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>
27 #include <linux/ima.h>
28 #include <linux/cred.h>
32 * inode locking rules.
34 * inode->i_lock protects:
35 * inode->i_state, inode->i_hash, __iget()
36 * inode_lru_lock protects:
37 * 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
63 * This is needed for the following functions:
67 * FIXME: remove all knowledge of the buffer layer from this file
69 #include <linux/buffer_head.h>
72 * New inode.c implementation.
74 * This implementation has the basic premise of trying
75 * to be extremely low-overhead and SMP-safe, yet be
76 * simple enough to be "obviously correct".
81 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
83 /* #define INODE_PARANOIA 1 */
84 /* #define INODE_DEBUG 1 */
87 * Inode lookup is no longer as critical as it used to be:
88 * most of the lookups are going to be through the dcache.
90 #define I_HASHBITS i_hash_shift
91 #define I_HASHMASK i_hash_mask
93 static unsigned int i_hash_mask __read_mostly
;
94 static unsigned int i_hash_shift __read_mostly
;
95 static struct hlist_head
*inode_hashtable __read_mostly
;
96 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_hash_lock
);
99 * Each inode can be on two separate lists. One is
100 * the hash list of the inode, used for lookups. The
101 * other linked list is the "type" list:
102 * "in_use" - valid inode, i_count > 0, i_nlink > 0
103 * "dirty" - as "in_use" but also dirty
104 * "unused" - valid inode, i_count = 0
106 * A "dirty" list is maintained for each super block,
107 * allowing for low-overhead inode sync() operations.
110 static LIST_HEAD(inode_lru
);
111 static DEFINE_SPINLOCK(inode_lru_lock
);
113 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_sb_list_lock
);
114 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_wb_list_lock
);
117 * iprune_sem provides exclusion between the icache shrinking and the
120 * We don't actually need it to protect anything in the umount path,
121 * but only need to cycle through it to make sure any inode that
122 * prune_icache took off the LRU list has been fully torn down by the
123 * time we are past evict_inodes.
125 static DECLARE_RWSEM(iprune_sem
);
128 * Statistics gathering..
130 struct inodes_stat_t inodes_stat
;
132 static DEFINE_PER_CPU(unsigned int, nr_inodes
);
134 static struct kmem_cache
*inode_cachep __read_mostly
;
136 static int get_nr_inodes(void)
140 for_each_possible_cpu(i
)
141 sum
+= per_cpu(nr_inodes
, i
);
142 return sum
< 0 ? 0 : sum
;
145 static inline int get_nr_inodes_unused(void)
147 return inodes_stat
.nr_unused
;
150 int get_nr_dirty_inodes(void)
152 /* not actually dirty inodes, but a wild approximation */
153 int nr_dirty
= get_nr_inodes() - get_nr_inodes_unused();
154 return nr_dirty
> 0 ? nr_dirty
: 0;
158 * Handle nr_inode sysctl
161 int proc_nr_inodes(ctl_table
*table
, int write
,
162 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
164 inodes_stat
.nr_inodes
= get_nr_inodes();
165 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
170 * inode_init_always - perform inode structure intialisation
171 * @sb: superblock inode belongs to
172 * @inode: inode to initialise
174 * These are initializations that need to be done on every inode
175 * allocation as the fields are not initialised by slab allocation.
177 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
179 static const struct address_space_operations empty_aops
;
180 static const struct inode_operations empty_iops
;
181 static const struct file_operations empty_fops
;
182 struct address_space
*const mapping
= &inode
->i_data
;
185 inode
->i_blkbits
= sb
->s_blocksize_bits
;
187 atomic_set(&inode
->i_count
, 1);
188 inode
->i_op
= &empty_iops
;
189 inode
->i_fop
= &empty_fops
;
193 atomic_set(&inode
->i_writecount
, 0);
197 inode
->i_generation
= 0;
199 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
201 inode
->i_pipe
= NULL
;
202 inode
->i_bdev
= NULL
;
203 inode
->i_cdev
= NULL
;
205 inode
->dirtied_when
= 0;
207 if (security_inode_alloc(inode
))
209 spin_lock_init(&inode
->i_lock
);
210 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
212 mutex_init(&inode
->i_mutex
);
213 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
215 init_rwsem(&inode
->i_alloc_sem
);
216 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
218 mapping
->a_ops
= &empty_aops
;
219 mapping
->host
= inode
;
221 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
222 mapping
->assoc_mapping
= NULL
;
223 mapping
->backing_dev_info
= &default_backing_dev_info
;
224 mapping
->writeback_index
= 0;
227 * If the block_device provides a backing_dev_info for client
228 * inodes then use that. Otherwise the inode share the bdev's
232 struct backing_dev_info
*bdi
;
234 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
235 mapping
->backing_dev_info
= bdi
;
237 inode
->i_private
= NULL
;
238 inode
->i_mapping
= mapping
;
239 #ifdef CONFIG_FS_POSIX_ACL
240 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
243 #ifdef CONFIG_FSNOTIFY
244 inode
->i_fsnotify_mask
= 0;
247 this_cpu_inc(nr_inodes
);
253 EXPORT_SYMBOL(inode_init_always
);
255 static struct inode
*alloc_inode(struct super_block
*sb
)
259 if (sb
->s_op
->alloc_inode
)
260 inode
= sb
->s_op
->alloc_inode(sb
);
262 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
267 if (unlikely(inode_init_always(sb
, inode
))) {
268 if (inode
->i_sb
->s_op
->destroy_inode
)
269 inode
->i_sb
->s_op
->destroy_inode(inode
);
271 kmem_cache_free(inode_cachep
, inode
);
278 void free_inode_nonrcu(struct inode
*inode
)
280 kmem_cache_free(inode_cachep
, inode
);
282 EXPORT_SYMBOL(free_inode_nonrcu
);
284 void __destroy_inode(struct inode
*inode
)
286 BUG_ON(inode_has_buffers(inode
));
287 security_inode_free(inode
);
288 fsnotify_inode_delete(inode
);
289 #ifdef CONFIG_FS_POSIX_ACL
290 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
291 posix_acl_release(inode
->i_acl
);
292 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
293 posix_acl_release(inode
->i_default_acl
);
295 this_cpu_dec(nr_inodes
);
297 EXPORT_SYMBOL(__destroy_inode
);
299 static void i_callback(struct rcu_head
*head
)
301 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
302 INIT_LIST_HEAD(&inode
->i_dentry
);
303 kmem_cache_free(inode_cachep
, inode
);
306 static void destroy_inode(struct inode
*inode
)
308 BUG_ON(!list_empty(&inode
->i_lru
));
309 __destroy_inode(inode
);
310 if (inode
->i_sb
->s_op
->destroy_inode
)
311 inode
->i_sb
->s_op
->destroy_inode(inode
);
313 call_rcu(&inode
->i_rcu
, i_callback
);
316 void address_space_init_once(struct address_space
*mapping
)
318 memset(mapping
, 0, sizeof(*mapping
));
319 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
320 spin_lock_init(&mapping
->tree_lock
);
321 spin_lock_init(&mapping
->i_mmap_lock
);
322 INIT_LIST_HEAD(&mapping
->private_list
);
323 spin_lock_init(&mapping
->private_lock
);
324 INIT_RAW_PRIO_TREE_ROOT(&mapping
->i_mmap
);
325 INIT_LIST_HEAD(&mapping
->i_mmap_nonlinear
);
326 mutex_init(&mapping
->unmap_mutex
);
328 EXPORT_SYMBOL(address_space_init_once
);
331 * These are initializations that only need to be done
332 * once, because the fields are idempotent across use
333 * of the inode, so let the slab aware of that.
335 void inode_init_once(struct inode
*inode
)
337 memset(inode
, 0, sizeof(*inode
));
338 INIT_HLIST_NODE(&inode
->i_hash
);
339 INIT_LIST_HEAD(&inode
->i_dentry
);
340 INIT_LIST_HEAD(&inode
->i_devices
);
341 INIT_LIST_HEAD(&inode
->i_wb_list
);
342 INIT_LIST_HEAD(&inode
->i_lru
);
343 address_space_init_once(&inode
->i_data
);
344 i_size_ordered_init(inode
);
345 #ifdef CONFIG_FSNOTIFY
346 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
349 EXPORT_SYMBOL(inode_init_once
);
351 static void init_once(void *foo
)
353 struct inode
*inode
= (struct inode
*) foo
;
355 inode_init_once(inode
);
359 * inode->i_lock must be held
361 void __iget(struct inode
*inode
)
363 atomic_inc(&inode
->i_count
);
367 * get additional reference to inode; caller must already hold one.
369 void ihold(struct inode
*inode
)
371 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
373 EXPORT_SYMBOL(ihold
);
375 static void inode_lru_list_add(struct inode
*inode
)
377 spin_lock(&inode_lru_lock
);
378 if (list_empty(&inode
->i_lru
)) {
379 list_add(&inode
->i_lru
, &inode_lru
);
380 inodes_stat
.nr_unused
++;
382 spin_unlock(&inode_lru_lock
);
385 static void inode_lru_list_del(struct inode
*inode
)
387 spin_lock(&inode_lru_lock
);
388 if (!list_empty(&inode
->i_lru
)) {
389 list_del_init(&inode
->i_lru
);
390 inodes_stat
.nr_unused
--;
392 spin_unlock(&inode_lru_lock
);
396 * inode_sb_list_add - add inode to the superblock list of inodes
397 * @inode: inode to add
399 void inode_sb_list_add(struct inode
*inode
)
401 spin_lock(&inode_sb_list_lock
);
402 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
403 spin_unlock(&inode_sb_list_lock
);
405 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
407 static inline void inode_sb_list_del(struct inode
*inode
)
409 spin_lock(&inode_sb_list_lock
);
410 list_del_init(&inode
->i_sb_list
);
411 spin_unlock(&inode_sb_list_lock
);
414 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
418 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
420 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
421 return tmp
& I_HASHMASK
;
425 * __insert_inode_hash - hash an inode
426 * @inode: unhashed inode
427 * @hashval: unsigned long value used to locate this object in the
430 * Add an inode to the inode hash for this superblock.
432 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
434 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
436 spin_lock(&inode_hash_lock
);
437 spin_lock(&inode
->i_lock
);
438 hlist_add_head(&inode
->i_hash
, b
);
439 spin_unlock(&inode
->i_lock
);
440 spin_unlock(&inode_hash_lock
);
442 EXPORT_SYMBOL(__insert_inode_hash
);
445 * remove_inode_hash - remove an inode from the hash
446 * @inode: inode to unhash
448 * Remove an inode from the superblock.
450 void remove_inode_hash(struct inode
*inode
)
452 spin_lock(&inode_hash_lock
);
453 spin_lock(&inode
->i_lock
);
454 hlist_del_init(&inode
->i_hash
);
455 spin_unlock(&inode
->i_lock
);
456 spin_unlock(&inode_hash_lock
);
458 EXPORT_SYMBOL(remove_inode_hash
);
460 void end_writeback(struct inode
*inode
)
463 BUG_ON(inode
->i_data
.nrpages
);
464 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
465 BUG_ON(!(inode
->i_state
& I_FREEING
));
466 BUG_ON(inode
->i_state
& I_CLEAR
);
467 inode_sync_wait(inode
);
468 /* don't need i_lock here, no concurrent mods to i_state */
469 inode
->i_state
= I_FREEING
| I_CLEAR
;
471 EXPORT_SYMBOL(end_writeback
);
474 * Free the inode passed in, removing it from the lists it is still connected
475 * to. We remove any pages still attached to the inode and wait for any IO that
476 * is still in progress before finally destroying the inode.
478 * An inode must already be marked I_FREEING so that we avoid the inode being
479 * moved back onto lists if we race with other code that manipulates the lists
480 * (e.g. writeback_single_inode). The caller is responsible for setting this.
482 * An inode must already be removed from the LRU list before being evicted from
483 * the cache. This should occur atomically with setting the I_FREEING state
484 * flag, so no inodes here should ever be on the LRU when being evicted.
486 static void evict(struct inode
*inode
)
488 const struct super_operations
*op
= inode
->i_sb
->s_op
;
490 BUG_ON(!(inode
->i_state
& I_FREEING
));
491 BUG_ON(!list_empty(&inode
->i_lru
));
493 inode_wb_list_del(inode
);
494 inode_sb_list_del(inode
);
496 if (op
->evict_inode
) {
497 op
->evict_inode(inode
);
499 if (inode
->i_data
.nrpages
)
500 truncate_inode_pages(&inode
->i_data
, 0);
501 end_writeback(inode
);
503 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
505 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
508 remove_inode_hash(inode
);
510 spin_lock(&inode
->i_lock
);
511 wake_up_bit(&inode
->i_state
, __I_NEW
);
512 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
513 spin_unlock(&inode
->i_lock
);
515 destroy_inode(inode
);
519 * dispose_list - dispose of the contents of a local list
520 * @head: the head of the list to free
522 * Dispose-list gets a local list with local inodes in it, so it doesn't
523 * need to worry about list corruption and SMP locks.
525 static void dispose_list(struct list_head
*head
)
527 while (!list_empty(head
)) {
530 inode
= list_first_entry(head
, struct inode
, i_lru
);
531 list_del_init(&inode
->i_lru
);
538 * evict_inodes - evict all evictable inodes for a superblock
539 * @sb: superblock to operate on
541 * Make sure that no inodes with zero refcount are retained. This is
542 * called by superblock shutdown after having MS_ACTIVE flag removed,
543 * so any inode reaching zero refcount during or after that call will
544 * be immediately evicted.
546 void evict_inodes(struct super_block
*sb
)
548 struct inode
*inode
, *next
;
551 spin_lock(&inode_sb_list_lock
);
552 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
553 if (atomic_read(&inode
->i_count
))
556 spin_lock(&inode
->i_lock
);
557 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
558 spin_unlock(&inode
->i_lock
);
562 inode
->i_state
|= I_FREEING
;
563 inode_lru_list_del(inode
);
564 spin_unlock(&inode
->i_lock
);
565 list_add(&inode
->i_lru
, &dispose
);
567 spin_unlock(&inode_sb_list_lock
);
569 dispose_list(&dispose
);
572 * Cycle through iprune_sem to make sure any inode that prune_icache
573 * moved off the list before we took the lock has been fully torn
576 down_write(&iprune_sem
);
577 up_write(&iprune_sem
);
581 * invalidate_inodes - attempt to free all inodes on a superblock
582 * @sb: superblock to operate on
583 * @kill_dirty: flag to guide handling of dirty inodes
585 * Attempts to free all inodes for a given superblock. If there were any
586 * busy inodes return a non-zero value, else zero.
587 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
590 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
593 struct inode
*inode
, *next
;
596 spin_lock(&inode_sb_list_lock
);
597 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
598 spin_lock(&inode
->i_lock
);
599 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
600 spin_unlock(&inode
->i_lock
);
603 if (inode
->i_state
& I_DIRTY
&& !kill_dirty
) {
604 spin_unlock(&inode
->i_lock
);
608 if (atomic_read(&inode
->i_count
)) {
609 spin_unlock(&inode
->i_lock
);
614 inode
->i_state
|= I_FREEING
;
615 inode_lru_list_del(inode
);
616 spin_unlock(&inode
->i_lock
);
617 list_add(&inode
->i_lru
, &dispose
);
619 spin_unlock(&inode_sb_list_lock
);
621 dispose_list(&dispose
);
626 static int can_unuse(struct inode
*inode
)
628 if (inode
->i_state
& ~I_REFERENCED
)
630 if (inode_has_buffers(inode
))
632 if (atomic_read(&inode
->i_count
))
634 if (inode
->i_data
.nrpages
)
640 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
641 * temporary list and then are freed outside inode_lru_lock by dispose_list().
643 * Any inodes which are pinned purely because of attached pagecache have their
644 * pagecache removed. If the inode has metadata buffers attached to
645 * mapping->private_list then try to remove them.
647 * If the inode has the I_REFERENCED flag set, then it means that it has been
648 * used recently - the flag is set in iput_final(). When we encounter such an
649 * inode, clear the flag and move it to the back of the LRU so it gets another
650 * pass through the LRU before it gets reclaimed. This is necessary because of
651 * the fact we are doing lazy LRU updates to minimise lock contention so the
652 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
653 * with this flag set because they are the inodes that are out of order.
655 static void prune_icache(int nr_to_scan
)
659 unsigned long reap
= 0;
661 down_read(&iprune_sem
);
662 spin_lock(&inode_lru_lock
);
663 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
666 if (list_empty(&inode_lru
))
669 inode
= list_entry(inode_lru
.prev
, struct inode
, i_lru
);
672 * we are inverting the inode_lru_lock/inode->i_lock here,
673 * so use a trylock. If we fail to get the lock, just move the
674 * inode to the back of the list so we don't spin on it.
676 if (!spin_trylock(&inode
->i_lock
)) {
677 list_move(&inode
->i_lru
, &inode_lru
);
682 * Referenced or dirty inodes are still in use. Give them
683 * another pass through the LRU as we canot reclaim them now.
685 if (atomic_read(&inode
->i_count
) ||
686 (inode
->i_state
& ~I_REFERENCED
)) {
687 list_del_init(&inode
->i_lru
);
688 spin_unlock(&inode
->i_lock
);
689 inodes_stat
.nr_unused
--;
693 /* recently referenced inodes get one more pass */
694 if (inode
->i_state
& I_REFERENCED
) {
695 inode
->i_state
&= ~I_REFERENCED
;
696 list_move(&inode
->i_lru
, &inode_lru
);
697 spin_unlock(&inode
->i_lock
);
700 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
702 spin_unlock(&inode
->i_lock
);
703 spin_unlock(&inode_lru_lock
);
704 if (remove_inode_buffers(inode
))
705 reap
+= invalidate_mapping_pages(&inode
->i_data
,
708 spin_lock(&inode_lru_lock
);
710 if (inode
!= list_entry(inode_lru
.next
,
711 struct inode
, i_lru
))
712 continue; /* wrong inode or list_empty */
713 /* avoid lock inversions with trylock */
714 if (!spin_trylock(&inode
->i_lock
))
716 if (!can_unuse(inode
)) {
717 spin_unlock(&inode
->i_lock
);
721 WARN_ON(inode
->i_state
& I_NEW
);
722 inode
->i_state
|= I_FREEING
;
723 spin_unlock(&inode
->i_lock
);
725 list_move(&inode
->i_lru
, &freeable
);
726 inodes_stat
.nr_unused
--;
728 if (current_is_kswapd())
729 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
731 __count_vm_events(PGINODESTEAL
, reap
);
732 spin_unlock(&inode_lru_lock
);
734 dispose_list(&freeable
);
735 up_read(&iprune_sem
);
739 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
740 * "unused" means that no dentries are referring to the inodes: the files are
741 * not open and the dcache references to those inodes have already been
744 * This function is passed the number of inodes to scan, and it returns the
745 * total number of remaining possibly-reclaimable inodes.
747 static int shrink_icache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
751 * Nasty deadlock avoidance. We may hold various FS locks,
752 * and we don't want to recurse into the FS that called us
753 * in clear_inode() and friends..
755 if (!(gfp_mask
& __GFP_FS
))
759 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure
;
762 static struct shrinker icache_shrinker
= {
763 .shrink
= shrink_icache_memory
,
764 .seeks
= DEFAULT_SEEKS
,
767 static void __wait_on_freeing_inode(struct inode
*inode
);
769 * Called with the inode lock held.
771 static struct inode
*find_inode(struct super_block
*sb
,
772 struct hlist_head
*head
,
773 int (*test
)(struct inode
*, void *),
776 struct hlist_node
*node
;
777 struct inode
*inode
= NULL
;
780 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
781 spin_lock(&inode
->i_lock
);
782 if (inode
->i_sb
!= sb
) {
783 spin_unlock(&inode
->i_lock
);
786 if (!test(inode
, data
)) {
787 spin_unlock(&inode
->i_lock
);
790 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
791 __wait_on_freeing_inode(inode
);
795 spin_unlock(&inode
->i_lock
);
802 * find_inode_fast is the fast path version of find_inode, see the comment at
803 * iget_locked for details.
805 static struct inode
*find_inode_fast(struct super_block
*sb
,
806 struct hlist_head
*head
, unsigned long ino
)
808 struct hlist_node
*node
;
809 struct inode
*inode
= NULL
;
812 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
813 spin_lock(&inode
->i_lock
);
814 if (inode
->i_ino
!= ino
) {
815 spin_unlock(&inode
->i_lock
);
818 if (inode
->i_sb
!= sb
) {
819 spin_unlock(&inode
->i_lock
);
822 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
823 __wait_on_freeing_inode(inode
);
827 spin_unlock(&inode
->i_lock
);
834 * Each cpu owns a range of LAST_INO_BATCH numbers.
835 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
836 * to renew the exhausted range.
838 * This does not significantly increase overflow rate because every CPU can
839 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
840 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
841 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
842 * overflow rate by 2x, which does not seem too significant.
844 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
845 * error if st_ino won't fit in target struct field. Use 32bit counter
846 * here to attempt to avoid that.
848 #define LAST_INO_BATCH 1024
849 static DEFINE_PER_CPU(unsigned int, last_ino
);
851 unsigned int get_next_ino(void)
853 unsigned int *p
= &get_cpu_var(last_ino
);
854 unsigned int res
= *p
;
857 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
858 static atomic_t shared_last_ino
;
859 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
861 res
= next
- LAST_INO_BATCH
;
866 put_cpu_var(last_ino
);
869 EXPORT_SYMBOL(get_next_ino
);
872 * new_inode - obtain an inode
875 * Allocates a new inode for given superblock. The default gfp_mask
876 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
877 * If HIGHMEM pages are unsuitable or it is known that pages allocated
878 * for the page cache are not reclaimable or migratable,
879 * mapping_set_gfp_mask() must be called with suitable flags on the
880 * newly created inode's mapping
883 struct inode
*new_inode(struct super_block
*sb
)
887 spin_lock_prefetch(&inode_sb_list_lock
);
889 inode
= alloc_inode(sb
);
891 spin_lock(&inode
->i_lock
);
893 spin_unlock(&inode
->i_lock
);
894 inode_sb_list_add(inode
);
898 EXPORT_SYMBOL(new_inode
);
901 * unlock_new_inode - clear the I_NEW state and wake up any waiters
902 * @inode: new inode to unlock
904 * Called when the inode is fully initialised to clear the new state of the
905 * inode and wake up anyone waiting for the inode to finish initialisation.
907 void unlock_new_inode(struct inode
*inode
)
909 #ifdef CONFIG_DEBUG_LOCK_ALLOC
910 if (S_ISDIR(inode
->i_mode
)) {
911 struct file_system_type
*type
= inode
->i_sb
->s_type
;
913 /* Set new key only if filesystem hasn't already changed it */
914 if (!lockdep_match_class(&inode
->i_mutex
,
915 &type
->i_mutex_key
)) {
917 * ensure nobody is actually holding i_mutex
919 mutex_destroy(&inode
->i_mutex
);
920 mutex_init(&inode
->i_mutex
);
921 lockdep_set_class(&inode
->i_mutex
,
922 &type
->i_mutex_dir_key
);
926 spin_lock(&inode
->i_lock
);
927 WARN_ON(!(inode
->i_state
& I_NEW
));
928 inode
->i_state
&= ~I_NEW
;
929 wake_up_bit(&inode
->i_state
, __I_NEW
);
930 spin_unlock(&inode
->i_lock
);
932 EXPORT_SYMBOL(unlock_new_inode
);
935 * This is called without the inode hash lock held.. Be careful.
937 * We no longer cache the sb_flags in i_flags - see fs.h
938 * -- rmk@arm.uk.linux.org
940 static struct inode
*get_new_inode(struct super_block
*sb
,
941 struct hlist_head
*head
,
942 int (*test
)(struct inode
*, void *),
943 int (*set
)(struct inode
*, void *),
948 inode
= alloc_inode(sb
);
952 spin_lock(&inode_hash_lock
);
953 /* We released the lock, so.. */
954 old
= find_inode(sb
, head
, test
, data
);
956 if (set(inode
, data
))
959 spin_lock(&inode
->i_lock
);
960 inode
->i_state
= I_NEW
;
961 hlist_add_head(&inode
->i_hash
, head
);
962 spin_unlock(&inode
->i_lock
);
963 inode_sb_list_add(inode
);
964 spin_unlock(&inode_hash_lock
);
966 /* Return the locked inode with I_NEW set, the
967 * caller is responsible for filling in the contents
973 * Uhhuh, somebody else created the same inode under
974 * us. Use the old inode instead of the one we just
977 spin_unlock(&inode_hash_lock
);
978 destroy_inode(inode
);
980 wait_on_inode(inode
);
985 spin_unlock(&inode_hash_lock
);
986 destroy_inode(inode
);
991 * get_new_inode_fast is the fast path version of get_new_inode, see the
992 * comment at iget_locked for details.
994 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
995 struct hlist_head
*head
, unsigned long ino
)
999 inode
= alloc_inode(sb
);
1003 spin_lock(&inode_hash_lock
);
1004 /* We released the lock, so.. */
1005 old
= find_inode_fast(sb
, head
, ino
);
1008 spin_lock(&inode
->i_lock
);
1009 inode
->i_state
= I_NEW
;
1010 hlist_add_head(&inode
->i_hash
, head
);
1011 spin_unlock(&inode
->i_lock
);
1012 inode_sb_list_add(inode
);
1013 spin_unlock(&inode_hash_lock
);
1015 /* Return the locked inode with I_NEW set, the
1016 * caller is responsible for filling in the contents
1022 * Uhhuh, somebody else created the same inode under
1023 * us. Use the old inode instead of the one we just
1026 spin_unlock(&inode_hash_lock
);
1027 destroy_inode(inode
);
1029 wait_on_inode(inode
);
1035 * search the inode cache for a matching inode number.
1036 * If we find one, then the inode number we are trying to
1037 * allocate is not unique and so we should not use it.
1039 * Returns 1 if the inode number is unique, 0 if it is not.
1041 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1043 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1044 struct hlist_node
*node
;
1045 struct inode
*inode
;
1047 spin_lock(&inode_hash_lock
);
1048 hlist_for_each_entry(inode
, node
, b
, i_hash
) {
1049 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
) {
1050 spin_unlock(&inode_hash_lock
);
1054 spin_unlock(&inode_hash_lock
);
1060 * iunique - get a unique inode number
1062 * @max_reserved: highest reserved inode number
1064 * Obtain an inode number that is unique on the system for a given
1065 * superblock. This is used by file systems that have no natural
1066 * permanent inode numbering system. An inode number is returned that
1067 * is higher than the reserved limit but unique.
1070 * With a large number of inodes live on the file system this function
1071 * currently becomes quite slow.
1073 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1076 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1077 * error if st_ino won't fit in target struct field. Use 32bit counter
1078 * here to attempt to avoid that.
1080 static DEFINE_SPINLOCK(iunique_lock
);
1081 static unsigned int counter
;
1084 spin_lock(&iunique_lock
);
1086 if (counter
<= max_reserved
)
1087 counter
= max_reserved
+ 1;
1089 } while (!test_inode_iunique(sb
, res
));
1090 spin_unlock(&iunique_lock
);
1094 EXPORT_SYMBOL(iunique
);
1096 struct inode
*igrab(struct inode
*inode
)
1098 spin_lock(&inode
->i_lock
);
1099 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1101 spin_unlock(&inode
->i_lock
);
1103 spin_unlock(&inode
->i_lock
);
1105 * Handle the case where s_op->clear_inode is not been
1106 * called yet, and somebody is calling igrab
1107 * while the inode is getting freed.
1113 EXPORT_SYMBOL(igrab
);
1116 * ifind - internal function, you want ilookup5() or iget5().
1117 * @sb: super block of file system to search
1118 * @head: the head of the list to search
1119 * @test: callback used for comparisons between inodes
1120 * @data: opaque data pointer to pass to @test
1121 * @wait: if true wait for the inode to be unlocked, if false do not
1123 * ifind() searches for the inode specified by @data in the inode
1124 * cache. This is a generalized version of ifind_fast() for file systems where
1125 * the inode number is not sufficient for unique identification of an inode.
1127 * If the inode is in the cache, the inode is returned with an incremented
1130 * Otherwise NULL is returned.
1132 * Note, @test is called with the inode_hash_lock held, so can't sleep.
1134 static struct inode
*ifind(struct super_block
*sb
,
1135 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
1136 void *data
, const int wait
)
1138 struct inode
*inode
;
1140 spin_lock(&inode_hash_lock
);
1141 inode
= find_inode(sb
, head
, test
, data
);
1143 spin_unlock(&inode_hash_lock
);
1145 wait_on_inode(inode
);
1148 spin_unlock(&inode_hash_lock
);
1153 * ifind_fast - internal function, you want ilookup() or iget().
1154 * @sb: super block of file system to search
1155 * @head: head of the list to search
1156 * @ino: inode number to search for
1158 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1159 * file systems where the inode number is sufficient for unique identification
1162 * If the inode is in the cache, the inode is returned with an incremented
1165 * Otherwise NULL is returned.
1167 static struct inode
*ifind_fast(struct super_block
*sb
,
1168 struct hlist_head
*head
, unsigned long ino
)
1170 struct inode
*inode
;
1172 spin_lock(&inode_hash_lock
);
1173 inode
= find_inode_fast(sb
, head
, ino
);
1175 spin_unlock(&inode_hash_lock
);
1176 wait_on_inode(inode
);
1179 spin_unlock(&inode_hash_lock
);
1184 * ilookup5_nowait - search for an inode in the inode cache
1185 * @sb: super block of file system to search
1186 * @hashval: hash value (usually inode number) to search for
1187 * @test: callback used for comparisons between inodes
1188 * @data: opaque data pointer to pass to @test
1190 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1191 * @data in the inode cache. This is a generalized version of ilookup() for
1192 * file systems where the inode number is not sufficient for unique
1193 * identification of an inode.
1195 * If the inode is in the cache, the inode is returned with an incremented
1196 * reference count. Note, the inode lock is not waited upon so you have to be
1197 * very careful what you do with the returned inode. You probably should be
1198 * using ilookup5() instead.
1200 * Otherwise NULL is returned.
1202 * Note, @test is called with the inode_hash_lock held, so can't sleep.
1204 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1205 int (*test
)(struct inode
*, void *), void *data
)
1207 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1209 return ifind(sb
, head
, test
, data
, 0);
1211 EXPORT_SYMBOL(ilookup5_nowait
);
1214 * ilookup5 - search for an inode in the inode cache
1215 * @sb: super block of file system to search
1216 * @hashval: hash value (usually inode number) to search for
1217 * @test: callback used for comparisons between inodes
1218 * @data: opaque data pointer to pass to @test
1220 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1221 * @data in the inode cache. This is a generalized version of ilookup() for
1222 * file systems where the inode number is not sufficient for unique
1223 * identification of an inode.
1225 * If the inode is in the cache, the inode lock is waited upon and the inode is
1226 * returned with an incremented reference count.
1228 * Otherwise NULL is returned.
1230 * Note, @test is called with the inode_hash_lock held, so can't sleep.
1232 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1233 int (*test
)(struct inode
*, void *), void *data
)
1235 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1237 return ifind(sb
, head
, test
, data
, 1);
1239 EXPORT_SYMBOL(ilookup5
);
1242 * ilookup - search for an inode in the inode cache
1243 * @sb: super block of file system to search
1244 * @ino: inode number to search for
1246 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1247 * This is for file systems where the inode number is sufficient for unique
1248 * identification of an inode.
1250 * If the inode is in the cache, the inode is returned with an incremented
1253 * Otherwise NULL is returned.
1255 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1257 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1259 return ifind_fast(sb
, head
, ino
);
1261 EXPORT_SYMBOL(ilookup
);
1264 * iget5_locked - obtain an inode from a mounted file system
1265 * @sb: super block of file system
1266 * @hashval: hash value (usually inode number) to get
1267 * @test: callback used for comparisons between inodes
1268 * @set: callback used to initialize a new struct inode
1269 * @data: opaque data pointer to pass to @test and @set
1271 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1272 * and @data in the inode cache and if present it is returned with an increased
1273 * reference count. This is a generalized version of iget_locked() for file
1274 * systems where the inode number is not sufficient for unique identification
1277 * If the inode is not in cache, get_new_inode() is called to allocate a new
1278 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1279 * file system gets to fill it in before unlocking it via unlock_new_inode().
1281 * Note both @test and @set are called with the inode_hash_lock held, so can't
1284 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1285 int (*test
)(struct inode
*, void *),
1286 int (*set
)(struct inode
*, void *), void *data
)
1288 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1289 struct inode
*inode
;
1291 inode
= ifind(sb
, head
, test
, data
, 1);
1295 * get_new_inode() will do the right thing, re-trying the search
1296 * in case it had to block at any point.
1298 return get_new_inode(sb
, head
, test
, set
, data
);
1300 EXPORT_SYMBOL(iget5_locked
);
1303 * iget_locked - obtain an inode from a mounted file system
1304 * @sb: super block of file system
1305 * @ino: inode number to get
1307 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1308 * the inode cache and if present it is returned with an increased reference
1309 * count. This is for file systems where the inode number is sufficient for
1310 * unique identification of an inode.
1312 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1313 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1314 * The file system gets to fill it in before unlocking it via
1315 * unlock_new_inode().
1317 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1319 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1320 struct inode
*inode
;
1322 inode
= ifind_fast(sb
, head
, ino
);
1326 * get_new_inode_fast() will do the right thing, re-trying the search
1327 * in case it had to block at any point.
1329 return get_new_inode_fast(sb
, head
, ino
);
1331 EXPORT_SYMBOL(iget_locked
);
1333 int insert_inode_locked(struct inode
*inode
)
1335 struct super_block
*sb
= inode
->i_sb
;
1336 ino_t ino
= inode
->i_ino
;
1337 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1340 struct hlist_node
*node
;
1341 struct inode
*old
= NULL
;
1342 spin_lock(&inode_hash_lock
);
1343 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1344 if (old
->i_ino
!= ino
)
1346 if (old
->i_sb
!= sb
)
1348 spin_lock(&old
->i_lock
);
1349 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1350 spin_unlock(&old
->i_lock
);
1355 if (likely(!node
)) {
1356 spin_lock(&inode
->i_lock
);
1357 inode
->i_state
|= I_NEW
;
1358 hlist_add_head(&inode
->i_hash
, head
);
1359 spin_unlock(&inode
->i_lock
);
1360 spin_unlock(&inode_hash_lock
);
1364 spin_unlock(&old
->i_lock
);
1365 spin_unlock(&inode_hash_lock
);
1367 if (unlikely(!inode_unhashed(old
))) {
1374 EXPORT_SYMBOL(insert_inode_locked
);
1376 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1377 int (*test
)(struct inode
*, void *), void *data
)
1379 struct super_block
*sb
= inode
->i_sb
;
1380 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1383 struct hlist_node
*node
;
1384 struct inode
*old
= NULL
;
1386 spin_lock(&inode_hash_lock
);
1387 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1388 if (old
->i_sb
!= sb
)
1390 if (!test(old
, data
))
1392 spin_lock(&old
->i_lock
);
1393 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1394 spin_unlock(&old
->i_lock
);
1399 if (likely(!node
)) {
1400 spin_lock(&inode
->i_lock
);
1401 inode
->i_state
|= I_NEW
;
1402 hlist_add_head(&inode
->i_hash
, head
);
1403 spin_unlock(&inode
->i_lock
);
1404 spin_unlock(&inode_hash_lock
);
1408 spin_unlock(&old
->i_lock
);
1409 spin_unlock(&inode_hash_lock
);
1411 if (unlikely(!inode_unhashed(old
))) {
1418 EXPORT_SYMBOL(insert_inode_locked4
);
1421 int generic_delete_inode(struct inode
*inode
)
1425 EXPORT_SYMBOL(generic_delete_inode
);
1428 * Normal UNIX filesystem behaviour: delete the
1429 * inode when the usage count drops to zero, and
1432 int generic_drop_inode(struct inode
*inode
)
1434 return !inode
->i_nlink
|| inode_unhashed(inode
);
1436 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1439 * Called when we're dropping the last reference
1442 * Call the FS "drop_inode()" function, defaulting to
1443 * the legacy UNIX filesystem behaviour. If it tells
1444 * us to evict inode, do so. Otherwise, retain inode
1445 * in cache if fs is alive, sync and evict if fs is
1448 static void iput_final(struct inode
*inode
)
1450 struct super_block
*sb
= inode
->i_sb
;
1451 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1454 WARN_ON(inode
->i_state
& I_NEW
);
1456 if (op
&& op
->drop_inode
)
1457 drop
= op
->drop_inode(inode
);
1459 drop
= generic_drop_inode(inode
);
1461 if (!drop
&& (sb
->s_flags
& MS_ACTIVE
)) {
1462 inode
->i_state
|= I_REFERENCED
;
1463 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1464 inode_lru_list_add(inode
);
1465 spin_unlock(&inode
->i_lock
);
1470 inode
->i_state
|= I_WILL_FREE
;
1471 spin_unlock(&inode
->i_lock
);
1472 write_inode_now(inode
, 1);
1473 spin_lock(&inode
->i_lock
);
1474 WARN_ON(inode
->i_state
& I_NEW
);
1475 inode
->i_state
&= ~I_WILL_FREE
;
1478 inode
->i_state
|= I_FREEING
;
1479 inode_lru_list_del(inode
);
1480 spin_unlock(&inode
->i_lock
);
1486 * iput - put an inode
1487 * @inode: inode to put
1489 * Puts an inode, dropping its usage count. If the inode use count hits
1490 * zero, the inode is then freed and may also be destroyed.
1492 * Consequently, iput() can sleep.
1494 void iput(struct inode
*inode
)
1497 BUG_ON(inode
->i_state
& I_CLEAR
);
1499 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
))
1503 EXPORT_SYMBOL(iput
);
1506 * bmap - find a block number in a file
1507 * @inode: inode of file
1508 * @block: block to find
1510 * Returns the block number on the device holding the inode that
1511 * is the disk block number for the block of the file requested.
1512 * That is, asked for block 4 of inode 1 the function will return the
1513 * disk block relative to the disk start that holds that block of the
1516 sector_t
bmap(struct inode
*inode
, sector_t block
)
1519 if (inode
->i_mapping
->a_ops
->bmap
)
1520 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1523 EXPORT_SYMBOL(bmap
);
1526 * With relative atime, only update atime if the previous atime is
1527 * earlier than either the ctime or mtime or if at least a day has
1528 * passed since the last atime update.
1530 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1531 struct timespec now
)
1534 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1537 * Is mtime younger than atime? If yes, update atime:
1539 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1542 * Is ctime younger than atime? If yes, update atime:
1544 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1548 * Is the previous atime value older than a day? If yes,
1551 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1554 * Good, we can skip the atime update:
1560 * touch_atime - update the access time
1561 * @mnt: mount the inode is accessed on
1562 * @dentry: dentry accessed
1564 * Update the accessed time on an inode and mark it for writeback.
1565 * This function automatically handles read only file systems and media,
1566 * as well as the "noatime" flag and inode specific "noatime" markers.
1568 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1570 struct inode
*inode
= dentry
->d_inode
;
1571 struct timespec now
;
1573 if (inode
->i_flags
& S_NOATIME
)
1575 if (IS_NOATIME(inode
))
1577 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1580 if (mnt
->mnt_flags
& MNT_NOATIME
)
1582 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1585 now
= current_fs_time(inode
->i_sb
);
1587 if (!relatime_need_update(mnt
, inode
, now
))
1590 if (timespec_equal(&inode
->i_atime
, &now
))
1593 if (mnt_want_write(mnt
))
1596 inode
->i_atime
= now
;
1597 mark_inode_dirty_sync(inode
);
1598 mnt_drop_write(mnt
);
1600 EXPORT_SYMBOL(touch_atime
);
1603 * file_update_time - update mtime and ctime time
1604 * @file: file accessed
1606 * Update the mtime and ctime members of an inode and mark the inode
1607 * for writeback. Note that this function is meant exclusively for
1608 * usage in the file write path of filesystems, and filesystems may
1609 * choose to explicitly ignore update via this function with the
1610 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1611 * timestamps are handled by the server.
1614 void file_update_time(struct file
*file
)
1616 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1617 struct timespec now
;
1618 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1620 /* First try to exhaust all avenues to not sync */
1621 if (IS_NOCMTIME(inode
))
1624 now
= current_fs_time(inode
->i_sb
);
1625 if (!timespec_equal(&inode
->i_mtime
, &now
))
1628 if (!timespec_equal(&inode
->i_ctime
, &now
))
1631 if (IS_I_VERSION(inode
))
1632 sync_it
|= S_VERSION
;
1637 /* Finally allowed to write? Takes lock. */
1638 if (mnt_want_write_file(file
))
1641 /* Only change inode inside the lock region */
1642 if (sync_it
& S_VERSION
)
1643 inode_inc_iversion(inode
);
1644 if (sync_it
& S_CTIME
)
1645 inode
->i_ctime
= now
;
1646 if (sync_it
& S_MTIME
)
1647 inode
->i_mtime
= now
;
1648 mark_inode_dirty_sync(inode
);
1649 mnt_drop_write(file
->f_path
.mnt
);
1651 EXPORT_SYMBOL(file_update_time
);
1653 int inode_needs_sync(struct inode
*inode
)
1657 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1661 EXPORT_SYMBOL(inode_needs_sync
);
1663 int inode_wait(void *word
)
1668 EXPORT_SYMBOL(inode_wait
);
1671 * If we try to find an inode in the inode hash while it is being
1672 * deleted, we have to wait until the filesystem completes its
1673 * deletion before reporting that it isn't found. This function waits
1674 * until the deletion _might_ have completed. Callers are responsible
1675 * to recheck inode state.
1677 * It doesn't matter if I_NEW is not set initially, a call to
1678 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1681 static void __wait_on_freeing_inode(struct inode
*inode
)
1683 wait_queue_head_t
*wq
;
1684 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1685 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1686 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1687 spin_unlock(&inode
->i_lock
);
1688 spin_unlock(&inode_hash_lock
);
1690 finish_wait(wq
, &wait
.wait
);
1691 spin_lock(&inode_hash_lock
);
1694 static __initdata
unsigned long ihash_entries
;
1695 static int __init
set_ihash_entries(char *str
)
1699 ihash_entries
= simple_strtoul(str
, &str
, 0);
1702 __setup("ihash_entries=", set_ihash_entries
);
1705 * Initialize the waitqueues and inode hash table.
1707 void __init
inode_init_early(void)
1711 /* If hashes are distributed across NUMA nodes, defer
1712 * hash allocation until vmalloc space is available.
1718 alloc_large_system_hash("Inode-cache",
1719 sizeof(struct hlist_head
),
1727 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1728 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1731 void __init
inode_init(void)
1735 /* inode slab cache */
1736 inode_cachep
= kmem_cache_create("inode_cache",
1737 sizeof(struct inode
),
1739 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1742 register_shrinker(&icache_shrinker
);
1744 /* Hash may have been set up in inode_init_early */
1749 alloc_large_system_hash("Inode-cache",
1750 sizeof(struct hlist_head
),
1758 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1759 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1762 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1764 inode
->i_mode
= mode
;
1765 if (S_ISCHR(mode
)) {
1766 inode
->i_fop
= &def_chr_fops
;
1767 inode
->i_rdev
= rdev
;
1768 } else if (S_ISBLK(mode
)) {
1769 inode
->i_fop
= &def_blk_fops
;
1770 inode
->i_rdev
= rdev
;
1771 } else if (S_ISFIFO(mode
))
1772 inode
->i_fop
= &def_fifo_fops
;
1773 else if (S_ISSOCK(mode
))
1774 inode
->i_fop
= &bad_sock_fops
;
1776 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1777 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1780 EXPORT_SYMBOL(init_special_inode
);
1783 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1785 * @dir: Directory inode
1786 * @mode: mode of the new inode
1788 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1791 inode
->i_uid
= current_fsuid();
1792 if (dir
&& dir
->i_mode
& S_ISGID
) {
1793 inode
->i_gid
= dir
->i_gid
;
1797 inode
->i_gid
= current_fsgid();
1798 inode
->i_mode
= mode
;
1800 EXPORT_SYMBOL(inode_init_owner
);
1803 * inode_owner_or_capable - check current task permissions to inode
1804 * @inode: inode being checked
1806 * Return true if current either has CAP_FOWNER to the inode, or
1809 bool inode_owner_or_capable(const struct inode
*inode
)
1811 struct user_namespace
*ns
= inode_userns(inode
);
1813 if (current_user_ns() == ns
&& current_fsuid() == inode
->i_uid
)
1815 if (ns_capable(ns
, CAP_FOWNER
))
1819 EXPORT_SYMBOL(inode_owner_or_capable
);