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/fsnotify.h>
24 #include <linux/mount.h>
25 #include <linux/async.h>
26 #include <linux/posix_acl.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 static LIST_HEAD(inode_unused
);
76 static struct hlist_head
*inode_hashtable __read_mostly
;
79 * A simple spinlock to protect the list manipulations.
81 * NOTE! You also have to own the lock if you change
82 * the i_state of an inode while it is in use..
84 DEFINE_SPINLOCK(inode_lock
);
87 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
88 * icache shrinking path, and the umount path. Without this exclusion,
89 * by the time prune_icache calls iput for the inode whose pages it has
90 * been invalidating, or by the time it calls clear_inode & destroy_inode
91 * from its final dispose_list, the struct super_block they refer to
92 * (for inode->i_sb->s_op) may already have been freed and reused.
94 * We make this an rwsem because the fastpath is icache shrinking. In
95 * some cases a filesystem may be doing a significant amount of work in
96 * its inode reclaim code, so this should improve parallelism.
98 static DECLARE_RWSEM(iprune_sem
);
101 * Statistics gathering..
103 struct inodes_stat_t inodes_stat
;
105 static struct percpu_counter nr_inodes __cacheline_aligned_in_smp
;
106 static struct percpu_counter nr_inodes_unused __cacheline_aligned_in_smp
;
108 static struct kmem_cache
*inode_cachep __read_mostly
;
110 static inline int get_nr_inodes(void)
112 return percpu_counter_sum_positive(&nr_inodes
);
115 static inline int get_nr_inodes_unused(void)
117 return percpu_counter_sum_positive(&nr_inodes_unused
);
120 int get_nr_dirty_inodes(void)
122 int nr_dirty
= get_nr_inodes() - get_nr_inodes_unused();
123 return nr_dirty
> 0 ? nr_dirty
: 0;
128 * Handle nr_inode sysctl
131 int proc_nr_inodes(ctl_table
*table
, int write
,
132 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
134 inodes_stat
.nr_inodes
= get_nr_inodes();
135 inodes_stat
.nr_unused
= get_nr_inodes_unused();
136 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
140 static void wake_up_inode(struct inode
*inode
)
143 * Prevent speculative execution through spin_unlock(&inode_lock);
146 wake_up_bit(&inode
->i_state
, __I_NEW
);
150 * inode_init_always - perform inode structure intialisation
151 * @sb: superblock inode belongs to
152 * @inode: inode to initialise
154 * These are initializations that need to be done on every inode
155 * allocation as the fields are not initialised by slab allocation.
157 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
159 static const struct address_space_operations empty_aops
;
160 static const struct inode_operations empty_iops
;
161 static const struct file_operations empty_fops
;
162 struct address_space
*const mapping
= &inode
->i_data
;
165 inode
->i_blkbits
= sb
->s_blocksize_bits
;
167 atomic_set(&inode
->i_count
, 1);
168 inode
->i_op
= &empty_iops
;
169 inode
->i_fop
= &empty_fops
;
173 atomic_set(&inode
->i_writecount
, 0);
177 inode
->i_generation
= 0;
179 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
181 inode
->i_pipe
= NULL
;
182 inode
->i_bdev
= NULL
;
183 inode
->i_cdev
= NULL
;
185 inode
->dirtied_when
= 0;
187 if (security_inode_alloc(inode
))
189 spin_lock_init(&inode
->i_lock
);
190 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
192 mutex_init(&inode
->i_mutex
);
193 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
195 init_rwsem(&inode
->i_alloc_sem
);
196 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
198 mapping
->a_ops
= &empty_aops
;
199 mapping
->host
= inode
;
201 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
202 mapping
->assoc_mapping
= NULL
;
203 mapping
->backing_dev_info
= &default_backing_dev_info
;
204 mapping
->writeback_index
= 0;
207 * If the block_device provides a backing_dev_info for client
208 * inodes then use that. Otherwise the inode share the bdev's
212 struct backing_dev_info
*bdi
;
214 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
215 mapping
->backing_dev_info
= bdi
;
217 inode
->i_private
= NULL
;
218 inode
->i_mapping
= mapping
;
219 #ifdef CONFIG_FS_POSIX_ACL
220 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
223 #ifdef CONFIG_FSNOTIFY
224 inode
->i_fsnotify_mask
= 0;
227 percpu_counter_inc(&nr_inodes
);
233 EXPORT_SYMBOL(inode_init_always
);
235 static struct inode
*alloc_inode(struct super_block
*sb
)
239 if (sb
->s_op
->alloc_inode
)
240 inode
= sb
->s_op
->alloc_inode(sb
);
242 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
247 if (unlikely(inode_init_always(sb
, inode
))) {
248 if (inode
->i_sb
->s_op
->destroy_inode
)
249 inode
->i_sb
->s_op
->destroy_inode(inode
);
251 kmem_cache_free(inode_cachep
, inode
);
258 void __destroy_inode(struct inode
*inode
)
260 BUG_ON(inode_has_buffers(inode
));
261 security_inode_free(inode
);
262 fsnotify_inode_delete(inode
);
263 #ifdef CONFIG_FS_POSIX_ACL
264 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
265 posix_acl_release(inode
->i_acl
);
266 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
267 posix_acl_release(inode
->i_default_acl
);
269 percpu_counter_dec(&nr_inodes
);
271 EXPORT_SYMBOL(__destroy_inode
);
273 static void destroy_inode(struct inode
*inode
)
275 __destroy_inode(inode
);
276 if (inode
->i_sb
->s_op
->destroy_inode
)
277 inode
->i_sb
->s_op
->destroy_inode(inode
);
279 kmem_cache_free(inode_cachep
, (inode
));
283 * These are initializations that only need to be done
284 * once, because the fields are idempotent across use
285 * of the inode, so let the slab aware of that.
287 void inode_init_once(struct inode
*inode
)
289 memset(inode
, 0, sizeof(*inode
));
290 INIT_HLIST_NODE(&inode
->i_hash
);
291 INIT_LIST_HEAD(&inode
->i_dentry
);
292 INIT_LIST_HEAD(&inode
->i_devices
);
293 INIT_LIST_HEAD(&inode
->i_list
);
294 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
295 spin_lock_init(&inode
->i_data
.tree_lock
);
296 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
297 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
298 spin_lock_init(&inode
->i_data
.private_lock
);
299 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
300 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
301 i_size_ordered_init(inode
);
302 #ifdef CONFIG_FSNOTIFY
303 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
306 EXPORT_SYMBOL(inode_init_once
);
308 static void init_once(void *foo
)
310 struct inode
*inode
= (struct inode
*) foo
;
312 inode_init_once(inode
);
316 * inode_lock must be held
318 void __iget(struct inode
*inode
)
320 atomic_inc(&inode
->i_count
);
323 static void inode_lru_list_add(struct inode
*inode
)
325 if (list_empty(&inode
->i_list
)) {
326 list_add(&inode
->i_list
, &inode_unused
);
327 percpu_counter_inc(&nr_inodes_unused
);
331 static void inode_lru_list_del(struct inode
*inode
)
333 if (!list_empty(&inode
->i_list
)) {
334 list_del_init(&inode
->i_list
);
335 percpu_counter_dec(&nr_inodes_unused
);
339 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
343 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
345 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
346 return tmp
& I_HASHMASK
;
350 * __insert_inode_hash - hash an inode
351 * @inode: unhashed inode
352 * @hashval: unsigned long value used to locate this object in the
355 * Add an inode to the inode hash for this superblock.
357 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
359 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
360 spin_lock(&inode_lock
);
361 hlist_add_head(&inode
->i_hash
, head
);
362 spin_unlock(&inode_lock
);
364 EXPORT_SYMBOL(__insert_inode_hash
);
367 * __remove_inode_hash - remove an inode from the hash
368 * @inode: inode to unhash
370 * Remove an inode from the superblock.
372 static void __remove_inode_hash(struct inode
*inode
)
374 hlist_del_init(&inode
->i_hash
);
378 * remove_inode_hash - remove an inode from the hash
379 * @inode: inode to unhash
381 * Remove an inode from the superblock.
383 void remove_inode_hash(struct inode
*inode
)
385 spin_lock(&inode_lock
);
386 hlist_del_init(&inode
->i_hash
);
387 spin_unlock(&inode_lock
);
389 EXPORT_SYMBOL(remove_inode_hash
);
391 void end_writeback(struct inode
*inode
)
394 BUG_ON(inode
->i_data
.nrpages
);
395 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
396 BUG_ON(!(inode
->i_state
& I_FREEING
));
397 BUG_ON(inode
->i_state
& I_CLEAR
);
398 inode_sync_wait(inode
);
399 inode
->i_state
= I_FREEING
| I_CLEAR
;
401 EXPORT_SYMBOL(end_writeback
);
403 static void evict(struct inode
*inode
)
405 const struct super_operations
*op
= inode
->i_sb
->s_op
;
407 if (op
->evict_inode
) {
408 op
->evict_inode(inode
);
410 if (inode
->i_data
.nrpages
)
411 truncate_inode_pages(&inode
->i_data
, 0);
412 end_writeback(inode
);
414 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
416 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
421 * dispose_list - dispose of the contents of a local list
422 * @head: the head of the list to free
424 * Dispose-list gets a local list with local inodes in it, so it doesn't
425 * need to worry about list corruption and SMP locks.
427 static void dispose_list(struct list_head
*head
)
429 while (!list_empty(head
)) {
432 inode
= list_first_entry(head
, struct inode
, i_list
);
433 list_del_init(&inode
->i_list
);
437 spin_lock(&inode_lock
);
438 __remove_inode_hash(inode
);
439 list_del_init(&inode
->i_sb_list
);
440 spin_unlock(&inode_lock
);
442 wake_up_inode(inode
);
443 destroy_inode(inode
);
448 * Invalidate all inodes for a device.
450 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
452 struct list_head
*next
;
457 struct list_head
*tmp
= next
;
461 * We can reschedule here without worrying about the list's
462 * consistency because the per-sb list of inodes must not
463 * change during umount anymore, and because iprune_sem keeps
464 * shrink_icache_memory() away.
466 cond_resched_lock(&inode_lock
);
471 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
472 if (inode
->i_state
& I_NEW
)
474 invalidate_inode_buffers(inode
);
475 if (!atomic_read(&inode
->i_count
)) {
476 list_move(&inode
->i_list
, dispose
);
477 WARN_ON(inode
->i_state
& I_NEW
);
478 inode
->i_state
|= I_FREEING
;
479 if (!(inode
->i_state
& (I_DIRTY
| I_SYNC
)))
480 percpu_counter_dec(&nr_inodes_unused
);
489 * invalidate_inodes - discard the inodes on a device
492 * Discard all of the inodes for a given superblock. If the discard
493 * fails because there are busy inodes then a non zero value is returned.
494 * If the discard is successful all the inodes have been discarded.
496 int invalidate_inodes(struct super_block
*sb
)
499 LIST_HEAD(throw_away
);
501 down_write(&iprune_sem
);
502 spin_lock(&inode_lock
);
503 fsnotify_unmount_inodes(&sb
->s_inodes
);
504 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
505 spin_unlock(&inode_lock
);
507 dispose_list(&throw_away
);
508 up_write(&iprune_sem
);
513 static int can_unuse(struct inode
*inode
)
515 if (inode
->i_state
& ~I_REFERENCED
)
517 if (inode_has_buffers(inode
))
519 if (atomic_read(&inode
->i_count
))
521 if (inode
->i_data
.nrpages
)
527 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
528 * temporary list and then are freed outside inode_lock by dispose_list().
530 * Any inodes which are pinned purely because of attached pagecache have their
531 * pagecache removed. If the inode has metadata buffers attached to
532 * mapping->private_list then try to remove them.
534 * If the inode has the I_REFERENCED flag set, then it means that it has been
535 * used recently - the flag is set in iput_final(). When we encounter such an
536 * inode, clear the flag and move it to the back of the LRU so it gets another
537 * pass through the LRU before it gets reclaimed. This is necessary because of
538 * the fact we are doing lazy LRU updates to minimise lock contention so the
539 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
540 * with this flag set because they are the inodes that are out of order.
542 static void prune_icache(int nr_to_scan
)
546 unsigned long reap
= 0;
548 down_read(&iprune_sem
);
549 spin_lock(&inode_lock
);
550 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
553 if (list_empty(&inode_unused
))
556 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
559 * Referenced or dirty inodes are still in use. Give them
560 * another pass through the LRU as we canot reclaim them now.
562 if (atomic_read(&inode
->i_count
) ||
563 (inode
->i_state
& ~I_REFERENCED
)) {
564 list_del_init(&inode
->i_list
);
565 percpu_counter_dec(&nr_inodes_unused
);
569 /* recently referenced inodes get one more pass */
570 if (inode
->i_state
& I_REFERENCED
) {
571 list_move(&inode
->i_list
, &inode_unused
);
572 inode
->i_state
&= ~I_REFERENCED
;
575 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
577 spin_unlock(&inode_lock
);
578 if (remove_inode_buffers(inode
))
579 reap
+= invalidate_mapping_pages(&inode
->i_data
,
582 spin_lock(&inode_lock
);
584 if (inode
!= list_entry(inode_unused
.next
,
585 struct inode
, i_list
))
586 continue; /* wrong inode or list_empty */
587 if (!can_unuse(inode
))
590 list_move(&inode
->i_list
, &freeable
);
591 WARN_ON(inode
->i_state
& I_NEW
);
592 inode
->i_state
|= I_FREEING
;
593 percpu_counter_dec(&nr_inodes_unused
);
595 if (current_is_kswapd())
596 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
598 __count_vm_events(PGINODESTEAL
, reap
);
599 spin_unlock(&inode_lock
);
601 dispose_list(&freeable
);
602 up_read(&iprune_sem
);
606 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
607 * "unused" means that no dentries are referring to the inodes: the files are
608 * not open and the dcache references to those inodes have already been
611 * This function is passed the number of inodes to scan, and it returns the
612 * total number of remaining possibly-reclaimable inodes.
614 static int shrink_icache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
618 * Nasty deadlock avoidance. We may hold various FS locks,
619 * and we don't want to recurse into the FS that called us
620 * in clear_inode() and friends..
622 if (!(gfp_mask
& __GFP_FS
))
626 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure
;
629 static struct shrinker icache_shrinker
= {
630 .shrink
= shrink_icache_memory
,
631 .seeks
= DEFAULT_SEEKS
,
634 static void __wait_on_freeing_inode(struct inode
*inode
);
636 * Called with the inode lock held.
637 * NOTE: we are not increasing the inode-refcount, you must call __iget()
638 * by hand after calling find_inode now! This simplifies iunique and won't
639 * add any additional branch in the common code.
641 static struct inode
*find_inode(struct super_block
*sb
,
642 struct hlist_head
*head
,
643 int (*test
)(struct inode
*, void *),
646 struct hlist_node
*node
;
647 struct inode
*inode
= NULL
;
650 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
651 if (inode
->i_sb
!= sb
)
653 if (!test(inode
, data
))
655 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
656 __wait_on_freeing_inode(inode
);
661 return node
? inode
: NULL
;
665 * find_inode_fast is the fast path version of find_inode, see the comment at
666 * iget_locked for details.
668 static struct inode
*find_inode_fast(struct super_block
*sb
,
669 struct hlist_head
*head
, unsigned long ino
)
671 struct hlist_node
*node
;
672 struct inode
*inode
= NULL
;
675 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
676 if (inode
->i_ino
!= ino
)
678 if (inode
->i_sb
!= sb
)
680 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
681 __wait_on_freeing_inode(inode
);
686 return node
? inode
: NULL
;
690 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
693 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
695 hlist_add_head(&inode
->i_hash
, head
);
699 * inode_add_to_lists - add a new inode to relevant lists
700 * @sb: superblock inode belongs to
701 * @inode: inode to mark in use
703 * When an inode is allocated it needs to be accounted for, added to the in use
704 * list, the owning superblock and the inode hash. This needs to be done under
705 * the inode_lock, so export a function to do this rather than the inode lock
706 * itself. We calculate the hash list to add to here so it is all internal
707 * which requires the caller to have already set up the inode number in the
710 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
712 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
714 spin_lock(&inode_lock
);
715 __inode_add_to_lists(sb
, head
, inode
);
716 spin_unlock(&inode_lock
);
718 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
721 * new_inode - obtain an inode
724 * Allocates a new inode for given superblock. The default gfp_mask
725 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
726 * If HIGHMEM pages are unsuitable or it is known that pages allocated
727 * for the page cache are not reclaimable or migratable,
728 * mapping_set_gfp_mask() must be called with suitable flags on the
729 * newly created inode's mapping
732 struct inode
*new_inode(struct super_block
*sb
)
735 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
736 * error if st_ino won't fit in target struct field. Use 32bit counter
737 * here to attempt to avoid that.
739 static unsigned int last_ino
;
742 spin_lock_prefetch(&inode_lock
);
744 inode
= alloc_inode(sb
);
746 spin_lock(&inode_lock
);
747 __inode_add_to_lists(sb
, NULL
, inode
);
748 inode
->i_ino
= ++last_ino
;
750 spin_unlock(&inode_lock
);
754 EXPORT_SYMBOL(new_inode
);
756 void unlock_new_inode(struct inode
*inode
)
758 #ifdef CONFIG_DEBUG_LOCK_ALLOC
759 if (S_ISDIR(inode
->i_mode
)) {
760 struct file_system_type
*type
= inode
->i_sb
->s_type
;
762 /* Set new key only if filesystem hasn't already changed it */
763 if (!lockdep_match_class(&inode
->i_mutex
,
764 &type
->i_mutex_key
)) {
766 * ensure nobody is actually holding i_mutex
768 mutex_destroy(&inode
->i_mutex
);
769 mutex_init(&inode
->i_mutex
);
770 lockdep_set_class(&inode
->i_mutex
,
771 &type
->i_mutex_dir_key
);
776 * This is special! We do not need the spinlock when clearing I_NEW,
777 * because we're guaranteed that nobody else tries to do anything about
778 * the state of the inode when it is locked, as we just created it (so
779 * there can be no old holders that haven't tested I_NEW).
780 * However we must emit the memory barrier so that other CPUs reliably
781 * see the clearing of I_NEW after the other inode initialisation has
785 WARN_ON(!(inode
->i_state
& I_NEW
));
786 inode
->i_state
&= ~I_NEW
;
787 wake_up_inode(inode
);
789 EXPORT_SYMBOL(unlock_new_inode
);
792 * This is called without the inode lock held.. Be careful.
794 * We no longer cache the sb_flags in i_flags - see fs.h
795 * -- rmk@arm.uk.linux.org
797 static struct inode
*get_new_inode(struct super_block
*sb
,
798 struct hlist_head
*head
,
799 int (*test
)(struct inode
*, void *),
800 int (*set
)(struct inode
*, void *),
805 inode
= alloc_inode(sb
);
809 spin_lock(&inode_lock
);
810 /* We released the lock, so.. */
811 old
= find_inode(sb
, head
, test
, data
);
813 if (set(inode
, data
))
816 __inode_add_to_lists(sb
, head
, inode
);
817 inode
->i_state
= I_NEW
;
818 spin_unlock(&inode_lock
);
820 /* Return the locked inode with I_NEW set, the
821 * caller is responsible for filling in the contents
827 * Uhhuh, somebody else created the same inode under
828 * us. Use the old inode instead of the one we just
832 spin_unlock(&inode_lock
);
833 destroy_inode(inode
);
835 wait_on_inode(inode
);
840 spin_unlock(&inode_lock
);
841 destroy_inode(inode
);
846 * get_new_inode_fast is the fast path version of get_new_inode, see the
847 * comment at iget_locked for details.
849 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
850 struct hlist_head
*head
, unsigned long ino
)
854 inode
= alloc_inode(sb
);
858 spin_lock(&inode_lock
);
859 /* We released the lock, so.. */
860 old
= find_inode_fast(sb
, head
, ino
);
863 __inode_add_to_lists(sb
, head
, inode
);
864 inode
->i_state
= I_NEW
;
865 spin_unlock(&inode_lock
);
867 /* Return the locked inode with I_NEW set, the
868 * caller is responsible for filling in the contents
874 * Uhhuh, somebody else created the same inode under
875 * us. Use the old inode instead of the one we just
879 spin_unlock(&inode_lock
);
880 destroy_inode(inode
);
882 wait_on_inode(inode
);
888 * iunique - get a unique inode number
890 * @max_reserved: highest reserved inode number
892 * Obtain an inode number that is unique on the system for a given
893 * superblock. This is used by file systems that have no natural
894 * permanent inode numbering system. An inode number is returned that
895 * is higher than the reserved limit but unique.
898 * With a large number of inodes live on the file system this function
899 * currently becomes quite slow.
901 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
904 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
905 * error if st_ino won't fit in target struct field. Use 32bit counter
906 * here to attempt to avoid that.
908 static unsigned int counter
;
910 struct hlist_head
*head
;
913 spin_lock(&inode_lock
);
915 if (counter
<= max_reserved
)
916 counter
= max_reserved
+ 1;
918 head
= inode_hashtable
+ hash(sb
, res
);
919 inode
= find_inode_fast(sb
, head
, res
);
920 } while (inode
!= NULL
);
921 spin_unlock(&inode_lock
);
925 EXPORT_SYMBOL(iunique
);
927 struct inode
*igrab(struct inode
*inode
)
929 spin_lock(&inode_lock
);
930 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
)))
934 * Handle the case where s_op->clear_inode is not been
935 * called yet, and somebody is calling igrab
936 * while the inode is getting freed.
939 spin_unlock(&inode_lock
);
942 EXPORT_SYMBOL(igrab
);
945 * ifind - internal function, you want ilookup5() or iget5().
946 * @sb: super block of file system to search
947 * @head: the head of the list to search
948 * @test: callback used for comparisons between inodes
949 * @data: opaque data pointer to pass to @test
950 * @wait: if true wait for the inode to be unlocked, if false do not
952 * ifind() searches for the inode specified by @data in the inode
953 * cache. This is a generalized version of ifind_fast() for file systems where
954 * the inode number is not sufficient for unique identification of an inode.
956 * If the inode is in the cache, the inode is returned with an incremented
959 * Otherwise NULL is returned.
961 * Note, @test is called with the inode_lock held, so can't sleep.
963 static struct inode
*ifind(struct super_block
*sb
,
964 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
965 void *data
, const int wait
)
969 spin_lock(&inode_lock
);
970 inode
= find_inode(sb
, head
, test
, data
);
973 spin_unlock(&inode_lock
);
975 wait_on_inode(inode
);
978 spin_unlock(&inode_lock
);
983 * ifind_fast - internal function, you want ilookup() or iget().
984 * @sb: super block of file system to search
985 * @head: head of the list to search
986 * @ino: inode number to search for
988 * ifind_fast() searches for the inode @ino in the inode cache. This is for
989 * file systems where the inode number is sufficient for unique identification
992 * If the inode is in the cache, the inode is returned with an incremented
995 * Otherwise NULL is returned.
997 static struct inode
*ifind_fast(struct super_block
*sb
,
998 struct hlist_head
*head
, unsigned long ino
)
1000 struct inode
*inode
;
1002 spin_lock(&inode_lock
);
1003 inode
= find_inode_fast(sb
, head
, ino
);
1006 spin_unlock(&inode_lock
);
1007 wait_on_inode(inode
);
1010 spin_unlock(&inode_lock
);
1015 * ilookup5_nowait - search for an inode in the inode cache
1016 * @sb: super block of file system to search
1017 * @hashval: hash value (usually inode number) to search for
1018 * @test: callback used for comparisons between inodes
1019 * @data: opaque data pointer to pass to @test
1021 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1022 * @data in the inode cache. This is a generalized version of ilookup() for
1023 * file systems where the inode number is not sufficient for unique
1024 * identification of an inode.
1026 * If the inode is in the cache, the inode is returned with an incremented
1027 * reference count. Note, the inode lock is not waited upon so you have to be
1028 * very careful what you do with the returned inode. You probably should be
1029 * using ilookup5() instead.
1031 * Otherwise NULL is returned.
1033 * Note, @test is called with the inode_lock held, so can't sleep.
1035 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1036 int (*test
)(struct inode
*, void *), void *data
)
1038 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1040 return ifind(sb
, head
, test
, data
, 0);
1042 EXPORT_SYMBOL(ilookup5_nowait
);
1045 * ilookup5 - search for an inode in the inode cache
1046 * @sb: super block of file system to search
1047 * @hashval: hash value (usually inode number) to search for
1048 * @test: callback used for comparisons between inodes
1049 * @data: opaque data pointer to pass to @test
1051 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1052 * @data in the inode cache. This is a generalized version of ilookup() for
1053 * file systems where the inode number is not sufficient for unique
1054 * identification of an inode.
1056 * If the inode is in the cache, the inode lock is waited upon and the inode is
1057 * returned with an incremented reference count.
1059 * Otherwise NULL is returned.
1061 * Note, @test is called with the inode_lock held, so can't sleep.
1063 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1064 int (*test
)(struct inode
*, void *), void *data
)
1066 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1068 return ifind(sb
, head
, test
, data
, 1);
1070 EXPORT_SYMBOL(ilookup5
);
1073 * ilookup - search for an inode in the inode cache
1074 * @sb: super block of file system to search
1075 * @ino: inode number to search for
1077 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1078 * This is for file systems where the inode number is sufficient for unique
1079 * identification of an inode.
1081 * If the inode is in the cache, the inode is returned with an incremented
1084 * Otherwise NULL is returned.
1086 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1088 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1090 return ifind_fast(sb
, head
, ino
);
1092 EXPORT_SYMBOL(ilookup
);
1095 * iget5_locked - obtain an inode from a mounted file system
1096 * @sb: super block of file system
1097 * @hashval: hash value (usually inode number) to get
1098 * @test: callback used for comparisons between inodes
1099 * @set: callback used to initialize a new struct inode
1100 * @data: opaque data pointer to pass to @test and @set
1102 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1103 * and @data in the inode cache and if present it is returned with an increased
1104 * reference count. This is a generalized version of iget_locked() for file
1105 * systems where the inode number is not sufficient for unique identification
1108 * If the inode is not in cache, get_new_inode() is called to allocate a new
1109 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1110 * file system gets to fill it in before unlocking it via unlock_new_inode().
1112 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1114 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1115 int (*test
)(struct inode
*, void *),
1116 int (*set
)(struct inode
*, void *), void *data
)
1118 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1119 struct inode
*inode
;
1121 inode
= ifind(sb
, head
, test
, data
, 1);
1125 * get_new_inode() will do the right thing, re-trying the search
1126 * in case it had to block at any point.
1128 return get_new_inode(sb
, head
, test
, set
, data
);
1130 EXPORT_SYMBOL(iget5_locked
);
1133 * iget_locked - obtain an inode from a mounted file system
1134 * @sb: super block of file system
1135 * @ino: inode number to get
1137 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1138 * the inode cache and if present it is returned with an increased reference
1139 * count. This is for file systems where the inode number is sufficient for
1140 * unique identification of an inode.
1142 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1143 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1144 * The file system gets to fill it in before unlocking it via
1145 * unlock_new_inode().
1147 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1149 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1150 struct inode
*inode
;
1152 inode
= ifind_fast(sb
, head
, ino
);
1156 * get_new_inode_fast() will do the right thing, re-trying the search
1157 * in case it had to block at any point.
1159 return get_new_inode_fast(sb
, head
, ino
);
1161 EXPORT_SYMBOL(iget_locked
);
1163 int insert_inode_locked(struct inode
*inode
)
1165 struct super_block
*sb
= inode
->i_sb
;
1166 ino_t ino
= inode
->i_ino
;
1167 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1169 inode
->i_state
|= I_NEW
;
1171 struct hlist_node
*node
;
1172 struct inode
*old
= NULL
;
1173 spin_lock(&inode_lock
);
1174 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1175 if (old
->i_ino
!= ino
)
1177 if (old
->i_sb
!= sb
)
1179 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
))
1183 if (likely(!node
)) {
1184 hlist_add_head(&inode
->i_hash
, head
);
1185 spin_unlock(&inode_lock
);
1189 spin_unlock(&inode_lock
);
1191 if (unlikely(!inode_unhashed(old
))) {
1198 EXPORT_SYMBOL(insert_inode_locked
);
1200 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1201 int (*test
)(struct inode
*, void *), void *data
)
1203 struct super_block
*sb
= inode
->i_sb
;
1204 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1206 inode
->i_state
|= I_NEW
;
1209 struct hlist_node
*node
;
1210 struct inode
*old
= NULL
;
1212 spin_lock(&inode_lock
);
1213 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1214 if (old
->i_sb
!= sb
)
1216 if (!test(old
, data
))
1218 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
))
1222 if (likely(!node
)) {
1223 hlist_add_head(&inode
->i_hash
, head
);
1224 spin_unlock(&inode_lock
);
1228 spin_unlock(&inode_lock
);
1230 if (unlikely(!inode_unhashed(old
))) {
1237 EXPORT_SYMBOL(insert_inode_locked4
);
1240 int generic_delete_inode(struct inode
*inode
)
1244 EXPORT_SYMBOL(generic_delete_inode
);
1247 * Normal UNIX filesystem behaviour: delete the
1248 * inode when the usage count drops to zero, and
1251 int generic_drop_inode(struct inode
*inode
)
1253 return !inode
->i_nlink
|| inode_unhashed(inode
);
1255 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1258 * Called when we're dropping the last reference
1261 * Call the FS "drop_inode()" function, defaulting to
1262 * the legacy UNIX filesystem behaviour. If it tells
1263 * us to evict inode, do so. Otherwise, retain inode
1264 * in cache if fs is alive, sync and evict if fs is
1267 static void iput_final(struct inode
*inode
)
1269 struct super_block
*sb
= inode
->i_sb
;
1270 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1273 if (op
&& op
->drop_inode
)
1274 drop
= op
->drop_inode(inode
);
1276 drop
= generic_drop_inode(inode
);
1279 if (sb
->s_flags
& MS_ACTIVE
) {
1280 inode
->i_state
|= I_REFERENCED
;
1281 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
))) {
1282 inode_lru_list_add(inode
);
1284 spin_unlock(&inode_lock
);
1287 WARN_ON(inode
->i_state
& I_NEW
);
1288 inode
->i_state
|= I_WILL_FREE
;
1289 spin_unlock(&inode_lock
);
1290 write_inode_now(inode
, 1);
1291 spin_lock(&inode_lock
);
1292 WARN_ON(inode
->i_state
& I_NEW
);
1293 inode
->i_state
&= ~I_WILL_FREE
;
1294 __remove_inode_hash(inode
);
1296 WARN_ON(inode
->i_state
& I_NEW
);
1297 inode
->i_state
|= I_FREEING
;
1300 * After we delete the inode from the LRU here, we avoid moving dirty
1301 * inodes back onto the LRU now because I_FREEING is set and hence
1302 * writeback_single_inode() won't move the inode around.
1304 inode_lru_list_del(inode
);
1306 list_del_init(&inode
->i_sb_list
);
1307 spin_unlock(&inode_lock
);
1309 remove_inode_hash(inode
);
1310 wake_up_inode(inode
);
1311 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
1312 destroy_inode(inode
);
1316 * iput - put an inode
1317 * @inode: inode to put
1319 * Puts an inode, dropping its usage count. If the inode use count hits
1320 * zero, the inode is then freed and may also be destroyed.
1322 * Consequently, iput() can sleep.
1324 void iput(struct inode
*inode
)
1327 BUG_ON(inode
->i_state
& I_CLEAR
);
1329 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1333 EXPORT_SYMBOL(iput
);
1336 * bmap - find a block number in a file
1337 * @inode: inode of file
1338 * @block: block to find
1340 * Returns the block number on the device holding the inode that
1341 * is the disk block number for the block of the file requested.
1342 * That is, asked for block 4 of inode 1 the function will return the
1343 * disk block relative to the disk start that holds that block of the
1346 sector_t
bmap(struct inode
*inode
, sector_t block
)
1349 if (inode
->i_mapping
->a_ops
->bmap
)
1350 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1353 EXPORT_SYMBOL(bmap
);
1356 * With relative atime, only update atime if the previous atime is
1357 * earlier than either the ctime or mtime or if at least a day has
1358 * passed since the last atime update.
1360 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1361 struct timespec now
)
1364 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1367 * Is mtime younger than atime? If yes, update atime:
1369 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1372 * Is ctime younger than atime? If yes, update atime:
1374 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1378 * Is the previous atime value older than a day? If yes,
1381 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1384 * Good, we can skip the atime update:
1390 * touch_atime - update the access time
1391 * @mnt: mount the inode is accessed on
1392 * @dentry: dentry accessed
1394 * Update the accessed time on an inode and mark it for writeback.
1395 * This function automatically handles read only file systems and media,
1396 * as well as the "noatime" flag and inode specific "noatime" markers.
1398 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1400 struct inode
*inode
= dentry
->d_inode
;
1401 struct timespec now
;
1403 if (inode
->i_flags
& S_NOATIME
)
1405 if (IS_NOATIME(inode
))
1407 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1410 if (mnt
->mnt_flags
& MNT_NOATIME
)
1412 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1415 now
= current_fs_time(inode
->i_sb
);
1417 if (!relatime_need_update(mnt
, inode
, now
))
1420 if (timespec_equal(&inode
->i_atime
, &now
))
1423 if (mnt_want_write(mnt
))
1426 inode
->i_atime
= now
;
1427 mark_inode_dirty_sync(inode
);
1428 mnt_drop_write(mnt
);
1430 EXPORT_SYMBOL(touch_atime
);
1433 * file_update_time - update mtime and ctime time
1434 * @file: file accessed
1436 * Update the mtime and ctime members of an inode and mark the inode
1437 * for writeback. Note that this function is meant exclusively for
1438 * usage in the file write path of filesystems, and filesystems may
1439 * choose to explicitly ignore update via this function with the
1440 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1441 * timestamps are handled by the server.
1444 void file_update_time(struct file
*file
)
1446 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1447 struct timespec now
;
1448 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1450 /* First try to exhaust all avenues to not sync */
1451 if (IS_NOCMTIME(inode
))
1454 now
= current_fs_time(inode
->i_sb
);
1455 if (!timespec_equal(&inode
->i_mtime
, &now
))
1458 if (!timespec_equal(&inode
->i_ctime
, &now
))
1461 if (IS_I_VERSION(inode
))
1462 sync_it
|= S_VERSION
;
1467 /* Finally allowed to write? Takes lock. */
1468 if (mnt_want_write_file(file
))
1471 /* Only change inode inside the lock region */
1472 if (sync_it
& S_VERSION
)
1473 inode_inc_iversion(inode
);
1474 if (sync_it
& S_CTIME
)
1475 inode
->i_ctime
= now
;
1476 if (sync_it
& S_MTIME
)
1477 inode
->i_mtime
= now
;
1478 mark_inode_dirty_sync(inode
);
1479 mnt_drop_write(file
->f_path
.mnt
);
1481 EXPORT_SYMBOL(file_update_time
);
1483 int inode_needs_sync(struct inode
*inode
)
1487 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1491 EXPORT_SYMBOL(inode_needs_sync
);
1493 int inode_wait(void *word
)
1498 EXPORT_SYMBOL(inode_wait
);
1501 * If we try to find an inode in the inode hash while it is being
1502 * deleted, we have to wait until the filesystem completes its
1503 * deletion before reporting that it isn't found. This function waits
1504 * until the deletion _might_ have completed. Callers are responsible
1505 * to recheck inode state.
1507 * It doesn't matter if I_NEW is not set initially, a call to
1508 * wake_up_inode() after removing from the hash list will DTRT.
1510 * This is called with inode_lock held.
1512 static void __wait_on_freeing_inode(struct inode
*inode
)
1514 wait_queue_head_t
*wq
;
1515 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1516 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1517 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1518 spin_unlock(&inode_lock
);
1520 finish_wait(wq
, &wait
.wait
);
1521 spin_lock(&inode_lock
);
1524 static __initdata
unsigned long ihash_entries
;
1525 static int __init
set_ihash_entries(char *str
)
1529 ihash_entries
= simple_strtoul(str
, &str
, 0);
1532 __setup("ihash_entries=", set_ihash_entries
);
1535 * Initialize the waitqueues and inode hash table.
1537 void __init
inode_init_early(void)
1541 /* If hashes are distributed across NUMA nodes, defer
1542 * hash allocation until vmalloc space is available.
1548 alloc_large_system_hash("Inode-cache",
1549 sizeof(struct hlist_head
),
1557 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1558 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1561 void __init
inode_init(void)
1565 /* inode slab cache */
1566 inode_cachep
= kmem_cache_create("inode_cache",
1567 sizeof(struct inode
),
1569 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1572 register_shrinker(&icache_shrinker
);
1573 percpu_counter_init(&nr_inodes
, 0);
1574 percpu_counter_init(&nr_inodes_unused
, 0);
1576 /* Hash may have been set up in inode_init_early */
1581 alloc_large_system_hash("Inode-cache",
1582 sizeof(struct hlist_head
),
1590 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1591 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1594 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1596 inode
->i_mode
= mode
;
1597 if (S_ISCHR(mode
)) {
1598 inode
->i_fop
= &def_chr_fops
;
1599 inode
->i_rdev
= rdev
;
1600 } else if (S_ISBLK(mode
)) {
1601 inode
->i_fop
= &def_blk_fops
;
1602 inode
->i_rdev
= rdev
;
1603 } else if (S_ISFIFO(mode
))
1604 inode
->i_fop
= &def_fifo_fops
;
1605 else if (S_ISSOCK(mode
))
1606 inode
->i_fop
= &bad_sock_fops
;
1608 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1609 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1612 EXPORT_SYMBOL(init_special_inode
);
1615 * Init uid,gid,mode for new inode according to posix standards
1617 * @dir: Directory inode
1618 * @mode: mode of the new inode
1620 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1623 inode
->i_uid
= current_fsuid();
1624 if (dir
&& dir
->i_mode
& S_ISGID
) {
1625 inode
->i_gid
= dir
->i_gid
;
1629 inode
->i_gid
= current_fsgid();
1630 inode
->i_mode
= mode
;
1632 EXPORT_SYMBOL(inode_init_owner
);