache using mb_cache_entry_insert(). Returns NULL
499  * if no more memory was available.
500  */
501 struct mb_cache_entry *
502 mb_cache_entry_alloc(struct mb_cache *cache, gfp_t gfp_flags)
503 {
504 	struct mb_cache_entry *ce;
505 
506 	if (atomic_read(&cache->c_entry_count) >= cache->c_max_entries) {
507 		struct list_head *l;
508 
509 		l = &mb_cache_lru_list;
510 		spin_lock(&mb_cache_spinlock);
511 		while (!list_is_last(l, &mb_cache_lru_list)) {
512 			l = l->next;
513 			ce = list_entry(l, struct mb_cache_entry, e_lru_list);
514 			if (ce->e_cache == cache) {
515 				list_del_init(&ce->e_lru_list);
516 				if (ce->e_used || ce->e_queued ||
517 					atomic_read(&ce->e_refcnt))
518 					continue;
519 				spin_unlock(&mb_cache_spinlock);
520 				/*
521 				 * Prevent any find or get operation on the
522 				 * entry.
523 				 */
524 				hlist_bl_lock(ce->e_block_hash_p);
525 				hlist_bl_lock(ce->e_index_hash_p);
526 				/* Ignore if it is touched by a find/get */
527 				if (ce->e_used || ce->e_queued ||
528 					atomic_read(&ce->e_refcnt) ||
529 					!list_empty(&ce->e_lru_list)) {
530 					hlist_bl_unlock(ce->e_index_hash_p);
531 					hlist_bl_unlock(ce->e_block_hash_p);
532 					l = &mb_cache_lru_list;
533 					spin_lock(&mb_cache_spinlock);
534 					continue;
535 				}
536 				mb_assert(list_empty(&ce->e_lru_list));
537 				mb_assert(!(ce->e_used || ce->e_queued ||
538 					atomic_read(&ce->e_refcnt)));
539 				__mb_cache_entry_unhash_unlock(ce);
540 				goto found;
541 			}
542 		}
543 		spin_unlock(&mb_cache_spinlock);
544 	}
545 
546 	ce = kmem_cache_alloc(cache->c_entry_cache, gfp_flags);
547 	if (!ce)
548 		return NULL;
549 	atomic_inc(&cache->c_entry_count);
550 	INIT_LIST_HEAD(&ce->e_lru_list);
551 	INIT_HLIST_BL_NODE(&ce->e_block_list);
552 	INIT_HLIST_BL_NODE(&ce->e_index.o_list);
553 	ce->e_cache = cache;
554 	ce->e_queued = 0;
555 	atomic_set(&ce->e_refcnt, 0);
556 found:
557 	ce->e_block_hash_p = &cache->c_block_hash[0];
558 	ce->e_index_hash_p = &cache->c_index_hash[0];
559 	ce->e_used = 1 + MB_CACHE_WRITER;
560 	return ce;
561 }
562 
563 
564 /*
565  * mb_cache_entry_insert()
566  *
567  * Inserts an entry that was allocated using mb_cache_entry_alloc() into
568  * the cache. After this, the cache entry can be looked up, but is not yet
569  * in the lru list as the caller still holds a handle to it. Returns 0 on
570  * success, or -EBUSY if a cache entry for that device + inode exists
571  * already (this may happen after a failed lookup, but when another process
572  * has inserted the same cache entry in the meantime).
573  *
574  * @bdev: device the cache entry belongs to
575  * @block: block number
576  * @key: lookup key
577  */
578 int
579 mb_cache_entry_insert(struct mb_cache_entry *ce, struct block_device *bdev,
580 		      sector_t block, unsigned int key)
581 {
582 	struct mb_cache *cache = ce->e_cache;
583 	unsigned int bucket;
584 	struct hlist_bl_node *l;
585 	struct hlist_bl_head *block_hash_p;
586 	struct hlist_bl_head *index_hash_p;
587 	struct mb_cache_entry *lce;
588 
589 	mb_assert(ce);
590 	bucket = hash_long((unsigned long)bdev + (block & 0xffffffff),
591 			   cache->c_bucket_bits);
592 	block_hash_p = &cache->c_block_hash[bucket];
593 	hlist_bl_lock(block_hash_p);
594 	hlist_bl_for_each_entry(lce, l, block_hash_p, e_block_list) {
595 		if (lce->e_bdev == bdev && lce->e_block == block) {
596 			hlist_bl_unlock(block_hash_p);
597 			return -EBUSY;
598 		}
599 	}
600 	mb_assert(!__mb_cache_entry_is_block_hashed(ce));
601 	__mb_cache_entry_unhash_block(ce);
602 	__mb_cache_entry_unhash_index(ce);
603 	ce->e_bdev = bdev;
604 	ce->e_block = block;
605 	ce->e_block_hash_p = block_hash_p;
606 	ce->e_index.o_key = key;
607 	hlist_bl_add_head(&ce->e_block_list, block_hash_p);
608 	hlist_bl_unlock(block_hash_p);
609 	bucket = hash_long(key, cache->c_bucket_bits);
610 	index_hash_p = &cache->c_index_hash[bucket];
611 	hlist_bl_lock(index_hash_p);
612 	ce->e_index_hash_p = index_hash_p;
613 	hlist_bl_add_head(&ce->e_index.o_list, index_hash_p);
614 	hlist_bl_unlock(index_hash_p);
615 	return 0;
616 }
617 
618 
619 /*
620  * mb_cache_entry_release()
621  *
622  * Release a handle to a cache entry. When the last handle to a cache entry
623  * is released it is either freed (if it is invalid) or otherwise inserted
624  * in to the lru list.
625  */
626 void
627 mb_cache_entry_release(struct mb_cache_entry *ce)
628 {
629 	__mb_cache_entry_release(ce);
630 }
631 
632 
633 /*
634  * mb_cache_entry_free()
635  *
636  */
637 void
638 mb_cache_entry_free(struct mb_cache_entry *ce)
639 {
640 	mb_assert(ce);
641 	mb_assert(list_empty(&ce->e_lru_list));
642 	hlist_bl_lock(ce->e_index_hash_p);
643 	__mb_cache_entry_unhash_index(ce);
644 	hlist_bl_unlock(ce->e_index_hash_p);
645 	hlist_bl_lock(ce->e_block_hash_p);
646 	__mb_cache_entry_unhash_block(ce);
647 	hlist_bl_unlock(ce->e_block_hash_p);
648 	__mb_cache_entry_release(ce);
649 }
650 
651 
652 /*
653  * mb_cache_entry_get()
654  *
655  * Get a cache entry  by device / block number. (There can only be one entry
656  * in the cache per device and block.) Returns NULL if no such cache entry
657  * exists. The returned cache entry is locked for exclusive access ("single
658  * writer").
659  */
660 struct mb_cache_entry *
661 mb_cache_entry_get(struct mb_cache *cache, struct block_device *bdev,
662 		   sector_t block)
663 {
664 	unsigned int bucket;
665 	struct hlist_bl_node *l;
666 	struct mb_cache_entry *ce;
667 	struct hlist_bl_head *block_hash_p;
668 
669 	bucket = hash_long((unsigned long)bdev + (block & 0xffffffff),
670 			   cache->c_bucket_bits);
671 	block_hash_p = &cache->c_block_hash[bucket];
672 	/* First serialize access to the block corresponding hash chain. */
673 	hlist_bl_lock(block_hash_p);
674 	hlist_bl_for_each_entry(ce, l, block_hash_p, e_block_list) {
675 		mb_assert(ce->e_block_hash_p == block_hash_p);
676 		if (ce->e_bdev == bdev && ce->e_block == block) {
677 			/*
678 			 * Prevent a free from removing the entry.
679 			 */
680 			atomic_inc(&ce->e_refcnt);
681 			hlist_bl_unlock(block_hash_p);
682 			__spin_lock_mb_cache_entry(ce);
683 			atomic_dec(&ce->e_refcnt);
684 			if (ce->e_used > 0) {
685 				DEFINE_WAIT(wait);
686 				while (ce->e_used > 0) {
687 					ce->e_queued++;
688 					prepare_to_wait(&mb_cache_queue, &wait,
689 							TASK_UNINTERRUPTIBLE);
690 					__spin_unlock_mb_cache_entry(ce);
691 					schedule();
692 					__spin_lock_mb_cache_entry(ce);
693 					ce->e_queued--;
694 				}
695 				finish_wait(&mb_cache_queue, &wait);
696 			}
697 			ce->e_used += 1 + MB_CACHE_WRITER;
698 			__spin_unlock_mb_cache_entry(ce);
699 
700 			if (!list_empty(&ce->e_lru_list)) {
701 				spin_lock(&mb_cache_spinlock);
702 				list_del_init(&ce->e_lru_list);
703 				spin_unlock(&mb_cache_spinlock);
704 			}
705 			if (!__mb_cache_entry_is_block_hashed(ce)) {
706 				__mb_cache_entry_release(ce);
707 				return NULL;
708 			}
709 			return ce;
710 		}
711 	}
712 	hlist_bl_unlock(block_hash_p);
713 	return NULL;
714 }
715 
716 #if !defined(MB_CACHE_INDEXES_COUNT) || (MB_CACHE_INDEXES_COUNT > 0)
717 
718 static struct mb_cache_entry *
719 __mb_cache_entry_find(struct hlist_bl_node *l, struct hlist_bl_head *head,
720 		      struct block_device *bdev, unsigned int key)
721 {
722 
723 	/* The index hash chain is alredy acquire by caller. */
724 	while (l != NULL) {
725 		struct mb_cache_entry *ce =
726 			hlist_bl_entry(l, struct mb_cache_entry,
727 				e_index.o_list);
728 		mb_assert(ce->e_index_hash_p == head);
729 		if (ce->e_bdev == bdev && ce->e_index.o_key == key) {
730 			/*
731 			 * Prevent a free from removing the entry.
732 			 */
733 			atomic_inc(&ce->e_refcnt);
734 			hlist_bl_unlock(head);
735 			__spin_lock_mb_cache_entry(ce);
736 			atomic_dec(&ce->e_refcnt);
737 			ce->e_used++;
738 			/* Incrementing before holding the lock gives readers
739 			   priority over writers. */
740 			if (ce->e_used >= MB_CACHE_WRITER) {
741 				DEFINE_WAIT(wait);
742 
743 				while (ce->e_used >= MB_CACHE_WRITER) {
744 					ce->e_queued++;
745 					prepare_to_wait(&mb_cache_queue, &wait,
746 							TASK_UNINTERRUPTIBLE);
747 					__spin_unlock_mb_cache_entry(ce);
748 					schedule();
749 					__spin_lock_mb_cache_entry(ce);
750 					ce->e_queued--;
751 				}
752 				finish_wait(&mb_cache_queue, &wait);
753 			}
754 			__spin_unlock_mb_cache_entry(ce);
755 			if (!list_empty(&ce->e_lru_list)) {
756 				spin_lock(&mb_cache_spinlock);
757 				list_del_init(&ce->e_lru_list);
758 				spin_unlock(&mb_cache_spinlock);
759 			}
760 			if (!__mb_cache_entry_is_block_hashed(ce)) {
761 				__mb_cache_entry_release(ce);
762 				return ERR_PTR(-EAGAIN);
763 			}
764 			return ce;
765 		}
766 		l = l->next;
767 	}
768 	hlist_bl_unlock(head);
769 	return NULL;
770 }
771 
772 
773 /*
774  * mb_cache_entry_find_first()
775  *
776  * Find the first cache entry on a given device with a certain key in
777  * an additional index. Additional matches can be found with
778  * mb_cache_entry_find_next(). Returns NULL if no match was found. The
779  * returned cache entry is locked for shared access ("multiple readers").
780  *
781  * @cache: the cache to search
782  * @bdev: the device the cache entry should belong to
783  * @key: the key in the index
784  */
785 struct mb_cache_entry *
786 mb_cache_entry_find_first(struct mb_cache *cache, struct block_device *bdev,
787 			  unsigned int key)
788 {
789 	unsigned int bucket = hash_long(key, cache->c_bucket_bits);
790 	struct hlist_bl_node *l;
791 	struct mb_cache_entry *ce = NULL;
792 	struct hlist_bl_head *index_hash_p;
793 
794 	index_hash_p = &cache->c_index_hash[bucket];
795 	hlist_bl_lock(index_hash_p);
796 	if (!hlist_bl_empty(index_hash_p)) {
797 		l = hlist_bl_first(index_hash_p);
798 		ce = __mb_cache_entry_find(l, index_hash_p, bdev, key);
799 	} else
800 		hlist_bl_unlock(index_hash_p);
801 	return ce;
802 }
803 
804 
805 /*
806  * mb_cache_entry_find_next()
807  *
808  * Find the next cache entry on a given device with a certain key in an
809  * additional index. Returns NULL if no match could be found. The previous
810  * entry is atomatically released, so that mb_cache_entry_find_next() can
811  * be called like this:
812  *
813  * entry = mb_cache_entry_find_first();
814  * while (entry) {
815  * 	...
816  *	entry = mb_cache_entry_find_next(entry, ...);
817  * }
818  *
819  * @prev: The previous match
820  * @bdev: the device the cache entry should belong to
821  * @key: the key in the index
822  */
823 struct mb_cache_entry *
824 mb_cache_entry_find_next(struct mb_cache_entry *prev,
825 			 struct block_device *bdev, unsigned int key)
826 {
827 	struct mb_cache *cache = prev->e_cache;
828 	unsigned int bucket = hash_long(key, cache->c_bucket_bits);
829 	struct hlist_bl_node *l;
830 	struct mb_cache_entry *ce;
831 	struct hlist_bl_head *index_hash_p;
832 
833 	index_hash_p = &cache->c_index_hash[bucket];
834 	mb_assert(prev->e_index_hash_p == index_hash_p);
835 	hlist_bl_lock(index_hash_p);
836 	mb_assert(!hlist_bl_empty(index_hash_p));
837 	l = prev->e_index.o_list.next;
838 	ce = __mb_cache_entry_find(l, index_hash_p, bdev, key);
839 	__mb_cache_entry_release(prev);
840 	return ce;
841 }
842 
843 #endif  /* !defined(MB_CACHE_INDEXES_COUNT) || (MB_CACHE_INDEXES_COUNT > 0) */
844 
845 static int __init init_mbcache(void)
846 {
847 	register_shrinker(&mb_cache_shrinker);
848 	return 0;
849 }
850 
851 static void __exit exit_mbcache(void)
852 {
853 	unregister_shrinker(&mb_cache_shrinker);
854 }
855 
856 module_init(init_mbcache)
857 module_exit(exit_mbcache)
858 
859