4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/slab.h>
12 #include <linux/writeback.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/wait.h>
16 #include <linux/rwsem.h>
17 #include <linux/hash.h>
18 #include <linux/swap.h>
19 #include <linux/security.h>
20 #include <linux/pagemap.h>
21 #include <linux/cdev.h>
22 #include <linux/bootmem.h>
23 #include <linux/inotify.h>
24 #include <linux/fsnotify.h>
25 #include <linux/mount.h>
26 #include <linux/async.h>
27 #include <linux/posix_acl.h>
30 * This is needed for the following functions:
32 * - invalidate_inode_buffers
35 * FIXME: remove all knowledge of the buffer layer from this file
37 #include <linux/buffer_head.h>
40 * New inode.c implementation.
42 * This implementation has the basic premise of trying
43 * to be extremely low-overhead and SMP-safe, yet be
44 * simple enough to be "obviously correct".
49 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
51 /* #define INODE_PARANOIA 1 */
52 /* #define INODE_DEBUG 1 */
55 * Inode lookup is no longer as critical as it used to be:
56 * most of the lookups are going to be through the dcache.
58 #define I_HASHBITS i_hash_shift
59 #define I_HASHMASK i_hash_mask
61 static unsigned int i_hash_mask __read_mostly
;
62 static unsigned int i_hash_shift __read_mostly
;
65 * Each inode can be on two separate lists. One is
66 * the hash list of the inode, used for lookups. The
67 * other linked list is the "type" list:
68 * "in_use" - valid inode, i_count > 0, i_nlink > 0
69 * "dirty" - as "in_use" but also dirty
70 * "unused" - valid inode, i_count = 0
72 * A "dirty" list is maintained for each super block,
73 * allowing for low-overhead inode sync() operations.
76 LIST_HEAD(inode_in_use
);
77 LIST_HEAD(inode_unused
);
78 static struct hlist_head
*inode_hashtable __read_mostly
;
81 * A simple spinlock to protect the list manipulations.
83 * NOTE! You also have to own the lock if you change
84 * the i_state of an inode while it is in use..
86 DEFINE_SPINLOCK(inode_lock
);
89 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
90 * icache shrinking path, and the umount path. Without this exclusion,
91 * by the time prune_icache calls iput for the inode whose pages it has
92 * been invalidating, or by the time it calls clear_inode & destroy_inode
93 * from its final dispose_list, the struct super_block they refer to
94 * (for inode->i_sb->s_op) may already have been freed and reused.
96 * We make this an rwsem because the fastpath is icache shrinking. In
97 * some cases a filesystem may be doing a significant amount of work in
98 * its inode reclaim code, so this should improve parallelism.
100 static DECLARE_RWSEM(iprune_sem
);
103 * Statistics gathering..
105 struct inodes_stat_t inodes_stat
;
107 static struct kmem_cache
*inode_cachep __read_mostly
;
109 static void wake_up_inode(struct inode
*inode
)
112 * Prevent speculative execution through spin_unlock(&inode_lock);
115 wake_up_bit(&inode
->i_state
, __I_NEW
);
119 * inode_init_always - perform inode structure intialisation
120 * @sb: superblock inode belongs to
121 * @inode: inode to initialise
123 * These are initializations that need to be done on every inode
124 * allocation as the fields are not initialised by slab allocation.
126 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
128 static const struct address_space_operations empty_aops
;
129 static const struct inode_operations empty_iops
;
130 static const struct file_operations empty_fops
;
131 struct address_space
*const mapping
= &inode
->i_data
;
134 inode
->i_blkbits
= sb
->s_blocksize_bits
;
136 atomic_set(&inode
->i_count
, 1);
137 inode
->i_op
= &empty_iops
;
138 inode
->i_fop
= &empty_fops
;
142 atomic_set(&inode
->i_writecount
, 0);
146 inode
->i_generation
= 0;
148 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
150 inode
->i_pipe
= NULL
;
151 inode
->i_bdev
= NULL
;
152 inode
->i_cdev
= NULL
;
154 inode
->dirtied_when
= 0;
156 if (security_inode_alloc(inode
))
158 spin_lock_init(&inode
->i_lock
);
159 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
161 mutex_init(&inode
->i_mutex
);
162 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
164 init_rwsem(&inode
->i_alloc_sem
);
165 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
167 mapping
->a_ops
= &empty_aops
;
168 mapping
->host
= inode
;
170 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
171 mapping
->assoc_mapping
= NULL
;
172 mapping
->backing_dev_info
= &default_backing_dev_info
;
173 mapping
->writeback_index
= 0;
176 * If the block_device provides a backing_dev_info for client
177 * inodes then use that. Otherwise the inode share the bdev's
181 struct backing_dev_info
*bdi
;
183 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
184 mapping
->backing_dev_info
= bdi
;
186 inode
->i_private
= NULL
;
187 inode
->i_mapping
= mapping
;
188 #ifdef CONFIG_FS_POSIX_ACL
189 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
192 #ifdef CONFIG_FSNOTIFY
193 inode
->i_fsnotify_mask
= 0;
200 EXPORT_SYMBOL(inode_init_always
);
202 static struct inode
*alloc_inode(struct super_block
*sb
)
206 if (sb
->s_op
->alloc_inode
)
207 inode
= sb
->s_op
->alloc_inode(sb
);
209 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
214 if (unlikely(inode_init_always(sb
, inode
))) {
215 if (inode
->i_sb
->s_op
->destroy_inode
)
216 inode
->i_sb
->s_op
->destroy_inode(inode
);
218 kmem_cache_free(inode_cachep
, inode
);
225 void __destroy_inode(struct inode
*inode
)
227 BUG_ON(inode_has_buffers(inode
));
228 security_inode_free(inode
);
229 fsnotify_inode_delete(inode
);
230 #ifdef CONFIG_FS_POSIX_ACL
231 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
232 posix_acl_release(inode
->i_acl
);
233 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
234 posix_acl_release(inode
->i_default_acl
);
237 EXPORT_SYMBOL(__destroy_inode
);
239 void destroy_inode(struct inode
*inode
)
241 __destroy_inode(inode
);
242 if (inode
->i_sb
->s_op
->destroy_inode
)
243 inode
->i_sb
->s_op
->destroy_inode(inode
);
245 kmem_cache_free(inode_cachep
, (inode
));
249 * These are initializations that only need to be done
250 * once, because the fields are idempotent across use
251 * of the inode, so let the slab aware of that.
253 void inode_init_once(struct inode
*inode
)
255 memset(inode
, 0, sizeof(*inode
));
256 INIT_HLIST_NODE(&inode
->i_hash
);
257 INIT_LIST_HEAD(&inode
->i_dentry
);
258 INIT_LIST_HEAD(&inode
->i_devices
);
259 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
260 spin_lock_init(&inode
->i_data
.tree_lock
);
261 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
262 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
263 spin_lock_init(&inode
->i_data
.private_lock
);
264 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
265 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
266 i_size_ordered_init(inode
);
267 #ifdef CONFIG_INOTIFY
268 INIT_LIST_HEAD(&inode
->inotify_watches
);
269 mutex_init(&inode
->inotify_mutex
);
271 #ifdef CONFIG_FSNOTIFY
272 INIT_HLIST_HEAD(&inode
->i_fsnotify_mark_entries
);
275 EXPORT_SYMBOL(inode_init_once
);
277 static void init_once(void *foo
)
279 struct inode
*inode
= (struct inode
*) foo
;
281 inode_init_once(inode
);
285 * inode_lock must be held
287 void __iget(struct inode
*inode
)
289 if (atomic_inc_return(&inode
->i_count
) != 1)
292 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
293 list_move(&inode
->i_list
, &inode_in_use
);
294 inodes_stat
.nr_unused
--;
297 void end_writeback(struct inode
*inode
)
300 BUG_ON(inode
->i_data
.nrpages
);
301 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
302 BUG_ON(!(inode
->i_state
& I_FREEING
));
303 BUG_ON(inode
->i_state
& I_CLEAR
);
304 inode_sync_wait(inode
);
305 inode
->i_state
= I_FREEING
| I_CLEAR
;
307 EXPORT_SYMBOL(end_writeback
);
310 * clear_inode - clear an inode
311 * @inode: inode to clear
313 * This is called by the filesystem to tell us
314 * that the inode is no longer useful. We just
315 * terminate it with extreme prejudice.
317 void clear_inode(struct inode
*inode
)
320 invalidate_inode_buffers(inode
);
322 BUG_ON(inode
->i_data
.nrpages
);
323 BUG_ON(!(inode
->i_state
& I_FREEING
));
324 BUG_ON(inode
->i_state
& I_CLEAR
);
325 inode_sync_wait(inode
);
326 if (inode
->i_sb
->s_op
->clear_inode
)
327 inode
->i_sb
->s_op
->clear_inode(inode
);
328 inode
->i_state
= I_FREEING
| I_CLEAR
;
330 EXPORT_SYMBOL(clear_inode
);
332 static void evict(struct inode
*inode
, int delete)
334 const struct super_operations
*op
= inode
->i_sb
->s_op
;
336 if (op
->evict_inode
) {
337 op
->evict_inode(inode
);
338 } else if (delete && op
->delete_inode
) {
339 op
->delete_inode(inode
);
341 if (inode
->i_data
.nrpages
)
342 truncate_inode_pages(&inode
->i_data
, 0);
345 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
347 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
352 * dispose_list - dispose of the contents of a local list
353 * @head: the head of the list to free
355 * Dispose-list gets a local list with local inodes in it, so it doesn't
356 * need to worry about list corruption and SMP locks.
358 static void dispose_list(struct list_head
*head
)
362 while (!list_empty(head
)) {
365 inode
= list_first_entry(head
, struct inode
, i_list
);
366 list_del(&inode
->i_list
);
370 spin_lock(&inode_lock
);
371 hlist_del_init(&inode
->i_hash
);
372 list_del_init(&inode
->i_sb_list
);
373 spin_unlock(&inode_lock
);
375 wake_up_inode(inode
);
376 destroy_inode(inode
);
379 spin_lock(&inode_lock
);
380 inodes_stat
.nr_inodes
-= nr_disposed
;
381 spin_unlock(&inode_lock
);
385 * Invalidate all inodes for a device.
387 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
389 struct list_head
*next
;
390 int busy
= 0, count
= 0;
394 struct list_head
*tmp
= next
;
398 * We can reschedule here without worrying about the list's
399 * consistency because the per-sb list of inodes must not
400 * change during umount anymore, and because iprune_sem keeps
401 * shrink_icache_memory() away.
403 cond_resched_lock(&inode_lock
);
408 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
409 if (inode
->i_state
& I_NEW
)
411 invalidate_inode_buffers(inode
);
412 if (!atomic_read(&inode
->i_count
)) {
413 list_move(&inode
->i_list
, dispose
);
414 WARN_ON(inode
->i_state
& I_NEW
);
415 inode
->i_state
|= I_FREEING
;
421 /* only unused inodes may be cached with i_count zero */
422 inodes_stat
.nr_unused
-= count
;
427 * invalidate_inodes - discard the inodes on a device
430 * Discard all of the inodes for a given superblock. If the discard
431 * fails because there are busy inodes then a non zero value is returned.
432 * If the discard is successful all the inodes have been discarded.
434 int invalidate_inodes(struct super_block
*sb
)
437 LIST_HEAD(throw_away
);
439 down_write(&iprune_sem
);
440 spin_lock(&inode_lock
);
441 inotify_unmount_inodes(&sb
->s_inodes
);
442 fsnotify_unmount_inodes(&sb
->s_inodes
);
443 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
444 spin_unlock(&inode_lock
);
446 dispose_list(&throw_away
);
447 up_write(&iprune_sem
);
451 EXPORT_SYMBOL(invalidate_inodes
);
453 static int can_unuse(struct inode
*inode
)
457 if (inode_has_buffers(inode
))
459 if (atomic_read(&inode
->i_count
))
461 if (inode
->i_data
.nrpages
)
467 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
468 * a temporary list and then are freed outside inode_lock by dispose_list().
470 * Any inodes which are pinned purely because of attached pagecache have their
471 * pagecache removed. We expect the final iput() on that inode to add it to
472 * the front of the inode_unused list. So look for it there and if the
473 * inode is still freeable, proceed. The right inode is found 99.9% of the
474 * time in testing on a 4-way.
476 * If the inode has metadata buffers attached to mapping->private_list then
477 * try to remove them.
479 static void prune_icache(int nr_to_scan
)
484 unsigned long reap
= 0;
486 down_read(&iprune_sem
);
487 spin_lock(&inode_lock
);
488 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
491 if (list_empty(&inode_unused
))
494 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
496 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
497 list_move(&inode
->i_list
, &inode_unused
);
500 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
502 spin_unlock(&inode_lock
);
503 if (remove_inode_buffers(inode
))
504 reap
+= invalidate_mapping_pages(&inode
->i_data
,
507 spin_lock(&inode_lock
);
509 if (inode
!= list_entry(inode_unused
.next
,
510 struct inode
, i_list
))
511 continue; /* wrong inode or list_empty */
512 if (!can_unuse(inode
))
515 list_move(&inode
->i_list
, &freeable
);
516 WARN_ON(inode
->i_state
& I_NEW
);
517 inode
->i_state
|= I_FREEING
;
520 inodes_stat
.nr_unused
-= nr_pruned
;
521 if (current_is_kswapd())
522 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
524 __count_vm_events(PGINODESTEAL
, reap
);
525 spin_unlock(&inode_lock
);
527 dispose_list(&freeable
);
528 up_read(&iprune_sem
);
532 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
533 * "unused" means that no dentries are referring to the inodes: the files are
534 * not open and the dcache references to those inodes have already been
537 * This function is passed the number of inodes to scan, and it returns the
538 * total number of remaining possibly-reclaimable inodes.
540 static int shrink_icache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
544 * Nasty deadlock avoidance. We may hold various FS locks,
545 * and we don't want to recurse into the FS that called us
546 * in clear_inode() and friends..
548 if (!(gfp_mask
& __GFP_FS
))
552 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
555 static struct shrinker icache_shrinker
= {
556 .shrink
= shrink_icache_memory
,
557 .seeks
= DEFAULT_SEEKS
,
560 static void __wait_on_freeing_inode(struct inode
*inode
);
562 * Called with the inode lock held.
563 * NOTE: we are not increasing the inode-refcount, you must call __iget()
564 * by hand after calling find_inode now! This simplifies iunique and won't
565 * add any additional branch in the common code.
567 static struct inode
*find_inode(struct super_block
*sb
,
568 struct hlist_head
*head
,
569 int (*test
)(struct inode
*, void *),
572 struct hlist_node
*node
;
573 struct inode
*inode
= NULL
;
576 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
577 if (inode
->i_sb
!= sb
)
579 if (!test(inode
, data
))
581 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
582 __wait_on_freeing_inode(inode
);
587 return node
? inode
: NULL
;
591 * find_inode_fast is the fast path version of find_inode, see the comment at
592 * iget_locked for details.
594 static struct inode
*find_inode_fast(struct super_block
*sb
,
595 struct hlist_head
*head
, unsigned long ino
)
597 struct hlist_node
*node
;
598 struct inode
*inode
= NULL
;
601 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
602 if (inode
->i_ino
!= ino
)
604 if (inode
->i_sb
!= sb
)
606 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
607 __wait_on_freeing_inode(inode
);
612 return node
? inode
: NULL
;
615 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
619 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
621 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
622 return tmp
& I_HASHMASK
;
626 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
629 inodes_stat
.nr_inodes
++;
630 list_add(&inode
->i_list
, &inode_in_use
);
631 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
633 hlist_add_head(&inode
->i_hash
, head
);
637 * inode_add_to_lists - add a new inode to relevant lists
638 * @sb: superblock inode belongs to
639 * @inode: inode to mark in use
641 * When an inode is allocated it needs to be accounted for, added to the in use
642 * list, the owning superblock and the inode hash. This needs to be done under
643 * the inode_lock, so export a function to do this rather than the inode lock
644 * itself. We calculate the hash list to add to here so it is all internal
645 * which requires the caller to have already set up the inode number in the
648 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
650 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
652 spin_lock(&inode_lock
);
653 __inode_add_to_lists(sb
, head
, inode
);
654 spin_unlock(&inode_lock
);
656 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
659 * new_inode - obtain an inode
662 * Allocates a new inode for given superblock. The default gfp_mask
663 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
664 * If HIGHMEM pages are unsuitable or it is known that pages allocated
665 * for the page cache are not reclaimable or migratable,
666 * mapping_set_gfp_mask() must be called with suitable flags on the
667 * newly created inode's mapping
670 struct inode
*new_inode(struct super_block
*sb
)
673 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
674 * error if st_ino won't fit in target struct field. Use 32bit counter
675 * here to attempt to avoid that.
677 static unsigned int last_ino
;
680 spin_lock_prefetch(&inode_lock
);
682 inode
= alloc_inode(sb
);
684 spin_lock(&inode_lock
);
685 __inode_add_to_lists(sb
, NULL
, inode
);
686 inode
->i_ino
= ++last_ino
;
688 spin_unlock(&inode_lock
);
692 EXPORT_SYMBOL(new_inode
);
694 void unlock_new_inode(struct inode
*inode
)
696 #ifdef CONFIG_DEBUG_LOCK_ALLOC
697 if (inode
->i_mode
& S_IFDIR
) {
698 struct file_system_type
*type
= inode
->i_sb
->s_type
;
700 /* Set new key only if filesystem hasn't already changed it */
701 if (!lockdep_match_class(&inode
->i_mutex
,
702 &type
->i_mutex_key
)) {
704 * ensure nobody is actually holding i_mutex
706 mutex_destroy(&inode
->i_mutex
);
707 mutex_init(&inode
->i_mutex
);
708 lockdep_set_class(&inode
->i_mutex
,
709 &type
->i_mutex_dir_key
);
714 * This is special! We do not need the spinlock when clearing I_NEW,
715 * because we're guaranteed that nobody else tries to do anything about
716 * the state of the inode when it is locked, as we just created it (so
717 * there can be no old holders that haven't tested I_NEW).
718 * However we must emit the memory barrier so that other CPUs reliably
719 * see the clearing of I_NEW after the other inode initialisation has
723 WARN_ON(!(inode
->i_state
& I_NEW
));
724 inode
->i_state
&= ~I_NEW
;
725 wake_up_inode(inode
);
727 EXPORT_SYMBOL(unlock_new_inode
);
730 * This is called without the inode lock held.. Be careful.
732 * We no longer cache the sb_flags in i_flags - see fs.h
733 * -- rmk@arm.uk.linux.org
735 static struct inode
*get_new_inode(struct super_block
*sb
,
736 struct hlist_head
*head
,
737 int (*test
)(struct inode
*, void *),
738 int (*set
)(struct inode
*, void *),
743 inode
= alloc_inode(sb
);
747 spin_lock(&inode_lock
);
748 /* We released the lock, so.. */
749 old
= find_inode(sb
, head
, test
, data
);
751 if (set(inode
, data
))
754 __inode_add_to_lists(sb
, head
, inode
);
755 inode
->i_state
= I_NEW
;
756 spin_unlock(&inode_lock
);
758 /* Return the locked inode with I_NEW set, the
759 * caller is responsible for filling in the contents
765 * Uhhuh, somebody else created the same inode under
766 * us. Use the old inode instead of the one we just
770 spin_unlock(&inode_lock
);
771 destroy_inode(inode
);
773 wait_on_inode(inode
);
778 spin_unlock(&inode_lock
);
779 destroy_inode(inode
);
784 * get_new_inode_fast is the fast path version of get_new_inode, see the
785 * comment at iget_locked for details.
787 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
788 struct hlist_head
*head
, unsigned long ino
)
792 inode
= alloc_inode(sb
);
796 spin_lock(&inode_lock
);
797 /* We released the lock, so.. */
798 old
= find_inode_fast(sb
, head
, ino
);
801 __inode_add_to_lists(sb
, head
, inode
);
802 inode
->i_state
= I_NEW
;
803 spin_unlock(&inode_lock
);
805 /* Return the locked inode with I_NEW set, the
806 * caller is responsible for filling in the contents
812 * Uhhuh, somebody else created the same inode under
813 * us. Use the old inode instead of the one we just
817 spin_unlock(&inode_lock
);
818 destroy_inode(inode
);
820 wait_on_inode(inode
);
826 * iunique - get a unique inode number
828 * @max_reserved: highest reserved inode number
830 * Obtain an inode number that is unique on the system for a given
831 * superblock. This is used by file systems that have no natural
832 * permanent inode numbering system. An inode number is returned that
833 * is higher than the reserved limit but unique.
836 * With a large number of inodes live on the file system this function
837 * currently becomes quite slow.
839 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
842 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
843 * error if st_ino won't fit in target struct field. Use 32bit counter
844 * here to attempt to avoid that.
846 static unsigned int counter
;
848 struct hlist_head
*head
;
851 spin_lock(&inode_lock
);
853 if (counter
<= max_reserved
)
854 counter
= max_reserved
+ 1;
856 head
= inode_hashtable
+ hash(sb
, res
);
857 inode
= find_inode_fast(sb
, head
, res
);
858 } while (inode
!= NULL
);
859 spin_unlock(&inode_lock
);
863 EXPORT_SYMBOL(iunique
);
865 struct inode
*igrab(struct inode
*inode
)
867 spin_lock(&inode_lock
);
868 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
)))
872 * Handle the case where s_op->clear_inode is not been
873 * called yet, and somebody is calling igrab
874 * while the inode is getting freed.
877 spin_unlock(&inode_lock
);
880 EXPORT_SYMBOL(igrab
);
883 * ifind - internal function, you want ilookup5() or iget5().
884 * @sb: super block of file system to search
885 * @head: the head of the list to search
886 * @test: callback used for comparisons between inodes
887 * @data: opaque data pointer to pass to @test
888 * @wait: if true wait for the inode to be unlocked, if false do not
890 * ifind() searches for the inode specified by @data in the inode
891 * cache. This is a generalized version of ifind_fast() for file systems where
892 * the inode number is not sufficient for unique identification of an inode.
894 * If the inode is in the cache, the inode is returned with an incremented
897 * Otherwise NULL is returned.
899 * Note, @test is called with the inode_lock held, so can't sleep.
901 static struct inode
*ifind(struct super_block
*sb
,
902 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
903 void *data
, const int wait
)
907 spin_lock(&inode_lock
);
908 inode
= find_inode(sb
, head
, test
, data
);
911 spin_unlock(&inode_lock
);
913 wait_on_inode(inode
);
916 spin_unlock(&inode_lock
);
921 * ifind_fast - internal function, you want ilookup() or iget().
922 * @sb: super block of file system to search
923 * @head: head of the list to search
924 * @ino: inode number to search for
926 * ifind_fast() searches for the inode @ino in the inode cache. This is for
927 * file systems where the inode number is sufficient for unique identification
930 * If the inode is in the cache, the inode is returned with an incremented
933 * Otherwise NULL is returned.
935 static struct inode
*ifind_fast(struct super_block
*sb
,
936 struct hlist_head
*head
, unsigned long ino
)
940 spin_lock(&inode_lock
);
941 inode
= find_inode_fast(sb
, head
, ino
);
944 spin_unlock(&inode_lock
);
945 wait_on_inode(inode
);
948 spin_unlock(&inode_lock
);
953 * ilookup5_nowait - search for an inode in the inode cache
954 * @sb: super block of file system to search
955 * @hashval: hash value (usually inode number) to search for
956 * @test: callback used for comparisons between inodes
957 * @data: opaque data pointer to pass to @test
959 * ilookup5() uses ifind() to search for the inode specified by @hashval and
960 * @data in the inode cache. This is a generalized version of ilookup() for
961 * file systems where the inode number is not sufficient for unique
962 * identification of an inode.
964 * If the inode is in the cache, the inode is returned with an incremented
965 * reference count. Note, the inode lock is not waited upon so you have to be
966 * very careful what you do with the returned inode. You probably should be
967 * using ilookup5() instead.
969 * Otherwise NULL is returned.
971 * Note, @test is called with the inode_lock held, so can't sleep.
973 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
974 int (*test
)(struct inode
*, void *), void *data
)
976 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
978 return ifind(sb
, head
, test
, data
, 0);
980 EXPORT_SYMBOL(ilookup5_nowait
);
983 * ilookup5 - search for an inode in the inode cache
984 * @sb: super block of file system to search
985 * @hashval: hash value (usually inode number) to search for
986 * @test: callback used for comparisons between inodes
987 * @data: opaque data pointer to pass to @test
989 * ilookup5() uses ifind() to search for the inode specified by @hashval and
990 * @data in the inode cache. This is a generalized version of ilookup() for
991 * file systems where the inode number is not sufficient for unique
992 * identification of an inode.
994 * If the inode is in the cache, the inode lock is waited upon and the inode is
995 * returned with an incremented reference count.
997 * Otherwise NULL is returned.
999 * Note, @test is called with the inode_lock held, so can't sleep.
1001 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1002 int (*test
)(struct inode
*, void *), void *data
)
1004 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1006 return ifind(sb
, head
, test
, data
, 1);
1008 EXPORT_SYMBOL(ilookup5
);
1011 * ilookup - search for an inode in the inode cache
1012 * @sb: super block of file system to search
1013 * @ino: inode number to search for
1015 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1016 * This is for file systems where the inode number is sufficient for unique
1017 * identification of an inode.
1019 * If the inode is in the cache, the inode is returned with an incremented
1022 * Otherwise NULL is returned.
1024 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1026 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1028 return ifind_fast(sb
, head
, ino
);
1030 EXPORT_SYMBOL(ilookup
);
1033 * iget5_locked - obtain an inode from a mounted file system
1034 * @sb: super block of file system
1035 * @hashval: hash value (usually inode number) to get
1036 * @test: callback used for comparisons between inodes
1037 * @set: callback used to initialize a new struct inode
1038 * @data: opaque data pointer to pass to @test and @set
1040 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1041 * and @data in the inode cache and if present it is returned with an increased
1042 * reference count. This is a generalized version of iget_locked() for file
1043 * systems where the inode number is not sufficient for unique identification
1046 * If the inode is not in cache, get_new_inode() is called to allocate a new
1047 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1048 * file system gets to fill it in before unlocking it via unlock_new_inode().
1050 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1052 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1053 int (*test
)(struct inode
*, void *),
1054 int (*set
)(struct inode
*, void *), void *data
)
1056 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1057 struct inode
*inode
;
1059 inode
= ifind(sb
, head
, test
, data
, 1);
1063 * get_new_inode() will do the right thing, re-trying the search
1064 * in case it had to block at any point.
1066 return get_new_inode(sb
, head
, test
, set
, data
);
1068 EXPORT_SYMBOL(iget5_locked
);
1071 * iget_locked - obtain an inode from a mounted file system
1072 * @sb: super block of file system
1073 * @ino: inode number to get
1075 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1076 * the inode cache and if present it is returned with an increased reference
1077 * count. This is for file systems where the inode number is sufficient for
1078 * unique identification of an inode.
1080 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1081 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1082 * The file system gets to fill it in before unlocking it via
1083 * unlock_new_inode().
1085 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1087 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1088 struct inode
*inode
;
1090 inode
= ifind_fast(sb
, head
, ino
);
1094 * get_new_inode_fast() will do the right thing, re-trying the search
1095 * in case it had to block at any point.
1097 return get_new_inode_fast(sb
, head
, ino
);
1099 EXPORT_SYMBOL(iget_locked
);
1101 int insert_inode_locked(struct inode
*inode
)
1103 struct super_block
*sb
= inode
->i_sb
;
1104 ino_t ino
= inode
->i_ino
;
1105 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1107 inode
->i_state
|= I_NEW
;
1109 struct hlist_node
*node
;
1110 struct inode
*old
= NULL
;
1111 spin_lock(&inode_lock
);
1112 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1113 if (old
->i_ino
!= ino
)
1115 if (old
->i_sb
!= sb
)
1117 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
))
1121 if (likely(!node
)) {
1122 hlist_add_head(&inode
->i_hash
, head
);
1123 spin_unlock(&inode_lock
);
1127 spin_unlock(&inode_lock
);
1129 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1136 EXPORT_SYMBOL(insert_inode_locked
);
1138 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1139 int (*test
)(struct inode
*, void *), void *data
)
1141 struct super_block
*sb
= inode
->i_sb
;
1142 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1144 inode
->i_state
|= I_NEW
;
1147 struct hlist_node
*node
;
1148 struct inode
*old
= NULL
;
1150 spin_lock(&inode_lock
);
1151 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1152 if (old
->i_sb
!= sb
)
1154 if (!test(old
, data
))
1156 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
))
1160 if (likely(!node
)) {
1161 hlist_add_head(&inode
->i_hash
, head
);
1162 spin_unlock(&inode_lock
);
1166 spin_unlock(&inode_lock
);
1168 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1175 EXPORT_SYMBOL(insert_inode_locked4
);
1178 * __insert_inode_hash - hash an inode
1179 * @inode: unhashed inode
1180 * @hashval: unsigned long value used to locate this object in the
1183 * Add an inode to the inode hash for this superblock.
1185 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1187 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1188 spin_lock(&inode_lock
);
1189 hlist_add_head(&inode
->i_hash
, head
);
1190 spin_unlock(&inode_lock
);
1192 EXPORT_SYMBOL(__insert_inode_hash
);
1195 * remove_inode_hash - remove an inode from the hash
1196 * @inode: inode to unhash
1198 * Remove an inode from the superblock.
1200 void remove_inode_hash(struct inode
*inode
)
1202 spin_lock(&inode_lock
);
1203 hlist_del_init(&inode
->i_hash
);
1204 spin_unlock(&inode_lock
);
1206 EXPORT_SYMBOL(remove_inode_hash
);
1209 * Tell the filesystem that this inode is no longer of any interest and should
1210 * be completely destroyed.
1212 * We leave the inode in the inode hash table until *after* the filesystem's
1213 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1214 * instigate) will always find up-to-date information either in the hash or on
1217 * I_FREEING is set so that no-one will take a new reference to the inode while
1218 * it is being deleted.
1220 void generic_delete_inode(struct inode
*inode
)
1222 list_del_init(&inode
->i_list
);
1223 list_del_init(&inode
->i_sb_list
);
1224 WARN_ON(inode
->i_state
& I_NEW
);
1225 inode
->i_state
|= I_FREEING
;
1226 inodes_stat
.nr_inodes
--;
1227 spin_unlock(&inode_lock
);
1231 spin_lock(&inode_lock
);
1232 hlist_del_init(&inode
->i_hash
);
1233 spin_unlock(&inode_lock
);
1234 wake_up_inode(inode
);
1235 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
1236 destroy_inode(inode
);
1238 EXPORT_SYMBOL(generic_delete_inode
);
1241 * generic_detach_inode - remove inode from inode lists
1242 * @inode: inode to remove
1244 * Remove inode from inode lists, write it if it's dirty. This is just an
1245 * internal VFS helper exported for hugetlbfs. Do not use!
1247 * Returns 1 if inode should be completely destroyed.
1249 static int generic_detach_inode(struct inode
*inode
)
1251 struct super_block
*sb
= inode
->i_sb
;
1253 if (!hlist_unhashed(&inode
->i_hash
)) {
1254 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1255 list_move(&inode
->i_list
, &inode_unused
);
1256 inodes_stat
.nr_unused
++;
1257 if (sb
->s_flags
& MS_ACTIVE
) {
1258 spin_unlock(&inode_lock
);
1261 WARN_ON(inode
->i_state
& I_NEW
);
1262 inode
->i_state
|= I_WILL_FREE
;
1263 spin_unlock(&inode_lock
);
1264 write_inode_now(inode
, 1);
1265 spin_lock(&inode_lock
);
1266 WARN_ON(inode
->i_state
& I_NEW
);
1267 inode
->i_state
&= ~I_WILL_FREE
;
1268 inodes_stat
.nr_unused
--;
1269 hlist_del_init(&inode
->i_hash
);
1271 list_del_init(&inode
->i_list
);
1272 list_del_init(&inode
->i_sb_list
);
1273 WARN_ON(inode
->i_state
& I_NEW
);
1274 inode
->i_state
|= I_FREEING
;
1275 inodes_stat
.nr_inodes
--;
1276 spin_unlock(&inode_lock
);
1280 static void generic_forget_inode(struct inode
*inode
)
1282 if (!generic_detach_inode(inode
))
1285 wake_up_inode(inode
);
1286 destroy_inode(inode
);
1290 * Normal UNIX filesystem behaviour: delete the
1291 * inode when the usage count drops to zero, and
1294 void generic_drop_inode(struct inode
*inode
)
1296 if (!inode
->i_nlink
)
1297 generic_delete_inode(inode
);
1299 generic_forget_inode(inode
);
1301 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1304 * Called when we're dropping the last reference
1307 * Call the FS "drop()" function, defaulting to
1308 * the legacy UNIX filesystem behaviour..
1310 * NOTE! NOTE! NOTE! We're called with the inode lock
1311 * held, and the drop function is supposed to release
1314 static inline void iput_final(struct inode
*inode
)
1316 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1317 void (*drop
)(struct inode
*) = generic_drop_inode
;
1319 if (op
&& op
->drop_inode
)
1320 drop
= op
->drop_inode
;
1325 * iput - put an inode
1326 * @inode: inode to put
1328 * Puts an inode, dropping its usage count. If the inode use count hits
1329 * zero, the inode is then freed and may also be destroyed.
1331 * Consequently, iput() can sleep.
1333 void iput(struct inode
*inode
)
1336 BUG_ON(inode
->i_state
& I_CLEAR
);
1338 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1342 EXPORT_SYMBOL(iput
);
1345 * bmap - find a block number in a file
1346 * @inode: inode of file
1347 * @block: block to find
1349 * Returns the block number on the device holding the inode that
1350 * is the disk block number for the block of the file requested.
1351 * That is, asked for block 4 of inode 1 the function will return the
1352 * disk block relative to the disk start that holds that block of the
1355 sector_t
bmap(struct inode
*inode
, sector_t block
)
1358 if (inode
->i_mapping
->a_ops
->bmap
)
1359 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1362 EXPORT_SYMBOL(bmap
);
1365 * With relative atime, only update atime if the previous atime is
1366 * earlier than either the ctime or mtime or if at least a day has
1367 * passed since the last atime update.
1369 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1370 struct timespec now
)
1373 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1376 * Is mtime younger than atime? If yes, update atime:
1378 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1381 * Is ctime younger than atime? If yes, update atime:
1383 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1387 * Is the previous atime value older than a day? If yes,
1390 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1393 * Good, we can skip the atime update:
1399 * touch_atime - update the access time
1400 * @mnt: mount the inode is accessed on
1401 * @dentry: dentry accessed
1403 * Update the accessed time on an inode and mark it for writeback.
1404 * This function automatically handles read only file systems and media,
1405 * as well as the "noatime" flag and inode specific "noatime" markers.
1407 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1409 struct inode
*inode
= dentry
->d_inode
;
1410 struct timespec now
;
1412 if (inode
->i_flags
& S_NOATIME
)
1414 if (IS_NOATIME(inode
))
1416 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1419 if (mnt
->mnt_flags
& MNT_NOATIME
)
1421 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1424 now
= current_fs_time(inode
->i_sb
);
1426 if (!relatime_need_update(mnt
, inode
, now
))
1429 if (timespec_equal(&inode
->i_atime
, &now
))
1432 if (mnt_want_write(mnt
))
1435 inode
->i_atime
= now
;
1436 mark_inode_dirty_sync(inode
);
1437 mnt_drop_write(mnt
);
1439 EXPORT_SYMBOL(touch_atime
);
1442 * file_update_time - update mtime and ctime time
1443 * @file: file accessed
1445 * Update the mtime and ctime members of an inode and mark the inode
1446 * for writeback. Note that this function is meant exclusively for
1447 * usage in the file write path of filesystems, and filesystems may
1448 * choose to explicitly ignore update via this function with the
1449 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1450 * timestamps are handled by the server.
1453 void file_update_time(struct file
*file
)
1455 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1456 struct timespec now
;
1457 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1459 /* First try to exhaust all avenues to not sync */
1460 if (IS_NOCMTIME(inode
))
1463 now
= current_fs_time(inode
->i_sb
);
1464 if (!timespec_equal(&inode
->i_mtime
, &now
))
1467 if (!timespec_equal(&inode
->i_ctime
, &now
))
1470 if (IS_I_VERSION(inode
))
1471 sync_it
|= S_VERSION
;
1476 /* Finally allowed to write? Takes lock. */
1477 if (mnt_want_write_file(file
))
1480 /* Only change inode inside the lock region */
1481 if (sync_it
& S_VERSION
)
1482 inode_inc_iversion(inode
);
1483 if (sync_it
& S_CTIME
)
1484 inode
->i_ctime
= now
;
1485 if (sync_it
& S_MTIME
)
1486 inode
->i_mtime
= now
;
1487 mark_inode_dirty_sync(inode
);
1488 mnt_drop_write(file
->f_path
.mnt
);
1490 EXPORT_SYMBOL(file_update_time
);
1492 int inode_needs_sync(struct inode
*inode
)
1496 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1500 EXPORT_SYMBOL(inode_needs_sync
);
1502 int inode_wait(void *word
)
1507 EXPORT_SYMBOL(inode_wait
);
1510 * If we try to find an inode in the inode hash while it is being
1511 * deleted, we have to wait until the filesystem completes its
1512 * deletion before reporting that it isn't found. This function waits
1513 * until the deletion _might_ have completed. Callers are responsible
1514 * to recheck inode state.
1516 * It doesn't matter if I_NEW is not set initially, a call to
1517 * wake_up_inode() after removing from the hash list will DTRT.
1519 * This is called with inode_lock held.
1521 static void __wait_on_freeing_inode(struct inode
*inode
)
1523 wait_queue_head_t
*wq
;
1524 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1525 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1526 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1527 spin_unlock(&inode_lock
);
1529 finish_wait(wq
, &wait
.wait
);
1530 spin_lock(&inode_lock
);
1533 static __initdata
unsigned long ihash_entries
;
1534 static int __init
set_ihash_entries(char *str
)
1538 ihash_entries
= simple_strtoul(str
, &str
, 0);
1541 __setup("ihash_entries=", set_ihash_entries
);
1544 * Initialize the waitqueues and inode hash table.
1546 void __init
inode_init_early(void)
1550 /* If hashes are distributed across NUMA nodes, defer
1551 * hash allocation until vmalloc space is available.
1557 alloc_large_system_hash("Inode-cache",
1558 sizeof(struct hlist_head
),
1566 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1567 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1570 void __init
inode_init(void)
1574 /* inode slab cache */
1575 inode_cachep
= kmem_cache_create("inode_cache",
1576 sizeof(struct inode
),
1578 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1581 register_shrinker(&icache_shrinker
);
1583 /* Hash may have been set up in inode_init_early */
1588 alloc_large_system_hash("Inode-cache",
1589 sizeof(struct hlist_head
),
1597 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1598 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1601 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1603 inode
->i_mode
= mode
;
1604 if (S_ISCHR(mode
)) {
1605 inode
->i_fop
= &def_chr_fops
;
1606 inode
->i_rdev
= rdev
;
1607 } else if (S_ISBLK(mode
)) {
1608 inode
->i_fop
= &def_blk_fops
;
1609 inode
->i_rdev
= rdev
;
1610 } else if (S_ISFIFO(mode
))
1611 inode
->i_fop
= &def_fifo_fops
;
1612 else if (S_ISSOCK(mode
))
1613 inode
->i_fop
= &bad_sock_fops
;
1615 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1616 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1619 EXPORT_SYMBOL(init_special_inode
);
1622 * Init uid,gid,mode for new inode according to posix standards
1624 * @dir: Directory inode
1625 * @mode: mode of the new inode
1627 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1630 inode
->i_uid
= current_fsuid();
1631 if (dir
&& dir
->i_mode
& S_ISGID
) {
1632 inode
->i_gid
= dir
->i_gid
;
1636 inode
->i_gid
= current_fsgid();
1637 inode
->i_mode
= mode
;
1639 EXPORT_SYMBOL(inode_init_owner
);