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/inotify.h>
24 #include <linux/fsnotify.h>
25 #include <linux/mount.h>
26 #include <linux/async.h>
27 #include <linux/posix_acl.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
));
248 void address_space_init_once(struct address_space
*mapping
)
250 memset(mapping
, 0, sizeof(*mapping
));
251 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
252 spin_lock_init(&mapping
->tree_lock
);
253 spin_lock_init(&mapping
->i_mmap_lock
);
254 INIT_LIST_HEAD(&mapping
->private_list
);
255 spin_lock_init(&mapping
->private_lock
);
256 INIT_RAW_PRIO_TREE_ROOT(&mapping
->i_mmap
);
257 INIT_LIST_HEAD(&mapping
->i_mmap_nonlinear
);
258 mutex_init(&mapping
->unmap_mutex
);
260 EXPORT_SYMBOL(address_space_init_once
);
263 * These are initializations that only need to be done
264 * once, because the fields are idempotent across use
265 * of the inode, so let the slab aware of that.
267 void inode_init_once(struct inode
*inode
)
269 memset(inode
, 0, sizeof(*inode
));
270 INIT_HLIST_NODE(&inode
->i_hash
);
271 INIT_LIST_HEAD(&inode
->i_dentry
);
272 INIT_LIST_HEAD(&inode
->i_devices
);
273 address_space_init_once(&inode
->i_data
);
274 i_size_ordered_init(inode
);
275 #ifdef CONFIG_INOTIFY
276 INIT_LIST_HEAD(&inode
->inotify_watches
);
277 mutex_init(&inode
->inotify_mutex
);
279 #ifdef CONFIG_FSNOTIFY
280 INIT_HLIST_HEAD(&inode
->i_fsnotify_mark_entries
);
283 EXPORT_SYMBOL(inode_init_once
);
285 static void init_once(void *foo
)
287 struct inode
*inode
= (struct inode
*) foo
;
289 inode_init_once(inode
);
293 * inode_lock must be held
295 void __iget(struct inode
*inode
)
297 if (atomic_read(&inode
->i_count
)) {
298 atomic_inc(&inode
->i_count
);
301 atomic_inc(&inode
->i_count
);
302 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
303 list_move(&inode
->i_list
, &inode_in_use
);
304 inodes_stat
.nr_unused
--;
308 * clear_inode - clear an inode
309 * @inode: inode to clear
311 * This is called by the filesystem to tell us
312 * that the inode is no longer useful. We just
313 * terminate it with extreme prejudice.
315 void clear_inode(struct inode
*inode
)
318 invalidate_inode_buffers(inode
);
320 BUG_ON(inode
->i_data
.nrpages
);
321 BUG_ON(!(inode
->i_state
& I_FREEING
));
322 BUG_ON(inode
->i_state
& I_CLEAR
);
323 inode_sync_wait(inode
);
324 if (inode
->i_sb
->s_op
->clear_inode
)
325 inode
->i_sb
->s_op
->clear_inode(inode
);
326 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
328 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
330 inode
->i_state
= I_CLEAR
;
332 EXPORT_SYMBOL(clear_inode
);
335 * dispose_list - dispose of the contents of a local list
336 * @head: the head of the list to free
338 * Dispose-list gets a local list with local inodes in it, so it doesn't
339 * need to worry about list corruption and SMP locks.
341 static void dispose_list(struct list_head
*head
)
345 while (!list_empty(head
)) {
348 inode
= list_first_entry(head
, struct inode
, i_list
);
349 list_del(&inode
->i_list
);
351 if (inode
->i_data
.nrpages
)
352 truncate_inode_pages(&inode
->i_data
, 0);
355 spin_lock(&inode_lock
);
356 hlist_del_init(&inode
->i_hash
);
357 list_del_init(&inode
->i_sb_list
);
358 spin_unlock(&inode_lock
);
360 wake_up_inode(inode
);
361 destroy_inode(inode
);
364 spin_lock(&inode_lock
);
365 inodes_stat
.nr_inodes
-= nr_disposed
;
366 spin_unlock(&inode_lock
);
370 * Invalidate all inodes for a device.
372 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
374 struct list_head
*next
;
375 int busy
= 0, count
= 0;
379 struct list_head
*tmp
= next
;
383 * We can reschedule here without worrying about the list's
384 * consistency because the per-sb list of inodes must not
385 * change during umount anymore, and because iprune_sem keeps
386 * shrink_icache_memory() away.
388 cond_resched_lock(&inode_lock
);
393 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
394 if (inode
->i_state
& I_NEW
)
396 invalidate_inode_buffers(inode
);
397 if (!atomic_read(&inode
->i_count
)) {
398 list_move(&inode
->i_list
, dispose
);
399 WARN_ON(inode
->i_state
& I_NEW
);
400 inode
->i_state
|= I_FREEING
;
406 /* only unused inodes may be cached with i_count zero */
407 inodes_stat
.nr_unused
-= count
;
412 * invalidate_inodes - discard the inodes on a device
415 * Discard all of the inodes for a given superblock. If the discard
416 * fails because there are busy inodes then a non zero value is returned.
417 * If the discard is successful all the inodes have been discarded.
419 int invalidate_inodes(struct super_block
*sb
)
422 LIST_HEAD(throw_away
);
424 down_write(&iprune_sem
);
425 spin_lock(&inode_lock
);
426 inotify_unmount_inodes(&sb
->s_inodes
);
427 fsnotify_unmount_inodes(&sb
->s_inodes
);
428 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
429 spin_unlock(&inode_lock
);
431 dispose_list(&throw_away
);
432 up_write(&iprune_sem
);
436 EXPORT_SYMBOL(invalidate_inodes
);
438 static int can_unuse(struct inode
*inode
)
442 if (inode_has_buffers(inode
))
444 if (atomic_read(&inode
->i_count
))
446 if (inode
->i_data
.nrpages
)
452 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
453 * a temporary list and then are freed outside inode_lock by dispose_list().
455 * Any inodes which are pinned purely because of attached pagecache have their
456 * pagecache removed. We expect the final iput() on that inode to add it to
457 * the front of the inode_unused list. So look for it there and if the
458 * inode is still freeable, proceed. The right inode is found 99.9% of the
459 * time in testing on a 4-way.
461 * If the inode has metadata buffers attached to mapping->private_list then
462 * try to remove them.
464 static void prune_icache(int nr_to_scan
)
469 unsigned long reap
= 0;
471 down_read(&iprune_sem
);
472 spin_lock(&inode_lock
);
473 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
476 if (list_empty(&inode_unused
))
479 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
481 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
482 list_move(&inode
->i_list
, &inode_unused
);
485 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
487 spin_unlock(&inode_lock
);
488 if (remove_inode_buffers(inode
))
489 reap
+= invalidate_mapping_pages(&inode
->i_data
,
492 spin_lock(&inode_lock
);
494 if (inode
!= list_entry(inode_unused
.next
,
495 struct inode
, i_list
))
496 continue; /* wrong inode or list_empty */
497 if (!can_unuse(inode
))
500 list_move(&inode
->i_list
, &freeable
);
501 WARN_ON(inode
->i_state
& I_NEW
);
502 inode
->i_state
|= I_FREEING
;
505 inodes_stat
.nr_unused
-= nr_pruned
;
506 if (current_is_kswapd())
507 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
509 __count_vm_events(PGINODESTEAL
, reap
);
510 spin_unlock(&inode_lock
);
512 dispose_list(&freeable
);
513 up_read(&iprune_sem
);
517 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
518 * "unused" means that no dentries are referring to the inodes: the files are
519 * not open and the dcache references to those inodes have already been
522 * This function is passed the number of inodes to scan, and it returns the
523 * total number of remaining possibly-reclaimable inodes.
525 static int shrink_icache_memory(int nr
, gfp_t gfp_mask
)
529 * Nasty deadlock avoidance. We may hold various FS locks,
530 * and we don't want to recurse into the FS that called us
531 * in clear_inode() and friends..
533 if (!(gfp_mask
& __GFP_FS
))
537 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
540 static struct shrinker icache_shrinker
= {
541 .shrink
= shrink_icache_memory
,
542 .seeks
= DEFAULT_SEEKS
,
545 static void __wait_on_freeing_inode(struct inode
*inode
);
547 * Called with the inode lock held.
548 * NOTE: we are not increasing the inode-refcount, you must call __iget()
549 * by hand after calling find_inode now! This simplifies iunique and won't
550 * add any additional branch in the common code.
552 static struct inode
*find_inode(struct super_block
*sb
,
553 struct hlist_head
*head
,
554 int (*test
)(struct inode
*, void *),
557 struct hlist_node
*node
;
558 struct inode
*inode
= NULL
;
561 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
562 if (inode
->i_sb
!= sb
)
564 if (!test(inode
, data
))
566 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
567 __wait_on_freeing_inode(inode
);
572 return node
? inode
: NULL
;
576 * find_inode_fast is the fast path version of find_inode, see the comment at
577 * iget_locked for details.
579 static struct inode
*find_inode_fast(struct super_block
*sb
,
580 struct hlist_head
*head
, unsigned long ino
)
582 struct hlist_node
*node
;
583 struct inode
*inode
= NULL
;
586 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
587 if (inode
->i_ino
!= ino
)
589 if (inode
->i_sb
!= sb
)
591 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
592 __wait_on_freeing_inode(inode
);
597 return node
? inode
: NULL
;
600 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
604 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
606 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
607 return tmp
& I_HASHMASK
;
611 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
614 inodes_stat
.nr_inodes
++;
615 list_add(&inode
->i_list
, &inode_in_use
);
616 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
618 hlist_add_head(&inode
->i_hash
, head
);
622 * inode_add_to_lists - add a new inode to relevant lists
623 * @sb: superblock inode belongs to
624 * @inode: inode to mark in use
626 * When an inode is allocated it needs to be accounted for, added to the in use
627 * list, the owning superblock and the inode hash. This needs to be done under
628 * the inode_lock, so export a function to do this rather than the inode lock
629 * itself. We calculate the hash list to add to here so it is all internal
630 * which requires the caller to have already set up the inode number in the
633 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
635 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
637 spin_lock(&inode_lock
);
638 __inode_add_to_lists(sb
, head
, inode
);
639 spin_unlock(&inode_lock
);
641 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
644 * new_inode - obtain an inode
647 * Allocates a new inode for given superblock. The default gfp_mask
648 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
649 * If HIGHMEM pages are unsuitable or it is known that pages allocated
650 * for the page cache are not reclaimable or migratable,
651 * mapping_set_gfp_mask() must be called with suitable flags on the
652 * newly created inode's mapping
655 struct inode
*new_inode(struct super_block
*sb
)
658 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
659 * error if st_ino won't fit in target struct field. Use 32bit counter
660 * here to attempt to avoid that.
662 static unsigned int last_ino
;
665 spin_lock_prefetch(&inode_lock
);
667 inode
= alloc_inode(sb
);
669 spin_lock(&inode_lock
);
670 __inode_add_to_lists(sb
, NULL
, inode
);
671 inode
->i_ino
= ++last_ino
;
673 spin_unlock(&inode_lock
);
677 EXPORT_SYMBOL(new_inode
);
679 void unlock_new_inode(struct inode
*inode
)
681 #ifdef CONFIG_DEBUG_LOCK_ALLOC
682 if (inode
->i_mode
& S_IFDIR
) {
683 struct file_system_type
*type
= inode
->i_sb
->s_type
;
685 /* Set new key only if filesystem hasn't already changed it */
686 if (!lockdep_match_class(&inode
->i_mutex
,
687 &type
->i_mutex_key
)) {
689 * ensure nobody is actually holding i_mutex
691 mutex_destroy(&inode
->i_mutex
);
692 mutex_init(&inode
->i_mutex
);
693 lockdep_set_class(&inode
->i_mutex
,
694 &type
->i_mutex_dir_key
);
699 * This is special! We do not need the spinlock when clearing I_NEW,
700 * because we're guaranteed that nobody else tries to do anything about
701 * the state of the inode when it is locked, as we just created it (so
702 * there can be no old holders that haven't tested I_NEW).
703 * However we must emit the memory barrier so that other CPUs reliably
704 * see the clearing of I_NEW after the other inode initialisation has
708 WARN_ON(!(inode
->i_state
& I_NEW
));
709 inode
->i_state
&= ~I_NEW
;
710 wake_up_inode(inode
);
712 EXPORT_SYMBOL(unlock_new_inode
);
715 * This is called without the inode lock held.. Be careful.
717 * We no longer cache the sb_flags in i_flags - see fs.h
718 * -- rmk@arm.uk.linux.org
720 static struct inode
*get_new_inode(struct super_block
*sb
,
721 struct hlist_head
*head
,
722 int (*test
)(struct inode
*, void *),
723 int (*set
)(struct inode
*, void *),
728 inode
= alloc_inode(sb
);
732 spin_lock(&inode_lock
);
733 /* We released the lock, so.. */
734 old
= find_inode(sb
, head
, test
, data
);
736 if (set(inode
, data
))
739 __inode_add_to_lists(sb
, head
, inode
);
740 inode
->i_state
= I_NEW
;
741 spin_unlock(&inode_lock
);
743 /* Return the locked inode with I_NEW set, the
744 * caller is responsible for filling in the contents
750 * Uhhuh, somebody else created the same inode under
751 * us. Use the old inode instead of the one we just
755 spin_unlock(&inode_lock
);
756 destroy_inode(inode
);
758 wait_on_inode(inode
);
763 spin_unlock(&inode_lock
);
764 destroy_inode(inode
);
769 * get_new_inode_fast is the fast path version of get_new_inode, see the
770 * comment at iget_locked for details.
772 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
773 struct hlist_head
*head
, unsigned long ino
)
777 inode
= alloc_inode(sb
);
781 spin_lock(&inode_lock
);
782 /* We released the lock, so.. */
783 old
= find_inode_fast(sb
, head
, ino
);
786 __inode_add_to_lists(sb
, head
, inode
);
787 inode
->i_state
= I_NEW
;
788 spin_unlock(&inode_lock
);
790 /* Return the locked inode with I_NEW set, the
791 * caller is responsible for filling in the contents
797 * Uhhuh, somebody else created the same inode under
798 * us. Use the old inode instead of the one we just
802 spin_unlock(&inode_lock
);
803 destroy_inode(inode
);
805 wait_on_inode(inode
);
811 * iunique - get a unique inode number
813 * @max_reserved: highest reserved inode number
815 * Obtain an inode number that is unique on the system for a given
816 * superblock. This is used by file systems that have no natural
817 * permanent inode numbering system. An inode number is returned that
818 * is higher than the reserved limit but unique.
821 * With a large number of inodes live on the file system this function
822 * currently becomes quite slow.
824 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
827 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
828 * error if st_ino won't fit in target struct field. Use 32bit counter
829 * here to attempt to avoid that.
831 static unsigned int counter
;
833 struct hlist_head
*head
;
836 spin_lock(&inode_lock
);
838 if (counter
<= max_reserved
)
839 counter
= max_reserved
+ 1;
841 head
= inode_hashtable
+ hash(sb
, res
);
842 inode
= find_inode_fast(sb
, head
, res
);
843 } while (inode
!= NULL
);
844 spin_unlock(&inode_lock
);
848 EXPORT_SYMBOL(iunique
);
850 struct inode
*igrab(struct inode
*inode
)
852 spin_lock(&inode_lock
);
853 if (!(inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)))
857 * Handle the case where s_op->clear_inode is not been
858 * called yet, and somebody is calling igrab
859 * while the inode is getting freed.
862 spin_unlock(&inode_lock
);
865 EXPORT_SYMBOL(igrab
);
868 * ifind - internal function, you want ilookup5() or iget5().
869 * @sb: super block of file system to search
870 * @head: the head of the list to search
871 * @test: callback used for comparisons between inodes
872 * @data: opaque data pointer to pass to @test
873 * @wait: if true wait for the inode to be unlocked, if false do not
875 * ifind() searches for the inode specified by @data in the inode
876 * cache. This is a generalized version of ifind_fast() for file systems where
877 * the inode number is not sufficient for unique identification of an inode.
879 * If the inode is in the cache, the inode is returned with an incremented
882 * Otherwise NULL is returned.
884 * Note, @test is called with the inode_lock held, so can't sleep.
886 static struct inode
*ifind(struct super_block
*sb
,
887 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
888 void *data
, const int wait
)
892 spin_lock(&inode_lock
);
893 inode
= find_inode(sb
, head
, test
, data
);
896 spin_unlock(&inode_lock
);
898 wait_on_inode(inode
);
901 spin_unlock(&inode_lock
);
906 * ifind_fast - internal function, you want ilookup() or iget().
907 * @sb: super block of file system to search
908 * @head: head of the list to search
909 * @ino: inode number to search for
911 * ifind_fast() searches for the inode @ino in the inode cache. This is for
912 * file systems where the inode number is sufficient for unique identification
915 * If the inode is in the cache, the inode is returned with an incremented
918 * Otherwise NULL is returned.
920 static struct inode
*ifind_fast(struct super_block
*sb
,
921 struct hlist_head
*head
, unsigned long ino
)
925 spin_lock(&inode_lock
);
926 inode
= find_inode_fast(sb
, head
, ino
);
929 spin_unlock(&inode_lock
);
930 wait_on_inode(inode
);
933 spin_unlock(&inode_lock
);
938 * ilookup5_nowait - search for an inode in the inode cache
939 * @sb: super block of file system to search
940 * @hashval: hash value (usually inode number) to search for
941 * @test: callback used for comparisons between inodes
942 * @data: opaque data pointer to pass to @test
944 * ilookup5() uses ifind() to search for the inode specified by @hashval and
945 * @data in the inode cache. This is a generalized version of ilookup() for
946 * file systems where the inode number is not sufficient for unique
947 * identification of an inode.
949 * If the inode is in the cache, the inode is returned with an incremented
950 * reference count. Note, the inode lock is not waited upon so you have to be
951 * very careful what you do with the returned inode. You probably should be
952 * using ilookup5() instead.
954 * Otherwise NULL is returned.
956 * Note, @test is called with the inode_lock held, so can't sleep.
958 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
959 int (*test
)(struct inode
*, void *), void *data
)
961 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
963 return ifind(sb
, head
, test
, data
, 0);
965 EXPORT_SYMBOL(ilookup5_nowait
);
968 * ilookup5 - search for an inode in the inode cache
969 * @sb: super block of file system to search
970 * @hashval: hash value (usually inode number) to search for
971 * @test: callback used for comparisons between inodes
972 * @data: opaque data pointer to pass to @test
974 * ilookup5() uses ifind() to search for the inode specified by @hashval and
975 * @data in the inode cache. This is a generalized version of ilookup() for
976 * file systems where the inode number is not sufficient for unique
977 * identification of an inode.
979 * If the inode is in the cache, the inode lock is waited upon and the inode is
980 * returned with an incremented reference count.
982 * Otherwise NULL is returned.
984 * Note, @test is called with the inode_lock held, so can't sleep.
986 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
987 int (*test
)(struct inode
*, void *), void *data
)
989 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
991 return ifind(sb
, head
, test
, data
, 1);
993 EXPORT_SYMBOL(ilookup5
);
996 * ilookup - search for an inode in the inode cache
997 * @sb: super block of file system to search
998 * @ino: inode number to search for
1000 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1001 * This is for file systems where the inode number is sufficient for unique
1002 * identification of an inode.
1004 * If the inode is in the cache, the inode is returned with an incremented
1007 * Otherwise NULL is returned.
1009 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1011 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1013 return ifind_fast(sb
, head
, ino
);
1015 EXPORT_SYMBOL(ilookup
);
1018 * iget5_locked - obtain an inode from a mounted file system
1019 * @sb: super block of file system
1020 * @hashval: hash value (usually inode number) to get
1021 * @test: callback used for comparisons between inodes
1022 * @set: callback used to initialize a new struct inode
1023 * @data: opaque data pointer to pass to @test and @set
1025 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1026 * and @data in the inode cache and if present it is returned with an increased
1027 * reference count. This is a generalized version of iget_locked() for file
1028 * systems where the inode number is not sufficient for unique identification
1031 * If the inode is not in cache, get_new_inode() is called to allocate a new
1032 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1033 * file system gets to fill it in before unlocking it via unlock_new_inode().
1035 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1037 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1038 int (*test
)(struct inode
*, void *),
1039 int (*set
)(struct inode
*, void *), void *data
)
1041 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1042 struct inode
*inode
;
1044 inode
= ifind(sb
, head
, test
, data
, 1);
1048 * get_new_inode() will do the right thing, re-trying the search
1049 * in case it had to block at any point.
1051 return get_new_inode(sb
, head
, test
, set
, data
);
1053 EXPORT_SYMBOL(iget5_locked
);
1056 * iget_locked - obtain an inode from a mounted file system
1057 * @sb: super block of file system
1058 * @ino: inode number to get
1060 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1061 * the inode cache and if present it is returned with an increased reference
1062 * count. This is for file systems where the inode number is sufficient for
1063 * unique identification of an inode.
1065 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1066 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1067 * The file system gets to fill it in before unlocking it via
1068 * unlock_new_inode().
1070 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1072 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1073 struct inode
*inode
;
1075 inode
= ifind_fast(sb
, head
, ino
);
1079 * get_new_inode_fast() will do the right thing, re-trying the search
1080 * in case it had to block at any point.
1082 return get_new_inode_fast(sb
, head
, ino
);
1084 EXPORT_SYMBOL(iget_locked
);
1086 int insert_inode_locked(struct inode
*inode
)
1088 struct super_block
*sb
= inode
->i_sb
;
1089 ino_t ino
= inode
->i_ino
;
1090 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1092 inode
->i_state
|= I_NEW
;
1094 struct hlist_node
*node
;
1095 struct inode
*old
= NULL
;
1096 spin_lock(&inode_lock
);
1097 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1098 if (old
->i_ino
!= ino
)
1100 if (old
->i_sb
!= sb
)
1102 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1106 if (likely(!node
)) {
1107 hlist_add_head(&inode
->i_hash
, head
);
1108 spin_unlock(&inode_lock
);
1112 spin_unlock(&inode_lock
);
1114 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1121 EXPORT_SYMBOL(insert_inode_locked
);
1123 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1124 int (*test
)(struct inode
*, void *), void *data
)
1126 struct super_block
*sb
= inode
->i_sb
;
1127 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1129 inode
->i_state
|= I_NEW
;
1132 struct hlist_node
*node
;
1133 struct inode
*old
= NULL
;
1135 spin_lock(&inode_lock
);
1136 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1137 if (old
->i_sb
!= sb
)
1139 if (!test(old
, data
))
1141 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1145 if (likely(!node
)) {
1146 hlist_add_head(&inode
->i_hash
, head
);
1147 spin_unlock(&inode_lock
);
1151 spin_unlock(&inode_lock
);
1153 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1160 EXPORT_SYMBOL(insert_inode_locked4
);
1163 * __insert_inode_hash - hash an inode
1164 * @inode: unhashed inode
1165 * @hashval: unsigned long value used to locate this object in the
1168 * Add an inode to the inode hash for this superblock.
1170 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1172 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1173 spin_lock(&inode_lock
);
1174 hlist_add_head(&inode
->i_hash
, head
);
1175 spin_unlock(&inode_lock
);
1177 EXPORT_SYMBOL(__insert_inode_hash
);
1180 * remove_inode_hash - remove an inode from the hash
1181 * @inode: inode to unhash
1183 * Remove an inode from the superblock.
1185 void remove_inode_hash(struct inode
*inode
)
1187 spin_lock(&inode_lock
);
1188 hlist_del_init(&inode
->i_hash
);
1189 spin_unlock(&inode_lock
);
1191 EXPORT_SYMBOL(remove_inode_hash
);
1194 * Tell the filesystem that this inode is no longer of any interest and should
1195 * be completely destroyed.
1197 * We leave the inode in the inode hash table until *after* the filesystem's
1198 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1199 * instigate) will always find up-to-date information either in the hash or on
1202 * I_FREEING is set so that no-one will take a new reference to the inode while
1203 * it is being deleted.
1205 void generic_delete_inode(struct inode
*inode
)
1207 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1209 list_del_init(&inode
->i_list
);
1210 list_del_init(&inode
->i_sb_list
);
1211 WARN_ON(inode
->i_state
& I_NEW
);
1212 inode
->i_state
|= I_FREEING
;
1213 inodes_stat
.nr_inodes
--;
1214 spin_unlock(&inode_lock
);
1216 security_inode_delete(inode
);
1218 if (op
->delete_inode
) {
1219 void (*delete)(struct inode
*) = op
->delete_inode
;
1220 /* Filesystems implementing their own
1221 * s_op->delete_inode are required to call
1222 * truncate_inode_pages and clear_inode()
1226 truncate_inode_pages(&inode
->i_data
, 0);
1229 spin_lock(&inode_lock
);
1230 hlist_del_init(&inode
->i_hash
);
1231 spin_unlock(&inode_lock
);
1232 wake_up_inode(inode
);
1233 BUG_ON(inode
->i_state
!= I_CLEAR
);
1234 destroy_inode(inode
);
1236 EXPORT_SYMBOL(generic_delete_inode
);
1239 * generic_detach_inode - remove inode from inode lists
1240 * @inode: inode to remove
1242 * Remove inode from inode lists, write it if it's dirty. This is just an
1243 * internal VFS helper exported for hugetlbfs. Do not use!
1245 * Returns 1 if inode should be completely destroyed.
1247 int generic_detach_inode(struct inode
*inode
)
1249 struct super_block
*sb
= inode
->i_sb
;
1251 if (!hlist_unhashed(&inode
->i_hash
)) {
1252 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1253 list_move(&inode
->i_list
, &inode_unused
);
1254 inodes_stat
.nr_unused
++;
1255 if (sb
->s_flags
& MS_ACTIVE
) {
1256 spin_unlock(&inode_lock
);
1259 WARN_ON(inode
->i_state
& I_NEW
);
1260 inode
->i_state
|= I_WILL_FREE
;
1261 spin_unlock(&inode_lock
);
1262 write_inode_now(inode
, 1);
1263 spin_lock(&inode_lock
);
1264 WARN_ON(inode
->i_state
& I_NEW
);
1265 inode
->i_state
&= ~I_WILL_FREE
;
1266 inodes_stat
.nr_unused
--;
1267 hlist_del_init(&inode
->i_hash
);
1269 list_del_init(&inode
->i_list
);
1270 list_del_init(&inode
->i_sb_list
);
1271 WARN_ON(inode
->i_state
& I_NEW
);
1272 inode
->i_state
|= I_FREEING
;
1273 inodes_stat
.nr_inodes
--;
1274 spin_unlock(&inode_lock
);
1277 EXPORT_SYMBOL_GPL(generic_detach_inode
);
1279 static void generic_forget_inode(struct inode
*inode
)
1281 if (!generic_detach_inode(inode
))
1283 if (inode
->i_data
.nrpages
)
1284 truncate_inode_pages(&inode
->i_data
, 0);
1286 wake_up_inode(inode
);
1287 destroy_inode(inode
);
1291 * Normal UNIX filesystem behaviour: delete the
1292 * inode when the usage count drops to zero, and
1295 void generic_drop_inode(struct inode
*inode
)
1297 if (!inode
->i_nlink
)
1298 generic_delete_inode(inode
);
1300 generic_forget_inode(inode
);
1302 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1305 * Called when we're dropping the last reference
1308 * Call the FS "drop()" function, defaulting to
1309 * the legacy UNIX filesystem behaviour..
1311 * NOTE! NOTE! NOTE! We're called with the inode lock
1312 * held, and the drop function is supposed to release
1315 static inline void iput_final(struct inode
*inode
)
1317 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1318 void (*drop
)(struct inode
*) = generic_drop_inode
;
1320 if (op
&& op
->drop_inode
)
1321 drop
= op
->drop_inode
;
1326 * iput - put an inode
1327 * @inode: inode to put
1329 * Puts an inode, dropping its usage count. If the inode use count hits
1330 * zero, the inode is then freed and may also be destroyed.
1332 * Consequently, iput() can sleep.
1334 void iput(struct inode
*inode
)
1337 BUG_ON(inode
->i_state
== I_CLEAR
);
1339 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1343 EXPORT_SYMBOL(iput
);
1346 * bmap - find a block number in a file
1347 * @inode: inode of file
1348 * @block: block to find
1350 * Returns the block number on the device holding the inode that
1351 * is the disk block number for the block of the file requested.
1352 * That is, asked for block 4 of inode 1 the function will return the
1353 * disk block relative to the disk start that holds that block of the
1356 sector_t
bmap(struct inode
*inode
, sector_t block
)
1359 if (inode
->i_mapping
->a_ops
->bmap
)
1360 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1363 EXPORT_SYMBOL(bmap
);
1366 * With relative atime, only update atime if the previous atime is
1367 * earlier than either the ctime or mtime or if at least a day has
1368 * passed since the last atime update.
1370 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1371 struct timespec now
)
1374 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1377 * Is mtime younger than atime? If yes, update atime:
1379 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1382 * Is ctime younger than atime? If yes, update atime:
1384 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1388 * Is the previous atime value older than a day? If yes,
1391 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1394 * Good, we can skip the atime update:
1400 * touch_atime - update the access time
1401 * @mnt: mount the inode is accessed on
1402 * @dentry: dentry accessed
1404 * Update the accessed time on an inode and mark it for writeback.
1405 * This function automatically handles read only file systems and media,
1406 * as well as the "noatime" flag and inode specific "noatime" markers.
1408 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1410 struct inode
*inode
= dentry
->d_inode
;
1411 struct timespec now
;
1413 if (inode
->i_flags
& S_NOATIME
)
1415 if (IS_NOATIME(inode
))
1417 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1420 if (mnt
->mnt_flags
& MNT_NOATIME
)
1422 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1425 now
= current_fs_time(inode
->i_sb
);
1427 if (!relatime_need_update(mnt
, inode
, now
))
1430 if (timespec_equal(&inode
->i_atime
, &now
))
1433 if (mnt_want_write(mnt
))
1436 inode
->i_atime
= now
;
1437 mark_inode_dirty_sync(inode
);
1438 mnt_drop_write(mnt
);
1440 EXPORT_SYMBOL(touch_atime
);
1443 * file_update_time - update mtime and ctime time
1444 * @file: file accessed
1446 * Update the mtime and ctime members of an inode and mark the inode
1447 * for writeback. Note that this function is meant exclusively for
1448 * usage in the file write path of filesystems, and filesystems may
1449 * choose to explicitly ignore update via this function with the
1450 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1451 * timestamps are handled by the server.
1454 void file_update_time(struct file
*file
)
1456 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1457 struct timespec now
;
1458 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1460 /* First try to exhaust all avenues to not sync */
1461 if (IS_NOCMTIME(inode
))
1464 now
= current_fs_time(inode
->i_sb
);
1465 if (!timespec_equal(&inode
->i_mtime
, &now
))
1468 if (!timespec_equal(&inode
->i_ctime
, &now
))
1471 if (IS_I_VERSION(inode
))
1472 sync_it
|= S_VERSION
;
1477 /* Finally allowed to write? Takes lock. */
1478 if (mnt_want_write_file(file
))
1481 /* Only change inode inside the lock region */
1482 if (sync_it
& S_VERSION
)
1483 inode_inc_iversion(inode
);
1484 if (sync_it
& S_CTIME
)
1485 inode
->i_ctime
= now
;
1486 if (sync_it
& S_MTIME
)
1487 inode
->i_mtime
= now
;
1488 mark_inode_dirty_sync(inode
);
1489 mnt_drop_write(file
->f_path
.mnt
);
1491 EXPORT_SYMBOL(file_update_time
);
1493 int inode_needs_sync(struct inode
*inode
)
1497 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1501 EXPORT_SYMBOL(inode_needs_sync
);
1503 int inode_wait(void *word
)
1508 EXPORT_SYMBOL(inode_wait
);
1511 * If we try to find an inode in the inode hash while it is being
1512 * deleted, we have to wait until the filesystem completes its
1513 * deletion before reporting that it isn't found. This function waits
1514 * until the deletion _might_ have completed. Callers are responsible
1515 * to recheck inode state.
1517 * It doesn't matter if I_NEW is not set initially, a call to
1518 * wake_up_inode() after removing from the hash list will DTRT.
1520 * This is called with inode_lock held.
1522 static void __wait_on_freeing_inode(struct inode
*inode
)
1524 wait_queue_head_t
*wq
;
1525 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1526 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1527 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1528 spin_unlock(&inode_lock
);
1530 finish_wait(wq
, &wait
.wait
);
1531 spin_lock(&inode_lock
);
1534 static __initdata
unsigned long ihash_entries
;
1535 static int __init
set_ihash_entries(char *str
)
1539 ihash_entries
= simple_strtoul(str
, &str
, 0);
1542 __setup("ihash_entries=", set_ihash_entries
);
1545 * Initialize the waitqueues and inode hash table.
1547 void __init
inode_init_early(void)
1551 /* If hashes are distributed across NUMA nodes, defer
1552 * hash allocation until vmalloc space is available.
1558 alloc_large_system_hash("Inode-cache",
1559 sizeof(struct hlist_head
),
1567 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1568 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1571 void __init
inode_init(void)
1575 /* inode slab cache */
1576 inode_cachep
= kmem_cache_create("inode_cache",
1577 sizeof(struct inode
),
1579 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1582 register_shrinker(&icache_shrinker
);
1584 /* Hash may have been set up in inode_init_early */
1589 alloc_large_system_hash("Inode-cache",
1590 sizeof(struct hlist_head
),
1598 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1599 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1602 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1604 inode
->i_mode
= mode
;
1605 if (S_ISCHR(mode
)) {
1606 inode
->i_fop
= &def_chr_fops
;
1607 inode
->i_rdev
= rdev
;
1608 } else if (S_ISBLK(mode
)) {
1609 inode
->i_fop
= &def_blk_fops
;
1610 inode
->i_rdev
= rdev
;
1611 } else if (S_ISFIFO(mode
))
1612 inode
->i_fop
= &def_fifo_fops
;
1613 else if (S_ISSOCK(mode
))
1614 inode
->i_fop
= &bad_sock_fops
;
1616 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1617 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1620 EXPORT_SYMBOL(init_special_inode
);