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>
28 #include <linux/buffer_head.h>
31 * New inode.c implementation.
33 * This implementation has the basic premise of trying
34 * to be extremely low-overhead and SMP-safe, yet be
35 * simple enough to be "obviously correct".
40 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
42 /* #define INODE_PARANOIA 1 */
43 /* #define INODE_DEBUG 1 */
46 * Inode lookup is no longer as critical as it used to be:
47 * most of the lookups are going to be through the dcache.
49 #define I_HASHBITS i_hash_shift
50 #define I_HASHMASK i_hash_mask
52 static unsigned int i_hash_mask __read_mostly
;
53 static unsigned int i_hash_shift __read_mostly
;
56 * Each inode can be on two separate lists. One is
57 * the hash list of the inode, used for lookups. The
58 * other linked list is the "type" list:
59 * "in_use" - valid inode, i_count > 0, i_nlink > 0
60 * "dirty" - as "in_use" but also dirty
61 * "unused" - valid inode, i_count = 0
63 * A "dirty" list is maintained for each super block,
64 * allowing for low-overhead inode sync() operations.
67 LIST_HEAD(inode_in_use
);
68 LIST_HEAD(inode_unused
);
69 static struct hlist_head
*inode_hashtable __read_mostly
;
72 * A simple spinlock to protect the list manipulations.
74 * NOTE! You also have to own the lock if you change
75 * the i_state of an inode while it is in use..
77 DEFINE_SPINLOCK(inode_lock
);
80 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
81 * icache shrinking path, and the umount path. Without this exclusion,
82 * by the time prune_icache calls iput for the inode whose pages it has
83 * been invalidating, or by the time it calls clear_inode & destroy_inode
84 * from its final dispose_list, the struct super_block they refer to
85 * (for inode->i_sb->s_op) may already have been freed and reused.
87 * We make this an rwsem because the fastpath is icache shrinking. In
88 * some cases a filesystem may be doing a significant amount of work in
89 * its inode reclaim code, so this should improve parallelism.
91 static DECLARE_RWSEM(iprune_sem
);
94 * Statistics gathering..
96 struct inodes_stat_t inodes_stat
;
98 static struct kmem_cache
*inode_cachep __read_mostly
;
100 static void wake_up_inode(struct inode
*inode
)
103 * Prevent speculative execution through spin_unlock(&inode_lock);
106 wake_up_bit(&inode
->i_state
, __I_NEW
);
110 * inode_init_always - perform inode structure intialisation
111 * @sb: superblock inode belongs to
112 * @inode: inode to initialise
114 * These are initializations that need to be done on every inode
115 * allocation as the fields are not initialised by slab allocation.
117 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
119 static const struct address_space_operations empty_aops
;
120 static const struct inode_operations empty_iops
;
121 static const struct file_operations empty_fops
;
122 struct address_space
*const mapping
= &inode
->i_data
;
125 inode
->i_blkbits
= sb
->s_blocksize_bits
;
127 atomic_set(&inode
->i_count
, 1);
128 inode
->i_op
= &empty_iops
;
129 inode
->i_fop
= &empty_fops
;
133 atomic_set(&inode
->i_writecount
, 0);
137 inode
->i_generation
= 0;
139 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
141 inode
->i_pipe
= NULL
;
142 inode
->i_bdev
= NULL
;
143 inode
->i_cdev
= NULL
;
145 inode
->dirtied_when
= 0;
147 if (security_inode_alloc(inode
))
149 spin_lock_init(&inode
->i_lock
);
150 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
152 mutex_init(&inode
->i_mutex
);
153 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
155 init_rwsem(&inode
->i_alloc_sem
);
156 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
158 mapping
->a_ops
= &empty_aops
;
159 mapping
->host
= inode
;
161 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
162 mapping
->assoc_mapping
= NULL
;
163 mapping
->backing_dev_info
= &default_backing_dev_info
;
164 mapping
->writeback_index
= 0;
167 * If the block_device provides a backing_dev_info for client
168 * inodes then use that. Otherwise the inode share the bdev's
172 struct backing_dev_info
*bdi
;
174 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
175 mapping
->backing_dev_info
= bdi
;
177 inode
->i_private
= NULL
;
178 inode
->i_mapping
= mapping
;
179 #ifdef CONFIG_FS_POSIX_ACL
180 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
183 #ifdef CONFIG_FSNOTIFY
184 inode
->i_fsnotify_mask
= 0;
191 EXPORT_SYMBOL(inode_init_always
);
193 static struct inode
*alloc_inode(struct super_block
*sb
)
197 if (sb
->s_op
->alloc_inode
)
198 inode
= sb
->s_op
->alloc_inode(sb
);
200 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
205 if (unlikely(inode_init_always(sb
, inode
))) {
206 if (inode
->i_sb
->s_op
->destroy_inode
)
207 inode
->i_sb
->s_op
->destroy_inode(inode
);
209 kmem_cache_free(inode_cachep
, inode
);
216 void __destroy_inode(struct inode
*inode
)
218 BUG_ON(inode_has_buffers(inode
));
219 security_inode_free(inode
);
220 fsnotify_inode_delete(inode
);
221 #ifdef CONFIG_FS_POSIX_ACL
222 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
223 posix_acl_release(inode
->i_acl
);
224 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
225 posix_acl_release(inode
->i_default_acl
);
228 EXPORT_SYMBOL(__destroy_inode
);
230 void destroy_inode(struct inode
*inode
)
232 __destroy_inode(inode
);
233 if (inode
->i_sb
->s_op
->destroy_inode
)
234 inode
->i_sb
->s_op
->destroy_inode(inode
);
236 kmem_cache_free(inode_cachep
, (inode
));
240 * These are initializations that only need to be done
241 * once, because the fields are idempotent across use
242 * of the inode, so let the slab aware of that.
244 void inode_init_once(struct inode
*inode
)
246 memset(inode
, 0, sizeof(*inode
));
247 INIT_HLIST_NODE(&inode
->i_hash
);
248 INIT_LIST_HEAD(&inode
->i_dentry
);
249 INIT_LIST_HEAD(&inode
->i_devices
);
250 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
251 spin_lock_init(&inode
->i_data
.tree_lock
);
252 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
253 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
254 spin_lock_init(&inode
->i_data
.private_lock
);
255 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
256 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
257 i_size_ordered_init(inode
);
258 #ifdef CONFIG_FSNOTIFY
259 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
262 EXPORT_SYMBOL(inode_init_once
);
264 static void init_once(void *foo
)
266 struct inode
*inode
= (struct inode
*) foo
;
268 inode_init_once(inode
);
272 * inode_lock must be held
274 void __iget(struct inode
*inode
)
276 if (atomic_inc_return(&inode
->i_count
) != 1)
279 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
280 list_move(&inode
->i_list
, &inode_in_use
);
281 inodes_stat
.nr_unused
--;
284 void end_writeback(struct inode
*inode
)
287 BUG_ON(inode
->i_data
.nrpages
);
288 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
289 BUG_ON(!(inode
->i_state
& I_FREEING
));
290 BUG_ON(inode
->i_state
& I_CLEAR
);
291 inode_sync_wait(inode
);
292 inode
->i_state
= I_FREEING
| I_CLEAR
;
294 EXPORT_SYMBOL(end_writeback
);
296 static void evict(struct inode
*inode
)
298 const struct super_operations
*op
= inode
->i_sb
->s_op
;
300 if (op
->evict_inode
) {
301 op
->evict_inode(inode
);
303 if (inode
->i_data
.nrpages
)
304 truncate_inode_pages(&inode
->i_data
, 0);
305 end_writeback(inode
);
307 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
309 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
314 * dispose_list - dispose of the contents of a local list
315 * @head: the head of the list to free
317 * Dispose-list gets a local list with local inodes in it, so it doesn't
318 * need to worry about list corruption and SMP locks.
320 static void dispose_list(struct list_head
*head
)
324 while (!list_empty(head
)) {
327 inode
= list_first_entry(head
, struct inode
, i_list
);
328 list_del(&inode
->i_list
);
332 spin_lock(&inode_lock
);
333 hlist_del_init(&inode
->i_hash
);
334 list_del_init(&inode
->i_sb_list
);
335 spin_unlock(&inode_lock
);
337 wake_up_inode(inode
);
338 destroy_inode(inode
);
341 spin_lock(&inode_lock
);
342 inodes_stat
.nr_inodes
-= nr_disposed
;
343 spin_unlock(&inode_lock
);
347 * Invalidate all inodes for a device.
349 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
351 struct list_head
*next
;
352 int busy
= 0, count
= 0;
356 struct list_head
*tmp
= next
;
360 * We can reschedule here without worrying about the list's
361 * consistency because the per-sb list of inodes must not
362 * change during umount anymore, and because iprune_sem keeps
363 * shrink_icache_memory() away.
365 cond_resched_lock(&inode_lock
);
370 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
371 if (inode
->i_state
& I_NEW
)
373 invalidate_inode_buffers(inode
);
374 if (!atomic_read(&inode
->i_count
)) {
375 list_move(&inode
->i_list
, dispose
);
376 WARN_ON(inode
->i_state
& I_NEW
);
377 inode
->i_state
|= I_FREEING
;
383 /* only unused inodes may be cached with i_count zero */
384 inodes_stat
.nr_unused
-= count
;
389 * invalidate_inodes - discard the inodes on a device
392 * Discard all of the inodes for a given superblock. If the discard
393 * fails because there are busy inodes then a non zero value is returned.
394 * If the discard is successful all the inodes have been discarded.
396 int invalidate_inodes(struct super_block
*sb
)
399 LIST_HEAD(throw_away
);
401 down_write(&iprune_sem
);
402 spin_lock(&inode_lock
);
403 fsnotify_unmount_inodes(&sb
->s_inodes
);
404 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
405 spin_unlock(&inode_lock
);
407 dispose_list(&throw_away
);
408 up_write(&iprune_sem
);
412 EXPORT_SYMBOL(invalidate_inodes
);
414 static int can_unuse(struct inode
*inode
)
418 if (inode_has_buffers(inode
))
420 if (atomic_read(&inode
->i_count
))
422 if (inode
->i_data
.nrpages
)
428 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
429 * a temporary list and then are freed outside inode_lock by dispose_list().
431 * Any inodes which are pinned purely because of attached pagecache have their
432 * pagecache removed. We expect the final iput() on that inode to add it to
433 * the front of the inode_unused list. So look for it there and if the
434 * inode is still freeable, proceed. The right inode is found 99.9% of the
435 * time in testing on a 4-way.
437 * If the inode has metadata buffers attached to mapping->private_list then
438 * try to remove them.
440 static void prune_icache(int nr_to_scan
)
445 unsigned long reap
= 0;
447 down_read(&iprune_sem
);
448 spin_lock(&inode_lock
);
449 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
452 if (list_empty(&inode_unused
))
455 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
457 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
458 list_move(&inode
->i_list
, &inode_unused
);
461 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
463 spin_unlock(&inode_lock
);
464 if (remove_inode_buffers(inode
))
465 reap
+= invalidate_mapping_pages(&inode
->i_data
,
468 spin_lock(&inode_lock
);
470 if (inode
!= list_entry(inode_unused
.next
,
471 struct inode
, i_list
))
472 continue; /* wrong inode or list_empty */
473 if (!can_unuse(inode
))
476 list_move(&inode
->i_list
, &freeable
);
477 WARN_ON(inode
->i_state
& I_NEW
);
478 inode
->i_state
|= I_FREEING
;
481 inodes_stat
.nr_unused
-= nr_pruned
;
482 if (current_is_kswapd())
483 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
485 __count_vm_events(PGINODESTEAL
, reap
);
486 spin_unlock(&inode_lock
);
488 dispose_list(&freeable
);
489 up_read(&iprune_sem
);
493 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
494 * "unused" means that no dentries are referring to the inodes: the files are
495 * not open and the dcache references to those inodes have already been
498 * This function is passed the number of inodes to scan, and it returns the
499 * total number of remaining possibly-reclaimable inodes.
501 static int shrink_icache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
505 * Nasty deadlock avoidance. We may hold various FS locks,
506 * and we don't want to recurse into the FS that called us
507 * in clear_inode() and friends..
509 if (!(gfp_mask
& __GFP_FS
))
513 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
516 static struct shrinker icache_shrinker
= {
517 .shrink
= shrink_icache_memory
,
518 .seeks
= DEFAULT_SEEKS
,
521 static void __wait_on_freeing_inode(struct inode
*inode
);
523 * Called with the inode lock held.
524 * NOTE: we are not increasing the inode-refcount, you must call __iget()
525 * by hand after calling find_inode now! This simplifies iunique and won't
526 * add any additional branch in the common code.
528 static struct inode
*find_inode(struct super_block
*sb
,
529 struct hlist_head
*head
,
530 int (*test
)(struct inode
*, void *),
533 struct hlist_node
*node
;
534 struct inode
*inode
= NULL
;
537 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
538 if (inode
->i_sb
!= sb
)
540 if (!test(inode
, data
))
542 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
543 __wait_on_freeing_inode(inode
);
548 return node
? inode
: NULL
;
552 * find_inode_fast is the fast path version of find_inode, see the comment at
553 * iget_locked for details.
555 static struct inode
*find_inode_fast(struct super_block
*sb
,
556 struct hlist_head
*head
, unsigned long ino
)
558 struct hlist_node
*node
;
559 struct inode
*inode
= NULL
;
562 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
563 if (inode
->i_ino
!= ino
)
565 if (inode
->i_sb
!= sb
)
567 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
568 __wait_on_freeing_inode(inode
);
573 return node
? inode
: NULL
;
576 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
580 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
582 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
583 return tmp
& I_HASHMASK
;
587 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
590 inodes_stat
.nr_inodes
++;
591 list_add(&inode
->i_list
, &inode_in_use
);
592 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
594 hlist_add_head(&inode
->i_hash
, head
);
598 * inode_add_to_lists - add a new inode to relevant lists
599 * @sb: superblock inode belongs to
600 * @inode: inode to mark in use
602 * When an inode is allocated it needs to be accounted for, added to the in use
603 * list, the owning superblock and the inode hash. This needs to be done under
604 * the inode_lock, so export a function to do this rather than the inode lock
605 * itself. We calculate the hash list to add to here so it is all internal
606 * which requires the caller to have already set up the inode number in the
609 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
611 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
613 spin_lock(&inode_lock
);
614 __inode_add_to_lists(sb
, head
, inode
);
615 spin_unlock(&inode_lock
);
617 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
620 * new_inode - obtain an inode
623 * Allocates a new inode for given superblock. The default gfp_mask
624 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
625 * If HIGHMEM pages are unsuitable or it is known that pages allocated
626 * for the page cache are not reclaimable or migratable,
627 * mapping_set_gfp_mask() must be called with suitable flags on the
628 * newly created inode's mapping
631 struct inode
*new_inode(struct super_block
*sb
)
634 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
635 * error if st_ino won't fit in target struct field. Use 32bit counter
636 * here to attempt to avoid that.
638 static unsigned int last_ino
;
641 spin_lock_prefetch(&inode_lock
);
643 inode
= alloc_inode(sb
);
645 spin_lock(&inode_lock
);
646 __inode_add_to_lists(sb
, NULL
, inode
);
647 inode
->i_ino
= ++last_ino
;
649 spin_unlock(&inode_lock
);
653 EXPORT_SYMBOL(new_inode
);
655 void unlock_new_inode(struct inode
*inode
)
657 #ifdef CONFIG_DEBUG_LOCK_ALLOC
658 if (inode
->i_mode
& S_IFDIR
) {
659 struct file_system_type
*type
= inode
->i_sb
->s_type
;
661 /* Set new key only if filesystem hasn't already changed it */
662 if (!lockdep_match_class(&inode
->i_mutex
,
663 &type
->i_mutex_key
)) {
665 * ensure nobody is actually holding i_mutex
667 mutex_destroy(&inode
->i_mutex
);
668 mutex_init(&inode
->i_mutex
);
669 lockdep_set_class(&inode
->i_mutex
,
670 &type
->i_mutex_dir_key
);
675 * This is special! We do not need the spinlock when clearing I_NEW,
676 * because we're guaranteed that nobody else tries to do anything about
677 * the state of the inode when it is locked, as we just created it (so
678 * there can be no old holders that haven't tested I_NEW).
679 * However we must emit the memory barrier so that other CPUs reliably
680 * see the clearing of I_NEW after the other inode initialisation has
684 WARN_ON(!(inode
->i_state
& I_NEW
));
685 inode
->i_state
&= ~I_NEW
;
686 wake_up_inode(inode
);
688 EXPORT_SYMBOL(unlock_new_inode
);
691 * This is called without the inode lock held.. Be careful.
693 * We no longer cache the sb_flags in i_flags - see fs.h
694 * -- rmk@arm.uk.linux.org
696 static struct inode
*get_new_inode(struct super_block
*sb
,
697 struct hlist_head
*head
,
698 int (*test
)(struct inode
*, void *),
699 int (*set
)(struct inode
*, void *),
704 inode
= alloc_inode(sb
);
708 spin_lock(&inode_lock
);
709 /* We released the lock, so.. */
710 old
= find_inode(sb
, head
, test
, data
);
712 if (set(inode
, data
))
715 __inode_add_to_lists(sb
, head
, inode
);
716 inode
->i_state
= I_NEW
;
717 spin_unlock(&inode_lock
);
719 /* Return the locked inode with I_NEW set, the
720 * caller is responsible for filling in the contents
726 * Uhhuh, somebody else created the same inode under
727 * us. Use the old inode instead of the one we just
731 spin_unlock(&inode_lock
);
732 destroy_inode(inode
);
734 wait_on_inode(inode
);
739 spin_unlock(&inode_lock
);
740 destroy_inode(inode
);
745 * get_new_inode_fast is the fast path version of get_new_inode, see the
746 * comment at iget_locked for details.
748 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
749 struct hlist_head
*head
, unsigned long ino
)
753 inode
= alloc_inode(sb
);
757 spin_lock(&inode_lock
);
758 /* We released the lock, so.. */
759 old
= find_inode_fast(sb
, head
, ino
);
762 __inode_add_to_lists(sb
, head
, inode
);
763 inode
->i_state
= I_NEW
;
764 spin_unlock(&inode_lock
);
766 /* Return the locked inode with I_NEW set, the
767 * caller is responsible for filling in the contents
773 * Uhhuh, somebody else created the same inode under
774 * us. Use the old inode instead of the one we just
778 spin_unlock(&inode_lock
);
779 destroy_inode(inode
);
781 wait_on_inode(inode
);
787 * iunique - get a unique inode number
789 * @max_reserved: highest reserved inode number
791 * Obtain an inode number that is unique on the system for a given
792 * superblock. This is used by file systems that have no natural
793 * permanent inode numbering system. An inode number is returned that
794 * is higher than the reserved limit but unique.
797 * With a large number of inodes live on the file system this function
798 * currently becomes quite slow.
800 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
803 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
804 * error if st_ino won't fit in target struct field. Use 32bit counter
805 * here to attempt to avoid that.
807 static unsigned int counter
;
809 struct hlist_head
*head
;
812 spin_lock(&inode_lock
);
814 if (counter
<= max_reserved
)
815 counter
= max_reserved
+ 1;
817 head
= inode_hashtable
+ hash(sb
, res
);
818 inode
= find_inode_fast(sb
, head
, res
);
819 } while (inode
!= NULL
);
820 spin_unlock(&inode_lock
);
824 EXPORT_SYMBOL(iunique
);
826 struct inode
*igrab(struct inode
*inode
)
828 spin_lock(&inode_lock
);
829 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
)))
833 * Handle the case where s_op->clear_inode is not been
834 * called yet, and somebody is calling igrab
835 * while the inode is getting freed.
838 spin_unlock(&inode_lock
);
841 EXPORT_SYMBOL(igrab
);
844 * ifind - internal function, you want ilookup5() or iget5().
845 * @sb: super block of file system to search
846 * @head: the head of the list to search
847 * @test: callback used for comparisons between inodes
848 * @data: opaque data pointer to pass to @test
849 * @wait: if true wait for the inode to be unlocked, if false do not
851 * ifind() searches for the inode specified by @data in the inode
852 * cache. This is a generalized version of ifind_fast() for file systems where
853 * the inode number is not sufficient for unique identification of an inode.
855 * If the inode is in the cache, the inode is returned with an incremented
858 * Otherwise NULL is returned.
860 * Note, @test is called with the inode_lock held, so can't sleep.
862 static struct inode
*ifind(struct super_block
*sb
,
863 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
864 void *data
, const int wait
)
868 spin_lock(&inode_lock
);
869 inode
= find_inode(sb
, head
, test
, data
);
872 spin_unlock(&inode_lock
);
874 wait_on_inode(inode
);
877 spin_unlock(&inode_lock
);
882 * ifind_fast - internal function, you want ilookup() or iget().
883 * @sb: super block of file system to search
884 * @head: head of the list to search
885 * @ino: inode number to search for
887 * ifind_fast() searches for the inode @ino in the inode cache. This is for
888 * file systems where the inode number is sufficient for unique identification
891 * If the inode is in the cache, the inode is returned with an incremented
894 * Otherwise NULL is returned.
896 static struct inode
*ifind_fast(struct super_block
*sb
,
897 struct hlist_head
*head
, unsigned long ino
)
901 spin_lock(&inode_lock
);
902 inode
= find_inode_fast(sb
, head
, ino
);
905 spin_unlock(&inode_lock
);
906 wait_on_inode(inode
);
909 spin_unlock(&inode_lock
);
914 * ilookup5_nowait - search for an inode in the inode cache
915 * @sb: super block of file system to search
916 * @hashval: hash value (usually inode number) to search for
917 * @test: callback used for comparisons between inodes
918 * @data: opaque data pointer to pass to @test
920 * ilookup5() uses ifind() to search for the inode specified by @hashval and
921 * @data in the inode cache. This is a generalized version of ilookup() for
922 * file systems where the inode number is not sufficient for unique
923 * identification of an inode.
925 * If the inode is in the cache, the inode is returned with an incremented
926 * reference count. Note, the inode lock is not waited upon so you have to be
927 * very careful what you do with the returned inode. You probably should be
928 * using ilookup5() instead.
930 * Otherwise NULL is returned.
932 * Note, @test is called with the inode_lock held, so can't sleep.
934 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
935 int (*test
)(struct inode
*, void *), void *data
)
937 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
939 return ifind(sb
, head
, test
, data
, 0);
941 EXPORT_SYMBOL(ilookup5_nowait
);
944 * ilookup5 - search for an inode in the inode cache
945 * @sb: super block of file system to search
946 * @hashval: hash value (usually inode number) to search for
947 * @test: callback used for comparisons between inodes
948 * @data: opaque data pointer to pass to @test
950 * ilookup5() uses ifind() to search for the inode specified by @hashval and
951 * @data in the inode cache. This is a generalized version of ilookup() for
952 * file systems where the inode number is not sufficient for unique
953 * identification of an inode.
955 * If the inode is in the cache, the inode lock is waited upon and the inode is
956 * returned with an incremented reference count.
958 * Otherwise NULL is returned.
960 * Note, @test is called with the inode_lock held, so can't sleep.
962 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
963 int (*test
)(struct inode
*, void *), void *data
)
965 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
967 return ifind(sb
, head
, test
, data
, 1);
969 EXPORT_SYMBOL(ilookup5
);
972 * ilookup - search for an inode in the inode cache
973 * @sb: super block of file system to search
974 * @ino: inode number to search for
976 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
977 * This is for file systems where the inode number is sufficient for unique
978 * identification of an inode.
980 * If the inode is in the cache, the inode is returned with an incremented
983 * Otherwise NULL is returned.
985 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
987 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
989 return ifind_fast(sb
, head
, ino
);
991 EXPORT_SYMBOL(ilookup
);
994 * iget5_locked - obtain an inode from a mounted file system
995 * @sb: super block of file system
996 * @hashval: hash value (usually inode number) to get
997 * @test: callback used for comparisons between inodes
998 * @set: callback used to initialize a new struct inode
999 * @data: opaque data pointer to pass to @test and @set
1001 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1002 * and @data in the inode cache and if present it is returned with an increased
1003 * reference count. This is a generalized version of iget_locked() for file
1004 * systems where the inode number is not sufficient for unique identification
1007 * If the inode is not in cache, get_new_inode() is called to allocate a new
1008 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1009 * file system gets to fill it in before unlocking it via unlock_new_inode().
1011 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1013 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1014 int (*test
)(struct inode
*, void *),
1015 int (*set
)(struct inode
*, void *), void *data
)
1017 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1018 struct inode
*inode
;
1020 inode
= ifind(sb
, head
, test
, data
, 1);
1024 * get_new_inode() will do the right thing, re-trying the search
1025 * in case it had to block at any point.
1027 return get_new_inode(sb
, head
, test
, set
, data
);
1029 EXPORT_SYMBOL(iget5_locked
);
1032 * iget_locked - obtain an inode from a mounted file system
1033 * @sb: super block of file system
1034 * @ino: inode number to get
1036 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1037 * the inode cache and if present it is returned with an increased reference
1038 * count. This is for file systems where the inode number is sufficient for
1039 * unique identification of an inode.
1041 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1042 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1043 * The file system gets to fill it in before unlocking it via
1044 * unlock_new_inode().
1046 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1048 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1049 struct inode
*inode
;
1051 inode
= ifind_fast(sb
, head
, ino
);
1055 * get_new_inode_fast() will do the right thing, re-trying the search
1056 * in case it had to block at any point.
1058 return get_new_inode_fast(sb
, head
, ino
);
1060 EXPORT_SYMBOL(iget_locked
);
1062 int insert_inode_locked(struct inode
*inode
)
1064 struct super_block
*sb
= inode
->i_sb
;
1065 ino_t ino
= inode
->i_ino
;
1066 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1068 inode
->i_state
|= I_NEW
;
1070 struct hlist_node
*node
;
1071 struct inode
*old
= NULL
;
1072 spin_lock(&inode_lock
);
1073 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1074 if (old
->i_ino
!= ino
)
1076 if (old
->i_sb
!= sb
)
1078 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
))
1082 if (likely(!node
)) {
1083 hlist_add_head(&inode
->i_hash
, head
);
1084 spin_unlock(&inode_lock
);
1088 spin_unlock(&inode_lock
);
1090 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1097 EXPORT_SYMBOL(insert_inode_locked
);
1099 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1100 int (*test
)(struct inode
*, void *), void *data
)
1102 struct super_block
*sb
= inode
->i_sb
;
1103 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1105 inode
->i_state
|= I_NEW
;
1108 struct hlist_node
*node
;
1109 struct inode
*old
= NULL
;
1111 spin_lock(&inode_lock
);
1112 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1113 if (old
->i_sb
!= sb
)
1115 if (!test(old
, data
))
1117 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
))
1121 if (likely(!node
)) {
1122 hlist_add_head(&inode
->i_hash
, head
);
1123 spin_unlock(&inode_lock
);
1127 spin_unlock(&inode_lock
);
1129 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1136 EXPORT_SYMBOL(insert_inode_locked4
);
1139 * __insert_inode_hash - hash an inode
1140 * @inode: unhashed inode
1141 * @hashval: unsigned long value used to locate this object in the
1144 * Add an inode to the inode hash for this superblock.
1146 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1148 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1149 spin_lock(&inode_lock
);
1150 hlist_add_head(&inode
->i_hash
, head
);
1151 spin_unlock(&inode_lock
);
1153 EXPORT_SYMBOL(__insert_inode_hash
);
1156 * remove_inode_hash - remove an inode from the hash
1157 * @inode: inode to unhash
1159 * Remove an inode from the superblock.
1161 void remove_inode_hash(struct inode
*inode
)
1163 spin_lock(&inode_lock
);
1164 hlist_del_init(&inode
->i_hash
);
1165 spin_unlock(&inode_lock
);
1167 EXPORT_SYMBOL(remove_inode_hash
);
1169 int generic_delete_inode(struct inode
*inode
)
1173 EXPORT_SYMBOL(generic_delete_inode
);
1176 * Normal UNIX filesystem behaviour: delete the
1177 * inode when the usage count drops to zero, and
1180 int generic_drop_inode(struct inode
*inode
)
1182 return !inode
->i_nlink
|| hlist_unhashed(&inode
->i_hash
);
1184 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1187 * Called when we're dropping the last reference
1190 * Call the FS "drop_inode()" function, defaulting to
1191 * the legacy UNIX filesystem behaviour. If it tells
1192 * us to evict inode, do so. Otherwise, retain inode
1193 * in cache if fs is alive, sync and evict if fs is
1196 static void iput_final(struct inode
*inode
)
1198 struct super_block
*sb
= inode
->i_sb
;
1199 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1202 if (op
&& op
->drop_inode
)
1203 drop
= op
->drop_inode(inode
);
1205 drop
= generic_drop_inode(inode
);
1208 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1209 list_move(&inode
->i_list
, &inode_unused
);
1210 inodes_stat
.nr_unused
++;
1211 if (sb
->s_flags
& MS_ACTIVE
) {
1212 spin_unlock(&inode_lock
);
1215 WARN_ON(inode
->i_state
& I_NEW
);
1216 inode
->i_state
|= I_WILL_FREE
;
1217 spin_unlock(&inode_lock
);
1218 write_inode_now(inode
, 1);
1219 spin_lock(&inode_lock
);
1220 WARN_ON(inode
->i_state
& I_NEW
);
1221 inode
->i_state
&= ~I_WILL_FREE
;
1222 inodes_stat
.nr_unused
--;
1223 hlist_del_init(&inode
->i_hash
);
1225 list_del_init(&inode
->i_list
);
1226 list_del_init(&inode
->i_sb_list
);
1227 WARN_ON(inode
->i_state
& I_NEW
);
1228 inode
->i_state
|= I_FREEING
;
1229 inodes_stat
.nr_inodes
--;
1230 spin_unlock(&inode_lock
);
1232 spin_lock(&inode_lock
);
1233 hlist_del_init(&inode
->i_hash
);
1234 spin_unlock(&inode_lock
);
1235 wake_up_inode(inode
);
1236 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
1237 destroy_inode(inode
);
1241 * iput - put an inode
1242 * @inode: inode to put
1244 * Puts an inode, dropping its usage count. If the inode use count hits
1245 * zero, the inode is then freed and may also be destroyed.
1247 * Consequently, iput() can sleep.
1249 void iput(struct inode
*inode
)
1252 BUG_ON(inode
->i_state
& I_CLEAR
);
1254 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1258 EXPORT_SYMBOL(iput
);
1261 * bmap - find a block number in a file
1262 * @inode: inode of file
1263 * @block: block to find
1265 * Returns the block number on the device holding the inode that
1266 * is the disk block number for the block of the file requested.
1267 * That is, asked for block 4 of inode 1 the function will return the
1268 * disk block relative to the disk start that holds that block of the
1271 sector_t
bmap(struct inode
*inode
, sector_t block
)
1274 if (inode
->i_mapping
->a_ops
->bmap
)
1275 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1278 EXPORT_SYMBOL(bmap
);
1281 * With relative atime, only update atime if the previous atime is
1282 * earlier than either the ctime or mtime or if at least a day has
1283 * passed since the last atime update.
1285 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1286 struct timespec now
)
1289 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1292 * Is mtime younger than atime? If yes, update atime:
1294 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1297 * Is ctime younger than atime? If yes, update atime:
1299 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1303 * Is the previous atime value older than a day? If yes,
1306 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1309 * Good, we can skip the atime update:
1315 * touch_atime - update the access time
1316 * @mnt: mount the inode is accessed on
1317 * @dentry: dentry accessed
1319 * Update the accessed time on an inode and mark it for writeback.
1320 * This function automatically handles read only file systems and media,
1321 * as well as the "noatime" flag and inode specific "noatime" markers.
1323 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1325 struct inode
*inode
= dentry
->d_inode
;
1326 struct timespec now
;
1328 if (inode
->i_flags
& S_NOATIME
)
1330 if (IS_NOATIME(inode
))
1332 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1335 if (mnt
->mnt_flags
& MNT_NOATIME
)
1337 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1340 now
= current_fs_time(inode
->i_sb
);
1342 if (!relatime_need_update(mnt
, inode
, now
))
1345 if (timespec_equal(&inode
->i_atime
, &now
))
1348 if (mnt_want_write(mnt
))
1351 inode
->i_atime
= now
;
1352 mark_inode_dirty_sync(inode
);
1353 mnt_drop_write(mnt
);
1355 EXPORT_SYMBOL(touch_atime
);
1358 * file_update_time - update mtime and ctime time
1359 * @file: file accessed
1361 * Update the mtime and ctime members of an inode and mark the inode
1362 * for writeback. Note that this function is meant exclusively for
1363 * usage in the file write path of filesystems, and filesystems may
1364 * choose to explicitly ignore update via this function with the
1365 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1366 * timestamps are handled by the server.
1369 void file_update_time(struct file
*file
)
1371 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1372 struct timespec now
;
1373 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1375 /* First try to exhaust all avenues to not sync */
1376 if (IS_NOCMTIME(inode
))
1379 now
= current_fs_time(inode
->i_sb
);
1380 if (!timespec_equal(&inode
->i_mtime
, &now
))
1383 if (!timespec_equal(&inode
->i_ctime
, &now
))
1386 if (IS_I_VERSION(inode
))
1387 sync_it
|= S_VERSION
;
1392 /* Finally allowed to write? Takes lock. */
1393 if (mnt_want_write_file(file
))
1396 /* Only change inode inside the lock region */
1397 if (sync_it
& S_VERSION
)
1398 inode_inc_iversion(inode
);
1399 if (sync_it
& S_CTIME
)
1400 inode
->i_ctime
= now
;
1401 if (sync_it
& S_MTIME
)
1402 inode
->i_mtime
= now
;
1403 mark_inode_dirty_sync(inode
);
1404 mnt_drop_write(file
->f_path
.mnt
);
1406 EXPORT_SYMBOL(file_update_time
);
1408 int inode_needs_sync(struct inode
*inode
)
1412 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1416 EXPORT_SYMBOL(inode_needs_sync
);
1418 int inode_wait(void *word
)
1423 EXPORT_SYMBOL(inode_wait
);
1426 * If we try to find an inode in the inode hash while it is being
1427 * deleted, we have to wait until the filesystem completes its
1428 * deletion before reporting that it isn't found. This function waits
1429 * until the deletion _might_ have completed. Callers are responsible
1430 * to recheck inode state.
1432 * It doesn't matter if I_NEW is not set initially, a call to
1433 * wake_up_inode() after removing from the hash list will DTRT.
1435 * This is called with inode_lock held.
1437 static void __wait_on_freeing_inode(struct inode
*inode
)
1439 wait_queue_head_t
*wq
;
1440 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1441 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1442 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1443 spin_unlock(&inode_lock
);
1445 finish_wait(wq
, &wait
.wait
);
1446 spin_lock(&inode_lock
);
1449 static __initdata
unsigned long ihash_entries
;
1450 static int __init
set_ihash_entries(char *str
)
1454 ihash_entries
= simple_strtoul(str
, &str
, 0);
1457 __setup("ihash_entries=", set_ihash_entries
);
1460 * Initialize the waitqueues and inode hash table.
1462 void __init
inode_init_early(void)
1466 /* If hashes are distributed across NUMA nodes, defer
1467 * hash allocation until vmalloc space is available.
1473 alloc_large_system_hash("Inode-cache",
1474 sizeof(struct hlist_head
),
1482 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1483 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1486 void __init
inode_init(void)
1490 /* inode slab cache */
1491 inode_cachep
= kmem_cache_create("inode_cache",
1492 sizeof(struct inode
),
1494 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1497 register_shrinker(&icache_shrinker
);
1499 /* Hash may have been set up in inode_init_early */
1504 alloc_large_system_hash("Inode-cache",
1505 sizeof(struct hlist_head
),
1513 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1514 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1517 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1519 inode
->i_mode
= mode
;
1520 if (S_ISCHR(mode
)) {
1521 inode
->i_fop
= &def_chr_fops
;
1522 inode
->i_rdev
= rdev
;
1523 } else if (S_ISBLK(mode
)) {
1524 inode
->i_fop
= &def_blk_fops
;
1525 inode
->i_rdev
= rdev
;
1526 } else if (S_ISFIFO(mode
))
1527 inode
->i_fop
= &def_fifo_fops
;
1528 else if (S_ISSOCK(mode
))
1529 inode
->i_fop
= &bad_sock_fops
;
1531 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1532 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1535 EXPORT_SYMBOL(init_special_inode
);
1538 * Init uid,gid,mode for new inode according to posix standards
1540 * @dir: Directory inode
1541 * @mode: mode of the new inode
1543 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1546 inode
->i_uid
= current_fsuid();
1547 if (dir
&& dir
->i_mode
& S_ISGID
) {
1548 inode
->i_gid
= dir
->i_gid
;
1552 inode
->i_gid
= current_fsgid();
1553 inode
->i_mode
= mode
;
1555 EXPORT_SYMBOL(inode_init_owner
);