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>
29 * This is needed for the following functions:
31 * - invalidate_inode_buffers
34 * FIXME: remove all knowledge of the buffer layer from this file
36 #include <linux/buffer_head.h>
39 * New inode.c implementation.
41 * This implementation has the basic premise of trying
42 * to be extremely low-overhead and SMP-safe, yet be
43 * simple enough to be "obviously correct".
48 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
50 /* #define INODE_PARANOIA 1 */
51 /* #define INODE_DEBUG 1 */
54 * Inode lookup is no longer as critical as it used to be:
55 * most of the lookups are going to be through the dcache.
57 #define I_HASHBITS i_hash_shift
58 #define I_HASHMASK i_hash_mask
60 static unsigned int i_hash_mask __read_mostly
;
61 static unsigned int i_hash_shift __read_mostly
;
64 * Each inode can be on two separate lists. One is
65 * the hash list of the inode, used for lookups. The
66 * other linked list is the "type" list:
67 * "in_use" - valid inode, i_count > 0, i_nlink > 0
68 * "dirty" - as "in_use" but also dirty
69 * "unused" - valid inode, i_count = 0
71 * A "dirty" list is maintained for each super block,
72 * allowing for low-overhead inode sync() operations.
75 LIST_HEAD(inode_in_use
);
76 LIST_HEAD(inode_unused
);
77 static struct hlist_head
*inode_hashtable __read_mostly
;
80 * A simple spinlock to protect the list manipulations.
82 * NOTE! You also have to own the lock if you change
83 * the i_state of an inode while it is in use..
85 DEFINE_SPINLOCK(inode_lock
);
88 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
89 * icache shrinking path, and the umount path. Without this exclusion,
90 * by the time prune_icache calls iput for the inode whose pages it has
91 * been invalidating, or by the time it calls clear_inode & destroy_inode
92 * from its final dispose_list, the struct super_block they refer to
93 * (for inode->i_sb->s_op) may already have been freed and reused.
95 * We make this an rwsem because the fastpath is icache shrinking. In
96 * some cases a filesystem may be doing a significant amount of work in
97 * its inode reclaim code, so this should improve parallelism.
99 static DECLARE_RWSEM(iprune_sem
);
102 * Statistics gathering..
104 struct inodes_stat_t inodes_stat
;
106 static struct kmem_cache
*inode_cachep __read_mostly
;
108 static void wake_up_inode(struct inode
*inode
)
111 * Prevent speculative execution through spin_unlock(&inode_lock);
114 wake_up_bit(&inode
->i_state
, __I_NEW
);
118 * inode_init_always - perform inode structure intialisation
119 * @sb: superblock inode belongs to
120 * @inode: inode to initialise
122 * These are initializations that need to be done on every inode
123 * allocation as the fields are not initialised by slab allocation.
125 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
127 static const struct address_space_operations empty_aops
;
128 static const struct inode_operations empty_iops
;
129 static const struct file_operations empty_fops
;
130 struct address_space
*const mapping
= &inode
->i_data
;
133 inode
->i_blkbits
= sb
->s_blocksize_bits
;
135 atomic_set(&inode
->i_count
, 1);
136 inode
->i_op
= &empty_iops
;
137 inode
->i_fop
= &empty_fops
;
141 atomic_set(&inode
->i_writecount
, 0);
145 inode
->i_generation
= 0;
147 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
149 inode
->i_pipe
= NULL
;
150 inode
->i_bdev
= NULL
;
151 inode
->i_cdev
= NULL
;
153 inode
->dirtied_when
= 0;
155 if (security_inode_alloc(inode
))
157 spin_lock_init(&inode
->i_lock
);
158 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
160 mutex_init(&inode
->i_mutex
);
161 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
163 init_rwsem(&inode
->i_alloc_sem
);
164 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
166 mapping
->a_ops
= &empty_aops
;
167 mapping
->host
= inode
;
169 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
170 mapping
->assoc_mapping
= NULL
;
171 mapping
->backing_dev_info
= &default_backing_dev_info
;
172 mapping
->writeback_index
= 0;
175 * If the block_device provides a backing_dev_info for client
176 * inodes then use that. Otherwise the inode share the bdev's
180 struct backing_dev_info
*bdi
;
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
;
187 #ifdef CONFIG_FS_POSIX_ACL
188 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
191 #ifdef CONFIG_FSNOTIFY
192 inode
->i_fsnotify_mask
= 0;
199 EXPORT_SYMBOL(inode_init_always
);
201 static struct inode
*alloc_inode(struct super_block
*sb
)
205 if (sb
->s_op
->alloc_inode
)
206 inode
= sb
->s_op
->alloc_inode(sb
);
208 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
213 if (unlikely(inode_init_always(sb
, inode
))) {
214 if (inode
->i_sb
->s_op
->destroy_inode
)
215 inode
->i_sb
->s_op
->destroy_inode(inode
);
217 kmem_cache_free(inode_cachep
, inode
);
224 void __destroy_inode(struct inode
*inode
)
226 BUG_ON(inode_has_buffers(inode
));
227 security_inode_free(inode
);
228 fsnotify_inode_delete(inode
);
229 #ifdef CONFIG_FS_POSIX_ACL
230 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
231 posix_acl_release(inode
->i_acl
);
232 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
233 posix_acl_release(inode
->i_default_acl
);
236 EXPORT_SYMBOL(__destroy_inode
);
238 static void destroy_inode(struct inode
*inode
)
240 __destroy_inode(inode
);
241 if (inode
->i_sb
->s_op
->destroy_inode
)
242 inode
->i_sb
->s_op
->destroy_inode(inode
);
244 kmem_cache_free(inode_cachep
, (inode
));
248 * These are initializations that only need to be done
249 * once, because the fields are idempotent across use
250 * of the inode, so let the slab aware of that.
252 void inode_init_once(struct inode
*inode
)
254 memset(inode
, 0, sizeof(*inode
));
255 INIT_HLIST_NODE(&inode
->i_hash
);
256 INIT_LIST_HEAD(&inode
->i_dentry
);
257 INIT_LIST_HEAD(&inode
->i_devices
);
258 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
259 spin_lock_init(&inode
->i_data
.tree_lock
);
260 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
261 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
262 spin_lock_init(&inode
->i_data
.private_lock
);
263 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
264 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
265 i_size_ordered_init(inode
);
266 #ifdef CONFIG_FSNOTIFY
267 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
270 EXPORT_SYMBOL(inode_init_once
);
272 static void init_once(void *foo
)
274 struct inode
*inode
= (struct inode
*) foo
;
276 inode_init_once(inode
);
280 * inode_lock must be held
282 void __iget(struct inode
*inode
)
284 if (atomic_inc_return(&inode
->i_count
) != 1)
287 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
288 list_move(&inode
->i_list
, &inode_in_use
);
289 inodes_stat
.nr_unused
--;
292 void end_writeback(struct inode
*inode
)
295 BUG_ON(inode
->i_data
.nrpages
);
296 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
297 BUG_ON(!(inode
->i_state
& I_FREEING
));
298 BUG_ON(inode
->i_state
& I_CLEAR
);
299 inode_sync_wait(inode
);
300 inode
->i_state
= I_FREEING
| I_CLEAR
;
302 EXPORT_SYMBOL(end_writeback
);
304 static void evict(struct inode
*inode
)
306 const struct super_operations
*op
= inode
->i_sb
->s_op
;
308 if (op
->evict_inode
) {
309 op
->evict_inode(inode
);
311 if (inode
->i_data
.nrpages
)
312 truncate_inode_pages(&inode
->i_data
, 0);
313 end_writeback(inode
);
315 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
317 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
322 * dispose_list - dispose of the contents of a local list
323 * @head: the head of the list to free
325 * Dispose-list gets a local list with local inodes in it, so it doesn't
326 * need to worry about list corruption and SMP locks.
328 static void dispose_list(struct list_head
*head
)
332 while (!list_empty(head
)) {
335 inode
= list_first_entry(head
, struct inode
, i_list
);
336 list_del(&inode
->i_list
);
340 spin_lock(&inode_lock
);
341 hlist_del_init(&inode
->i_hash
);
342 list_del_init(&inode
->i_sb_list
);
343 spin_unlock(&inode_lock
);
345 wake_up_inode(inode
);
346 destroy_inode(inode
);
349 spin_lock(&inode_lock
);
350 inodes_stat
.nr_inodes
-= nr_disposed
;
351 spin_unlock(&inode_lock
);
355 * Invalidate all inodes for a device.
357 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
359 struct list_head
*next
;
360 int busy
= 0, count
= 0;
364 struct list_head
*tmp
= next
;
368 * We can reschedule here without worrying about the list's
369 * consistency because the per-sb list of inodes must not
370 * change during umount anymore, and because iprune_sem keeps
371 * shrink_icache_memory() away.
373 cond_resched_lock(&inode_lock
);
378 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
379 if (inode
->i_state
& I_NEW
)
381 invalidate_inode_buffers(inode
);
382 if (!atomic_read(&inode
->i_count
)) {
383 list_move(&inode
->i_list
, dispose
);
384 WARN_ON(inode
->i_state
& I_NEW
);
385 inode
->i_state
|= I_FREEING
;
391 /* only unused inodes may be cached with i_count zero */
392 inodes_stat
.nr_unused
-= count
;
397 * invalidate_inodes - discard the inodes on a device
400 * Discard all of the inodes for a given superblock. If the discard
401 * fails because there are busy inodes then a non zero value is returned.
402 * If the discard is successful all the inodes have been discarded.
404 int invalidate_inodes(struct super_block
*sb
)
407 LIST_HEAD(throw_away
);
409 down_write(&iprune_sem
);
410 spin_lock(&inode_lock
);
411 fsnotify_unmount_inodes(&sb
->s_inodes
);
412 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
413 spin_unlock(&inode_lock
);
415 dispose_list(&throw_away
);
416 up_write(&iprune_sem
);
421 static int can_unuse(struct inode
*inode
)
425 if (inode_has_buffers(inode
))
427 if (atomic_read(&inode
->i_count
))
429 if (inode
->i_data
.nrpages
)
435 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
436 * a temporary list and then are freed outside inode_lock by dispose_list().
438 * Any inodes which are pinned purely because of attached pagecache have their
439 * pagecache removed. We expect the final iput() on that inode to add it to
440 * the front of the inode_unused list. So look for it there and if the
441 * inode is still freeable, proceed. The right inode is found 99.9% of the
442 * time in testing on a 4-way.
444 * If the inode has metadata buffers attached to mapping->private_list then
445 * try to remove them.
447 static void prune_icache(int nr_to_scan
)
452 unsigned long reap
= 0;
454 down_read(&iprune_sem
);
455 spin_lock(&inode_lock
);
456 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
459 if (list_empty(&inode_unused
))
462 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
464 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
465 list_move(&inode
->i_list
, &inode_unused
);
468 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
470 spin_unlock(&inode_lock
);
471 if (remove_inode_buffers(inode
))
472 reap
+= invalidate_mapping_pages(&inode
->i_data
,
475 spin_lock(&inode_lock
);
477 if (inode
!= list_entry(inode_unused
.next
,
478 struct inode
, i_list
))
479 continue; /* wrong inode or list_empty */
480 if (!can_unuse(inode
))
483 list_move(&inode
->i_list
, &freeable
);
484 WARN_ON(inode
->i_state
& I_NEW
);
485 inode
->i_state
|= I_FREEING
;
488 inodes_stat
.nr_unused
-= nr_pruned
;
489 if (current_is_kswapd())
490 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
492 __count_vm_events(PGINODESTEAL
, reap
);
493 spin_unlock(&inode_lock
);
495 dispose_list(&freeable
);
496 up_read(&iprune_sem
);
500 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
501 * "unused" means that no dentries are referring to the inodes: the files are
502 * not open and the dcache references to those inodes have already been
505 * This function is passed the number of inodes to scan, and it returns the
506 * total number of remaining possibly-reclaimable inodes.
508 static int shrink_icache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
512 * Nasty deadlock avoidance. We may hold various FS locks,
513 * and we don't want to recurse into the FS that called us
514 * in clear_inode() and friends..
516 if (!(gfp_mask
& __GFP_FS
))
520 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
523 static struct shrinker icache_shrinker
= {
524 .shrink
= shrink_icache_memory
,
525 .seeks
= DEFAULT_SEEKS
,
528 static void __wait_on_freeing_inode(struct inode
*inode
);
530 * Called with the inode lock held.
531 * NOTE: we are not increasing the inode-refcount, you must call __iget()
532 * by hand after calling find_inode now! This simplifies iunique and won't
533 * add any additional branch in the common code.
535 static struct inode
*find_inode(struct super_block
*sb
,
536 struct hlist_head
*head
,
537 int (*test
)(struct inode
*, void *),
540 struct hlist_node
*node
;
541 struct inode
*inode
= NULL
;
544 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
545 if (inode
->i_sb
!= sb
)
547 if (!test(inode
, data
))
549 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
550 __wait_on_freeing_inode(inode
);
555 return node
? inode
: NULL
;
559 * find_inode_fast is the fast path version of find_inode, see the comment at
560 * iget_locked for details.
562 static struct inode
*find_inode_fast(struct super_block
*sb
,
563 struct hlist_head
*head
, unsigned long ino
)
565 struct hlist_node
*node
;
566 struct inode
*inode
= NULL
;
569 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
570 if (inode
->i_ino
!= ino
)
572 if (inode
->i_sb
!= sb
)
574 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
575 __wait_on_freeing_inode(inode
);
580 return node
? inode
: NULL
;
583 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
587 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
589 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
590 return tmp
& I_HASHMASK
;
594 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
597 inodes_stat
.nr_inodes
++;
598 list_add(&inode
->i_list
, &inode_in_use
);
599 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
601 hlist_add_head(&inode
->i_hash
, head
);
605 * inode_add_to_lists - add a new inode to relevant lists
606 * @sb: superblock inode belongs to
607 * @inode: inode to mark in use
609 * When an inode is allocated it needs to be accounted for, added to the in use
610 * list, the owning superblock and the inode hash. This needs to be done under
611 * the inode_lock, so export a function to do this rather than the inode lock
612 * itself. We calculate the hash list to add to here so it is all internal
613 * which requires the caller to have already set up the inode number in the
616 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
618 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
620 spin_lock(&inode_lock
);
621 __inode_add_to_lists(sb
, head
, inode
);
622 spin_unlock(&inode_lock
);
624 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
627 * new_inode - obtain an inode
630 * Allocates a new inode for given superblock. The default gfp_mask
631 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
632 * If HIGHMEM pages are unsuitable or it is known that pages allocated
633 * for the page cache are not reclaimable or migratable,
634 * mapping_set_gfp_mask() must be called with suitable flags on the
635 * newly created inode's mapping
638 struct inode
*new_inode(struct super_block
*sb
)
641 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
642 * error if st_ino won't fit in target struct field. Use 32bit counter
643 * here to attempt to avoid that.
645 static unsigned int last_ino
;
648 spin_lock_prefetch(&inode_lock
);
650 inode
= alloc_inode(sb
);
652 spin_lock(&inode_lock
);
653 __inode_add_to_lists(sb
, NULL
, inode
);
654 inode
->i_ino
= ++last_ino
;
656 spin_unlock(&inode_lock
);
660 EXPORT_SYMBOL(new_inode
);
662 void unlock_new_inode(struct inode
*inode
)
664 #ifdef CONFIG_DEBUG_LOCK_ALLOC
665 if (S_ISDIR(inode
->i_mode
)) {
666 struct file_system_type
*type
= inode
->i_sb
->s_type
;
668 /* Set new key only if filesystem hasn't already changed it */
669 if (!lockdep_match_class(&inode
->i_mutex
,
670 &type
->i_mutex_key
)) {
672 * ensure nobody is actually holding i_mutex
674 mutex_destroy(&inode
->i_mutex
);
675 mutex_init(&inode
->i_mutex
);
676 lockdep_set_class(&inode
->i_mutex
,
677 &type
->i_mutex_dir_key
);
682 * This is special! We do not need the spinlock when clearing I_NEW,
683 * because we're guaranteed that nobody else tries to do anything about
684 * the state of the inode when it is locked, as we just created it (so
685 * there can be no old holders that haven't tested I_NEW).
686 * However we must emit the memory barrier so that other CPUs reliably
687 * see the clearing of I_NEW after the other inode initialisation has
691 WARN_ON(!(inode
->i_state
& I_NEW
));
692 inode
->i_state
&= ~I_NEW
;
693 wake_up_inode(inode
);
695 EXPORT_SYMBOL(unlock_new_inode
);
698 * This is called without the inode lock held.. Be careful.
700 * We no longer cache the sb_flags in i_flags - see fs.h
701 * -- rmk@arm.uk.linux.org
703 static struct inode
*get_new_inode(struct super_block
*sb
,
704 struct hlist_head
*head
,
705 int (*test
)(struct inode
*, void *),
706 int (*set
)(struct inode
*, void *),
711 inode
= alloc_inode(sb
);
715 spin_lock(&inode_lock
);
716 /* We released the lock, so.. */
717 old
= find_inode(sb
, head
, test
, data
);
719 if (set(inode
, data
))
722 __inode_add_to_lists(sb
, head
, inode
);
723 inode
->i_state
= I_NEW
;
724 spin_unlock(&inode_lock
);
726 /* Return the locked inode with I_NEW set, the
727 * caller is responsible for filling in the contents
733 * Uhhuh, somebody else created the same inode under
734 * us. Use the old inode instead of the one we just
738 spin_unlock(&inode_lock
);
739 destroy_inode(inode
);
741 wait_on_inode(inode
);
746 spin_unlock(&inode_lock
);
747 destroy_inode(inode
);
752 * get_new_inode_fast is the fast path version of get_new_inode, see the
753 * comment at iget_locked for details.
755 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
756 struct hlist_head
*head
, unsigned long ino
)
760 inode
= alloc_inode(sb
);
764 spin_lock(&inode_lock
);
765 /* We released the lock, so.. */
766 old
= find_inode_fast(sb
, head
, ino
);
769 __inode_add_to_lists(sb
, head
, inode
);
770 inode
->i_state
= I_NEW
;
771 spin_unlock(&inode_lock
);
773 /* Return the locked inode with I_NEW set, the
774 * caller is responsible for filling in the contents
780 * Uhhuh, somebody else created the same inode under
781 * us. Use the old inode instead of the one we just
785 spin_unlock(&inode_lock
);
786 destroy_inode(inode
);
788 wait_on_inode(inode
);
794 * iunique - get a unique inode number
796 * @max_reserved: highest reserved inode number
798 * Obtain an inode number that is unique on the system for a given
799 * superblock. This is used by file systems that have no natural
800 * permanent inode numbering system. An inode number is returned that
801 * is higher than the reserved limit but unique.
804 * With a large number of inodes live on the file system this function
805 * currently becomes quite slow.
807 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
810 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
811 * error if st_ino won't fit in target struct field. Use 32bit counter
812 * here to attempt to avoid that.
814 static unsigned int counter
;
816 struct hlist_head
*head
;
819 spin_lock(&inode_lock
);
821 if (counter
<= max_reserved
)
822 counter
= max_reserved
+ 1;
824 head
= inode_hashtable
+ hash(sb
, res
);
825 inode
= find_inode_fast(sb
, head
, res
);
826 } while (inode
!= NULL
);
827 spin_unlock(&inode_lock
);
831 EXPORT_SYMBOL(iunique
);
833 struct inode
*igrab(struct inode
*inode
)
835 spin_lock(&inode_lock
);
836 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
)))
840 * Handle the case where s_op->clear_inode is not been
841 * called yet, and somebody is calling igrab
842 * while the inode is getting freed.
845 spin_unlock(&inode_lock
);
848 EXPORT_SYMBOL(igrab
);
851 * ifind - internal function, you want ilookup5() or iget5().
852 * @sb: super block of file system to search
853 * @head: the head of the list to search
854 * @test: callback used for comparisons between inodes
855 * @data: opaque data pointer to pass to @test
856 * @wait: if true wait for the inode to be unlocked, if false do not
858 * ifind() searches for the inode specified by @data in the inode
859 * cache. This is a generalized version of ifind_fast() for file systems where
860 * the inode number is not sufficient for unique identification of an inode.
862 * If the inode is in the cache, the inode is returned with an incremented
865 * Otherwise NULL is returned.
867 * Note, @test is called with the inode_lock held, so can't sleep.
869 static struct inode
*ifind(struct super_block
*sb
,
870 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
871 void *data
, const int wait
)
875 spin_lock(&inode_lock
);
876 inode
= find_inode(sb
, head
, test
, data
);
879 spin_unlock(&inode_lock
);
881 wait_on_inode(inode
);
884 spin_unlock(&inode_lock
);
889 * ifind_fast - internal function, you want ilookup() or iget().
890 * @sb: super block of file system to search
891 * @head: head of the list to search
892 * @ino: inode number to search for
894 * ifind_fast() searches for the inode @ino in the inode cache. This is for
895 * file systems where the inode number is sufficient for unique identification
898 * If the inode is in the cache, the inode is returned with an incremented
901 * Otherwise NULL is returned.
903 static struct inode
*ifind_fast(struct super_block
*sb
,
904 struct hlist_head
*head
, unsigned long ino
)
908 spin_lock(&inode_lock
);
909 inode
= find_inode_fast(sb
, head
, ino
);
912 spin_unlock(&inode_lock
);
913 wait_on_inode(inode
);
916 spin_unlock(&inode_lock
);
921 * ilookup5_nowait - search for an inode in the inode cache
922 * @sb: super block of file system to search
923 * @hashval: hash value (usually inode number) to search for
924 * @test: callback used for comparisons between inodes
925 * @data: opaque data pointer to pass to @test
927 * ilookup5() uses ifind() to search for the inode specified by @hashval and
928 * @data in the inode cache. This is a generalized version of ilookup() for
929 * file systems where the inode number is not sufficient for unique
930 * identification of an inode.
932 * If the inode is in the cache, the inode is returned with an incremented
933 * reference count. Note, the inode lock is not waited upon so you have to be
934 * very careful what you do with the returned inode. You probably should be
935 * using ilookup5() instead.
937 * Otherwise NULL is returned.
939 * Note, @test is called with the inode_lock held, so can't sleep.
941 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
942 int (*test
)(struct inode
*, void *), void *data
)
944 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
946 return ifind(sb
, head
, test
, data
, 0);
948 EXPORT_SYMBOL(ilookup5_nowait
);
951 * ilookup5 - search for an inode in the inode cache
952 * @sb: super block of file system to search
953 * @hashval: hash value (usually inode number) to search for
954 * @test: callback used for comparisons between inodes
955 * @data: opaque data pointer to pass to @test
957 * ilookup5() uses ifind() to search for the inode specified by @hashval and
958 * @data in the inode cache. This is a generalized version of ilookup() for
959 * file systems where the inode number is not sufficient for unique
960 * identification of an inode.
962 * If the inode is in the cache, the inode lock is waited upon and the inode is
963 * returned with an incremented reference count.
965 * Otherwise NULL is returned.
967 * Note, @test is called with the inode_lock held, so can't sleep.
969 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
970 int (*test
)(struct inode
*, void *), void *data
)
972 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
974 return ifind(sb
, head
, test
, data
, 1);
976 EXPORT_SYMBOL(ilookup5
);
979 * ilookup - search for an inode in the inode cache
980 * @sb: super block of file system to search
981 * @ino: inode number to search for
983 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
984 * This is for file systems where the inode number is sufficient for unique
985 * identification of an inode.
987 * If the inode is in the cache, the inode is returned with an incremented
990 * Otherwise NULL is returned.
992 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
994 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
996 return ifind_fast(sb
, head
, ino
);
998 EXPORT_SYMBOL(ilookup
);
1001 * iget5_locked - obtain an inode from a mounted file system
1002 * @sb: super block of file system
1003 * @hashval: hash value (usually inode number) to get
1004 * @test: callback used for comparisons between inodes
1005 * @set: callback used to initialize a new struct inode
1006 * @data: opaque data pointer to pass to @test and @set
1008 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1009 * and @data in the inode cache and if present it is returned with an increased
1010 * reference count. This is a generalized version of iget_locked() for file
1011 * systems where the inode number is not sufficient for unique identification
1014 * If the inode is not in cache, get_new_inode() is called to allocate a new
1015 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1016 * file system gets to fill it in before unlocking it via unlock_new_inode().
1018 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1020 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1021 int (*test
)(struct inode
*, void *),
1022 int (*set
)(struct inode
*, void *), void *data
)
1024 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1025 struct inode
*inode
;
1027 inode
= ifind(sb
, head
, test
, data
, 1);
1031 * get_new_inode() will do the right thing, re-trying the search
1032 * in case it had to block at any point.
1034 return get_new_inode(sb
, head
, test
, set
, data
);
1036 EXPORT_SYMBOL(iget5_locked
);
1039 * iget_locked - obtain an inode from a mounted file system
1040 * @sb: super block of file system
1041 * @ino: inode number to get
1043 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1044 * the inode cache and if present it is returned with an increased reference
1045 * count. This is for file systems where the inode number is sufficient for
1046 * unique identification of an inode.
1048 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1049 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1050 * The file system gets to fill it in before unlocking it via
1051 * unlock_new_inode().
1053 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1055 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1056 struct inode
*inode
;
1058 inode
= ifind_fast(sb
, head
, ino
);
1062 * get_new_inode_fast() will do the right thing, re-trying the search
1063 * in case it had to block at any point.
1065 return get_new_inode_fast(sb
, head
, ino
);
1067 EXPORT_SYMBOL(iget_locked
);
1069 int insert_inode_locked(struct inode
*inode
)
1071 struct super_block
*sb
= inode
->i_sb
;
1072 ino_t ino
= inode
->i_ino
;
1073 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1075 inode
->i_state
|= I_NEW
;
1077 struct hlist_node
*node
;
1078 struct inode
*old
= NULL
;
1079 spin_lock(&inode_lock
);
1080 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1081 if (old
->i_ino
!= ino
)
1083 if (old
->i_sb
!= sb
)
1085 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
))
1089 if (likely(!node
)) {
1090 hlist_add_head(&inode
->i_hash
, head
);
1091 spin_unlock(&inode_lock
);
1095 spin_unlock(&inode_lock
);
1097 if (unlikely(!inode_unhashed(old
))) {
1104 EXPORT_SYMBOL(insert_inode_locked
);
1106 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1107 int (*test
)(struct inode
*, void *), void *data
)
1109 struct super_block
*sb
= inode
->i_sb
;
1110 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1112 inode
->i_state
|= I_NEW
;
1115 struct hlist_node
*node
;
1116 struct inode
*old
= NULL
;
1118 spin_lock(&inode_lock
);
1119 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1120 if (old
->i_sb
!= sb
)
1122 if (!test(old
, data
))
1124 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
))
1128 if (likely(!node
)) {
1129 hlist_add_head(&inode
->i_hash
, head
);
1130 spin_unlock(&inode_lock
);
1134 spin_unlock(&inode_lock
);
1136 if (unlikely(!inode_unhashed(old
))) {
1143 EXPORT_SYMBOL(insert_inode_locked4
);
1146 * __insert_inode_hash - hash an inode
1147 * @inode: unhashed inode
1148 * @hashval: unsigned long value used to locate this object in the
1151 * Add an inode to the inode hash for this superblock.
1153 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1155 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1156 spin_lock(&inode_lock
);
1157 hlist_add_head(&inode
->i_hash
, head
);
1158 spin_unlock(&inode_lock
);
1160 EXPORT_SYMBOL(__insert_inode_hash
);
1163 * remove_inode_hash - remove an inode from the hash
1164 * @inode: inode to unhash
1166 * Remove an inode from the superblock.
1168 void remove_inode_hash(struct inode
*inode
)
1170 spin_lock(&inode_lock
);
1171 hlist_del_init(&inode
->i_hash
);
1172 spin_unlock(&inode_lock
);
1174 EXPORT_SYMBOL(remove_inode_hash
);
1176 int generic_delete_inode(struct inode
*inode
)
1180 EXPORT_SYMBOL(generic_delete_inode
);
1183 * Normal UNIX filesystem behaviour: delete the
1184 * inode when the usage count drops to zero, and
1187 int generic_drop_inode(struct inode
*inode
)
1189 return !inode
->i_nlink
|| inode_unhashed(inode
);
1191 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1194 * Called when we're dropping the last reference
1197 * Call the FS "drop_inode()" function, defaulting to
1198 * the legacy UNIX filesystem behaviour. If it tells
1199 * us to evict inode, do so. Otherwise, retain inode
1200 * in cache if fs is alive, sync and evict if fs is
1203 static void iput_final(struct inode
*inode
)
1205 struct super_block
*sb
= inode
->i_sb
;
1206 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1209 if (op
&& op
->drop_inode
)
1210 drop
= op
->drop_inode(inode
);
1212 drop
= generic_drop_inode(inode
);
1215 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1216 list_move(&inode
->i_list
, &inode_unused
);
1217 inodes_stat
.nr_unused
++;
1218 if (sb
->s_flags
& MS_ACTIVE
) {
1219 spin_unlock(&inode_lock
);
1222 WARN_ON(inode
->i_state
& I_NEW
);
1223 inode
->i_state
|= I_WILL_FREE
;
1224 spin_unlock(&inode_lock
);
1225 write_inode_now(inode
, 1);
1226 spin_lock(&inode_lock
);
1227 WARN_ON(inode
->i_state
& I_NEW
);
1228 inode
->i_state
&= ~I_WILL_FREE
;
1229 inodes_stat
.nr_unused
--;
1230 hlist_del_init(&inode
->i_hash
);
1232 list_del_init(&inode
->i_list
);
1233 list_del_init(&inode
->i_sb_list
);
1234 WARN_ON(inode
->i_state
& I_NEW
);
1235 inode
->i_state
|= I_FREEING
;
1236 inodes_stat
.nr_inodes
--;
1237 spin_unlock(&inode_lock
);
1239 spin_lock(&inode_lock
);
1240 hlist_del_init(&inode
->i_hash
);
1241 spin_unlock(&inode_lock
);
1242 wake_up_inode(inode
);
1243 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
1244 destroy_inode(inode
);
1248 * iput - put an inode
1249 * @inode: inode to put
1251 * Puts an inode, dropping its usage count. If the inode use count hits
1252 * zero, the inode is then freed and may also be destroyed.
1254 * Consequently, iput() can sleep.
1256 void iput(struct inode
*inode
)
1259 BUG_ON(inode
->i_state
& I_CLEAR
);
1261 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1265 EXPORT_SYMBOL(iput
);
1268 * bmap - find a block number in a file
1269 * @inode: inode of file
1270 * @block: block to find
1272 * Returns the block number on the device holding the inode that
1273 * is the disk block number for the block of the file requested.
1274 * That is, asked for block 4 of inode 1 the function will return the
1275 * disk block relative to the disk start that holds that block of the
1278 sector_t
bmap(struct inode
*inode
, sector_t block
)
1281 if (inode
->i_mapping
->a_ops
->bmap
)
1282 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1285 EXPORT_SYMBOL(bmap
);
1288 * With relative atime, only update atime if the previous atime is
1289 * earlier than either the ctime or mtime or if at least a day has
1290 * passed since the last atime update.
1292 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1293 struct timespec now
)
1296 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1299 * Is mtime younger than atime? If yes, update atime:
1301 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1304 * Is ctime younger than atime? If yes, update atime:
1306 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1310 * Is the previous atime value older than a day? If yes,
1313 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1316 * Good, we can skip the atime update:
1322 * touch_atime - update the access time
1323 * @mnt: mount the inode is accessed on
1324 * @dentry: dentry accessed
1326 * Update the accessed time on an inode and mark it for writeback.
1327 * This function automatically handles read only file systems and media,
1328 * as well as the "noatime" flag and inode specific "noatime" markers.
1330 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1332 struct inode
*inode
= dentry
->d_inode
;
1333 struct timespec now
;
1335 if (inode
->i_flags
& S_NOATIME
)
1337 if (IS_NOATIME(inode
))
1339 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1342 if (mnt
->mnt_flags
& MNT_NOATIME
)
1344 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1347 now
= current_fs_time(inode
->i_sb
);
1349 if (!relatime_need_update(mnt
, inode
, now
))
1352 if (timespec_equal(&inode
->i_atime
, &now
))
1355 if (mnt_want_write(mnt
))
1358 inode
->i_atime
= now
;
1359 mark_inode_dirty_sync(inode
);
1360 mnt_drop_write(mnt
);
1362 EXPORT_SYMBOL(touch_atime
);
1365 * file_update_time - update mtime and ctime time
1366 * @file: file accessed
1368 * Update the mtime and ctime members of an inode and mark the inode
1369 * for writeback. Note that this function is meant exclusively for
1370 * usage in the file write path of filesystems, and filesystems may
1371 * choose to explicitly ignore update via this function with the
1372 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1373 * timestamps are handled by the server.
1376 void file_update_time(struct file
*file
)
1378 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1379 struct timespec now
;
1380 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1382 /* First try to exhaust all avenues to not sync */
1383 if (IS_NOCMTIME(inode
))
1386 now
= current_fs_time(inode
->i_sb
);
1387 if (!timespec_equal(&inode
->i_mtime
, &now
))
1390 if (!timespec_equal(&inode
->i_ctime
, &now
))
1393 if (IS_I_VERSION(inode
))
1394 sync_it
|= S_VERSION
;
1399 /* Finally allowed to write? Takes lock. */
1400 if (mnt_want_write_file(file
))
1403 /* Only change inode inside the lock region */
1404 if (sync_it
& S_VERSION
)
1405 inode_inc_iversion(inode
);
1406 if (sync_it
& S_CTIME
)
1407 inode
->i_ctime
= now
;
1408 if (sync_it
& S_MTIME
)
1409 inode
->i_mtime
= now
;
1410 mark_inode_dirty_sync(inode
);
1411 mnt_drop_write(file
->f_path
.mnt
);
1413 EXPORT_SYMBOL(file_update_time
);
1415 int inode_needs_sync(struct inode
*inode
)
1419 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1423 EXPORT_SYMBOL(inode_needs_sync
);
1425 int inode_wait(void *word
)
1430 EXPORT_SYMBOL(inode_wait
);
1433 * If we try to find an inode in the inode hash while it is being
1434 * deleted, we have to wait until the filesystem completes its
1435 * deletion before reporting that it isn't found. This function waits
1436 * until the deletion _might_ have completed. Callers are responsible
1437 * to recheck inode state.
1439 * It doesn't matter if I_NEW is not set initially, a call to
1440 * wake_up_inode() after removing from the hash list will DTRT.
1442 * This is called with inode_lock held.
1444 static void __wait_on_freeing_inode(struct inode
*inode
)
1446 wait_queue_head_t
*wq
;
1447 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1448 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1449 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1450 spin_unlock(&inode_lock
);
1452 finish_wait(wq
, &wait
.wait
);
1453 spin_lock(&inode_lock
);
1456 static __initdata
unsigned long ihash_entries
;
1457 static int __init
set_ihash_entries(char *str
)
1461 ihash_entries
= simple_strtoul(str
, &str
, 0);
1464 __setup("ihash_entries=", set_ihash_entries
);
1467 * Initialize the waitqueues and inode hash table.
1469 void __init
inode_init_early(void)
1473 /* If hashes are distributed across NUMA nodes, defer
1474 * hash allocation until vmalloc space is available.
1480 alloc_large_system_hash("Inode-cache",
1481 sizeof(struct hlist_head
),
1489 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1490 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1493 void __init
inode_init(void)
1497 /* inode slab cache */
1498 inode_cachep
= kmem_cache_create("inode_cache",
1499 sizeof(struct inode
),
1501 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1504 register_shrinker(&icache_shrinker
);
1506 /* Hash may have been set up in inode_init_early */
1511 alloc_large_system_hash("Inode-cache",
1512 sizeof(struct hlist_head
),
1520 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1521 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1524 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1526 inode
->i_mode
= mode
;
1527 if (S_ISCHR(mode
)) {
1528 inode
->i_fop
= &def_chr_fops
;
1529 inode
->i_rdev
= rdev
;
1530 } else if (S_ISBLK(mode
)) {
1531 inode
->i_fop
= &def_blk_fops
;
1532 inode
->i_rdev
= rdev
;
1533 } else if (S_ISFIFO(mode
))
1534 inode
->i_fop
= &def_fifo_fops
;
1535 else if (S_ISSOCK(mode
))
1536 inode
->i_fop
= &bad_sock_fops
;
1538 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1539 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1542 EXPORT_SYMBOL(init_special_inode
);
1545 * Init uid,gid,mode for new inode according to posix standards
1547 * @dir: Directory inode
1548 * @mode: mode of the new inode
1550 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1553 inode
->i_uid
= current_fsuid();
1554 if (dir
&& dir
->i_mode
& S_ISGID
) {
1555 inode
->i_gid
= dir
->i_gid
;
1559 inode
->i_gid
= current_fsgid();
1560 inode
->i_mode
= mode
;
1562 EXPORT_SYMBOL(inode_init_owner
);