4 * (C) 1997 Linus Torvalds
7 #include <linux/config.h>
10 #include <linux/dcache.h>
11 #include <linux/init.h>
12 #include <linux/quotaops.h>
13 #include <linux/slab.h>
14 #include <linux/writeback.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/wait.h>
18 #include <linux/hash.h>
19 #include <linux/swap.h>
20 #include <linux/security.h>
21 #include <linux/pagemap.h>
22 #include <linux/cdev.h>
23 #include <linux/bootmem.h>
26 * This is needed for the following functions:
28 * - invalidate_inode_buffers
31 * FIXME: remove all knowledge of the buffer layer from this file
33 #include <linux/buffer_head.h>
36 * New inode.c implementation.
38 * This implementation has the basic premise of trying
39 * to be extremely low-overhead and SMP-safe, yet be
40 * simple enough to be "obviously correct".
45 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
47 /* #define INODE_PARANOIA 1 */
48 /* #define INODE_DEBUG 1 */
51 * Inode lookup is no longer as critical as it used to be:
52 * most of the lookups are going to be through the dcache.
54 #define I_HASHBITS i_hash_shift
55 #define I_HASHMASK i_hash_mask
57 static unsigned int i_hash_mask
;
58 static unsigned int i_hash_shift
;
61 * Each inode can be on two separate lists. One is
62 * the hash list of the inode, used for lookups. The
63 * other linked list is the "type" list:
64 * "in_use" - valid inode, i_count > 0, i_nlink > 0
65 * "dirty" - as "in_use" but also dirty
66 * "unused" - valid inode, i_count = 0
68 * A "dirty" list is maintained for each super block,
69 * allowing for low-overhead inode sync() operations.
72 LIST_HEAD(inode_in_use
);
73 LIST_HEAD(inode_unused
);
74 static struct hlist_head
*inode_hashtable
;
77 * A simple spinlock to protect the list manipulations.
79 * NOTE! You also have to own the lock if you change
80 * the i_state of an inode while it is in use..
82 DEFINE_SPINLOCK(inode_lock
);
85 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
86 * icache shrinking path, and the umount path. Without this exclusion,
87 * by the time prune_icache calls iput for the inode whose pages it has
88 * been invalidating, or by the time it calls clear_inode & destroy_inode
89 * from its final dispose_list, the struct super_block they refer to
90 * (for inode->i_sb->s_op) may already have been freed and reused.
92 DECLARE_MUTEX(iprune_sem
);
95 * Statistics gathering..
97 struct inodes_stat_t inodes_stat
;
99 static kmem_cache_t
* inode_cachep
;
101 static struct inode
*alloc_inode(struct super_block
*sb
)
103 static struct address_space_operations empty_aops
;
104 static struct inode_operations empty_iops
;
105 static struct file_operations empty_fops
;
108 if (sb
->s_op
->alloc_inode
)
109 inode
= sb
->s_op
->alloc_inode(sb
);
111 inode
= (struct inode
*) kmem_cache_alloc(inode_cachep
, SLAB_KERNEL
);
114 struct address_space
* const mapping
= &inode
->i_data
;
117 inode
->i_blkbits
= sb
->s_blocksize_bits
;
119 atomic_set(&inode
->i_count
, 1);
120 inode
->i_op
= &empty_iops
;
121 inode
->i_fop
= &empty_fops
;
123 atomic_set(&inode
->i_writecount
, 0);
127 inode
->i_generation
= 0;
129 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
131 inode
->i_pipe
= NULL
;
132 inode
->i_bdev
= NULL
;
133 inode
->i_cdev
= NULL
;
135 inode
->i_security
= NULL
;
136 inode
->dirtied_when
= 0;
137 if (security_inode_alloc(inode
)) {
138 if (inode
->i_sb
->s_op
->destroy_inode
)
139 inode
->i_sb
->s_op
->destroy_inode(inode
);
141 kmem_cache_free(inode_cachep
, (inode
));
145 mapping
->a_ops
= &empty_aops
;
146 mapping
->host
= inode
;
148 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER
);
149 mapping
->assoc_mapping
= NULL
;
150 mapping
->backing_dev_info
= &default_backing_dev_info
;
153 * If the block_device provides a backing_dev_info for client
154 * inodes then use that. Otherwise the inode share the bdev's
158 struct backing_dev_info
*bdi
;
160 bdi
= sb
->s_bdev
->bd_inode_backing_dev_info
;
162 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
163 mapping
->backing_dev_info
= bdi
;
165 memset(&inode
->u
, 0, sizeof(inode
->u
));
166 inode
->i_mapping
= mapping
;
171 void destroy_inode(struct inode
*inode
)
173 if (inode_has_buffers(inode
))
175 security_inode_free(inode
);
176 if (inode
->i_sb
->s_op
->destroy_inode
)
177 inode
->i_sb
->s_op
->destroy_inode(inode
);
179 kmem_cache_free(inode_cachep
, (inode
));
184 * These are initializations that only need to be done
185 * once, because the fields are idempotent across use
186 * of the inode, so let the slab aware of that.
188 void inode_init_once(struct inode
*inode
)
190 memset(inode
, 0, sizeof(*inode
));
191 INIT_HLIST_NODE(&inode
->i_hash
);
192 INIT_LIST_HEAD(&inode
->i_dentry
);
193 INIT_LIST_HEAD(&inode
->i_devices
);
194 sema_init(&inode
->i_sem
, 1);
195 init_rwsem(&inode
->i_alloc_sem
);
196 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
197 rwlock_init(&inode
->i_data
.tree_lock
);
198 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
199 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
200 spin_lock_init(&inode
->i_data
.private_lock
);
201 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
202 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
203 spin_lock_init(&inode
->i_lock
);
204 i_size_ordered_init(inode
);
207 EXPORT_SYMBOL(inode_init_once
);
209 static void init_once(void * foo
, kmem_cache_t
* cachep
, unsigned long flags
)
211 struct inode
* inode
= (struct inode
*) foo
;
213 if ((flags
& (SLAB_CTOR_VERIFY
|SLAB_CTOR_CONSTRUCTOR
)) ==
214 SLAB_CTOR_CONSTRUCTOR
)
215 inode_init_once(inode
);
219 * inode_lock must be held
221 void __iget(struct inode
* inode
)
223 if (atomic_read(&inode
->i_count
)) {
224 atomic_inc(&inode
->i_count
);
227 atomic_inc(&inode
->i_count
);
228 if (!(inode
->i_state
& (I_DIRTY
|I_LOCK
)))
229 list_move(&inode
->i_list
, &inode_in_use
);
230 inodes_stat
.nr_unused
--;
234 * clear_inode - clear an inode
235 * @inode: inode to clear
237 * This is called by the filesystem to tell us
238 * that the inode is no longer useful. We just
239 * terminate it with extreme prejudice.
241 void clear_inode(struct inode
*inode
)
244 invalidate_inode_buffers(inode
);
246 if (inode
->i_data
.nrpages
)
248 if (!(inode
->i_state
& I_FREEING
))
250 if (inode
->i_state
& I_CLEAR
)
252 wait_on_inode(inode
);
254 if (inode
->i_sb
&& inode
->i_sb
->s_op
->clear_inode
)
255 inode
->i_sb
->s_op
->clear_inode(inode
);
260 inode
->i_state
= I_CLEAR
;
263 EXPORT_SYMBOL(clear_inode
);
266 * dispose_list - dispose of the contents of a local list
267 * @head: the head of the list to free
269 * Dispose-list gets a local list with local inodes in it, so it doesn't
270 * need to worry about list corruption and SMP locks.
272 static void dispose_list(struct list_head
*head
)
276 while (!list_empty(head
)) {
279 inode
= list_entry(head
->next
, struct inode
, i_list
);
280 list_del(&inode
->i_list
);
282 if (inode
->i_data
.nrpages
)
283 truncate_inode_pages(&inode
->i_data
, 0);
286 spin_lock(&inode_lock
);
287 hlist_del_init(&inode
->i_hash
);
288 list_del_init(&inode
->i_sb_list
);
289 spin_unlock(&inode_lock
);
291 wake_up_inode(inode
);
292 destroy_inode(inode
);
295 spin_lock(&inode_lock
);
296 inodes_stat
.nr_inodes
-= nr_disposed
;
297 spin_unlock(&inode_lock
);
301 * Invalidate all inodes for a device.
303 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
305 struct list_head
*next
;
306 int busy
= 0, count
= 0;
310 struct list_head
* tmp
= next
;
311 struct inode
* inode
;
314 * We can reschedule here without worrying about the list's
315 * consistency because the per-sb list of inodes must not
316 * change during umount anymore, and because iprune_sem keeps
317 * shrink_icache_memory() away.
319 cond_resched_lock(&inode_lock
);
324 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
325 invalidate_inode_buffers(inode
);
326 if (!atomic_read(&inode
->i_count
)) {
327 list_move(&inode
->i_list
, dispose
);
328 inode
->i_state
|= I_FREEING
;
334 /* only unused inodes may be cached with i_count zero */
335 inodes_stat
.nr_unused
-= count
;
340 * invalidate_inodes - discard the inodes on a device
343 * Discard all of the inodes for a given superblock. If the discard
344 * fails because there are busy inodes then a non zero value is returned.
345 * If the discard is successful all the inodes have been discarded.
347 int invalidate_inodes(struct super_block
* sb
)
350 LIST_HEAD(throw_away
);
353 spin_lock(&inode_lock
);
354 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
355 spin_unlock(&inode_lock
);
357 dispose_list(&throw_away
);
363 EXPORT_SYMBOL(invalidate_inodes
);
365 int __invalidate_device(struct block_device
*bdev
)
367 struct super_block
*sb
= get_super(bdev
);
372 * no need to lock the super, get_super holds the
373 * read semaphore so the filesystem cannot go away
374 * under us (->put_super runs with the write lock
377 shrink_dcache_sb(sb
);
378 res
= invalidate_inodes(sb
);
381 invalidate_bdev(bdev
, 0);
384 EXPORT_SYMBOL(__invalidate_device
);
386 static int can_unuse(struct inode
*inode
)
390 if (inode_has_buffers(inode
))
392 if (atomic_read(&inode
->i_count
))
394 if (inode
->i_data
.nrpages
)
400 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
401 * a temporary list and then are freed outside inode_lock by dispose_list().
403 * Any inodes which are pinned purely because of attached pagecache have their
404 * pagecache removed. We expect the final iput() on that inode to add it to
405 * the front of the inode_unused list. So look for it there and if the
406 * inode is still freeable, proceed. The right inode is found 99.9% of the
407 * time in testing on a 4-way.
409 * If the inode has metadata buffers attached to mapping->private_list then
410 * try to remove them.
412 static void prune_icache(int nr_to_scan
)
417 unsigned long reap
= 0;
420 spin_lock(&inode_lock
);
421 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
424 if (list_empty(&inode_unused
))
427 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
429 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
430 list_move(&inode
->i_list
, &inode_unused
);
433 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
435 spin_unlock(&inode_lock
);
436 if (remove_inode_buffers(inode
))
437 reap
+= invalidate_inode_pages(&inode
->i_data
);
439 spin_lock(&inode_lock
);
441 if (inode
!= list_entry(inode_unused
.next
,
442 struct inode
, i_list
))
443 continue; /* wrong inode or list_empty */
444 if (!can_unuse(inode
))
447 list_move(&inode
->i_list
, &freeable
);
448 inode
->i_state
|= I_FREEING
;
451 inodes_stat
.nr_unused
-= nr_pruned
;
452 spin_unlock(&inode_lock
);
454 dispose_list(&freeable
);
457 if (current_is_kswapd())
458 mod_page_state(kswapd_inodesteal
, reap
);
460 mod_page_state(pginodesteal
, reap
);
464 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
465 * "unused" means that no dentries are referring to the inodes: the files are
466 * not open and the dcache references to those inodes have already been
469 * This function is passed the number of inodes to scan, and it returns the
470 * total number of remaining possibly-reclaimable inodes.
472 static int shrink_icache_memory(int nr
, unsigned int gfp_mask
)
476 * Nasty deadlock avoidance. We may hold various FS locks,
477 * and we don't want to recurse into the FS that called us
478 * in clear_inode() and friends..
480 if (!(gfp_mask
& __GFP_FS
))
484 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
487 static void __wait_on_freeing_inode(struct inode
*inode
);
489 * Called with the inode lock held.
490 * NOTE: we are not increasing the inode-refcount, you must call __iget()
491 * by hand after calling find_inode now! This simplifies iunique and won't
492 * add any additional branch in the common code.
494 static struct inode
* find_inode(struct super_block
* sb
, struct hlist_head
*head
, int (*test
)(struct inode
*, void *), void *data
)
496 struct hlist_node
*node
;
497 struct inode
* inode
= NULL
;
500 hlist_for_each (node
, head
) {
501 inode
= hlist_entry(node
, struct inode
, i_hash
);
502 if (inode
->i_sb
!= sb
)
504 if (!test(inode
, data
))
506 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
507 __wait_on_freeing_inode(inode
);
512 return node
? inode
: NULL
;
516 * find_inode_fast is the fast path version of find_inode, see the comment at
517 * iget_locked for details.
519 static struct inode
* find_inode_fast(struct super_block
* sb
, struct hlist_head
*head
, unsigned long ino
)
521 struct hlist_node
*node
;
522 struct inode
* inode
= NULL
;
525 hlist_for_each (node
, head
) {
526 inode
= hlist_entry(node
, struct inode
, i_hash
);
527 if (inode
->i_ino
!= ino
)
529 if (inode
->i_sb
!= sb
)
531 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
532 __wait_on_freeing_inode(inode
);
537 return node
? inode
: NULL
;
541 * new_inode - obtain an inode
544 * Allocates a new inode for given superblock.
546 struct inode
*new_inode(struct super_block
*sb
)
548 static unsigned long last_ino
;
549 struct inode
* inode
;
551 spin_lock_prefetch(&inode_lock
);
553 inode
= alloc_inode(sb
);
555 spin_lock(&inode_lock
);
556 inodes_stat
.nr_inodes
++;
557 list_add(&inode
->i_list
, &inode_in_use
);
558 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
559 inode
->i_ino
= ++last_ino
;
561 spin_unlock(&inode_lock
);
566 EXPORT_SYMBOL(new_inode
);
568 void unlock_new_inode(struct inode
*inode
)
571 * This is special! We do not need the spinlock
572 * when clearing I_LOCK, because we're guaranteed
573 * that nobody else tries to do anything about the
574 * state of the inode when it is locked, as we
575 * just created it (so there can be no old holders
576 * that haven't tested I_LOCK).
578 inode
->i_state
&= ~(I_LOCK
|I_NEW
);
579 wake_up_inode(inode
);
582 EXPORT_SYMBOL(unlock_new_inode
);
585 * This is called without the inode lock held.. Be careful.
587 * We no longer cache the sb_flags in i_flags - see fs.h
588 * -- rmk@arm.uk.linux.org
590 static struct inode
* get_new_inode(struct super_block
*sb
, struct hlist_head
*head
, int (*test
)(struct inode
*, void *), int (*set
)(struct inode
*, void *), void *data
)
592 struct inode
* inode
;
594 inode
= alloc_inode(sb
);
598 spin_lock(&inode_lock
);
599 /* We released the lock, so.. */
600 old
= find_inode(sb
, head
, test
, data
);
602 if (set(inode
, data
))
605 inodes_stat
.nr_inodes
++;
606 list_add(&inode
->i_list
, &inode_in_use
);
607 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
608 hlist_add_head(&inode
->i_hash
, head
);
609 inode
->i_state
= I_LOCK
|I_NEW
;
610 spin_unlock(&inode_lock
);
612 /* Return the locked inode with I_NEW set, the
613 * caller is responsible for filling in the contents
619 * Uhhuh, somebody else created the same inode under
620 * us. Use the old inode instead of the one we just
624 spin_unlock(&inode_lock
);
625 destroy_inode(inode
);
627 wait_on_inode(inode
);
632 spin_unlock(&inode_lock
);
633 destroy_inode(inode
);
638 * get_new_inode_fast is the fast path version of get_new_inode, see the
639 * comment at iget_locked for details.
641 static struct inode
* get_new_inode_fast(struct super_block
*sb
, struct hlist_head
*head
, unsigned long ino
)
643 struct inode
* inode
;
645 inode
= alloc_inode(sb
);
649 spin_lock(&inode_lock
);
650 /* We released the lock, so.. */
651 old
= find_inode_fast(sb
, head
, ino
);
654 inodes_stat
.nr_inodes
++;
655 list_add(&inode
->i_list
, &inode_in_use
);
656 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
657 hlist_add_head(&inode
->i_hash
, head
);
658 inode
->i_state
= I_LOCK
|I_NEW
;
659 spin_unlock(&inode_lock
);
661 /* Return the locked inode with I_NEW set, the
662 * caller is responsible for filling in the contents
668 * Uhhuh, somebody else created the same inode under
669 * us. Use the old inode instead of the one we just
673 spin_unlock(&inode_lock
);
674 destroy_inode(inode
);
676 wait_on_inode(inode
);
681 static inline unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
685 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
687 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
688 return tmp
& I_HASHMASK
;
692 * iunique - get a unique inode number
694 * @max_reserved: highest reserved inode number
696 * Obtain an inode number that is unique on the system for a given
697 * superblock. This is used by file systems that have no natural
698 * permanent inode numbering system. An inode number is returned that
699 * is higher than the reserved limit but unique.
702 * With a large number of inodes live on the file system this function
703 * currently becomes quite slow.
705 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
707 static ino_t counter
;
709 struct hlist_head
* head
;
711 spin_lock(&inode_lock
);
713 if (counter
> max_reserved
) {
714 head
= inode_hashtable
+ hash(sb
,counter
);
716 inode
= find_inode_fast(sb
, head
, res
);
718 spin_unlock(&inode_lock
);
722 counter
= max_reserved
+ 1;
728 EXPORT_SYMBOL(iunique
);
730 struct inode
*igrab(struct inode
*inode
)
732 spin_lock(&inode_lock
);
733 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
)))
737 * Handle the case where s_op->clear_inode is not been
738 * called yet, and somebody is calling igrab
739 * while the inode is getting freed.
742 spin_unlock(&inode_lock
);
746 EXPORT_SYMBOL(igrab
);
749 * ifind - internal function, you want ilookup5() or iget5().
750 * @sb: super block of file system to search
751 * @head: the head of the list to search
752 * @test: callback used for comparisons between inodes
753 * @data: opaque data pointer to pass to @test
755 * ifind() searches for the inode specified by @data in the inode
756 * cache. This is a generalized version of ifind_fast() for file systems where
757 * the inode number is not sufficient for unique identification of an inode.
759 * If the inode is in the cache, the inode is returned with an incremented
762 * Otherwise NULL is returned.
764 * Note, @test is called with the inode_lock held, so can't sleep.
766 static inline struct inode
*ifind(struct super_block
*sb
,
767 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
772 spin_lock(&inode_lock
);
773 inode
= find_inode(sb
, head
, test
, data
);
776 spin_unlock(&inode_lock
);
777 wait_on_inode(inode
);
780 spin_unlock(&inode_lock
);
785 * ifind_fast - internal function, you want ilookup() or iget().
786 * @sb: super block of file system to search
787 * @head: head of the list to search
788 * @ino: inode number to search for
790 * ifind_fast() searches for the inode @ino in the inode cache. This is for
791 * file systems where the inode number is sufficient for unique identification
794 * If the inode is in the cache, the inode is returned with an incremented
797 * Otherwise NULL is returned.
799 static inline struct inode
*ifind_fast(struct super_block
*sb
,
800 struct hlist_head
*head
, unsigned long ino
)
804 spin_lock(&inode_lock
);
805 inode
= find_inode_fast(sb
, head
, ino
);
808 spin_unlock(&inode_lock
);
809 wait_on_inode(inode
);
812 spin_unlock(&inode_lock
);
817 * ilookup5 - search for an inode in the inode cache
818 * @sb: super block of file system to search
819 * @hashval: hash value (usually inode number) to search for
820 * @test: callback used for comparisons between inodes
821 * @data: opaque data pointer to pass to @test
823 * ilookup5() uses ifind() to search for the inode specified by @hashval and
824 * @data in the inode cache. This is a generalized version of ilookup() for
825 * file systems where the inode number is not sufficient for unique
826 * identification of an inode.
828 * If the inode is in the cache, the inode is returned with an incremented
831 * Otherwise NULL is returned.
833 * Note, @test is called with the inode_lock held, so can't sleep.
835 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
836 int (*test
)(struct inode
*, void *), void *data
)
838 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
840 return ifind(sb
, head
, test
, data
);
843 EXPORT_SYMBOL(ilookup5
);
846 * ilookup - search for an inode in the inode cache
847 * @sb: super block of file system to search
848 * @ino: inode number to search for
850 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
851 * This is for file systems where the inode number is sufficient for unique
852 * identification of an inode.
854 * If the inode is in the cache, the inode is returned with an incremented
857 * Otherwise NULL is returned.
859 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
861 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
863 return ifind_fast(sb
, head
, ino
);
866 EXPORT_SYMBOL(ilookup
);
869 * iget5_locked - obtain an inode from a mounted file system
870 * @sb: super block of file system
871 * @hashval: hash value (usually inode number) to get
872 * @test: callback used for comparisons between inodes
873 * @set: callback used to initialize a new struct inode
874 * @data: opaque data pointer to pass to @test and @set
876 * This is iget() without the read_inode() portion of get_new_inode().
878 * iget5_locked() uses ifind() to search for the inode specified by @hashval
879 * and @data in the inode cache and if present it is returned with an increased
880 * reference count. This is a generalized version of iget_locked() for file
881 * systems where the inode number is not sufficient for unique identification
884 * If the inode is not in cache, get_new_inode() is called to allocate a new
885 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
886 * file system gets to fill it in before unlocking it via unlock_new_inode().
888 * Note both @test and @set are called with the inode_lock held, so can't sleep.
890 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
891 int (*test
)(struct inode
*, void *),
892 int (*set
)(struct inode
*, void *), void *data
)
894 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
897 inode
= ifind(sb
, head
, test
, data
);
901 * get_new_inode() will do the right thing, re-trying the search
902 * in case it had to block at any point.
904 return get_new_inode(sb
, head
, test
, set
, data
);
907 EXPORT_SYMBOL(iget5_locked
);
910 * iget_locked - obtain an inode from a mounted file system
911 * @sb: super block of file system
912 * @ino: inode number to get
914 * This is iget() without the read_inode() portion of get_new_inode_fast().
916 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
917 * the inode cache and if present it is returned with an increased reference
918 * count. This is for file systems where the inode number is sufficient for
919 * unique identification of an inode.
921 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
922 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
923 * The file system gets to fill it in before unlocking it via
924 * unlock_new_inode().
926 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
928 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
931 inode
= ifind_fast(sb
, head
, ino
);
935 * get_new_inode_fast() will do the right thing, re-trying the search
936 * in case it had to block at any point.
938 return get_new_inode_fast(sb
, head
, ino
);
941 EXPORT_SYMBOL(iget_locked
);
944 * __insert_inode_hash - hash an inode
945 * @inode: unhashed inode
946 * @hashval: unsigned long value used to locate this object in the
949 * Add an inode to the inode hash for this superblock.
951 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
953 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
954 spin_lock(&inode_lock
);
955 hlist_add_head(&inode
->i_hash
, head
);
956 spin_unlock(&inode_lock
);
959 EXPORT_SYMBOL(__insert_inode_hash
);
962 * remove_inode_hash - remove an inode from the hash
963 * @inode: inode to unhash
965 * Remove an inode from the superblock.
967 void remove_inode_hash(struct inode
*inode
)
969 spin_lock(&inode_lock
);
970 hlist_del_init(&inode
->i_hash
);
971 spin_unlock(&inode_lock
);
974 EXPORT_SYMBOL(remove_inode_hash
);
977 * Tell the filesystem that this inode is no longer of any interest and should
978 * be completely destroyed.
980 * We leave the inode in the inode hash table until *after* the filesystem's
981 * ->delete_inode completes. This ensures that an iget (such as nfsd might
982 * instigate) will always find up-to-date information either in the hash or on
985 * I_FREEING is set so that no-one will take a new reference to the inode while
986 * it is being deleted.
988 void generic_delete_inode(struct inode
*inode
)
990 struct super_operations
*op
= inode
->i_sb
->s_op
;
992 list_del_init(&inode
->i_list
);
993 list_del_init(&inode
->i_sb_list
);
994 inode
->i_state
|=I_FREEING
;
995 inodes_stat
.nr_inodes
--;
996 spin_unlock(&inode_lock
);
998 if (inode
->i_data
.nrpages
)
999 truncate_inode_pages(&inode
->i_data
, 0);
1001 security_inode_delete(inode
);
1003 if (op
->delete_inode
) {
1004 void (*delete)(struct inode
*) = op
->delete_inode
;
1005 if (!is_bad_inode(inode
))
1007 /* s_op->delete_inode internally recalls clear_inode() */
1011 spin_lock(&inode_lock
);
1012 hlist_del_init(&inode
->i_hash
);
1013 spin_unlock(&inode_lock
);
1014 wake_up_inode(inode
);
1015 if (inode
->i_state
!= I_CLEAR
)
1017 destroy_inode(inode
);
1020 EXPORT_SYMBOL(generic_delete_inode
);
1022 static void generic_forget_inode(struct inode
*inode
)
1024 struct super_block
*sb
= inode
->i_sb
;
1026 if (!hlist_unhashed(&inode
->i_hash
)) {
1027 if (!(inode
->i_state
& (I_DIRTY
|I_LOCK
)))
1028 list_move(&inode
->i_list
, &inode_unused
);
1029 inodes_stat
.nr_unused
++;
1030 if (!sb
|| (sb
->s_flags
& MS_ACTIVE
)) {
1031 spin_unlock(&inode_lock
);
1034 inode
->i_state
|= I_WILL_FREE
;
1035 spin_unlock(&inode_lock
);
1036 write_inode_now(inode
, 1);
1037 spin_lock(&inode_lock
);
1038 inode
->i_state
&= ~I_WILL_FREE
;
1039 inodes_stat
.nr_unused
--;
1040 hlist_del_init(&inode
->i_hash
);
1042 list_del_init(&inode
->i_list
);
1043 list_del_init(&inode
->i_sb_list
);
1044 inode
->i_state
|= I_FREEING
;
1045 inodes_stat
.nr_inodes
--;
1046 spin_unlock(&inode_lock
);
1047 if (inode
->i_data
.nrpages
)
1048 truncate_inode_pages(&inode
->i_data
, 0);
1050 destroy_inode(inode
);
1054 * Normal UNIX filesystem behaviour: delete the
1055 * inode when the usage count drops to zero, and
1058 void generic_drop_inode(struct inode
*inode
)
1060 if (!inode
->i_nlink
)
1061 generic_delete_inode(inode
);
1063 generic_forget_inode(inode
);
1066 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1069 * Called when we're dropping the last reference
1072 * Call the FS "drop()" function, defaulting to
1073 * the legacy UNIX filesystem behaviour..
1075 * NOTE! NOTE! NOTE! We're called with the inode lock
1076 * held, and the drop function is supposed to release
1079 static inline void iput_final(struct inode
*inode
)
1081 struct super_operations
*op
= inode
->i_sb
->s_op
;
1082 void (*drop
)(struct inode
*) = generic_drop_inode
;
1084 if (op
&& op
->drop_inode
)
1085 drop
= op
->drop_inode
;
1090 * iput - put an inode
1091 * @inode: inode to put
1093 * Puts an inode, dropping its usage count. If the inode use count hits
1094 * zero, the inode is then freed and may also be destroyed.
1096 * Consequently, iput() can sleep.
1098 void iput(struct inode
*inode
)
1101 struct super_operations
*op
= inode
->i_sb
->s_op
;
1103 BUG_ON(inode
->i_state
== I_CLEAR
);
1105 if (op
&& op
->put_inode
)
1106 op
->put_inode(inode
);
1108 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1113 EXPORT_SYMBOL(iput
);
1116 * bmap - find a block number in a file
1117 * @inode: inode of file
1118 * @block: block to find
1120 * Returns the block number on the device holding the inode that
1121 * is the disk block number for the block of the file requested.
1122 * That is, asked for block 4 of inode 1 the function will return the
1123 * disk block relative to the disk start that holds that block of the
1126 sector_t
bmap(struct inode
* inode
, sector_t block
)
1129 if (inode
->i_mapping
->a_ops
->bmap
)
1130 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1134 EXPORT_SYMBOL(bmap
);
1137 * update_atime - update the access time
1138 * @inode: inode accessed
1140 * Update the accessed time on an inode and mark it for writeback.
1141 * This function automatically handles read only file systems and media,
1142 * as well as the "noatime" flag and inode specific "noatime" markers.
1144 void update_atime(struct inode
*inode
)
1146 struct timespec now
;
1148 if (IS_NOATIME(inode
))
1150 if (IS_NODIRATIME(inode
) && S_ISDIR(inode
->i_mode
))
1152 if (IS_RDONLY(inode
))
1155 now
= current_fs_time(inode
->i_sb
);
1156 if (!timespec_equal(&inode
->i_atime
, &now
)) {
1157 inode
->i_atime
= now
;
1158 mark_inode_dirty_sync(inode
);
1160 if (!timespec_equal(&inode
->i_atime
, &now
))
1161 inode
->i_atime
= now
;
1165 EXPORT_SYMBOL(update_atime
);
1168 * inode_update_time - update mtime and ctime time
1169 * @inode: inode accessed
1170 * @ctime_too: update ctime too
1172 * Update the mtime time on an inode and mark it for writeback.
1173 * When ctime_too is specified update the ctime too.
1176 void inode_update_time(struct inode
*inode
, int ctime_too
)
1178 struct timespec now
;
1181 if (IS_NOCMTIME(inode
))
1183 if (IS_RDONLY(inode
))
1186 now
= current_fs_time(inode
->i_sb
);
1187 if (!timespec_equal(&inode
->i_mtime
, &now
))
1189 inode
->i_mtime
= now
;
1192 if (!timespec_equal(&inode
->i_ctime
, &now
))
1194 inode
->i_ctime
= now
;
1197 mark_inode_dirty_sync(inode
);
1200 EXPORT_SYMBOL(inode_update_time
);
1202 int inode_needs_sync(struct inode
*inode
)
1206 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1211 EXPORT_SYMBOL(inode_needs_sync
);
1214 * Quota functions that want to walk the inode lists..
1218 /* Function back in dquot.c */
1219 int remove_inode_dquot_ref(struct inode
*, int, struct list_head
*);
1221 void remove_dquot_ref(struct super_block
*sb
, int type
,
1222 struct list_head
*tofree_head
)
1224 struct inode
*inode
;
1227 return; /* nothing to do */
1228 spin_lock(&inode_lock
); /* This lock is for inodes code */
1231 * We don't have to lock against quota code - test IS_QUOTAINIT is
1232 * just for speedup...
1234 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
)
1235 if (!IS_NOQUOTA(inode
))
1236 remove_inode_dquot_ref(inode
, type
, tofree_head
);
1238 spin_unlock(&inode_lock
);
1243 int inode_wait(void *word
)
1250 * If we try to find an inode in the inode hash while it is being
1251 * deleted, we have to wait until the filesystem completes its
1252 * deletion before reporting that it isn't found. This function waits
1253 * until the deletion _might_ have completed. Callers are responsible
1254 * to recheck inode state.
1256 * It doesn't matter if I_LOCK is not set initially, a call to
1257 * wake_up_inode() after removing from the hash list will DTRT.
1259 * This is called with inode_lock held.
1261 static void __wait_on_freeing_inode(struct inode
*inode
)
1263 wait_queue_head_t
*wq
;
1264 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_LOCK
);
1265 wq
= bit_waitqueue(&inode
->i_state
, __I_LOCK
);
1266 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1267 spin_unlock(&inode_lock
);
1269 finish_wait(wq
, &wait
.wait
);
1270 spin_lock(&inode_lock
);
1273 void wake_up_inode(struct inode
*inode
)
1276 * Prevent speculative execution through spin_unlock(&inode_lock);
1279 wake_up_bit(&inode
->i_state
, __I_LOCK
);
1282 static __initdata
unsigned long ihash_entries
;
1283 static int __init
set_ihash_entries(char *str
)
1287 ihash_entries
= simple_strtoul(str
, &str
, 0);
1290 __setup("ihash_entries=", set_ihash_entries
);
1293 * Initialize the waitqueues and inode hash table.
1295 void __init
inode_init_early(void)
1299 /* If hashes are distributed across NUMA nodes, defer
1300 * hash allocation until vmalloc space is available.
1306 alloc_large_system_hash("Inode-cache",
1307 sizeof(struct hlist_head
),
1315 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1316 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1319 void __init
inode_init(unsigned long mempages
)
1323 /* inode slab cache */
1324 inode_cachep
= kmem_cache_create("inode_cache", sizeof(struct inode
),
1325 0, SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
, init_once
, NULL
);
1326 set_shrinker(DEFAULT_SEEKS
, shrink_icache_memory
);
1328 /* Hash may have been set up in inode_init_early */
1333 alloc_large_system_hash("Inode-cache",
1334 sizeof(struct hlist_head
),
1342 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1343 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1346 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1348 inode
->i_mode
= mode
;
1349 if (S_ISCHR(mode
)) {
1350 inode
->i_fop
= &def_chr_fops
;
1351 inode
->i_rdev
= rdev
;
1352 } else if (S_ISBLK(mode
)) {
1353 inode
->i_fop
= &def_blk_fops
;
1354 inode
->i_rdev
= rdev
;
1355 } else if (S_ISFIFO(mode
))
1356 inode
->i_fop
= &def_fifo_fops
;
1357 else if (S_ISSOCK(mode
))
1358 inode
->i_fop
= &bad_sock_fops
;
1360 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o)\n",
1363 EXPORT_SYMBOL(init_special_inode
);