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
);
229 EXPORT_SYMBOL(__destroy_inode
);
231 void destroy_inode(struct inode
*inode
)
233 __destroy_inode(inode
);
234 if (inode
->i_sb
->s_op
->destroy_inode
)
235 inode
->i_sb
->s_op
->destroy_inode(inode
);
237 kmem_cache_free(inode_cachep
, (inode
));
241 * These are initializations that only need to be done
242 * once, because the fields are idempotent across use
243 * of the inode, so let the slab aware of that.
245 void inode_init_once(struct inode
*inode
)
247 memset(inode
, 0, sizeof(*inode
));
248 INIT_HLIST_NODE(&inode
->i_hash
);
249 INIT_LIST_HEAD(&inode
->i_dentry
);
250 INIT_LIST_HEAD(&inode
->i_devices
);
251 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
252 spin_lock_init(&inode
->i_data
.tree_lock
);
253 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
254 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
255 spin_lock_init(&inode
->i_data
.private_lock
);
256 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
257 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
258 i_size_ordered_init(inode
);
259 #ifdef CONFIG_INOTIFY
260 INIT_LIST_HEAD(&inode
->inotify_watches
);
261 mutex_init(&inode
->inotify_mutex
);
264 EXPORT_SYMBOL(inode_init_once
);
266 static void init_once(void *foo
)
268 struct inode
*inode
= (struct inode
*) foo
;
270 inode_init_once(inode
);
274 * inode_lock must be held
276 void __iget(struct inode
*inode
)
278 if (atomic_read(&inode
->i_count
)) {
279 atomic_inc(&inode
->i_count
);
282 atomic_inc(&inode
->i_count
);
283 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
284 list_move(&inode
->i_list
, &inode_in_use
);
285 inodes_stat
.nr_unused
--;
289 * clear_inode - clear an inode
290 * @inode: inode to clear
292 * This is called by the filesystem to tell us
293 * that the inode is no longer useful. We just
294 * terminate it with extreme prejudice.
296 void clear_inode(struct inode
*inode
)
299 invalidate_inode_buffers(inode
);
301 BUG_ON(inode
->i_data
.nrpages
);
302 BUG_ON(!(inode
->i_state
& I_FREEING
));
303 BUG_ON(inode
->i_state
& I_CLEAR
);
304 inode_sync_wait(inode
);
306 if (inode
->i_sb
->s_op
->clear_inode
)
307 inode
->i_sb
->s_op
->clear_inode(inode
);
308 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
310 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
312 inode
->i_state
= I_CLEAR
;
314 EXPORT_SYMBOL(clear_inode
);
317 * dispose_list - dispose of the contents of a local list
318 * @head: the head of the list to free
320 * Dispose-list gets a local list with local inodes in it, so it doesn't
321 * need to worry about list corruption and SMP locks.
323 static void dispose_list(struct list_head
*head
)
327 while (!list_empty(head
)) {
330 inode
= list_first_entry(head
, struct inode
, i_list
);
331 list_del(&inode
->i_list
);
333 if (inode
->i_data
.nrpages
)
334 truncate_inode_pages(&inode
->i_data
, 0);
337 spin_lock(&inode_lock
);
338 hlist_del_init(&inode
->i_hash
);
339 list_del_init(&inode
->i_sb_list
);
340 spin_unlock(&inode_lock
);
342 wake_up_inode(inode
);
343 destroy_inode(inode
);
346 spin_lock(&inode_lock
);
347 inodes_stat
.nr_inodes
-= nr_disposed
;
348 spin_unlock(&inode_lock
);
352 * Invalidate all inodes for a device.
354 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
356 struct list_head
*next
;
357 int busy
= 0, count
= 0;
361 struct list_head
*tmp
= next
;
365 * We can reschedule here without worrying about the list's
366 * consistency because the per-sb list of inodes must not
367 * change during umount anymore, and because iprune_mutex keeps
368 * shrink_icache_memory() away.
370 cond_resched_lock(&inode_lock
);
375 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
376 if (inode
->i_state
& I_NEW
)
378 invalidate_inode_buffers(inode
);
379 if (!atomic_read(&inode
->i_count
)) {
380 list_move(&inode
->i_list
, dispose
);
381 WARN_ON(inode
->i_state
& I_NEW
);
382 inode
->i_state
|= I_FREEING
;
388 /* only unused inodes may be cached with i_count zero */
389 inodes_stat
.nr_unused
-= count
;
394 * invalidate_inodes - discard the inodes on a device
397 * Discard all of the inodes for a given superblock. If the discard
398 * fails because there are busy inodes then a non zero value is returned.
399 * If the discard is successful all the inodes have been discarded.
401 int invalidate_inodes(struct super_block
*sb
)
404 LIST_HEAD(throw_away
);
406 mutex_lock(&iprune_mutex
);
407 spin_lock(&inode_lock
);
408 inotify_unmount_inodes(&sb
->s_inodes
);
409 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
410 spin_unlock(&inode_lock
);
412 dispose_list(&throw_away
);
413 mutex_unlock(&iprune_mutex
);
417 EXPORT_SYMBOL(invalidate_inodes
);
419 static int can_unuse(struct inode
*inode
)
423 if (inode_has_buffers(inode
))
425 if (atomic_read(&inode
->i_count
))
427 if (inode
->i_data
.nrpages
)
433 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
434 * a temporary list and then are freed outside inode_lock by dispose_list().
436 * Any inodes which are pinned purely because of attached pagecache have their
437 * pagecache removed. We expect the final iput() on that inode to add it to
438 * the front of the inode_unused list. So look for it there and if the
439 * inode is still freeable, proceed. The right inode is found 99.9% of the
440 * time in testing on a 4-way.
442 * If the inode has metadata buffers attached to mapping->private_list then
443 * try to remove them.
445 static void prune_icache(int nr_to_scan
)
450 unsigned long reap
= 0;
452 mutex_lock(&iprune_mutex
);
453 spin_lock(&inode_lock
);
454 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
457 if (list_empty(&inode_unused
))
460 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
462 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
463 list_move(&inode
->i_list
, &inode_unused
);
466 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
468 spin_unlock(&inode_lock
);
469 if (remove_inode_buffers(inode
))
470 reap
+= invalidate_mapping_pages(&inode
->i_data
,
473 spin_lock(&inode_lock
);
475 if (inode
!= list_entry(inode_unused
.next
,
476 struct inode
, i_list
))
477 continue; /* wrong inode or list_empty */
478 if (!can_unuse(inode
))
481 list_move(&inode
->i_list
, &freeable
);
482 WARN_ON(inode
->i_state
& I_NEW
);
483 inode
->i_state
|= I_FREEING
;
486 inodes_stat
.nr_unused
-= nr_pruned
;
487 if (current_is_kswapd())
488 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
490 __count_vm_events(PGINODESTEAL
, reap
);
491 spin_unlock(&inode_lock
);
493 dispose_list(&freeable
);
494 mutex_unlock(&iprune_mutex
);
498 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
499 * "unused" means that no dentries are referring to the inodes: the files are
500 * not open and the dcache references to those inodes have already been
503 * This function is passed the number of inodes to scan, and it returns the
504 * total number of remaining possibly-reclaimable inodes.
506 static int shrink_icache_memory(int nr
, gfp_t gfp_mask
)
510 * Nasty deadlock avoidance. We may hold various FS locks,
511 * and we don't want to recurse into the FS that called us
512 * in clear_inode() and friends..
514 if (!(gfp_mask
& __GFP_FS
))
518 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
521 static struct shrinker icache_shrinker
= {
522 .shrink
= shrink_icache_memory
,
523 .seeks
= DEFAULT_SEEKS
,
526 static void __wait_on_freeing_inode(struct inode
*inode
);
528 * Called with the inode lock held.
529 * NOTE: we are not increasing the inode-refcount, you must call __iget()
530 * by hand after calling find_inode now! This simplifies iunique and won't
531 * add any additional branch in the common code.
533 static struct inode
*find_inode(struct super_block
*sb
,
534 struct hlist_head
*head
,
535 int (*test
)(struct inode
*, void *),
538 struct hlist_node
*node
;
539 struct inode
*inode
= NULL
;
542 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
543 if (inode
->i_sb
!= sb
)
545 if (!test(inode
, data
))
547 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
548 __wait_on_freeing_inode(inode
);
553 return node
? inode
: NULL
;
557 * find_inode_fast is the fast path version of find_inode, see the comment at
558 * iget_locked for details.
560 static struct inode
*find_inode_fast(struct super_block
*sb
,
561 struct hlist_head
*head
, unsigned long ino
)
563 struct hlist_node
*node
;
564 struct inode
*inode
= NULL
;
567 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
568 if (inode
->i_ino
!= ino
)
570 if (inode
->i_sb
!= sb
)
572 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
573 __wait_on_freeing_inode(inode
);
578 return node
? inode
: NULL
;
581 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
585 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
587 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
588 return tmp
& I_HASHMASK
;
592 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
595 inodes_stat
.nr_inodes
++;
596 list_add(&inode
->i_list
, &inode_in_use
);
597 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
599 hlist_add_head(&inode
->i_hash
, head
);
603 * inode_add_to_lists - add a new inode to relevant lists
604 * @sb: superblock inode belongs to
605 * @inode: inode to mark in use
607 * When an inode is allocated it needs to be accounted for, added to the in use
608 * list, the owning superblock and the inode hash. This needs to be done under
609 * the inode_lock, so export a function to do this rather than the inode lock
610 * itself. We calculate the hash list to add to here so it is all internal
611 * which requires the caller to have already set up the inode number in the
614 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
616 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
618 spin_lock(&inode_lock
);
619 __inode_add_to_lists(sb
, head
, inode
);
620 spin_unlock(&inode_lock
);
622 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
625 * new_inode - obtain an inode
628 * Allocates a new inode for given superblock. The default gfp_mask
629 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
630 * If HIGHMEM pages are unsuitable or it is known that pages allocated
631 * for the page cache are not reclaimable or migratable,
632 * mapping_set_gfp_mask() must be called with suitable flags on the
633 * newly created inode's mapping
636 struct inode
*new_inode(struct super_block
*sb
)
639 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
640 * error if st_ino won't fit in target struct field. Use 32bit counter
641 * here to attempt to avoid that.
643 static unsigned int last_ino
;
646 spin_lock_prefetch(&inode_lock
);
648 inode
= alloc_inode(sb
);
650 spin_lock(&inode_lock
);
651 __inode_add_to_lists(sb
, NULL
, inode
);
652 inode
->i_ino
= ++last_ino
;
654 spin_unlock(&inode_lock
);
658 EXPORT_SYMBOL(new_inode
);
660 void unlock_new_inode(struct inode
*inode
)
662 #ifdef CONFIG_DEBUG_LOCK_ALLOC
663 if (inode
->i_mode
& S_IFDIR
) {
664 struct file_system_type
*type
= inode
->i_sb
->s_type
;
667 * ensure nobody is actually holding i_mutex
669 mutex_destroy(&inode
->i_mutex
);
670 mutex_init(&inode
->i_mutex
);
671 lockdep_set_class(&inode
->i_mutex
, &type
->i_mutex_dir_key
);
675 * This is special! We do not need the spinlock
676 * when clearing I_LOCK, because we're guaranteed
677 * that nobody else tries to do anything about the
678 * state of the inode when it is locked, as we
679 * just created it (so there can be no old holders
680 * that haven't tested I_LOCK).
682 WARN_ON((inode
->i_state
& (I_LOCK
|I_NEW
)) != (I_LOCK
|I_NEW
));
683 inode
->i_state
&= ~(I_LOCK
|I_NEW
);
684 wake_up_inode(inode
);
686 EXPORT_SYMBOL(unlock_new_inode
);
689 * This is called without the inode lock held.. Be careful.
691 * We no longer cache the sb_flags in i_flags - see fs.h
692 * -- rmk@arm.uk.linux.org
694 static struct inode
*get_new_inode(struct super_block
*sb
,
695 struct hlist_head
*head
,
696 int (*test
)(struct inode
*, void *),
697 int (*set
)(struct inode
*, void *),
702 inode
= alloc_inode(sb
);
706 spin_lock(&inode_lock
);
707 /* We released the lock, so.. */
708 old
= find_inode(sb
, head
, test
, data
);
710 if (set(inode
, data
))
713 __inode_add_to_lists(sb
, head
, inode
);
714 inode
->i_state
= I_LOCK
|I_NEW
;
715 spin_unlock(&inode_lock
);
717 /* Return the locked inode with I_NEW set, the
718 * caller is responsible for filling in the contents
724 * Uhhuh, somebody else created the same inode under
725 * us. Use the old inode instead of the one we just
729 spin_unlock(&inode_lock
);
730 destroy_inode(inode
);
732 wait_on_inode(inode
);
737 spin_unlock(&inode_lock
);
738 destroy_inode(inode
);
743 * get_new_inode_fast is the fast path version of get_new_inode, see the
744 * comment at iget_locked for details.
746 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
747 struct hlist_head
*head
, unsigned long ino
)
751 inode
= alloc_inode(sb
);
755 spin_lock(&inode_lock
);
756 /* We released the lock, so.. */
757 old
= find_inode_fast(sb
, head
, ino
);
760 __inode_add_to_lists(sb
, head
, inode
);
761 inode
->i_state
= I_LOCK
|I_NEW
;
762 spin_unlock(&inode_lock
);
764 /* Return the locked inode with I_NEW set, the
765 * caller is responsible for filling in the contents
771 * Uhhuh, somebody else created the same inode under
772 * us. Use the old inode instead of the one we just
776 spin_unlock(&inode_lock
);
777 destroy_inode(inode
);
779 wait_on_inode(inode
);
785 * iunique - get a unique inode number
787 * @max_reserved: highest reserved inode number
789 * Obtain an inode number that is unique on the system for a given
790 * superblock. This is used by file systems that have no natural
791 * permanent inode numbering system. An inode number is returned that
792 * is higher than the reserved limit but unique.
795 * With a large number of inodes live on the file system this function
796 * currently becomes quite slow.
798 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
801 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
802 * error if st_ino won't fit in target struct field. Use 32bit counter
803 * here to attempt to avoid that.
805 static unsigned int counter
;
807 struct hlist_head
*head
;
810 spin_lock(&inode_lock
);
812 if (counter
<= max_reserved
)
813 counter
= max_reserved
+ 1;
815 head
= inode_hashtable
+ hash(sb
, res
);
816 inode
= find_inode_fast(sb
, head
, res
);
817 } while (inode
!= NULL
);
818 spin_unlock(&inode_lock
);
822 EXPORT_SYMBOL(iunique
);
824 struct inode
*igrab(struct inode
*inode
)
826 spin_lock(&inode_lock
);
827 if (!(inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)))
831 * Handle the case where s_op->clear_inode is not been
832 * called yet, and somebody is calling igrab
833 * while the inode is getting freed.
836 spin_unlock(&inode_lock
);
839 EXPORT_SYMBOL(igrab
);
842 * ifind - internal function, you want ilookup5() or iget5().
843 * @sb: super block of file system to search
844 * @head: the head of the list to search
845 * @test: callback used for comparisons between inodes
846 * @data: opaque data pointer to pass to @test
847 * @wait: if true wait for the inode to be unlocked, if false do not
849 * ifind() searches for the inode specified by @data in the inode
850 * cache. This is a generalized version of ifind_fast() for file systems where
851 * the inode number is not sufficient for unique identification of an inode.
853 * If the inode is in the cache, the inode is returned with an incremented
856 * Otherwise NULL is returned.
858 * Note, @test is called with the inode_lock held, so can't sleep.
860 static struct inode
*ifind(struct super_block
*sb
,
861 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
862 void *data
, const int wait
)
866 spin_lock(&inode_lock
);
867 inode
= find_inode(sb
, head
, test
, data
);
870 spin_unlock(&inode_lock
);
872 wait_on_inode(inode
);
875 spin_unlock(&inode_lock
);
880 * ifind_fast - internal function, you want ilookup() or iget().
881 * @sb: super block of file system to search
882 * @head: head of the list to search
883 * @ino: inode number to search for
885 * ifind_fast() searches for the inode @ino in the inode cache. This is for
886 * file systems where the inode number is sufficient for unique identification
889 * If the inode is in the cache, the inode is returned with an incremented
892 * Otherwise NULL is returned.
894 static struct inode
*ifind_fast(struct super_block
*sb
,
895 struct hlist_head
*head
, unsigned long ino
)
899 spin_lock(&inode_lock
);
900 inode
= find_inode_fast(sb
, head
, ino
);
903 spin_unlock(&inode_lock
);
904 wait_on_inode(inode
);
907 spin_unlock(&inode_lock
);
912 * ilookup5_nowait - search for an inode in the inode cache
913 * @sb: super block of file system to search
914 * @hashval: hash value (usually inode number) to search for
915 * @test: callback used for comparisons between inodes
916 * @data: opaque data pointer to pass to @test
918 * ilookup5() uses ifind() to search for the inode specified by @hashval and
919 * @data in the inode cache. This is a generalized version of ilookup() for
920 * file systems where the inode number is not sufficient for unique
921 * identification of an inode.
923 * If the inode is in the cache, the inode is returned with an incremented
924 * reference count. Note, the inode lock is not waited upon so you have to be
925 * very careful what you do with the returned inode. You probably should be
926 * using ilookup5() instead.
928 * Otherwise NULL is returned.
930 * Note, @test is called with the inode_lock held, so can't sleep.
932 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
933 int (*test
)(struct inode
*, void *), void *data
)
935 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
937 return ifind(sb
, head
, test
, data
, 0);
939 EXPORT_SYMBOL(ilookup5_nowait
);
942 * ilookup5 - search for an inode in the inode cache
943 * @sb: super block of file system to search
944 * @hashval: hash value (usually inode number) to search for
945 * @test: callback used for comparisons between inodes
946 * @data: opaque data pointer to pass to @test
948 * ilookup5() uses ifind() to search for the inode specified by @hashval and
949 * @data in the inode cache. This is a generalized version of ilookup() for
950 * file systems where the inode number is not sufficient for unique
951 * identification of an inode.
953 * If the inode is in the cache, the inode lock is waited upon and the inode is
954 * returned with an incremented reference count.
956 * Otherwise NULL is returned.
958 * Note, @test is called with the inode_lock held, so can't sleep.
960 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
961 int (*test
)(struct inode
*, void *), void *data
)
963 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
965 return ifind(sb
, head
, test
, data
, 1);
967 EXPORT_SYMBOL(ilookup5
);
970 * ilookup - search for an inode in the inode cache
971 * @sb: super block of file system to search
972 * @ino: inode number to search for
974 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
975 * This is for file systems where the inode number is sufficient for unique
976 * identification of an inode.
978 * If the inode is in the cache, the inode is returned with an incremented
981 * Otherwise NULL is returned.
983 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
985 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
987 return ifind_fast(sb
, head
, ino
);
989 EXPORT_SYMBOL(ilookup
);
992 * iget5_locked - obtain an inode from a mounted file system
993 * @sb: super block of file system
994 * @hashval: hash value (usually inode number) to get
995 * @test: callback used for comparisons between inodes
996 * @set: callback used to initialize a new struct inode
997 * @data: opaque data pointer to pass to @test and @set
999 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1000 * and @data in the inode cache and if present it is returned with an increased
1001 * reference count. This is a generalized version of iget_locked() for file
1002 * systems where the inode number is not sufficient for unique identification
1005 * If the inode is not in cache, get_new_inode() is called to allocate a new
1006 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1007 * file system gets to fill it in before unlocking it via unlock_new_inode().
1009 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1011 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1012 int (*test
)(struct inode
*, void *),
1013 int (*set
)(struct inode
*, void *), void *data
)
1015 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1016 struct inode
*inode
;
1018 inode
= ifind(sb
, head
, test
, data
, 1);
1022 * get_new_inode() will do the right thing, re-trying the search
1023 * in case it had to block at any point.
1025 return get_new_inode(sb
, head
, test
, set
, data
);
1027 EXPORT_SYMBOL(iget5_locked
);
1030 * iget_locked - obtain an inode from a mounted file system
1031 * @sb: super block of file system
1032 * @ino: inode number to get
1034 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1035 * the inode cache and if present it is returned with an increased reference
1036 * count. This is for file systems where the inode number is sufficient for
1037 * unique identification of an inode.
1039 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1040 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1041 * The file system gets to fill it in before unlocking it via
1042 * unlock_new_inode().
1044 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1046 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1047 struct inode
*inode
;
1049 inode
= ifind_fast(sb
, head
, ino
);
1053 * get_new_inode_fast() will do the right thing, re-trying the search
1054 * in case it had to block at any point.
1056 return get_new_inode_fast(sb
, head
, ino
);
1058 EXPORT_SYMBOL(iget_locked
);
1060 int insert_inode_locked(struct inode
*inode
)
1062 struct super_block
*sb
= inode
->i_sb
;
1063 ino_t ino
= inode
->i_ino
;
1064 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1066 inode
->i_state
|= I_LOCK
|I_NEW
;
1068 struct hlist_node
*node
;
1069 struct inode
*old
= NULL
;
1070 spin_lock(&inode_lock
);
1071 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1072 if (old
->i_ino
!= ino
)
1074 if (old
->i_sb
!= sb
)
1076 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1080 if (likely(!node
)) {
1081 hlist_add_head(&inode
->i_hash
, head
);
1082 spin_unlock(&inode_lock
);
1086 spin_unlock(&inode_lock
);
1088 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1095 EXPORT_SYMBOL(insert_inode_locked
);
1097 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1098 int (*test
)(struct inode
*, void *), void *data
)
1100 struct super_block
*sb
= inode
->i_sb
;
1101 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1103 inode
->i_state
|= I_LOCK
|I_NEW
;
1106 struct hlist_node
*node
;
1107 struct inode
*old
= NULL
;
1109 spin_lock(&inode_lock
);
1110 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1111 if (old
->i_sb
!= sb
)
1113 if (!test(old
, data
))
1115 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1119 if (likely(!node
)) {
1120 hlist_add_head(&inode
->i_hash
, head
);
1121 spin_unlock(&inode_lock
);
1125 spin_unlock(&inode_lock
);
1127 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1134 EXPORT_SYMBOL(insert_inode_locked4
);
1137 * __insert_inode_hash - hash an inode
1138 * @inode: unhashed inode
1139 * @hashval: unsigned long value used to locate this object in the
1142 * Add an inode to the inode hash for this superblock.
1144 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1146 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1147 spin_lock(&inode_lock
);
1148 hlist_add_head(&inode
->i_hash
, head
);
1149 spin_unlock(&inode_lock
);
1151 EXPORT_SYMBOL(__insert_inode_hash
);
1154 * remove_inode_hash - remove an inode from the hash
1155 * @inode: inode to unhash
1157 * Remove an inode from the superblock.
1159 void remove_inode_hash(struct inode
*inode
)
1161 spin_lock(&inode_lock
);
1162 hlist_del_init(&inode
->i_hash
);
1163 spin_unlock(&inode_lock
);
1165 EXPORT_SYMBOL(remove_inode_hash
);
1168 * Tell the filesystem that this inode is no longer of any interest and should
1169 * be completely destroyed.
1171 * We leave the inode in the inode hash table until *after* the filesystem's
1172 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1173 * instigate) will always find up-to-date information either in the hash or on
1176 * I_FREEING is set so that no-one will take a new reference to the inode while
1177 * it is being deleted.
1179 void generic_delete_inode(struct inode
*inode
)
1181 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1183 list_del_init(&inode
->i_list
);
1184 list_del_init(&inode
->i_sb_list
);
1185 WARN_ON(inode
->i_state
& I_NEW
);
1186 inode
->i_state
|= I_FREEING
;
1187 inodes_stat
.nr_inodes
--;
1188 spin_unlock(&inode_lock
);
1190 security_inode_delete(inode
);
1192 if (op
->delete_inode
) {
1193 void (*delete)(struct inode
*) = op
->delete_inode
;
1194 if (!is_bad_inode(inode
))
1196 /* Filesystems implementing their own
1197 * s_op->delete_inode are required to call
1198 * truncate_inode_pages and clear_inode()
1202 truncate_inode_pages(&inode
->i_data
, 0);
1205 spin_lock(&inode_lock
);
1206 hlist_del_init(&inode
->i_hash
);
1207 spin_unlock(&inode_lock
);
1208 wake_up_inode(inode
);
1209 BUG_ON(inode
->i_state
!= I_CLEAR
);
1210 destroy_inode(inode
);
1212 EXPORT_SYMBOL(generic_delete_inode
);
1214 static void generic_forget_inode(struct inode
*inode
)
1216 struct super_block
*sb
= inode
->i_sb
;
1218 if (!hlist_unhashed(&inode
->i_hash
)) {
1219 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1220 list_move(&inode
->i_list
, &inode_unused
);
1221 inodes_stat
.nr_unused
++;
1222 if (sb
->s_flags
& MS_ACTIVE
) {
1223 spin_unlock(&inode_lock
);
1226 WARN_ON(inode
->i_state
& I_NEW
);
1227 inode
->i_state
|= I_WILL_FREE
;
1228 spin_unlock(&inode_lock
);
1229 write_inode_now(inode
, 1);
1230 spin_lock(&inode_lock
);
1231 WARN_ON(inode
->i_state
& I_NEW
);
1232 inode
->i_state
&= ~I_WILL_FREE
;
1233 inodes_stat
.nr_unused
--;
1234 hlist_del_init(&inode
->i_hash
);
1236 list_del_init(&inode
->i_list
);
1237 list_del_init(&inode
->i_sb_list
);
1238 WARN_ON(inode
->i_state
& I_NEW
);
1239 inode
->i_state
|= I_FREEING
;
1240 inodes_stat
.nr_inodes
--;
1241 spin_unlock(&inode_lock
);
1242 if (inode
->i_data
.nrpages
)
1243 truncate_inode_pages(&inode
->i_data
, 0);
1245 wake_up_inode(inode
);
1246 destroy_inode(inode
);
1250 * Normal UNIX filesystem behaviour: delete the
1251 * inode when the usage count drops to zero, and
1254 void generic_drop_inode(struct inode
*inode
)
1256 if (!inode
->i_nlink
)
1257 generic_delete_inode(inode
);
1259 generic_forget_inode(inode
);
1261 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1264 * Called when we're dropping the last reference
1267 * Call the FS "drop()" function, defaulting to
1268 * the legacy UNIX filesystem behaviour..
1270 * NOTE! NOTE! NOTE! We're called with the inode lock
1271 * held, and the drop function is supposed to release
1274 static inline void iput_final(struct inode
*inode
)
1276 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1277 void (*drop
)(struct inode
*) = generic_drop_inode
;
1279 if (op
&& op
->drop_inode
)
1280 drop
= op
->drop_inode
;
1285 * iput - put an inode
1286 * @inode: inode to put
1288 * Puts an inode, dropping its usage count. If the inode use count hits
1289 * zero, the inode is then freed and may also be destroyed.
1291 * Consequently, iput() can sleep.
1293 void iput(struct inode
*inode
)
1296 BUG_ON(inode
->i_state
== I_CLEAR
);
1298 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1302 EXPORT_SYMBOL(iput
);
1305 * bmap - find a block number in a file
1306 * @inode: inode of file
1307 * @block: block to find
1309 * Returns the block number on the device holding the inode that
1310 * is the disk block number for the block of the file requested.
1311 * That is, asked for block 4 of inode 1 the function will return the
1312 * disk block relative to the disk start that holds that block of the
1315 sector_t
bmap(struct inode
*inode
, sector_t block
)
1318 if (inode
->i_mapping
->a_ops
->bmap
)
1319 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1322 EXPORT_SYMBOL(bmap
);
1325 * With relative atime, only update atime if the previous atime is
1326 * earlier than either the ctime or mtime or if at least a day has
1327 * passed since the last atime update.
1329 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1330 struct timespec now
)
1333 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1336 * Is mtime younger than atime? If yes, update atime:
1338 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1341 * Is ctime younger than atime? If yes, update atime:
1343 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1347 * Is the previous atime value older than a day? If yes,
1350 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1353 * Good, we can skip the atime update:
1359 * touch_atime - update the access time
1360 * @mnt: mount the inode is accessed on
1361 * @dentry: dentry accessed
1363 * Update the accessed time on an inode and mark it for writeback.
1364 * This function automatically handles read only file systems and media,
1365 * as well as the "noatime" flag and inode specific "noatime" markers.
1367 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1369 struct inode
*inode
= dentry
->d_inode
;
1370 struct timespec now
;
1372 if (mnt_want_write(mnt
))
1374 if (inode
->i_flags
& S_NOATIME
)
1376 if (IS_NOATIME(inode
))
1378 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1381 if (mnt
->mnt_flags
& MNT_NOATIME
)
1383 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1386 now
= current_fs_time(inode
->i_sb
);
1388 if (!relatime_need_update(mnt
, inode
, now
))
1391 if (timespec_equal(&inode
->i_atime
, &now
))
1394 inode
->i_atime
= now
;
1395 mark_inode_dirty_sync(inode
);
1397 mnt_drop_write(mnt
);
1399 EXPORT_SYMBOL(touch_atime
);
1402 * file_update_time - update mtime and ctime time
1403 * @file: file accessed
1405 * Update the mtime and ctime members of an inode and mark the inode
1406 * for writeback. Note that this function is meant exclusively for
1407 * usage in the file write path of filesystems, and filesystems may
1408 * choose to explicitly ignore update via this function with the
1409 * S_NOCTIME inode flag, e.g. for network filesystem where these
1410 * timestamps are handled by the server.
1413 void file_update_time(struct file
*file
)
1415 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1416 struct timespec now
;
1420 if (IS_NOCMTIME(inode
))
1423 err
= mnt_want_write(file
->f_path
.mnt
);
1427 now
= current_fs_time(inode
->i_sb
);
1428 if (!timespec_equal(&inode
->i_mtime
, &now
)) {
1429 inode
->i_mtime
= now
;
1433 if (!timespec_equal(&inode
->i_ctime
, &now
)) {
1434 inode
->i_ctime
= now
;
1438 if (IS_I_VERSION(inode
)) {
1439 inode_inc_iversion(inode
);
1444 mark_inode_dirty_sync(inode
);
1445 mnt_drop_write(file
->f_path
.mnt
);
1447 EXPORT_SYMBOL(file_update_time
);
1449 int inode_needs_sync(struct inode
*inode
)
1453 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1457 EXPORT_SYMBOL(inode_needs_sync
);
1459 int inode_wait(void *word
)
1464 EXPORT_SYMBOL(inode_wait
);
1467 * If we try to find an inode in the inode hash while it is being
1468 * deleted, we have to wait until the filesystem completes its
1469 * deletion before reporting that it isn't found. This function waits
1470 * until the deletion _might_ have completed. Callers are responsible
1471 * to recheck inode state.
1473 * It doesn't matter if I_LOCK is not set initially, a call to
1474 * wake_up_inode() after removing from the hash list will DTRT.
1476 * This is called with inode_lock held.
1478 static void __wait_on_freeing_inode(struct inode
*inode
)
1480 wait_queue_head_t
*wq
;
1481 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_LOCK
);
1482 wq
= bit_waitqueue(&inode
->i_state
, __I_LOCK
);
1483 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1484 spin_unlock(&inode_lock
);
1486 finish_wait(wq
, &wait
.wait
);
1487 spin_lock(&inode_lock
);
1490 static __initdata
unsigned long ihash_entries
;
1491 static int __init
set_ihash_entries(char *str
)
1495 ihash_entries
= simple_strtoul(str
, &str
, 0);
1498 __setup("ihash_entries=", set_ihash_entries
);
1501 * Initialize the waitqueues and inode hash table.
1503 void __init
inode_init_early(void)
1507 /* If hashes are distributed across NUMA nodes, defer
1508 * hash allocation until vmalloc space is available.
1514 alloc_large_system_hash("Inode-cache",
1515 sizeof(struct hlist_head
),
1523 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1524 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1527 void __init
inode_init(void)
1531 /* inode slab cache */
1532 inode_cachep
= kmem_cache_create("inode_cache",
1533 sizeof(struct inode
),
1535 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1538 register_shrinker(&icache_shrinker
);
1540 /* Hash may have been set up in inode_init_early */
1545 alloc_large_system_hash("Inode-cache",
1546 sizeof(struct hlist_head
),
1554 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1555 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1558 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1560 inode
->i_mode
= mode
;
1561 if (S_ISCHR(mode
)) {
1562 inode
->i_fop
= &def_chr_fops
;
1563 inode
->i_rdev
= rdev
;
1564 } else if (S_ISBLK(mode
)) {
1565 inode
->i_fop
= &def_blk_fops
;
1566 inode
->i_rdev
= rdev
;
1567 } else if (S_ISFIFO(mode
))
1568 inode
->i_fop
= &def_fifo_fops
;
1569 else if (S_ISSOCK(mode
))
1570 inode
->i_fop
= &bad_sock_fops
;
1572 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o)\n",
1575 EXPORT_SYMBOL(init_special_inode
);