4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/slab.h>
12 #include <linux/writeback.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/wait.h>
16 #include <linux/rwsem.h>
17 #include <linux/hash.h>
18 #include <linux/swap.h>
19 #include <linux/security.h>
20 #include <linux/pagemap.h>
21 #include <linux/cdev.h>
22 #include <linux/bootmem.h>
23 #include <linux/inotify.h>
24 #include <linux/fsnotify.h>
25 #include <linux/mount.h>
26 #include <linux/async.h>
27 #include <linux/posix_acl.h>
30 * This is needed for the following functions:
32 * - invalidate_inode_buffers
35 * FIXME: remove all knowledge of the buffer layer from this file
37 #include <linux/buffer_head.h>
40 * New inode.c implementation.
42 * This implementation has the basic premise of trying
43 * to be extremely low-overhead and SMP-safe, yet be
44 * simple enough to be "obviously correct".
49 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
51 /* #define INODE_PARANOIA 1 */
52 /* #define INODE_DEBUG 1 */
55 * Inode lookup is no longer as critical as it used to be:
56 * most of the lookups are going to be through the dcache.
58 #define I_HASHBITS i_hash_shift
59 #define I_HASHMASK i_hash_mask
61 static unsigned int i_hash_mask __read_mostly
;
62 static unsigned int i_hash_shift __read_mostly
;
65 * Each inode can be on two separate lists. One is
66 * the hash list of the inode, used for lookups. The
67 * other linked list is the "type" list:
68 * "in_use" - valid inode, i_count > 0, i_nlink > 0
69 * "dirty" - as "in_use" but also dirty
70 * "unused" - valid inode, i_count = 0
72 * A "dirty" list is maintained for each super block,
73 * allowing for low-overhead inode sync() operations.
76 LIST_HEAD(inode_in_use
);
77 LIST_HEAD(inode_unused
);
78 static struct hlist_head
*inode_hashtable __read_mostly
;
81 * A simple spinlock to protect the list manipulations.
83 * NOTE! You also have to own the lock if you change
84 * the i_state of an inode while it is in use..
86 DEFINE_SPINLOCK(inode_lock
);
89 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
90 * icache shrinking path, and the umount path. Without this exclusion,
91 * by the time prune_icache calls iput for the inode whose pages it has
92 * been invalidating, or by the time it calls clear_inode & destroy_inode
93 * from its final dispose_list, the struct super_block they refer to
94 * (for inode->i_sb->s_op) may already have been freed and reused.
96 * We make this an rwsem because the fastpath is icache shrinking. In
97 * some cases a filesystem may be doing a significant amount of work in
98 * its inode reclaim code, so this should improve parallelism.
100 static DECLARE_RWSEM(iprune_sem
);
103 * Statistics gathering..
105 struct inodes_stat_t inodes_stat
;
107 static struct kmem_cache
*inode_cachep __read_mostly
;
109 static void wake_up_inode(struct inode
*inode
)
112 * Prevent speculative execution through spin_unlock(&inode_lock);
115 wake_up_bit(&inode
->i_state
, __I_NEW
);
119 * inode_init_always - perform inode structure intialisation
120 * @sb: superblock inode belongs to
121 * @inode: inode to initialise
123 * These are initializations that need to be done on every inode
124 * allocation as the fields are not initialised by slab allocation.
126 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
128 static const struct address_space_operations empty_aops
;
129 static const struct inode_operations empty_iops
;
130 static const struct file_operations empty_fops
;
131 struct address_space
*const mapping
= &inode
->i_data
;
134 inode
->i_blkbits
= sb
->s_blocksize_bits
;
136 atomic_set(&inode
->i_count
, 1);
137 inode
->i_op
= &empty_iops
;
138 inode
->i_fop
= &empty_fops
;
142 atomic_set(&inode
->i_writecount
, 0);
146 inode
->i_generation
= 0;
148 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
150 inode
->i_pipe
= NULL
;
151 inode
->i_bdev
= NULL
;
152 inode
->i_cdev
= NULL
;
154 inode
->dirtied_when
= 0;
156 if (security_inode_alloc(inode
))
158 spin_lock_init(&inode
->i_lock
);
159 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
161 mutex_init(&inode
->i_mutex
);
162 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
164 init_rwsem(&inode
->i_alloc_sem
);
165 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
167 mapping
->a_ops
= &empty_aops
;
168 mapping
->host
= inode
;
170 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
171 mapping
->assoc_mapping
= NULL
;
172 mapping
->backing_dev_info
= &default_backing_dev_info
;
173 mapping
->writeback_index
= 0;
176 * If the block_device provides a backing_dev_info for client
177 * inodes then use that. Otherwise the inode share the bdev's
181 struct backing_dev_info
*bdi
;
183 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
184 mapping
->backing_dev_info
= bdi
;
186 inode
->i_private
= NULL
;
187 inode
->i_mapping
= mapping
;
188 #ifdef CONFIG_FS_POSIX_ACL
189 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
192 #ifdef CONFIG_FSNOTIFY
193 inode
->i_fsnotify_mask
= 0;
200 EXPORT_SYMBOL(inode_init_always
);
202 static struct inode
*alloc_inode(struct super_block
*sb
)
206 if (sb
->s_op
->alloc_inode
)
207 inode
= sb
->s_op
->alloc_inode(sb
);
209 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
214 if (unlikely(inode_init_always(sb
, inode
))) {
215 if (inode
->i_sb
->s_op
->destroy_inode
)
216 inode
->i_sb
->s_op
->destroy_inode(inode
);
218 kmem_cache_free(inode_cachep
, inode
);
225 void __destroy_inode(struct inode
*inode
)
227 BUG_ON(inode_has_buffers(inode
));
228 security_inode_free(inode
);
229 fsnotify_inode_delete(inode
);
230 #ifdef CONFIG_FS_POSIX_ACL
231 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
232 posix_acl_release(inode
->i_acl
);
233 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
234 posix_acl_release(inode
->i_default_acl
);
237 EXPORT_SYMBOL(__destroy_inode
);
239 void destroy_inode(struct inode
*inode
)
241 __destroy_inode(inode
);
242 if (inode
->i_sb
->s_op
->destroy_inode
)
243 inode
->i_sb
->s_op
->destroy_inode(inode
);
245 kmem_cache_free(inode_cachep
, (inode
));
249 * These are initializations that only need to be done
250 * once, because the fields are idempotent across use
251 * of the inode, so let the slab aware of that.
253 void inode_init_once(struct inode
*inode
)
255 memset(inode
, 0, sizeof(*inode
));
256 INIT_HLIST_NODE(&inode
->i_hash
);
257 INIT_LIST_HEAD(&inode
->i_dentry
);
258 INIT_LIST_HEAD(&inode
->i_devices
);
259 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
260 spin_lock_init(&inode
->i_data
.tree_lock
);
261 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
262 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
263 spin_lock_init(&inode
->i_data
.private_lock
);
264 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
265 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
266 i_size_ordered_init(inode
);
267 #ifdef CONFIG_INOTIFY
268 INIT_LIST_HEAD(&inode
->inotify_watches
);
269 mutex_init(&inode
->inotify_mutex
);
271 #ifdef CONFIG_FSNOTIFY
272 INIT_HLIST_HEAD(&inode
->i_fsnotify_mark_entries
);
275 EXPORT_SYMBOL(inode_init_once
);
277 static void init_once(void *foo
)
279 struct inode
*inode
= (struct inode
*) foo
;
281 inode_init_once(inode
);
285 * inode_lock must be held
287 void __iget(struct inode
*inode
)
289 if (atomic_inc_return(&inode
->i_count
) != 1)
292 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
293 list_move(&inode
->i_list
, &inode_in_use
);
294 inodes_stat
.nr_unused
--;
298 * clear_inode - clear an inode
299 * @inode: inode to clear
301 * This is called by the filesystem to tell us
302 * that the inode is no longer useful. We just
303 * terminate it with extreme prejudice.
305 void clear_inode(struct inode
*inode
)
308 invalidate_inode_buffers(inode
);
310 BUG_ON(inode
->i_data
.nrpages
);
311 BUG_ON(!(inode
->i_state
& I_FREEING
));
312 BUG_ON(inode
->i_state
& I_CLEAR
);
313 inode_sync_wait(inode
);
314 if (inode
->i_sb
->s_op
->clear_inode
)
315 inode
->i_sb
->s_op
->clear_inode(inode
);
316 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
318 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
320 inode
->i_state
= I_FREEING
| I_CLEAR
;
322 EXPORT_SYMBOL(clear_inode
);
324 static void evict(struct inode
*inode
, int delete)
326 const struct super_operations
*op
= inode
->i_sb
->s_op
;
328 if (delete && op
->delete_inode
) {
329 op
->delete_inode(inode
);
331 if (inode
->i_data
.nrpages
)
332 truncate_inode_pages(&inode
->i_data
, 0);
338 * dispose_list - dispose of the contents of a local list
339 * @head: the head of the list to free
341 * Dispose-list gets a local list with local inodes in it, so it doesn't
342 * need to worry about list corruption and SMP locks.
344 static void dispose_list(struct list_head
*head
)
348 while (!list_empty(head
)) {
351 inode
= list_first_entry(head
, struct inode
, i_list
);
352 list_del(&inode
->i_list
);
356 spin_lock(&inode_lock
);
357 hlist_del_init(&inode
->i_hash
);
358 list_del_init(&inode
->i_sb_list
);
359 spin_unlock(&inode_lock
);
361 wake_up_inode(inode
);
362 destroy_inode(inode
);
365 spin_lock(&inode_lock
);
366 inodes_stat
.nr_inodes
-= nr_disposed
;
367 spin_unlock(&inode_lock
);
371 * Invalidate all inodes for a device.
373 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
375 struct list_head
*next
;
376 int busy
= 0, count
= 0;
380 struct list_head
*tmp
= next
;
384 * We can reschedule here without worrying about the list's
385 * consistency because the per-sb list of inodes must not
386 * change during umount anymore, and because iprune_sem keeps
387 * shrink_icache_memory() away.
389 cond_resched_lock(&inode_lock
);
394 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
395 if (inode
->i_state
& I_NEW
)
397 invalidate_inode_buffers(inode
);
398 if (!atomic_read(&inode
->i_count
)) {
399 list_move(&inode
->i_list
, dispose
);
400 WARN_ON(inode
->i_state
& I_NEW
);
401 inode
->i_state
|= I_FREEING
;
407 /* only unused inodes may be cached with i_count zero */
408 inodes_stat
.nr_unused
-= count
;
413 * invalidate_inodes - discard the inodes on a device
416 * Discard all of the inodes for a given superblock. If the discard
417 * fails because there are busy inodes then a non zero value is returned.
418 * If the discard is successful all the inodes have been discarded.
420 int invalidate_inodes(struct super_block
*sb
)
423 LIST_HEAD(throw_away
);
425 down_write(&iprune_sem
);
426 spin_lock(&inode_lock
);
427 inotify_unmount_inodes(&sb
->s_inodes
);
428 fsnotify_unmount_inodes(&sb
->s_inodes
);
429 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
430 spin_unlock(&inode_lock
);
432 dispose_list(&throw_away
);
433 up_write(&iprune_sem
);
437 EXPORT_SYMBOL(invalidate_inodes
);
439 static int can_unuse(struct inode
*inode
)
443 if (inode_has_buffers(inode
))
445 if (atomic_read(&inode
->i_count
))
447 if (inode
->i_data
.nrpages
)
453 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
454 * a temporary list and then are freed outside inode_lock by dispose_list().
456 * Any inodes which are pinned purely because of attached pagecache have their
457 * pagecache removed. We expect the final iput() on that inode to add it to
458 * the front of the inode_unused list. So look for it there and if the
459 * inode is still freeable, proceed. The right inode is found 99.9% of the
460 * time in testing on a 4-way.
462 * If the inode has metadata buffers attached to mapping->private_list then
463 * try to remove them.
465 static void prune_icache(int nr_to_scan
)
470 unsigned long reap
= 0;
472 down_read(&iprune_sem
);
473 spin_lock(&inode_lock
);
474 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
477 if (list_empty(&inode_unused
))
480 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
482 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
483 list_move(&inode
->i_list
, &inode_unused
);
486 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
488 spin_unlock(&inode_lock
);
489 if (remove_inode_buffers(inode
))
490 reap
+= invalidate_mapping_pages(&inode
->i_data
,
493 spin_lock(&inode_lock
);
495 if (inode
!= list_entry(inode_unused
.next
,
496 struct inode
, i_list
))
497 continue; /* wrong inode or list_empty */
498 if (!can_unuse(inode
))
501 list_move(&inode
->i_list
, &freeable
);
502 WARN_ON(inode
->i_state
& I_NEW
);
503 inode
->i_state
|= I_FREEING
;
506 inodes_stat
.nr_unused
-= nr_pruned
;
507 if (current_is_kswapd())
508 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
510 __count_vm_events(PGINODESTEAL
, reap
);
511 spin_unlock(&inode_lock
);
513 dispose_list(&freeable
);
514 up_read(&iprune_sem
);
518 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
519 * "unused" means that no dentries are referring to the inodes: the files are
520 * not open and the dcache references to those inodes have already been
523 * This function is passed the number of inodes to scan, and it returns the
524 * total number of remaining possibly-reclaimable inodes.
526 static int shrink_icache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
530 * Nasty deadlock avoidance. We may hold various FS locks,
531 * and we don't want to recurse into the FS that called us
532 * in clear_inode() and friends..
534 if (!(gfp_mask
& __GFP_FS
))
538 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
541 static struct shrinker icache_shrinker
= {
542 .shrink
= shrink_icache_memory
,
543 .seeks
= DEFAULT_SEEKS
,
546 static void __wait_on_freeing_inode(struct inode
*inode
);
548 * Called with the inode lock held.
549 * NOTE: we are not increasing the inode-refcount, you must call __iget()
550 * by hand after calling find_inode now! This simplifies iunique and won't
551 * add any additional branch in the common code.
553 static struct inode
*find_inode(struct super_block
*sb
,
554 struct hlist_head
*head
,
555 int (*test
)(struct inode
*, void *),
558 struct hlist_node
*node
;
559 struct inode
*inode
= NULL
;
562 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
563 if (inode
->i_sb
!= sb
)
565 if (!test(inode
, data
))
567 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
568 __wait_on_freeing_inode(inode
);
573 return node
? inode
: NULL
;
577 * find_inode_fast is the fast path version of find_inode, see the comment at
578 * iget_locked for details.
580 static struct inode
*find_inode_fast(struct super_block
*sb
,
581 struct hlist_head
*head
, unsigned long ino
)
583 struct hlist_node
*node
;
584 struct inode
*inode
= NULL
;
587 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
588 if (inode
->i_ino
!= ino
)
590 if (inode
->i_sb
!= sb
)
592 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
593 __wait_on_freeing_inode(inode
);
598 return node
? inode
: NULL
;
601 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
605 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
607 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
608 return tmp
& I_HASHMASK
;
612 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
615 inodes_stat
.nr_inodes
++;
616 list_add(&inode
->i_list
, &inode_in_use
);
617 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
619 hlist_add_head(&inode
->i_hash
, head
);
623 * inode_add_to_lists - add a new inode to relevant lists
624 * @sb: superblock inode belongs to
625 * @inode: inode to mark in use
627 * When an inode is allocated it needs to be accounted for, added to the in use
628 * list, the owning superblock and the inode hash. This needs to be done under
629 * the inode_lock, so export a function to do this rather than the inode lock
630 * itself. We calculate the hash list to add to here so it is all internal
631 * which requires the caller to have already set up the inode number in the
634 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
636 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
638 spin_lock(&inode_lock
);
639 __inode_add_to_lists(sb
, head
, inode
);
640 spin_unlock(&inode_lock
);
642 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
645 * new_inode - obtain an inode
648 * Allocates a new inode for given superblock. The default gfp_mask
649 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
650 * If HIGHMEM pages are unsuitable or it is known that pages allocated
651 * for the page cache are not reclaimable or migratable,
652 * mapping_set_gfp_mask() must be called with suitable flags on the
653 * newly created inode's mapping
656 struct inode
*new_inode(struct super_block
*sb
)
659 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
660 * error if st_ino won't fit in target struct field. Use 32bit counter
661 * here to attempt to avoid that.
663 static unsigned int last_ino
;
666 spin_lock_prefetch(&inode_lock
);
668 inode
= alloc_inode(sb
);
670 spin_lock(&inode_lock
);
671 __inode_add_to_lists(sb
, NULL
, inode
);
672 inode
->i_ino
= ++last_ino
;
674 spin_unlock(&inode_lock
);
678 EXPORT_SYMBOL(new_inode
);
680 void unlock_new_inode(struct inode
*inode
)
682 #ifdef CONFIG_DEBUG_LOCK_ALLOC
683 if (inode
->i_mode
& S_IFDIR
) {
684 struct file_system_type
*type
= inode
->i_sb
->s_type
;
686 /* Set new key only if filesystem hasn't already changed it */
687 if (!lockdep_match_class(&inode
->i_mutex
,
688 &type
->i_mutex_key
)) {
690 * ensure nobody is actually holding i_mutex
692 mutex_destroy(&inode
->i_mutex
);
693 mutex_init(&inode
->i_mutex
);
694 lockdep_set_class(&inode
->i_mutex
,
695 &type
->i_mutex_dir_key
);
700 * This is special! We do not need the spinlock when clearing I_NEW,
701 * because we're guaranteed that nobody else tries to do anything about
702 * the state of the inode when it is locked, as we just created it (so
703 * there can be no old holders that haven't tested I_NEW).
704 * However we must emit the memory barrier so that other CPUs reliably
705 * see the clearing of I_NEW after the other inode initialisation has
709 WARN_ON(!(inode
->i_state
& I_NEW
));
710 inode
->i_state
&= ~I_NEW
;
711 wake_up_inode(inode
);
713 EXPORT_SYMBOL(unlock_new_inode
);
716 * This is called without the inode lock held.. Be careful.
718 * We no longer cache the sb_flags in i_flags - see fs.h
719 * -- rmk@arm.uk.linux.org
721 static struct inode
*get_new_inode(struct super_block
*sb
,
722 struct hlist_head
*head
,
723 int (*test
)(struct inode
*, void *),
724 int (*set
)(struct inode
*, void *),
729 inode
= alloc_inode(sb
);
733 spin_lock(&inode_lock
);
734 /* We released the lock, so.. */
735 old
= find_inode(sb
, head
, test
, data
);
737 if (set(inode
, data
))
740 __inode_add_to_lists(sb
, head
, inode
);
741 inode
->i_state
= I_NEW
;
742 spin_unlock(&inode_lock
);
744 /* Return the locked inode with I_NEW set, the
745 * caller is responsible for filling in the contents
751 * Uhhuh, somebody else created the same inode under
752 * us. Use the old inode instead of the one we just
756 spin_unlock(&inode_lock
);
757 destroy_inode(inode
);
759 wait_on_inode(inode
);
764 spin_unlock(&inode_lock
);
765 destroy_inode(inode
);
770 * get_new_inode_fast is the fast path version of get_new_inode, see the
771 * comment at iget_locked for details.
773 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
774 struct hlist_head
*head
, unsigned long ino
)
778 inode
= alloc_inode(sb
);
782 spin_lock(&inode_lock
);
783 /* We released the lock, so.. */
784 old
= find_inode_fast(sb
, head
, ino
);
787 __inode_add_to_lists(sb
, head
, inode
);
788 inode
->i_state
= I_NEW
;
789 spin_unlock(&inode_lock
);
791 /* Return the locked inode with I_NEW set, the
792 * caller is responsible for filling in the contents
798 * Uhhuh, somebody else created the same inode under
799 * us. Use the old inode instead of the one we just
803 spin_unlock(&inode_lock
);
804 destroy_inode(inode
);
806 wait_on_inode(inode
);
812 * iunique - get a unique inode number
814 * @max_reserved: highest reserved inode number
816 * Obtain an inode number that is unique on the system for a given
817 * superblock. This is used by file systems that have no natural
818 * permanent inode numbering system. An inode number is returned that
819 * is higher than the reserved limit but unique.
822 * With a large number of inodes live on the file system this function
823 * currently becomes quite slow.
825 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
828 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
829 * error if st_ino won't fit in target struct field. Use 32bit counter
830 * here to attempt to avoid that.
832 static unsigned int counter
;
834 struct hlist_head
*head
;
837 spin_lock(&inode_lock
);
839 if (counter
<= max_reserved
)
840 counter
= max_reserved
+ 1;
842 head
= inode_hashtable
+ hash(sb
, res
);
843 inode
= find_inode_fast(sb
, head
, res
);
844 } while (inode
!= NULL
);
845 spin_unlock(&inode_lock
);
849 EXPORT_SYMBOL(iunique
);
851 struct inode
*igrab(struct inode
*inode
)
853 spin_lock(&inode_lock
);
854 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
)))
858 * Handle the case where s_op->clear_inode is not been
859 * called yet, and somebody is calling igrab
860 * while the inode is getting freed.
863 spin_unlock(&inode_lock
);
866 EXPORT_SYMBOL(igrab
);
869 * ifind - internal function, you want ilookup5() or iget5().
870 * @sb: super block of file system to search
871 * @head: the head of the list to search
872 * @test: callback used for comparisons between inodes
873 * @data: opaque data pointer to pass to @test
874 * @wait: if true wait for the inode to be unlocked, if false do not
876 * ifind() searches for the inode specified by @data in the inode
877 * cache. This is a generalized version of ifind_fast() for file systems where
878 * the inode number is not sufficient for unique identification of an inode.
880 * If the inode is in the cache, the inode is returned with an incremented
883 * Otherwise NULL is returned.
885 * Note, @test is called with the inode_lock held, so can't sleep.
887 static struct inode
*ifind(struct super_block
*sb
,
888 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
889 void *data
, const int wait
)
893 spin_lock(&inode_lock
);
894 inode
= find_inode(sb
, head
, test
, data
);
897 spin_unlock(&inode_lock
);
899 wait_on_inode(inode
);
902 spin_unlock(&inode_lock
);
907 * ifind_fast - internal function, you want ilookup() or iget().
908 * @sb: super block of file system to search
909 * @head: head of the list to search
910 * @ino: inode number to search for
912 * ifind_fast() searches for the inode @ino in the inode cache. This is for
913 * file systems where the inode number is sufficient for unique identification
916 * If the inode is in the cache, the inode is returned with an incremented
919 * Otherwise NULL is returned.
921 static struct inode
*ifind_fast(struct super_block
*sb
,
922 struct hlist_head
*head
, unsigned long ino
)
926 spin_lock(&inode_lock
);
927 inode
= find_inode_fast(sb
, head
, ino
);
930 spin_unlock(&inode_lock
);
931 wait_on_inode(inode
);
934 spin_unlock(&inode_lock
);
939 * ilookup5_nowait - search for an inode in the inode cache
940 * @sb: super block of file system to search
941 * @hashval: hash value (usually inode number) to search for
942 * @test: callback used for comparisons between inodes
943 * @data: opaque data pointer to pass to @test
945 * ilookup5() uses ifind() to search for the inode specified by @hashval and
946 * @data in the inode cache. This is a generalized version of ilookup() for
947 * file systems where the inode number is not sufficient for unique
948 * identification of an inode.
950 * If the inode is in the cache, the inode is returned with an incremented
951 * reference count. Note, the inode lock is not waited upon so you have to be
952 * very careful what you do with the returned inode. You probably should be
953 * using ilookup5() instead.
955 * Otherwise NULL is returned.
957 * Note, @test is called with the inode_lock held, so can't sleep.
959 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
960 int (*test
)(struct inode
*, void *), void *data
)
962 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
964 return ifind(sb
, head
, test
, data
, 0);
966 EXPORT_SYMBOL(ilookup5_nowait
);
969 * ilookup5 - search for an inode in the inode cache
970 * @sb: super block of file system to search
971 * @hashval: hash value (usually inode number) to search for
972 * @test: callback used for comparisons between inodes
973 * @data: opaque data pointer to pass to @test
975 * ilookup5() uses ifind() to search for the inode specified by @hashval and
976 * @data in the inode cache. This is a generalized version of ilookup() for
977 * file systems where the inode number is not sufficient for unique
978 * identification of an inode.
980 * If the inode is in the cache, the inode lock is waited upon and the inode is
981 * returned with an incremented reference count.
983 * Otherwise NULL is returned.
985 * Note, @test is called with the inode_lock held, so can't sleep.
987 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
988 int (*test
)(struct inode
*, void *), void *data
)
990 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
992 return ifind(sb
, head
, test
, data
, 1);
994 EXPORT_SYMBOL(ilookup5
);
997 * ilookup - search for an inode in the inode cache
998 * @sb: super block of file system to search
999 * @ino: inode number to search for
1001 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1002 * This is for file systems where the inode number is sufficient for unique
1003 * identification of an inode.
1005 * If the inode is in the cache, the inode is returned with an incremented
1008 * Otherwise NULL is returned.
1010 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1012 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1014 return ifind_fast(sb
, head
, ino
);
1016 EXPORT_SYMBOL(ilookup
);
1019 * iget5_locked - obtain an inode from a mounted file system
1020 * @sb: super block of file system
1021 * @hashval: hash value (usually inode number) to get
1022 * @test: callback used for comparisons between inodes
1023 * @set: callback used to initialize a new struct inode
1024 * @data: opaque data pointer to pass to @test and @set
1026 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1027 * and @data in the inode cache and if present it is returned with an increased
1028 * reference count. This is a generalized version of iget_locked() for file
1029 * systems where the inode number is not sufficient for unique identification
1032 * If the inode is not in cache, get_new_inode() is called to allocate a new
1033 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1034 * file system gets to fill it in before unlocking it via unlock_new_inode().
1036 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1038 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1039 int (*test
)(struct inode
*, void *),
1040 int (*set
)(struct inode
*, void *), void *data
)
1042 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1043 struct inode
*inode
;
1045 inode
= ifind(sb
, head
, test
, data
, 1);
1049 * get_new_inode() will do the right thing, re-trying the search
1050 * in case it had to block at any point.
1052 return get_new_inode(sb
, head
, test
, set
, data
);
1054 EXPORT_SYMBOL(iget5_locked
);
1057 * iget_locked - obtain an inode from a mounted file system
1058 * @sb: super block of file system
1059 * @ino: inode number to get
1061 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1062 * the inode cache and if present it is returned with an increased reference
1063 * count. This is for file systems where the inode number is sufficient for
1064 * unique identification of an inode.
1066 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1067 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1068 * The file system gets to fill it in before unlocking it via
1069 * unlock_new_inode().
1071 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1073 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1074 struct inode
*inode
;
1076 inode
= ifind_fast(sb
, head
, ino
);
1080 * get_new_inode_fast() will do the right thing, re-trying the search
1081 * in case it had to block at any point.
1083 return get_new_inode_fast(sb
, head
, ino
);
1085 EXPORT_SYMBOL(iget_locked
);
1087 int insert_inode_locked(struct inode
*inode
)
1089 struct super_block
*sb
= inode
->i_sb
;
1090 ino_t ino
= inode
->i_ino
;
1091 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1093 inode
->i_state
|= I_NEW
;
1095 struct hlist_node
*node
;
1096 struct inode
*old
= NULL
;
1097 spin_lock(&inode_lock
);
1098 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1099 if (old
->i_ino
!= ino
)
1101 if (old
->i_sb
!= sb
)
1103 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
))
1107 if (likely(!node
)) {
1108 hlist_add_head(&inode
->i_hash
, head
);
1109 spin_unlock(&inode_lock
);
1113 spin_unlock(&inode_lock
);
1115 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1122 EXPORT_SYMBOL(insert_inode_locked
);
1124 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1125 int (*test
)(struct inode
*, void *), void *data
)
1127 struct super_block
*sb
= inode
->i_sb
;
1128 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1130 inode
->i_state
|= I_NEW
;
1133 struct hlist_node
*node
;
1134 struct inode
*old
= NULL
;
1136 spin_lock(&inode_lock
);
1137 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1138 if (old
->i_sb
!= sb
)
1140 if (!test(old
, data
))
1142 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
))
1146 if (likely(!node
)) {
1147 hlist_add_head(&inode
->i_hash
, head
);
1148 spin_unlock(&inode_lock
);
1152 spin_unlock(&inode_lock
);
1154 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1161 EXPORT_SYMBOL(insert_inode_locked4
);
1164 * __insert_inode_hash - hash an inode
1165 * @inode: unhashed inode
1166 * @hashval: unsigned long value used to locate this object in the
1169 * Add an inode to the inode hash for this superblock.
1171 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1173 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1174 spin_lock(&inode_lock
);
1175 hlist_add_head(&inode
->i_hash
, head
);
1176 spin_unlock(&inode_lock
);
1178 EXPORT_SYMBOL(__insert_inode_hash
);
1181 * remove_inode_hash - remove an inode from the hash
1182 * @inode: inode to unhash
1184 * Remove an inode from the superblock.
1186 void remove_inode_hash(struct inode
*inode
)
1188 spin_lock(&inode_lock
);
1189 hlist_del_init(&inode
->i_hash
);
1190 spin_unlock(&inode_lock
);
1192 EXPORT_SYMBOL(remove_inode_hash
);
1195 * Tell the filesystem that this inode is no longer of any interest and should
1196 * be completely destroyed.
1198 * We leave the inode in the inode hash table until *after* the filesystem's
1199 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1200 * instigate) will always find up-to-date information either in the hash or on
1203 * I_FREEING is set so that no-one will take a new reference to the inode while
1204 * it is being deleted.
1206 void generic_delete_inode(struct inode
*inode
)
1208 list_del_init(&inode
->i_list
);
1209 list_del_init(&inode
->i_sb_list
);
1210 WARN_ON(inode
->i_state
& I_NEW
);
1211 inode
->i_state
|= I_FREEING
;
1212 inodes_stat
.nr_inodes
--;
1213 spin_unlock(&inode_lock
);
1217 spin_lock(&inode_lock
);
1218 hlist_del_init(&inode
->i_hash
);
1219 spin_unlock(&inode_lock
);
1220 wake_up_inode(inode
);
1221 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
1222 destroy_inode(inode
);
1224 EXPORT_SYMBOL(generic_delete_inode
);
1227 * generic_detach_inode - remove inode from inode lists
1228 * @inode: inode to remove
1230 * Remove inode from inode lists, write it if it's dirty. This is just an
1231 * internal VFS helper exported for hugetlbfs. Do not use!
1233 * Returns 1 if inode should be completely destroyed.
1235 int generic_detach_inode(struct inode
*inode
)
1237 struct super_block
*sb
= inode
->i_sb
;
1239 if (!hlist_unhashed(&inode
->i_hash
)) {
1240 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1241 list_move(&inode
->i_list
, &inode_unused
);
1242 inodes_stat
.nr_unused
++;
1243 if (sb
->s_flags
& MS_ACTIVE
) {
1244 spin_unlock(&inode_lock
);
1247 WARN_ON(inode
->i_state
& I_NEW
);
1248 inode
->i_state
|= I_WILL_FREE
;
1249 spin_unlock(&inode_lock
);
1250 write_inode_now(inode
, 1);
1251 spin_lock(&inode_lock
);
1252 WARN_ON(inode
->i_state
& I_NEW
);
1253 inode
->i_state
&= ~I_WILL_FREE
;
1254 inodes_stat
.nr_unused
--;
1255 hlist_del_init(&inode
->i_hash
);
1257 list_del_init(&inode
->i_list
);
1258 list_del_init(&inode
->i_sb_list
);
1259 WARN_ON(inode
->i_state
& I_NEW
);
1260 inode
->i_state
|= I_FREEING
;
1261 inodes_stat
.nr_inodes
--;
1262 spin_unlock(&inode_lock
);
1265 EXPORT_SYMBOL_GPL(generic_detach_inode
);
1267 static void generic_forget_inode(struct inode
*inode
)
1269 if (!generic_detach_inode(inode
))
1272 wake_up_inode(inode
);
1273 destroy_inode(inode
);
1277 * Normal UNIX filesystem behaviour: delete the
1278 * inode when the usage count drops to zero, and
1281 void generic_drop_inode(struct inode
*inode
)
1283 if (!inode
->i_nlink
)
1284 generic_delete_inode(inode
);
1286 generic_forget_inode(inode
);
1288 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1291 * Called when we're dropping the last reference
1294 * Call the FS "drop()" function, defaulting to
1295 * the legacy UNIX filesystem behaviour..
1297 * NOTE! NOTE! NOTE! We're called with the inode lock
1298 * held, and the drop function is supposed to release
1301 static inline void iput_final(struct inode
*inode
)
1303 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1304 void (*drop
)(struct inode
*) = generic_drop_inode
;
1306 if (op
&& op
->drop_inode
)
1307 drop
= op
->drop_inode
;
1312 * iput - put an inode
1313 * @inode: inode to put
1315 * Puts an inode, dropping its usage count. If the inode use count hits
1316 * zero, the inode is then freed and may also be destroyed.
1318 * Consequently, iput() can sleep.
1320 void iput(struct inode
*inode
)
1323 BUG_ON(inode
->i_state
& I_CLEAR
);
1325 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1329 EXPORT_SYMBOL(iput
);
1332 * bmap - find a block number in a file
1333 * @inode: inode of file
1334 * @block: block to find
1336 * Returns the block number on the device holding the inode that
1337 * is the disk block number for the block of the file requested.
1338 * That is, asked for block 4 of inode 1 the function will return the
1339 * disk block relative to the disk start that holds that block of the
1342 sector_t
bmap(struct inode
*inode
, sector_t block
)
1345 if (inode
->i_mapping
->a_ops
->bmap
)
1346 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1349 EXPORT_SYMBOL(bmap
);
1352 * With relative atime, only update atime if the previous atime is
1353 * earlier than either the ctime or mtime or if at least a day has
1354 * passed since the last atime update.
1356 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1357 struct timespec now
)
1360 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1363 * Is mtime younger than atime? If yes, update atime:
1365 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1368 * Is ctime younger than atime? If yes, update atime:
1370 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1374 * Is the previous atime value older than a day? If yes,
1377 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1380 * Good, we can skip the atime update:
1386 * touch_atime - update the access time
1387 * @mnt: mount the inode is accessed on
1388 * @dentry: dentry accessed
1390 * Update the accessed time on an inode and mark it for writeback.
1391 * This function automatically handles read only file systems and media,
1392 * as well as the "noatime" flag and inode specific "noatime" markers.
1394 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1396 struct inode
*inode
= dentry
->d_inode
;
1397 struct timespec now
;
1399 if (inode
->i_flags
& S_NOATIME
)
1401 if (IS_NOATIME(inode
))
1403 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1406 if (mnt
->mnt_flags
& MNT_NOATIME
)
1408 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1411 now
= current_fs_time(inode
->i_sb
);
1413 if (!relatime_need_update(mnt
, inode
, now
))
1416 if (timespec_equal(&inode
->i_atime
, &now
))
1419 if (mnt_want_write(mnt
))
1422 inode
->i_atime
= now
;
1423 mark_inode_dirty_sync(inode
);
1424 mnt_drop_write(mnt
);
1426 EXPORT_SYMBOL(touch_atime
);
1429 * file_update_time - update mtime and ctime time
1430 * @file: file accessed
1432 * Update the mtime and ctime members of an inode and mark the inode
1433 * for writeback. Note that this function is meant exclusively for
1434 * usage in the file write path of filesystems, and filesystems may
1435 * choose to explicitly ignore update via this function with the
1436 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1437 * timestamps are handled by the server.
1440 void file_update_time(struct file
*file
)
1442 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1443 struct timespec now
;
1444 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1446 /* First try to exhaust all avenues to not sync */
1447 if (IS_NOCMTIME(inode
))
1450 now
= current_fs_time(inode
->i_sb
);
1451 if (!timespec_equal(&inode
->i_mtime
, &now
))
1454 if (!timespec_equal(&inode
->i_ctime
, &now
))
1457 if (IS_I_VERSION(inode
))
1458 sync_it
|= S_VERSION
;
1463 /* Finally allowed to write? Takes lock. */
1464 if (mnt_want_write_file(file
))
1467 /* Only change inode inside the lock region */
1468 if (sync_it
& S_VERSION
)
1469 inode_inc_iversion(inode
);
1470 if (sync_it
& S_CTIME
)
1471 inode
->i_ctime
= now
;
1472 if (sync_it
& S_MTIME
)
1473 inode
->i_mtime
= now
;
1474 mark_inode_dirty_sync(inode
);
1475 mnt_drop_write(file
->f_path
.mnt
);
1477 EXPORT_SYMBOL(file_update_time
);
1479 int inode_needs_sync(struct inode
*inode
)
1483 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1487 EXPORT_SYMBOL(inode_needs_sync
);
1489 int inode_wait(void *word
)
1494 EXPORT_SYMBOL(inode_wait
);
1497 * If we try to find an inode in the inode hash while it is being
1498 * deleted, we have to wait until the filesystem completes its
1499 * deletion before reporting that it isn't found. This function waits
1500 * until the deletion _might_ have completed. Callers are responsible
1501 * to recheck inode state.
1503 * It doesn't matter if I_NEW is not set initially, a call to
1504 * wake_up_inode() after removing from the hash list will DTRT.
1506 * This is called with inode_lock held.
1508 static void __wait_on_freeing_inode(struct inode
*inode
)
1510 wait_queue_head_t
*wq
;
1511 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1512 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1513 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1514 spin_unlock(&inode_lock
);
1516 finish_wait(wq
, &wait
.wait
);
1517 spin_lock(&inode_lock
);
1520 static __initdata
unsigned long ihash_entries
;
1521 static int __init
set_ihash_entries(char *str
)
1525 ihash_entries
= simple_strtoul(str
, &str
, 0);
1528 __setup("ihash_entries=", set_ihash_entries
);
1531 * Initialize the waitqueues and inode hash table.
1533 void __init
inode_init_early(void)
1537 /* If hashes are distributed across NUMA nodes, defer
1538 * hash allocation until vmalloc space is available.
1544 alloc_large_system_hash("Inode-cache",
1545 sizeof(struct hlist_head
),
1553 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1554 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1557 void __init
inode_init(void)
1561 /* inode slab cache */
1562 inode_cachep
= kmem_cache_create("inode_cache",
1563 sizeof(struct inode
),
1565 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1568 register_shrinker(&icache_shrinker
);
1570 /* Hash may have been set up in inode_init_early */
1575 alloc_large_system_hash("Inode-cache",
1576 sizeof(struct hlist_head
),
1584 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1585 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1588 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1590 inode
->i_mode
= mode
;
1591 if (S_ISCHR(mode
)) {
1592 inode
->i_fop
= &def_chr_fops
;
1593 inode
->i_rdev
= rdev
;
1594 } else if (S_ISBLK(mode
)) {
1595 inode
->i_fop
= &def_blk_fops
;
1596 inode
->i_rdev
= rdev
;
1597 } else if (S_ISFIFO(mode
))
1598 inode
->i_fop
= &def_fifo_fops
;
1599 else if (S_ISSOCK(mode
))
1600 inode
->i_fop
= &bad_sock_fops
;
1602 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1603 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1606 EXPORT_SYMBOL(init_special_inode
);
1609 * Init uid,gid,mode for new inode according to posix standards
1611 * @dir: Directory inode
1612 * @mode: mode of the new inode
1614 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1617 inode
->i_uid
= current_fsuid();
1618 if (dir
&& dir
->i_mode
& S_ISGID
) {
1619 inode
->i_gid
= dir
->i_gid
;
1623 inode
->i_gid
= current_fsgid();
1624 inode
->i_mode
= mode
;
1626 EXPORT_SYMBOL(inode_init_owner
);