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
));
249 * These are initializations that only need to be done
250 * once, because the fields are idempotent across use
251 * of the inode, so let the slab aware of that.
253 void inode_init_once(struct inode
*inode
)
255 memset(inode
, 0, sizeof(*inode
));
256 INIT_HLIST_NODE(&inode
->i_hash
);
257 INIT_LIST_HEAD(&inode
->i_dentry
);
258 INIT_LIST_HEAD(&inode
->i_devices
);
259 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
260 spin_lock_init(&inode
->i_data
.tree_lock
);
261 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
262 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
263 spin_lock_init(&inode
->i_data
.private_lock
);
264 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
265 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
266 i_size_ordered_init(inode
);
267 #ifdef CONFIG_INOTIFY
268 INIT_LIST_HEAD(&inode
->inotify_watches
);
269 mutex_init(&inode
->inotify_mutex
);
271 #ifdef CONFIG_FSNOTIFY
272 INIT_HLIST_HEAD(&inode
->i_fsnotify_mark_entries
);
275 EXPORT_SYMBOL(inode_init_once
);
277 static void init_once(void *foo
)
279 struct inode
*inode
= (struct inode
*) foo
;
281 inode_init_once(inode
);
285 * inode_lock must be held
287 void __iget(struct inode
*inode
)
289 if (atomic_inc_return(&inode
->i_count
) != 1)
292 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
293 list_move(&inode
->i_list
, &inode_in_use
);
294 inodes_stat
.nr_unused
--;
298 * clear_inode - clear an inode
299 * @inode: inode to clear
301 * This is called by the filesystem to tell us
302 * that the inode is no longer useful. We just
303 * terminate it with extreme prejudice.
305 void clear_inode(struct inode
*inode
)
308 invalidate_inode_buffers(inode
);
310 BUG_ON(inode
->i_data
.nrpages
);
311 BUG_ON(!(inode
->i_state
& I_FREEING
));
312 BUG_ON(inode
->i_state
& I_CLEAR
);
313 inode_sync_wait(inode
);
314 if (inode
->i_sb
->s_op
->clear_inode
)
315 inode
->i_sb
->s_op
->clear_inode(inode
);
316 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
318 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
320 inode
->i_state
= I_CLEAR
;
322 EXPORT_SYMBOL(clear_inode
);
325 * dispose_list - dispose of the contents of a local list
326 * @head: the head of the list to free
328 * Dispose-list gets a local list with local inodes in it, so it doesn't
329 * need to worry about list corruption and SMP locks.
331 static void dispose_list(struct list_head
*head
)
335 while (!list_empty(head
)) {
338 inode
= list_first_entry(head
, struct inode
, i_list
);
339 list_del(&inode
->i_list
);
341 if (inode
->i_data
.nrpages
)
342 truncate_inode_pages(&inode
->i_data
, 0);
345 spin_lock(&inode_lock
);
346 hlist_del_init(&inode
->i_hash
);
347 list_del_init(&inode
->i_sb_list
);
348 spin_unlock(&inode_lock
);
350 wake_up_inode(inode
);
351 destroy_inode(inode
);
354 spin_lock(&inode_lock
);
355 inodes_stat
.nr_inodes
-= nr_disposed
;
356 spin_unlock(&inode_lock
);
360 * Invalidate all inodes for a device.
362 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
364 struct list_head
*next
;
365 int busy
= 0, count
= 0;
369 struct list_head
*tmp
= next
;
373 * We can reschedule here without worrying about the list's
374 * consistency because the per-sb list of inodes must not
375 * change during umount anymore, and because iprune_sem keeps
376 * shrink_icache_memory() away.
378 cond_resched_lock(&inode_lock
);
383 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
384 if (inode
->i_state
& I_NEW
)
386 invalidate_inode_buffers(inode
);
387 if (!atomic_read(&inode
->i_count
)) {
388 list_move(&inode
->i_list
, dispose
);
389 WARN_ON(inode
->i_state
& I_NEW
);
390 inode
->i_state
|= I_FREEING
;
396 /* only unused inodes may be cached with i_count zero */
397 inodes_stat
.nr_unused
-= count
;
402 * invalidate_inodes - discard the inodes on a device
405 * Discard all of the inodes for a given superblock. If the discard
406 * fails because there are busy inodes then a non zero value is returned.
407 * If the discard is successful all the inodes have been discarded.
409 int invalidate_inodes(struct super_block
*sb
)
412 LIST_HEAD(throw_away
);
414 down_write(&iprune_sem
);
415 spin_lock(&inode_lock
);
416 inotify_unmount_inodes(&sb
->s_inodes
);
417 fsnotify_unmount_inodes(&sb
->s_inodes
);
418 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
419 spin_unlock(&inode_lock
);
421 dispose_list(&throw_away
);
422 up_write(&iprune_sem
);
426 EXPORT_SYMBOL(invalidate_inodes
);
428 static int can_unuse(struct inode
*inode
)
432 if (inode_has_buffers(inode
))
434 if (atomic_read(&inode
->i_count
))
436 if (inode
->i_data
.nrpages
)
442 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
443 * a temporary list and then are freed outside inode_lock by dispose_list().
445 * Any inodes which are pinned purely because of attached pagecache have their
446 * pagecache removed. We expect the final iput() on that inode to add it to
447 * the front of the inode_unused list. So look for it there and if the
448 * inode is still freeable, proceed. The right inode is found 99.9% of the
449 * time in testing on a 4-way.
451 * If the inode has metadata buffers attached to mapping->private_list then
452 * try to remove them.
454 static void prune_icache(int nr_to_scan
)
459 unsigned long reap
= 0;
461 down_read(&iprune_sem
);
462 spin_lock(&inode_lock
);
463 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
466 if (list_empty(&inode_unused
))
469 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
471 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
472 list_move(&inode
->i_list
, &inode_unused
);
475 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
477 spin_unlock(&inode_lock
);
478 if (remove_inode_buffers(inode
))
479 reap
+= invalidate_mapping_pages(&inode
->i_data
,
482 spin_lock(&inode_lock
);
484 if (inode
!= list_entry(inode_unused
.next
,
485 struct inode
, i_list
))
486 continue; /* wrong inode or list_empty */
487 if (!can_unuse(inode
))
490 list_move(&inode
->i_list
, &freeable
);
491 WARN_ON(inode
->i_state
& I_NEW
);
492 inode
->i_state
|= I_FREEING
;
495 inodes_stat
.nr_unused
-= nr_pruned
;
496 if (current_is_kswapd())
497 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
499 __count_vm_events(PGINODESTEAL
, reap
);
500 spin_unlock(&inode_lock
);
502 dispose_list(&freeable
);
503 up_read(&iprune_sem
);
507 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
508 * "unused" means that no dentries are referring to the inodes: the files are
509 * not open and the dcache references to those inodes have already been
512 * This function is passed the number of inodes to scan, and it returns the
513 * total number of remaining possibly-reclaimable inodes.
515 static int shrink_icache_memory(int nr
, gfp_t gfp_mask
)
519 * Nasty deadlock avoidance. We may hold various FS locks,
520 * and we don't want to recurse into the FS that called us
521 * in clear_inode() and friends..
523 if (!(gfp_mask
& __GFP_FS
))
527 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
530 static struct shrinker icache_shrinker
= {
531 .shrink
= shrink_icache_memory
,
532 .seeks
= DEFAULT_SEEKS
,
535 static void __wait_on_freeing_inode(struct inode
*inode
);
537 * Called with the inode lock held.
538 * NOTE: we are not increasing the inode-refcount, you must call __iget()
539 * by hand after calling find_inode now! This simplifies iunique and won't
540 * add any additional branch in the common code.
542 static struct inode
*find_inode(struct super_block
*sb
,
543 struct hlist_head
*head
,
544 int (*test
)(struct inode
*, void *),
547 struct hlist_node
*node
;
548 struct inode
*inode
= NULL
;
551 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
552 if (inode
->i_sb
!= sb
)
554 if (!test(inode
, data
))
556 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
557 __wait_on_freeing_inode(inode
);
562 return node
? inode
: NULL
;
566 * find_inode_fast is the fast path version of find_inode, see the comment at
567 * iget_locked for details.
569 static struct inode
*find_inode_fast(struct super_block
*sb
,
570 struct hlist_head
*head
, unsigned long ino
)
572 struct hlist_node
*node
;
573 struct inode
*inode
= NULL
;
576 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
577 if (inode
->i_ino
!= ino
)
579 if (inode
->i_sb
!= sb
)
581 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
582 __wait_on_freeing_inode(inode
);
587 return node
? inode
: NULL
;
590 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
594 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
596 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
597 return tmp
& I_HASHMASK
;
601 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
604 inodes_stat
.nr_inodes
++;
605 list_add(&inode
->i_list
, &inode_in_use
);
606 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
608 hlist_add_head(&inode
->i_hash
, head
);
612 * inode_add_to_lists - add a new inode to relevant lists
613 * @sb: superblock inode belongs to
614 * @inode: inode to mark in use
616 * When an inode is allocated it needs to be accounted for, added to the in use
617 * list, the owning superblock and the inode hash. This needs to be done under
618 * the inode_lock, so export a function to do this rather than the inode lock
619 * itself. We calculate the hash list to add to here so it is all internal
620 * which requires the caller to have already set up the inode number in the
623 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
625 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
627 spin_lock(&inode_lock
);
628 __inode_add_to_lists(sb
, head
, inode
);
629 spin_unlock(&inode_lock
);
631 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
634 * new_inode - obtain an inode
637 * Allocates a new inode for given superblock. The default gfp_mask
638 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
639 * If HIGHMEM pages are unsuitable or it is known that pages allocated
640 * for the page cache are not reclaimable or migratable,
641 * mapping_set_gfp_mask() must be called with suitable flags on the
642 * newly created inode's mapping
645 struct inode
*new_inode(struct super_block
*sb
)
648 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
649 * error if st_ino won't fit in target struct field. Use 32bit counter
650 * here to attempt to avoid that.
652 static unsigned int last_ino
;
655 spin_lock_prefetch(&inode_lock
);
657 inode
= alloc_inode(sb
);
659 spin_lock(&inode_lock
);
660 __inode_add_to_lists(sb
, NULL
, inode
);
661 inode
->i_ino
= ++last_ino
;
663 spin_unlock(&inode_lock
);
667 EXPORT_SYMBOL(new_inode
);
669 void unlock_new_inode(struct inode
*inode
)
671 #ifdef CONFIG_DEBUG_LOCK_ALLOC
672 if (inode
->i_mode
& S_IFDIR
) {
673 struct file_system_type
*type
= inode
->i_sb
->s_type
;
675 /* Set new key only if filesystem hasn't already changed it */
676 if (!lockdep_match_class(&inode
->i_mutex
,
677 &type
->i_mutex_key
)) {
679 * ensure nobody is actually holding i_mutex
681 mutex_destroy(&inode
->i_mutex
);
682 mutex_init(&inode
->i_mutex
);
683 lockdep_set_class(&inode
->i_mutex
,
684 &type
->i_mutex_dir_key
);
689 * This is special! We do not need the spinlock when clearing I_NEW,
690 * because we're guaranteed that nobody else tries to do anything about
691 * the state of the inode when it is locked, as we just created it (so
692 * there can be no old holders that haven't tested I_NEW).
693 * However we must emit the memory barrier so that other CPUs reliably
694 * see the clearing of I_NEW after the other inode initialisation has
698 WARN_ON(!(inode
->i_state
& I_NEW
));
699 inode
->i_state
&= ~I_NEW
;
700 wake_up_inode(inode
);
702 EXPORT_SYMBOL(unlock_new_inode
);
705 * This is called without the inode lock held.. Be careful.
707 * We no longer cache the sb_flags in i_flags - see fs.h
708 * -- rmk@arm.uk.linux.org
710 static struct inode
*get_new_inode(struct super_block
*sb
,
711 struct hlist_head
*head
,
712 int (*test
)(struct inode
*, void *),
713 int (*set
)(struct inode
*, void *),
718 inode
= alloc_inode(sb
);
722 spin_lock(&inode_lock
);
723 /* We released the lock, so.. */
724 old
= find_inode(sb
, head
, test
, data
);
726 if (set(inode
, data
))
729 __inode_add_to_lists(sb
, head
, inode
);
730 inode
->i_state
= I_NEW
;
731 spin_unlock(&inode_lock
);
733 /* Return the locked inode with I_NEW set, the
734 * caller is responsible for filling in the contents
740 * Uhhuh, somebody else created the same inode under
741 * us. Use the old inode instead of the one we just
745 spin_unlock(&inode_lock
);
746 destroy_inode(inode
);
748 wait_on_inode(inode
);
753 spin_unlock(&inode_lock
);
754 destroy_inode(inode
);
759 * get_new_inode_fast is the fast path version of get_new_inode, see the
760 * comment at iget_locked for details.
762 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
763 struct hlist_head
*head
, unsigned long ino
)
767 inode
= alloc_inode(sb
);
771 spin_lock(&inode_lock
);
772 /* We released the lock, so.. */
773 old
= find_inode_fast(sb
, head
, ino
);
776 __inode_add_to_lists(sb
, head
, inode
);
777 inode
->i_state
= I_NEW
;
778 spin_unlock(&inode_lock
);
780 /* Return the locked inode with I_NEW set, the
781 * caller is responsible for filling in the contents
787 * Uhhuh, somebody else created the same inode under
788 * us. Use the old inode instead of the one we just
792 spin_unlock(&inode_lock
);
793 destroy_inode(inode
);
795 wait_on_inode(inode
);
801 * iunique - get a unique inode number
803 * @max_reserved: highest reserved inode number
805 * Obtain an inode number that is unique on the system for a given
806 * superblock. This is used by file systems that have no natural
807 * permanent inode numbering system. An inode number is returned that
808 * is higher than the reserved limit but unique.
811 * With a large number of inodes live on the file system this function
812 * currently becomes quite slow.
814 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
817 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
818 * error if st_ino won't fit in target struct field. Use 32bit counter
819 * here to attempt to avoid that.
821 static unsigned int counter
;
823 struct hlist_head
*head
;
826 spin_lock(&inode_lock
);
828 if (counter
<= max_reserved
)
829 counter
= max_reserved
+ 1;
831 head
= inode_hashtable
+ hash(sb
, res
);
832 inode
= find_inode_fast(sb
, head
, res
);
833 } while (inode
!= NULL
);
834 spin_unlock(&inode_lock
);
838 EXPORT_SYMBOL(iunique
);
840 struct inode
*igrab(struct inode
*inode
)
842 spin_lock(&inode_lock
);
843 if (!(inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)))
847 * Handle the case where s_op->clear_inode is not been
848 * called yet, and somebody is calling igrab
849 * while the inode is getting freed.
852 spin_unlock(&inode_lock
);
855 EXPORT_SYMBOL(igrab
);
858 * ifind - internal function, you want ilookup5() or iget5().
859 * @sb: super block of file system to search
860 * @head: the head of the list to search
861 * @test: callback used for comparisons between inodes
862 * @data: opaque data pointer to pass to @test
863 * @wait: if true wait for the inode to be unlocked, if false do not
865 * ifind() searches for the inode specified by @data in the inode
866 * cache. This is a generalized version of ifind_fast() for file systems where
867 * the inode number is not sufficient for unique identification of an inode.
869 * If the inode is in the cache, the inode is returned with an incremented
872 * Otherwise NULL is returned.
874 * Note, @test is called with the inode_lock held, so can't sleep.
876 static struct inode
*ifind(struct super_block
*sb
,
877 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
878 void *data
, const int wait
)
882 spin_lock(&inode_lock
);
883 inode
= find_inode(sb
, head
, test
, data
);
886 spin_unlock(&inode_lock
);
888 wait_on_inode(inode
);
891 spin_unlock(&inode_lock
);
896 * ifind_fast - internal function, you want ilookup() or iget().
897 * @sb: super block of file system to search
898 * @head: head of the list to search
899 * @ino: inode number to search for
901 * ifind_fast() searches for the inode @ino in the inode cache. This is for
902 * file systems where the inode number is sufficient for unique identification
905 * If the inode is in the cache, the inode is returned with an incremented
908 * Otherwise NULL is returned.
910 static struct inode
*ifind_fast(struct super_block
*sb
,
911 struct hlist_head
*head
, unsigned long ino
)
915 spin_lock(&inode_lock
);
916 inode
= find_inode_fast(sb
, head
, ino
);
919 spin_unlock(&inode_lock
);
920 wait_on_inode(inode
);
923 spin_unlock(&inode_lock
);
928 * ilookup5_nowait - search for an inode in the inode cache
929 * @sb: super block of file system to search
930 * @hashval: hash value (usually inode number) to search for
931 * @test: callback used for comparisons between inodes
932 * @data: opaque data pointer to pass to @test
934 * ilookup5() uses ifind() to search for the inode specified by @hashval and
935 * @data in the inode cache. This is a generalized version of ilookup() for
936 * file systems where the inode number is not sufficient for unique
937 * identification of an inode.
939 * If the inode is in the cache, the inode is returned with an incremented
940 * reference count. Note, the inode lock is not waited upon so you have to be
941 * very careful what you do with the returned inode. You probably should be
942 * using ilookup5() instead.
944 * Otherwise NULL is returned.
946 * Note, @test is called with the inode_lock held, so can't sleep.
948 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
949 int (*test
)(struct inode
*, void *), void *data
)
951 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
953 return ifind(sb
, head
, test
, data
, 0);
955 EXPORT_SYMBOL(ilookup5_nowait
);
958 * ilookup5 - search for an inode in the inode cache
959 * @sb: super block of file system to search
960 * @hashval: hash value (usually inode number) to search for
961 * @test: callback used for comparisons between inodes
962 * @data: opaque data pointer to pass to @test
964 * ilookup5() uses ifind() to search for the inode specified by @hashval and
965 * @data in the inode cache. This is a generalized version of ilookup() for
966 * file systems where the inode number is not sufficient for unique
967 * identification of an inode.
969 * If the inode is in the cache, the inode lock is waited upon and the inode is
970 * returned with an incremented reference count.
972 * Otherwise NULL is returned.
974 * Note, @test is called with the inode_lock held, so can't sleep.
976 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
977 int (*test
)(struct inode
*, void *), void *data
)
979 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
981 return ifind(sb
, head
, test
, data
, 1);
983 EXPORT_SYMBOL(ilookup5
);
986 * ilookup - search for an inode in the inode cache
987 * @sb: super block of file system to search
988 * @ino: inode number to search for
990 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
991 * This is for file systems where the inode number is sufficient for unique
992 * identification of an inode.
994 * If the inode is in the cache, the inode is returned with an incremented
997 * Otherwise NULL is returned.
999 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1001 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1003 return ifind_fast(sb
, head
, ino
);
1005 EXPORT_SYMBOL(ilookup
);
1008 * iget5_locked - obtain an inode from a mounted file system
1009 * @sb: super block of file system
1010 * @hashval: hash value (usually inode number) to get
1011 * @test: callback used for comparisons between inodes
1012 * @set: callback used to initialize a new struct inode
1013 * @data: opaque data pointer to pass to @test and @set
1015 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1016 * and @data in the inode cache and if present it is returned with an increased
1017 * reference count. This is a generalized version of iget_locked() for file
1018 * systems where the inode number is not sufficient for unique identification
1021 * If the inode is not in cache, get_new_inode() is called to allocate a new
1022 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1023 * file system gets to fill it in before unlocking it via unlock_new_inode().
1025 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1027 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1028 int (*test
)(struct inode
*, void *),
1029 int (*set
)(struct inode
*, void *), void *data
)
1031 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1032 struct inode
*inode
;
1034 inode
= ifind(sb
, head
, test
, data
, 1);
1038 * get_new_inode() will do the right thing, re-trying the search
1039 * in case it had to block at any point.
1041 return get_new_inode(sb
, head
, test
, set
, data
);
1043 EXPORT_SYMBOL(iget5_locked
);
1046 * iget_locked - obtain an inode from a mounted file system
1047 * @sb: super block of file system
1048 * @ino: inode number to get
1050 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1051 * the inode cache and if present it is returned with an increased reference
1052 * count. This is for file systems where the inode number is sufficient for
1053 * unique identification of an inode.
1055 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1056 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1057 * The file system gets to fill it in before unlocking it via
1058 * unlock_new_inode().
1060 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1062 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1063 struct inode
*inode
;
1065 inode
= ifind_fast(sb
, head
, ino
);
1069 * get_new_inode_fast() will do the right thing, re-trying the search
1070 * in case it had to block at any point.
1072 return get_new_inode_fast(sb
, head
, ino
);
1074 EXPORT_SYMBOL(iget_locked
);
1076 int insert_inode_locked(struct inode
*inode
)
1078 struct super_block
*sb
= inode
->i_sb
;
1079 ino_t ino
= inode
->i_ino
;
1080 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1082 inode
->i_state
|= I_NEW
;
1084 struct hlist_node
*node
;
1085 struct inode
*old
= NULL
;
1086 spin_lock(&inode_lock
);
1087 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1088 if (old
->i_ino
!= ino
)
1090 if (old
->i_sb
!= sb
)
1092 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1096 if (likely(!node
)) {
1097 hlist_add_head(&inode
->i_hash
, head
);
1098 spin_unlock(&inode_lock
);
1102 spin_unlock(&inode_lock
);
1104 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1111 EXPORT_SYMBOL(insert_inode_locked
);
1113 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1114 int (*test
)(struct inode
*, void *), void *data
)
1116 struct super_block
*sb
= inode
->i_sb
;
1117 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1119 inode
->i_state
|= I_NEW
;
1122 struct hlist_node
*node
;
1123 struct inode
*old
= NULL
;
1125 spin_lock(&inode_lock
);
1126 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1127 if (old
->i_sb
!= sb
)
1129 if (!test(old
, data
))
1131 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1135 if (likely(!node
)) {
1136 hlist_add_head(&inode
->i_hash
, head
);
1137 spin_unlock(&inode_lock
);
1141 spin_unlock(&inode_lock
);
1143 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1150 EXPORT_SYMBOL(insert_inode_locked4
);
1153 * __insert_inode_hash - hash an inode
1154 * @inode: unhashed inode
1155 * @hashval: unsigned long value used to locate this object in the
1158 * Add an inode to the inode hash for this superblock.
1160 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1162 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1163 spin_lock(&inode_lock
);
1164 hlist_add_head(&inode
->i_hash
, head
);
1165 spin_unlock(&inode_lock
);
1167 EXPORT_SYMBOL(__insert_inode_hash
);
1170 * remove_inode_hash - remove an inode from the hash
1171 * @inode: inode to unhash
1173 * Remove an inode from the superblock.
1175 void remove_inode_hash(struct inode
*inode
)
1177 spin_lock(&inode_lock
);
1178 hlist_del_init(&inode
->i_hash
);
1179 spin_unlock(&inode_lock
);
1181 EXPORT_SYMBOL(remove_inode_hash
);
1184 * Tell the filesystem that this inode is no longer of any interest and should
1185 * be completely destroyed.
1187 * We leave the inode in the inode hash table until *after* the filesystem's
1188 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1189 * instigate) will always find up-to-date information either in the hash or on
1192 * I_FREEING is set so that no-one will take a new reference to the inode while
1193 * it is being deleted.
1195 void generic_delete_inode(struct inode
*inode
)
1197 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1199 list_del_init(&inode
->i_list
);
1200 list_del_init(&inode
->i_sb_list
);
1201 WARN_ON(inode
->i_state
& I_NEW
);
1202 inode
->i_state
|= I_FREEING
;
1203 inodes_stat
.nr_inodes
--;
1204 spin_unlock(&inode_lock
);
1206 if (op
->delete_inode
) {
1207 void (*delete)(struct inode
*) = op
->delete_inode
;
1208 /* Filesystems implementing their own
1209 * s_op->delete_inode are required to call
1210 * truncate_inode_pages and clear_inode()
1214 truncate_inode_pages(&inode
->i_data
, 0);
1217 spin_lock(&inode_lock
);
1218 hlist_del_init(&inode
->i_hash
);
1219 spin_unlock(&inode_lock
);
1220 wake_up_inode(inode
);
1221 BUG_ON(inode
->i_state
!= I_CLEAR
);
1222 destroy_inode(inode
);
1224 EXPORT_SYMBOL(generic_delete_inode
);
1227 * generic_detach_inode - remove inode from inode lists
1228 * @inode: inode to remove
1230 * Remove inode from inode lists, write it if it's dirty. This is just an
1231 * internal VFS helper exported for hugetlbfs. Do not use!
1233 * Returns 1 if inode should be completely destroyed.
1235 int generic_detach_inode(struct inode
*inode
)
1237 struct super_block
*sb
= inode
->i_sb
;
1239 if (!hlist_unhashed(&inode
->i_hash
)) {
1240 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1241 list_move(&inode
->i_list
, &inode_unused
);
1242 inodes_stat
.nr_unused
++;
1243 if (sb
->s_flags
& MS_ACTIVE
) {
1244 spin_unlock(&inode_lock
);
1247 WARN_ON(inode
->i_state
& I_NEW
);
1248 inode
->i_state
|= I_WILL_FREE
;
1249 spin_unlock(&inode_lock
);
1250 write_inode_now(inode
, 1);
1251 spin_lock(&inode_lock
);
1252 WARN_ON(inode
->i_state
& I_NEW
);
1253 inode
->i_state
&= ~I_WILL_FREE
;
1254 inodes_stat
.nr_unused
--;
1255 hlist_del_init(&inode
->i_hash
);
1257 list_del_init(&inode
->i_list
);
1258 list_del_init(&inode
->i_sb_list
);
1259 WARN_ON(inode
->i_state
& I_NEW
);
1260 inode
->i_state
|= I_FREEING
;
1261 inodes_stat
.nr_inodes
--;
1262 spin_unlock(&inode_lock
);
1265 EXPORT_SYMBOL_GPL(generic_detach_inode
);
1267 static void generic_forget_inode(struct inode
*inode
)
1269 if (!generic_detach_inode(inode
))
1271 if (inode
->i_data
.nrpages
)
1272 truncate_inode_pages(&inode
->i_data
, 0);
1274 wake_up_inode(inode
);
1275 destroy_inode(inode
);
1279 * Normal UNIX filesystem behaviour: delete the
1280 * inode when the usage count drops to zero, and
1283 void generic_drop_inode(struct inode
*inode
)
1285 if (!inode
->i_nlink
)
1286 generic_delete_inode(inode
);
1288 generic_forget_inode(inode
);
1290 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1293 * Called when we're dropping the last reference
1296 * Call the FS "drop()" function, defaulting to
1297 * the legacy UNIX filesystem behaviour..
1299 * NOTE! NOTE! NOTE! We're called with the inode lock
1300 * held, and the drop function is supposed to release
1303 static inline void iput_final(struct inode
*inode
)
1305 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1306 void (*drop
)(struct inode
*) = generic_drop_inode
;
1308 if (op
&& op
->drop_inode
)
1309 drop
= op
->drop_inode
;
1314 * iput - put an inode
1315 * @inode: inode to put
1317 * Puts an inode, dropping its usage count. If the inode use count hits
1318 * zero, the inode is then freed and may also be destroyed.
1320 * Consequently, iput() can sleep.
1322 void iput(struct inode
*inode
)
1325 BUG_ON(inode
->i_state
== I_CLEAR
);
1327 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1331 EXPORT_SYMBOL(iput
);
1334 * bmap - find a block number in a file
1335 * @inode: inode of file
1336 * @block: block to find
1338 * Returns the block number on the device holding the inode that
1339 * is the disk block number for the block of the file requested.
1340 * That is, asked for block 4 of inode 1 the function will return the
1341 * disk block relative to the disk start that holds that block of the
1344 sector_t
bmap(struct inode
*inode
, sector_t block
)
1347 if (inode
->i_mapping
->a_ops
->bmap
)
1348 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1351 EXPORT_SYMBOL(bmap
);
1354 * With relative atime, only update atime if the previous atime is
1355 * earlier than either the ctime or mtime or if at least a day has
1356 * passed since the last atime update.
1358 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1359 struct timespec now
)
1362 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1365 * Is mtime younger than atime? If yes, update atime:
1367 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1370 * Is ctime younger than atime? If yes, update atime:
1372 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1376 * Is the previous atime value older than a day? If yes,
1379 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1382 * Good, we can skip the atime update:
1388 * touch_atime - update the access time
1389 * @mnt: mount the inode is accessed on
1390 * @dentry: dentry accessed
1392 * Update the accessed time on an inode and mark it for writeback.
1393 * This function automatically handles read only file systems and media,
1394 * as well as the "noatime" flag and inode specific "noatime" markers.
1396 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1398 struct inode
*inode
= dentry
->d_inode
;
1399 struct timespec now
;
1401 if (inode
->i_flags
& S_NOATIME
)
1403 if (IS_NOATIME(inode
))
1405 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1408 if (mnt
->mnt_flags
& MNT_NOATIME
)
1410 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1413 now
= current_fs_time(inode
->i_sb
);
1415 if (!relatime_need_update(mnt
, inode
, now
))
1418 if (timespec_equal(&inode
->i_atime
, &now
))
1421 if (mnt_want_write(mnt
))
1424 inode
->i_atime
= now
;
1425 mark_inode_dirty_sync(inode
);
1426 mnt_drop_write(mnt
);
1428 EXPORT_SYMBOL(touch_atime
);
1431 * file_update_time - update mtime and ctime time
1432 * @file: file accessed
1434 * Update the mtime and ctime members of an inode and mark the inode
1435 * for writeback. Note that this function is meant exclusively for
1436 * usage in the file write path of filesystems, and filesystems may
1437 * choose to explicitly ignore update via this function with the
1438 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1439 * timestamps are handled by the server.
1442 void file_update_time(struct file
*file
)
1444 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1445 struct timespec now
;
1446 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1448 /* First try to exhaust all avenues to not sync */
1449 if (IS_NOCMTIME(inode
))
1452 now
= current_fs_time(inode
->i_sb
);
1453 if (!timespec_equal(&inode
->i_mtime
, &now
))
1456 if (!timespec_equal(&inode
->i_ctime
, &now
))
1459 if (IS_I_VERSION(inode
))
1460 sync_it
|= S_VERSION
;
1465 /* Finally allowed to write? Takes lock. */
1466 if (mnt_want_write_file(file
))
1469 /* Only change inode inside the lock region */
1470 if (sync_it
& S_VERSION
)
1471 inode_inc_iversion(inode
);
1472 if (sync_it
& S_CTIME
)
1473 inode
->i_ctime
= now
;
1474 if (sync_it
& S_MTIME
)
1475 inode
->i_mtime
= now
;
1476 mark_inode_dirty_sync(inode
);
1477 mnt_drop_write(file
->f_path
.mnt
);
1479 EXPORT_SYMBOL(file_update_time
);
1481 int inode_needs_sync(struct inode
*inode
)
1485 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1489 EXPORT_SYMBOL(inode_needs_sync
);
1491 int inode_wait(void *word
)
1496 EXPORT_SYMBOL(inode_wait
);
1499 * If we try to find an inode in the inode hash while it is being
1500 * deleted, we have to wait until the filesystem completes its
1501 * deletion before reporting that it isn't found. This function waits
1502 * until the deletion _might_ have completed. Callers are responsible
1503 * to recheck inode state.
1505 * It doesn't matter if I_NEW is not set initially, a call to
1506 * wake_up_inode() after removing from the hash list will DTRT.
1508 * This is called with inode_lock held.
1510 static void __wait_on_freeing_inode(struct inode
*inode
)
1512 wait_queue_head_t
*wq
;
1513 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1514 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1515 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1516 spin_unlock(&inode_lock
);
1518 finish_wait(wq
, &wait
.wait
);
1519 spin_lock(&inode_lock
);
1522 static __initdata
unsigned long ihash_entries
;
1523 static int __init
set_ihash_entries(char *str
)
1527 ihash_entries
= simple_strtoul(str
, &str
, 0);
1530 __setup("ihash_entries=", set_ihash_entries
);
1533 * Initialize the waitqueues and inode hash table.
1535 void __init
inode_init_early(void)
1539 /* If hashes are distributed across NUMA nodes, defer
1540 * hash allocation until vmalloc space is available.
1546 alloc_large_system_hash("Inode-cache",
1547 sizeof(struct hlist_head
),
1555 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1556 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1559 void __init
inode_init(void)
1563 /* inode slab cache */
1564 inode_cachep
= kmem_cache_create("inode_cache",
1565 sizeof(struct inode
),
1567 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1570 register_shrinker(&icache_shrinker
);
1572 /* Hash may have been set up in inode_init_early */
1577 alloc_large_system_hash("Inode-cache",
1578 sizeof(struct hlist_head
),
1586 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1587 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1590 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1592 inode
->i_mode
= mode
;
1593 if (S_ISCHR(mode
)) {
1594 inode
->i_fop
= &def_chr_fops
;
1595 inode
->i_rdev
= rdev
;
1596 } else if (S_ISBLK(mode
)) {
1597 inode
->i_fop
= &def_blk_fops
;
1598 inode
->i_rdev
= rdev
;
1599 } else if (S_ISFIFO(mode
))
1600 inode
->i_fop
= &def_fifo_fops
;
1601 else if (S_ISSOCK(mode
))
1602 inode
->i_fop
= &bad_sock_fops
;
1604 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1605 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1608 EXPORT_SYMBOL(init_special_inode
);