4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/quotaops.h>
12 #include <linux/slab.h>
13 #include <linux/writeback.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/wait.h>
17 #include <linux/hash.h>
18 #include <linux/swap.h>
19 #include <linux/security.h>
20 #include <linux/ima.h>
21 #include <linux/pagemap.h>
22 #include <linux/cdev.h>
23 #include <linux/bootmem.h>
24 #include <linux/inotify.h>
25 #include <linux/mount.h>
26 #include <linux/async.h>
29 * This is needed for the following functions:
31 * - invalidate_inode_buffers
34 * FIXME: remove all knowledge of the buffer layer from this file
36 #include <linux/buffer_head.h>
39 * New inode.c implementation.
41 * This implementation has the basic premise of trying
42 * to be extremely low-overhead and SMP-safe, yet be
43 * simple enough to be "obviously correct".
48 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
50 /* #define INODE_PARANOIA 1 */
51 /* #define INODE_DEBUG 1 */
54 * Inode lookup is no longer as critical as it used to be:
55 * most of the lookups are going to be through the dcache.
57 #define I_HASHBITS i_hash_shift
58 #define I_HASHMASK i_hash_mask
60 static unsigned int i_hash_mask __read_mostly
;
61 static unsigned int i_hash_shift __read_mostly
;
64 * Each inode can be on two separate lists. One is
65 * the hash list of the inode, used for lookups. The
66 * other linked list is the "type" list:
67 * "in_use" - valid inode, i_count > 0, i_nlink > 0
68 * "dirty" - as "in_use" but also dirty
69 * "unused" - valid inode, i_count = 0
71 * A "dirty" list is maintained for each super block,
72 * allowing for low-overhead inode sync() operations.
75 LIST_HEAD(inode_in_use
);
76 LIST_HEAD(inode_unused
);
77 static struct hlist_head
*inode_hashtable __read_mostly
;
80 * A simple spinlock to protect the list manipulations.
82 * NOTE! You also have to own the lock if you change
83 * the i_state of an inode while it is in use..
85 DEFINE_SPINLOCK(inode_lock
);
88 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
89 * icache shrinking path, and the umount path. Without this exclusion,
90 * by the time prune_icache calls iput for the inode whose pages it has
91 * been invalidating, or by the time it calls clear_inode & destroy_inode
92 * from its final dispose_list, the struct super_block they refer to
93 * (for inode->i_sb->s_op) may already have been freed and reused.
95 static DEFINE_MUTEX(iprune_mutex
);
98 * Statistics gathering..
100 struct inodes_stat_t inodes_stat
;
102 static struct kmem_cache
*inode_cachep __read_mostly
;
104 static void wake_up_inode(struct inode
*inode
)
107 * Prevent speculative execution through spin_unlock(&inode_lock);
110 wake_up_bit(&inode
->i_state
, __I_LOCK
);
114 * inode_init_always - perform inode structure intialisation
115 * @sb: superblock inode belongs to
116 * @inode: inode to initialise
118 * These are initializations that need to be done on every inode
119 * allocation as the fields are not initialised by slab allocation.
121 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
123 static const struct address_space_operations empty_aops
;
124 static struct inode_operations empty_iops
;
125 static const struct file_operations empty_fops
;
126 struct address_space
*const mapping
= &inode
->i_data
;
129 inode
->i_blkbits
= sb
->s_blocksize_bits
;
131 atomic_set(&inode
->i_count
, 1);
132 inode
->i_op
= &empty_iops
;
133 inode
->i_fop
= &empty_fops
;
137 atomic_set(&inode
->i_writecount
, 0);
141 inode
->i_generation
= 0;
143 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
145 inode
->i_pipe
= NULL
;
146 inode
->i_bdev
= NULL
;
147 inode
->i_cdev
= NULL
;
149 inode
->dirtied_when
= 0;
151 if (security_inode_alloc(inode
))
154 /* allocate and initialize an i_integrity */
155 if (ima_inode_alloc(inode
))
156 goto out_free_security
;
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_backing_dev_info
;
185 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
186 mapping
->backing_dev_info
= bdi
;
188 inode
->i_private
= NULL
;
189 inode
->i_mapping
= mapping
;
194 security_inode_free(inode
);
198 EXPORT_SYMBOL(inode_init_always
);
200 static struct inode
*alloc_inode(struct super_block
*sb
)
204 if (sb
->s_op
->alloc_inode
)
205 inode
= sb
->s_op
->alloc_inode(sb
);
207 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
212 if (unlikely(inode_init_always(sb
, inode
))) {
213 if (inode
->i_sb
->s_op
->destroy_inode
)
214 inode
->i_sb
->s_op
->destroy_inode(inode
);
216 kmem_cache_free(inode_cachep
, inode
);
223 void destroy_inode(struct inode
*inode
)
225 BUG_ON(inode_has_buffers(inode
));
226 ima_inode_free(inode
);
227 security_inode_free(inode
);
228 if (inode
->i_sb
->s_op
->destroy_inode
)
229 inode
->i_sb
->s_op
->destroy_inode(inode
);
231 kmem_cache_free(inode_cachep
, (inode
));
233 EXPORT_SYMBOL(destroy_inode
);
237 * These are initializations that only need to be done
238 * once, because the fields are idempotent across use
239 * of the inode, so let the slab aware of that.
241 void inode_init_once(struct inode
*inode
)
243 memset(inode
, 0, sizeof(*inode
));
244 INIT_HLIST_NODE(&inode
->i_hash
);
245 INIT_LIST_HEAD(&inode
->i_dentry
);
246 INIT_LIST_HEAD(&inode
->i_devices
);
247 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
248 spin_lock_init(&inode
->i_data
.tree_lock
);
249 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
250 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
251 spin_lock_init(&inode
->i_data
.private_lock
);
252 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
253 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
254 i_size_ordered_init(inode
);
255 #ifdef CONFIG_INOTIFY
256 INIT_LIST_HEAD(&inode
->inotify_watches
);
257 mutex_init(&inode
->inotify_mutex
);
260 EXPORT_SYMBOL(inode_init_once
);
262 static void init_once(void *foo
)
264 struct inode
*inode
= (struct inode
*) foo
;
266 inode_init_once(inode
);
270 * inode_lock must be held
272 void __iget(struct inode
*inode
)
274 if (atomic_read(&inode
->i_count
)) {
275 atomic_inc(&inode
->i_count
);
278 atomic_inc(&inode
->i_count
);
279 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
280 list_move(&inode
->i_list
, &inode_in_use
);
281 inodes_stat
.nr_unused
--;
285 * clear_inode - clear an inode
286 * @inode: inode to clear
288 * This is called by the filesystem to tell us
289 * that the inode is no longer useful. We just
290 * terminate it with extreme prejudice.
292 void clear_inode(struct inode
*inode
)
295 invalidate_inode_buffers(inode
);
297 BUG_ON(inode
->i_data
.nrpages
);
298 BUG_ON(!(inode
->i_state
& I_FREEING
));
299 BUG_ON(inode
->i_state
& I_CLEAR
);
300 inode_sync_wait(inode
);
302 if (inode
->i_sb
->s_op
->clear_inode
)
303 inode
->i_sb
->s_op
->clear_inode(inode
);
304 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
306 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
308 inode
->i_state
= I_CLEAR
;
310 EXPORT_SYMBOL(clear_inode
);
313 * dispose_list - dispose of the contents of a local list
314 * @head: the head of the list to free
316 * Dispose-list gets a local list with local inodes in it, so it doesn't
317 * need to worry about list corruption and SMP locks.
319 static void dispose_list(struct list_head
*head
)
323 while (!list_empty(head
)) {
326 inode
= list_first_entry(head
, struct inode
, i_list
);
327 list_del(&inode
->i_list
);
329 if (inode
->i_data
.nrpages
)
330 truncate_inode_pages(&inode
->i_data
, 0);
333 spin_lock(&inode_lock
);
334 hlist_del_init(&inode
->i_hash
);
335 list_del_init(&inode
->i_sb_list
);
336 spin_unlock(&inode_lock
);
338 wake_up_inode(inode
);
339 destroy_inode(inode
);
342 spin_lock(&inode_lock
);
343 inodes_stat
.nr_inodes
-= nr_disposed
;
344 spin_unlock(&inode_lock
);
348 * Invalidate all inodes for a device.
350 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
352 struct list_head
*next
;
353 int busy
= 0, count
= 0;
357 struct list_head
*tmp
= next
;
361 * We can reschedule here without worrying about the list's
362 * consistency because the per-sb list of inodes must not
363 * change during umount anymore, and because iprune_mutex keeps
364 * shrink_icache_memory() away.
366 cond_resched_lock(&inode_lock
);
371 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
372 if (inode
->i_state
& I_NEW
)
374 invalidate_inode_buffers(inode
);
375 if (!atomic_read(&inode
->i_count
)) {
376 list_move(&inode
->i_list
, dispose
);
377 WARN_ON(inode
->i_state
& I_NEW
);
378 inode
->i_state
|= I_FREEING
;
384 /* only unused inodes may be cached with i_count zero */
385 inodes_stat
.nr_unused
-= count
;
390 * invalidate_inodes - discard the inodes on a device
393 * Discard all of the inodes for a given superblock. If the discard
394 * fails because there are busy inodes then a non zero value is returned.
395 * If the discard is successful all the inodes have been discarded.
397 int invalidate_inodes(struct super_block
*sb
)
400 LIST_HEAD(throw_away
);
402 mutex_lock(&iprune_mutex
);
403 spin_lock(&inode_lock
);
404 inotify_unmount_inodes(&sb
->s_inodes
);
405 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
406 spin_unlock(&inode_lock
);
408 dispose_list(&throw_away
);
409 mutex_unlock(&iprune_mutex
);
413 EXPORT_SYMBOL(invalidate_inodes
);
415 static int can_unuse(struct inode
*inode
)
419 if (inode_has_buffers(inode
))
421 if (atomic_read(&inode
->i_count
))
423 if (inode
->i_data
.nrpages
)
429 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
430 * a temporary list and then are freed outside inode_lock by dispose_list().
432 * Any inodes which are pinned purely because of attached pagecache have their
433 * pagecache removed. We expect the final iput() on that inode to add it to
434 * the front of the inode_unused list. So look for it there and if the
435 * inode is still freeable, proceed. The right inode is found 99.9% of the
436 * time in testing on a 4-way.
438 * If the inode has metadata buffers attached to mapping->private_list then
439 * try to remove them.
441 static void prune_icache(int nr_to_scan
)
446 unsigned long reap
= 0;
448 mutex_lock(&iprune_mutex
);
449 spin_lock(&inode_lock
);
450 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
453 if (list_empty(&inode_unused
))
456 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
458 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
459 list_move(&inode
->i_list
, &inode_unused
);
462 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
464 spin_unlock(&inode_lock
);
465 if (remove_inode_buffers(inode
))
466 reap
+= invalidate_mapping_pages(&inode
->i_data
,
469 spin_lock(&inode_lock
);
471 if (inode
!= list_entry(inode_unused
.next
,
472 struct inode
, i_list
))
473 continue; /* wrong inode or list_empty */
474 if (!can_unuse(inode
))
477 list_move(&inode
->i_list
, &freeable
);
478 WARN_ON(inode
->i_state
& I_NEW
);
479 inode
->i_state
|= I_FREEING
;
482 inodes_stat
.nr_unused
-= nr_pruned
;
483 if (current_is_kswapd())
484 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
486 __count_vm_events(PGINODESTEAL
, reap
);
487 spin_unlock(&inode_lock
);
489 dispose_list(&freeable
);
490 mutex_unlock(&iprune_mutex
);
494 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
495 * "unused" means that no dentries are referring to the inodes: the files are
496 * not open and the dcache references to those inodes have already been
499 * This function is passed the number of inodes to scan, and it returns the
500 * total number of remaining possibly-reclaimable inodes.
502 static int shrink_icache_memory(int nr
, gfp_t gfp_mask
)
506 * Nasty deadlock avoidance. We may hold various FS locks,
507 * and we don't want to recurse into the FS that called us
508 * in clear_inode() and friends..
510 if (!(gfp_mask
& __GFP_FS
))
514 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
517 static struct shrinker icache_shrinker
= {
518 .shrink
= shrink_icache_memory
,
519 .seeks
= DEFAULT_SEEKS
,
522 static void __wait_on_freeing_inode(struct inode
*inode
);
524 * Called with the inode lock held.
525 * NOTE: we are not increasing the inode-refcount, you must call __iget()
526 * by hand after calling find_inode now! This simplifies iunique and won't
527 * add any additional branch in the common code.
529 static struct inode
*find_inode(struct super_block
*sb
,
530 struct hlist_head
*head
,
531 int (*test
)(struct inode
*, void *),
534 struct hlist_node
*node
;
535 struct inode
*inode
= NULL
;
538 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
539 if (inode
->i_sb
!= sb
)
541 if (!test(inode
, data
))
543 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
544 __wait_on_freeing_inode(inode
);
549 return node
? inode
: NULL
;
553 * find_inode_fast is the fast path version of find_inode, see the comment at
554 * iget_locked for details.
556 static struct inode
*find_inode_fast(struct super_block
*sb
,
557 struct hlist_head
*head
, unsigned long ino
)
559 struct hlist_node
*node
;
560 struct inode
*inode
= NULL
;
563 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
564 if (inode
->i_ino
!= ino
)
566 if (inode
->i_sb
!= sb
)
568 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
569 __wait_on_freeing_inode(inode
);
574 return node
? inode
: NULL
;
577 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
581 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
583 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
584 return tmp
& I_HASHMASK
;
588 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
591 inodes_stat
.nr_inodes
++;
592 list_add(&inode
->i_list
, &inode_in_use
);
593 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
595 hlist_add_head(&inode
->i_hash
, head
);
599 * inode_add_to_lists - add a new inode to relevant lists
600 * @sb: superblock inode belongs to
601 * @inode: inode to mark in use
603 * When an inode is allocated it needs to be accounted for, added to the in use
604 * list, the owning superblock and the inode hash. This needs to be done under
605 * the inode_lock, so export a function to do this rather than the inode lock
606 * itself. We calculate the hash list to add to here so it is all internal
607 * which requires the caller to have already set up the inode number in the
610 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
612 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
614 spin_lock(&inode_lock
);
615 __inode_add_to_lists(sb
, head
, inode
);
616 spin_unlock(&inode_lock
);
618 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
621 * new_inode - obtain an inode
624 * Allocates a new inode for given superblock. The default gfp_mask
625 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
626 * If HIGHMEM pages are unsuitable or it is known that pages allocated
627 * for the page cache are not reclaimable or migratable,
628 * mapping_set_gfp_mask() must be called with suitable flags on the
629 * newly created inode's mapping
632 struct inode
*new_inode(struct super_block
*sb
)
635 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
636 * error if st_ino won't fit in target struct field. Use 32bit counter
637 * here to attempt to avoid that.
639 static unsigned int last_ino
;
642 spin_lock_prefetch(&inode_lock
);
644 inode
= alloc_inode(sb
);
646 spin_lock(&inode_lock
);
647 __inode_add_to_lists(sb
, NULL
, inode
);
648 inode
->i_ino
= ++last_ino
;
650 spin_unlock(&inode_lock
);
654 EXPORT_SYMBOL(new_inode
);
656 void unlock_new_inode(struct inode
*inode
)
658 #ifdef CONFIG_DEBUG_LOCK_ALLOC
659 if (inode
->i_mode
& S_IFDIR
) {
660 struct file_system_type
*type
= inode
->i_sb
->s_type
;
663 * ensure nobody is actually holding i_mutex
665 mutex_destroy(&inode
->i_mutex
);
666 mutex_init(&inode
->i_mutex
);
667 lockdep_set_class(&inode
->i_mutex
, &type
->i_mutex_dir_key
);
671 * This is special! We do not need the spinlock
672 * when clearing I_LOCK, because we're guaranteed
673 * that nobody else tries to do anything about the
674 * state of the inode when it is locked, as we
675 * just created it (so there can be no old holders
676 * that haven't tested I_LOCK).
678 WARN_ON((inode
->i_state
& (I_LOCK
|I_NEW
)) != (I_LOCK
|I_NEW
));
679 inode
->i_state
&= ~(I_LOCK
|I_NEW
);
680 wake_up_inode(inode
);
682 EXPORT_SYMBOL(unlock_new_inode
);
685 * This is called without the inode lock held.. Be careful.
687 * We no longer cache the sb_flags in i_flags - see fs.h
688 * -- rmk@arm.uk.linux.org
690 static struct inode
*get_new_inode(struct super_block
*sb
,
691 struct hlist_head
*head
,
692 int (*test
)(struct inode
*, void *),
693 int (*set
)(struct inode
*, void *),
698 inode
= alloc_inode(sb
);
702 spin_lock(&inode_lock
);
703 /* We released the lock, so.. */
704 old
= find_inode(sb
, head
, test
, data
);
706 if (set(inode
, data
))
709 __inode_add_to_lists(sb
, head
, inode
);
710 inode
->i_state
= I_LOCK
|I_NEW
;
711 spin_unlock(&inode_lock
);
713 /* Return the locked inode with I_NEW set, the
714 * caller is responsible for filling in the contents
720 * Uhhuh, somebody else created the same inode under
721 * us. Use the old inode instead of the one we just
725 spin_unlock(&inode_lock
);
726 destroy_inode(inode
);
728 wait_on_inode(inode
);
733 spin_unlock(&inode_lock
);
734 destroy_inode(inode
);
739 * get_new_inode_fast is the fast path version of get_new_inode, see the
740 * comment at iget_locked for details.
742 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
743 struct hlist_head
*head
, unsigned long ino
)
747 inode
= alloc_inode(sb
);
751 spin_lock(&inode_lock
);
752 /* We released the lock, so.. */
753 old
= find_inode_fast(sb
, head
, ino
);
756 __inode_add_to_lists(sb
, head
, inode
);
757 inode
->i_state
= I_LOCK
|I_NEW
;
758 spin_unlock(&inode_lock
);
760 /* Return the locked inode with I_NEW set, the
761 * caller is responsible for filling in the contents
767 * Uhhuh, somebody else created the same inode under
768 * us. Use the old inode instead of the one we just
772 spin_unlock(&inode_lock
);
773 destroy_inode(inode
);
775 wait_on_inode(inode
);
781 * iunique - get a unique inode number
783 * @max_reserved: highest reserved inode number
785 * Obtain an inode number that is unique on the system for a given
786 * superblock. This is used by file systems that have no natural
787 * permanent inode numbering system. An inode number is returned that
788 * is higher than the reserved limit but unique.
791 * With a large number of inodes live on the file system this function
792 * currently becomes quite slow.
794 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
797 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
798 * error if st_ino won't fit in target struct field. Use 32bit counter
799 * here to attempt to avoid that.
801 static unsigned int counter
;
803 struct hlist_head
*head
;
806 spin_lock(&inode_lock
);
808 if (counter
<= max_reserved
)
809 counter
= max_reserved
+ 1;
811 head
= inode_hashtable
+ hash(sb
, res
);
812 inode
= find_inode_fast(sb
, head
, res
);
813 } while (inode
!= NULL
);
814 spin_unlock(&inode_lock
);
818 EXPORT_SYMBOL(iunique
);
820 struct inode
*igrab(struct inode
*inode
)
822 spin_lock(&inode_lock
);
823 if (!(inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)))
827 * Handle the case where s_op->clear_inode is not been
828 * called yet, and somebody is calling igrab
829 * while the inode is getting freed.
832 spin_unlock(&inode_lock
);
835 EXPORT_SYMBOL(igrab
);
838 * ifind - internal function, you want ilookup5() or iget5().
839 * @sb: super block of file system to search
840 * @head: the head of the list to search
841 * @test: callback used for comparisons between inodes
842 * @data: opaque data pointer to pass to @test
843 * @wait: if true wait for the inode to be unlocked, if false do not
845 * ifind() searches for the inode specified by @data in the inode
846 * cache. This is a generalized version of ifind_fast() for file systems where
847 * the inode number is not sufficient for unique identification of an inode.
849 * If the inode is in the cache, the inode is returned with an incremented
852 * Otherwise NULL is returned.
854 * Note, @test is called with the inode_lock held, so can't sleep.
856 static struct inode
*ifind(struct super_block
*sb
,
857 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
858 void *data
, const int wait
)
862 spin_lock(&inode_lock
);
863 inode
= find_inode(sb
, head
, test
, data
);
866 spin_unlock(&inode_lock
);
868 wait_on_inode(inode
);
871 spin_unlock(&inode_lock
);
876 * ifind_fast - internal function, you want ilookup() or iget().
877 * @sb: super block of file system to search
878 * @head: head of the list to search
879 * @ino: inode number to search for
881 * ifind_fast() searches for the inode @ino in the inode cache. This is for
882 * file systems where the inode number is sufficient for unique identification
885 * If the inode is in the cache, the inode is returned with an incremented
888 * Otherwise NULL is returned.
890 static struct inode
*ifind_fast(struct super_block
*sb
,
891 struct hlist_head
*head
, unsigned long ino
)
895 spin_lock(&inode_lock
);
896 inode
= find_inode_fast(sb
, head
, ino
);
899 spin_unlock(&inode_lock
);
900 wait_on_inode(inode
);
903 spin_unlock(&inode_lock
);
908 * ilookup5_nowait - search for an inode in the inode cache
909 * @sb: super block of file system to search
910 * @hashval: hash value (usually inode number) to search for
911 * @test: callback used for comparisons between inodes
912 * @data: opaque data pointer to pass to @test
914 * ilookup5() uses ifind() to search for the inode specified by @hashval and
915 * @data in the inode cache. This is a generalized version of ilookup() for
916 * file systems where the inode number is not sufficient for unique
917 * identification of an inode.
919 * If the inode is in the cache, the inode is returned with an incremented
920 * reference count. Note, the inode lock is not waited upon so you have to be
921 * very careful what you do with the returned inode. You probably should be
922 * using ilookup5() instead.
924 * Otherwise NULL is returned.
926 * Note, @test is called with the inode_lock held, so can't sleep.
928 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
929 int (*test
)(struct inode
*, void *), void *data
)
931 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
933 return ifind(sb
, head
, test
, data
, 0);
935 EXPORT_SYMBOL(ilookup5_nowait
);
938 * ilookup5 - 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 lock is waited upon and the inode is
950 * returned with an incremented reference count.
952 * Otherwise NULL is returned.
954 * Note, @test is called with the inode_lock held, so can't sleep.
956 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
957 int (*test
)(struct inode
*, void *), void *data
)
959 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
961 return ifind(sb
, head
, test
, data
, 1);
963 EXPORT_SYMBOL(ilookup5
);
966 * ilookup - search for an inode in the inode cache
967 * @sb: super block of file system to search
968 * @ino: inode number to search for
970 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
971 * This is for file systems where the inode number is sufficient for unique
972 * identification of an inode.
974 * If the inode is in the cache, the inode is returned with an incremented
977 * Otherwise NULL is returned.
979 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
981 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
983 return ifind_fast(sb
, head
, ino
);
985 EXPORT_SYMBOL(ilookup
);
988 * iget5_locked - obtain an inode from a mounted file system
989 * @sb: super block of file system
990 * @hashval: hash value (usually inode number) to get
991 * @test: callback used for comparisons between inodes
992 * @set: callback used to initialize a new struct inode
993 * @data: opaque data pointer to pass to @test and @set
995 * iget5_locked() uses ifind() to search for the inode specified by @hashval
996 * and @data in the inode cache and if present it is returned with an increased
997 * reference count. This is a generalized version of iget_locked() for file
998 * systems where the inode number is not sufficient for unique identification
1001 * If the inode is not in cache, get_new_inode() is called to allocate a new
1002 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1003 * file system gets to fill it in before unlocking it via unlock_new_inode().
1005 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1007 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1008 int (*test
)(struct inode
*, void *),
1009 int (*set
)(struct inode
*, void *), void *data
)
1011 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1012 struct inode
*inode
;
1014 inode
= ifind(sb
, head
, test
, data
, 1);
1018 * get_new_inode() will do the right thing, re-trying the search
1019 * in case it had to block at any point.
1021 return get_new_inode(sb
, head
, test
, set
, data
);
1023 EXPORT_SYMBOL(iget5_locked
);
1026 * iget_locked - obtain an inode from a mounted file system
1027 * @sb: super block of file system
1028 * @ino: inode number to get
1030 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1031 * the inode cache and if present it is returned with an increased reference
1032 * count. This is for file systems where the inode number is sufficient for
1033 * unique identification of an inode.
1035 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1036 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1037 * The file system gets to fill it in before unlocking it via
1038 * unlock_new_inode().
1040 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1042 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1043 struct inode
*inode
;
1045 inode
= ifind_fast(sb
, head
, ino
);
1049 * get_new_inode_fast() will do the right thing, re-trying the search
1050 * in case it had to block at any point.
1052 return get_new_inode_fast(sb
, head
, ino
);
1054 EXPORT_SYMBOL(iget_locked
);
1056 int insert_inode_locked(struct inode
*inode
)
1058 struct super_block
*sb
= inode
->i_sb
;
1059 ino_t ino
= inode
->i_ino
;
1060 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1062 inode
->i_state
|= I_LOCK
|I_NEW
;
1064 struct hlist_node
*node
;
1065 struct inode
*old
= NULL
;
1066 spin_lock(&inode_lock
);
1067 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1068 if (old
->i_ino
!= ino
)
1070 if (old
->i_sb
!= sb
)
1072 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1076 if (likely(!node
)) {
1077 hlist_add_head(&inode
->i_hash
, head
);
1078 spin_unlock(&inode_lock
);
1082 spin_unlock(&inode_lock
);
1084 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1091 EXPORT_SYMBOL(insert_inode_locked
);
1093 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1094 int (*test
)(struct inode
*, void *), void *data
)
1096 struct super_block
*sb
= inode
->i_sb
;
1097 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1099 inode
->i_state
|= I_LOCK
|I_NEW
;
1102 struct hlist_node
*node
;
1103 struct inode
*old
= NULL
;
1105 spin_lock(&inode_lock
);
1106 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1107 if (old
->i_sb
!= sb
)
1109 if (!test(old
, data
))
1111 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1115 if (likely(!node
)) {
1116 hlist_add_head(&inode
->i_hash
, head
);
1117 spin_unlock(&inode_lock
);
1121 spin_unlock(&inode_lock
);
1123 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1130 EXPORT_SYMBOL(insert_inode_locked4
);
1133 * __insert_inode_hash - hash an inode
1134 * @inode: unhashed inode
1135 * @hashval: unsigned long value used to locate this object in the
1138 * Add an inode to the inode hash for this superblock.
1140 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1142 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1143 spin_lock(&inode_lock
);
1144 hlist_add_head(&inode
->i_hash
, head
);
1145 spin_unlock(&inode_lock
);
1147 EXPORT_SYMBOL(__insert_inode_hash
);
1150 * remove_inode_hash - remove an inode from the hash
1151 * @inode: inode to unhash
1153 * Remove an inode from the superblock.
1155 void remove_inode_hash(struct inode
*inode
)
1157 spin_lock(&inode_lock
);
1158 hlist_del_init(&inode
->i_hash
);
1159 spin_unlock(&inode_lock
);
1161 EXPORT_SYMBOL(remove_inode_hash
);
1164 * Tell the filesystem that this inode is no longer of any interest and should
1165 * be completely destroyed.
1167 * We leave the inode in the inode hash table until *after* the filesystem's
1168 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1169 * instigate) will always find up-to-date information either in the hash or on
1172 * I_FREEING is set so that no-one will take a new reference to the inode while
1173 * it is being deleted.
1175 void generic_delete_inode(struct inode
*inode
)
1177 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1179 list_del_init(&inode
->i_list
);
1180 list_del_init(&inode
->i_sb_list
);
1181 WARN_ON(inode
->i_state
& I_NEW
);
1182 inode
->i_state
|= I_FREEING
;
1183 inodes_stat
.nr_inodes
--;
1184 spin_unlock(&inode_lock
);
1186 security_inode_delete(inode
);
1188 if (op
->delete_inode
) {
1189 void (*delete)(struct inode
*) = op
->delete_inode
;
1190 if (!is_bad_inode(inode
))
1192 /* Filesystems implementing their own
1193 * s_op->delete_inode are required to call
1194 * truncate_inode_pages and clear_inode()
1198 truncate_inode_pages(&inode
->i_data
, 0);
1201 spin_lock(&inode_lock
);
1202 hlist_del_init(&inode
->i_hash
);
1203 spin_unlock(&inode_lock
);
1204 wake_up_inode(inode
);
1205 BUG_ON(inode
->i_state
!= I_CLEAR
);
1206 destroy_inode(inode
);
1208 EXPORT_SYMBOL(generic_delete_inode
);
1210 static void generic_forget_inode(struct inode
*inode
)
1212 struct super_block
*sb
= inode
->i_sb
;
1214 if (!hlist_unhashed(&inode
->i_hash
)) {
1215 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1216 list_move(&inode
->i_list
, &inode_unused
);
1217 inodes_stat
.nr_unused
++;
1218 if (sb
->s_flags
& MS_ACTIVE
) {
1219 spin_unlock(&inode_lock
);
1222 WARN_ON(inode
->i_state
& I_NEW
);
1223 inode
->i_state
|= I_WILL_FREE
;
1224 spin_unlock(&inode_lock
);
1225 write_inode_now(inode
, 1);
1226 spin_lock(&inode_lock
);
1227 WARN_ON(inode
->i_state
& I_NEW
);
1228 inode
->i_state
&= ~I_WILL_FREE
;
1229 inodes_stat
.nr_unused
--;
1230 hlist_del_init(&inode
->i_hash
);
1232 list_del_init(&inode
->i_list
);
1233 list_del_init(&inode
->i_sb_list
);
1234 WARN_ON(inode
->i_state
& I_NEW
);
1235 inode
->i_state
|= I_FREEING
;
1236 inodes_stat
.nr_inodes
--;
1237 spin_unlock(&inode_lock
);
1238 if (inode
->i_data
.nrpages
)
1239 truncate_inode_pages(&inode
->i_data
, 0);
1241 wake_up_inode(inode
);
1242 destroy_inode(inode
);
1246 * Normal UNIX filesystem behaviour: delete the
1247 * inode when the usage count drops to zero, and
1250 void generic_drop_inode(struct inode
*inode
)
1252 if (!inode
->i_nlink
)
1253 generic_delete_inode(inode
);
1255 generic_forget_inode(inode
);
1257 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1260 * Called when we're dropping the last reference
1263 * Call the FS "drop()" function, defaulting to
1264 * the legacy UNIX filesystem behaviour..
1266 * NOTE! NOTE! NOTE! We're called with the inode lock
1267 * held, and the drop function is supposed to release
1270 static inline void iput_final(struct inode
*inode
)
1272 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1273 void (*drop
)(struct inode
*) = generic_drop_inode
;
1275 if (op
&& op
->drop_inode
)
1276 drop
= op
->drop_inode
;
1281 * iput - put an inode
1282 * @inode: inode to put
1284 * Puts an inode, dropping its usage count. If the inode use count hits
1285 * zero, the inode is then freed and may also be destroyed.
1287 * Consequently, iput() can sleep.
1289 void iput(struct inode
*inode
)
1292 BUG_ON(inode
->i_state
== I_CLEAR
);
1294 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1298 EXPORT_SYMBOL(iput
);
1301 * bmap - find a block number in a file
1302 * @inode: inode of file
1303 * @block: block to find
1305 * Returns the block number on the device holding the inode that
1306 * is the disk block number for the block of the file requested.
1307 * That is, asked for block 4 of inode 1 the function will return the
1308 * disk block relative to the disk start that holds that block of the
1311 sector_t
bmap(struct inode
*inode
, sector_t block
)
1314 if (inode
->i_mapping
->a_ops
->bmap
)
1315 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1318 EXPORT_SYMBOL(bmap
);
1321 * With relative atime, only update atime if the previous atime is
1322 * earlier than either the ctime or mtime or if at least a day has
1323 * passed since the last atime update.
1325 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1326 struct timespec now
)
1329 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1332 * Is mtime younger than atime? If yes, update atime:
1334 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1337 * Is ctime younger than atime? If yes, update atime:
1339 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1343 * Is the previous atime value older than a day? If yes,
1346 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1349 * Good, we can skip the atime update:
1355 * touch_atime - update the access time
1356 * @mnt: mount the inode is accessed on
1357 * @dentry: dentry accessed
1359 * Update the accessed time on an inode and mark it for writeback.
1360 * This function automatically handles read only file systems and media,
1361 * as well as the "noatime" flag and inode specific "noatime" markers.
1363 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1365 struct inode
*inode
= dentry
->d_inode
;
1366 struct timespec now
;
1368 if (mnt_want_write(mnt
))
1370 if (inode
->i_flags
& S_NOATIME
)
1372 if (IS_NOATIME(inode
))
1374 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1377 if (mnt
->mnt_flags
& MNT_NOATIME
)
1379 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1382 now
= current_fs_time(inode
->i_sb
);
1384 if (!relatime_need_update(mnt
, inode
, now
))
1387 if (timespec_equal(&inode
->i_atime
, &now
))
1390 inode
->i_atime
= now
;
1391 mark_inode_dirty_sync(inode
);
1393 mnt_drop_write(mnt
);
1395 EXPORT_SYMBOL(touch_atime
);
1398 * file_update_time - update mtime and ctime time
1399 * @file: file accessed
1401 * Update the mtime and ctime members of an inode and mark the inode
1402 * for writeback. Note that this function is meant exclusively for
1403 * usage in the file write path of filesystems, and filesystems may
1404 * choose to explicitly ignore update via this function with the
1405 * S_NOCTIME inode flag, e.g. for network filesystem where these
1406 * timestamps are handled by the server.
1409 void file_update_time(struct file
*file
)
1411 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1412 struct timespec now
;
1416 if (IS_NOCMTIME(inode
))
1419 err
= mnt_want_write(file
->f_path
.mnt
);
1423 now
= current_fs_time(inode
->i_sb
);
1424 if (!timespec_equal(&inode
->i_mtime
, &now
)) {
1425 inode
->i_mtime
= now
;
1429 if (!timespec_equal(&inode
->i_ctime
, &now
)) {
1430 inode
->i_ctime
= now
;
1434 if (IS_I_VERSION(inode
)) {
1435 inode_inc_iversion(inode
);
1440 mark_inode_dirty_sync(inode
);
1441 mnt_drop_write(file
->f_path
.mnt
);
1443 EXPORT_SYMBOL(file_update_time
);
1445 int inode_needs_sync(struct inode
*inode
)
1449 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1453 EXPORT_SYMBOL(inode_needs_sync
);
1455 int inode_wait(void *word
)
1460 EXPORT_SYMBOL(inode_wait
);
1463 * If we try to find an inode in the inode hash while it is being
1464 * deleted, we have to wait until the filesystem completes its
1465 * deletion before reporting that it isn't found. This function waits
1466 * until the deletion _might_ have completed. Callers are responsible
1467 * to recheck inode state.
1469 * It doesn't matter if I_LOCK is not set initially, a call to
1470 * wake_up_inode() after removing from the hash list will DTRT.
1472 * This is called with inode_lock held.
1474 static void __wait_on_freeing_inode(struct inode
*inode
)
1476 wait_queue_head_t
*wq
;
1477 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_LOCK
);
1478 wq
= bit_waitqueue(&inode
->i_state
, __I_LOCK
);
1479 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1480 spin_unlock(&inode_lock
);
1482 finish_wait(wq
, &wait
.wait
);
1483 spin_lock(&inode_lock
);
1486 static __initdata
unsigned long ihash_entries
;
1487 static int __init
set_ihash_entries(char *str
)
1491 ihash_entries
= simple_strtoul(str
, &str
, 0);
1494 __setup("ihash_entries=", set_ihash_entries
);
1497 * Initialize the waitqueues and inode hash table.
1499 void __init
inode_init_early(void)
1503 /* If hashes are distributed across NUMA nodes, defer
1504 * hash allocation until vmalloc space is available.
1510 alloc_large_system_hash("Inode-cache",
1511 sizeof(struct hlist_head
),
1519 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1520 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1523 void __init
inode_init(void)
1527 /* inode slab cache */
1528 inode_cachep
= kmem_cache_create("inode_cache",
1529 sizeof(struct inode
),
1531 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1534 register_shrinker(&icache_shrinker
);
1536 /* Hash may have been set up in inode_init_early */
1541 alloc_large_system_hash("Inode-cache",
1542 sizeof(struct hlist_head
),
1550 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1551 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1554 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1556 inode
->i_mode
= mode
;
1557 if (S_ISCHR(mode
)) {
1558 inode
->i_fop
= &def_chr_fops
;
1559 inode
->i_rdev
= rdev
;
1560 } else if (S_ISBLK(mode
)) {
1561 inode
->i_fop
= &def_blk_fops
;
1562 inode
->i_rdev
= rdev
;
1563 } else if (S_ISFIFO(mode
))
1564 inode
->i_fop
= &def_fifo_fops
;
1565 else if (S_ISSOCK(mode
))
1566 inode
->i_fop
= &bad_sock_fops
;
1568 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o)\n",
1571 EXPORT_SYMBOL(init_special_inode
);