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>
31 * inode locking rules.
33 * inode->i_lock protects:
34 * inode->i_state, inode->i_hash, __iget()
35 * inode_lru_lock protects:
36 * inode_lru, inode->i_lru
37 * inode_sb_list_lock protects:
38 * sb->s_inodes, inode->i_sb_list
50 * This is needed for the following functions:
54 * FIXME: remove all knowledge of the buffer layer from this file
56 #include <linux/buffer_head.h>
59 * New inode.c implementation.
61 * This implementation has the basic premise of trying
62 * to be extremely low-overhead and SMP-safe, yet be
63 * simple enough to be "obviously correct".
68 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
70 /* #define INODE_PARANOIA 1 */
71 /* #define INODE_DEBUG 1 */
74 * Inode lookup is no longer as critical as it used to be:
75 * most of the lookups are going to be through the dcache.
77 #define I_HASHBITS i_hash_shift
78 #define I_HASHMASK i_hash_mask
80 static unsigned int i_hash_mask __read_mostly
;
81 static unsigned int i_hash_shift __read_mostly
;
84 * Each inode can be on two separate lists. One is
85 * the hash list of the inode, used for lookups. The
86 * other linked list is the "type" list:
87 * "in_use" - valid inode, i_count > 0, i_nlink > 0
88 * "dirty" - as "in_use" but also dirty
89 * "unused" - valid inode, i_count = 0
91 * A "dirty" list is maintained for each super block,
92 * allowing for low-overhead inode sync() operations.
95 static LIST_HEAD(inode_lru
);
96 static DEFINE_SPINLOCK(inode_lru_lock
);
97 static struct hlist_head
*inode_hashtable __read_mostly
;
100 * A simple spinlock to protect the list manipulations.
102 * NOTE! You also have to own the lock if you change
103 * the i_state of an inode while it is in use..
105 DEFINE_SPINLOCK(inode_lock
);
107 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_sb_list_lock
);
110 * iprune_sem provides exclusion between the icache shrinking and the
113 * We don't actually need it to protect anything in the umount path,
114 * but only need to cycle through it to make sure any inode that
115 * prune_icache took off the LRU list has been fully torn down by the
116 * time we are past evict_inodes.
118 static DECLARE_RWSEM(iprune_sem
);
121 * Statistics gathering..
123 struct inodes_stat_t inodes_stat
;
125 static DEFINE_PER_CPU(unsigned int, nr_inodes
);
127 static struct kmem_cache
*inode_cachep __read_mostly
;
129 static int get_nr_inodes(void)
133 for_each_possible_cpu(i
)
134 sum
+= per_cpu(nr_inodes
, i
);
135 return sum
< 0 ? 0 : sum
;
138 static inline int get_nr_inodes_unused(void)
140 return inodes_stat
.nr_unused
;
143 int get_nr_dirty_inodes(void)
145 /* not actually dirty inodes, but a wild approximation */
146 int nr_dirty
= get_nr_inodes() - get_nr_inodes_unused();
147 return nr_dirty
> 0 ? nr_dirty
: 0;
151 * Handle nr_inode sysctl
154 int proc_nr_inodes(ctl_table
*table
, int write
,
155 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
157 inodes_stat
.nr_inodes
= get_nr_inodes();
158 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
163 * inode_init_always - perform inode structure intialisation
164 * @sb: superblock inode belongs to
165 * @inode: inode to initialise
167 * These are initializations that need to be done on every inode
168 * allocation as the fields are not initialised by slab allocation.
170 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
172 static const struct address_space_operations empty_aops
;
173 static const struct inode_operations empty_iops
;
174 static const struct file_operations empty_fops
;
175 struct address_space
*const mapping
= &inode
->i_data
;
178 inode
->i_blkbits
= sb
->s_blocksize_bits
;
180 atomic_set(&inode
->i_count
, 1);
181 inode
->i_op
= &empty_iops
;
182 inode
->i_fop
= &empty_fops
;
186 atomic_set(&inode
->i_writecount
, 0);
190 inode
->i_generation
= 0;
192 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
194 inode
->i_pipe
= NULL
;
195 inode
->i_bdev
= NULL
;
196 inode
->i_cdev
= NULL
;
198 inode
->dirtied_when
= 0;
200 if (security_inode_alloc(inode
))
202 spin_lock_init(&inode
->i_lock
);
203 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
205 mutex_init(&inode
->i_mutex
);
206 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
208 init_rwsem(&inode
->i_alloc_sem
);
209 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
211 mapping
->a_ops
= &empty_aops
;
212 mapping
->host
= inode
;
214 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
215 mapping
->assoc_mapping
= NULL
;
216 mapping
->backing_dev_info
= &default_backing_dev_info
;
217 mapping
->writeback_index
= 0;
220 * If the block_device provides a backing_dev_info for client
221 * inodes then use that. Otherwise the inode share the bdev's
225 struct backing_dev_info
*bdi
;
227 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
228 mapping
->backing_dev_info
= bdi
;
230 inode
->i_private
= NULL
;
231 inode
->i_mapping
= mapping
;
232 #ifdef CONFIG_FS_POSIX_ACL
233 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
236 #ifdef CONFIG_FSNOTIFY
237 inode
->i_fsnotify_mask
= 0;
240 this_cpu_inc(nr_inodes
);
246 EXPORT_SYMBOL(inode_init_always
);
248 static struct inode
*alloc_inode(struct super_block
*sb
)
252 if (sb
->s_op
->alloc_inode
)
253 inode
= sb
->s_op
->alloc_inode(sb
);
255 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
260 if (unlikely(inode_init_always(sb
, inode
))) {
261 if (inode
->i_sb
->s_op
->destroy_inode
)
262 inode
->i_sb
->s_op
->destroy_inode(inode
);
264 kmem_cache_free(inode_cachep
, inode
);
271 void free_inode_nonrcu(struct inode
*inode
)
273 kmem_cache_free(inode_cachep
, inode
);
275 EXPORT_SYMBOL(free_inode_nonrcu
);
277 void __destroy_inode(struct inode
*inode
)
279 BUG_ON(inode_has_buffers(inode
));
280 security_inode_free(inode
);
281 fsnotify_inode_delete(inode
);
282 #ifdef CONFIG_FS_POSIX_ACL
283 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
284 posix_acl_release(inode
->i_acl
);
285 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
286 posix_acl_release(inode
->i_default_acl
);
288 this_cpu_dec(nr_inodes
);
290 EXPORT_SYMBOL(__destroy_inode
);
292 static void i_callback(struct rcu_head
*head
)
294 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
295 INIT_LIST_HEAD(&inode
->i_dentry
);
296 kmem_cache_free(inode_cachep
, inode
);
299 static void destroy_inode(struct inode
*inode
)
301 BUG_ON(!list_empty(&inode
->i_lru
));
302 __destroy_inode(inode
);
303 if (inode
->i_sb
->s_op
->destroy_inode
)
304 inode
->i_sb
->s_op
->destroy_inode(inode
);
306 call_rcu(&inode
->i_rcu
, i_callback
);
309 void address_space_init_once(struct address_space
*mapping
)
311 memset(mapping
, 0, sizeof(*mapping
));
312 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
313 spin_lock_init(&mapping
->tree_lock
);
314 spin_lock_init(&mapping
->i_mmap_lock
);
315 INIT_LIST_HEAD(&mapping
->private_list
);
316 spin_lock_init(&mapping
->private_lock
);
317 INIT_RAW_PRIO_TREE_ROOT(&mapping
->i_mmap
);
318 INIT_LIST_HEAD(&mapping
->i_mmap_nonlinear
);
319 mutex_init(&mapping
->unmap_mutex
);
321 EXPORT_SYMBOL(address_space_init_once
);
324 * These are initializations that only need to be done
325 * once, because the fields are idempotent across use
326 * of the inode, so let the slab aware of that.
328 void inode_init_once(struct inode
*inode
)
330 memset(inode
, 0, sizeof(*inode
));
331 INIT_HLIST_NODE(&inode
->i_hash
);
332 INIT_LIST_HEAD(&inode
->i_dentry
);
333 INIT_LIST_HEAD(&inode
->i_devices
);
334 INIT_LIST_HEAD(&inode
->i_wb_list
);
335 INIT_LIST_HEAD(&inode
->i_lru
);
336 address_space_init_once(&inode
->i_data
);
337 i_size_ordered_init(inode
);
338 #ifdef CONFIG_FSNOTIFY
339 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
342 EXPORT_SYMBOL(inode_init_once
);
344 static void init_once(void *foo
)
346 struct inode
*inode
= (struct inode
*) foo
;
348 inode_init_once(inode
);
352 * inode->i_lock must be held
354 void __iget(struct inode
*inode
)
356 atomic_inc(&inode
->i_count
);
360 * get additional reference to inode; caller must already hold one.
362 void ihold(struct inode
*inode
)
364 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
366 EXPORT_SYMBOL(ihold
);
368 static void inode_lru_list_add(struct inode
*inode
)
370 spin_lock(&inode_lru_lock
);
371 if (list_empty(&inode
->i_lru
)) {
372 list_add(&inode
->i_lru
, &inode_lru
);
373 inodes_stat
.nr_unused
++;
375 spin_unlock(&inode_lru_lock
);
378 static void inode_lru_list_del(struct inode
*inode
)
380 spin_lock(&inode_lru_lock
);
381 if (!list_empty(&inode
->i_lru
)) {
382 list_del_init(&inode
->i_lru
);
383 inodes_stat
.nr_unused
--;
385 spin_unlock(&inode_lru_lock
);
389 * inode_sb_list_add - add inode to the superblock list of inodes
390 * @inode: inode to add
392 void inode_sb_list_add(struct inode
*inode
)
394 spin_lock(&inode_sb_list_lock
);
395 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
396 spin_unlock(&inode_sb_list_lock
);
398 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
400 static inline void inode_sb_list_del(struct inode
*inode
)
402 spin_lock(&inode_sb_list_lock
);
403 list_del_init(&inode
->i_sb_list
);
404 spin_unlock(&inode_sb_list_lock
);
407 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
411 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
413 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
414 return tmp
& I_HASHMASK
;
418 * __insert_inode_hash - hash an inode
419 * @inode: unhashed inode
420 * @hashval: unsigned long value used to locate this object in the
423 * Add an inode to the inode hash for this superblock.
425 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
427 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
429 spin_lock(&inode_lock
);
430 spin_lock(&inode
->i_lock
);
431 hlist_add_head(&inode
->i_hash
, b
);
432 spin_unlock(&inode
->i_lock
);
433 spin_unlock(&inode_lock
);
435 EXPORT_SYMBOL(__insert_inode_hash
);
438 * remove_inode_hash - remove an inode from the hash
439 * @inode: inode to unhash
441 * Remove an inode from the superblock.
443 void remove_inode_hash(struct inode
*inode
)
445 spin_lock(&inode_lock
);
446 spin_lock(&inode
->i_lock
);
447 hlist_del_init(&inode
->i_hash
);
448 spin_unlock(&inode
->i_lock
);
449 spin_unlock(&inode_lock
);
451 EXPORT_SYMBOL(remove_inode_hash
);
453 void end_writeback(struct inode
*inode
)
456 BUG_ON(inode
->i_data
.nrpages
);
457 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
458 BUG_ON(!(inode
->i_state
& I_FREEING
));
459 BUG_ON(inode
->i_state
& I_CLEAR
);
460 inode_sync_wait(inode
);
461 /* don't need i_lock here, no concurrent mods to i_state */
462 inode
->i_state
= I_FREEING
| I_CLEAR
;
464 EXPORT_SYMBOL(end_writeback
);
467 * Free the inode passed in, removing it from the lists it is still connected
468 * to. We remove any pages still attached to the inode and wait for any IO that
469 * is still in progress before finally destroying the inode.
471 * An inode must already be marked I_FREEING so that we avoid the inode being
472 * moved back onto lists if we race with other code that manipulates the lists
473 * (e.g. writeback_single_inode). The caller is responsible for setting this.
475 * An inode must already be removed from the LRU list before being evicted from
476 * the cache. This should occur atomically with setting the I_FREEING state
477 * flag, so no inodes here should ever be on the LRU when being evicted.
479 static void evict(struct inode
*inode
)
481 const struct super_operations
*op
= inode
->i_sb
->s_op
;
483 BUG_ON(!(inode
->i_state
& I_FREEING
));
484 BUG_ON(!list_empty(&inode
->i_lru
));
486 spin_lock(&inode_lock
);
487 list_del_init(&inode
->i_wb_list
);
488 spin_unlock(&inode_lock
);
490 inode_sb_list_del(inode
);
492 if (op
->evict_inode
) {
493 op
->evict_inode(inode
);
495 if (inode
->i_data
.nrpages
)
496 truncate_inode_pages(&inode
->i_data
, 0);
497 end_writeback(inode
);
499 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
501 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
504 remove_inode_hash(inode
);
506 spin_lock(&inode
->i_lock
);
507 wake_up_bit(&inode
->i_state
, __I_NEW
);
508 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
509 spin_unlock(&inode
->i_lock
);
511 destroy_inode(inode
);
515 * dispose_list - dispose of the contents of a local list
516 * @head: the head of the list to free
518 * Dispose-list gets a local list with local inodes in it, so it doesn't
519 * need to worry about list corruption and SMP locks.
521 static void dispose_list(struct list_head
*head
)
523 while (!list_empty(head
)) {
526 inode
= list_first_entry(head
, struct inode
, i_lru
);
527 list_del_init(&inode
->i_lru
);
534 * evict_inodes - evict all evictable inodes for a superblock
535 * @sb: superblock to operate on
537 * Make sure that no inodes with zero refcount are retained. This is
538 * called by superblock shutdown after having MS_ACTIVE flag removed,
539 * so any inode reaching zero refcount during or after that call will
540 * be immediately evicted.
542 void evict_inodes(struct super_block
*sb
)
544 struct inode
*inode
, *next
;
547 spin_lock(&inode_sb_list_lock
);
548 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
549 if (atomic_read(&inode
->i_count
))
552 spin_lock(&inode
->i_lock
);
553 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
554 spin_unlock(&inode
->i_lock
);
558 inode
->i_state
|= I_FREEING
;
559 inode_lru_list_del(inode
);
560 spin_unlock(&inode
->i_lock
);
561 list_add(&inode
->i_lru
, &dispose
);
563 spin_unlock(&inode_sb_list_lock
);
565 dispose_list(&dispose
);
568 * Cycle through iprune_sem to make sure any inode that prune_icache
569 * moved off the list before we took the lock has been fully torn
572 down_write(&iprune_sem
);
573 up_write(&iprune_sem
);
577 * invalidate_inodes - attempt to free all inodes on a superblock
578 * @sb: superblock to operate on
579 * @kill_dirty: flag to guide handling of dirty inodes
581 * Attempts to free all inodes for a given superblock. If there were any
582 * busy inodes return a non-zero value, else zero.
583 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
586 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
589 struct inode
*inode
, *next
;
592 spin_lock(&inode_sb_list_lock
);
593 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
594 spin_lock(&inode
->i_lock
);
595 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
596 spin_unlock(&inode
->i_lock
);
599 if (inode
->i_state
& I_DIRTY
&& !kill_dirty
) {
600 spin_unlock(&inode
->i_lock
);
604 if (atomic_read(&inode
->i_count
)) {
605 spin_unlock(&inode
->i_lock
);
610 inode
->i_state
|= I_FREEING
;
611 inode_lru_list_del(inode
);
612 spin_unlock(&inode
->i_lock
);
613 list_add(&inode
->i_lru
, &dispose
);
615 spin_unlock(&inode_sb_list_lock
);
617 dispose_list(&dispose
);
622 static int can_unuse(struct inode
*inode
)
624 if (inode
->i_state
& ~I_REFERENCED
)
626 if (inode_has_buffers(inode
))
628 if (atomic_read(&inode
->i_count
))
630 if (inode
->i_data
.nrpages
)
636 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
637 * temporary list and then are freed outside inode_lru_lock by dispose_list().
639 * Any inodes which are pinned purely because of attached pagecache have their
640 * pagecache removed. If the inode has metadata buffers attached to
641 * mapping->private_list then try to remove them.
643 * If the inode has the I_REFERENCED flag set, then it means that it has been
644 * used recently - the flag is set in iput_final(). When we encounter such an
645 * inode, clear the flag and move it to the back of the LRU so it gets another
646 * pass through the LRU before it gets reclaimed. This is necessary because of
647 * the fact we are doing lazy LRU updates to minimise lock contention so the
648 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
649 * with this flag set because they are the inodes that are out of order.
651 static void prune_icache(int nr_to_scan
)
655 unsigned long reap
= 0;
657 down_read(&iprune_sem
);
658 spin_lock(&inode_lru_lock
);
659 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
662 if (list_empty(&inode_lru
))
665 inode
= list_entry(inode_lru
.prev
, struct inode
, i_lru
);
668 * we are inverting the inode_lru_lock/inode->i_lock here,
669 * so use a trylock. If we fail to get the lock, just move the
670 * inode to the back of the list so we don't spin on it.
672 if (!spin_trylock(&inode
->i_lock
)) {
673 list_move(&inode
->i_lru
, &inode_lru
);
678 * Referenced or dirty inodes are still in use. Give them
679 * another pass through the LRU as we canot reclaim them now.
681 if (atomic_read(&inode
->i_count
) ||
682 (inode
->i_state
& ~I_REFERENCED
)) {
683 list_del_init(&inode
->i_lru
);
684 spin_unlock(&inode
->i_lock
);
685 inodes_stat
.nr_unused
--;
689 /* recently referenced inodes get one more pass */
690 if (inode
->i_state
& I_REFERENCED
) {
691 inode
->i_state
&= ~I_REFERENCED
;
692 list_move(&inode
->i_lru
, &inode_lru
);
693 spin_unlock(&inode
->i_lock
);
696 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
698 spin_unlock(&inode
->i_lock
);
699 spin_unlock(&inode_lru_lock
);
700 if (remove_inode_buffers(inode
))
701 reap
+= invalidate_mapping_pages(&inode
->i_data
,
704 spin_lock(&inode_lru_lock
);
706 if (inode
!= list_entry(inode_lru
.next
,
707 struct inode
, i_lru
))
708 continue; /* wrong inode or list_empty */
709 /* avoid lock inversions with trylock */
710 if (!spin_trylock(&inode
->i_lock
))
712 if (!can_unuse(inode
)) {
713 spin_unlock(&inode
->i_lock
);
717 WARN_ON(inode
->i_state
& I_NEW
);
718 inode
->i_state
|= I_FREEING
;
719 spin_unlock(&inode
->i_lock
);
721 list_move(&inode
->i_lru
, &freeable
);
722 inodes_stat
.nr_unused
--;
724 if (current_is_kswapd())
725 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
727 __count_vm_events(PGINODESTEAL
, reap
);
728 spin_unlock(&inode_lru_lock
);
730 dispose_list(&freeable
);
731 up_read(&iprune_sem
);
735 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
736 * "unused" means that no dentries are referring to the inodes: the files are
737 * not open and the dcache references to those inodes have already been
740 * This function is passed the number of inodes to scan, and it returns the
741 * total number of remaining possibly-reclaimable inodes.
743 static int shrink_icache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
747 * Nasty deadlock avoidance. We may hold various FS locks,
748 * and we don't want to recurse into the FS that called us
749 * in clear_inode() and friends..
751 if (!(gfp_mask
& __GFP_FS
))
755 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure
;
758 static struct shrinker icache_shrinker
= {
759 .shrink
= shrink_icache_memory
,
760 .seeks
= DEFAULT_SEEKS
,
763 static void __wait_on_freeing_inode(struct inode
*inode
);
765 * Called with the inode lock held.
767 static struct inode
*find_inode(struct super_block
*sb
,
768 struct hlist_head
*head
,
769 int (*test
)(struct inode
*, void *),
772 struct hlist_node
*node
;
773 struct inode
*inode
= NULL
;
776 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
777 if (inode
->i_sb
!= sb
)
779 if (!test(inode
, data
))
781 spin_lock(&inode
->i_lock
);
782 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
783 __wait_on_freeing_inode(inode
);
787 spin_unlock(&inode
->i_lock
);
794 * find_inode_fast is the fast path version of find_inode, see the comment at
795 * iget_locked for details.
797 static struct inode
*find_inode_fast(struct super_block
*sb
,
798 struct hlist_head
*head
, unsigned long ino
)
800 struct hlist_node
*node
;
801 struct inode
*inode
= NULL
;
804 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
805 if (inode
->i_ino
!= ino
)
807 if (inode
->i_sb
!= sb
)
809 spin_lock(&inode
->i_lock
);
810 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
811 __wait_on_freeing_inode(inode
);
815 spin_unlock(&inode
->i_lock
);
822 * Each cpu owns a range of LAST_INO_BATCH numbers.
823 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
824 * to renew the exhausted range.
826 * This does not significantly increase overflow rate because every CPU can
827 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
828 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
829 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
830 * overflow rate by 2x, which does not seem too significant.
832 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
833 * error if st_ino won't fit in target struct field. Use 32bit counter
834 * here to attempt to avoid that.
836 #define LAST_INO_BATCH 1024
837 static DEFINE_PER_CPU(unsigned int, last_ino
);
839 unsigned int get_next_ino(void)
841 unsigned int *p
= &get_cpu_var(last_ino
);
842 unsigned int res
= *p
;
845 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
846 static atomic_t shared_last_ino
;
847 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
849 res
= next
- LAST_INO_BATCH
;
854 put_cpu_var(last_ino
);
857 EXPORT_SYMBOL(get_next_ino
);
860 * new_inode - obtain an inode
863 * Allocates a new inode for given superblock. The default gfp_mask
864 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
865 * If HIGHMEM pages are unsuitable or it is known that pages allocated
866 * for the page cache are not reclaimable or migratable,
867 * mapping_set_gfp_mask() must be called with suitable flags on the
868 * newly created inode's mapping
871 struct inode
*new_inode(struct super_block
*sb
)
875 spin_lock_prefetch(&inode_sb_list_lock
);
877 inode
= alloc_inode(sb
);
879 spin_lock(&inode
->i_lock
);
881 spin_unlock(&inode
->i_lock
);
882 inode_sb_list_add(inode
);
886 EXPORT_SYMBOL(new_inode
);
889 * unlock_new_inode - clear the I_NEW state and wake up any waiters
890 * @inode: new inode to unlock
892 * Called when the inode is fully initialised to clear the new state of the
893 * inode and wake up anyone waiting for the inode to finish initialisation.
895 void unlock_new_inode(struct inode
*inode
)
897 #ifdef CONFIG_DEBUG_LOCK_ALLOC
898 if (S_ISDIR(inode
->i_mode
)) {
899 struct file_system_type
*type
= inode
->i_sb
->s_type
;
901 /* Set new key only if filesystem hasn't already changed it */
902 if (!lockdep_match_class(&inode
->i_mutex
,
903 &type
->i_mutex_key
)) {
905 * ensure nobody is actually holding i_mutex
907 mutex_destroy(&inode
->i_mutex
);
908 mutex_init(&inode
->i_mutex
);
909 lockdep_set_class(&inode
->i_mutex
,
910 &type
->i_mutex_dir_key
);
914 spin_lock(&inode
->i_lock
);
915 WARN_ON(!(inode
->i_state
& I_NEW
));
916 inode
->i_state
&= ~I_NEW
;
917 wake_up_bit(&inode
->i_state
, __I_NEW
);
918 spin_unlock(&inode
->i_lock
);
920 EXPORT_SYMBOL(unlock_new_inode
);
923 * This is called without the inode lock held.. Be careful.
925 * We no longer cache the sb_flags in i_flags - see fs.h
926 * -- rmk@arm.uk.linux.org
928 static struct inode
*get_new_inode(struct super_block
*sb
,
929 struct hlist_head
*head
,
930 int (*test
)(struct inode
*, void *),
931 int (*set
)(struct inode
*, void *),
936 inode
= alloc_inode(sb
);
940 spin_lock(&inode_lock
);
941 /* We released the lock, so.. */
942 old
= find_inode(sb
, head
, test
, data
);
944 if (set(inode
, data
))
947 spin_lock(&inode
->i_lock
);
948 inode
->i_state
= I_NEW
;
949 hlist_add_head(&inode
->i_hash
, head
);
950 spin_unlock(&inode
->i_lock
);
951 inode_sb_list_add(inode
);
952 spin_unlock(&inode_lock
);
954 /* Return the locked inode with I_NEW set, the
955 * caller is responsible for filling in the contents
961 * Uhhuh, somebody else created the same inode under
962 * us. Use the old inode instead of the one we just
965 spin_unlock(&inode_lock
);
966 destroy_inode(inode
);
968 wait_on_inode(inode
);
973 spin_unlock(&inode_lock
);
974 destroy_inode(inode
);
979 * get_new_inode_fast is the fast path version of get_new_inode, see the
980 * comment at iget_locked for details.
982 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
983 struct hlist_head
*head
, unsigned long ino
)
987 inode
= alloc_inode(sb
);
991 spin_lock(&inode_lock
);
992 /* We released the lock, so.. */
993 old
= find_inode_fast(sb
, head
, ino
);
996 spin_lock(&inode
->i_lock
);
997 inode
->i_state
= I_NEW
;
998 hlist_add_head(&inode
->i_hash
, head
);
999 spin_unlock(&inode
->i_lock
);
1000 inode_sb_list_add(inode
);
1001 spin_unlock(&inode_lock
);
1003 /* Return the locked inode with I_NEW set, the
1004 * caller is responsible for filling in the contents
1010 * Uhhuh, somebody else created the same inode under
1011 * us. Use the old inode instead of the one we just
1014 spin_unlock(&inode_lock
);
1015 destroy_inode(inode
);
1017 wait_on_inode(inode
);
1023 * search the inode cache for a matching inode number.
1024 * If we find one, then the inode number we are trying to
1025 * allocate is not unique and so we should not use it.
1027 * Returns 1 if the inode number is unique, 0 if it is not.
1029 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1031 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1032 struct hlist_node
*node
;
1033 struct inode
*inode
;
1035 hlist_for_each_entry(inode
, node
, b
, i_hash
) {
1036 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
)
1044 * iunique - get a unique inode number
1046 * @max_reserved: highest reserved inode number
1048 * Obtain an inode number that is unique on the system for a given
1049 * superblock. This is used by file systems that have no natural
1050 * permanent inode numbering system. An inode number is returned that
1051 * is higher than the reserved limit but unique.
1054 * With a large number of inodes live on the file system this function
1055 * currently becomes quite slow.
1057 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1060 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1061 * error if st_ino won't fit in target struct field. Use 32bit counter
1062 * here to attempt to avoid that.
1064 static DEFINE_SPINLOCK(iunique_lock
);
1065 static unsigned int counter
;
1068 spin_lock(&inode_lock
);
1069 spin_lock(&iunique_lock
);
1071 if (counter
<= max_reserved
)
1072 counter
= max_reserved
+ 1;
1074 } while (!test_inode_iunique(sb
, res
));
1075 spin_unlock(&iunique_lock
);
1076 spin_unlock(&inode_lock
);
1080 EXPORT_SYMBOL(iunique
);
1082 struct inode
*igrab(struct inode
*inode
)
1084 spin_lock(&inode
->i_lock
);
1085 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1087 spin_unlock(&inode
->i_lock
);
1089 spin_unlock(&inode
->i_lock
);
1091 * Handle the case where s_op->clear_inode is not been
1092 * called yet, and somebody is calling igrab
1093 * while the inode is getting freed.
1099 EXPORT_SYMBOL(igrab
);
1102 * ifind - internal function, you want ilookup5() or iget5().
1103 * @sb: super block of file system to search
1104 * @head: the head of the list to search
1105 * @test: callback used for comparisons between inodes
1106 * @data: opaque data pointer to pass to @test
1107 * @wait: if true wait for the inode to be unlocked, if false do not
1109 * ifind() searches for the inode specified by @data in the inode
1110 * cache. This is a generalized version of ifind_fast() for file systems where
1111 * the inode number is not sufficient for unique identification of an inode.
1113 * If the inode is in the cache, the inode is returned with an incremented
1116 * Otherwise NULL is returned.
1118 * Note, @test is called with the inode_lock held, so can't sleep.
1120 static struct inode
*ifind(struct super_block
*sb
,
1121 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
1122 void *data
, const int wait
)
1124 struct inode
*inode
;
1126 spin_lock(&inode_lock
);
1127 inode
= find_inode(sb
, head
, test
, data
);
1129 spin_unlock(&inode_lock
);
1131 wait_on_inode(inode
);
1134 spin_unlock(&inode_lock
);
1139 * ifind_fast - internal function, you want ilookup() or iget().
1140 * @sb: super block of file system to search
1141 * @head: head of the list to search
1142 * @ino: inode number to search for
1144 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1145 * file systems where the inode number is sufficient for unique identification
1148 * If the inode is in the cache, the inode is returned with an incremented
1151 * Otherwise NULL is returned.
1153 static struct inode
*ifind_fast(struct super_block
*sb
,
1154 struct hlist_head
*head
, unsigned long ino
)
1156 struct inode
*inode
;
1158 spin_lock(&inode_lock
);
1159 inode
= find_inode_fast(sb
, head
, ino
);
1161 spin_unlock(&inode_lock
);
1162 wait_on_inode(inode
);
1165 spin_unlock(&inode_lock
);
1170 * ilookup5_nowait - search for an inode in the inode cache
1171 * @sb: super block of file system to search
1172 * @hashval: hash value (usually inode number) to search for
1173 * @test: callback used for comparisons between inodes
1174 * @data: opaque data pointer to pass to @test
1176 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1177 * @data in the inode cache. This is a generalized version of ilookup() for
1178 * file systems where the inode number is not sufficient for unique
1179 * identification of an inode.
1181 * If the inode is in the cache, the inode is returned with an incremented
1182 * reference count. Note, the inode lock is not waited upon so you have to be
1183 * very careful what you do with the returned inode. You probably should be
1184 * using ilookup5() instead.
1186 * Otherwise NULL is returned.
1188 * Note, @test is called with the inode_lock held, so can't sleep.
1190 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1191 int (*test
)(struct inode
*, void *), void *data
)
1193 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1195 return ifind(sb
, head
, test
, data
, 0);
1197 EXPORT_SYMBOL(ilookup5_nowait
);
1200 * ilookup5 - search for an inode in the inode cache
1201 * @sb: super block of file system to search
1202 * @hashval: hash value (usually inode number) to search for
1203 * @test: callback used for comparisons between inodes
1204 * @data: opaque data pointer to pass to @test
1206 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1207 * @data in the inode cache. This is a generalized version of ilookup() for
1208 * file systems where the inode number is not sufficient for unique
1209 * identification of an inode.
1211 * If the inode is in the cache, the inode lock is waited upon and the inode is
1212 * returned with an incremented reference count.
1214 * Otherwise NULL is returned.
1216 * Note, @test is called with the inode_lock held, so can't sleep.
1218 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1219 int (*test
)(struct inode
*, void *), void *data
)
1221 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1223 return ifind(sb
, head
, test
, data
, 1);
1225 EXPORT_SYMBOL(ilookup5
);
1228 * ilookup - search for an inode in the inode cache
1229 * @sb: super block of file system to search
1230 * @ino: inode number to search for
1232 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1233 * This is for file systems where the inode number is sufficient for unique
1234 * identification of an inode.
1236 * If the inode is in the cache, the inode is returned with an incremented
1239 * Otherwise NULL is returned.
1241 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1243 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1245 return ifind_fast(sb
, head
, ino
);
1247 EXPORT_SYMBOL(ilookup
);
1250 * iget5_locked - obtain an inode from a mounted file system
1251 * @sb: super block of file system
1252 * @hashval: hash value (usually inode number) to get
1253 * @test: callback used for comparisons between inodes
1254 * @set: callback used to initialize a new struct inode
1255 * @data: opaque data pointer to pass to @test and @set
1257 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1258 * and @data in the inode cache and if present it is returned with an increased
1259 * reference count. This is a generalized version of iget_locked() for file
1260 * systems where the inode number is not sufficient for unique identification
1263 * If the inode is not in cache, get_new_inode() is called to allocate a new
1264 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1265 * file system gets to fill it in before unlocking it via unlock_new_inode().
1267 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1269 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1270 int (*test
)(struct inode
*, void *),
1271 int (*set
)(struct inode
*, void *), void *data
)
1273 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1274 struct inode
*inode
;
1276 inode
= ifind(sb
, head
, test
, data
, 1);
1280 * get_new_inode() will do the right thing, re-trying the search
1281 * in case it had to block at any point.
1283 return get_new_inode(sb
, head
, test
, set
, data
);
1285 EXPORT_SYMBOL(iget5_locked
);
1288 * iget_locked - obtain an inode from a mounted file system
1289 * @sb: super block of file system
1290 * @ino: inode number to get
1292 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1293 * the inode cache and if present it is returned with an increased reference
1294 * count. This is for file systems where the inode number is sufficient for
1295 * unique identification of an inode.
1297 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1298 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1299 * The file system gets to fill it in before unlocking it via
1300 * unlock_new_inode().
1302 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1304 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1305 struct inode
*inode
;
1307 inode
= ifind_fast(sb
, head
, ino
);
1311 * get_new_inode_fast() will do the right thing, re-trying the search
1312 * in case it had to block at any point.
1314 return get_new_inode_fast(sb
, head
, ino
);
1316 EXPORT_SYMBOL(iget_locked
);
1318 int insert_inode_locked(struct inode
*inode
)
1320 struct super_block
*sb
= inode
->i_sb
;
1321 ino_t ino
= inode
->i_ino
;
1322 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1325 struct hlist_node
*node
;
1326 struct inode
*old
= NULL
;
1327 spin_lock(&inode_lock
);
1328 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1329 if (old
->i_ino
!= ino
)
1331 if (old
->i_sb
!= sb
)
1333 spin_lock(&old
->i_lock
);
1334 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1335 spin_unlock(&old
->i_lock
);
1340 if (likely(!node
)) {
1341 spin_lock(&inode
->i_lock
);
1342 inode
->i_state
|= I_NEW
;
1343 hlist_add_head(&inode
->i_hash
, head
);
1344 spin_unlock(&inode
->i_lock
);
1345 spin_unlock(&inode_lock
);
1349 spin_unlock(&old
->i_lock
);
1350 spin_unlock(&inode_lock
);
1352 if (unlikely(!inode_unhashed(old
))) {
1359 EXPORT_SYMBOL(insert_inode_locked
);
1361 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1362 int (*test
)(struct inode
*, void *), void *data
)
1364 struct super_block
*sb
= inode
->i_sb
;
1365 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1368 struct hlist_node
*node
;
1369 struct inode
*old
= NULL
;
1371 spin_lock(&inode_lock
);
1372 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1373 if (old
->i_sb
!= sb
)
1375 if (!test(old
, data
))
1377 spin_lock(&old
->i_lock
);
1378 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1379 spin_unlock(&old
->i_lock
);
1384 if (likely(!node
)) {
1385 spin_lock(&inode
->i_lock
);
1386 inode
->i_state
|= I_NEW
;
1387 hlist_add_head(&inode
->i_hash
, head
);
1388 spin_unlock(&inode
->i_lock
);
1389 spin_unlock(&inode_lock
);
1393 spin_unlock(&old
->i_lock
);
1394 spin_unlock(&inode_lock
);
1396 if (unlikely(!inode_unhashed(old
))) {
1403 EXPORT_SYMBOL(insert_inode_locked4
);
1406 int generic_delete_inode(struct inode
*inode
)
1410 EXPORT_SYMBOL(generic_delete_inode
);
1413 * Normal UNIX filesystem behaviour: delete the
1414 * inode when the usage count drops to zero, and
1417 int generic_drop_inode(struct inode
*inode
)
1419 return !inode
->i_nlink
|| inode_unhashed(inode
);
1421 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1424 * Called when we're dropping the last reference
1427 * Call the FS "drop_inode()" function, defaulting to
1428 * the legacy UNIX filesystem behaviour. If it tells
1429 * us to evict inode, do so. Otherwise, retain inode
1430 * in cache if fs is alive, sync and evict if fs is
1433 static void iput_final(struct inode
*inode
)
1435 struct super_block
*sb
= inode
->i_sb
;
1436 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1439 WARN_ON(inode
->i_state
& I_NEW
);
1441 if (op
&& op
->drop_inode
)
1442 drop
= op
->drop_inode(inode
);
1444 drop
= generic_drop_inode(inode
);
1446 if (!drop
&& (sb
->s_flags
& MS_ACTIVE
)) {
1447 inode
->i_state
|= I_REFERENCED
;
1448 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1449 inode_lru_list_add(inode
);
1450 spin_unlock(&inode
->i_lock
);
1455 inode
->i_state
|= I_WILL_FREE
;
1456 spin_unlock(&inode
->i_lock
);
1457 write_inode_now(inode
, 1);
1458 spin_lock(&inode
->i_lock
);
1459 WARN_ON(inode
->i_state
& I_NEW
);
1460 inode
->i_state
&= ~I_WILL_FREE
;
1463 inode
->i_state
|= I_FREEING
;
1464 inode_lru_list_del(inode
);
1465 spin_unlock(&inode
->i_lock
);
1471 * iput - put an inode
1472 * @inode: inode to put
1474 * Puts an inode, dropping its usage count. If the inode use count hits
1475 * zero, the inode is then freed and may also be destroyed.
1477 * Consequently, iput() can sleep.
1479 void iput(struct inode
*inode
)
1482 BUG_ON(inode
->i_state
& I_CLEAR
);
1484 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
))
1488 EXPORT_SYMBOL(iput
);
1491 * bmap - find a block number in a file
1492 * @inode: inode of file
1493 * @block: block to find
1495 * Returns the block number on the device holding the inode that
1496 * is the disk block number for the block of the file requested.
1497 * That is, asked for block 4 of inode 1 the function will return the
1498 * disk block relative to the disk start that holds that block of the
1501 sector_t
bmap(struct inode
*inode
, sector_t block
)
1504 if (inode
->i_mapping
->a_ops
->bmap
)
1505 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1508 EXPORT_SYMBOL(bmap
);
1511 * With relative atime, only update atime if the previous atime is
1512 * earlier than either the ctime or mtime or if at least a day has
1513 * passed since the last atime update.
1515 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1516 struct timespec now
)
1519 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1522 * Is mtime younger than atime? If yes, update atime:
1524 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1527 * Is ctime younger than atime? If yes, update atime:
1529 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1533 * Is the previous atime value older than a day? If yes,
1536 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1539 * Good, we can skip the atime update:
1545 * touch_atime - update the access time
1546 * @mnt: mount the inode is accessed on
1547 * @dentry: dentry accessed
1549 * Update the accessed time on an inode and mark it for writeback.
1550 * This function automatically handles read only file systems and media,
1551 * as well as the "noatime" flag and inode specific "noatime" markers.
1553 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1555 struct inode
*inode
= dentry
->d_inode
;
1556 struct timespec now
;
1558 if (inode
->i_flags
& S_NOATIME
)
1560 if (IS_NOATIME(inode
))
1562 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1565 if (mnt
->mnt_flags
& MNT_NOATIME
)
1567 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1570 now
= current_fs_time(inode
->i_sb
);
1572 if (!relatime_need_update(mnt
, inode
, now
))
1575 if (timespec_equal(&inode
->i_atime
, &now
))
1578 if (mnt_want_write(mnt
))
1581 inode
->i_atime
= now
;
1582 mark_inode_dirty_sync(inode
);
1583 mnt_drop_write(mnt
);
1585 EXPORT_SYMBOL(touch_atime
);
1588 * file_update_time - update mtime and ctime time
1589 * @file: file accessed
1591 * Update the mtime and ctime members of an inode and mark the inode
1592 * for writeback. Note that this function is meant exclusively for
1593 * usage in the file write path of filesystems, and filesystems may
1594 * choose to explicitly ignore update via this function with the
1595 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1596 * timestamps are handled by the server.
1599 void file_update_time(struct file
*file
)
1601 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1602 struct timespec now
;
1603 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1605 /* First try to exhaust all avenues to not sync */
1606 if (IS_NOCMTIME(inode
))
1609 now
= current_fs_time(inode
->i_sb
);
1610 if (!timespec_equal(&inode
->i_mtime
, &now
))
1613 if (!timespec_equal(&inode
->i_ctime
, &now
))
1616 if (IS_I_VERSION(inode
))
1617 sync_it
|= S_VERSION
;
1622 /* Finally allowed to write? Takes lock. */
1623 if (mnt_want_write_file(file
))
1626 /* Only change inode inside the lock region */
1627 if (sync_it
& S_VERSION
)
1628 inode_inc_iversion(inode
);
1629 if (sync_it
& S_CTIME
)
1630 inode
->i_ctime
= now
;
1631 if (sync_it
& S_MTIME
)
1632 inode
->i_mtime
= now
;
1633 mark_inode_dirty_sync(inode
);
1634 mnt_drop_write(file
->f_path
.mnt
);
1636 EXPORT_SYMBOL(file_update_time
);
1638 int inode_needs_sync(struct inode
*inode
)
1642 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1646 EXPORT_SYMBOL(inode_needs_sync
);
1648 int inode_wait(void *word
)
1653 EXPORT_SYMBOL(inode_wait
);
1656 * If we try to find an inode in the inode hash while it is being
1657 * deleted, we have to wait until the filesystem completes its
1658 * deletion before reporting that it isn't found. This function waits
1659 * until the deletion _might_ have completed. Callers are responsible
1660 * to recheck inode state.
1662 * It doesn't matter if I_NEW is not set initially, a call to
1663 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1666 static void __wait_on_freeing_inode(struct inode
*inode
)
1668 wait_queue_head_t
*wq
;
1669 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1670 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1671 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1672 spin_unlock(&inode
->i_lock
);
1673 spin_unlock(&inode_lock
);
1675 finish_wait(wq
, &wait
.wait
);
1676 spin_lock(&inode_lock
);
1679 static __initdata
unsigned long ihash_entries
;
1680 static int __init
set_ihash_entries(char *str
)
1684 ihash_entries
= simple_strtoul(str
, &str
, 0);
1687 __setup("ihash_entries=", set_ihash_entries
);
1690 * Initialize the waitqueues and inode hash table.
1692 void __init
inode_init_early(void)
1696 /* If hashes are distributed across NUMA nodes, defer
1697 * hash allocation until vmalloc space is available.
1703 alloc_large_system_hash("Inode-cache",
1704 sizeof(struct hlist_head
),
1712 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1713 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1716 void __init
inode_init(void)
1720 /* inode slab cache */
1721 inode_cachep
= kmem_cache_create("inode_cache",
1722 sizeof(struct inode
),
1724 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1727 register_shrinker(&icache_shrinker
);
1729 /* Hash may have been set up in inode_init_early */
1734 alloc_large_system_hash("Inode-cache",
1735 sizeof(struct hlist_head
),
1743 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1744 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1747 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1749 inode
->i_mode
= mode
;
1750 if (S_ISCHR(mode
)) {
1751 inode
->i_fop
= &def_chr_fops
;
1752 inode
->i_rdev
= rdev
;
1753 } else if (S_ISBLK(mode
)) {
1754 inode
->i_fop
= &def_blk_fops
;
1755 inode
->i_rdev
= rdev
;
1756 } else if (S_ISFIFO(mode
))
1757 inode
->i_fop
= &def_fifo_fops
;
1758 else if (S_ISSOCK(mode
))
1759 inode
->i_fop
= &bad_sock_fops
;
1761 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1762 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1765 EXPORT_SYMBOL(init_special_inode
);
1768 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1770 * @dir: Directory inode
1771 * @mode: mode of the new inode
1773 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1776 inode
->i_uid
= current_fsuid();
1777 if (dir
&& dir
->i_mode
& S_ISGID
) {
1778 inode
->i_gid
= dir
->i_gid
;
1782 inode
->i_gid
= current_fsgid();
1783 inode
->i_mode
= mode
;
1785 EXPORT_SYMBOL(inode_init_owner
);
1788 * inode_owner_or_capable - check current task permissions to inode
1789 * @inode: inode being checked
1791 * Return true if current either has CAP_FOWNER to the inode, or
1794 bool inode_owner_or_capable(const struct inode
*inode
)
1796 struct user_namespace
*ns
= inode_userns(inode
);
1798 if (current_user_ns() == ns
&& current_fsuid() == inode
->i_uid
)
1800 if (ns_capable(ns
, CAP_FOWNER
))
1804 EXPORT_SYMBOL(inode_owner_or_capable
);