4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/slab.h>
12 #include <linux/writeback.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/wait.h>
16 #include <linux/rwsem.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/fsnotify.h>
24 #include <linux/mount.h>
25 #include <linux/async.h>
26 #include <linux/posix_acl.h>
27 #include <linux/ima.h>
30 * This is needed for the following functions:
32 * - invalidate_inode_buffers
35 * FIXME: remove all knowledge of the buffer layer from this file
37 #include <linux/buffer_head.h>
40 * New inode.c implementation.
42 * This implementation has the basic premise of trying
43 * to be extremely low-overhead and SMP-safe, yet be
44 * simple enough to be "obviously correct".
49 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
51 /* #define INODE_PARANOIA 1 */
52 /* #define INODE_DEBUG 1 */
55 * Inode lookup is no longer as critical as it used to be:
56 * most of the lookups are going to be through the dcache.
58 #define I_HASHBITS i_hash_shift
59 #define I_HASHMASK i_hash_mask
61 static unsigned int i_hash_mask __read_mostly
;
62 static unsigned int i_hash_shift __read_mostly
;
65 * Each inode can be on two separate lists. One is
66 * the hash list of the inode, used for lookups. The
67 * other linked list is the "type" list:
68 * "in_use" - valid inode, i_count > 0, i_nlink > 0
69 * "dirty" - as "in_use" but also dirty
70 * "unused" - valid inode, i_count = 0
72 * A "dirty" list is maintained for each super block,
73 * allowing for low-overhead inode sync() operations.
76 LIST_HEAD(inode_in_use
);
77 LIST_HEAD(inode_unused
);
78 static struct hlist_head
*inode_hashtable __read_mostly
;
81 * A simple spinlock to protect the list manipulations.
83 * NOTE! You also have to own the lock if you change
84 * the i_state of an inode while it is in use..
86 DEFINE_SPINLOCK(inode_lock
);
89 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
90 * icache shrinking path, and the umount path. Without this exclusion,
91 * by the time prune_icache calls iput for the inode whose pages it has
92 * been invalidating, or by the time it calls clear_inode & destroy_inode
93 * from its final dispose_list, the struct super_block they refer to
94 * (for inode->i_sb->s_op) may already have been freed and reused.
96 * We make this an rwsem because the fastpath is icache shrinking. In
97 * some cases a filesystem may be doing a significant amount of work in
98 * its inode reclaim code, so this should improve parallelism.
100 static DECLARE_RWSEM(iprune_sem
);
103 * Statistics gathering..
105 struct inodes_stat_t inodes_stat
;
107 static struct kmem_cache
*inode_cachep __read_mostly
;
109 static void wake_up_inode(struct inode
*inode
)
112 * Prevent speculative execution through spin_unlock(&inode_lock);
115 wake_up_bit(&inode
->i_state
, __I_NEW
);
119 * inode_init_always - perform inode structure intialisation
120 * @sb: superblock inode belongs to
121 * @inode: inode to initialise
123 * These are initializations that need to be done on every inode
124 * allocation as the fields are not initialised by slab allocation.
126 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
128 static const struct address_space_operations empty_aops
;
129 static const struct inode_operations empty_iops
;
130 static const struct file_operations empty_fops
;
131 struct address_space
*const mapping
= &inode
->i_data
;
134 inode
->i_blkbits
= sb
->s_blocksize_bits
;
136 atomic_set(&inode
->i_count
, 1);
137 inode
->i_op
= &empty_iops
;
138 inode
->i_fop
= &empty_fops
;
142 atomic_set(&inode
->i_writecount
, 0);
146 inode
->i_generation
= 0;
148 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
150 inode
->i_pipe
= NULL
;
151 inode
->i_bdev
= NULL
;
152 inode
->i_cdev
= NULL
;
154 inode
->dirtied_when
= 0;
156 if (security_inode_alloc(inode
))
158 spin_lock_init(&inode
->i_lock
);
159 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
161 mutex_init(&inode
->i_mutex
);
162 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
164 init_rwsem(&inode
->i_alloc_sem
);
165 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
167 mapping
->a_ops
= &empty_aops
;
168 mapping
->host
= inode
;
170 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
171 mapping
->assoc_mapping
= NULL
;
172 mapping
->backing_dev_info
= &default_backing_dev_info
;
173 mapping
->writeback_index
= 0;
176 * If the block_device provides a backing_dev_info for client
177 * inodes then use that. Otherwise the inode share the bdev's
181 struct backing_dev_info
*bdi
;
183 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
184 mapping
->backing_dev_info
= bdi
;
186 inode
->i_private
= NULL
;
187 inode
->i_mapping
= mapping
;
188 #ifdef CONFIG_FS_POSIX_ACL
189 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
192 #ifdef CONFIG_FSNOTIFY
193 inode
->i_fsnotify_mask
= 0;
200 EXPORT_SYMBOL(inode_init_always
);
202 static struct inode
*alloc_inode(struct super_block
*sb
)
206 if (sb
->s_op
->alloc_inode
)
207 inode
= sb
->s_op
->alloc_inode(sb
);
209 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
214 if (unlikely(inode_init_always(sb
, inode
))) {
215 if (inode
->i_sb
->s_op
->destroy_inode
)
216 inode
->i_sb
->s_op
->destroy_inode(inode
);
218 kmem_cache_free(inode_cachep
, inode
);
225 void __destroy_inode(struct inode
*inode
)
227 BUG_ON(inode_has_buffers(inode
));
228 security_inode_free(inode
);
229 fsnotify_inode_delete(inode
);
230 #ifdef CONFIG_FS_POSIX_ACL
231 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
232 posix_acl_release(inode
->i_acl
);
233 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
234 posix_acl_release(inode
->i_default_acl
);
237 EXPORT_SYMBOL(__destroy_inode
);
239 void destroy_inode(struct inode
*inode
)
241 __destroy_inode(inode
);
242 if (inode
->i_sb
->s_op
->destroy_inode
)
243 inode
->i_sb
->s_op
->destroy_inode(inode
);
245 kmem_cache_free(inode_cachep
, (inode
));
249 * These are initializations that only need to be done
250 * once, because the fields are idempotent across use
251 * of the inode, so let the slab aware of that.
253 void inode_init_once(struct inode
*inode
)
255 memset(inode
, 0, sizeof(*inode
));
256 INIT_HLIST_NODE(&inode
->i_hash
);
257 INIT_LIST_HEAD(&inode
->i_dentry
);
258 INIT_LIST_HEAD(&inode
->i_devices
);
259 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
260 spin_lock_init(&inode
->i_data
.tree_lock
);
261 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
262 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
263 spin_lock_init(&inode
->i_data
.private_lock
);
264 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
265 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
266 i_size_ordered_init(inode
);
267 #ifdef CONFIG_FSNOTIFY
268 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
271 EXPORT_SYMBOL(inode_init_once
);
273 static void init_once(void *foo
)
275 struct inode
*inode
= (struct inode
*) foo
;
277 inode_init_once(inode
);
281 * inode_lock must be held
283 void __iget(struct inode
*inode
)
285 if (atomic_inc_return(&inode
->i_count
) != 1)
288 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
289 list_move(&inode
->i_list
, &inode_in_use
);
290 inodes_stat
.nr_unused
--;
293 void end_writeback(struct inode
*inode
)
296 BUG_ON(inode
->i_data
.nrpages
);
297 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
298 BUG_ON(!(inode
->i_state
& I_FREEING
));
299 BUG_ON(inode
->i_state
& I_CLEAR
);
300 inode_sync_wait(inode
);
301 inode
->i_state
= I_FREEING
| I_CLEAR
;
303 EXPORT_SYMBOL(end_writeback
);
305 static void evict(struct inode
*inode
)
307 const struct super_operations
*op
= inode
->i_sb
->s_op
;
309 if (op
->evict_inode
) {
310 op
->evict_inode(inode
);
312 if (inode
->i_data
.nrpages
)
313 truncate_inode_pages(&inode
->i_data
, 0);
314 end_writeback(inode
);
316 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
318 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
323 * dispose_list - dispose of the contents of a local list
324 * @head: the head of the list to free
326 * Dispose-list gets a local list with local inodes in it, so it doesn't
327 * need to worry about list corruption and SMP locks.
329 static void dispose_list(struct list_head
*head
)
333 while (!list_empty(head
)) {
336 inode
= list_first_entry(head
, struct inode
, i_list
);
337 list_del(&inode
->i_list
);
341 spin_lock(&inode_lock
);
342 hlist_del_init(&inode
->i_hash
);
343 list_del_init(&inode
->i_sb_list
);
344 spin_unlock(&inode_lock
);
346 wake_up_inode(inode
);
347 destroy_inode(inode
);
350 spin_lock(&inode_lock
);
351 inodes_stat
.nr_inodes
-= nr_disposed
;
352 spin_unlock(&inode_lock
);
356 * Invalidate all inodes for a device.
358 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
360 struct list_head
*next
;
361 int busy
= 0, count
= 0;
365 struct list_head
*tmp
= next
;
369 * We can reschedule here without worrying about the list's
370 * consistency because the per-sb list of inodes must not
371 * change during umount anymore, and because iprune_sem keeps
372 * shrink_icache_memory() away.
374 cond_resched_lock(&inode_lock
);
379 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
380 if (inode
->i_state
& I_NEW
)
382 invalidate_inode_buffers(inode
);
383 if (!atomic_read(&inode
->i_count
)) {
384 list_move(&inode
->i_list
, dispose
);
385 WARN_ON(inode
->i_state
& I_NEW
);
386 inode
->i_state
|= I_FREEING
;
392 /* only unused inodes may be cached with i_count zero */
393 inodes_stat
.nr_unused
-= count
;
398 * invalidate_inodes - discard the inodes on a device
401 * Discard all of the inodes for a given superblock. If the discard
402 * fails because there are busy inodes then a non zero value is returned.
403 * If the discard is successful all the inodes have been discarded.
405 int invalidate_inodes(struct super_block
*sb
)
408 LIST_HEAD(throw_away
);
410 down_write(&iprune_sem
);
411 spin_lock(&inode_lock
);
412 fsnotify_unmount_inodes(&sb
->s_inodes
);
413 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
414 spin_unlock(&inode_lock
);
416 dispose_list(&throw_away
);
417 up_write(&iprune_sem
);
421 EXPORT_SYMBOL(invalidate_inodes
);
423 static int can_unuse(struct inode
*inode
)
427 if (inode_has_buffers(inode
))
429 if (atomic_read(&inode
->i_count
))
431 if (inode
->i_data
.nrpages
)
437 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
438 * a temporary list and then are freed outside inode_lock by dispose_list().
440 * Any inodes which are pinned purely because of attached pagecache have their
441 * pagecache removed. We expect the final iput() on that inode to add it to
442 * the front of the inode_unused list. So look for it there and if the
443 * inode is still freeable, proceed. The right inode is found 99.9% of the
444 * time in testing on a 4-way.
446 * If the inode has metadata buffers attached to mapping->private_list then
447 * try to remove them.
449 static void prune_icache(int nr_to_scan
)
454 unsigned long reap
= 0;
456 down_read(&iprune_sem
);
457 spin_lock(&inode_lock
);
458 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
461 if (list_empty(&inode_unused
))
464 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
466 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
467 list_move(&inode
->i_list
, &inode_unused
);
470 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
472 spin_unlock(&inode_lock
);
473 if (remove_inode_buffers(inode
))
474 reap
+= invalidate_mapping_pages(&inode
->i_data
,
477 spin_lock(&inode_lock
);
479 if (inode
!= list_entry(inode_unused
.next
,
480 struct inode
, i_list
))
481 continue; /* wrong inode or list_empty */
482 if (!can_unuse(inode
))
485 list_move(&inode
->i_list
, &freeable
);
486 WARN_ON(inode
->i_state
& I_NEW
);
487 inode
->i_state
|= I_FREEING
;
490 inodes_stat
.nr_unused
-= nr_pruned
;
491 if (current_is_kswapd())
492 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
494 __count_vm_events(PGINODESTEAL
, reap
);
495 spin_unlock(&inode_lock
);
497 dispose_list(&freeable
);
498 up_read(&iprune_sem
);
502 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
503 * "unused" means that no dentries are referring to the inodes: the files are
504 * not open and the dcache references to those inodes have already been
507 * This function is passed the number of inodes to scan, and it returns the
508 * total number of remaining possibly-reclaimable inodes.
510 static int shrink_icache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
514 * Nasty deadlock avoidance. We may hold various FS locks,
515 * and we don't want to recurse into the FS that called us
516 * in clear_inode() and friends..
518 if (!(gfp_mask
& __GFP_FS
))
522 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
525 static struct shrinker icache_shrinker
= {
526 .shrink
= shrink_icache_memory
,
527 .seeks
= DEFAULT_SEEKS
,
530 static void __wait_on_freeing_inode(struct inode
*inode
);
532 * Called with the inode lock held.
533 * NOTE: we are not increasing the inode-refcount, you must call __iget()
534 * by hand after calling find_inode now! This simplifies iunique and won't
535 * add any additional branch in the common code.
537 static struct inode
*find_inode(struct super_block
*sb
,
538 struct hlist_head
*head
,
539 int (*test
)(struct inode
*, void *),
542 struct hlist_node
*node
;
543 struct inode
*inode
= NULL
;
546 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
547 if (inode
->i_sb
!= sb
)
549 if (!test(inode
, data
))
551 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
552 __wait_on_freeing_inode(inode
);
557 return node
? inode
: NULL
;
561 * find_inode_fast is the fast path version of find_inode, see the comment at
562 * iget_locked for details.
564 static struct inode
*find_inode_fast(struct super_block
*sb
,
565 struct hlist_head
*head
, unsigned long ino
)
567 struct hlist_node
*node
;
568 struct inode
*inode
= NULL
;
571 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
572 if (inode
->i_ino
!= ino
)
574 if (inode
->i_sb
!= sb
)
576 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
577 __wait_on_freeing_inode(inode
);
582 return node
? inode
: NULL
;
585 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
589 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
591 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
592 return tmp
& I_HASHMASK
;
596 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
599 inodes_stat
.nr_inodes
++;
600 list_add(&inode
->i_list
, &inode_in_use
);
601 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
603 hlist_add_head(&inode
->i_hash
, head
);
607 * inode_add_to_lists - add a new inode to relevant lists
608 * @sb: superblock inode belongs to
609 * @inode: inode to mark in use
611 * When an inode is allocated it needs to be accounted for, added to the in use
612 * list, the owning superblock and the inode hash. This needs to be done under
613 * the inode_lock, so export a function to do this rather than the inode lock
614 * itself. We calculate the hash list to add to here so it is all internal
615 * which requires the caller to have already set up the inode number in the
618 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
620 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
622 spin_lock(&inode_lock
);
623 __inode_add_to_lists(sb
, head
, inode
);
624 spin_unlock(&inode_lock
);
626 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
629 * new_inode - obtain an inode
632 * Allocates a new inode for given superblock. The default gfp_mask
633 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
634 * If HIGHMEM pages are unsuitable or it is known that pages allocated
635 * for the page cache are not reclaimable or migratable,
636 * mapping_set_gfp_mask() must be called with suitable flags on the
637 * newly created inode's mapping
640 struct inode
*new_inode(struct super_block
*sb
)
643 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
644 * error if st_ino won't fit in target struct field. Use 32bit counter
645 * here to attempt to avoid that.
647 static unsigned int last_ino
;
650 spin_lock_prefetch(&inode_lock
);
652 inode
= alloc_inode(sb
);
654 spin_lock(&inode_lock
);
655 __inode_add_to_lists(sb
, NULL
, inode
);
656 inode
->i_ino
= ++last_ino
;
658 spin_unlock(&inode_lock
);
662 EXPORT_SYMBOL(new_inode
);
664 void unlock_new_inode(struct inode
*inode
)
666 #ifdef CONFIG_DEBUG_LOCK_ALLOC
667 if (inode
->i_mode
& S_IFDIR
) {
668 struct file_system_type
*type
= inode
->i_sb
->s_type
;
670 /* Set new key only if filesystem hasn't already changed it */
671 if (!lockdep_match_class(&inode
->i_mutex
,
672 &type
->i_mutex_key
)) {
674 * ensure nobody is actually holding i_mutex
676 mutex_destroy(&inode
->i_mutex
);
677 mutex_init(&inode
->i_mutex
);
678 lockdep_set_class(&inode
->i_mutex
,
679 &type
->i_mutex_dir_key
);
684 * This is special! We do not need the spinlock when clearing I_NEW,
685 * because we're guaranteed that nobody else tries to do anything about
686 * the state of the inode when it is locked, as we just created it (so
687 * there can be no old holders that haven't tested I_NEW).
688 * However we must emit the memory barrier so that other CPUs reliably
689 * see the clearing of I_NEW after the other inode initialisation has
693 WARN_ON(!(inode
->i_state
& I_NEW
));
694 inode
->i_state
&= ~I_NEW
;
695 wake_up_inode(inode
);
697 EXPORT_SYMBOL(unlock_new_inode
);
700 * This is called without the inode lock held.. Be careful.
702 * We no longer cache the sb_flags in i_flags - see fs.h
703 * -- rmk@arm.uk.linux.org
705 static struct inode
*get_new_inode(struct super_block
*sb
,
706 struct hlist_head
*head
,
707 int (*test
)(struct inode
*, void *),
708 int (*set
)(struct inode
*, void *),
713 inode
= alloc_inode(sb
);
717 spin_lock(&inode_lock
);
718 /* We released the lock, so.. */
719 old
= find_inode(sb
, head
, test
, data
);
721 if (set(inode
, data
))
724 __inode_add_to_lists(sb
, head
, inode
);
725 inode
->i_state
= I_NEW
;
726 spin_unlock(&inode_lock
);
728 /* Return the locked inode with I_NEW set, the
729 * caller is responsible for filling in the contents
735 * Uhhuh, somebody else created the same inode under
736 * us. Use the old inode instead of the one we just
740 spin_unlock(&inode_lock
);
741 destroy_inode(inode
);
743 wait_on_inode(inode
);
748 spin_unlock(&inode_lock
);
749 destroy_inode(inode
);
754 * get_new_inode_fast is the fast path version of get_new_inode, see the
755 * comment at iget_locked for details.
757 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
758 struct hlist_head
*head
, unsigned long ino
)
762 inode
= alloc_inode(sb
);
766 spin_lock(&inode_lock
);
767 /* We released the lock, so.. */
768 old
= find_inode_fast(sb
, head
, ino
);
771 __inode_add_to_lists(sb
, head
, inode
);
772 inode
->i_state
= I_NEW
;
773 spin_unlock(&inode_lock
);
775 /* Return the locked inode with I_NEW set, the
776 * caller is responsible for filling in the contents
782 * Uhhuh, somebody else created the same inode under
783 * us. Use the old inode instead of the one we just
787 spin_unlock(&inode_lock
);
788 destroy_inode(inode
);
790 wait_on_inode(inode
);
796 * iunique - get a unique inode number
798 * @max_reserved: highest reserved inode number
800 * Obtain an inode number that is unique on the system for a given
801 * superblock. This is used by file systems that have no natural
802 * permanent inode numbering system. An inode number is returned that
803 * is higher than the reserved limit but unique.
806 * With a large number of inodes live on the file system this function
807 * currently becomes quite slow.
809 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
812 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
813 * error if st_ino won't fit in target struct field. Use 32bit counter
814 * here to attempt to avoid that.
816 static unsigned int counter
;
818 struct hlist_head
*head
;
821 spin_lock(&inode_lock
);
823 if (counter
<= max_reserved
)
824 counter
= max_reserved
+ 1;
826 head
= inode_hashtable
+ hash(sb
, res
);
827 inode
= find_inode_fast(sb
, head
, res
);
828 } while (inode
!= NULL
);
829 spin_unlock(&inode_lock
);
833 EXPORT_SYMBOL(iunique
);
835 struct inode
*igrab(struct inode
*inode
)
837 spin_lock(&inode_lock
);
838 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
)))
842 * Handle the case where s_op->clear_inode is not been
843 * called yet, and somebody is calling igrab
844 * while the inode is getting freed.
847 spin_unlock(&inode_lock
);
850 EXPORT_SYMBOL(igrab
);
853 * ifind - internal function, you want ilookup5() or iget5().
854 * @sb: super block of file system to search
855 * @head: the head of the list to search
856 * @test: callback used for comparisons between inodes
857 * @data: opaque data pointer to pass to @test
858 * @wait: if true wait for the inode to be unlocked, if false do not
860 * ifind() searches for the inode specified by @data in the inode
861 * cache. This is a generalized version of ifind_fast() for file systems where
862 * the inode number is not sufficient for unique identification of an inode.
864 * If the inode is in the cache, the inode is returned with an incremented
867 * Otherwise NULL is returned.
869 * Note, @test is called with the inode_lock held, so can't sleep.
871 static struct inode
*ifind(struct super_block
*sb
,
872 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
873 void *data
, const int wait
)
877 spin_lock(&inode_lock
);
878 inode
= find_inode(sb
, head
, test
, data
);
881 spin_unlock(&inode_lock
);
883 wait_on_inode(inode
);
886 spin_unlock(&inode_lock
);
891 * ifind_fast - internal function, you want ilookup() or iget().
892 * @sb: super block of file system to search
893 * @head: head of the list to search
894 * @ino: inode number to search for
896 * ifind_fast() searches for the inode @ino in the inode cache. This is for
897 * file systems where the inode number is sufficient for unique identification
900 * If the inode is in the cache, the inode is returned with an incremented
903 * Otherwise NULL is returned.
905 static struct inode
*ifind_fast(struct super_block
*sb
,
906 struct hlist_head
*head
, unsigned long ino
)
910 spin_lock(&inode_lock
);
911 inode
= find_inode_fast(sb
, head
, ino
);
914 spin_unlock(&inode_lock
);
915 wait_on_inode(inode
);
918 spin_unlock(&inode_lock
);
923 * ilookup5_nowait - search for an inode in the inode cache
924 * @sb: super block of file system to search
925 * @hashval: hash value (usually inode number) to search for
926 * @test: callback used for comparisons between inodes
927 * @data: opaque data pointer to pass to @test
929 * ilookup5() uses ifind() to search for the inode specified by @hashval and
930 * @data in the inode cache. This is a generalized version of ilookup() for
931 * file systems where the inode number is not sufficient for unique
932 * identification of an inode.
934 * If the inode is in the cache, the inode is returned with an incremented
935 * reference count. Note, the inode lock is not waited upon so you have to be
936 * very careful what you do with the returned inode. You probably should be
937 * using ilookup5() instead.
939 * Otherwise NULL is returned.
941 * Note, @test is called with the inode_lock held, so can't sleep.
943 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
944 int (*test
)(struct inode
*, void *), void *data
)
946 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
948 return ifind(sb
, head
, test
, data
, 0);
950 EXPORT_SYMBOL(ilookup5_nowait
);
953 * ilookup5 - search for an inode in the inode cache
954 * @sb: super block of file system to search
955 * @hashval: hash value (usually inode number) to search for
956 * @test: callback used for comparisons between inodes
957 * @data: opaque data pointer to pass to @test
959 * ilookup5() uses ifind() to search for the inode specified by @hashval and
960 * @data in the inode cache. This is a generalized version of ilookup() for
961 * file systems where the inode number is not sufficient for unique
962 * identification of an inode.
964 * If the inode is in the cache, the inode lock is waited upon and the inode is
965 * returned with an incremented reference count.
967 * Otherwise NULL is returned.
969 * Note, @test is called with the inode_lock held, so can't sleep.
971 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
972 int (*test
)(struct inode
*, void *), void *data
)
974 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
976 return ifind(sb
, head
, test
, data
, 1);
978 EXPORT_SYMBOL(ilookup5
);
981 * ilookup - search for an inode in the inode cache
982 * @sb: super block of file system to search
983 * @ino: inode number to search for
985 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
986 * This is for file systems where the inode number is sufficient for unique
987 * identification of an inode.
989 * If the inode is in the cache, the inode is returned with an incremented
992 * Otherwise NULL is returned.
994 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
996 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
998 return ifind_fast(sb
, head
, ino
);
1000 EXPORT_SYMBOL(ilookup
);
1003 * iget5_locked - obtain an inode from a mounted file system
1004 * @sb: super block of file system
1005 * @hashval: hash value (usually inode number) to get
1006 * @test: callback used for comparisons between inodes
1007 * @set: callback used to initialize a new struct inode
1008 * @data: opaque data pointer to pass to @test and @set
1010 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1011 * and @data in the inode cache and if present it is returned with an increased
1012 * reference count. This is a generalized version of iget_locked() for file
1013 * systems where the inode number is not sufficient for unique identification
1016 * If the inode is not in cache, get_new_inode() is called to allocate a new
1017 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1018 * file system gets to fill it in before unlocking it via unlock_new_inode().
1020 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1022 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1023 int (*test
)(struct inode
*, void *),
1024 int (*set
)(struct inode
*, void *), void *data
)
1026 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1027 struct inode
*inode
;
1029 inode
= ifind(sb
, head
, test
, data
, 1);
1033 * get_new_inode() will do the right thing, re-trying the search
1034 * in case it had to block at any point.
1036 return get_new_inode(sb
, head
, test
, set
, data
);
1038 EXPORT_SYMBOL(iget5_locked
);
1041 * iget_locked - obtain an inode from a mounted file system
1042 * @sb: super block of file system
1043 * @ino: inode number to get
1045 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1046 * the inode cache and if present it is returned with an increased reference
1047 * count. This is for file systems where the inode number is sufficient for
1048 * unique identification of an inode.
1050 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1051 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1052 * The file system gets to fill it in before unlocking it via
1053 * unlock_new_inode().
1055 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1057 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1058 struct inode
*inode
;
1060 inode
= ifind_fast(sb
, head
, ino
);
1064 * get_new_inode_fast() will do the right thing, re-trying the search
1065 * in case it had to block at any point.
1067 return get_new_inode_fast(sb
, head
, ino
);
1069 EXPORT_SYMBOL(iget_locked
);
1071 int insert_inode_locked(struct inode
*inode
)
1073 struct super_block
*sb
= inode
->i_sb
;
1074 ino_t ino
= inode
->i_ino
;
1075 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1077 inode
->i_state
|= I_NEW
;
1079 struct hlist_node
*node
;
1080 struct inode
*old
= NULL
;
1081 spin_lock(&inode_lock
);
1082 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1083 if (old
->i_ino
!= ino
)
1085 if (old
->i_sb
!= sb
)
1087 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
))
1091 if (likely(!node
)) {
1092 hlist_add_head(&inode
->i_hash
, head
);
1093 spin_unlock(&inode_lock
);
1097 spin_unlock(&inode_lock
);
1099 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1106 EXPORT_SYMBOL(insert_inode_locked
);
1108 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1109 int (*test
)(struct inode
*, void *), void *data
)
1111 struct super_block
*sb
= inode
->i_sb
;
1112 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1114 inode
->i_state
|= I_NEW
;
1117 struct hlist_node
*node
;
1118 struct inode
*old
= NULL
;
1120 spin_lock(&inode_lock
);
1121 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1122 if (old
->i_sb
!= sb
)
1124 if (!test(old
, data
))
1126 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
))
1130 if (likely(!node
)) {
1131 hlist_add_head(&inode
->i_hash
, head
);
1132 spin_unlock(&inode_lock
);
1136 spin_unlock(&inode_lock
);
1138 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1145 EXPORT_SYMBOL(insert_inode_locked4
);
1148 * __insert_inode_hash - hash an inode
1149 * @inode: unhashed inode
1150 * @hashval: unsigned long value used to locate this object in the
1153 * Add an inode to the inode hash for this superblock.
1155 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1157 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1158 spin_lock(&inode_lock
);
1159 hlist_add_head(&inode
->i_hash
, head
);
1160 spin_unlock(&inode_lock
);
1162 EXPORT_SYMBOL(__insert_inode_hash
);
1165 * remove_inode_hash - remove an inode from the hash
1166 * @inode: inode to unhash
1168 * Remove an inode from the superblock.
1170 void remove_inode_hash(struct inode
*inode
)
1172 spin_lock(&inode_lock
);
1173 hlist_del_init(&inode
->i_hash
);
1174 spin_unlock(&inode_lock
);
1176 EXPORT_SYMBOL(remove_inode_hash
);
1178 int generic_delete_inode(struct inode
*inode
)
1182 EXPORT_SYMBOL(generic_delete_inode
);
1185 * Normal UNIX filesystem behaviour: delete the
1186 * inode when the usage count drops to zero, and
1189 int generic_drop_inode(struct inode
*inode
)
1191 return !inode
->i_nlink
|| hlist_unhashed(&inode
->i_hash
);
1193 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1196 * Called when we're dropping the last reference
1199 * Call the FS "drop_inode()" function, defaulting to
1200 * the legacy UNIX filesystem behaviour. If it tells
1201 * us to evict inode, do so. Otherwise, retain inode
1202 * in cache if fs is alive, sync and evict if fs is
1205 static void iput_final(struct inode
*inode
)
1207 struct super_block
*sb
= inode
->i_sb
;
1208 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1211 if (op
&& op
->drop_inode
)
1212 drop
= op
->drop_inode(inode
);
1214 drop
= generic_drop_inode(inode
);
1217 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1218 list_move(&inode
->i_list
, &inode_unused
);
1219 inodes_stat
.nr_unused
++;
1220 if (sb
->s_flags
& MS_ACTIVE
) {
1221 spin_unlock(&inode_lock
);
1224 WARN_ON(inode
->i_state
& I_NEW
);
1225 inode
->i_state
|= I_WILL_FREE
;
1226 spin_unlock(&inode_lock
);
1227 write_inode_now(inode
, 1);
1228 spin_lock(&inode_lock
);
1229 WARN_ON(inode
->i_state
& I_NEW
);
1230 inode
->i_state
&= ~I_WILL_FREE
;
1231 inodes_stat
.nr_unused
--;
1232 hlist_del_init(&inode
->i_hash
);
1234 list_del_init(&inode
->i_list
);
1235 list_del_init(&inode
->i_sb_list
);
1236 WARN_ON(inode
->i_state
& I_NEW
);
1237 inode
->i_state
|= I_FREEING
;
1238 inodes_stat
.nr_inodes
--;
1239 spin_unlock(&inode_lock
);
1241 spin_lock(&inode_lock
);
1242 hlist_del_init(&inode
->i_hash
);
1243 spin_unlock(&inode_lock
);
1244 wake_up_inode(inode
);
1245 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
1246 destroy_inode(inode
);
1250 * iput - put an inode
1251 * @inode: inode to put
1253 * Puts an inode, dropping its usage count. If the inode use count hits
1254 * zero, the inode is then freed and may also be destroyed.
1256 * Consequently, iput() can sleep.
1258 void iput(struct inode
*inode
)
1261 BUG_ON(inode
->i_state
& I_CLEAR
);
1263 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1267 EXPORT_SYMBOL(iput
);
1270 * bmap - find a block number in a file
1271 * @inode: inode of file
1272 * @block: block to find
1274 * Returns the block number on the device holding the inode that
1275 * is the disk block number for the block of the file requested.
1276 * That is, asked for block 4 of inode 1 the function will return the
1277 * disk block relative to the disk start that holds that block of the
1280 sector_t
bmap(struct inode
*inode
, sector_t block
)
1283 if (inode
->i_mapping
->a_ops
->bmap
)
1284 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1287 EXPORT_SYMBOL(bmap
);
1290 * With relative atime, only update atime if the previous atime is
1291 * earlier than either the ctime or mtime or if at least a day has
1292 * passed since the last atime update.
1294 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1295 struct timespec now
)
1298 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1301 * Is mtime younger than atime? If yes, update atime:
1303 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1306 * Is ctime younger than atime? If yes, update atime:
1308 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1312 * Is the previous atime value older than a day? If yes,
1315 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1318 * Good, we can skip the atime update:
1324 * touch_atime - update the access time
1325 * @mnt: mount the inode is accessed on
1326 * @dentry: dentry accessed
1328 * Update the accessed time on an inode and mark it for writeback.
1329 * This function automatically handles read only file systems and media,
1330 * as well as the "noatime" flag and inode specific "noatime" markers.
1332 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1334 struct inode
*inode
= dentry
->d_inode
;
1335 struct timespec now
;
1337 if (inode
->i_flags
& S_NOATIME
)
1339 if (IS_NOATIME(inode
))
1341 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1344 if (mnt
->mnt_flags
& MNT_NOATIME
)
1346 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1349 now
= current_fs_time(inode
->i_sb
);
1351 if (!relatime_need_update(mnt
, inode
, now
))
1354 if (timespec_equal(&inode
->i_atime
, &now
))
1357 if (mnt_want_write(mnt
))
1360 inode
->i_atime
= now
;
1361 mark_inode_dirty_sync(inode
);
1362 mnt_drop_write(mnt
);
1364 EXPORT_SYMBOL(touch_atime
);
1367 * file_update_time - update mtime and ctime time
1368 * @file: file accessed
1370 * Update the mtime and ctime members of an inode and mark the inode
1371 * for writeback. Note that this function is meant exclusively for
1372 * usage in the file write path of filesystems, and filesystems may
1373 * choose to explicitly ignore update via this function with the
1374 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1375 * timestamps are handled by the server.
1378 void file_update_time(struct file
*file
)
1380 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1381 struct timespec now
;
1382 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1384 /* First try to exhaust all avenues to not sync */
1385 if (IS_NOCMTIME(inode
))
1388 now
= current_fs_time(inode
->i_sb
);
1389 if (!timespec_equal(&inode
->i_mtime
, &now
))
1392 if (!timespec_equal(&inode
->i_ctime
, &now
))
1395 if (IS_I_VERSION(inode
))
1396 sync_it
|= S_VERSION
;
1401 /* Finally allowed to write? Takes lock. */
1402 if (mnt_want_write_file(file
))
1405 /* Only change inode inside the lock region */
1406 if (sync_it
& S_VERSION
)
1407 inode_inc_iversion(inode
);
1408 if (sync_it
& S_CTIME
)
1409 inode
->i_ctime
= now
;
1410 if (sync_it
& S_MTIME
)
1411 inode
->i_mtime
= now
;
1412 mark_inode_dirty_sync(inode
);
1413 mnt_drop_write(file
->f_path
.mnt
);
1415 EXPORT_SYMBOL(file_update_time
);
1417 int inode_needs_sync(struct inode
*inode
)
1421 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1425 EXPORT_SYMBOL(inode_needs_sync
);
1427 int inode_wait(void *word
)
1432 EXPORT_SYMBOL(inode_wait
);
1435 * If we try to find an inode in the inode hash while it is being
1436 * deleted, we have to wait until the filesystem completes its
1437 * deletion before reporting that it isn't found. This function waits
1438 * until the deletion _might_ have completed. Callers are responsible
1439 * to recheck inode state.
1441 * It doesn't matter if I_NEW is not set initially, a call to
1442 * wake_up_inode() after removing from the hash list will DTRT.
1444 * This is called with inode_lock held.
1446 static void __wait_on_freeing_inode(struct inode
*inode
)
1448 wait_queue_head_t
*wq
;
1449 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1450 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1451 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1452 spin_unlock(&inode_lock
);
1454 finish_wait(wq
, &wait
.wait
);
1455 spin_lock(&inode_lock
);
1458 static __initdata
unsigned long ihash_entries
;
1459 static int __init
set_ihash_entries(char *str
)
1463 ihash_entries
= simple_strtoul(str
, &str
, 0);
1466 __setup("ihash_entries=", set_ihash_entries
);
1469 * Initialize the waitqueues and inode hash table.
1471 void __init
inode_init_early(void)
1475 /* If hashes are distributed across NUMA nodes, defer
1476 * hash allocation until vmalloc space is available.
1482 alloc_large_system_hash("Inode-cache",
1483 sizeof(struct hlist_head
),
1491 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1492 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1495 void __init
inode_init(void)
1499 /* inode slab cache */
1500 inode_cachep
= kmem_cache_create("inode_cache",
1501 sizeof(struct inode
),
1503 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1506 register_shrinker(&icache_shrinker
);
1508 /* Hash may have been set up in inode_init_early */
1513 alloc_large_system_hash("Inode-cache",
1514 sizeof(struct hlist_head
),
1522 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1523 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1526 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1528 inode
->i_mode
= mode
;
1529 if (S_ISCHR(mode
)) {
1530 inode
->i_fop
= &def_chr_fops
;
1531 inode
->i_rdev
= rdev
;
1532 } else if (S_ISBLK(mode
)) {
1533 inode
->i_fop
= &def_blk_fops
;
1534 inode
->i_rdev
= rdev
;
1535 } else if (S_ISFIFO(mode
))
1536 inode
->i_fop
= &def_fifo_fops
;
1537 else if (S_ISSOCK(mode
))
1538 inode
->i_fop
= &bad_sock_fops
;
1540 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1541 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1544 EXPORT_SYMBOL(init_special_inode
);
1547 * Init uid,gid,mode for new inode according to posix standards
1549 * @dir: Directory inode
1550 * @mode: mode of the new inode
1552 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1555 inode
->i_uid
= current_fsuid();
1556 if (dir
&& dir
->i_mode
& S_ISGID
) {
1557 inode
->i_gid
= dir
->i_gid
;
1561 inode
->i_gid
= current_fsgid();
1562 inode
->i_mode
= mode
;
1564 EXPORT_SYMBOL(inode_init_owner
);