xref: /openbmc/linux/fs/mbcache.c (revision b34e08d5)
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 
77 #ifdef MB_CACHE_DEBUG
78 # define mb_debug(f...) do { \
79 		printk(KERN_DEBUG f); \
80 		printk("\n"); \
81 	} while (0)
82 #define mb_assert(c) do { if (!(c)) \
83 		printk(KERN_ERR "assertion " #c " failed\n"); \
84 	} while(0)
85 #else
86 # define mb_debug(f...) do { } while(0)
87 # define mb_assert(c) do { } while(0)
88 #endif
89 #define mb_error(f...) do { \
90 		printk(KERN_ERR f); \
91 		printk("\n"); \
92 	} while(0)
93 
94 #define MB_CACHE_WRITER ((unsigned short)~0U >> 1)
95 
96 #define MB_CACHE_ENTRY_LOCK_BITS	__builtin_log2(NR_BG_LOCKS)
97 #define	MB_CACHE_ENTRY_LOCK_INDEX(ce)			\
98 	(hash_long((unsigned long)ce, MB_CACHE_ENTRY_LOCK_BITS))
99 
100 static DECLARE_WAIT_QUEUE_HEAD(mb_cache_queue);
101 static struct blockgroup_lock *mb_cache_bg_lock;
102 static struct kmem_cache *mb_cache_kmem_cache;
103 
104 MODULE_AUTHOR("Andreas Gruenbacher <a.gruenbacher@computer.org>");
105 MODULE_DESCRIPTION("Meta block cache (for extended attributes)");
106 MODULE_LICENSE("GPL");
107 
108 EXPORT_SYMBOL(mb_cache_create);
109 EXPORT_SYMBOL(mb_cache_shrink);
110 EXPORT_SYMBOL(mb_cache_destroy);
111 EXPORT_SYMBOL(mb_cache_entry_alloc);
112 EXPORT_SYMBOL(mb_cache_entry_insert);
113 EXPORT_SYMBOL(mb_cache_entry_release);
114 EXPORT_SYMBOL(mb_cache_entry_free);
115 EXPORT_SYMBOL(mb_cache_entry_get);
116 #if !defined(MB_CACHE_INDEXES_COUNT) || (MB_CACHE_INDEXES_COUNT > 0)
117 EXPORT_SYMBOL(mb_cache_entry_find_first);
118 EXPORT_SYMBOL(mb_cache_entry_find_next);
119 #endif
120 
121 /*
122  * Global data: list of all mbcache's, lru list, and a spinlock for
123  * accessing cache data structures on SMP machines. The lru list is
124  * global across all mbcaches.
125  */
126 
127 static LIST_HEAD(mb_cache_list);
128 static LIST_HEAD(mb_cache_lru_list);
129 static DEFINE_SPINLOCK(mb_cache_spinlock);
130 
131 static inline void
132 __spin_lock_mb_cache_entry(struct mb_cache_entry *ce)
133 {
134 	spin_lock(bgl_lock_ptr(mb_cache_bg_lock,
135 		MB_CACHE_ENTRY_LOCK_INDEX(ce)));
136 }
137 
138 static inline void
139 __spin_unlock_mb_cache_entry(struct mb_cache_entry *ce)
140 {
141 	spin_unlock(bgl_lock_ptr(mb_cache_bg_lock,
142 		MB_CACHE_ENTRY_LOCK_INDEX(ce)));
143 }
144 
145 static inline int
146 __mb_cache_entry_is_block_hashed(struct mb_cache_entry *ce)
147 {
148 	return !hlist_bl_unhashed(&ce->e_block_list);
149 }
150 
151 
152 static inline void
153 __mb_cache_entry_unhash_block(struct mb_cache_entry *ce)
154 {
155 	if (__mb_cache_entry_is_block_hashed(ce))
156 		hlist_bl_del_init(&ce->e_block_list);
157 }
158 
159 static inline int
160 __mb_cache_entry_is_index_hashed(struct mb_cache_entry *ce)
161 {
162 	return !hlist_bl_unhashed(&ce->e_index.o_list);
163 }
164 
165 static inline void
166 __mb_cache_entry_unhash_index(struct mb_cache_entry *ce)
167 {
168 	if (__mb_cache_entry_is_index_hashed(ce))
169 		hlist_bl_del_init(&ce->e_index.o_list);
170 }
171 
172 /*
173  * __mb_cache_entry_unhash_unlock()
174  *
175  * This function is called to unhash both the block and index hash
176  * chain.
177  * It assumes both the block and index hash chain is locked upon entry.
178  * It also unlock both hash chains both exit
179  */
180 static inline void
181 __mb_cache_entry_unhash_unlock(struct mb_cache_entry *ce)
182 {
183 	__mb_cache_entry_unhash_index(ce);
184 	hlist_bl_unlock(ce->e_index_hash_p);
185 	__mb_cache_entry_unhash_block(ce);
186 	hlist_bl_unlock(ce->e_block_hash_p);
187 }
188 
189 static void
190 __mb_cache_entry_forget(struct mb_cache_entry *ce, gfp_t gfp_mask)
191 {
192 	struct mb_cache *cache = ce->e_cache;
193 
194 	mb_assert(!(ce->e_used || ce->e_queued || atomic_read(&ce->e_refcnt)));
195 	kmem_cache_free(cache->c_entry_cache, ce);
196 	atomic_dec(&cache->c_entry_count);
197 }
198 
199 static void
200 __mb_cache_entry_release(struct mb_cache_entry *ce)
201 {
202 	/* First lock the entry to serialize access to its local data. */
203 	__spin_lock_mb_cache_entry(ce);
204 	/* Wake up all processes queuing for this cache entry. */
205 	if (ce->e_queued)
206 		wake_up_all(&mb_cache_queue);
207 	if (ce->e_used >= MB_CACHE_WRITER)
208 		ce->e_used -= MB_CACHE_WRITER;
209 	/*
210 	 * Make sure that all cache entries on lru_list have
211 	 * both e_used and e_qued of 0s.
212 	 */
213 	ce->e_used--;
214 	if (!(ce->e_used || ce->e_queued || atomic_read(&ce->e_refcnt))) {
215 		if (!__mb_cache_entry_is_block_hashed(ce)) {
216 			__spin_unlock_mb_cache_entry(ce);
217 			goto forget;
218 		}
219 		/*
220 		 * Need access to lru list, first drop entry lock,
221 		 * then reacquire the lock in the proper order.
222 		 */
223 		spin_lock(&mb_cache_spinlock);
224 		if (list_empty(&ce->e_lru_list))
225 			list_add_tail(&ce->e_lru_list, &mb_cache_lru_list);
226 		spin_unlock(&mb_cache_spinlock);
227 	}
228 	__spin_unlock_mb_cache_entry(ce);
229 	return;
230 forget:
231 	mb_assert(list_empty(&ce->e_lru_list));
232 	__mb_cache_entry_forget(ce, GFP_KERNEL);
233 }
234 
235 /*
236  * mb_cache_shrink_scan()  memory pressure callback
237  *
238  * This function is called by the kernel memory management when memory
239  * gets low.
240  *
241  * @shrink: (ignored)
242  * @sc: shrink_control passed from reclaim
243  *
244  * Returns the number of objects freed.
245  */
246 static unsigned long
247 mb_cache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
248 {
249 	LIST_HEAD(free_list);
250 	struct mb_cache_entry *entry, *tmp;
251 	int nr_to_scan = sc->nr_to_scan;
252 	gfp_t gfp_mask = sc->gfp_mask;
253 	unsigned long freed = 0;
254 
255 	mb_debug("trying to free %d entries", nr_to_scan);
256 	spin_lock(&mb_cache_spinlock);
257 	while ((nr_to_scan-- > 0) && !list_empty(&mb_cache_lru_list)) {
258 		struct mb_cache_entry *ce =
259 			list_entry(mb_cache_lru_list.next,
260 				struct mb_cache_entry, e_lru_list);
261 		list_del_init(&ce->e_lru_list);
262 		if (ce->e_used || ce->e_queued || atomic_read(&ce->e_refcnt))
263 			continue;
264 		spin_unlock(&mb_cache_spinlock);
265 		/* Prevent any find or get operation on the entry */
266 		hlist_bl_lock(ce->e_block_hash_p);
267 		hlist_bl_lock(ce->e_index_hash_p);
268 		/* Ignore if it is touched by a find/get */
269 		if (ce->e_used || ce->e_queued || atomic_read(&ce->e_refcnt) ||
270 			!list_empty(&ce->e_lru_list)) {
271 			hlist_bl_unlock(ce->e_index_hash_p);
272 			hlist_bl_unlock(ce->e_block_hash_p);
273 			spin_lock(&mb_cache_spinlock);
274 			continue;
275 		}
276 		__mb_cache_entry_unhash_unlock(ce);
277 		list_add_tail(&ce->e_lru_list, &free_list);
278 		spin_lock(&mb_cache_spinlock);
279 	}
280 	spin_unlock(&mb_cache_spinlock);
281 
282 	list_for_each_entry_safe(entry, tmp, &free_list, e_lru_list) {
283 		__mb_cache_entry_forget(entry, gfp_mask);
284 		freed++;
285 	}
286 	return freed;
287 }
288 
289 static unsigned long
290 mb_cache_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
291 {
292 	struct mb_cache *cache;
293 	unsigned long count = 0;
294 
295 	spin_lock(&mb_cache_spinlock);
296 	list_for_each_entry(cache, &mb_cache_list, c_cache_list) {
297 		mb_debug("cache %s (%d)", cache->c_name,
298 			  atomic_read(&cache->c_entry_count));
299 		count += atomic_read(&cache->c_entry_count);
300 	}
301 	spin_unlock(&mb_cache_spinlock);
302 
303 	return vfs_pressure_ratio(count);
304 }
305 
306 static struct shrinker mb_cache_shrinker = {
307 	.count_objects = mb_cache_shrink_count,
308 	.scan_objects = mb_cache_shrink_scan,
309 	.seeks = DEFAULT_SEEKS,
310 };
311 
312 /*
313  * mb_cache_create()  create a new cache
314  *
315  * All entries in one cache are equal size. Cache entries may be from
316  * multiple devices. If this is the first mbcache created, registers
317  * the cache with kernel memory management. Returns NULL if no more
318  * memory was available.
319  *
320  * @name: name of the cache (informal)
321  * @bucket_bits: log2(number of hash buckets)
322  */
323 struct mb_cache *
324 mb_cache_create(const char *name, int bucket_bits)
325 {
326 	int n, bucket_count = 1 << bucket_bits;
327 	struct mb_cache *cache = NULL;
328 
329 	if (!mb_cache_bg_lock) {
330 		mb_cache_bg_lock = kmalloc(sizeof(struct blockgroup_lock),
331 			GFP_KERNEL);
332 		if (!mb_cache_bg_lock)
333 			return NULL;
334 		bgl_lock_init(mb_cache_bg_lock);
335 	}
336 
337 	cache = kmalloc(sizeof(struct mb_cache), GFP_KERNEL);
338 	if (!cache)
339 		return NULL;
340 	cache->c_name = name;
341 	atomic_set(&cache->c_entry_count, 0);
342 	cache->c_bucket_bits = bucket_bits;
343 	cache->c_block_hash = kmalloc(bucket_count *
344 		sizeof(struct hlist_bl_head), GFP_KERNEL);
345 	if (!cache->c_block_hash)
346 		goto fail;
347 	for (n=0; n<bucket_count; n++)
348 		INIT_HLIST_BL_HEAD(&cache->c_block_hash[n]);
349 	cache->c_index_hash = kmalloc(bucket_count *
350 		sizeof(struct hlist_bl_head), GFP_KERNEL);
351 	if (!cache->c_index_hash)
352 		goto fail;
353 	for (n=0; n<bucket_count; n++)
354 		INIT_HLIST_BL_HEAD(&cache->c_index_hash[n]);
355 	if (!mb_cache_kmem_cache) {
356 		mb_cache_kmem_cache = kmem_cache_create(name,
357 			sizeof(struct mb_cache_entry), 0,
358 			SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD, NULL);
359 		if (!mb_cache_kmem_cache)
360 			goto fail2;
361 	}
362 	cache->c_entry_cache = mb_cache_kmem_cache;
363 
364 	/*
365 	 * Set an upper limit on the number of cache entries so that the hash
366 	 * chains won't grow too long.
367 	 */
368 	cache->c_max_entries = bucket_count << 4;
369 
370 	spin_lock(&mb_cache_spinlock);
371 	list_add(&cache->c_cache_list, &mb_cache_list);
372 	spin_unlock(&mb_cache_spinlock);
373 	return cache;
374 
375 fail2:
376 	kfree(cache->c_index_hash);
377 
378 fail:
379 	kfree(cache->c_block_hash);
380 	kfree(cache);
381 	return NULL;
382 }
383 
384 
385 /*
386  * mb_cache_shrink()
387  *
388  * Removes all cache entries of a device from the cache. All cache entries
389  * currently in use cannot be freed, and thus remain in the cache. All others
390  * are freed.
391  *
392  * @bdev: which device's cache entries to shrink
393  */
394 void
395 mb_cache_shrink(struct block_device *bdev)
396 {
397 	LIST_HEAD(free_list);
398 	struct list_head *l;
399 	struct mb_cache_entry *ce, *tmp;
400 
401 	l = &mb_cache_lru_list;
402 	spin_lock(&mb_cache_spinlock);
403 	while (!list_is_last(l, &mb_cache_lru_list)) {
404 		l = l->next;
405 		ce = list_entry(l, struct mb_cache_entry, e_lru_list);
406 		if (ce->e_bdev == bdev) {
407 			list_del_init(&ce->e_lru_list);
408 			if (ce->e_used || ce->e_queued ||
409 				atomic_read(&ce->e_refcnt))
410 				continue;
411 			spin_unlock(&mb_cache_spinlock);
412 			/*
413 			 * Prevent any find or get operation on the entry.
414 			 */
415 			hlist_bl_lock(ce->e_block_hash_p);
416 			hlist_bl_lock(ce->e_index_hash_p);
417 			/* Ignore if it is touched by a find/get */
418 			if (ce->e_used || ce->e_queued ||
419 				atomic_read(&ce->e_refcnt) ||
420 				!list_empty(&ce->e_lru_list)) {
421 				hlist_bl_unlock(ce->e_index_hash_p);
422 				hlist_bl_unlock(ce->e_block_hash_p);
423 				l = &mb_cache_lru_list;
424 				spin_lock(&mb_cache_spinlock);
425 				continue;
426 			}
427 			__mb_cache_entry_unhash_unlock(ce);
428 			mb_assert(!(ce->e_used || ce->e_queued ||
429 				atomic_read(&ce->e_refcnt)));
430 			list_add_tail(&ce->e_lru_list, &free_list);
431 			l = &mb_cache_lru_list;
432 			spin_lock(&mb_cache_spinlock);
433 		}
434 	}
435 	spin_unlock(&mb_cache_spinlock);
436 
437 	list_for_each_entry_safe(ce, tmp, &free_list, e_lru_list) {
438 		__mb_cache_entry_forget(ce, GFP_KERNEL);
439 	}
440 }
441 
442 
443 /*
444  * mb_cache_destroy()
445  *
446  * Shrinks the cache to its minimum possible size (hopefully 0 entries),
447  * and then destroys it. If this was the last mbcache, un-registers the
448  * mbcache from kernel memory management.
449  */
450 void
451 mb_cache_destroy(struct mb_cache *cache)
452 {
453 	LIST_HEAD(free_list);
454 	struct mb_cache_entry *ce, *tmp;
455 
456 	spin_lock(&mb_cache_spinlock);
457 	list_for_each_entry_safe(ce, tmp, &mb_cache_lru_list, e_lru_list) {
458 		if (ce->e_cache == cache)
459 			list_move_tail(&ce->e_lru_list, &free_list);
460 	}
461 	list_del(&cache->c_cache_list);
462 	spin_unlock(&mb_cache_spinlock);
463 
464 	list_for_each_entry_safe(ce, tmp, &free_list, e_lru_list) {
465 		list_del_init(&ce->e_lru_list);
466 		/*
467 		 * Prevent any find or get operation on the entry.
468 		 */
469 		hlist_bl_lock(ce->e_block_hash_p);
470 		hlist_bl_lock(ce->e_index_hash_p);
471 		mb_assert(!(ce->e_used || ce->e_queued ||
472 			atomic_read(&ce->e_refcnt)));
473 		__mb_cache_entry_unhash_unlock(ce);
474 		__mb_cache_entry_forget(ce, GFP_KERNEL);
475 	}
476 
477 	if (atomic_read(&cache->c_entry_count) > 0) {
478 		mb_error("cache %s: %d orphaned entries",
479 			  cache->c_name,
480 			  atomic_read(&cache->c_entry_count));
481 	}
482 
483 	if (list_empty(&mb_cache_list)) {
484 		kmem_cache_destroy(mb_cache_kmem_cache);
485 		mb_cache_kmem_cache = NULL;
486 	}
487 	kfree(cache->c_index_hash);
488 	kfree(cache->c_block_hash);
489 	kfree(cache);
490 }
491 
492 /*
493  * mb_cache_entry_alloc()
494  *
495  * Allocates a new cache entry. The new entry will not be valid initially,
496  * and thus cannot be looked up yet. It should be filled with data, and
497  * then inserted into the cache using mb_cache_entry_insert(). Returns NULL
498  * if no more memory was available.
499  */
500 struct mb_cache_entry *
501 mb_cache_entry_alloc(struct mb_cache *cache, gfp_t gfp_flags)
502 {
503 	struct mb_cache_entry *ce;
504 
505 	if (atomic_read(&cache->c_entry_count) >= cache->c_max_entries) {
506 		struct list_head *l;
507 
508 		l = &mb_cache_lru_list;
509 		spin_lock(&mb_cache_spinlock);
510 		while (!list_is_last(l, &mb_cache_lru_list)) {
511 			l = l->next;
512 			ce = list_entry(l, struct mb_cache_entry, e_lru_list);
513 			if (ce->e_cache == cache) {
514 				list_del_init(&ce->e_lru_list);
515 				if (ce->e_used || ce->e_queued ||
516 					atomic_read(&ce->e_refcnt))
517 					continue;
518 				spin_unlock(&mb_cache_spinlock);
519 				/*
520 				 * Prevent any find or get operation on the
521 				 * entry.
522 				 */
523 				hlist_bl_lock(ce->e_block_hash_p);
524 				hlist_bl_lock(ce->e_index_hash_p);
525 				/* Ignore if it is touched by a find/get */
526 				if (ce->e_used || ce->e_queued ||
527 					atomic_read(&ce->e_refcnt) ||
528 					!list_empty(&ce->e_lru_list)) {
529 					hlist_bl_unlock(ce->e_index_hash_p);
530 					hlist_bl_unlock(ce->e_block_hash_p);
531 					l = &mb_cache_lru_list;
532 					spin_lock(&mb_cache_spinlock);
533 					continue;
534 				}
535 				mb_assert(list_empty(&ce->e_lru_list));
536 				mb_assert(!(ce->e_used || ce->e_queued ||
537 					atomic_read(&ce->e_refcnt)));
538 				__mb_cache_entry_unhash_unlock(ce);
539 				goto found;
540 			}
541 		}
542 		spin_unlock(&mb_cache_spinlock);
543 	}
544 
545 	ce = kmem_cache_alloc(cache->c_entry_cache, gfp_flags);
546 	if (!ce)
547 		return NULL;
548 	atomic_inc(&cache->c_entry_count);
549 	INIT_LIST_HEAD(&ce->e_lru_list);
550 	INIT_HLIST_BL_NODE(&ce->e_block_list);
551 	INIT_HLIST_BL_NODE(&ce->e_index.o_list);
552 	ce->e_cache = cache;
553 	ce->e_queued = 0;
554 	atomic_set(&ce->e_refcnt, 0);
555 found:
556 	ce->e_block_hash_p = &cache->c_block_hash[0];
557 	ce->e_index_hash_p = &cache->c_index_hash[0];
558 	ce->e_used = 1 + MB_CACHE_WRITER;
559 	return ce;
560 }
561 
562 
563 /*
564  * mb_cache_entry_insert()
565  *
566  * Inserts an entry that was allocated using mb_cache_entry_alloc() into
567  * the cache. After this, the cache entry can be looked up, but is not yet
568  * in the lru list as the caller still holds a handle to it. Returns 0 on
569  * success, or -EBUSY if a cache entry for that device + inode exists
570  * already (this may happen after a failed lookup, but when another process
571  * has inserted the same cache entry in the meantime).
572  *
573  * @bdev: device the cache entry belongs to
574  * @block: block number
575  * @key: lookup key
576  */
577 int
578 mb_cache_entry_insert(struct mb_cache_entry *ce, struct block_device *bdev,
579 		      sector_t block, unsigned int key)
580 {
581 	struct mb_cache *cache = ce->e_cache;
582 	unsigned int bucket;
583 	struct hlist_bl_node *l;
584 	struct hlist_bl_head *block_hash_p;
585 	struct hlist_bl_head *index_hash_p;
586 	struct mb_cache_entry *lce;
587 
588 	mb_assert(ce);
589 	bucket = hash_long((unsigned long)bdev + (block & 0xffffffff),
590 			   cache->c_bucket_bits);
591 	block_hash_p = &cache->c_block_hash[bucket];
592 	hlist_bl_lock(block_hash_p);
593 	hlist_bl_for_each_entry(lce, l, block_hash_p, e_block_list) {
594 		if (lce->e_bdev == bdev && lce->e_block == block) {
595 			hlist_bl_unlock(block_hash_p);
596 			return -EBUSY;
597 		}
598 	}
599 	mb_assert(!__mb_cache_entry_is_block_hashed(ce));
600 	__mb_cache_entry_unhash_block(ce);
601 	__mb_cache_entry_unhash_index(ce);
602 	ce->e_bdev = bdev;
603 	ce->e_block = block;
604 	ce->e_block_hash_p = block_hash_p;
605 	ce->e_index.o_key = key;
606 	hlist_bl_add_head(&ce->e_block_list, block_hash_p);
607 	hlist_bl_unlock(block_hash_p);
608 	bucket = hash_long(key, cache->c_bucket_bits);
609 	index_hash_p = &cache->c_index_hash[bucket];
610 	hlist_bl_lock(index_hash_p);
611 	ce->e_index_hash_p = index_hash_p;
612 	hlist_bl_add_head(&ce->e_index.o_list, index_hash_p);
613 	hlist_bl_unlock(index_hash_p);
614 	return 0;
615 }
616 
617 
618 /*
619  * mb_cache_entry_release()
620  *
621  * Release a handle to a cache entry. When the last handle to a cache entry
622  * is released it is either freed (if it is invalid) or otherwise inserted
623  * in to the lru list.
624  */
625 void
626 mb_cache_entry_release(struct mb_cache_entry *ce)
627 {
628 	__mb_cache_entry_release(ce);
629 }
630 
631 
632 /*
633  * mb_cache_entry_free()
634  *
635  */
636 void
637 mb_cache_entry_free(struct mb_cache_entry *ce)
638 {
639 	mb_assert(ce);
640 	mb_assert(list_empty(&ce->e_lru_list));
641 	hlist_bl_lock(ce->e_index_hash_p);
642 	__mb_cache_entry_unhash_index(ce);
643 	hlist_bl_unlock(ce->e_index_hash_p);
644 	hlist_bl_lock(ce->e_block_hash_p);
645 	__mb_cache_entry_unhash_block(ce);
646 	hlist_bl_unlock(ce->e_block_hash_p);
647 	__mb_cache_entry_release(ce);
648 }
649 
650 
651 /*
652  * mb_cache_entry_get()
653  *
654  * Get a cache entry  by device / block number. (There can only be one entry
655  * in the cache per device and block.) Returns NULL if no such cache entry
656  * exists. The returned cache entry is locked for exclusive access ("single
657  * writer").
658  */
659 struct mb_cache_entry *
660 mb_cache_entry_get(struct mb_cache *cache, struct block_device *bdev,
661 		   sector_t block)
662 {
663 	unsigned int bucket;
664 	struct hlist_bl_node *l;
665 	struct mb_cache_entry *ce;
666 	struct hlist_bl_head *block_hash_p;
667 
668 	bucket = hash_long((unsigned long)bdev + (block & 0xffffffff),
669 			   cache->c_bucket_bits);
670 	block_hash_p = &cache->c_block_hash[bucket];
671 	/* First serialize access to the block corresponding hash chain. */
672 	hlist_bl_lock(block_hash_p);
673 	hlist_bl_for_each_entry(ce, l, block_hash_p, e_block_list) {
674 		mb_assert(ce->e_block_hash_p == block_hash_p);
675 		if (ce->e_bdev == bdev && ce->e_block == block) {
676 			/*
677 			 * Prevent a free from removing the entry.
678 			 */
679 			atomic_inc(&ce->e_refcnt);
680 			hlist_bl_unlock(block_hash_p);
681 			__spin_lock_mb_cache_entry(ce);
682 			atomic_dec(&ce->e_refcnt);
683 			if (ce->e_used > 0) {
684 				DEFINE_WAIT(wait);
685 				while (ce->e_used > 0) {
686 					ce->e_queued++;
687 					prepare_to_wait(&mb_cache_queue, &wait,
688 							TASK_UNINTERRUPTIBLE);
689 					__spin_unlock_mb_cache_entry(ce);
690 					schedule();
691 					__spin_lock_mb_cache_entry(ce);
692 					ce->e_queued--;
693 				}
694 				finish_wait(&mb_cache_queue, &wait);
695 			}
696 			ce->e_used += 1 + MB_CACHE_WRITER;
697 			__spin_unlock_mb_cache_entry(ce);
698 
699 			if (!list_empty(&ce->e_lru_list)) {
700 				spin_lock(&mb_cache_spinlock);
701 				list_del_init(&ce->e_lru_list);
702 				spin_unlock(&mb_cache_spinlock);
703 			}
704 			if (!__mb_cache_entry_is_block_hashed(ce)) {
705 				__mb_cache_entry_release(ce);
706 				return NULL;
707 			}
708 			return ce;
709 		}
710 	}
711 	hlist_bl_unlock(block_hash_p);
712 	return NULL;
713 }
714 
715 #if !defined(MB_CACHE_INDEXES_COUNT) || (MB_CACHE_INDEXES_COUNT > 0)
716 
717 static struct mb_cache_entry *
718 __mb_cache_entry_find(struct hlist_bl_node *l, struct hlist_bl_head *head,
719 		      struct block_device *bdev, unsigned int key)
720 {
721 
722 	/* The index hash chain is alredy acquire by caller. */
723 	while (l != NULL) {
724 		struct mb_cache_entry *ce =
725 			hlist_bl_entry(l, struct mb_cache_entry,
726 				e_index.o_list);
727 		mb_assert(ce->e_index_hash_p == head);
728 		if (ce->e_bdev == bdev && ce->e_index.o_key == key) {
729 			/*
730 			 * Prevent a free from removing the entry.
731 			 */
732 			atomic_inc(&ce->e_refcnt);
733 			hlist_bl_unlock(head);
734 			__spin_lock_mb_cache_entry(ce);
735 			atomic_dec(&ce->e_refcnt);
736 			ce->e_used++;
737 			/* Incrementing before holding the lock gives readers
738 			   priority over writers. */
739 			if (ce->e_used >= MB_CACHE_WRITER) {
740 				DEFINE_WAIT(wait);
741 
742 				while (ce->e_used >= MB_CACHE_WRITER) {
743 					ce->e_queued++;
744 					prepare_to_wait(&mb_cache_queue, &wait,
745 							TASK_UNINTERRUPTIBLE);
746 					__spin_unlock_mb_cache_entry(ce);
747 					schedule();
748 					__spin_lock_mb_cache_entry(ce);
749 					ce->e_queued--;
750 				}
751 				finish_wait(&mb_cache_queue, &wait);
752 			}
753 			__spin_unlock_mb_cache_entry(ce);
754 			if (!list_empty(&ce->e_lru_list)) {
755 				spin_lock(&mb_cache_spinlock);
756 				list_del_init(&ce->e_lru_list);
757 				spin_unlock(&mb_cache_spinlock);
758 			}
759 			if (!__mb_cache_entry_is_block_hashed(ce)) {
760 				__mb_cache_entry_release(ce);
761 				return ERR_PTR(-EAGAIN);
762 			}
763 			return ce;
764 		}
765 		l = l->next;
766 	}
767 	hlist_bl_unlock(head);
768 	return NULL;
769 }
770 
771 
772 /*
773  * mb_cache_entry_find_first()
774  *
775  * Find the first cache entry on a given device with a certain key in
776  * an additional index. Additional matches can be found with
777  * mb_cache_entry_find_next(). Returns NULL if no match was found. The
778  * returned cache entry is locked for shared access ("multiple readers").
779  *
780  * @cache: the cache to search
781  * @bdev: the device the cache entry should belong to
782  * @key: the key in the index
783  */
784 struct mb_cache_entry *
785 mb_cache_entry_find_first(struct mb_cache *cache, struct block_device *bdev,
786 			  unsigned int key)
787 {
788 	unsigned int bucket = hash_long(key, cache->c_bucket_bits);
789 	struct hlist_bl_node *l;
790 	struct mb_cache_entry *ce = NULL;
791 	struct hlist_bl_head *index_hash_p;
792 
793 	index_hash_p = &cache->c_index_hash[bucket];
794 	hlist_bl_lock(index_hash_p);
795 	if (!hlist_bl_empty(index_hash_p)) {
796 		l = hlist_bl_first(index_hash_p);
797 		ce = __mb_cache_entry_find(l, index_hash_p, bdev, key);
798 	} else
799 		hlist_bl_unlock(index_hash_p);
800 	return ce;
801 }
802 
803 
804 /*
805  * mb_cache_entry_find_next()
806  *
807  * Find the next cache entry on a given device with a certain key in an
808  * additional index. Returns NULL if no match could be found. The previous
809  * entry is atomatically released, so that mb_cache_entry_find_next() can
810  * be called like this:
811  *
812  * entry = mb_cache_entry_find_first();
813  * while (entry) {
814  * 	...
815  *	entry = mb_cache_entry_find_next(entry, ...);
816  * }
817  *
818  * @prev: The previous match
819  * @bdev: the device the cache entry should belong to
820  * @key: the key in the index
821  */
822 struct mb_cache_entry *
823 mb_cache_entry_find_next(struct mb_cache_entry *prev,
824 			 struct block_device *bdev, unsigned int key)
825 {
826 	struct mb_cache *cache = prev->e_cache;
827 	unsigned int bucket = hash_long(key, cache->c_bucket_bits);
828 	struct hlist_bl_node *l;
829 	struct mb_cache_entry *ce;
830 	struct hlist_bl_head *index_hash_p;
831 
832 	index_hash_p = &cache->c_index_hash[bucket];
833 	mb_assert(prev->e_index_hash_p == index_hash_p);
834 	hlist_bl_lock(index_hash_p);
835 	mb_assert(!hlist_bl_empty(index_hash_p));
836 	l = prev->e_index.o_list.next;
837 	ce = __mb_cache_entry_find(l, index_hash_p, bdev, key);
838 	__mb_cache_entry_release(prev);
839 	return ce;
840 }
841 
842 #endif  /* !defined(MB_CACHE_INDEXES_COUNT) || (MB_CACHE_INDEXES_COUNT > 0) */
843 
844 static int __init init_mbcache(void)
845 {
846 	register_shrinker(&mb_cache_shrinker);
847 	return 0;
848 }
849 
850 static void __exit exit_mbcache(void)
851 {
852 	unregister_shrinker(&mb_cache_shrinker);
853 }
854 
855 module_init(init_mbcache)
856 module_exit(exit_mbcache)
857 
858