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()
42 * This is needed for the following functions:
46 * FIXME: remove all knowledge of the buffer layer from this file
48 #include <linux/buffer_head.h>
51 * New inode.c implementation.
53 * This implementation has the basic premise of trying
54 * to be extremely low-overhead and SMP-safe, yet be
55 * simple enough to be "obviously correct".
60 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
62 /* #define INODE_PARANOIA 1 */
63 /* #define INODE_DEBUG 1 */
66 * Inode lookup is no longer as critical as it used to be:
67 * most of the lookups are going to be through the dcache.
69 #define I_HASHBITS i_hash_shift
70 #define I_HASHMASK i_hash_mask
72 static unsigned int i_hash_mask __read_mostly
;
73 static unsigned int i_hash_shift __read_mostly
;
76 * Each inode can be on two separate lists. One is
77 * the hash list of the inode, used for lookups. The
78 * other linked list is the "type" list:
79 * "in_use" - valid inode, i_count > 0, i_nlink > 0
80 * "dirty" - as "in_use" but also dirty
81 * "unused" - valid inode, i_count = 0
83 * A "dirty" list is maintained for each super block,
84 * allowing for low-overhead inode sync() operations.
87 static LIST_HEAD(inode_lru
);
88 static struct hlist_head
*inode_hashtable __read_mostly
;
91 * A simple spinlock to protect the list manipulations.
93 * NOTE! You also have to own the lock if you change
94 * the i_state of an inode while it is in use..
96 DEFINE_SPINLOCK(inode_lock
);
99 * iprune_sem provides exclusion between the icache shrinking and the
102 * We don't actually need it to protect anything in the umount path,
103 * but only need to cycle through it to make sure any inode that
104 * prune_icache took off the LRU list has been fully torn down by the
105 * time we are past evict_inodes.
107 static DECLARE_RWSEM(iprune_sem
);
110 * Statistics gathering..
112 struct inodes_stat_t inodes_stat
;
114 static DEFINE_PER_CPU(unsigned int, nr_inodes
);
116 static struct kmem_cache
*inode_cachep __read_mostly
;
118 static int get_nr_inodes(void)
122 for_each_possible_cpu(i
)
123 sum
+= per_cpu(nr_inodes
, i
);
124 return sum
< 0 ? 0 : sum
;
127 static inline int get_nr_inodes_unused(void)
129 return inodes_stat
.nr_unused
;
132 int get_nr_dirty_inodes(void)
134 /* not actually dirty inodes, but a wild approximation */
135 int nr_dirty
= get_nr_inodes() - get_nr_inodes_unused();
136 return nr_dirty
> 0 ? nr_dirty
: 0;
140 * Handle nr_inode sysctl
143 int proc_nr_inodes(ctl_table
*table
, int write
,
144 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
146 inodes_stat
.nr_inodes
= get_nr_inodes();
147 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
152 * inode_init_always - perform inode structure intialisation
153 * @sb: superblock inode belongs to
154 * @inode: inode to initialise
156 * These are initializations that need to be done on every inode
157 * allocation as the fields are not initialised by slab allocation.
159 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
161 static const struct address_space_operations empty_aops
;
162 static const struct inode_operations empty_iops
;
163 static const struct file_operations empty_fops
;
164 struct address_space
*const mapping
= &inode
->i_data
;
167 inode
->i_blkbits
= sb
->s_blocksize_bits
;
169 atomic_set(&inode
->i_count
, 1);
170 inode
->i_op
= &empty_iops
;
171 inode
->i_fop
= &empty_fops
;
175 atomic_set(&inode
->i_writecount
, 0);
179 inode
->i_generation
= 0;
181 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
183 inode
->i_pipe
= NULL
;
184 inode
->i_bdev
= NULL
;
185 inode
->i_cdev
= NULL
;
187 inode
->dirtied_when
= 0;
189 if (security_inode_alloc(inode
))
191 spin_lock_init(&inode
->i_lock
);
192 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
194 mutex_init(&inode
->i_mutex
);
195 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
197 init_rwsem(&inode
->i_alloc_sem
);
198 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
200 mapping
->a_ops
= &empty_aops
;
201 mapping
->host
= inode
;
203 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
204 mapping
->assoc_mapping
= NULL
;
205 mapping
->backing_dev_info
= &default_backing_dev_info
;
206 mapping
->writeback_index
= 0;
209 * If the block_device provides a backing_dev_info for client
210 * inodes then use that. Otherwise the inode share the bdev's
214 struct backing_dev_info
*bdi
;
216 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
217 mapping
->backing_dev_info
= bdi
;
219 inode
->i_private
= NULL
;
220 inode
->i_mapping
= mapping
;
221 #ifdef CONFIG_FS_POSIX_ACL
222 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
225 #ifdef CONFIG_FSNOTIFY
226 inode
->i_fsnotify_mask
= 0;
229 this_cpu_inc(nr_inodes
);
235 EXPORT_SYMBOL(inode_init_always
);
237 static struct inode
*alloc_inode(struct super_block
*sb
)
241 if (sb
->s_op
->alloc_inode
)
242 inode
= sb
->s_op
->alloc_inode(sb
);
244 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
249 if (unlikely(inode_init_always(sb
, inode
))) {
250 if (inode
->i_sb
->s_op
->destroy_inode
)
251 inode
->i_sb
->s_op
->destroy_inode(inode
);
253 kmem_cache_free(inode_cachep
, inode
);
260 void free_inode_nonrcu(struct inode
*inode
)
262 kmem_cache_free(inode_cachep
, inode
);
264 EXPORT_SYMBOL(free_inode_nonrcu
);
266 void __destroy_inode(struct inode
*inode
)
268 BUG_ON(inode_has_buffers(inode
));
269 security_inode_free(inode
);
270 fsnotify_inode_delete(inode
);
271 #ifdef CONFIG_FS_POSIX_ACL
272 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
273 posix_acl_release(inode
->i_acl
);
274 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
275 posix_acl_release(inode
->i_default_acl
);
277 this_cpu_dec(nr_inodes
);
279 EXPORT_SYMBOL(__destroy_inode
);
281 static void i_callback(struct rcu_head
*head
)
283 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
284 INIT_LIST_HEAD(&inode
->i_dentry
);
285 kmem_cache_free(inode_cachep
, inode
);
288 static void destroy_inode(struct inode
*inode
)
290 BUG_ON(!list_empty(&inode
->i_lru
));
291 __destroy_inode(inode
);
292 if (inode
->i_sb
->s_op
->destroy_inode
)
293 inode
->i_sb
->s_op
->destroy_inode(inode
);
295 call_rcu(&inode
->i_rcu
, i_callback
);
298 void address_space_init_once(struct address_space
*mapping
)
300 memset(mapping
, 0, sizeof(*mapping
));
301 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
302 spin_lock_init(&mapping
->tree_lock
);
303 spin_lock_init(&mapping
->i_mmap_lock
);
304 INIT_LIST_HEAD(&mapping
->private_list
);
305 spin_lock_init(&mapping
->private_lock
);
306 INIT_RAW_PRIO_TREE_ROOT(&mapping
->i_mmap
);
307 INIT_LIST_HEAD(&mapping
->i_mmap_nonlinear
);
308 mutex_init(&mapping
->unmap_mutex
);
310 EXPORT_SYMBOL(address_space_init_once
);
313 * These are initializations that only need to be done
314 * once, because the fields are idempotent across use
315 * of the inode, so let the slab aware of that.
317 void inode_init_once(struct inode
*inode
)
319 memset(inode
, 0, sizeof(*inode
));
320 INIT_HLIST_NODE(&inode
->i_hash
);
321 INIT_LIST_HEAD(&inode
->i_dentry
);
322 INIT_LIST_HEAD(&inode
->i_devices
);
323 INIT_LIST_HEAD(&inode
->i_wb_list
);
324 INIT_LIST_HEAD(&inode
->i_lru
);
325 address_space_init_once(&inode
->i_data
);
326 i_size_ordered_init(inode
);
327 #ifdef CONFIG_FSNOTIFY
328 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
331 EXPORT_SYMBOL(inode_init_once
);
333 static void init_once(void *foo
)
335 struct inode
*inode
= (struct inode
*) foo
;
337 inode_init_once(inode
);
341 * inode->i_lock must be held
343 void __iget(struct inode
*inode
)
345 atomic_inc(&inode
->i_count
);
349 * get additional reference to inode; caller must already hold one.
351 void ihold(struct inode
*inode
)
353 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
355 EXPORT_SYMBOL(ihold
);
357 static void inode_lru_list_add(struct inode
*inode
)
359 if (list_empty(&inode
->i_lru
)) {
360 list_add(&inode
->i_lru
, &inode_lru
);
361 inodes_stat
.nr_unused
++;
365 static void inode_lru_list_del(struct inode
*inode
)
367 if (!list_empty(&inode
->i_lru
)) {
368 list_del_init(&inode
->i_lru
);
369 inodes_stat
.nr_unused
--;
373 static inline void __inode_sb_list_add(struct inode
*inode
)
375 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
379 * inode_sb_list_add - add inode to the superblock list of inodes
380 * @inode: inode to add
382 void inode_sb_list_add(struct inode
*inode
)
384 spin_lock(&inode_lock
);
385 __inode_sb_list_add(inode
);
386 spin_unlock(&inode_lock
);
388 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
390 static inline void __inode_sb_list_del(struct inode
*inode
)
392 list_del_init(&inode
->i_sb_list
);
395 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
399 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
401 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
402 return tmp
& I_HASHMASK
;
406 * __insert_inode_hash - hash an inode
407 * @inode: unhashed inode
408 * @hashval: unsigned long value used to locate this object in the
411 * Add an inode to the inode hash for this superblock.
413 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
415 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
417 spin_lock(&inode_lock
);
418 spin_lock(&inode
->i_lock
);
419 hlist_add_head(&inode
->i_hash
, b
);
420 spin_unlock(&inode
->i_lock
);
421 spin_unlock(&inode_lock
);
423 EXPORT_SYMBOL(__insert_inode_hash
);
426 * __remove_inode_hash - remove an inode from the hash
427 * @inode: inode to unhash
429 * Remove an inode from the superblock.
431 static void __remove_inode_hash(struct inode
*inode
)
433 hlist_del_init(&inode
->i_hash
);
437 * remove_inode_hash - remove an inode from the hash
438 * @inode: inode to unhash
440 * Remove an inode from the superblock.
442 void remove_inode_hash(struct inode
*inode
)
444 spin_lock(&inode_lock
);
445 spin_lock(&inode
->i_lock
);
446 hlist_del_init(&inode
->i_hash
);
447 spin_unlock(&inode
->i_lock
);
448 spin_unlock(&inode_lock
);
450 EXPORT_SYMBOL(remove_inode_hash
);
452 void end_writeback(struct inode
*inode
)
455 BUG_ON(inode
->i_data
.nrpages
);
456 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
457 BUG_ON(!(inode
->i_state
& I_FREEING
));
458 BUG_ON(inode
->i_state
& I_CLEAR
);
459 inode_sync_wait(inode
);
460 /* don't need i_lock here, no concurrent mods to i_state */
461 inode
->i_state
= I_FREEING
| I_CLEAR
;
463 EXPORT_SYMBOL(end_writeback
);
465 static void evict(struct inode
*inode
)
467 const struct super_operations
*op
= inode
->i_sb
->s_op
;
469 if (op
->evict_inode
) {
470 op
->evict_inode(inode
);
472 if (inode
->i_data
.nrpages
)
473 truncate_inode_pages(&inode
->i_data
, 0);
474 end_writeback(inode
);
476 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
478 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
483 * dispose_list - dispose of the contents of a local list
484 * @head: the head of the list to free
486 * Dispose-list gets a local list with local inodes in it, so it doesn't
487 * need to worry about list corruption and SMP locks.
489 static void dispose_list(struct list_head
*head
)
491 while (!list_empty(head
)) {
494 inode
= list_first_entry(head
, struct inode
, i_lru
);
495 list_del_init(&inode
->i_lru
);
499 spin_lock(&inode_lock
);
500 __remove_inode_hash(inode
);
501 __inode_sb_list_del(inode
);
502 spin_unlock(&inode_lock
);
504 spin_lock(&inode
->i_lock
);
505 wake_up_bit(&inode
->i_state
, __I_NEW
);
506 spin_unlock(&inode
->i_lock
);
507 destroy_inode(inode
);
512 * evict_inodes - evict all evictable inodes for a superblock
513 * @sb: superblock to operate on
515 * Make sure that no inodes with zero refcount are retained. This is
516 * called by superblock shutdown after having MS_ACTIVE flag removed,
517 * so any inode reaching zero refcount during or after that call will
518 * be immediately evicted.
520 void evict_inodes(struct super_block
*sb
)
522 struct inode
*inode
, *next
;
525 spin_lock(&inode_lock
);
526 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
527 if (atomic_read(&inode
->i_count
))
530 spin_lock(&inode
->i_lock
);
531 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
532 spin_unlock(&inode
->i_lock
);
536 inode
->i_state
|= I_FREEING
;
537 if (!(inode
->i_state
& (I_DIRTY
| I_SYNC
)))
538 inodes_stat
.nr_unused
--;
539 spin_unlock(&inode
->i_lock
);
542 * Move the inode off the IO lists and LRU once I_FREEING is
543 * set so that it won't get moved back on there if it is dirty.
545 list_move(&inode
->i_lru
, &dispose
);
546 list_del_init(&inode
->i_wb_list
);
548 spin_unlock(&inode_lock
);
550 dispose_list(&dispose
);
553 * Cycle through iprune_sem to make sure any inode that prune_icache
554 * moved off the list before we took the lock has been fully torn
557 down_write(&iprune_sem
);
558 up_write(&iprune_sem
);
562 * invalidate_inodes - attempt to free all inodes on a superblock
563 * @sb: superblock to operate on
564 * @kill_dirty: flag to guide handling of dirty inodes
566 * Attempts to free all inodes for a given superblock. If there were any
567 * busy inodes return a non-zero value, else zero.
568 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
571 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
574 struct inode
*inode
, *next
;
577 spin_lock(&inode_lock
);
578 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
579 spin_lock(&inode
->i_lock
);
580 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
581 spin_unlock(&inode
->i_lock
);
584 if (inode
->i_state
& I_DIRTY
&& !kill_dirty
) {
585 spin_unlock(&inode
->i_lock
);
589 if (atomic_read(&inode
->i_count
)) {
590 spin_unlock(&inode
->i_lock
);
595 inode
->i_state
|= I_FREEING
;
596 if (!(inode
->i_state
& (I_DIRTY
| I_SYNC
)))
597 inodes_stat
.nr_unused
--;
598 spin_unlock(&inode
->i_lock
);
601 * Move the inode off the IO lists and LRU once I_FREEING is
602 * set so that it won't get moved back on there if it is dirty.
604 list_move(&inode
->i_lru
, &dispose
);
605 list_del_init(&inode
->i_wb_list
);
607 spin_unlock(&inode_lock
);
609 dispose_list(&dispose
);
614 static int can_unuse(struct inode
*inode
)
616 if (inode
->i_state
& ~I_REFERENCED
)
618 if (inode_has_buffers(inode
))
620 if (atomic_read(&inode
->i_count
))
622 if (inode
->i_data
.nrpages
)
628 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
629 * temporary list and then are freed outside inode_lock by dispose_list().
631 * Any inodes which are pinned purely because of attached pagecache have their
632 * pagecache removed. If the inode has metadata buffers attached to
633 * mapping->private_list then try to remove them.
635 * If the inode has the I_REFERENCED flag set, then it means that it has been
636 * used recently - the flag is set in iput_final(). When we encounter such an
637 * inode, clear the flag and move it to the back of the LRU so it gets another
638 * pass through the LRU before it gets reclaimed. This is necessary because of
639 * the fact we are doing lazy LRU updates to minimise lock contention so the
640 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
641 * with this flag set because they are the inodes that are out of order.
643 static void prune_icache(int nr_to_scan
)
647 unsigned long reap
= 0;
649 down_read(&iprune_sem
);
650 spin_lock(&inode_lock
);
651 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
654 if (list_empty(&inode_lru
))
657 inode
= list_entry(inode_lru
.prev
, struct inode
, i_lru
);
660 * Referenced or dirty inodes are still in use. Give them
661 * another pass through the LRU as we canot reclaim them now.
663 spin_lock(&inode
->i_lock
);
664 if (atomic_read(&inode
->i_count
) ||
665 (inode
->i_state
& ~I_REFERENCED
)) {
666 spin_unlock(&inode
->i_lock
);
667 list_del_init(&inode
->i_lru
);
668 inodes_stat
.nr_unused
--;
672 /* recently referenced inodes get one more pass */
673 if (inode
->i_state
& I_REFERENCED
) {
674 inode
->i_state
&= ~I_REFERENCED
;
675 spin_unlock(&inode
->i_lock
);
676 list_move(&inode
->i_lru
, &inode_lru
);
679 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
681 spin_unlock(&inode
->i_lock
);
682 spin_unlock(&inode_lock
);
683 if (remove_inode_buffers(inode
))
684 reap
+= invalidate_mapping_pages(&inode
->i_data
,
687 spin_lock(&inode_lock
);
689 if (inode
!= list_entry(inode_lru
.next
,
690 struct inode
, i_lru
))
691 continue; /* wrong inode or list_empty */
692 spin_lock(&inode
->i_lock
);
693 if (!can_unuse(inode
)) {
694 spin_unlock(&inode
->i_lock
);
698 WARN_ON(inode
->i_state
& I_NEW
);
699 inode
->i_state
|= I_FREEING
;
700 spin_unlock(&inode
->i_lock
);
703 * Move the inode off the IO lists and LRU once I_FREEING is
704 * set so that it won't get moved back on there if it is dirty.
706 list_move(&inode
->i_lru
, &freeable
);
707 list_del_init(&inode
->i_wb_list
);
708 inodes_stat
.nr_unused
--;
710 if (current_is_kswapd())
711 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
713 __count_vm_events(PGINODESTEAL
, reap
);
714 spin_unlock(&inode_lock
);
716 dispose_list(&freeable
);
717 up_read(&iprune_sem
);
721 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
722 * "unused" means that no dentries are referring to the inodes: the files are
723 * not open and the dcache references to those inodes have already been
726 * This function is passed the number of inodes to scan, and it returns the
727 * total number of remaining possibly-reclaimable inodes.
729 static int shrink_icache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
733 * Nasty deadlock avoidance. We may hold various FS locks,
734 * and we don't want to recurse into the FS that called us
735 * in clear_inode() and friends..
737 if (!(gfp_mask
& __GFP_FS
))
741 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure
;
744 static struct shrinker icache_shrinker
= {
745 .shrink
= shrink_icache_memory
,
746 .seeks
= DEFAULT_SEEKS
,
749 static void __wait_on_freeing_inode(struct inode
*inode
);
751 * Called with the inode lock held.
753 static struct inode
*find_inode(struct super_block
*sb
,
754 struct hlist_head
*head
,
755 int (*test
)(struct inode
*, void *),
758 struct hlist_node
*node
;
759 struct inode
*inode
= NULL
;
762 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
763 if (inode
->i_sb
!= sb
)
765 if (!test(inode
, data
))
767 spin_lock(&inode
->i_lock
);
768 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
769 __wait_on_freeing_inode(inode
);
773 spin_unlock(&inode
->i_lock
);
780 * find_inode_fast is the fast path version of find_inode, see the comment at
781 * iget_locked for details.
783 static struct inode
*find_inode_fast(struct super_block
*sb
,
784 struct hlist_head
*head
, unsigned long ino
)
786 struct hlist_node
*node
;
787 struct inode
*inode
= NULL
;
790 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
791 if (inode
->i_ino
!= ino
)
793 if (inode
->i_sb
!= sb
)
795 spin_lock(&inode
->i_lock
);
796 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
797 __wait_on_freeing_inode(inode
);
801 spin_unlock(&inode
->i_lock
);
808 * Each cpu owns a range of LAST_INO_BATCH numbers.
809 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
810 * to renew the exhausted range.
812 * This does not significantly increase overflow rate because every CPU can
813 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
814 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
815 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
816 * overflow rate by 2x, which does not seem too significant.
818 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
819 * error if st_ino won't fit in target struct field. Use 32bit counter
820 * here to attempt to avoid that.
822 #define LAST_INO_BATCH 1024
823 static DEFINE_PER_CPU(unsigned int, last_ino
);
825 unsigned int get_next_ino(void)
827 unsigned int *p
= &get_cpu_var(last_ino
);
828 unsigned int res
= *p
;
831 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
832 static atomic_t shared_last_ino
;
833 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
835 res
= next
- LAST_INO_BATCH
;
840 put_cpu_var(last_ino
);
843 EXPORT_SYMBOL(get_next_ino
);
846 * new_inode - obtain an inode
849 * Allocates a new inode for given superblock. The default gfp_mask
850 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
851 * If HIGHMEM pages are unsuitable or it is known that pages allocated
852 * for the page cache are not reclaimable or migratable,
853 * mapping_set_gfp_mask() must be called with suitable flags on the
854 * newly created inode's mapping
857 struct inode
*new_inode(struct super_block
*sb
)
861 spin_lock_prefetch(&inode_lock
);
863 inode
= alloc_inode(sb
);
865 spin_lock(&inode_lock
);
866 spin_lock(&inode
->i_lock
);
868 spin_unlock(&inode
->i_lock
);
869 __inode_sb_list_add(inode
);
870 spin_unlock(&inode_lock
);
874 EXPORT_SYMBOL(new_inode
);
877 * unlock_new_inode - clear the I_NEW state and wake up any waiters
878 * @inode: new inode to unlock
880 * Called when the inode is fully initialised to clear the new state of the
881 * inode and wake up anyone waiting for the inode to finish initialisation.
883 void unlock_new_inode(struct inode
*inode
)
885 #ifdef CONFIG_DEBUG_LOCK_ALLOC
886 if (S_ISDIR(inode
->i_mode
)) {
887 struct file_system_type
*type
= inode
->i_sb
->s_type
;
889 /* Set new key only if filesystem hasn't already changed it */
890 if (!lockdep_match_class(&inode
->i_mutex
,
891 &type
->i_mutex_key
)) {
893 * ensure nobody is actually holding i_mutex
895 mutex_destroy(&inode
->i_mutex
);
896 mutex_init(&inode
->i_mutex
);
897 lockdep_set_class(&inode
->i_mutex
,
898 &type
->i_mutex_dir_key
);
902 spin_lock(&inode
->i_lock
);
903 WARN_ON(!(inode
->i_state
& I_NEW
));
904 inode
->i_state
&= ~I_NEW
;
905 wake_up_bit(&inode
->i_state
, __I_NEW
);
906 spin_unlock(&inode
->i_lock
);
908 EXPORT_SYMBOL(unlock_new_inode
);
911 * This is called without the inode lock held.. Be careful.
913 * We no longer cache the sb_flags in i_flags - see fs.h
914 * -- rmk@arm.uk.linux.org
916 static struct inode
*get_new_inode(struct super_block
*sb
,
917 struct hlist_head
*head
,
918 int (*test
)(struct inode
*, void *),
919 int (*set
)(struct inode
*, void *),
924 inode
= alloc_inode(sb
);
928 spin_lock(&inode_lock
);
929 /* We released the lock, so.. */
930 old
= find_inode(sb
, head
, test
, data
);
932 if (set(inode
, data
))
935 spin_lock(&inode
->i_lock
);
936 inode
->i_state
= I_NEW
;
937 hlist_add_head(&inode
->i_hash
, head
);
938 spin_unlock(&inode
->i_lock
);
939 __inode_sb_list_add(inode
);
940 spin_unlock(&inode_lock
);
942 /* Return the locked inode with I_NEW set, the
943 * caller is responsible for filling in the contents
949 * Uhhuh, somebody else created the same inode under
950 * us. Use the old inode instead of the one we just
953 spin_unlock(&inode_lock
);
954 destroy_inode(inode
);
956 wait_on_inode(inode
);
961 spin_unlock(&inode_lock
);
962 destroy_inode(inode
);
967 * get_new_inode_fast is the fast path version of get_new_inode, see the
968 * comment at iget_locked for details.
970 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
971 struct hlist_head
*head
, unsigned long ino
)
975 inode
= alloc_inode(sb
);
979 spin_lock(&inode_lock
);
980 /* We released the lock, so.. */
981 old
= find_inode_fast(sb
, head
, ino
);
984 spin_lock(&inode
->i_lock
);
985 inode
->i_state
= I_NEW
;
986 hlist_add_head(&inode
->i_hash
, head
);
987 spin_unlock(&inode
->i_lock
);
988 __inode_sb_list_add(inode
);
989 spin_unlock(&inode_lock
);
991 /* Return the locked inode with I_NEW set, the
992 * caller is responsible for filling in the contents
998 * Uhhuh, somebody else created the same inode under
999 * us. Use the old inode instead of the one we just
1002 spin_unlock(&inode_lock
);
1003 destroy_inode(inode
);
1005 wait_on_inode(inode
);
1011 * search the inode cache for a matching inode number.
1012 * If we find one, then the inode number we are trying to
1013 * allocate is not unique and so we should not use it.
1015 * Returns 1 if the inode number is unique, 0 if it is not.
1017 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1019 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1020 struct hlist_node
*node
;
1021 struct inode
*inode
;
1023 hlist_for_each_entry(inode
, node
, b
, i_hash
) {
1024 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
)
1032 * iunique - get a unique inode number
1034 * @max_reserved: highest reserved inode number
1036 * Obtain an inode number that is unique on the system for a given
1037 * superblock. This is used by file systems that have no natural
1038 * permanent inode numbering system. An inode number is returned that
1039 * is higher than the reserved limit but unique.
1042 * With a large number of inodes live on the file system this function
1043 * currently becomes quite slow.
1045 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1048 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1049 * error if st_ino won't fit in target struct field. Use 32bit counter
1050 * here to attempt to avoid that.
1052 static DEFINE_SPINLOCK(iunique_lock
);
1053 static unsigned int counter
;
1056 spin_lock(&inode_lock
);
1057 spin_lock(&iunique_lock
);
1059 if (counter
<= max_reserved
)
1060 counter
= max_reserved
+ 1;
1062 } while (!test_inode_iunique(sb
, res
));
1063 spin_unlock(&iunique_lock
);
1064 spin_unlock(&inode_lock
);
1068 EXPORT_SYMBOL(iunique
);
1070 struct inode
*igrab(struct inode
*inode
)
1072 spin_lock(&inode_lock
);
1073 spin_lock(&inode
->i_lock
);
1074 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1076 spin_unlock(&inode
->i_lock
);
1078 spin_unlock(&inode
->i_lock
);
1080 * Handle the case where s_op->clear_inode is not been
1081 * called yet, and somebody is calling igrab
1082 * while the inode is getting freed.
1086 spin_unlock(&inode_lock
);
1089 EXPORT_SYMBOL(igrab
);
1092 * ifind - internal function, you want ilookup5() or iget5().
1093 * @sb: super block of file system to search
1094 * @head: the head of the list to search
1095 * @test: callback used for comparisons between inodes
1096 * @data: opaque data pointer to pass to @test
1097 * @wait: if true wait for the inode to be unlocked, if false do not
1099 * ifind() searches for the inode specified by @data in the inode
1100 * cache. This is a generalized version of ifind_fast() for file systems where
1101 * the inode number is not sufficient for unique identification of an inode.
1103 * If the inode is in the cache, the inode is returned with an incremented
1106 * Otherwise NULL is returned.
1108 * Note, @test is called with the inode_lock held, so can't sleep.
1110 static struct inode
*ifind(struct super_block
*sb
,
1111 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
1112 void *data
, const int wait
)
1114 struct inode
*inode
;
1116 spin_lock(&inode_lock
);
1117 inode
= find_inode(sb
, head
, test
, data
);
1119 spin_unlock(&inode_lock
);
1121 wait_on_inode(inode
);
1124 spin_unlock(&inode_lock
);
1129 * ifind_fast - internal function, you want ilookup() or iget().
1130 * @sb: super block of file system to search
1131 * @head: head of the list to search
1132 * @ino: inode number to search for
1134 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1135 * file systems where the inode number is sufficient for unique identification
1138 * If the inode is in the cache, the inode is returned with an incremented
1141 * Otherwise NULL is returned.
1143 static struct inode
*ifind_fast(struct super_block
*sb
,
1144 struct hlist_head
*head
, unsigned long ino
)
1146 struct inode
*inode
;
1148 spin_lock(&inode_lock
);
1149 inode
= find_inode_fast(sb
, head
, ino
);
1151 spin_unlock(&inode_lock
);
1152 wait_on_inode(inode
);
1155 spin_unlock(&inode_lock
);
1160 * ilookup5_nowait - search for an inode in the inode cache
1161 * @sb: super block of file system to search
1162 * @hashval: hash value (usually inode number) to search for
1163 * @test: callback used for comparisons between inodes
1164 * @data: opaque data pointer to pass to @test
1166 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1167 * @data in the inode cache. This is a generalized version of ilookup() for
1168 * file systems where the inode number is not sufficient for unique
1169 * identification of an inode.
1171 * If the inode is in the cache, the inode is returned with an incremented
1172 * reference count. Note, the inode lock is not waited upon so you have to be
1173 * very careful what you do with the returned inode. You probably should be
1174 * using ilookup5() instead.
1176 * Otherwise NULL is returned.
1178 * Note, @test is called with the inode_lock held, so can't sleep.
1180 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1181 int (*test
)(struct inode
*, void *), void *data
)
1183 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1185 return ifind(sb
, head
, test
, data
, 0);
1187 EXPORT_SYMBOL(ilookup5_nowait
);
1190 * ilookup5 - search for an inode in the inode cache
1191 * @sb: super block of file system to search
1192 * @hashval: hash value (usually inode number) to search for
1193 * @test: callback used for comparisons between inodes
1194 * @data: opaque data pointer to pass to @test
1196 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1197 * @data in the inode cache. This is a generalized version of ilookup() for
1198 * file systems where the inode number is not sufficient for unique
1199 * identification of an inode.
1201 * If the inode is in the cache, the inode lock is waited upon and the inode is
1202 * returned with an incremented reference count.
1204 * Otherwise NULL is returned.
1206 * Note, @test is called with the inode_lock held, so can't sleep.
1208 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1209 int (*test
)(struct inode
*, void *), void *data
)
1211 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1213 return ifind(sb
, head
, test
, data
, 1);
1215 EXPORT_SYMBOL(ilookup5
);
1218 * ilookup - search for an inode in the inode cache
1219 * @sb: super block of file system to search
1220 * @ino: inode number to search for
1222 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1223 * This is for file systems where the inode number is sufficient for unique
1224 * identification of an inode.
1226 * If the inode is in the cache, the inode is returned with an incremented
1229 * Otherwise NULL is returned.
1231 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1233 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1235 return ifind_fast(sb
, head
, ino
);
1237 EXPORT_SYMBOL(ilookup
);
1240 * iget5_locked - obtain an inode from a mounted file system
1241 * @sb: super block of file system
1242 * @hashval: hash value (usually inode number) to get
1243 * @test: callback used for comparisons between inodes
1244 * @set: callback used to initialize a new struct inode
1245 * @data: opaque data pointer to pass to @test and @set
1247 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1248 * and @data in the inode cache and if present it is returned with an increased
1249 * reference count. This is a generalized version of iget_locked() for file
1250 * systems where the inode number is not sufficient for unique identification
1253 * If the inode is not in cache, get_new_inode() is called to allocate a new
1254 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1255 * file system gets to fill it in before unlocking it via unlock_new_inode().
1257 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1259 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1260 int (*test
)(struct inode
*, void *),
1261 int (*set
)(struct inode
*, void *), void *data
)
1263 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1264 struct inode
*inode
;
1266 inode
= ifind(sb
, head
, test
, data
, 1);
1270 * get_new_inode() will do the right thing, re-trying the search
1271 * in case it had to block at any point.
1273 return get_new_inode(sb
, head
, test
, set
, data
);
1275 EXPORT_SYMBOL(iget5_locked
);
1278 * iget_locked - obtain an inode from a mounted file system
1279 * @sb: super block of file system
1280 * @ino: inode number to get
1282 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1283 * the inode cache and if present it is returned with an increased reference
1284 * count. This is for file systems where the inode number is sufficient for
1285 * unique identification of an inode.
1287 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1288 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1289 * The file system gets to fill it in before unlocking it via
1290 * unlock_new_inode().
1292 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1294 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1295 struct inode
*inode
;
1297 inode
= ifind_fast(sb
, head
, ino
);
1301 * get_new_inode_fast() will do the right thing, re-trying the search
1302 * in case it had to block at any point.
1304 return get_new_inode_fast(sb
, head
, ino
);
1306 EXPORT_SYMBOL(iget_locked
);
1308 int insert_inode_locked(struct inode
*inode
)
1310 struct super_block
*sb
= inode
->i_sb
;
1311 ino_t ino
= inode
->i_ino
;
1312 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1315 struct hlist_node
*node
;
1316 struct inode
*old
= NULL
;
1317 spin_lock(&inode_lock
);
1318 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1319 if (old
->i_ino
!= ino
)
1321 if (old
->i_sb
!= sb
)
1323 spin_lock(&old
->i_lock
);
1324 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1325 spin_unlock(&old
->i_lock
);
1330 if (likely(!node
)) {
1331 spin_lock(&inode
->i_lock
);
1332 inode
->i_state
|= I_NEW
;
1333 hlist_add_head(&inode
->i_hash
, head
);
1334 spin_unlock(&inode
->i_lock
);
1335 spin_unlock(&inode_lock
);
1339 spin_unlock(&old
->i_lock
);
1340 spin_unlock(&inode_lock
);
1342 if (unlikely(!inode_unhashed(old
))) {
1349 EXPORT_SYMBOL(insert_inode_locked
);
1351 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1352 int (*test
)(struct inode
*, void *), void *data
)
1354 struct super_block
*sb
= inode
->i_sb
;
1355 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1358 struct hlist_node
*node
;
1359 struct inode
*old
= NULL
;
1361 spin_lock(&inode_lock
);
1362 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1363 if (old
->i_sb
!= sb
)
1365 if (!test(old
, data
))
1367 spin_lock(&old
->i_lock
);
1368 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1369 spin_unlock(&old
->i_lock
);
1374 if (likely(!node
)) {
1375 spin_lock(&inode
->i_lock
);
1376 inode
->i_state
|= I_NEW
;
1377 hlist_add_head(&inode
->i_hash
, head
);
1378 spin_unlock(&inode
->i_lock
);
1379 spin_unlock(&inode_lock
);
1383 spin_unlock(&old
->i_lock
);
1384 spin_unlock(&inode_lock
);
1386 if (unlikely(!inode_unhashed(old
))) {
1393 EXPORT_SYMBOL(insert_inode_locked4
);
1396 int generic_delete_inode(struct inode
*inode
)
1400 EXPORT_SYMBOL(generic_delete_inode
);
1403 * Normal UNIX filesystem behaviour: delete the
1404 * inode when the usage count drops to zero, and
1407 int generic_drop_inode(struct inode
*inode
)
1409 return !inode
->i_nlink
|| inode_unhashed(inode
);
1411 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1414 * Called when we're dropping the last reference
1417 * Call the FS "drop_inode()" function, defaulting to
1418 * the legacy UNIX filesystem behaviour. If it tells
1419 * us to evict inode, do so. Otherwise, retain inode
1420 * in cache if fs is alive, sync and evict if fs is
1423 static void iput_final(struct inode
*inode
)
1425 struct super_block
*sb
= inode
->i_sb
;
1426 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1429 spin_lock(&inode
->i_lock
);
1430 WARN_ON(inode
->i_state
& I_NEW
);
1432 if (op
&& op
->drop_inode
)
1433 drop
= op
->drop_inode(inode
);
1435 drop
= generic_drop_inode(inode
);
1438 if (sb
->s_flags
& MS_ACTIVE
) {
1439 inode
->i_state
|= I_REFERENCED
;
1440 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
))) {
1441 inode_lru_list_add(inode
);
1443 spin_unlock(&inode
->i_lock
);
1444 spin_unlock(&inode_lock
);
1447 inode
->i_state
|= I_WILL_FREE
;
1448 spin_unlock(&inode
->i_lock
);
1449 spin_unlock(&inode_lock
);
1450 write_inode_now(inode
, 1);
1451 spin_lock(&inode_lock
);
1452 spin_lock(&inode
->i_lock
);
1453 WARN_ON(inode
->i_state
& I_NEW
);
1454 inode
->i_state
&= ~I_WILL_FREE
;
1455 __remove_inode_hash(inode
);
1458 inode
->i_state
|= I_FREEING
;
1459 spin_unlock(&inode
->i_lock
);
1462 * Move the inode off the IO lists and LRU once I_FREEING is
1463 * set so that it won't get moved back on there if it is dirty.
1465 inode_lru_list_del(inode
);
1466 list_del_init(&inode
->i_wb_list
);
1468 __inode_sb_list_del(inode
);
1469 spin_unlock(&inode_lock
);
1471 remove_inode_hash(inode
);
1472 spin_lock(&inode
->i_lock
);
1473 wake_up_bit(&inode
->i_state
, __I_NEW
);
1474 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
1475 spin_unlock(&inode
->i_lock
);
1476 destroy_inode(inode
);
1480 * iput - put an inode
1481 * @inode: inode to put
1483 * Puts an inode, dropping its usage count. If the inode use count hits
1484 * zero, the inode is then freed and may also be destroyed.
1486 * Consequently, iput() can sleep.
1488 void iput(struct inode
*inode
)
1491 BUG_ON(inode
->i_state
& I_CLEAR
);
1493 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1497 EXPORT_SYMBOL(iput
);
1500 * bmap - find a block number in a file
1501 * @inode: inode of file
1502 * @block: block to find
1504 * Returns the block number on the device holding the inode that
1505 * is the disk block number for the block of the file requested.
1506 * That is, asked for block 4 of inode 1 the function will return the
1507 * disk block relative to the disk start that holds that block of the
1510 sector_t
bmap(struct inode
*inode
, sector_t block
)
1513 if (inode
->i_mapping
->a_ops
->bmap
)
1514 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1517 EXPORT_SYMBOL(bmap
);
1520 * With relative atime, only update atime if the previous atime is
1521 * earlier than either the ctime or mtime or if at least a day has
1522 * passed since the last atime update.
1524 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1525 struct timespec now
)
1528 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1531 * Is mtime younger than atime? If yes, update atime:
1533 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1536 * Is ctime younger than atime? If yes, update atime:
1538 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1542 * Is the previous atime value older than a day? If yes,
1545 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1548 * Good, we can skip the atime update:
1554 * touch_atime - update the access time
1555 * @mnt: mount the inode is accessed on
1556 * @dentry: dentry accessed
1558 * Update the accessed time on an inode and mark it for writeback.
1559 * This function automatically handles read only file systems and media,
1560 * as well as the "noatime" flag and inode specific "noatime" markers.
1562 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1564 struct inode
*inode
= dentry
->d_inode
;
1565 struct timespec now
;
1567 if (inode
->i_flags
& S_NOATIME
)
1569 if (IS_NOATIME(inode
))
1571 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1574 if (mnt
->mnt_flags
& MNT_NOATIME
)
1576 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1579 now
= current_fs_time(inode
->i_sb
);
1581 if (!relatime_need_update(mnt
, inode
, now
))
1584 if (timespec_equal(&inode
->i_atime
, &now
))
1587 if (mnt_want_write(mnt
))
1590 inode
->i_atime
= now
;
1591 mark_inode_dirty_sync(inode
);
1592 mnt_drop_write(mnt
);
1594 EXPORT_SYMBOL(touch_atime
);
1597 * file_update_time - update mtime and ctime time
1598 * @file: file accessed
1600 * Update the mtime and ctime members of an inode and mark the inode
1601 * for writeback. Note that this function is meant exclusively for
1602 * usage in the file write path of filesystems, and filesystems may
1603 * choose to explicitly ignore update via this function with the
1604 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1605 * timestamps are handled by the server.
1608 void file_update_time(struct file
*file
)
1610 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1611 struct timespec now
;
1612 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1614 /* First try to exhaust all avenues to not sync */
1615 if (IS_NOCMTIME(inode
))
1618 now
= current_fs_time(inode
->i_sb
);
1619 if (!timespec_equal(&inode
->i_mtime
, &now
))
1622 if (!timespec_equal(&inode
->i_ctime
, &now
))
1625 if (IS_I_VERSION(inode
))
1626 sync_it
|= S_VERSION
;
1631 /* Finally allowed to write? Takes lock. */
1632 if (mnt_want_write_file(file
))
1635 /* Only change inode inside the lock region */
1636 if (sync_it
& S_VERSION
)
1637 inode_inc_iversion(inode
);
1638 if (sync_it
& S_CTIME
)
1639 inode
->i_ctime
= now
;
1640 if (sync_it
& S_MTIME
)
1641 inode
->i_mtime
= now
;
1642 mark_inode_dirty_sync(inode
);
1643 mnt_drop_write(file
->f_path
.mnt
);
1645 EXPORT_SYMBOL(file_update_time
);
1647 int inode_needs_sync(struct inode
*inode
)
1651 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1655 EXPORT_SYMBOL(inode_needs_sync
);
1657 int inode_wait(void *word
)
1662 EXPORT_SYMBOL(inode_wait
);
1665 * If we try to find an inode in the inode hash while it is being
1666 * deleted, we have to wait until the filesystem completes its
1667 * deletion before reporting that it isn't found. This function waits
1668 * until the deletion _might_ have completed. Callers are responsible
1669 * to recheck inode state.
1671 * It doesn't matter if I_NEW is not set initially, a call to
1672 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1675 static void __wait_on_freeing_inode(struct inode
*inode
)
1677 wait_queue_head_t
*wq
;
1678 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1679 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1680 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1681 spin_unlock(&inode
->i_lock
);
1682 spin_unlock(&inode_lock
);
1684 finish_wait(wq
, &wait
.wait
);
1685 spin_lock(&inode_lock
);
1688 static __initdata
unsigned long ihash_entries
;
1689 static int __init
set_ihash_entries(char *str
)
1693 ihash_entries
= simple_strtoul(str
, &str
, 0);
1696 __setup("ihash_entries=", set_ihash_entries
);
1699 * Initialize the waitqueues and inode hash table.
1701 void __init
inode_init_early(void)
1705 /* If hashes are distributed across NUMA nodes, defer
1706 * hash allocation until vmalloc space is available.
1712 alloc_large_system_hash("Inode-cache",
1713 sizeof(struct hlist_head
),
1721 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1722 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1725 void __init
inode_init(void)
1729 /* inode slab cache */
1730 inode_cachep
= kmem_cache_create("inode_cache",
1731 sizeof(struct inode
),
1733 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1736 register_shrinker(&icache_shrinker
);
1738 /* Hash may have been set up in inode_init_early */
1743 alloc_large_system_hash("Inode-cache",
1744 sizeof(struct hlist_head
),
1752 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1753 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1756 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1758 inode
->i_mode
= mode
;
1759 if (S_ISCHR(mode
)) {
1760 inode
->i_fop
= &def_chr_fops
;
1761 inode
->i_rdev
= rdev
;
1762 } else if (S_ISBLK(mode
)) {
1763 inode
->i_fop
= &def_blk_fops
;
1764 inode
->i_rdev
= rdev
;
1765 } else if (S_ISFIFO(mode
))
1766 inode
->i_fop
= &def_fifo_fops
;
1767 else if (S_ISSOCK(mode
))
1768 inode
->i_fop
= &bad_sock_fops
;
1770 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1771 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1774 EXPORT_SYMBOL(init_special_inode
);
1777 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1779 * @dir: Directory inode
1780 * @mode: mode of the new inode
1782 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1785 inode
->i_uid
= current_fsuid();
1786 if (dir
&& dir
->i_mode
& S_ISGID
) {
1787 inode
->i_gid
= dir
->i_gid
;
1791 inode
->i_gid
= current_fsgid();
1792 inode
->i_mode
= mode
;
1794 EXPORT_SYMBOL(inode_init_owner
);
1797 * inode_owner_or_capable - check current task permissions to inode
1798 * @inode: inode being checked
1800 * Return true if current either has CAP_FOWNER to the inode, or
1803 bool inode_owner_or_capable(const struct inode
*inode
)
1805 struct user_namespace
*ns
= inode_userns(inode
);
1807 if (current_user_ns() == ns
&& current_fsuid() == inode
->i_uid
)
1809 if (ns_capable(ns
, CAP_FOWNER
))
1813 EXPORT_SYMBOL(inode_owner_or_capable
);