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/fsnotify.h>
26 #include <linux/mount.h>
27 #include <linux/async.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_mutex 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 static DEFINE_MUTEX(iprune_mutex
);
99 * Statistics gathering..
101 struct inodes_stat_t inodes_stat
;
103 static struct kmem_cache
*inode_cachep __read_mostly
;
105 static void wake_up_inode(struct inode
*inode
)
108 * Prevent speculative execution through spin_unlock(&inode_lock);
111 wake_up_bit(&inode
->i_state
, __I_LOCK
);
115 * inode_init_always - perform inode structure intialisation
116 * @sb: superblock inode belongs to
117 * @inode: inode to initialise
119 * These are initializations that need to be done on every inode
120 * allocation as the fields are not initialised by slab allocation.
122 struct inode
*inode_init_always(struct super_block
*sb
, struct inode
*inode
)
124 static const struct address_space_operations empty_aops
;
125 static struct inode_operations empty_iops
;
126 static const struct file_operations empty_fops
;
128 struct address_space
*const mapping
= &inode
->i_data
;
131 inode
->i_blkbits
= sb
->s_blocksize_bits
;
133 atomic_set(&inode
->i_count
, 1);
134 inode
->i_op
= &empty_iops
;
135 inode
->i_fop
= &empty_fops
;
139 atomic_set(&inode
->i_writecount
, 0);
143 inode
->i_generation
= 0;
145 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
147 inode
->i_pipe
= NULL
;
148 inode
->i_bdev
= NULL
;
149 inode
->i_cdev
= NULL
;
151 inode
->dirtied_when
= 0;
153 if (security_inode_alloc(inode
))
156 /* allocate and initialize an i_integrity */
157 if (ima_inode_alloc(inode
))
158 goto out_free_security
;
160 spin_lock_init(&inode
->i_lock
);
161 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
163 mutex_init(&inode
->i_mutex
);
164 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
166 init_rwsem(&inode
->i_alloc_sem
);
167 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
169 mapping
->a_ops
= &empty_aops
;
170 mapping
->host
= inode
;
172 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
173 mapping
->assoc_mapping
= NULL
;
174 mapping
->backing_dev_info
= &default_backing_dev_info
;
175 mapping
->writeback_index
= 0;
178 * If the block_device provides a backing_dev_info for client
179 * inodes then use that. Otherwise the inode share the bdev's
183 struct backing_dev_info
*bdi
;
185 bdi
= sb
->s_bdev
->bd_inode_backing_dev_info
;
187 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
188 mapping
->backing_dev_info
= bdi
;
190 inode
->i_private
= NULL
;
191 inode
->i_mapping
= mapping
;
193 #ifdef CONFIG_FSNOTIFY
194 inode
->i_fsnotify_mask
= 0;
200 security_inode_free(inode
);
202 if (inode
->i_sb
->s_op
->destroy_inode
)
203 inode
->i_sb
->s_op
->destroy_inode(inode
);
205 kmem_cache_free(inode_cachep
, (inode
));
208 EXPORT_SYMBOL(inode_init_always
);
210 static struct inode
*alloc_inode(struct super_block
*sb
)
214 if (sb
->s_op
->alloc_inode
)
215 inode
= sb
->s_op
->alloc_inode(sb
);
217 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
220 return inode_init_always(sb
, inode
);
224 void destroy_inode(struct inode
*inode
)
226 BUG_ON(inode_has_buffers(inode
));
227 ima_inode_free(inode
);
228 security_inode_free(inode
);
229 fsnotify_inode_delete(inode
);
230 if (inode
->i_sb
->s_op
->destroy_inode
)
231 inode
->i_sb
->s_op
->destroy_inode(inode
);
233 kmem_cache_free(inode_cachep
, (inode
));
235 EXPORT_SYMBOL(destroy_inode
);
239 * These are initializations that only need to be done
240 * once, because the fields are idempotent across use
241 * of the inode, so let the slab aware of that.
243 void inode_init_once(struct inode
*inode
)
245 memset(inode
, 0, sizeof(*inode
));
246 INIT_HLIST_NODE(&inode
->i_hash
);
247 INIT_LIST_HEAD(&inode
->i_dentry
);
248 INIT_LIST_HEAD(&inode
->i_devices
);
249 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
250 spin_lock_init(&inode
->i_data
.tree_lock
);
251 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
252 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
253 spin_lock_init(&inode
->i_data
.private_lock
);
254 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
255 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
256 i_size_ordered_init(inode
);
257 #ifdef CONFIG_INOTIFY
258 INIT_LIST_HEAD(&inode
->inotify_watches
);
259 mutex_init(&inode
->inotify_mutex
);
261 #ifdef CONFIG_FSNOTIFY
262 INIT_HLIST_HEAD(&inode
->i_fsnotify_mark_entries
);
265 EXPORT_SYMBOL(inode_init_once
);
267 static void init_once(void *foo
)
269 struct inode
*inode
= (struct inode
*) foo
;
271 inode_init_once(inode
);
275 * inode_lock must be held
277 void __iget(struct inode
*inode
)
279 if (atomic_read(&inode
->i_count
)) {
280 atomic_inc(&inode
->i_count
);
283 atomic_inc(&inode
->i_count
);
284 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
285 list_move(&inode
->i_list
, &inode_in_use
);
286 inodes_stat
.nr_unused
--;
290 * clear_inode - clear an inode
291 * @inode: inode to clear
293 * This is called by the filesystem to tell us
294 * that the inode is no longer useful. We just
295 * terminate it with extreme prejudice.
297 void clear_inode(struct inode
*inode
)
300 invalidate_inode_buffers(inode
);
302 BUG_ON(inode
->i_data
.nrpages
);
303 BUG_ON(!(inode
->i_state
& I_FREEING
));
304 BUG_ON(inode
->i_state
& I_CLEAR
);
305 inode_sync_wait(inode
);
307 if (inode
->i_sb
->s_op
->clear_inode
)
308 inode
->i_sb
->s_op
->clear_inode(inode
);
309 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
311 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
313 inode
->i_state
= I_CLEAR
;
315 EXPORT_SYMBOL(clear_inode
);
318 * dispose_list - dispose of the contents of a local list
319 * @head: the head of the list to free
321 * Dispose-list gets a local list with local inodes in it, so it doesn't
322 * need to worry about list corruption and SMP locks.
324 static void dispose_list(struct list_head
*head
)
328 while (!list_empty(head
)) {
331 inode
= list_first_entry(head
, struct inode
, i_list
);
332 list_del(&inode
->i_list
);
334 if (inode
->i_data
.nrpages
)
335 truncate_inode_pages(&inode
->i_data
, 0);
338 spin_lock(&inode_lock
);
339 hlist_del_init(&inode
->i_hash
);
340 list_del_init(&inode
->i_sb_list
);
341 spin_unlock(&inode_lock
);
343 wake_up_inode(inode
);
344 destroy_inode(inode
);
347 spin_lock(&inode_lock
);
348 inodes_stat
.nr_inodes
-= nr_disposed
;
349 spin_unlock(&inode_lock
);
353 * Invalidate all inodes for a device.
355 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
357 struct list_head
*next
;
358 int busy
= 0, count
= 0;
362 struct list_head
*tmp
= next
;
366 * We can reschedule here without worrying about the list's
367 * consistency because the per-sb list of inodes must not
368 * change during umount anymore, and because iprune_mutex keeps
369 * shrink_icache_memory() away.
371 cond_resched_lock(&inode_lock
);
376 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
377 if (inode
->i_state
& I_NEW
)
379 invalidate_inode_buffers(inode
);
380 if (!atomic_read(&inode
->i_count
)) {
381 list_move(&inode
->i_list
, dispose
);
382 WARN_ON(inode
->i_state
& I_NEW
);
383 inode
->i_state
|= I_FREEING
;
389 /* only unused inodes may be cached with i_count zero */
390 inodes_stat
.nr_unused
-= count
;
395 * invalidate_inodes - discard the inodes on a device
398 * Discard all of the inodes for a given superblock. If the discard
399 * fails because there are busy inodes then a non zero value is returned.
400 * If the discard is successful all the inodes have been discarded.
402 int invalidate_inodes(struct super_block
*sb
)
405 LIST_HEAD(throw_away
);
407 mutex_lock(&iprune_mutex
);
408 spin_lock(&inode_lock
);
409 inotify_unmount_inodes(&sb
->s_inodes
);
410 fsnotify_unmount_inodes(&sb
->s_inodes
);
411 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
412 spin_unlock(&inode_lock
);
414 dispose_list(&throw_away
);
415 mutex_unlock(&iprune_mutex
);
419 EXPORT_SYMBOL(invalidate_inodes
);
421 static int can_unuse(struct inode
*inode
)
425 if (inode_has_buffers(inode
))
427 if (atomic_read(&inode
->i_count
))
429 if (inode
->i_data
.nrpages
)
435 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
436 * a temporary list and then are freed outside inode_lock by dispose_list().
438 * Any inodes which are pinned purely because of attached pagecache have their
439 * pagecache removed. We expect the final iput() on that inode to add it to
440 * the front of the inode_unused list. So look for it there and if the
441 * inode is still freeable, proceed. The right inode is found 99.9% of the
442 * time in testing on a 4-way.
444 * If the inode has metadata buffers attached to mapping->private_list then
445 * try to remove them.
447 static void prune_icache(int nr_to_scan
)
452 unsigned long reap
= 0;
454 mutex_lock(&iprune_mutex
);
455 spin_lock(&inode_lock
);
456 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
459 if (list_empty(&inode_unused
))
462 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
464 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
465 list_move(&inode
->i_list
, &inode_unused
);
468 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
470 spin_unlock(&inode_lock
);
471 if (remove_inode_buffers(inode
))
472 reap
+= invalidate_mapping_pages(&inode
->i_data
,
475 spin_lock(&inode_lock
);
477 if (inode
!= list_entry(inode_unused
.next
,
478 struct inode
, i_list
))
479 continue; /* wrong inode or list_empty */
480 if (!can_unuse(inode
))
483 list_move(&inode
->i_list
, &freeable
);
484 WARN_ON(inode
->i_state
& I_NEW
);
485 inode
->i_state
|= I_FREEING
;
488 inodes_stat
.nr_unused
-= nr_pruned
;
489 if (current_is_kswapd())
490 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
492 __count_vm_events(PGINODESTEAL
, reap
);
493 spin_unlock(&inode_lock
);
495 dispose_list(&freeable
);
496 mutex_unlock(&iprune_mutex
);
500 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
501 * "unused" means that no dentries are referring to the inodes: the files are
502 * not open and the dcache references to those inodes have already been
505 * This function is passed the number of inodes to scan, and it returns the
506 * total number of remaining possibly-reclaimable inodes.
508 static int shrink_icache_memory(int nr
, gfp_t gfp_mask
)
512 * Nasty deadlock avoidance. We may hold various FS locks,
513 * and we don't want to recurse into the FS that called us
514 * in clear_inode() and friends..
516 if (!(gfp_mask
& __GFP_FS
))
520 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
523 static struct shrinker icache_shrinker
= {
524 .shrink
= shrink_icache_memory
,
525 .seeks
= DEFAULT_SEEKS
,
528 static void __wait_on_freeing_inode(struct inode
*inode
);
530 * Called with the inode lock held.
531 * NOTE: we are not increasing the inode-refcount, you must call __iget()
532 * by hand after calling find_inode now! This simplifies iunique and won't
533 * add any additional branch in the common code.
535 static struct inode
*find_inode(struct super_block
*sb
,
536 struct hlist_head
*head
,
537 int (*test
)(struct inode
*, void *),
540 struct hlist_node
*node
;
541 struct inode
*inode
= NULL
;
544 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
545 if (inode
->i_sb
!= sb
)
547 if (!test(inode
, data
))
549 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
550 __wait_on_freeing_inode(inode
);
555 return node
? inode
: NULL
;
559 * find_inode_fast is the fast path version of find_inode, see the comment at
560 * iget_locked for details.
562 static struct inode
*find_inode_fast(struct super_block
*sb
,
563 struct hlist_head
*head
, unsigned long ino
)
565 struct hlist_node
*node
;
566 struct inode
*inode
= NULL
;
569 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
570 if (inode
->i_ino
!= ino
)
572 if (inode
->i_sb
!= sb
)
574 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
575 __wait_on_freeing_inode(inode
);
580 return node
? inode
: NULL
;
583 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
587 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
589 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
590 return tmp
& I_HASHMASK
;
594 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
597 inodes_stat
.nr_inodes
++;
598 list_add(&inode
->i_list
, &inode_in_use
);
599 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
601 hlist_add_head(&inode
->i_hash
, head
);
605 * inode_add_to_lists - add a new inode to relevant lists
606 * @sb: superblock inode belongs to
607 * @inode: inode to mark in use
609 * When an inode is allocated it needs to be accounted for, added to the in use
610 * list, the owning superblock and the inode hash. This needs to be done under
611 * the inode_lock, so export a function to do this rather than the inode lock
612 * itself. We calculate the hash list to add to here so it is all internal
613 * which requires the caller to have already set up the inode number in the
616 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
618 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
620 spin_lock(&inode_lock
);
621 __inode_add_to_lists(sb
, head
, inode
);
622 spin_unlock(&inode_lock
);
624 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
627 * new_inode - obtain an inode
630 * Allocates a new inode for given superblock. The default gfp_mask
631 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
632 * If HIGHMEM pages are unsuitable or it is known that pages allocated
633 * for the page cache are not reclaimable or migratable,
634 * mapping_set_gfp_mask() must be called with suitable flags on the
635 * newly created inode's mapping
638 struct inode
*new_inode(struct super_block
*sb
)
641 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
642 * error if st_ino won't fit in target struct field. Use 32bit counter
643 * here to attempt to avoid that.
645 static unsigned int last_ino
;
648 spin_lock_prefetch(&inode_lock
);
650 inode
= alloc_inode(sb
);
652 spin_lock(&inode_lock
);
653 __inode_add_to_lists(sb
, NULL
, inode
);
654 inode
->i_ino
= ++last_ino
;
656 spin_unlock(&inode_lock
);
660 EXPORT_SYMBOL(new_inode
);
662 void unlock_new_inode(struct inode
*inode
)
664 #ifdef CONFIG_DEBUG_LOCK_ALLOC
665 if (inode
->i_mode
& S_IFDIR
) {
666 struct file_system_type
*type
= inode
->i_sb
->s_type
;
668 /* Set new key only if filesystem hasn't already changed it */
669 if (!lockdep_match_class(&inode
->i_mutex
,
670 &type
->i_mutex_key
)) {
672 * ensure nobody is actually holding i_mutex
674 mutex_destroy(&inode
->i_mutex
);
675 mutex_init(&inode
->i_mutex
);
676 lockdep_set_class(&inode
->i_mutex
,
677 &type
->i_mutex_dir_key
);
682 * This is special! We do not need the spinlock
683 * when clearing I_LOCK, because we're guaranteed
684 * that nobody else tries to do anything about the
685 * state of the inode when it is locked, as we
686 * just created it (so there can be no old holders
687 * that haven't tested I_LOCK).
689 WARN_ON((inode
->i_state
& (I_LOCK
|I_NEW
)) != (I_LOCK
|I_NEW
));
690 inode
->i_state
&= ~(I_LOCK
|I_NEW
);
691 wake_up_inode(inode
);
693 EXPORT_SYMBOL(unlock_new_inode
);
696 * This is called without the inode lock held.. Be careful.
698 * We no longer cache the sb_flags in i_flags - see fs.h
699 * -- rmk@arm.uk.linux.org
701 static struct inode
*get_new_inode(struct super_block
*sb
,
702 struct hlist_head
*head
,
703 int (*test
)(struct inode
*, void *),
704 int (*set
)(struct inode
*, void *),
709 inode
= alloc_inode(sb
);
713 spin_lock(&inode_lock
);
714 /* We released the lock, so.. */
715 old
= find_inode(sb
, head
, test
, data
);
717 if (set(inode
, data
))
720 __inode_add_to_lists(sb
, head
, inode
);
721 inode
->i_state
= I_LOCK
|I_NEW
;
722 spin_unlock(&inode_lock
);
724 /* Return the locked inode with I_NEW set, the
725 * caller is responsible for filling in the contents
731 * Uhhuh, somebody else created the same inode under
732 * us. Use the old inode instead of the one we just
736 spin_unlock(&inode_lock
);
737 destroy_inode(inode
);
739 wait_on_inode(inode
);
744 spin_unlock(&inode_lock
);
745 destroy_inode(inode
);
750 * get_new_inode_fast is the fast path version of get_new_inode, see the
751 * comment at iget_locked for details.
753 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
754 struct hlist_head
*head
, unsigned long ino
)
758 inode
= alloc_inode(sb
);
762 spin_lock(&inode_lock
);
763 /* We released the lock, so.. */
764 old
= find_inode_fast(sb
, head
, ino
);
767 __inode_add_to_lists(sb
, head
, inode
);
768 inode
->i_state
= I_LOCK
|I_NEW
;
769 spin_unlock(&inode_lock
);
771 /* Return the locked inode with I_NEW set, the
772 * caller is responsible for filling in the contents
778 * Uhhuh, somebody else created the same inode under
779 * us. Use the old inode instead of the one we just
783 spin_unlock(&inode_lock
);
784 destroy_inode(inode
);
786 wait_on_inode(inode
);
792 * iunique - get a unique inode number
794 * @max_reserved: highest reserved inode number
796 * Obtain an inode number that is unique on the system for a given
797 * superblock. This is used by file systems that have no natural
798 * permanent inode numbering system. An inode number is returned that
799 * is higher than the reserved limit but unique.
802 * With a large number of inodes live on the file system this function
803 * currently becomes quite slow.
805 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
808 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
809 * error if st_ino won't fit in target struct field. Use 32bit counter
810 * here to attempt to avoid that.
812 static unsigned int counter
;
814 struct hlist_head
*head
;
817 spin_lock(&inode_lock
);
819 if (counter
<= max_reserved
)
820 counter
= max_reserved
+ 1;
822 head
= inode_hashtable
+ hash(sb
, res
);
823 inode
= find_inode_fast(sb
, head
, res
);
824 } while (inode
!= NULL
);
825 spin_unlock(&inode_lock
);
829 EXPORT_SYMBOL(iunique
);
831 struct inode
*igrab(struct inode
*inode
)
833 spin_lock(&inode_lock
);
834 if (!(inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)))
838 * Handle the case where s_op->clear_inode is not been
839 * called yet, and somebody is calling igrab
840 * while the inode is getting freed.
843 spin_unlock(&inode_lock
);
846 EXPORT_SYMBOL(igrab
);
849 * ifind - internal function, you want ilookup5() or iget5().
850 * @sb: super block of file system to search
851 * @head: the head of the list to search
852 * @test: callback used for comparisons between inodes
853 * @data: opaque data pointer to pass to @test
854 * @wait: if true wait for the inode to be unlocked, if false do not
856 * ifind() searches for the inode specified by @data in the inode
857 * cache. This is a generalized version of ifind_fast() for file systems where
858 * the inode number is not sufficient for unique identification of an inode.
860 * If the inode is in the cache, the inode is returned with an incremented
863 * Otherwise NULL is returned.
865 * Note, @test is called with the inode_lock held, so can't sleep.
867 static struct inode
*ifind(struct super_block
*sb
,
868 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
869 void *data
, const int wait
)
873 spin_lock(&inode_lock
);
874 inode
= find_inode(sb
, head
, test
, data
);
877 spin_unlock(&inode_lock
);
879 wait_on_inode(inode
);
882 spin_unlock(&inode_lock
);
887 * ifind_fast - internal function, you want ilookup() or iget().
888 * @sb: super block of file system to search
889 * @head: head of the list to search
890 * @ino: inode number to search for
892 * ifind_fast() searches for the inode @ino in the inode cache. This is for
893 * file systems where the inode number is sufficient for unique identification
896 * If the inode is in the cache, the inode is returned with an incremented
899 * Otherwise NULL is returned.
901 static struct inode
*ifind_fast(struct super_block
*sb
,
902 struct hlist_head
*head
, unsigned long ino
)
906 spin_lock(&inode_lock
);
907 inode
= find_inode_fast(sb
, head
, ino
);
910 spin_unlock(&inode_lock
);
911 wait_on_inode(inode
);
914 spin_unlock(&inode_lock
);
919 * ilookup5_nowait - search for an inode in the inode cache
920 * @sb: super block of file system to search
921 * @hashval: hash value (usually inode number) to search for
922 * @test: callback used for comparisons between inodes
923 * @data: opaque data pointer to pass to @test
925 * ilookup5() uses ifind() to search for the inode specified by @hashval and
926 * @data in the inode cache. This is a generalized version of ilookup() for
927 * file systems where the inode number is not sufficient for unique
928 * identification of an inode.
930 * If the inode is in the cache, the inode is returned with an incremented
931 * reference count. Note, the inode lock is not waited upon so you have to be
932 * very careful what you do with the returned inode. You probably should be
933 * using ilookup5() instead.
935 * Otherwise NULL is returned.
937 * Note, @test is called with the inode_lock held, so can't sleep.
939 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
940 int (*test
)(struct inode
*, void *), void *data
)
942 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
944 return ifind(sb
, head
, test
, data
, 0);
946 EXPORT_SYMBOL(ilookup5_nowait
);
949 * ilookup5 - search for an inode in the inode cache
950 * @sb: super block of file system to search
951 * @hashval: hash value (usually inode number) to search for
952 * @test: callback used for comparisons between inodes
953 * @data: opaque data pointer to pass to @test
955 * ilookup5() uses ifind() to search for the inode specified by @hashval and
956 * @data in the inode cache. This is a generalized version of ilookup() for
957 * file systems where the inode number is not sufficient for unique
958 * identification of an inode.
960 * If the inode is in the cache, the inode lock is waited upon and the inode is
961 * returned with an incremented reference count.
963 * Otherwise NULL is returned.
965 * Note, @test is called with the inode_lock held, so can't sleep.
967 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
968 int (*test
)(struct inode
*, void *), void *data
)
970 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
972 return ifind(sb
, head
, test
, data
, 1);
974 EXPORT_SYMBOL(ilookup5
);
977 * ilookup - search for an inode in the inode cache
978 * @sb: super block of file system to search
979 * @ino: inode number to search for
981 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
982 * This is for file systems where the inode number is sufficient for unique
983 * identification of an inode.
985 * If the inode is in the cache, the inode is returned with an incremented
988 * Otherwise NULL is returned.
990 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
992 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
994 return ifind_fast(sb
, head
, ino
);
996 EXPORT_SYMBOL(ilookup
);
999 * iget5_locked - obtain an inode from a mounted file system
1000 * @sb: super block of file system
1001 * @hashval: hash value (usually inode number) to get
1002 * @test: callback used for comparisons between inodes
1003 * @set: callback used to initialize a new struct inode
1004 * @data: opaque data pointer to pass to @test and @set
1006 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1007 * and @data in the inode cache and if present it is returned with an increased
1008 * reference count. This is a generalized version of iget_locked() for file
1009 * systems where the inode number is not sufficient for unique identification
1012 * If the inode is not in cache, get_new_inode() is called to allocate a new
1013 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1014 * file system gets to fill it in before unlocking it via unlock_new_inode().
1016 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1018 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1019 int (*test
)(struct inode
*, void *),
1020 int (*set
)(struct inode
*, void *), void *data
)
1022 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1023 struct inode
*inode
;
1025 inode
= ifind(sb
, head
, test
, data
, 1);
1029 * get_new_inode() will do the right thing, re-trying the search
1030 * in case it had to block at any point.
1032 return get_new_inode(sb
, head
, test
, set
, data
);
1034 EXPORT_SYMBOL(iget5_locked
);
1037 * iget_locked - obtain an inode from a mounted file system
1038 * @sb: super block of file system
1039 * @ino: inode number to get
1041 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1042 * the inode cache and if present it is returned with an increased reference
1043 * count. This is for file systems where the inode number is sufficient for
1044 * unique identification of an inode.
1046 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1047 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1048 * The file system gets to fill it in before unlocking it via
1049 * unlock_new_inode().
1051 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1053 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1054 struct inode
*inode
;
1056 inode
= ifind_fast(sb
, head
, ino
);
1060 * get_new_inode_fast() will do the right thing, re-trying the search
1061 * in case it had to block at any point.
1063 return get_new_inode_fast(sb
, head
, ino
);
1065 EXPORT_SYMBOL(iget_locked
);
1067 int insert_inode_locked(struct inode
*inode
)
1069 struct super_block
*sb
= inode
->i_sb
;
1070 ino_t ino
= inode
->i_ino
;
1071 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1073 inode
->i_state
|= I_LOCK
|I_NEW
;
1075 struct hlist_node
*node
;
1076 struct inode
*old
= NULL
;
1077 spin_lock(&inode_lock
);
1078 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1079 if (old
->i_ino
!= ino
)
1081 if (old
->i_sb
!= sb
)
1083 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1087 if (likely(!node
)) {
1088 hlist_add_head(&inode
->i_hash
, head
);
1089 spin_unlock(&inode_lock
);
1093 spin_unlock(&inode_lock
);
1095 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1102 EXPORT_SYMBOL(insert_inode_locked
);
1104 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1105 int (*test
)(struct inode
*, void *), void *data
)
1107 struct super_block
*sb
= inode
->i_sb
;
1108 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1110 inode
->i_state
|= I_LOCK
|I_NEW
;
1113 struct hlist_node
*node
;
1114 struct inode
*old
= NULL
;
1116 spin_lock(&inode_lock
);
1117 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1118 if (old
->i_sb
!= sb
)
1120 if (!test(old
, data
))
1122 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1126 if (likely(!node
)) {
1127 hlist_add_head(&inode
->i_hash
, head
);
1128 spin_unlock(&inode_lock
);
1132 spin_unlock(&inode_lock
);
1134 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1141 EXPORT_SYMBOL(insert_inode_locked4
);
1144 * __insert_inode_hash - hash an inode
1145 * @inode: unhashed inode
1146 * @hashval: unsigned long value used to locate this object in the
1149 * Add an inode to the inode hash for this superblock.
1151 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1153 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1154 spin_lock(&inode_lock
);
1155 hlist_add_head(&inode
->i_hash
, head
);
1156 spin_unlock(&inode_lock
);
1158 EXPORT_SYMBOL(__insert_inode_hash
);
1161 * remove_inode_hash - remove an inode from the hash
1162 * @inode: inode to unhash
1164 * Remove an inode from the superblock.
1166 void remove_inode_hash(struct inode
*inode
)
1168 spin_lock(&inode_lock
);
1169 hlist_del_init(&inode
->i_hash
);
1170 spin_unlock(&inode_lock
);
1172 EXPORT_SYMBOL(remove_inode_hash
);
1175 * Tell the filesystem that this inode is no longer of any interest and should
1176 * be completely destroyed.
1178 * We leave the inode in the inode hash table until *after* the filesystem's
1179 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1180 * instigate) will always find up-to-date information either in the hash or on
1183 * I_FREEING is set so that no-one will take a new reference to the inode while
1184 * it is being deleted.
1186 void generic_delete_inode(struct inode
*inode
)
1188 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1190 list_del_init(&inode
->i_list
);
1191 list_del_init(&inode
->i_sb_list
);
1192 WARN_ON(inode
->i_state
& I_NEW
);
1193 inode
->i_state
|= I_FREEING
;
1194 inodes_stat
.nr_inodes
--;
1195 spin_unlock(&inode_lock
);
1197 security_inode_delete(inode
);
1199 if (op
->delete_inode
) {
1200 void (*delete)(struct inode
*) = op
->delete_inode
;
1201 if (!is_bad_inode(inode
))
1203 /* Filesystems implementing their own
1204 * s_op->delete_inode are required to call
1205 * truncate_inode_pages and clear_inode()
1209 truncate_inode_pages(&inode
->i_data
, 0);
1212 spin_lock(&inode_lock
);
1213 hlist_del_init(&inode
->i_hash
);
1214 spin_unlock(&inode_lock
);
1215 wake_up_inode(inode
);
1216 BUG_ON(inode
->i_state
!= I_CLEAR
);
1217 destroy_inode(inode
);
1219 EXPORT_SYMBOL(generic_delete_inode
);
1221 static void generic_forget_inode(struct inode
*inode
)
1223 struct super_block
*sb
= inode
->i_sb
;
1225 if (!hlist_unhashed(&inode
->i_hash
)) {
1226 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1227 list_move(&inode
->i_list
, &inode_unused
);
1228 inodes_stat
.nr_unused
++;
1229 if (sb
->s_flags
& MS_ACTIVE
) {
1230 spin_unlock(&inode_lock
);
1233 WARN_ON(inode
->i_state
& I_NEW
);
1234 inode
->i_state
|= I_WILL_FREE
;
1235 spin_unlock(&inode_lock
);
1236 write_inode_now(inode
, 1);
1237 spin_lock(&inode_lock
);
1238 WARN_ON(inode
->i_state
& I_NEW
);
1239 inode
->i_state
&= ~I_WILL_FREE
;
1240 inodes_stat
.nr_unused
--;
1241 hlist_del_init(&inode
->i_hash
);
1243 list_del_init(&inode
->i_list
);
1244 list_del_init(&inode
->i_sb_list
);
1245 WARN_ON(inode
->i_state
& I_NEW
);
1246 inode
->i_state
|= I_FREEING
;
1247 inodes_stat
.nr_inodes
--;
1248 spin_unlock(&inode_lock
);
1249 if (inode
->i_data
.nrpages
)
1250 truncate_inode_pages(&inode
->i_data
, 0);
1252 wake_up_inode(inode
);
1253 destroy_inode(inode
);
1257 * Normal UNIX filesystem behaviour: delete the
1258 * inode when the usage count drops to zero, and
1261 void generic_drop_inode(struct inode
*inode
)
1263 if (!inode
->i_nlink
)
1264 generic_delete_inode(inode
);
1266 generic_forget_inode(inode
);
1268 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1271 * Called when we're dropping the last reference
1274 * Call the FS "drop()" function, defaulting to
1275 * the legacy UNIX filesystem behaviour..
1277 * NOTE! NOTE! NOTE! We're called with the inode lock
1278 * held, and the drop function is supposed to release
1281 static inline void iput_final(struct inode
*inode
)
1283 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1284 void (*drop
)(struct inode
*) = generic_drop_inode
;
1286 if (op
&& op
->drop_inode
)
1287 drop
= op
->drop_inode
;
1292 * iput - put an inode
1293 * @inode: inode to put
1295 * Puts an inode, dropping its usage count. If the inode use count hits
1296 * zero, the inode is then freed and may also be destroyed.
1298 * Consequently, iput() can sleep.
1300 void iput(struct inode
*inode
)
1303 BUG_ON(inode
->i_state
== I_CLEAR
);
1305 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1309 EXPORT_SYMBOL(iput
);
1312 * bmap - find a block number in a file
1313 * @inode: inode of file
1314 * @block: block to find
1316 * Returns the block number on the device holding the inode that
1317 * is the disk block number for the block of the file requested.
1318 * That is, asked for block 4 of inode 1 the function will return the
1319 * disk block relative to the disk start that holds that block of the
1322 sector_t
bmap(struct inode
*inode
, sector_t block
)
1325 if (inode
->i_mapping
->a_ops
->bmap
)
1326 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1329 EXPORT_SYMBOL(bmap
);
1332 * With relative atime, only update atime if the previous atime is
1333 * earlier than either the ctime or mtime or if at least a day has
1334 * passed since the last atime update.
1336 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1337 struct timespec now
)
1340 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1343 * Is mtime younger than atime? If yes, update atime:
1345 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1348 * Is ctime younger than atime? If yes, update atime:
1350 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1354 * Is the previous atime value older than a day? If yes,
1357 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1360 * Good, we can skip the atime update:
1366 * touch_atime - update the access time
1367 * @mnt: mount the inode is accessed on
1368 * @dentry: dentry accessed
1370 * Update the accessed time on an inode and mark it for writeback.
1371 * This function automatically handles read only file systems and media,
1372 * as well as the "noatime" flag and inode specific "noatime" markers.
1374 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1376 struct inode
*inode
= dentry
->d_inode
;
1377 struct timespec now
;
1379 if (mnt_want_write(mnt
))
1381 if (inode
->i_flags
& S_NOATIME
)
1383 if (IS_NOATIME(inode
))
1385 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1388 if (mnt
->mnt_flags
& MNT_NOATIME
)
1390 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1393 now
= current_fs_time(inode
->i_sb
);
1395 if (!relatime_need_update(mnt
, inode
, now
))
1398 if (timespec_equal(&inode
->i_atime
, &now
))
1401 inode
->i_atime
= now
;
1402 mark_inode_dirty_sync(inode
);
1404 mnt_drop_write(mnt
);
1406 EXPORT_SYMBOL(touch_atime
);
1409 * file_update_time - update mtime and ctime time
1410 * @file: file accessed
1412 * Update the mtime and ctime members of an inode and mark the inode
1413 * for writeback. Note that this function is meant exclusively for
1414 * usage in the file write path of filesystems, and filesystems may
1415 * choose to explicitly ignore update via this function with the
1416 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1417 * timestamps are handled by the server.
1420 void file_update_time(struct file
*file
)
1422 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1423 struct timespec now
;
1427 if (IS_NOCMTIME(inode
))
1430 err
= mnt_want_write_file(file
);
1434 now
= current_fs_time(inode
->i_sb
);
1435 if (!timespec_equal(&inode
->i_mtime
, &now
)) {
1436 inode
->i_mtime
= now
;
1440 if (!timespec_equal(&inode
->i_ctime
, &now
)) {
1441 inode
->i_ctime
= now
;
1445 if (IS_I_VERSION(inode
)) {
1446 inode_inc_iversion(inode
);
1451 mark_inode_dirty_sync(inode
);
1452 mnt_drop_write(file
->f_path
.mnt
);
1454 EXPORT_SYMBOL(file_update_time
);
1456 int inode_needs_sync(struct inode
*inode
)
1460 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1464 EXPORT_SYMBOL(inode_needs_sync
);
1466 int inode_wait(void *word
)
1471 EXPORT_SYMBOL(inode_wait
);
1474 * If we try to find an inode in the inode hash while it is being
1475 * deleted, we have to wait until the filesystem completes its
1476 * deletion before reporting that it isn't found. This function waits
1477 * until the deletion _might_ have completed. Callers are responsible
1478 * to recheck inode state.
1480 * It doesn't matter if I_LOCK is not set initially, a call to
1481 * wake_up_inode() after removing from the hash list will DTRT.
1483 * This is called with inode_lock held.
1485 static void __wait_on_freeing_inode(struct inode
*inode
)
1487 wait_queue_head_t
*wq
;
1488 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_LOCK
);
1489 wq
= bit_waitqueue(&inode
->i_state
, __I_LOCK
);
1490 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1491 spin_unlock(&inode_lock
);
1493 finish_wait(wq
, &wait
.wait
);
1494 spin_lock(&inode_lock
);
1497 static __initdata
unsigned long ihash_entries
;
1498 static int __init
set_ihash_entries(char *str
)
1502 ihash_entries
= simple_strtoul(str
, &str
, 0);
1505 __setup("ihash_entries=", set_ihash_entries
);
1508 * Initialize the waitqueues and inode hash table.
1510 void __init
inode_init_early(void)
1514 /* If hashes are distributed across NUMA nodes, defer
1515 * hash allocation until vmalloc space is available.
1521 alloc_large_system_hash("Inode-cache",
1522 sizeof(struct hlist_head
),
1530 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1531 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1534 void __init
inode_init(void)
1538 /* inode slab cache */
1539 inode_cachep
= kmem_cache_create("inode_cache",
1540 sizeof(struct inode
),
1542 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1545 register_shrinker(&icache_shrinker
);
1547 /* Hash may have been set up in inode_init_early */
1552 alloc_large_system_hash("Inode-cache",
1553 sizeof(struct hlist_head
),
1561 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1562 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1565 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1567 inode
->i_mode
= mode
;
1568 if (S_ISCHR(mode
)) {
1569 inode
->i_fop
= &def_chr_fops
;
1570 inode
->i_rdev
= rdev
;
1571 } else if (S_ISBLK(mode
)) {
1572 inode
->i_fop
= &def_blk_fops
;
1573 inode
->i_rdev
= rdev
;
1574 } else if (S_ISFIFO(mode
))
1575 inode
->i_fop
= &def_fifo_fops
;
1576 else if (S_ISSOCK(mode
))
1577 inode
->i_fop
= &bad_sock_fops
;
1579 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o)\n",
1582 EXPORT_SYMBOL(init_special_inode
);