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
);
112 * inode_init_always - perform inode structure intialisation
113 * @sb - superblock inode belongs to.
114 * @inode - inode to initialise
116 * These are initializations that need to be done on every inode
117 * allocation as the fields are not initialised by slab allocation.
119 struct inode
*inode_init_always(struct super_block
*sb
, struct inode
*inode
)
121 static const struct address_space_operations empty_aops
;
122 static struct inode_operations empty_iops
;
123 static const struct file_operations empty_fops
;
125 struct address_space
* const mapping
= &inode
->i_data
;
128 inode
->i_blkbits
= sb
->s_blocksize_bits
;
130 atomic_set(&inode
->i_count
, 1);
131 inode
->i_op
= &empty_iops
;
132 inode
->i_fop
= &empty_fops
;
134 atomic_set(&inode
->i_writecount
, 0);
138 inode
->i_generation
= 0;
140 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
142 inode
->i_pipe
= NULL
;
143 inode
->i_bdev
= NULL
;
144 inode
->i_cdev
= NULL
;
146 inode
->dirtied_when
= 0;
147 if (security_inode_alloc(inode
)) {
148 if (inode
->i_sb
->s_op
->destroy_inode
)
149 inode
->i_sb
->s_op
->destroy_inode(inode
);
151 kmem_cache_free(inode_cachep
, (inode
));
155 spin_lock_init(&inode
->i_lock
);
156 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
158 mutex_init(&inode
->i_mutex
);
159 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
161 init_rwsem(&inode
->i_alloc_sem
);
162 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
164 mapping
->a_ops
= &empty_aops
;
165 mapping
->host
= inode
;
167 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_PAGECACHE
);
168 mapping
->assoc_mapping
= NULL
;
169 mapping
->backing_dev_info
= &default_backing_dev_info
;
170 mapping
->writeback_index
= 0;
173 * If the block_device provides a backing_dev_info for client
174 * inodes then use that. Otherwise the inode share the bdev's
178 struct backing_dev_info
*bdi
;
180 bdi
= sb
->s_bdev
->bd_inode_backing_dev_info
;
182 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
183 mapping
->backing_dev_info
= bdi
;
185 inode
->i_private
= NULL
;
186 inode
->i_mapping
= mapping
;
190 EXPORT_SYMBOL(inode_init_always
);
192 static struct inode
*alloc_inode(struct super_block
*sb
)
196 if (sb
->s_op
->alloc_inode
)
197 inode
= sb
->s_op
->alloc_inode(sb
);
199 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
202 return inode_init_always(sb
, inode
);
206 void destroy_inode(struct inode
*inode
)
208 BUG_ON(inode_has_buffers(inode
));
209 security_inode_free(inode
);
210 if (inode
->i_sb
->s_op
->destroy_inode
)
211 inode
->i_sb
->s_op
->destroy_inode(inode
);
213 kmem_cache_free(inode_cachep
, (inode
));
218 * These are initializations that only need to be done
219 * once, because the fields are idempotent across use
220 * of the inode, so let the slab aware of that.
222 void inode_init_once(struct inode
*inode
)
224 memset(inode
, 0, sizeof(*inode
));
225 INIT_HLIST_NODE(&inode
->i_hash
);
226 INIT_LIST_HEAD(&inode
->i_dentry
);
227 INIT_LIST_HEAD(&inode
->i_devices
);
228 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
229 spin_lock_init(&inode
->i_data
.tree_lock
);
230 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
231 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
232 spin_lock_init(&inode
->i_data
.private_lock
);
233 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
234 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
235 i_size_ordered_init(inode
);
236 #ifdef CONFIG_INOTIFY
237 INIT_LIST_HEAD(&inode
->inotify_watches
);
238 mutex_init(&inode
->inotify_mutex
);
242 EXPORT_SYMBOL(inode_init_once
);
244 static void init_once(void *foo
)
246 struct inode
* inode
= (struct inode
*) foo
;
248 inode_init_once(inode
);
252 * inode_lock must be held
254 void __iget(struct inode
* inode
)
256 if (atomic_read(&inode
->i_count
)) {
257 atomic_inc(&inode
->i_count
);
260 atomic_inc(&inode
->i_count
);
261 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
262 list_move(&inode
->i_list
, &inode_in_use
);
263 inodes_stat
.nr_unused
--;
267 * clear_inode - clear an inode
268 * @inode: inode to clear
270 * This is called by the filesystem to tell us
271 * that the inode is no longer useful. We just
272 * terminate it with extreme prejudice.
274 void clear_inode(struct inode
*inode
)
277 invalidate_inode_buffers(inode
);
279 BUG_ON(inode
->i_data
.nrpages
);
280 BUG_ON(!(inode
->i_state
& I_FREEING
));
281 BUG_ON(inode
->i_state
& I_CLEAR
);
282 inode_sync_wait(inode
);
284 if (inode
->i_sb
->s_op
->clear_inode
)
285 inode
->i_sb
->s_op
->clear_inode(inode
);
286 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
288 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
290 inode
->i_state
= I_CLEAR
;
293 EXPORT_SYMBOL(clear_inode
);
296 * dispose_list - dispose of the contents of a local list
297 * @head: the head of the list to free
299 * Dispose-list gets a local list with local inodes in it, so it doesn't
300 * need to worry about list corruption and SMP locks.
302 static void dispose_list(struct list_head
*head
)
306 while (!list_empty(head
)) {
309 inode
= list_first_entry(head
, struct inode
, i_list
);
310 list_del(&inode
->i_list
);
312 if (inode
->i_data
.nrpages
)
313 truncate_inode_pages(&inode
->i_data
, 0);
316 spin_lock(&inode_lock
);
317 hlist_del_init(&inode
->i_hash
);
318 list_del_init(&inode
->i_sb_list
);
319 spin_unlock(&inode_lock
);
321 wake_up_inode(inode
);
322 destroy_inode(inode
);
325 spin_lock(&inode_lock
);
326 inodes_stat
.nr_inodes
-= nr_disposed
;
327 spin_unlock(&inode_lock
);
331 * Invalidate all inodes for a device.
333 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
335 struct list_head
*next
;
336 int busy
= 0, count
= 0;
340 struct list_head
* tmp
= next
;
341 struct inode
* inode
;
344 * We can reschedule here without worrying about the list's
345 * consistency because the per-sb list of inodes must not
346 * change during umount anymore, and because iprune_mutex keeps
347 * shrink_icache_memory() away.
349 cond_resched_lock(&inode_lock
);
354 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
355 invalidate_inode_buffers(inode
);
356 if (!atomic_read(&inode
->i_count
)) {
357 list_move(&inode
->i_list
, dispose
);
358 inode
->i_state
|= I_FREEING
;
364 /* only unused inodes may be cached with i_count zero */
365 inodes_stat
.nr_unused
-= count
;
370 * invalidate_inodes - discard the inodes on a device
373 * Discard all of the inodes for a given superblock. If the discard
374 * fails because there are busy inodes then a non zero value is returned.
375 * If the discard is successful all the inodes have been discarded.
377 int invalidate_inodes(struct super_block
* sb
)
380 LIST_HEAD(throw_away
);
382 mutex_lock(&iprune_mutex
);
383 spin_lock(&inode_lock
);
384 inotify_unmount_inodes(&sb
->s_inodes
);
385 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
386 spin_unlock(&inode_lock
);
388 dispose_list(&throw_away
);
389 mutex_unlock(&iprune_mutex
);
394 EXPORT_SYMBOL(invalidate_inodes
);
396 static int can_unuse(struct inode
*inode
)
400 if (inode_has_buffers(inode
))
402 if (atomic_read(&inode
->i_count
))
404 if (inode
->i_data
.nrpages
)
410 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
411 * a temporary list and then are freed outside inode_lock by dispose_list().
413 * Any inodes which are pinned purely because of attached pagecache have their
414 * pagecache removed. We expect the final iput() on that inode to add it to
415 * the front of the inode_unused list. So look for it there and if the
416 * inode is still freeable, proceed. The right inode is found 99.9% of the
417 * time in testing on a 4-way.
419 * If the inode has metadata buffers attached to mapping->private_list then
420 * try to remove them.
422 static void prune_icache(int nr_to_scan
)
427 unsigned long reap
= 0;
429 mutex_lock(&iprune_mutex
);
430 spin_lock(&inode_lock
);
431 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
434 if (list_empty(&inode_unused
))
437 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
439 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
440 list_move(&inode
->i_list
, &inode_unused
);
443 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
445 spin_unlock(&inode_lock
);
446 if (remove_inode_buffers(inode
))
447 reap
+= invalidate_mapping_pages(&inode
->i_data
,
450 spin_lock(&inode_lock
);
452 if (inode
!= list_entry(inode_unused
.next
,
453 struct inode
, i_list
))
454 continue; /* wrong inode or list_empty */
455 if (!can_unuse(inode
))
458 list_move(&inode
->i_list
, &freeable
);
459 inode
->i_state
|= I_FREEING
;
462 inodes_stat
.nr_unused
-= nr_pruned
;
463 if (current_is_kswapd())
464 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
466 __count_vm_events(PGINODESTEAL
, reap
);
467 spin_unlock(&inode_lock
);
469 dispose_list(&freeable
);
470 mutex_unlock(&iprune_mutex
);
474 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
475 * "unused" means that no dentries are referring to the inodes: the files are
476 * not open and the dcache references to those inodes have already been
479 * This function is passed the number of inodes to scan, and it returns the
480 * total number of remaining possibly-reclaimable inodes.
482 static int shrink_icache_memory(int nr
, gfp_t gfp_mask
)
486 * Nasty deadlock avoidance. We may hold various FS locks,
487 * and we don't want to recurse into the FS that called us
488 * in clear_inode() and friends..
490 if (!(gfp_mask
& __GFP_FS
))
494 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
497 static struct shrinker icache_shrinker
= {
498 .shrink
= shrink_icache_memory
,
499 .seeks
= DEFAULT_SEEKS
,
502 static void __wait_on_freeing_inode(struct inode
*inode
);
504 * Called with the inode lock held.
505 * NOTE: we are not increasing the inode-refcount, you must call __iget()
506 * by hand after calling find_inode now! This simplifies iunique and won't
507 * add any additional branch in the common code.
509 static struct inode
* find_inode(struct super_block
* sb
, struct hlist_head
*head
, int (*test
)(struct inode
*, void *), void *data
)
511 struct hlist_node
*node
;
512 struct inode
* inode
= NULL
;
515 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
516 if (inode
->i_sb
!= sb
)
518 if (!test(inode
, data
))
520 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
521 __wait_on_freeing_inode(inode
);
526 return node
? inode
: NULL
;
530 * find_inode_fast is the fast path version of find_inode, see the comment at
531 * iget_locked for details.
533 static struct inode
* find_inode_fast(struct super_block
* sb
, struct hlist_head
*head
, unsigned long ino
)
535 struct hlist_node
*node
;
536 struct inode
* inode
= NULL
;
539 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
540 if (inode
->i_ino
!= ino
)
542 if (inode
->i_sb
!= sb
)
544 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
545 __wait_on_freeing_inode(inode
);
550 return node
? inode
: NULL
;
553 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
557 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
559 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
560 return tmp
& I_HASHMASK
;
564 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
567 inodes_stat
.nr_inodes
++;
568 list_add(&inode
->i_list
, &inode_in_use
);
569 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
571 hlist_add_head(&inode
->i_hash
, head
);
575 * inode_add_to_lists - add a new inode to relevant lists
576 * @sb - superblock inode belongs to.
577 * @inode - inode to mark in use
579 * When an inode is allocated it needs to be accounted for, added to the in use
580 * list, the owning superblock and the inode hash. This needs to be done under
581 * the inode_lock, so export a function to do this rather than the inode lock
582 * itself. We calculate the hash list to add to here so it is all internal
583 * which requires the caller to have already set up the inode number in the
586 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
588 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
590 spin_lock(&inode_lock
);
591 __inode_add_to_lists(sb
, head
, inode
);
592 spin_unlock(&inode_lock
);
594 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
597 * new_inode - obtain an inode
600 * Allocates a new inode for given superblock. The default gfp_mask
601 * for allocations related to inode->i_mapping is GFP_HIGHUSER_PAGECACHE.
602 * If HIGHMEM pages are unsuitable or it is known that pages allocated
603 * for the page cache are not reclaimable or migratable,
604 * mapping_set_gfp_mask() must be called with suitable flags on the
605 * newly created inode's mapping
608 struct inode
*new_inode(struct super_block
*sb
)
611 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
612 * error if st_ino won't fit in target struct field. Use 32bit counter
613 * here to attempt to avoid that.
615 static unsigned int last_ino
;
616 struct inode
* inode
;
618 spin_lock_prefetch(&inode_lock
);
620 inode
= alloc_inode(sb
);
622 spin_lock(&inode_lock
);
623 __inode_add_to_lists(sb
, NULL
, inode
);
624 inode
->i_ino
= ++last_ino
;
626 spin_unlock(&inode_lock
);
631 EXPORT_SYMBOL(new_inode
);
633 void unlock_new_inode(struct inode
*inode
)
635 #ifdef CONFIG_DEBUG_LOCK_ALLOC
636 if (inode
->i_mode
& S_IFDIR
) {
637 struct file_system_type
*type
= inode
->i_sb
->s_type
;
640 * ensure nobody is actually holding i_mutex
642 mutex_destroy(&inode
->i_mutex
);
643 mutex_init(&inode
->i_mutex
);
644 lockdep_set_class(&inode
->i_mutex
, &type
->i_mutex_dir_key
);
648 * This is special! We do not need the spinlock
649 * when clearing I_LOCK, because we're guaranteed
650 * that nobody else tries to do anything about the
651 * state of the inode when it is locked, as we
652 * just created it (so there can be no old holders
653 * that haven't tested I_LOCK).
655 inode
->i_state
&= ~(I_LOCK
|I_NEW
);
656 wake_up_inode(inode
);
659 EXPORT_SYMBOL(unlock_new_inode
);
662 * This is called without the inode lock held.. Be careful.
664 * We no longer cache the sb_flags in i_flags - see fs.h
665 * -- rmk@arm.uk.linux.org
667 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
)
669 struct inode
* inode
;
671 inode
= alloc_inode(sb
);
675 spin_lock(&inode_lock
);
676 /* We released the lock, so.. */
677 old
= find_inode(sb
, head
, test
, data
);
679 if (set(inode
, data
))
682 __inode_add_to_lists(sb
, head
, inode
);
683 inode
->i_state
= I_LOCK
|I_NEW
;
684 spin_unlock(&inode_lock
);
686 /* Return the locked inode with I_NEW set, the
687 * caller is responsible for filling in the contents
693 * Uhhuh, somebody else created the same inode under
694 * us. Use the old inode instead of the one we just
698 spin_unlock(&inode_lock
);
699 destroy_inode(inode
);
701 wait_on_inode(inode
);
706 spin_unlock(&inode_lock
);
707 destroy_inode(inode
);
712 * get_new_inode_fast is the fast path version of get_new_inode, see the
713 * comment at iget_locked for details.
715 static struct inode
* get_new_inode_fast(struct super_block
*sb
, struct hlist_head
*head
, unsigned long ino
)
717 struct inode
* inode
;
719 inode
= alloc_inode(sb
);
723 spin_lock(&inode_lock
);
724 /* We released the lock, so.. */
725 old
= find_inode_fast(sb
, head
, ino
);
728 __inode_add_to_lists(sb
, head
, inode
);
729 inode
->i_state
= I_LOCK
|I_NEW
;
730 spin_unlock(&inode_lock
);
732 /* Return the locked inode with I_NEW set, the
733 * caller is responsible for filling in the contents
739 * Uhhuh, somebody else created the same inode under
740 * us. Use the old inode instead of the one we just
744 spin_unlock(&inode_lock
);
745 destroy_inode(inode
);
747 wait_on_inode(inode
);
753 * iunique - get a unique inode number
755 * @max_reserved: highest reserved inode number
757 * Obtain an inode number that is unique on the system for a given
758 * superblock. This is used by file systems that have no natural
759 * permanent inode numbering system. An inode number is returned that
760 * is higher than the reserved limit but unique.
763 * With a large number of inodes live on the file system this function
764 * currently becomes quite slow.
766 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
769 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
770 * error if st_ino won't fit in target struct field. Use 32bit counter
771 * here to attempt to avoid that.
773 static unsigned int counter
;
775 struct hlist_head
*head
;
778 spin_lock(&inode_lock
);
780 if (counter
<= max_reserved
)
781 counter
= max_reserved
+ 1;
783 head
= inode_hashtable
+ hash(sb
, res
);
784 inode
= find_inode_fast(sb
, head
, res
);
785 } while (inode
!= NULL
);
786 spin_unlock(&inode_lock
);
790 EXPORT_SYMBOL(iunique
);
792 struct inode
*igrab(struct inode
*inode
)
794 spin_lock(&inode_lock
);
795 if (!(inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)))
799 * Handle the case where s_op->clear_inode is not been
800 * called yet, and somebody is calling igrab
801 * while the inode is getting freed.
804 spin_unlock(&inode_lock
);
808 EXPORT_SYMBOL(igrab
);
811 * ifind - internal function, you want ilookup5() or iget5().
812 * @sb: super block of file system to search
813 * @head: the head of the list to search
814 * @test: callback used for comparisons between inodes
815 * @data: opaque data pointer to pass to @test
816 * @wait: if true wait for the inode to be unlocked, if false do not
818 * ifind() searches for the inode specified by @data in the inode
819 * cache. This is a generalized version of ifind_fast() for file systems where
820 * the inode number is not sufficient for unique identification of an inode.
822 * If the inode is in the cache, the inode is returned with an incremented
825 * Otherwise NULL is returned.
827 * Note, @test is called with the inode_lock held, so can't sleep.
829 static struct inode
*ifind(struct super_block
*sb
,
830 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
831 void *data
, const int wait
)
835 spin_lock(&inode_lock
);
836 inode
= find_inode(sb
, head
, test
, data
);
839 spin_unlock(&inode_lock
);
841 wait_on_inode(inode
);
844 spin_unlock(&inode_lock
);
849 * ifind_fast - internal function, you want ilookup() or iget().
850 * @sb: super block of file system to search
851 * @head: head of the list to search
852 * @ino: inode number to search for
854 * ifind_fast() searches for the inode @ino in the inode cache. This is for
855 * file systems where the inode number is sufficient for unique identification
858 * If the inode is in the cache, the inode is returned with an incremented
861 * Otherwise NULL is returned.
863 static struct inode
*ifind_fast(struct super_block
*sb
,
864 struct hlist_head
*head
, unsigned long ino
)
868 spin_lock(&inode_lock
);
869 inode
= find_inode_fast(sb
, head
, ino
);
872 spin_unlock(&inode_lock
);
873 wait_on_inode(inode
);
876 spin_unlock(&inode_lock
);
881 * ilookup5_nowait - search for an inode in the inode cache
882 * @sb: super block of file system to search
883 * @hashval: hash value (usually inode number) to search for
884 * @test: callback used for comparisons between inodes
885 * @data: opaque data pointer to pass to @test
887 * ilookup5() uses ifind() to search for the inode specified by @hashval and
888 * @data in the inode cache. This is a generalized version of ilookup() for
889 * file systems where the inode number is not sufficient for unique
890 * identification of an inode.
892 * If the inode is in the cache, the inode is returned with an incremented
893 * reference count. Note, the inode lock is not waited upon so you have to be
894 * very careful what you do with the returned inode. You probably should be
895 * using ilookup5() instead.
897 * Otherwise NULL is returned.
899 * Note, @test is called with the inode_lock held, so can't sleep.
901 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
902 int (*test
)(struct inode
*, void *), void *data
)
904 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
906 return ifind(sb
, head
, test
, data
, 0);
909 EXPORT_SYMBOL(ilookup5_nowait
);
912 * ilookup5 - search for an inode in the inode cache
913 * @sb: super block of file system to search
914 * @hashval: hash value (usually inode number) to search for
915 * @test: callback used for comparisons between inodes
916 * @data: opaque data pointer to pass to @test
918 * ilookup5() uses ifind() to search for the inode specified by @hashval and
919 * @data in the inode cache. This is a generalized version of ilookup() for
920 * file systems where the inode number is not sufficient for unique
921 * identification of an inode.
923 * If the inode is in the cache, the inode lock is waited upon and the inode is
924 * returned with an incremented reference count.
926 * Otherwise NULL is returned.
928 * Note, @test is called with the inode_lock held, so can't sleep.
930 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
931 int (*test
)(struct inode
*, void *), void *data
)
933 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
935 return ifind(sb
, head
, test
, data
, 1);
938 EXPORT_SYMBOL(ilookup5
);
941 * ilookup - search for an inode in the inode cache
942 * @sb: super block of file system to search
943 * @ino: inode number to search for
945 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
946 * This is for file systems where the inode number is sufficient for unique
947 * identification of an inode.
949 * If the inode is in the cache, the inode is returned with an incremented
952 * Otherwise NULL is returned.
954 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
956 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
958 return ifind_fast(sb
, head
, ino
);
961 EXPORT_SYMBOL(ilookup
);
964 * iget5_locked - obtain an inode from a mounted file system
965 * @sb: super block of file system
966 * @hashval: hash value (usually inode number) to get
967 * @test: callback used for comparisons between inodes
968 * @set: callback used to initialize a new struct inode
969 * @data: opaque data pointer to pass to @test and @set
971 * iget5_locked() uses ifind() to search for the inode specified by @hashval
972 * and @data in the inode cache and if present it is returned with an increased
973 * reference count. This is a generalized version of iget_locked() for file
974 * systems where the inode number is not sufficient for unique identification
977 * If the inode is not in cache, get_new_inode() is called to allocate a new
978 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
979 * file system gets to fill it in before unlocking it via unlock_new_inode().
981 * Note both @test and @set are called with the inode_lock held, so can't sleep.
983 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
984 int (*test
)(struct inode
*, void *),
985 int (*set
)(struct inode
*, void *), void *data
)
987 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
990 inode
= ifind(sb
, head
, test
, data
, 1);
994 * get_new_inode() will do the right thing, re-trying the search
995 * in case it had to block at any point.
997 return get_new_inode(sb
, head
, test
, set
, data
);
1000 EXPORT_SYMBOL(iget5_locked
);
1003 * iget_locked - obtain an inode from a mounted file system
1004 * @sb: super block of file system
1005 * @ino: inode number to get
1007 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1008 * the inode cache and if present it is returned with an increased reference
1009 * count. This is for file systems where the inode number is sufficient for
1010 * unique identification of an inode.
1012 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1013 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1014 * The file system gets to fill it in before unlocking it via
1015 * unlock_new_inode().
1017 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1019 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1020 struct inode
*inode
;
1022 inode
= ifind_fast(sb
, head
, ino
);
1026 * get_new_inode_fast() will do the right thing, re-trying the search
1027 * in case it had to block at any point.
1029 return get_new_inode_fast(sb
, head
, ino
);
1032 EXPORT_SYMBOL(iget_locked
);
1035 * __insert_inode_hash - hash an inode
1036 * @inode: unhashed inode
1037 * @hashval: unsigned long value used to locate this object in the
1040 * Add an inode to the inode hash for this superblock.
1042 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1044 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1045 spin_lock(&inode_lock
);
1046 hlist_add_head(&inode
->i_hash
, head
);
1047 spin_unlock(&inode_lock
);
1050 EXPORT_SYMBOL(__insert_inode_hash
);
1053 * remove_inode_hash - remove an inode from the hash
1054 * @inode: inode to unhash
1056 * Remove an inode from the superblock.
1058 void remove_inode_hash(struct inode
*inode
)
1060 spin_lock(&inode_lock
);
1061 hlist_del_init(&inode
->i_hash
);
1062 spin_unlock(&inode_lock
);
1065 EXPORT_SYMBOL(remove_inode_hash
);
1068 * Tell the filesystem that this inode is no longer of any interest and should
1069 * be completely destroyed.
1071 * We leave the inode in the inode hash table until *after* the filesystem's
1072 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1073 * instigate) will always find up-to-date information either in the hash or on
1076 * I_FREEING is set so that no-one will take a new reference to the inode while
1077 * it is being deleted.
1079 void generic_delete_inode(struct inode
*inode
)
1081 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1083 list_del_init(&inode
->i_list
);
1084 list_del_init(&inode
->i_sb_list
);
1085 inode
->i_state
|= I_FREEING
;
1086 inodes_stat
.nr_inodes
--;
1087 spin_unlock(&inode_lock
);
1089 security_inode_delete(inode
);
1091 if (op
->delete_inode
) {
1092 void (*delete)(struct inode
*) = op
->delete_inode
;
1093 if (!is_bad_inode(inode
))
1095 /* Filesystems implementing their own
1096 * s_op->delete_inode are required to call
1097 * truncate_inode_pages and clear_inode()
1101 truncate_inode_pages(&inode
->i_data
, 0);
1104 spin_lock(&inode_lock
);
1105 hlist_del_init(&inode
->i_hash
);
1106 spin_unlock(&inode_lock
);
1107 wake_up_inode(inode
);
1108 BUG_ON(inode
->i_state
!= I_CLEAR
);
1109 destroy_inode(inode
);
1112 EXPORT_SYMBOL(generic_delete_inode
);
1114 static void generic_forget_inode(struct inode
*inode
)
1116 struct super_block
*sb
= inode
->i_sb
;
1118 if (!hlist_unhashed(&inode
->i_hash
)) {
1119 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1120 list_move(&inode
->i_list
, &inode_unused
);
1121 inodes_stat
.nr_unused
++;
1122 if (sb
->s_flags
& MS_ACTIVE
) {
1123 spin_unlock(&inode_lock
);
1126 inode
->i_state
|= I_WILL_FREE
;
1127 spin_unlock(&inode_lock
);
1128 write_inode_now(inode
, 1);
1129 spin_lock(&inode_lock
);
1130 inode
->i_state
&= ~I_WILL_FREE
;
1131 inodes_stat
.nr_unused
--;
1132 hlist_del_init(&inode
->i_hash
);
1134 list_del_init(&inode
->i_list
);
1135 list_del_init(&inode
->i_sb_list
);
1136 inode
->i_state
|= I_FREEING
;
1137 inodes_stat
.nr_inodes
--;
1138 spin_unlock(&inode_lock
);
1139 if (inode
->i_data
.nrpages
)
1140 truncate_inode_pages(&inode
->i_data
, 0);
1142 wake_up_inode(inode
);
1143 destroy_inode(inode
);
1147 * Normal UNIX filesystem behaviour: delete the
1148 * inode when the usage count drops to zero, and
1151 void generic_drop_inode(struct inode
*inode
)
1153 if (!inode
->i_nlink
)
1154 generic_delete_inode(inode
);
1156 generic_forget_inode(inode
);
1159 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1162 * Called when we're dropping the last reference
1165 * Call the FS "drop()" function, defaulting to
1166 * the legacy UNIX filesystem behaviour..
1168 * NOTE! NOTE! NOTE! We're called with the inode lock
1169 * held, and the drop function is supposed to release
1172 static inline void iput_final(struct inode
*inode
)
1174 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1175 void (*drop
)(struct inode
*) = generic_drop_inode
;
1177 if (op
&& op
->drop_inode
)
1178 drop
= op
->drop_inode
;
1183 * iput - put an inode
1184 * @inode: inode to put
1186 * Puts an inode, dropping its usage count. If the inode use count hits
1187 * zero, the inode is then freed and may also be destroyed.
1189 * Consequently, iput() can sleep.
1191 void iput(struct inode
*inode
)
1194 BUG_ON(inode
->i_state
== I_CLEAR
);
1196 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1201 EXPORT_SYMBOL(iput
);
1204 * bmap - find a block number in a file
1205 * @inode: inode of file
1206 * @block: block to find
1208 * Returns the block number on the device holding the inode that
1209 * is the disk block number for the block of the file requested.
1210 * That is, asked for block 4 of inode 1 the function will return the
1211 * disk block relative to the disk start that holds that block of the
1214 sector_t
bmap(struct inode
* inode
, sector_t block
)
1217 if (inode
->i_mapping
->a_ops
->bmap
)
1218 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1221 EXPORT_SYMBOL(bmap
);
1224 * touch_atime - update the access time
1225 * @mnt: mount the inode is accessed on
1226 * @dentry: dentry accessed
1228 * Update the accessed time on an inode and mark it for writeback.
1229 * This function automatically handles read only file systems and media,
1230 * as well as the "noatime" flag and inode specific "noatime" markers.
1232 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1234 struct inode
*inode
= dentry
->d_inode
;
1235 struct timespec now
;
1237 if (mnt_want_write(mnt
))
1239 if (inode
->i_flags
& S_NOATIME
)
1241 if (IS_NOATIME(inode
))
1243 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1246 if (mnt
->mnt_flags
& MNT_NOATIME
)
1248 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1250 if (mnt
->mnt_flags
& MNT_RELATIME
) {
1252 * With relative atime, only update atime if the previous
1253 * atime is earlier than either the ctime or mtime.
1255 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) < 0 &&
1256 timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) < 0)
1260 now
= current_fs_time(inode
->i_sb
);
1261 if (timespec_equal(&inode
->i_atime
, &now
))
1264 inode
->i_atime
= now
;
1265 mark_inode_dirty_sync(inode
);
1267 mnt_drop_write(mnt
);
1269 EXPORT_SYMBOL(touch_atime
);
1272 * file_update_time - update mtime and ctime time
1273 * @file: file accessed
1275 * Update the mtime and ctime members of an inode and mark the inode
1276 * for writeback. Note that this function is meant exclusively for
1277 * usage in the file write path of filesystems, and filesystems may
1278 * choose to explicitly ignore update via this function with the
1279 * S_NOCTIME inode flag, e.g. for network filesystem where these
1280 * timestamps are handled by the server.
1283 void file_update_time(struct file
*file
)
1285 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1286 struct timespec now
;
1290 if (IS_NOCMTIME(inode
))
1293 err
= mnt_want_write(file
->f_path
.mnt
);
1297 now
= current_fs_time(inode
->i_sb
);
1298 if (!timespec_equal(&inode
->i_mtime
, &now
)) {
1299 inode
->i_mtime
= now
;
1303 if (!timespec_equal(&inode
->i_ctime
, &now
)) {
1304 inode
->i_ctime
= now
;
1308 if (IS_I_VERSION(inode
)) {
1309 inode_inc_iversion(inode
);
1314 mark_inode_dirty_sync(inode
);
1315 mnt_drop_write(file
->f_path
.mnt
);
1318 EXPORT_SYMBOL(file_update_time
);
1320 int inode_needs_sync(struct inode
*inode
)
1324 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1329 EXPORT_SYMBOL(inode_needs_sync
);
1331 int inode_wait(void *word
)
1338 * If we try to find an inode in the inode hash while it is being
1339 * deleted, we have to wait until the filesystem completes its
1340 * deletion before reporting that it isn't found. This function waits
1341 * until the deletion _might_ have completed. Callers are responsible
1342 * to recheck inode state.
1344 * It doesn't matter if I_LOCK is not set initially, a call to
1345 * wake_up_inode() after removing from the hash list will DTRT.
1347 * This is called with inode_lock held.
1349 static void __wait_on_freeing_inode(struct inode
*inode
)
1351 wait_queue_head_t
*wq
;
1352 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_LOCK
);
1353 wq
= bit_waitqueue(&inode
->i_state
, __I_LOCK
);
1354 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1355 spin_unlock(&inode_lock
);
1357 finish_wait(wq
, &wait
.wait
);
1358 spin_lock(&inode_lock
);
1362 * We rarely want to lock two inodes that do not have a parent/child
1363 * relationship (such as directory, child inode) simultaneously. The
1364 * vast majority of file systems should be able to get along fine
1365 * without this. Do not use these functions except as a last resort.
1367 void inode_double_lock(struct inode
*inode1
, struct inode
*inode2
)
1369 if (inode1
== NULL
|| inode2
== NULL
|| inode1
== inode2
) {
1371 mutex_lock(&inode1
->i_mutex
);
1373 mutex_lock(&inode2
->i_mutex
);
1377 if (inode1
< inode2
) {
1378 mutex_lock_nested(&inode1
->i_mutex
, I_MUTEX_PARENT
);
1379 mutex_lock_nested(&inode2
->i_mutex
, I_MUTEX_CHILD
);
1381 mutex_lock_nested(&inode2
->i_mutex
, I_MUTEX_PARENT
);
1382 mutex_lock_nested(&inode1
->i_mutex
, I_MUTEX_CHILD
);
1385 EXPORT_SYMBOL(inode_double_lock
);
1387 void inode_double_unlock(struct inode
*inode1
, struct inode
*inode2
)
1390 mutex_unlock(&inode1
->i_mutex
);
1392 if (inode2
&& inode2
!= inode1
)
1393 mutex_unlock(&inode2
->i_mutex
);
1395 EXPORT_SYMBOL(inode_double_unlock
);
1397 static __initdata
unsigned long ihash_entries
;
1398 static int __init
set_ihash_entries(char *str
)
1402 ihash_entries
= simple_strtoul(str
, &str
, 0);
1405 __setup("ihash_entries=", set_ihash_entries
);
1408 * Initialize the waitqueues and inode hash table.
1410 void __init
inode_init_early(void)
1414 /* If hashes are distributed across NUMA nodes, defer
1415 * hash allocation until vmalloc space is available.
1421 alloc_large_system_hash("Inode-cache",
1422 sizeof(struct hlist_head
),
1430 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1431 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1434 void __init
inode_init(void)
1438 /* inode slab cache */
1439 inode_cachep
= kmem_cache_create("inode_cache",
1440 sizeof(struct inode
),
1442 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1445 register_shrinker(&icache_shrinker
);
1447 /* Hash may have been set up in inode_init_early */
1452 alloc_large_system_hash("Inode-cache",
1453 sizeof(struct hlist_head
),
1461 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1462 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1465 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1467 inode
->i_mode
= mode
;
1468 if (S_ISCHR(mode
)) {
1469 inode
->i_fop
= &def_chr_fops
;
1470 inode
->i_rdev
= rdev
;
1471 } else if (S_ISBLK(mode
)) {
1472 inode
->i_fop
= &def_blk_fops
;
1473 inode
->i_rdev
= rdev
;
1474 } else if (S_ISFIFO(mode
))
1475 inode
->i_fop
= &def_fifo_fops
;
1476 else if (S_ISSOCK(mode
))
1477 inode
->i_fop
= &bad_sock_fops
;
1479 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o)\n",
1482 EXPORT_SYMBOL(init_special_inode
);