4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/quotaops.h>
12 #include <linux/slab.h>
13 #include <linux/writeback.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/wait.h>
17 #include <linux/hash.h>
18 #include <linux/swap.h>
19 #include <linux/security.h>
20 #include <linux/pagemap.h>
21 #include <linux/cdev.h>
22 #include <linux/bootmem.h>
23 #include <linux/inotify.h>
24 #include <linux/mount.h>
27 * This is needed for the following functions:
29 * - invalidate_inode_buffers
32 * FIXME: remove all knowledge of the buffer layer from this file
34 #include <linux/buffer_head.h>
37 * New inode.c implementation.
39 * This implementation has the basic premise of trying
40 * to be extremely low-overhead and SMP-safe, yet be
41 * simple enough to be "obviously correct".
46 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
48 /* #define INODE_PARANOIA 1 */
49 /* #define INODE_DEBUG 1 */
52 * Inode lookup is no longer as critical as it used to be:
53 * most of the lookups are going to be through the dcache.
55 #define I_HASHBITS i_hash_shift
56 #define I_HASHMASK i_hash_mask
58 static unsigned int i_hash_mask __read_mostly
;
59 static unsigned int i_hash_shift __read_mostly
;
62 * Each inode can be on two separate lists. One is
63 * the hash list of the inode, used for lookups. The
64 * other linked list is the "type" list:
65 * "in_use" - valid inode, i_count > 0, i_nlink > 0
66 * "dirty" - as "in_use" but also dirty
67 * "unused" - valid inode, i_count = 0
69 * A "dirty" list is maintained for each super block,
70 * allowing for low-overhead inode sync() operations.
73 LIST_HEAD(inode_in_use
);
74 LIST_HEAD(inode_unused
);
75 static struct hlist_head
*inode_hashtable __read_mostly
;
78 * A simple spinlock to protect the list manipulations.
80 * NOTE! You also have to own the lock if you change
81 * the i_state of an inode while it is in use..
83 DEFINE_SPINLOCK(inode_lock
);
86 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
87 * icache shrinking path, and the umount path. Without this exclusion,
88 * by the time prune_icache calls iput for the inode whose pages it has
89 * been invalidating, or by the time it calls clear_inode & destroy_inode
90 * from its final dispose_list, the struct super_block they refer to
91 * (for inode->i_sb->s_op) may already have been freed and reused.
93 static DEFINE_MUTEX(iprune_mutex
);
96 * Statistics gathering..
98 struct inodes_stat_t inodes_stat
;
100 static struct kmem_cache
* inode_cachep __read_mostly
;
102 static void wake_up_inode(struct inode
*inode
)
105 * Prevent speculative execution through spin_unlock(&inode_lock);
108 wake_up_bit(&inode
->i_state
, __I_LOCK
);
111 static struct inode
*alloc_inode(struct super_block
*sb
)
113 static const struct address_space_operations empty_aops
;
114 static struct inode_operations empty_iops
;
115 static const struct file_operations empty_fops
;
118 if (sb
->s_op
->alloc_inode
)
119 inode
= sb
->s_op
->alloc_inode(sb
);
121 inode
= (struct inode
*) kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
124 struct address_space
* const mapping
= &inode
->i_data
;
127 inode
->i_blkbits
= sb
->s_blocksize_bits
;
129 atomic_set(&inode
->i_count
, 1);
130 inode
->i_op
= &empty_iops
;
131 inode
->i_fop
= &empty_fops
;
133 atomic_set(&inode
->i_writecount
, 0);
137 inode
->i_generation
= 0;
139 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
141 inode
->i_pipe
= NULL
;
142 inode
->i_bdev
= NULL
;
143 inode
->i_cdev
= NULL
;
145 inode
->dirtied_when
= 0;
146 if (security_inode_alloc(inode
)) {
147 if (inode
->i_sb
->s_op
->destroy_inode
)
148 inode
->i_sb
->s_op
->destroy_inode(inode
);
150 kmem_cache_free(inode_cachep
, (inode
));
154 spin_lock_init(&inode
->i_lock
);
155 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
157 mutex_init(&inode
->i_mutex
);
158 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
160 init_rwsem(&inode
->i_alloc_sem
);
161 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
163 mapping
->a_ops
= &empty_aops
;
164 mapping
->host
= inode
;
166 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_PAGECACHE
);
167 mapping
->assoc_mapping
= NULL
;
168 mapping
->backing_dev_info
= &default_backing_dev_info
;
171 * If the block_device provides a backing_dev_info for client
172 * inodes then use that. Otherwise the inode share the bdev's
176 struct backing_dev_info
*bdi
;
178 bdi
= sb
->s_bdev
->bd_inode_backing_dev_info
;
180 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
181 mapping
->backing_dev_info
= bdi
;
183 inode
->i_private
= NULL
;
184 inode
->i_mapping
= mapping
;
189 void destroy_inode(struct inode
*inode
)
191 BUG_ON(inode_has_buffers(inode
));
192 security_inode_free(inode
);
193 if (inode
->i_sb
->s_op
->destroy_inode
)
194 inode
->i_sb
->s_op
->destroy_inode(inode
);
196 kmem_cache_free(inode_cachep
, (inode
));
201 * These are initializations that only need to be done
202 * once, because the fields are idempotent across use
203 * of the inode, so let the slab aware of that.
205 void inode_init_once(struct inode
*inode
)
207 memset(inode
, 0, sizeof(*inode
));
208 INIT_HLIST_NODE(&inode
->i_hash
);
209 INIT_LIST_HEAD(&inode
->i_dentry
);
210 INIT_LIST_HEAD(&inode
->i_devices
);
211 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
212 rwlock_init(&inode
->i_data
.tree_lock
);
213 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
214 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
215 spin_lock_init(&inode
->i_data
.private_lock
);
216 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
217 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
218 i_size_ordered_init(inode
);
219 #ifdef CONFIG_INOTIFY
220 INIT_LIST_HEAD(&inode
->inotify_watches
);
221 mutex_init(&inode
->inotify_mutex
);
225 EXPORT_SYMBOL(inode_init_once
);
227 static void init_once(struct kmem_cache
* cachep
, void *foo
)
229 struct inode
* inode
= (struct inode
*) foo
;
231 inode_init_once(inode
);
235 * inode_lock must be held
237 void __iget(struct inode
* inode
)
239 if (atomic_read(&inode
->i_count
)) {
240 atomic_inc(&inode
->i_count
);
243 atomic_inc(&inode
->i_count
);
244 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
245 list_move(&inode
->i_list
, &inode_in_use
);
246 inodes_stat
.nr_unused
--;
250 * clear_inode - clear an inode
251 * @inode: inode to clear
253 * This is called by the filesystem to tell us
254 * that the inode is no longer useful. We just
255 * terminate it with extreme prejudice.
257 void clear_inode(struct inode
*inode
)
260 invalidate_inode_buffers(inode
);
262 BUG_ON(inode
->i_data
.nrpages
);
263 BUG_ON(!(inode
->i_state
& I_FREEING
));
264 BUG_ON(inode
->i_state
& I_CLEAR
);
265 inode_sync_wait(inode
);
267 if (inode
->i_sb
->s_op
->clear_inode
)
268 inode
->i_sb
->s_op
->clear_inode(inode
);
269 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
271 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
273 inode
->i_state
= I_CLEAR
;
276 EXPORT_SYMBOL(clear_inode
);
279 * dispose_list - dispose of the contents of a local list
280 * @head: the head of the list to free
282 * Dispose-list gets a local list with local inodes in it, so it doesn't
283 * need to worry about list corruption and SMP locks.
285 static void dispose_list(struct list_head
*head
)
289 while (!list_empty(head
)) {
292 inode
= list_first_entry(head
, struct inode
, i_list
);
293 list_del(&inode
->i_list
);
295 if (inode
->i_data
.nrpages
)
296 truncate_inode_pages(&inode
->i_data
, 0);
299 spin_lock(&inode_lock
);
300 hlist_del_init(&inode
->i_hash
);
301 list_del_init(&inode
->i_sb_list
);
302 spin_unlock(&inode_lock
);
304 wake_up_inode(inode
);
305 destroy_inode(inode
);
308 spin_lock(&inode_lock
);
309 inodes_stat
.nr_inodes
-= nr_disposed
;
310 spin_unlock(&inode_lock
);
314 * Invalidate all inodes for a device.
316 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
318 struct list_head
*next
;
319 int busy
= 0, count
= 0;
323 struct list_head
* tmp
= next
;
324 struct inode
* inode
;
327 * We can reschedule here without worrying about the list's
328 * consistency because the per-sb list of inodes must not
329 * change during umount anymore, and because iprune_mutex keeps
330 * shrink_icache_memory() away.
332 cond_resched_lock(&inode_lock
);
337 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
338 invalidate_inode_buffers(inode
);
339 if (!atomic_read(&inode
->i_count
)) {
340 list_move(&inode
->i_list
, dispose
);
341 inode
->i_state
|= I_FREEING
;
347 /* only unused inodes may be cached with i_count zero */
348 inodes_stat
.nr_unused
-= count
;
353 * invalidate_inodes - discard the inodes on a device
356 * Discard all of the inodes for a given superblock. If the discard
357 * fails because there are busy inodes then a non zero value is returned.
358 * If the discard is successful all the inodes have been discarded.
360 int invalidate_inodes(struct super_block
* sb
)
363 LIST_HEAD(throw_away
);
365 mutex_lock(&iprune_mutex
);
366 spin_lock(&inode_lock
);
367 inotify_unmount_inodes(&sb
->s_inodes
);
368 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
369 spin_unlock(&inode_lock
);
371 dispose_list(&throw_away
);
372 mutex_unlock(&iprune_mutex
);
377 EXPORT_SYMBOL(invalidate_inodes
);
379 static int can_unuse(struct inode
*inode
)
383 if (inode_has_buffers(inode
))
385 if (atomic_read(&inode
->i_count
))
387 if (inode
->i_data
.nrpages
)
393 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
394 * a temporary list and then are freed outside inode_lock by dispose_list().
396 * Any inodes which are pinned purely because of attached pagecache have their
397 * pagecache removed. We expect the final iput() on that inode to add it to
398 * the front of the inode_unused list. So look for it there and if the
399 * inode is still freeable, proceed. The right inode is found 99.9% of the
400 * time in testing on a 4-way.
402 * If the inode has metadata buffers attached to mapping->private_list then
403 * try to remove them.
405 static void prune_icache(int nr_to_scan
)
410 unsigned long reap
= 0;
412 mutex_lock(&iprune_mutex
);
413 spin_lock(&inode_lock
);
414 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
417 if (list_empty(&inode_unused
))
420 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
422 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
423 list_move(&inode
->i_list
, &inode_unused
);
426 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
428 spin_unlock(&inode_lock
);
429 if (remove_inode_buffers(inode
))
430 reap
+= invalidate_mapping_pages(&inode
->i_data
,
433 spin_lock(&inode_lock
);
435 if (inode
!= list_entry(inode_unused
.next
,
436 struct inode
, i_list
))
437 continue; /* wrong inode or list_empty */
438 if (!can_unuse(inode
))
441 list_move(&inode
->i_list
, &freeable
);
442 inode
->i_state
|= I_FREEING
;
445 inodes_stat
.nr_unused
-= nr_pruned
;
446 if (current_is_kswapd())
447 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
449 __count_vm_events(PGINODESTEAL
, reap
);
450 spin_unlock(&inode_lock
);
452 dispose_list(&freeable
);
453 mutex_unlock(&iprune_mutex
);
457 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
458 * "unused" means that no dentries are referring to the inodes: the files are
459 * not open and the dcache references to those inodes have already been
462 * This function is passed the number of inodes to scan, and it returns the
463 * total number of remaining possibly-reclaimable inodes.
465 static int shrink_icache_memory(int nr
, gfp_t gfp_mask
)
469 * Nasty deadlock avoidance. We may hold various FS locks,
470 * and we don't want to recurse into the FS that called us
471 * in clear_inode() and friends..
473 if (!(gfp_mask
& __GFP_FS
))
477 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
480 static struct shrinker icache_shrinker
= {
481 .shrink
= shrink_icache_memory
,
482 .seeks
= DEFAULT_SEEKS
,
485 static void __wait_on_freeing_inode(struct inode
*inode
);
487 * Called with the inode lock held.
488 * NOTE: we are not increasing the inode-refcount, you must call __iget()
489 * by hand after calling find_inode now! This simplifies iunique and won't
490 * add any additional branch in the common code.
492 static struct inode
* find_inode(struct super_block
* sb
, struct hlist_head
*head
, int (*test
)(struct inode
*, void *), void *data
)
494 struct hlist_node
*node
;
495 struct inode
* inode
= NULL
;
498 hlist_for_each (node
, head
) {
499 inode
= hlist_entry(node
, struct inode
, i_hash
);
500 if (inode
->i_sb
!= sb
)
502 if (!test(inode
, data
))
504 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
505 __wait_on_freeing_inode(inode
);
510 return node
? inode
: NULL
;
514 * find_inode_fast is the fast path version of find_inode, see the comment at
515 * iget_locked for details.
517 static struct inode
* find_inode_fast(struct super_block
* sb
, struct hlist_head
*head
, unsigned long ino
)
519 struct hlist_node
*node
;
520 struct inode
* inode
= NULL
;
523 hlist_for_each (node
, head
) {
524 inode
= hlist_entry(node
, struct inode
, i_hash
);
525 if (inode
->i_ino
!= ino
)
527 if (inode
->i_sb
!= sb
)
529 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
530 __wait_on_freeing_inode(inode
);
535 return node
? inode
: NULL
;
539 * new_inode - obtain an inode
542 * Allocates a new inode for given superblock. The default gfp_mask
543 * for allocations related to inode->i_mapping is GFP_HIGHUSER_PAGECACHE.
544 * If HIGHMEM pages are unsuitable or it is known that pages allocated
545 * for the page cache are not reclaimable or migratable,
546 * mapping_set_gfp_mask() must be called with suitable flags on the
547 * newly created inode's mapping
550 struct inode
*new_inode(struct super_block
*sb
)
553 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
554 * error if st_ino won't fit in target struct field. Use 32bit counter
555 * here to attempt to avoid that.
557 static unsigned int last_ino
;
558 struct inode
* inode
;
560 spin_lock_prefetch(&inode_lock
);
562 inode
= alloc_inode(sb
);
564 spin_lock(&inode_lock
);
565 inodes_stat
.nr_inodes
++;
566 list_add(&inode
->i_list
, &inode_in_use
);
567 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
568 inode
->i_ino
= ++last_ino
;
570 spin_unlock(&inode_lock
);
575 EXPORT_SYMBOL(new_inode
);
577 void unlock_new_inode(struct inode
*inode
)
579 #ifdef CONFIG_DEBUG_LOCK_ALLOC
580 if (inode
->i_mode
& S_IFDIR
) {
581 struct file_system_type
*type
= inode
->i_sb
->s_type
;
584 * ensure nobody is actually holding i_mutex
586 mutex_destroy(&inode
->i_mutex
);
587 mutex_init(&inode
->i_mutex
);
588 lockdep_set_class(&inode
->i_mutex
, &type
->i_mutex_dir_key
);
592 * This is special! We do not need the spinlock
593 * when clearing I_LOCK, because we're guaranteed
594 * that nobody else tries to do anything about the
595 * state of the inode when it is locked, as we
596 * just created it (so there can be no old holders
597 * that haven't tested I_LOCK).
599 inode
->i_state
&= ~(I_LOCK
|I_NEW
);
600 wake_up_inode(inode
);
603 EXPORT_SYMBOL(unlock_new_inode
);
606 * This is called without the inode lock held.. Be careful.
608 * We no longer cache the sb_flags in i_flags - see fs.h
609 * -- rmk@arm.uk.linux.org
611 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
)
613 struct inode
* inode
;
615 inode
= alloc_inode(sb
);
619 spin_lock(&inode_lock
);
620 /* We released the lock, so.. */
621 old
= find_inode(sb
, head
, test
, data
);
623 if (set(inode
, data
))
626 inodes_stat
.nr_inodes
++;
627 list_add(&inode
->i_list
, &inode_in_use
);
628 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
629 hlist_add_head(&inode
->i_hash
, head
);
630 inode
->i_state
= I_LOCK
|I_NEW
;
631 spin_unlock(&inode_lock
);
633 /* Return the locked inode with I_NEW set, the
634 * caller is responsible for filling in the contents
640 * Uhhuh, somebody else created the same inode under
641 * us. Use the old inode instead of the one we just
645 spin_unlock(&inode_lock
);
646 destroy_inode(inode
);
648 wait_on_inode(inode
);
653 spin_unlock(&inode_lock
);
654 destroy_inode(inode
);
659 * get_new_inode_fast is the fast path version of get_new_inode, see the
660 * comment at iget_locked for details.
662 static struct inode
* get_new_inode_fast(struct super_block
*sb
, struct hlist_head
*head
, unsigned long ino
)
664 struct inode
* inode
;
666 inode
= alloc_inode(sb
);
670 spin_lock(&inode_lock
);
671 /* We released the lock, so.. */
672 old
= find_inode_fast(sb
, head
, ino
);
675 inodes_stat
.nr_inodes
++;
676 list_add(&inode
->i_list
, &inode_in_use
);
677 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
678 hlist_add_head(&inode
->i_hash
, head
);
679 inode
->i_state
= I_LOCK
|I_NEW
;
680 spin_unlock(&inode_lock
);
682 /* Return the locked inode with I_NEW set, the
683 * caller is responsible for filling in the contents
689 * Uhhuh, somebody else created the same inode under
690 * us. Use the old inode instead of the one we just
694 spin_unlock(&inode_lock
);
695 destroy_inode(inode
);
697 wait_on_inode(inode
);
702 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
706 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
708 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
709 return tmp
& I_HASHMASK
;
713 * iunique - get a unique inode number
715 * @max_reserved: highest reserved inode number
717 * Obtain an inode number that is unique on the system for a given
718 * superblock. This is used by file systems that have no natural
719 * permanent inode numbering system. An inode number is returned that
720 * is higher than the reserved limit but unique.
723 * With a large number of inodes live on the file system this function
724 * currently becomes quite slow.
726 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
729 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
730 * error if st_ino won't fit in target struct field. Use 32bit counter
731 * here to attempt to avoid that.
733 static unsigned int counter
;
735 struct hlist_head
*head
;
738 spin_lock(&inode_lock
);
740 if (counter
<= max_reserved
)
741 counter
= max_reserved
+ 1;
743 head
= inode_hashtable
+ hash(sb
, res
);
744 inode
= find_inode_fast(sb
, head
, res
);
745 } while (inode
!= NULL
);
746 spin_unlock(&inode_lock
);
750 EXPORT_SYMBOL(iunique
);
752 struct inode
*igrab(struct inode
*inode
)
754 spin_lock(&inode_lock
);
755 if (!(inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)))
759 * Handle the case where s_op->clear_inode is not been
760 * called yet, and somebody is calling igrab
761 * while the inode is getting freed.
764 spin_unlock(&inode_lock
);
768 EXPORT_SYMBOL(igrab
);
771 * ifind - internal function, you want ilookup5() or iget5().
772 * @sb: super block of file system to search
773 * @head: the head of the list to search
774 * @test: callback used for comparisons between inodes
775 * @data: opaque data pointer to pass to @test
776 * @wait: if true wait for the inode to be unlocked, if false do not
778 * ifind() searches for the inode specified by @data in the inode
779 * cache. This is a generalized version of ifind_fast() for file systems where
780 * the inode number is not sufficient for unique identification of an inode.
782 * If the inode is in the cache, the inode is returned with an incremented
785 * Otherwise NULL is returned.
787 * Note, @test is called with the inode_lock held, so can't sleep.
789 static struct inode
*ifind(struct super_block
*sb
,
790 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
791 void *data
, const int wait
)
795 spin_lock(&inode_lock
);
796 inode
= find_inode(sb
, head
, test
, data
);
799 spin_unlock(&inode_lock
);
801 wait_on_inode(inode
);
804 spin_unlock(&inode_lock
);
809 * ifind_fast - internal function, you want ilookup() or iget().
810 * @sb: super block of file system to search
811 * @head: head of the list to search
812 * @ino: inode number to search for
814 * ifind_fast() searches for the inode @ino in the inode cache. This is for
815 * file systems where the inode number is sufficient for unique identification
818 * If the inode is in the cache, the inode is returned with an incremented
821 * Otherwise NULL is returned.
823 static struct inode
*ifind_fast(struct super_block
*sb
,
824 struct hlist_head
*head
, unsigned long ino
)
828 spin_lock(&inode_lock
);
829 inode
= find_inode_fast(sb
, head
, ino
);
832 spin_unlock(&inode_lock
);
833 wait_on_inode(inode
);
836 spin_unlock(&inode_lock
);
841 * ilookup5_nowait - search for an inode in the inode cache
842 * @sb: super block of file system to search
843 * @hashval: hash value (usually inode number) to search for
844 * @test: callback used for comparisons between inodes
845 * @data: opaque data pointer to pass to @test
847 * ilookup5() uses ifind() to search for the inode specified by @hashval and
848 * @data in the inode cache. This is a generalized version of ilookup() for
849 * file systems where the inode number is not sufficient for unique
850 * identification of an inode.
852 * If the inode is in the cache, the inode is returned with an incremented
853 * reference count. Note, the inode lock is not waited upon so you have to be
854 * very careful what you do with the returned inode. You probably should be
855 * using ilookup5() instead.
857 * Otherwise NULL is returned.
859 * Note, @test is called with the inode_lock held, so can't sleep.
861 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
862 int (*test
)(struct inode
*, void *), void *data
)
864 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
866 return ifind(sb
, head
, test
, data
, 0);
869 EXPORT_SYMBOL(ilookup5_nowait
);
872 * ilookup5 - search for an inode in the inode cache
873 * @sb: super block of file system to search
874 * @hashval: hash value (usually inode number) to search for
875 * @test: callback used for comparisons between inodes
876 * @data: opaque data pointer to pass to @test
878 * ilookup5() uses ifind() to search for the inode specified by @hashval and
879 * @data in the inode cache. This is a generalized version of ilookup() for
880 * file systems where the inode number is not sufficient for unique
881 * identification of an inode.
883 * If the inode is in the cache, the inode lock is waited upon and the inode is
884 * returned with an incremented reference count.
886 * Otherwise NULL is returned.
888 * Note, @test is called with the inode_lock held, so can't sleep.
890 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
891 int (*test
)(struct inode
*, void *), void *data
)
893 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
895 return ifind(sb
, head
, test
, data
, 1);
898 EXPORT_SYMBOL(ilookup5
);
901 * ilookup - search for an inode in the inode cache
902 * @sb: super block of file system to search
903 * @ino: inode number to search for
905 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
906 * This is for file systems where the inode number is sufficient for unique
907 * identification of an inode.
909 * If the inode is in the cache, the inode is returned with an incremented
912 * Otherwise NULL is returned.
914 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
916 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
918 return ifind_fast(sb
, head
, ino
);
921 EXPORT_SYMBOL(ilookup
);
924 * iget5_locked - obtain an inode from a mounted file system
925 * @sb: super block of file system
926 * @hashval: hash value (usually inode number) to get
927 * @test: callback used for comparisons between inodes
928 * @set: callback used to initialize a new struct inode
929 * @data: opaque data pointer to pass to @test and @set
931 * iget5_locked() uses ifind() to search for the inode specified by @hashval
932 * and @data in the inode cache and if present it is returned with an increased
933 * reference count. This is a generalized version of iget_locked() for file
934 * systems where the inode number is not sufficient for unique identification
937 * If the inode is not in cache, get_new_inode() is called to allocate a new
938 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
939 * file system gets to fill it in before unlocking it via unlock_new_inode().
941 * Note both @test and @set are called with the inode_lock held, so can't sleep.
943 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
944 int (*test
)(struct inode
*, void *),
945 int (*set
)(struct inode
*, void *), void *data
)
947 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
950 inode
= ifind(sb
, head
, test
, data
, 1);
954 * get_new_inode() will do the right thing, re-trying the search
955 * in case it had to block at any point.
957 return get_new_inode(sb
, head
, test
, set
, data
);
960 EXPORT_SYMBOL(iget5_locked
);
963 * iget_locked - obtain an inode from a mounted file system
964 * @sb: super block of file system
965 * @ino: inode number to get
967 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
968 * the inode cache and if present it is returned with an increased reference
969 * count. This is for file systems where the inode number is sufficient for
970 * unique identification of an inode.
972 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
973 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
974 * The file system gets to fill it in before unlocking it via
975 * unlock_new_inode().
977 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
979 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
982 inode
= ifind_fast(sb
, head
, ino
);
986 * get_new_inode_fast() will do the right thing, re-trying the search
987 * in case it had to block at any point.
989 return get_new_inode_fast(sb
, head
, ino
);
992 EXPORT_SYMBOL(iget_locked
);
995 * __insert_inode_hash - hash an inode
996 * @inode: unhashed inode
997 * @hashval: unsigned long value used to locate this object in the
1000 * Add an inode to the inode hash for this superblock.
1002 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1004 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1005 spin_lock(&inode_lock
);
1006 hlist_add_head(&inode
->i_hash
, head
);
1007 spin_unlock(&inode_lock
);
1010 EXPORT_SYMBOL(__insert_inode_hash
);
1013 * remove_inode_hash - remove an inode from the hash
1014 * @inode: inode to unhash
1016 * Remove an inode from the superblock.
1018 void remove_inode_hash(struct inode
*inode
)
1020 spin_lock(&inode_lock
);
1021 hlist_del_init(&inode
->i_hash
);
1022 spin_unlock(&inode_lock
);
1025 EXPORT_SYMBOL(remove_inode_hash
);
1028 * Tell the filesystem that this inode is no longer of any interest and should
1029 * be completely destroyed.
1031 * We leave the inode in the inode hash table until *after* the filesystem's
1032 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1033 * instigate) will always find up-to-date information either in the hash or on
1036 * I_FREEING is set so that no-one will take a new reference to the inode while
1037 * it is being deleted.
1039 void generic_delete_inode(struct inode
*inode
)
1041 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1043 list_del_init(&inode
->i_list
);
1044 list_del_init(&inode
->i_sb_list
);
1045 inode
->i_state
|= I_FREEING
;
1046 inodes_stat
.nr_inodes
--;
1047 spin_unlock(&inode_lock
);
1049 security_inode_delete(inode
);
1051 if (op
->delete_inode
) {
1052 void (*delete)(struct inode
*) = op
->delete_inode
;
1053 if (!is_bad_inode(inode
))
1055 /* Filesystems implementing their own
1056 * s_op->delete_inode are required to call
1057 * truncate_inode_pages and clear_inode()
1061 truncate_inode_pages(&inode
->i_data
, 0);
1064 spin_lock(&inode_lock
);
1065 hlist_del_init(&inode
->i_hash
);
1066 spin_unlock(&inode_lock
);
1067 wake_up_inode(inode
);
1068 BUG_ON(inode
->i_state
!= I_CLEAR
);
1069 destroy_inode(inode
);
1072 EXPORT_SYMBOL(generic_delete_inode
);
1074 static void generic_forget_inode(struct inode
*inode
)
1076 struct super_block
*sb
= inode
->i_sb
;
1078 if (!hlist_unhashed(&inode
->i_hash
)) {
1079 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1080 list_move(&inode
->i_list
, &inode_unused
);
1081 inodes_stat
.nr_unused
++;
1082 if (sb
->s_flags
& MS_ACTIVE
) {
1083 spin_unlock(&inode_lock
);
1086 inode
->i_state
|= I_WILL_FREE
;
1087 spin_unlock(&inode_lock
);
1088 write_inode_now(inode
, 1);
1089 spin_lock(&inode_lock
);
1090 inode
->i_state
&= ~I_WILL_FREE
;
1091 inodes_stat
.nr_unused
--;
1092 hlist_del_init(&inode
->i_hash
);
1094 list_del_init(&inode
->i_list
);
1095 list_del_init(&inode
->i_sb_list
);
1096 inode
->i_state
|= I_FREEING
;
1097 inodes_stat
.nr_inodes
--;
1098 spin_unlock(&inode_lock
);
1099 if (inode
->i_data
.nrpages
)
1100 truncate_inode_pages(&inode
->i_data
, 0);
1102 wake_up_inode(inode
);
1103 destroy_inode(inode
);
1107 * Normal UNIX filesystem behaviour: delete the
1108 * inode when the usage count drops to zero, and
1111 void generic_drop_inode(struct inode
*inode
)
1113 if (!inode
->i_nlink
)
1114 generic_delete_inode(inode
);
1116 generic_forget_inode(inode
);
1119 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1122 * Called when we're dropping the last reference
1125 * Call the FS "drop()" function, defaulting to
1126 * the legacy UNIX filesystem behaviour..
1128 * NOTE! NOTE! NOTE! We're called with the inode lock
1129 * held, and the drop function is supposed to release
1132 static inline void iput_final(struct inode
*inode
)
1134 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1135 void (*drop
)(struct inode
*) = generic_drop_inode
;
1137 if (op
&& op
->drop_inode
)
1138 drop
= op
->drop_inode
;
1143 * iput - put an inode
1144 * @inode: inode to put
1146 * Puts an inode, dropping its usage count. If the inode use count hits
1147 * zero, the inode is then freed and may also be destroyed.
1149 * Consequently, iput() can sleep.
1151 void iput(struct inode
*inode
)
1154 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1156 BUG_ON(inode
->i_state
== I_CLEAR
);
1158 if (op
&& op
->put_inode
)
1159 op
->put_inode(inode
);
1161 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1166 EXPORT_SYMBOL(iput
);
1169 * bmap - find a block number in a file
1170 * @inode: inode of file
1171 * @block: block to find
1173 * Returns the block number on the device holding the inode that
1174 * is the disk block number for the block of the file requested.
1175 * That is, asked for block 4 of inode 1 the function will return the
1176 * disk block relative to the disk start that holds that block of the
1179 sector_t
bmap(struct inode
* inode
, sector_t block
)
1182 if (inode
->i_mapping
->a_ops
->bmap
)
1183 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1186 EXPORT_SYMBOL(bmap
);
1189 * touch_atime - update the access time
1190 * @mnt: mount the inode is accessed on
1191 * @dentry: dentry accessed
1193 * Update the accessed time on an inode and mark it for writeback.
1194 * This function automatically handles read only file systems and media,
1195 * as well as the "noatime" flag and inode specific "noatime" markers.
1197 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1199 struct inode
*inode
= dentry
->d_inode
;
1200 struct timespec now
;
1202 if (mnt
&& mnt_want_write(mnt
))
1204 if (inode
->i_flags
& S_NOATIME
)
1206 if (IS_NOATIME(inode
))
1208 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1212 * We may have a NULL vfsmount when coming from NFSD
1215 if (mnt
->mnt_flags
& MNT_NOATIME
)
1217 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1219 if (mnt
->mnt_flags
& MNT_RELATIME
) {
1221 * With relative atime, only update atime if the
1222 * previous atime is earlier than either the ctime or
1225 if (timespec_compare(&inode
->i_mtime
,
1226 &inode
->i_atime
) < 0 &&
1227 timespec_compare(&inode
->i_ctime
,
1228 &inode
->i_atime
) < 0)
1233 now
= current_fs_time(inode
->i_sb
);
1234 if (timespec_equal(&inode
->i_atime
, &now
))
1237 inode
->i_atime
= now
;
1238 mark_inode_dirty_sync(inode
);
1241 mnt_drop_write(mnt
);
1243 EXPORT_SYMBOL(touch_atime
);
1246 * file_update_time - update mtime and ctime time
1247 * @file: file accessed
1249 * Update the mtime and ctime members of an inode and mark the inode
1250 * for writeback. Note that this function is meant exclusively for
1251 * usage in the file write path of filesystems, and filesystems may
1252 * choose to explicitly ignore update via this function with the
1253 * S_NOCTIME inode flag, e.g. for network filesystem where these
1254 * timestamps are handled by the server.
1257 void file_update_time(struct file
*file
)
1259 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1260 struct timespec now
;
1264 if (IS_NOCMTIME(inode
))
1267 * Ideally, we want to guarantee that 'f_vfsmnt'
1268 * is non-NULL here. But, NFS exports need to
1269 * be fixed up before we can do that. So, check
1270 * it for now. - Dave Hansen
1273 err
= mnt_want_write(file
->f_vfsmnt
);
1277 now
= current_fs_time(inode
->i_sb
);
1278 if (!timespec_equal(&inode
->i_mtime
, &now
)) {
1279 inode
->i_mtime
= now
;
1283 if (!timespec_equal(&inode
->i_ctime
, &now
)) {
1284 inode
->i_ctime
= now
;
1289 mark_inode_dirty_sync(inode
);
1291 mnt_drop_write(file
->f_vfsmnt
);
1294 EXPORT_SYMBOL(file_update_time
);
1296 int inode_needs_sync(struct inode
*inode
)
1300 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1305 EXPORT_SYMBOL(inode_needs_sync
);
1307 int inode_wait(void *word
)
1314 * If we try to find an inode in the inode hash while it is being
1315 * deleted, we have to wait until the filesystem completes its
1316 * deletion before reporting that it isn't found. This function waits
1317 * until the deletion _might_ have completed. Callers are responsible
1318 * to recheck inode state.
1320 * It doesn't matter if I_LOCK is not set initially, a call to
1321 * wake_up_inode() after removing from the hash list will DTRT.
1323 * This is called with inode_lock held.
1325 static void __wait_on_freeing_inode(struct inode
*inode
)
1327 wait_queue_head_t
*wq
;
1328 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_LOCK
);
1329 wq
= bit_waitqueue(&inode
->i_state
, __I_LOCK
);
1330 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1331 spin_unlock(&inode_lock
);
1333 finish_wait(wq
, &wait
.wait
);
1334 spin_lock(&inode_lock
);
1338 * We rarely want to lock two inodes that do not have a parent/child
1339 * relationship (such as directory, child inode) simultaneously. The
1340 * vast majority of file systems should be able to get along fine
1341 * without this. Do not use these functions except as a last resort.
1343 void inode_double_lock(struct inode
*inode1
, struct inode
*inode2
)
1345 if (inode1
== NULL
|| inode2
== NULL
|| inode1
== inode2
) {
1347 mutex_lock(&inode1
->i_mutex
);
1349 mutex_lock(&inode2
->i_mutex
);
1353 if (inode1
< inode2
) {
1354 mutex_lock_nested(&inode1
->i_mutex
, I_MUTEX_PARENT
);
1355 mutex_lock_nested(&inode2
->i_mutex
, I_MUTEX_CHILD
);
1357 mutex_lock_nested(&inode2
->i_mutex
, I_MUTEX_PARENT
);
1358 mutex_lock_nested(&inode1
->i_mutex
, I_MUTEX_CHILD
);
1361 EXPORT_SYMBOL(inode_double_lock
);
1363 void inode_double_unlock(struct inode
*inode1
, struct inode
*inode2
)
1366 mutex_unlock(&inode1
->i_mutex
);
1368 if (inode2
&& inode2
!= inode1
)
1369 mutex_unlock(&inode2
->i_mutex
);
1371 EXPORT_SYMBOL(inode_double_unlock
);
1373 static __initdata
unsigned long ihash_entries
;
1374 static int __init
set_ihash_entries(char *str
)
1378 ihash_entries
= simple_strtoul(str
, &str
, 0);
1381 __setup("ihash_entries=", set_ihash_entries
);
1384 * Initialize the waitqueues and inode hash table.
1386 void __init
inode_init_early(void)
1390 /* If hashes are distributed across NUMA nodes, defer
1391 * hash allocation until vmalloc space is available.
1397 alloc_large_system_hash("Inode-cache",
1398 sizeof(struct hlist_head
),
1406 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1407 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1410 void __init
inode_init(void)
1414 /* inode slab cache */
1415 inode_cachep
= kmem_cache_create("inode_cache",
1416 sizeof(struct inode
),
1418 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1421 register_shrinker(&icache_shrinker
);
1423 /* Hash may have been set up in inode_init_early */
1428 alloc_large_system_hash("Inode-cache",
1429 sizeof(struct hlist_head
),
1437 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1438 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1441 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1443 inode
->i_mode
= mode
;
1444 if (S_ISCHR(mode
)) {
1445 inode
->i_fop
= &def_chr_fops
;
1446 inode
->i_rdev
= rdev
;
1447 } else if (S_ISBLK(mode
)) {
1448 inode
->i_fop
= &def_blk_fops
;
1449 inode
->i_rdev
= rdev
;
1450 } else if (S_ISFIFO(mode
))
1451 inode
->i_fop
= &def_fifo_fops
;
1452 else if (S_ISSOCK(mode
))
1453 inode
->i_fop
= &bad_sock_fops
;
1455 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o)\n",
1458 EXPORT_SYMBOL(init_special_inode
);