2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
7 #include <linux/dcache.h>
8 #include <linux/init.h>
9 #include <linux/slab.h>
10 #include <linux/writeback.h>
11 #include <linux/module.h>
12 #include <linux/backing-dev.h>
13 #include <linux/wait.h>
14 #include <linux/rwsem.h>
15 #include <linux/hash.h>
16 #include <linux/swap.h>
17 #include <linux/security.h>
18 #include <linux/pagemap.h>
19 #include <linux/cdev.h>
20 #include <linux/bootmem.h>
21 #include <linux/fsnotify.h>
22 #include <linux/mount.h>
23 #include <linux/async.h>
24 #include <linux/posix_acl.h>
25 #include <linux/prefetch.h>
26 #include <linux/ima.h>
27 #include <linux/cred.h>
28 #include <linux/buffer_head.h> /* for inode_has_buffers */
32 * Inode locking rules:
34 * inode->i_lock protects:
35 * inode->i_state, inode->i_hash, __iget()
36 * inode->i_sb->s_inode_lru_lock protects:
37 * inode->i_sb->s_inode_lru, inode->i_lru
38 * inode_sb_list_lock protects:
39 * sb->s_inodes, inode->i_sb_list
40 * inode_wb_list_lock protects:
41 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
42 * inode_hash_lock protects:
43 * inode_hashtable, inode->i_hash
49 * inode->i_sb->s_inode_lru_lock
62 static unsigned int i_hash_mask __read_mostly
;
63 static unsigned int i_hash_shift __read_mostly
;
64 static struct hlist_head
*inode_hashtable __read_mostly
;
65 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_hash_lock
);
67 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_sb_list_lock
);
68 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_wb_list_lock
);
71 * iprune_sem provides exclusion between the icache shrinking and the
74 * We don't actually need it to protect anything in the umount path,
75 * but only need to cycle through it to make sure any inode that
76 * prune_icache took off the LRU list has been fully torn down by the
77 * time we are past evict_inodes.
79 static DECLARE_RWSEM(iprune_sem
);
82 * Empty aops. Can be used for the cases where the user does not
83 * define any of the address_space operations.
85 const struct address_space_operations empty_aops
= {
87 EXPORT_SYMBOL(empty_aops
);
90 * Statistics gathering..
92 struct inodes_stat_t inodes_stat
;
94 static DEFINE_PER_CPU(unsigned int, nr_inodes
);
95 static DEFINE_PER_CPU(unsigned int, nr_unused
);
97 static struct kmem_cache
*inode_cachep __read_mostly
;
99 static int get_nr_inodes(void)
103 for_each_possible_cpu(i
)
104 sum
+= per_cpu(nr_inodes
, i
);
105 return sum
< 0 ? 0 : sum
;
108 static inline int get_nr_inodes_unused(void)
112 for_each_possible_cpu(i
)
113 sum
+= per_cpu(nr_unused
, i
);
114 return sum
< 0 ? 0 : sum
;
117 int get_nr_dirty_inodes(void)
119 /* not actually dirty inodes, but a wild approximation */
120 int nr_dirty
= get_nr_inodes() - get_nr_inodes_unused();
121 return nr_dirty
> 0 ? nr_dirty
: 0;
125 * Handle nr_inode sysctl
128 int proc_nr_inodes(ctl_table
*table
, int write
,
129 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
131 inodes_stat
.nr_inodes
= get_nr_inodes();
132 inodes_stat
.nr_unused
= get_nr_inodes_unused();
133 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
138 * inode_init_always - perform inode structure intialisation
139 * @sb: superblock inode belongs to
140 * @inode: inode to initialise
142 * These are initializations that need to be done on every inode
143 * allocation as the fields are not initialised by slab allocation.
145 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
147 static const struct inode_operations empty_iops
;
148 static const struct file_operations empty_fops
;
149 struct address_space
*const mapping
= &inode
->i_data
;
152 inode
->i_blkbits
= sb
->s_blocksize_bits
;
154 atomic_set(&inode
->i_count
, 1);
155 inode
->i_op
= &empty_iops
;
156 inode
->i_fop
= &empty_fops
;
160 atomic_set(&inode
->i_writecount
, 0);
164 inode
->i_generation
= 0;
166 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
168 inode
->i_pipe
= NULL
;
169 inode
->i_bdev
= NULL
;
170 inode
->i_cdev
= NULL
;
172 inode
->dirtied_when
= 0;
174 if (security_inode_alloc(inode
))
176 spin_lock_init(&inode
->i_lock
);
177 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
179 mutex_init(&inode
->i_mutex
);
180 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
182 init_rwsem(&inode
->i_alloc_sem
);
183 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
185 mapping
->a_ops
= &empty_aops
;
186 mapping
->host
= inode
;
188 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
189 mapping
->assoc_mapping
= NULL
;
190 mapping
->backing_dev_info
= &default_backing_dev_info
;
191 mapping
->writeback_index
= 0;
194 * If the block_device provides a backing_dev_info for client
195 * inodes then use that. Otherwise the inode share the bdev's
199 struct backing_dev_info
*bdi
;
201 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
202 mapping
->backing_dev_info
= bdi
;
204 inode
->i_private
= NULL
;
205 inode
->i_mapping
= mapping
;
206 #ifdef CONFIG_FS_POSIX_ACL
207 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
210 #ifdef CONFIG_FSNOTIFY
211 inode
->i_fsnotify_mask
= 0;
214 this_cpu_inc(nr_inodes
);
220 EXPORT_SYMBOL(inode_init_always
);
222 static struct inode
*alloc_inode(struct super_block
*sb
)
226 if (sb
->s_op
->alloc_inode
)
227 inode
= sb
->s_op
->alloc_inode(sb
);
229 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
234 if (unlikely(inode_init_always(sb
, inode
))) {
235 if (inode
->i_sb
->s_op
->destroy_inode
)
236 inode
->i_sb
->s_op
->destroy_inode(inode
);
238 kmem_cache_free(inode_cachep
, inode
);
245 void free_inode_nonrcu(struct inode
*inode
)
247 kmem_cache_free(inode_cachep
, inode
);
249 EXPORT_SYMBOL(free_inode_nonrcu
);
251 void __destroy_inode(struct inode
*inode
)
253 BUG_ON(inode_has_buffers(inode
));
254 security_inode_free(inode
);
255 fsnotify_inode_delete(inode
);
256 #ifdef CONFIG_FS_POSIX_ACL
257 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
258 posix_acl_release(inode
->i_acl
);
259 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
260 posix_acl_release(inode
->i_default_acl
);
262 this_cpu_dec(nr_inodes
);
264 EXPORT_SYMBOL(__destroy_inode
);
266 static void i_callback(struct rcu_head
*head
)
268 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
269 INIT_LIST_HEAD(&inode
->i_dentry
);
270 kmem_cache_free(inode_cachep
, inode
);
273 static void destroy_inode(struct inode
*inode
)
275 BUG_ON(!list_empty(&inode
->i_lru
));
276 __destroy_inode(inode
);
277 if (inode
->i_sb
->s_op
->destroy_inode
)
278 inode
->i_sb
->s_op
->destroy_inode(inode
);
280 call_rcu(&inode
->i_rcu
, i_callback
);
283 void address_space_init_once(struct address_space
*mapping
)
285 memset(mapping
, 0, sizeof(*mapping
));
286 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
287 spin_lock_init(&mapping
->tree_lock
);
288 mutex_init(&mapping
->i_mmap_mutex
);
289 INIT_LIST_HEAD(&mapping
->private_list
);
290 spin_lock_init(&mapping
->private_lock
);
291 INIT_RAW_PRIO_TREE_ROOT(&mapping
->i_mmap
);
292 INIT_LIST_HEAD(&mapping
->i_mmap_nonlinear
);
294 EXPORT_SYMBOL(address_space_init_once
);
297 * These are initializations that only need to be done
298 * once, because the fields are idempotent across use
299 * of the inode, so let the slab aware of that.
301 void inode_init_once(struct inode
*inode
)
303 memset(inode
, 0, sizeof(*inode
));
304 INIT_HLIST_NODE(&inode
->i_hash
);
305 INIT_LIST_HEAD(&inode
->i_dentry
);
306 INIT_LIST_HEAD(&inode
->i_devices
);
307 INIT_LIST_HEAD(&inode
->i_wb_list
);
308 INIT_LIST_HEAD(&inode
->i_lru
);
309 address_space_init_once(&inode
->i_data
);
310 i_size_ordered_init(inode
);
311 #ifdef CONFIG_FSNOTIFY
312 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
315 EXPORT_SYMBOL(inode_init_once
);
317 static void init_once(void *foo
)
319 struct inode
*inode
= (struct inode
*) foo
;
321 inode_init_once(inode
);
325 * inode->i_lock must be held
327 void __iget(struct inode
*inode
)
329 atomic_inc(&inode
->i_count
);
333 * get additional reference to inode; caller must already hold one.
335 void ihold(struct inode
*inode
)
337 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
339 EXPORT_SYMBOL(ihold
);
341 static void inode_lru_list_add(struct inode
*inode
)
343 spin_lock(&inode
->i_sb
->s_inode_lru_lock
);
344 if (list_empty(&inode
->i_lru
)) {
345 list_add(&inode
->i_lru
, &inode
->i_sb
->s_inode_lru
);
346 inode
->i_sb
->s_nr_inodes_unused
++;
347 this_cpu_inc(nr_unused
);
349 spin_unlock(&inode
->i_sb
->s_inode_lru_lock
);
352 static void inode_lru_list_del(struct inode
*inode
)
354 spin_lock(&inode
->i_sb
->s_inode_lru_lock
);
355 if (!list_empty(&inode
->i_lru
)) {
356 list_del_init(&inode
->i_lru
);
357 inode
->i_sb
->s_nr_inodes_unused
--;
358 this_cpu_dec(nr_unused
);
360 spin_unlock(&inode
->i_sb
->s_inode_lru_lock
);
364 * inode_sb_list_add - add inode to the superblock list of inodes
365 * @inode: inode to add
367 void inode_sb_list_add(struct inode
*inode
)
369 spin_lock(&inode_sb_list_lock
);
370 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
371 spin_unlock(&inode_sb_list_lock
);
373 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
375 static inline void inode_sb_list_del(struct inode
*inode
)
377 spin_lock(&inode_sb_list_lock
);
378 list_del_init(&inode
->i_sb_list
);
379 spin_unlock(&inode_sb_list_lock
);
382 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
386 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
388 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> i_hash_shift
);
389 return tmp
& i_hash_mask
;
393 * __insert_inode_hash - hash an inode
394 * @inode: unhashed inode
395 * @hashval: unsigned long value used to locate this object in the
398 * Add an inode to the inode hash for this superblock.
400 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
402 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
404 spin_lock(&inode_hash_lock
);
405 spin_lock(&inode
->i_lock
);
406 hlist_add_head(&inode
->i_hash
, b
);
407 spin_unlock(&inode
->i_lock
);
408 spin_unlock(&inode_hash_lock
);
410 EXPORT_SYMBOL(__insert_inode_hash
);
413 * remove_inode_hash - remove an inode from the hash
414 * @inode: inode to unhash
416 * Remove an inode from the superblock.
418 void remove_inode_hash(struct inode
*inode
)
420 spin_lock(&inode_hash_lock
);
421 spin_lock(&inode
->i_lock
);
422 hlist_del_init(&inode
->i_hash
);
423 spin_unlock(&inode
->i_lock
);
424 spin_unlock(&inode_hash_lock
);
426 EXPORT_SYMBOL(remove_inode_hash
);
428 void end_writeback(struct inode
*inode
)
432 * We have to cycle tree_lock here because reclaim can be still in the
433 * process of removing the last page (in __delete_from_page_cache())
434 * and we must not free mapping under it.
436 spin_lock_irq(&inode
->i_data
.tree_lock
);
437 BUG_ON(inode
->i_data
.nrpages
);
438 spin_unlock_irq(&inode
->i_data
.tree_lock
);
439 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
440 BUG_ON(!(inode
->i_state
& I_FREEING
));
441 BUG_ON(inode
->i_state
& I_CLEAR
);
442 inode_sync_wait(inode
);
443 /* don't need i_lock here, no concurrent mods to i_state */
444 inode
->i_state
= I_FREEING
| I_CLEAR
;
446 EXPORT_SYMBOL(end_writeback
);
449 * Free the inode passed in, removing it from the lists it is still connected
450 * to. We remove any pages still attached to the inode and wait for any IO that
451 * is still in progress before finally destroying the inode.
453 * An inode must already be marked I_FREEING so that we avoid the inode being
454 * moved back onto lists if we race with other code that manipulates the lists
455 * (e.g. writeback_single_inode). The caller is responsible for setting this.
457 * An inode must already be removed from the LRU list before being evicted from
458 * the cache. This should occur atomically with setting the I_FREEING state
459 * flag, so no inodes here should ever be on the LRU when being evicted.
461 static void evict(struct inode
*inode
)
463 const struct super_operations
*op
= inode
->i_sb
->s_op
;
465 BUG_ON(!(inode
->i_state
& I_FREEING
));
466 BUG_ON(!list_empty(&inode
->i_lru
));
468 inode_wb_list_del(inode
);
469 inode_sb_list_del(inode
);
471 if (op
->evict_inode
) {
472 op
->evict_inode(inode
);
474 if (inode
->i_data
.nrpages
)
475 truncate_inode_pages(&inode
->i_data
, 0);
476 end_writeback(inode
);
478 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
480 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
483 remove_inode_hash(inode
);
485 spin_lock(&inode
->i_lock
);
486 wake_up_bit(&inode
->i_state
, __I_NEW
);
487 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
488 spin_unlock(&inode
->i_lock
);
490 destroy_inode(inode
);
494 * dispose_list - dispose of the contents of a local list
495 * @head: the head of the list to free
497 * Dispose-list gets a local list with local inodes in it, so it doesn't
498 * need to worry about list corruption and SMP locks.
500 static void dispose_list(struct list_head
*head
)
502 while (!list_empty(head
)) {
505 inode
= list_first_entry(head
, struct inode
, i_lru
);
506 list_del_init(&inode
->i_lru
);
513 * evict_inodes - evict all evictable inodes for a superblock
514 * @sb: superblock to operate on
516 * Make sure that no inodes with zero refcount are retained. This is
517 * called by superblock shutdown after having MS_ACTIVE flag removed,
518 * so any inode reaching zero refcount during or after that call will
519 * be immediately evicted.
521 void evict_inodes(struct super_block
*sb
)
523 struct inode
*inode
, *next
;
526 spin_lock(&inode_sb_list_lock
);
527 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
528 if (atomic_read(&inode
->i_count
))
531 spin_lock(&inode
->i_lock
);
532 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
533 spin_unlock(&inode
->i_lock
);
537 inode
->i_state
|= I_FREEING
;
538 inode_lru_list_del(inode
);
539 spin_unlock(&inode
->i_lock
);
540 list_add(&inode
->i_lru
, &dispose
);
542 spin_unlock(&inode_sb_list_lock
);
544 dispose_list(&dispose
);
547 * Cycle through iprune_sem to make sure any inode that prune_icache
548 * moved off the list before we took the lock has been fully torn
551 down_write(&iprune_sem
);
552 up_write(&iprune_sem
);
556 * invalidate_inodes - attempt to free all inodes on a superblock
557 * @sb: superblock to operate on
558 * @kill_dirty: flag to guide handling of dirty inodes
560 * Attempts to free all inodes for a given superblock. If there were any
561 * busy inodes return a non-zero value, else zero.
562 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
565 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
568 struct inode
*inode
, *next
;
571 spin_lock(&inode_sb_list_lock
);
572 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
573 spin_lock(&inode
->i_lock
);
574 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
575 spin_unlock(&inode
->i_lock
);
578 if (inode
->i_state
& I_DIRTY
&& !kill_dirty
) {
579 spin_unlock(&inode
->i_lock
);
583 if (atomic_read(&inode
->i_count
)) {
584 spin_unlock(&inode
->i_lock
);
589 inode
->i_state
|= I_FREEING
;
590 inode_lru_list_del(inode
);
591 spin_unlock(&inode
->i_lock
);
592 list_add(&inode
->i_lru
, &dispose
);
594 spin_unlock(&inode_sb_list_lock
);
596 dispose_list(&dispose
);
601 static int can_unuse(struct inode
*inode
)
603 if (inode
->i_state
& ~I_REFERENCED
)
605 if (inode_has_buffers(inode
))
607 if (atomic_read(&inode
->i_count
))
609 if (inode
->i_data
.nrpages
)
615 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
616 * temporary list and then are freed outside sb->s_inode_lru_lock by
619 * Any inodes which are pinned purely because of attached pagecache have their
620 * pagecache removed. If the inode has metadata buffers attached to
621 * mapping->private_list then try to remove them.
623 * If the inode has the I_REFERENCED flag set, then it means that it has been
624 * used recently - the flag is set in iput_final(). When we encounter such an
625 * inode, clear the flag and move it to the back of the LRU so it gets another
626 * pass through the LRU before it gets reclaimed. This is necessary because of
627 * the fact we are doing lazy LRU updates to minimise lock contention so the
628 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
629 * with this flag set because they are the inodes that are out of order.
631 static void shrink_icache_sb(struct super_block
*sb
, int *nr_to_scan
)
635 unsigned long reap
= 0;
637 spin_lock(&sb
->s_inode_lru_lock
);
638 for (nr_scanned
= *nr_to_scan
; nr_scanned
>= 0; nr_scanned
--) {
641 if (list_empty(&sb
->s_inode_lru
))
644 inode
= list_entry(sb
->s_inode_lru
.prev
, struct inode
, i_lru
);
647 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
648 * so use a trylock. If we fail to get the lock, just move the
649 * inode to the back of the list so we don't spin on it.
651 if (!spin_trylock(&inode
->i_lock
)) {
652 list_move(&inode
->i_lru
, &sb
->s_inode_lru
);
657 * Referenced or dirty inodes are still in use. Give them
658 * another pass through the LRU as we canot reclaim them now.
660 if (atomic_read(&inode
->i_count
) ||
661 (inode
->i_state
& ~I_REFERENCED
)) {
662 list_del_init(&inode
->i_lru
);
663 spin_unlock(&inode
->i_lock
);
664 sb
->s_nr_inodes_unused
--;
665 this_cpu_dec(nr_unused
);
669 /* recently referenced inodes get one more pass */
670 if (inode
->i_state
& I_REFERENCED
) {
671 inode
->i_state
&= ~I_REFERENCED
;
672 list_move(&inode
->i_lru
, &sb
->s_inode_lru
);
673 spin_unlock(&inode
->i_lock
);
676 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
678 spin_unlock(&inode
->i_lock
);
679 spin_unlock(&sb
->s_inode_lru_lock
);
680 if (remove_inode_buffers(inode
))
681 reap
+= invalidate_mapping_pages(&inode
->i_data
,
684 spin_lock(&sb
->s_inode_lru_lock
);
686 if (inode
!= list_entry(sb
->s_inode_lru
.next
,
687 struct inode
, i_lru
))
688 continue; /* wrong inode or list_empty */
689 /* avoid lock inversions with trylock */
690 if (!spin_trylock(&inode
->i_lock
))
692 if (!can_unuse(inode
)) {
693 spin_unlock(&inode
->i_lock
);
697 WARN_ON(inode
->i_state
& I_NEW
);
698 inode
->i_state
|= I_FREEING
;
699 spin_unlock(&inode
->i_lock
);
701 list_move(&inode
->i_lru
, &freeable
);
702 sb
->s_nr_inodes_unused
--;
703 this_cpu_dec(nr_unused
);
705 if (current_is_kswapd())
706 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
708 __count_vm_events(PGINODESTEAL
, reap
);
709 spin_unlock(&sb
->s_inode_lru_lock
);
710 *nr_to_scan
= nr_scanned
;
712 dispose_list(&freeable
);
715 static void prune_icache(int count
)
717 struct super_block
*sb
, *p
= NULL
;
719 int unused
= inodes_stat
.nr_unused
;
723 if (unused
== 0 || count
== 0)
725 down_read(&iprune_sem
);
729 prune_ratio
= unused
/ count
;
731 list_for_each_entry(sb
, &super_blocks
, s_list
) {
732 if (list_empty(&sb
->s_instances
))
734 if (sb
->s_nr_inodes_unused
== 0)
737 /* Now, we reclaim unused dentrins with fairness.
738 * We reclaim them same percentage from each superblock.
739 * We calculate number of dentries to scan on this sb
740 * as follows, but the implementation is arranged to avoid
742 * number of dentries to scan on this sb =
743 * count * (number of dentries on this sb /
744 * number of dentries in the machine)
746 spin_unlock(&sb_lock
);
747 if (prune_ratio
!= 1)
748 w_count
= (sb
->s_nr_inodes_unused
/ prune_ratio
) + 1;
750 w_count
= sb
->s_nr_inodes_unused
;
753 * We need to be sure this filesystem isn't being unmounted,
754 * otherwise we could race with generic_shutdown_super(), and
755 * end up holding a reference to an inode while the filesystem
756 * is unmounted. So we try to get s_umount, and make sure
759 if (down_read_trylock(&sb
->s_umount
)) {
760 if ((sb
->s_root
!= NULL
) &&
761 (!list_empty(&sb
->s_dentry_lru
))) {
762 shrink_icache_sb(sb
, &w_count
);
765 up_read(&sb
->s_umount
);
772 /* more work left to do? */
778 spin_unlock(&sb_lock
);
779 up_read(&iprune_sem
);
783 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
784 * "unused" means that no dentries are referring to the inodes: the files are
785 * not open and the dcache references to those inodes have already been
788 * This function is passed the number of inodes to scan, and it returns the
789 * total number of remaining possibly-reclaimable inodes.
791 static int shrink_icache_memory(struct shrinker
*shrink
,
792 struct shrink_control
*sc
)
794 int nr
= sc
->nr_to_scan
;
795 gfp_t gfp_mask
= sc
->gfp_mask
;
799 * Nasty deadlock avoidance. We may hold various FS locks,
800 * and we don't want to recurse into the FS that called us
801 * in clear_inode() and friends..
803 if (!(gfp_mask
& __GFP_FS
))
807 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure
;
810 static struct shrinker icache_shrinker
= {
811 .shrink
= shrink_icache_memory
,
812 .seeks
= DEFAULT_SEEKS
,
815 static void __wait_on_freeing_inode(struct inode
*inode
);
817 * Called with the inode lock held.
819 static struct inode
*find_inode(struct super_block
*sb
,
820 struct hlist_head
*head
,
821 int (*test
)(struct inode
*, void *),
824 struct hlist_node
*node
;
825 struct inode
*inode
= NULL
;
828 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
829 spin_lock(&inode
->i_lock
);
830 if (inode
->i_sb
!= sb
) {
831 spin_unlock(&inode
->i_lock
);
834 if (!test(inode
, data
)) {
835 spin_unlock(&inode
->i_lock
);
838 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
839 __wait_on_freeing_inode(inode
);
843 spin_unlock(&inode
->i_lock
);
850 * find_inode_fast is the fast path version of find_inode, see the comment at
851 * iget_locked for details.
853 static struct inode
*find_inode_fast(struct super_block
*sb
,
854 struct hlist_head
*head
, unsigned long ino
)
856 struct hlist_node
*node
;
857 struct inode
*inode
= NULL
;
860 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
861 spin_lock(&inode
->i_lock
);
862 if (inode
->i_ino
!= ino
) {
863 spin_unlock(&inode
->i_lock
);
866 if (inode
->i_sb
!= sb
) {
867 spin_unlock(&inode
->i_lock
);
870 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
871 __wait_on_freeing_inode(inode
);
875 spin_unlock(&inode
->i_lock
);
882 * Each cpu owns a range of LAST_INO_BATCH numbers.
883 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
884 * to renew the exhausted range.
886 * This does not significantly increase overflow rate because every CPU can
887 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
888 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
889 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
890 * overflow rate by 2x, which does not seem too significant.
892 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
893 * error if st_ino won't fit in target struct field. Use 32bit counter
894 * here to attempt to avoid that.
896 #define LAST_INO_BATCH 1024
897 static DEFINE_PER_CPU(unsigned int, last_ino
);
899 unsigned int get_next_ino(void)
901 unsigned int *p
= &get_cpu_var(last_ino
);
902 unsigned int res
= *p
;
905 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
906 static atomic_t shared_last_ino
;
907 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
909 res
= next
- LAST_INO_BATCH
;
914 put_cpu_var(last_ino
);
917 EXPORT_SYMBOL(get_next_ino
);
920 * new_inode - obtain an inode
923 * Allocates a new inode for given superblock. The default gfp_mask
924 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
925 * If HIGHMEM pages are unsuitable or it is known that pages allocated
926 * for the page cache are not reclaimable or migratable,
927 * mapping_set_gfp_mask() must be called with suitable flags on the
928 * newly created inode's mapping
931 struct inode
*new_inode(struct super_block
*sb
)
935 spin_lock_prefetch(&inode_sb_list_lock
);
937 inode
= alloc_inode(sb
);
939 spin_lock(&inode
->i_lock
);
941 spin_unlock(&inode
->i_lock
);
942 inode_sb_list_add(inode
);
946 EXPORT_SYMBOL(new_inode
);
949 * unlock_new_inode - clear the I_NEW state and wake up any waiters
950 * @inode: new inode to unlock
952 * Called when the inode is fully initialised to clear the new state of the
953 * inode and wake up anyone waiting for the inode to finish initialisation.
955 void unlock_new_inode(struct inode
*inode
)
957 #ifdef CONFIG_DEBUG_LOCK_ALLOC
958 if (S_ISDIR(inode
->i_mode
)) {
959 struct file_system_type
*type
= inode
->i_sb
->s_type
;
961 /* Set new key only if filesystem hasn't already changed it */
962 if (!lockdep_match_class(&inode
->i_mutex
,
963 &type
->i_mutex_key
)) {
965 * ensure nobody is actually holding i_mutex
967 mutex_destroy(&inode
->i_mutex
);
968 mutex_init(&inode
->i_mutex
);
969 lockdep_set_class(&inode
->i_mutex
,
970 &type
->i_mutex_dir_key
);
974 spin_lock(&inode
->i_lock
);
975 WARN_ON(!(inode
->i_state
& I_NEW
));
976 inode
->i_state
&= ~I_NEW
;
977 wake_up_bit(&inode
->i_state
, __I_NEW
);
978 spin_unlock(&inode
->i_lock
);
980 EXPORT_SYMBOL(unlock_new_inode
);
983 * iget5_locked - obtain an inode from a mounted file system
984 * @sb: super block of file system
985 * @hashval: hash value (usually inode number) to get
986 * @test: callback used for comparisons between inodes
987 * @set: callback used to initialize a new struct inode
988 * @data: opaque data pointer to pass to @test and @set
990 * Search for the inode specified by @hashval and @data in the inode cache,
991 * and if present it is return it with an increased reference count. This is
992 * a generalized version of iget_locked() for file systems where the inode
993 * number is not sufficient for unique identification of an inode.
995 * If the inode is not in cache, allocate a new inode and return it locked,
996 * hashed, and with the I_NEW flag set. The file system gets to fill it in
997 * before unlocking it via unlock_new_inode().
999 * Note both @test and @set are called with the inode_hash_lock held, so can't
1002 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1003 int (*test
)(struct inode
*, void *),
1004 int (*set
)(struct inode
*, void *), void *data
)
1006 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1007 struct inode
*inode
;
1009 spin_lock(&inode_hash_lock
);
1010 inode
= find_inode(sb
, head
, test
, data
);
1011 spin_unlock(&inode_hash_lock
);
1014 wait_on_inode(inode
);
1018 inode
= alloc_inode(sb
);
1022 spin_lock(&inode_hash_lock
);
1023 /* We released the lock, so.. */
1024 old
= find_inode(sb
, head
, test
, data
);
1026 if (set(inode
, data
))
1029 spin_lock(&inode
->i_lock
);
1030 inode
->i_state
= I_NEW
;
1031 hlist_add_head(&inode
->i_hash
, head
);
1032 spin_unlock(&inode
->i_lock
);
1033 inode_sb_list_add(inode
);
1034 spin_unlock(&inode_hash_lock
);
1036 /* Return the locked inode with I_NEW set, the
1037 * caller is responsible for filling in the contents
1043 * Uhhuh, somebody else created the same inode under
1044 * us. Use the old inode instead of the one we just
1047 spin_unlock(&inode_hash_lock
);
1048 destroy_inode(inode
);
1050 wait_on_inode(inode
);
1055 spin_unlock(&inode_hash_lock
);
1056 destroy_inode(inode
);
1059 EXPORT_SYMBOL(iget5_locked
);
1062 * iget_locked - obtain an inode from a mounted file system
1063 * @sb: super block of file system
1064 * @ino: inode number to get
1066 * Search for the inode specified by @ino in the inode cache and if present
1067 * return it with an increased reference count. This is for file systems
1068 * where the inode number is sufficient for unique identification of an inode.
1070 * If the inode is not in cache, allocate a new inode and return it locked,
1071 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1072 * before unlocking it via unlock_new_inode().
1074 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1076 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1077 struct inode
*inode
;
1079 spin_lock(&inode_hash_lock
);
1080 inode
= find_inode_fast(sb
, head
, ino
);
1081 spin_unlock(&inode_hash_lock
);
1083 wait_on_inode(inode
);
1087 inode
= alloc_inode(sb
);
1091 spin_lock(&inode_hash_lock
);
1092 /* We released the lock, so.. */
1093 old
= find_inode_fast(sb
, head
, ino
);
1096 spin_lock(&inode
->i_lock
);
1097 inode
->i_state
= I_NEW
;
1098 hlist_add_head(&inode
->i_hash
, head
);
1099 spin_unlock(&inode
->i_lock
);
1100 inode_sb_list_add(inode
);
1101 spin_unlock(&inode_hash_lock
);
1103 /* Return the locked inode with I_NEW set, the
1104 * caller is responsible for filling in the contents
1110 * Uhhuh, somebody else created the same inode under
1111 * us. Use the old inode instead of the one we just
1114 spin_unlock(&inode_hash_lock
);
1115 destroy_inode(inode
);
1117 wait_on_inode(inode
);
1121 EXPORT_SYMBOL(iget_locked
);
1124 * search the inode cache for a matching inode number.
1125 * If we find one, then the inode number we are trying to
1126 * allocate is not unique and so we should not use it.
1128 * Returns 1 if the inode number is unique, 0 if it is not.
1130 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1132 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1133 struct hlist_node
*node
;
1134 struct inode
*inode
;
1136 spin_lock(&inode_hash_lock
);
1137 hlist_for_each_entry(inode
, node
, b
, i_hash
) {
1138 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
) {
1139 spin_unlock(&inode_hash_lock
);
1143 spin_unlock(&inode_hash_lock
);
1149 * iunique - get a unique inode number
1151 * @max_reserved: highest reserved inode number
1153 * Obtain an inode number that is unique on the system for a given
1154 * superblock. This is used by file systems that have no natural
1155 * permanent inode numbering system. An inode number is returned that
1156 * is higher than the reserved limit but unique.
1159 * With a large number of inodes live on the file system this function
1160 * currently becomes quite slow.
1162 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1165 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1166 * error if st_ino won't fit in target struct field. Use 32bit counter
1167 * here to attempt to avoid that.
1169 static DEFINE_SPINLOCK(iunique_lock
);
1170 static unsigned int counter
;
1173 spin_lock(&iunique_lock
);
1175 if (counter
<= max_reserved
)
1176 counter
= max_reserved
+ 1;
1178 } while (!test_inode_iunique(sb
, res
));
1179 spin_unlock(&iunique_lock
);
1183 EXPORT_SYMBOL(iunique
);
1185 struct inode
*igrab(struct inode
*inode
)
1187 spin_lock(&inode
->i_lock
);
1188 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1190 spin_unlock(&inode
->i_lock
);
1192 spin_unlock(&inode
->i_lock
);
1194 * Handle the case where s_op->clear_inode is not been
1195 * called yet, and somebody is calling igrab
1196 * while the inode is getting freed.
1202 EXPORT_SYMBOL(igrab
);
1205 * ilookup5_nowait - search for an inode in the inode cache
1206 * @sb: super block of file system to search
1207 * @hashval: hash value (usually inode number) to search for
1208 * @test: callback used for comparisons between inodes
1209 * @data: opaque data pointer to pass to @test
1211 * Search for the inode specified by @hashval and @data in the inode cache.
1212 * If the inode is in the cache, the inode is returned with an incremented
1215 * Note: I_NEW is not waited upon so you have to be very careful what you do
1216 * with the returned inode. You probably should be using ilookup5() instead.
1218 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1220 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1221 int (*test
)(struct inode
*, void *), void *data
)
1223 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1224 struct inode
*inode
;
1226 spin_lock(&inode_hash_lock
);
1227 inode
= find_inode(sb
, head
, test
, data
);
1228 spin_unlock(&inode_hash_lock
);
1232 EXPORT_SYMBOL(ilookup5_nowait
);
1235 * ilookup5 - search for an inode in the inode cache
1236 * @sb: super block of file system to search
1237 * @hashval: hash value (usually inode number) to search for
1238 * @test: callback used for comparisons between inodes
1239 * @data: opaque data pointer to pass to @test
1241 * Search for the inode specified by @hashval and @data in the inode cache,
1242 * and if the inode is in the cache, return the inode with an incremented
1243 * reference count. Waits on I_NEW before returning the inode.
1244 * returned with an incremented reference count.
1246 * This is a generalized version of ilookup() for file systems where the
1247 * inode number is not sufficient for unique identification of an inode.
1249 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1251 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1252 int (*test
)(struct inode
*, void *), void *data
)
1254 struct inode
*inode
= ilookup5_nowait(sb
, hashval
, test
, data
);
1257 wait_on_inode(inode
);
1260 EXPORT_SYMBOL(ilookup5
);
1263 * ilookup - search for an inode in the inode cache
1264 * @sb: super block of file system to search
1265 * @ino: inode number to search for
1267 * Search for the inode @ino in the inode cache, and if the inode is in the
1268 * cache, the inode is returned with an incremented reference count.
1270 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1272 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1273 struct inode
*inode
;
1275 spin_lock(&inode_hash_lock
);
1276 inode
= find_inode_fast(sb
, head
, ino
);
1277 spin_unlock(&inode_hash_lock
);
1280 wait_on_inode(inode
);
1283 EXPORT_SYMBOL(ilookup
);
1285 int insert_inode_locked(struct inode
*inode
)
1287 struct super_block
*sb
= inode
->i_sb
;
1288 ino_t ino
= inode
->i_ino
;
1289 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1292 struct hlist_node
*node
;
1293 struct inode
*old
= NULL
;
1294 spin_lock(&inode_hash_lock
);
1295 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1296 if (old
->i_ino
!= ino
)
1298 if (old
->i_sb
!= sb
)
1300 spin_lock(&old
->i_lock
);
1301 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1302 spin_unlock(&old
->i_lock
);
1307 if (likely(!node
)) {
1308 spin_lock(&inode
->i_lock
);
1309 inode
->i_state
|= I_NEW
;
1310 hlist_add_head(&inode
->i_hash
, head
);
1311 spin_unlock(&inode
->i_lock
);
1312 spin_unlock(&inode_hash_lock
);
1316 spin_unlock(&old
->i_lock
);
1317 spin_unlock(&inode_hash_lock
);
1319 if (unlikely(!inode_unhashed(old
))) {
1326 EXPORT_SYMBOL(insert_inode_locked
);
1328 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1329 int (*test
)(struct inode
*, void *), void *data
)
1331 struct super_block
*sb
= inode
->i_sb
;
1332 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1335 struct hlist_node
*node
;
1336 struct inode
*old
= NULL
;
1338 spin_lock(&inode_hash_lock
);
1339 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1340 if (old
->i_sb
!= sb
)
1342 if (!test(old
, data
))
1344 spin_lock(&old
->i_lock
);
1345 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1346 spin_unlock(&old
->i_lock
);
1351 if (likely(!node
)) {
1352 spin_lock(&inode
->i_lock
);
1353 inode
->i_state
|= I_NEW
;
1354 hlist_add_head(&inode
->i_hash
, head
);
1355 spin_unlock(&inode
->i_lock
);
1356 spin_unlock(&inode_hash_lock
);
1360 spin_unlock(&old
->i_lock
);
1361 spin_unlock(&inode_hash_lock
);
1363 if (unlikely(!inode_unhashed(old
))) {
1370 EXPORT_SYMBOL(insert_inode_locked4
);
1373 int generic_delete_inode(struct inode
*inode
)
1377 EXPORT_SYMBOL(generic_delete_inode
);
1380 * Normal UNIX filesystem behaviour: delete the
1381 * inode when the usage count drops to zero, and
1384 int generic_drop_inode(struct inode
*inode
)
1386 return !inode
->i_nlink
|| inode_unhashed(inode
);
1388 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1391 * Called when we're dropping the last reference
1394 * Call the FS "drop_inode()" function, defaulting to
1395 * the legacy UNIX filesystem behaviour. If it tells
1396 * us to evict inode, do so. Otherwise, retain inode
1397 * in cache if fs is alive, sync and evict if fs is
1400 static void iput_final(struct inode
*inode
)
1402 struct super_block
*sb
= inode
->i_sb
;
1403 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1406 WARN_ON(inode
->i_state
& I_NEW
);
1409 drop
= op
->drop_inode(inode
);
1411 drop
= generic_drop_inode(inode
);
1413 if (!drop
&& (sb
->s_flags
& MS_ACTIVE
)) {
1414 inode
->i_state
|= I_REFERENCED
;
1415 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1416 inode_lru_list_add(inode
);
1417 spin_unlock(&inode
->i_lock
);
1422 inode
->i_state
|= I_WILL_FREE
;
1423 spin_unlock(&inode
->i_lock
);
1424 write_inode_now(inode
, 1);
1425 spin_lock(&inode
->i_lock
);
1426 WARN_ON(inode
->i_state
& I_NEW
);
1427 inode
->i_state
&= ~I_WILL_FREE
;
1430 inode
->i_state
|= I_FREEING
;
1431 inode_lru_list_del(inode
);
1432 spin_unlock(&inode
->i_lock
);
1438 * iput - put an inode
1439 * @inode: inode to put
1441 * Puts an inode, dropping its usage count. If the inode use count hits
1442 * zero, the inode is then freed and may also be destroyed.
1444 * Consequently, iput() can sleep.
1446 void iput(struct inode
*inode
)
1449 BUG_ON(inode
->i_state
& I_CLEAR
);
1451 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
))
1455 EXPORT_SYMBOL(iput
);
1458 * bmap - find a block number in a file
1459 * @inode: inode of file
1460 * @block: block to find
1462 * Returns the block number on the device holding the inode that
1463 * is the disk block number for the block of the file requested.
1464 * That is, asked for block 4 of inode 1 the function will return the
1465 * disk block relative to the disk start that holds that block of the
1468 sector_t
bmap(struct inode
*inode
, sector_t block
)
1471 if (inode
->i_mapping
->a_ops
->bmap
)
1472 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1475 EXPORT_SYMBOL(bmap
);
1478 * With relative atime, only update atime if the previous atime is
1479 * earlier than either the ctime or mtime or if at least a day has
1480 * passed since the last atime update.
1482 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1483 struct timespec now
)
1486 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1489 * Is mtime younger than atime? If yes, update atime:
1491 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1494 * Is ctime younger than atime? If yes, update atime:
1496 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1500 * Is the previous atime value older than a day? If yes,
1503 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1506 * Good, we can skip the atime update:
1512 * touch_atime - update the access time
1513 * @mnt: mount the inode is accessed on
1514 * @dentry: dentry accessed
1516 * Update the accessed time on an inode and mark it for writeback.
1517 * This function automatically handles read only file systems and media,
1518 * as well as the "noatime" flag and inode specific "noatime" markers.
1520 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1522 struct inode
*inode
= dentry
->d_inode
;
1523 struct timespec now
;
1525 if (inode
->i_flags
& S_NOATIME
)
1527 if (IS_NOATIME(inode
))
1529 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1532 if (mnt
->mnt_flags
& MNT_NOATIME
)
1534 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1537 now
= current_fs_time(inode
->i_sb
);
1539 if (!relatime_need_update(mnt
, inode
, now
))
1542 if (timespec_equal(&inode
->i_atime
, &now
))
1545 if (mnt_want_write(mnt
))
1548 inode
->i_atime
= now
;
1549 mark_inode_dirty_sync(inode
);
1550 mnt_drop_write(mnt
);
1552 EXPORT_SYMBOL(touch_atime
);
1555 * file_update_time - update mtime and ctime time
1556 * @file: file accessed
1558 * Update the mtime and ctime members of an inode and mark the inode
1559 * for writeback. Note that this function is meant exclusively for
1560 * usage in the file write path of filesystems, and filesystems may
1561 * choose to explicitly ignore update via this function with the
1562 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1563 * timestamps are handled by the server.
1566 void file_update_time(struct file
*file
)
1568 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1569 struct timespec now
;
1570 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1572 /* First try to exhaust all avenues to not sync */
1573 if (IS_NOCMTIME(inode
))
1576 now
= current_fs_time(inode
->i_sb
);
1577 if (!timespec_equal(&inode
->i_mtime
, &now
))
1580 if (!timespec_equal(&inode
->i_ctime
, &now
))
1583 if (IS_I_VERSION(inode
))
1584 sync_it
|= S_VERSION
;
1589 /* Finally allowed to write? Takes lock. */
1590 if (mnt_want_write_file(file
))
1593 /* Only change inode inside the lock region */
1594 if (sync_it
& S_VERSION
)
1595 inode_inc_iversion(inode
);
1596 if (sync_it
& S_CTIME
)
1597 inode
->i_ctime
= now
;
1598 if (sync_it
& S_MTIME
)
1599 inode
->i_mtime
= now
;
1600 mark_inode_dirty_sync(inode
);
1601 mnt_drop_write(file
->f_path
.mnt
);
1603 EXPORT_SYMBOL(file_update_time
);
1605 int inode_needs_sync(struct inode
*inode
)
1609 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1613 EXPORT_SYMBOL(inode_needs_sync
);
1615 int inode_wait(void *word
)
1620 EXPORT_SYMBOL(inode_wait
);
1623 * If we try to find an inode in the inode hash while it is being
1624 * deleted, we have to wait until the filesystem completes its
1625 * deletion before reporting that it isn't found. This function waits
1626 * until the deletion _might_ have completed. Callers are responsible
1627 * to recheck inode state.
1629 * It doesn't matter if I_NEW is not set initially, a call to
1630 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1633 static void __wait_on_freeing_inode(struct inode
*inode
)
1635 wait_queue_head_t
*wq
;
1636 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1637 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1638 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1639 spin_unlock(&inode
->i_lock
);
1640 spin_unlock(&inode_hash_lock
);
1642 finish_wait(wq
, &wait
.wait
);
1643 spin_lock(&inode_hash_lock
);
1646 static __initdata
unsigned long ihash_entries
;
1647 static int __init
set_ihash_entries(char *str
)
1651 ihash_entries
= simple_strtoul(str
, &str
, 0);
1654 __setup("ihash_entries=", set_ihash_entries
);
1657 * Initialize the waitqueues and inode hash table.
1659 void __init
inode_init_early(void)
1663 /* If hashes are distributed across NUMA nodes, defer
1664 * hash allocation until vmalloc space is available.
1670 alloc_large_system_hash("Inode-cache",
1671 sizeof(struct hlist_head
),
1679 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1680 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1683 void __init
inode_init(void)
1687 /* inode slab cache */
1688 inode_cachep
= kmem_cache_create("inode_cache",
1689 sizeof(struct inode
),
1691 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1694 register_shrinker(&icache_shrinker
);
1696 /* Hash may have been set up in inode_init_early */
1701 alloc_large_system_hash("Inode-cache",
1702 sizeof(struct hlist_head
),
1710 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1711 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1714 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1716 inode
->i_mode
= mode
;
1717 if (S_ISCHR(mode
)) {
1718 inode
->i_fop
= &def_chr_fops
;
1719 inode
->i_rdev
= rdev
;
1720 } else if (S_ISBLK(mode
)) {
1721 inode
->i_fop
= &def_blk_fops
;
1722 inode
->i_rdev
= rdev
;
1723 } else if (S_ISFIFO(mode
))
1724 inode
->i_fop
= &def_fifo_fops
;
1725 else if (S_ISSOCK(mode
))
1726 inode
->i_fop
= &bad_sock_fops
;
1728 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1729 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1732 EXPORT_SYMBOL(init_special_inode
);
1735 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1737 * @dir: Directory inode
1738 * @mode: mode of the new inode
1740 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1743 inode
->i_uid
= current_fsuid();
1744 if (dir
&& dir
->i_mode
& S_ISGID
) {
1745 inode
->i_gid
= dir
->i_gid
;
1749 inode
->i_gid
= current_fsgid();
1750 inode
->i_mode
= mode
;
1752 EXPORT_SYMBOL(inode_init_owner
);
1755 * inode_owner_or_capable - check current task permissions to inode
1756 * @inode: inode being checked
1758 * Return true if current either has CAP_FOWNER to the inode, or
1761 bool inode_owner_or_capable(const struct inode
*inode
)
1763 struct user_namespace
*ns
= inode_userns(inode
);
1765 if (current_user_ns() == ns
&& current_fsuid() == inode
->i_uid
)
1767 if (ns_capable(ns
, CAP_FOWNER
))
1771 EXPORT_SYMBOL(inode_owner_or_capable
);