2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
5 #include <linux/export.h>
8 #include <linux/backing-dev.h>
9 #include <linux/hash.h>
10 #include <linux/swap.h>
11 #include <linux/security.h>
12 #include <linux/cdev.h>
13 #include <linux/bootmem.h>
14 #include <linux/fsnotify.h>
15 #include <linux/mount.h>
16 #include <linux/posix_acl.h>
17 #include <linux/prefetch.h>
18 #include <linux/buffer_head.h> /* for inode_has_buffers */
19 #include <linux/ratelimit.h>
20 #include <linux/list_lru.h>
24 * Inode locking rules:
26 * inode->i_lock protects:
27 * inode->i_state, inode->i_hash, __iget()
28 * Inode LRU list locks protect:
29 * inode->i_sb->s_inode_lru, inode->i_lru
30 * inode_sb_list_lock protects:
31 * sb->s_inodes, inode->i_sb_list
32 * bdi->wb.list_lock protects:
33 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
34 * inode_hash_lock protects:
35 * inode_hashtable, inode->i_hash
41 * Inode LRU list locks
54 static unsigned int i_hash_mask __read_mostly
;
55 static unsigned int i_hash_shift __read_mostly
;
56 static struct hlist_head
*inode_hashtable __read_mostly
;
57 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_hash_lock
);
59 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_sb_list_lock
);
62 * Empty aops. Can be used for the cases where the user does not
63 * define any of the address_space operations.
65 const struct address_space_operations empty_aops
= {
67 EXPORT_SYMBOL(empty_aops
);
70 * Statistics gathering..
72 struct inodes_stat_t inodes_stat
;
74 static DEFINE_PER_CPU(unsigned long, nr_inodes
);
75 static DEFINE_PER_CPU(unsigned long, nr_unused
);
77 static struct kmem_cache
*inode_cachep __read_mostly
;
79 static long get_nr_inodes(void)
83 for_each_possible_cpu(i
)
84 sum
+= per_cpu(nr_inodes
, i
);
85 return sum
< 0 ? 0 : sum
;
88 static inline long get_nr_inodes_unused(void)
92 for_each_possible_cpu(i
)
93 sum
+= per_cpu(nr_unused
, i
);
94 return sum
< 0 ? 0 : sum
;
97 long get_nr_dirty_inodes(void)
99 /* not actually dirty inodes, but a wild approximation */
100 long nr_dirty
= get_nr_inodes() - get_nr_inodes_unused();
101 return nr_dirty
> 0 ? nr_dirty
: 0;
105 * Handle nr_inode sysctl
108 int proc_nr_inodes(ctl_table
*table
, int write
,
109 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
111 inodes_stat
.nr_inodes
= get_nr_inodes();
112 inodes_stat
.nr_unused
= get_nr_inodes_unused();
113 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
118 * inode_init_always - perform inode structure intialisation
119 * @sb: superblock inode belongs to
120 * @inode: inode to initialise
122 * These are initializations that need to be done on every inode
123 * allocation as the fields are not initialised by slab allocation.
125 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
127 static const struct inode_operations empty_iops
;
128 static const struct file_operations empty_fops
;
129 struct address_space
*const mapping
= &inode
->i_data
;
132 inode
->i_blkbits
= sb
->s_blocksize_bits
;
134 atomic_set(&inode
->i_count
, 1);
135 inode
->i_op
= &empty_iops
;
136 inode
->i_fop
= &empty_fops
;
137 inode
->__i_nlink
= 1;
138 inode
->i_opflags
= 0;
139 i_uid_write(inode
, 0);
140 i_gid_write(inode
, 0);
141 atomic_set(&inode
->i_writecount
, 0);
145 inode
->i_generation
= 0;
147 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
149 inode
->i_pipe
= NULL
;
150 inode
->i_bdev
= NULL
;
151 inode
->i_cdev
= NULL
;
153 inode
->dirtied_when
= 0;
155 if (security_inode_alloc(inode
))
157 spin_lock_init(&inode
->i_lock
);
158 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
160 mutex_init(&inode
->i_mutex
);
161 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
163 atomic_set(&inode
->i_dio_count
, 0);
165 mapping
->a_ops
= &empty_aops
;
166 mapping
->host
= inode
;
168 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
169 mapping
->private_data
= NULL
;
170 mapping
->backing_dev_info
= &default_backing_dev_info
;
171 mapping
->writeback_index
= 0;
174 * If the block_device provides a backing_dev_info for client
175 * inodes then use that. Otherwise the inode share the bdev's
179 struct backing_dev_info
*bdi
;
181 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
182 mapping
->backing_dev_info
= bdi
;
184 inode
->i_private
= NULL
;
185 inode
->i_mapping
= mapping
;
186 INIT_HLIST_HEAD(&inode
->i_dentry
); /* buggered by rcu freeing */
187 #ifdef CONFIG_FS_POSIX_ACL
188 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
191 #ifdef CONFIG_FSNOTIFY
192 inode
->i_fsnotify_mask
= 0;
195 this_cpu_inc(nr_inodes
);
201 EXPORT_SYMBOL(inode_init_always
);
203 static struct inode
*alloc_inode(struct super_block
*sb
)
207 if (sb
->s_op
->alloc_inode
)
208 inode
= sb
->s_op
->alloc_inode(sb
);
210 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
215 if (unlikely(inode_init_always(sb
, inode
))) {
216 if (inode
->i_sb
->s_op
->destroy_inode
)
217 inode
->i_sb
->s_op
->destroy_inode(inode
);
219 kmem_cache_free(inode_cachep
, inode
);
226 void free_inode_nonrcu(struct inode
*inode
)
228 kmem_cache_free(inode_cachep
, inode
);
230 EXPORT_SYMBOL(free_inode_nonrcu
);
232 void __destroy_inode(struct inode
*inode
)
234 BUG_ON(inode_has_buffers(inode
));
235 security_inode_free(inode
);
236 fsnotify_inode_delete(inode
);
237 if (!inode
->i_nlink
) {
238 WARN_ON(atomic_long_read(&inode
->i_sb
->s_remove_count
) == 0);
239 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
242 #ifdef CONFIG_FS_POSIX_ACL
243 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
244 posix_acl_release(inode
->i_acl
);
245 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
246 posix_acl_release(inode
->i_default_acl
);
248 this_cpu_dec(nr_inodes
);
250 EXPORT_SYMBOL(__destroy_inode
);
252 static void i_callback(struct rcu_head
*head
)
254 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
255 kmem_cache_free(inode_cachep
, inode
);
258 static void destroy_inode(struct inode
*inode
)
260 BUG_ON(!list_empty(&inode
->i_lru
));
261 __destroy_inode(inode
);
262 if (inode
->i_sb
->s_op
->destroy_inode
)
263 inode
->i_sb
->s_op
->destroy_inode(inode
);
265 call_rcu(&inode
->i_rcu
, i_callback
);
269 * drop_nlink - directly drop an inode's link count
272 * This is a low-level filesystem helper to replace any
273 * direct filesystem manipulation of i_nlink. In cases
274 * where we are attempting to track writes to the
275 * filesystem, a decrement to zero means an imminent
276 * write when the file is truncated and actually unlinked
279 void drop_nlink(struct inode
*inode
)
281 WARN_ON(inode
->i_nlink
== 0);
284 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
286 EXPORT_SYMBOL(drop_nlink
);
289 * clear_nlink - directly zero an inode's link count
292 * This is a low-level filesystem helper to replace any
293 * direct filesystem manipulation of i_nlink. See
294 * drop_nlink() for why we care about i_nlink hitting zero.
296 void clear_nlink(struct inode
*inode
)
298 if (inode
->i_nlink
) {
299 inode
->__i_nlink
= 0;
300 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
303 EXPORT_SYMBOL(clear_nlink
);
306 * set_nlink - directly set an inode's link count
308 * @nlink: new nlink (should be non-zero)
310 * This is a low-level filesystem helper to replace any
311 * direct filesystem manipulation of i_nlink.
313 void set_nlink(struct inode
*inode
, unsigned int nlink
)
318 /* Yes, some filesystems do change nlink from zero to one */
319 if (inode
->i_nlink
== 0)
320 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
322 inode
->__i_nlink
= nlink
;
325 EXPORT_SYMBOL(set_nlink
);
328 * inc_nlink - directly increment an inode's link count
331 * This is a low-level filesystem helper to replace any
332 * direct filesystem manipulation of i_nlink. Currently,
333 * it is only here for parity with dec_nlink().
335 void inc_nlink(struct inode
*inode
)
337 if (unlikely(inode
->i_nlink
== 0)) {
338 WARN_ON(!(inode
->i_state
& I_LINKABLE
));
339 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
344 EXPORT_SYMBOL(inc_nlink
);
346 void address_space_init_once(struct address_space
*mapping
)
348 memset(mapping
, 0, sizeof(*mapping
));
349 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
350 spin_lock_init(&mapping
->tree_lock
);
351 mutex_init(&mapping
->i_mmap_mutex
);
352 INIT_LIST_HEAD(&mapping
->private_list
);
353 spin_lock_init(&mapping
->private_lock
);
354 mapping
->i_mmap
= RB_ROOT
;
355 INIT_LIST_HEAD(&mapping
->i_mmap_nonlinear
);
357 EXPORT_SYMBOL(address_space_init_once
);
360 * These are initializations that only need to be done
361 * once, because the fields are idempotent across use
362 * of the inode, so let the slab aware of that.
364 void inode_init_once(struct inode
*inode
)
366 memset(inode
, 0, sizeof(*inode
));
367 INIT_HLIST_NODE(&inode
->i_hash
);
368 INIT_LIST_HEAD(&inode
->i_devices
);
369 INIT_LIST_HEAD(&inode
->i_wb_list
);
370 INIT_LIST_HEAD(&inode
->i_lru
);
371 address_space_init_once(&inode
->i_data
);
372 i_size_ordered_init(inode
);
373 #ifdef CONFIG_FSNOTIFY
374 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
377 EXPORT_SYMBOL(inode_init_once
);
379 static void init_once(void *foo
)
381 struct inode
*inode
= (struct inode
*) foo
;
383 inode_init_once(inode
);
387 * inode->i_lock must be held
389 void __iget(struct inode
*inode
)
391 atomic_inc(&inode
->i_count
);
395 * get additional reference to inode; caller must already hold one.
397 void ihold(struct inode
*inode
)
399 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
401 EXPORT_SYMBOL(ihold
);
403 static void inode_lru_list_add(struct inode
*inode
)
405 if (list_lru_add(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
406 this_cpu_inc(nr_unused
);
410 * Add inode to LRU if needed (inode is unused and clean).
412 * Needs inode->i_lock held.
414 void inode_add_lru(struct inode
*inode
)
416 if (!(inode
->i_state
& (I_DIRTY
| I_SYNC
| I_FREEING
| I_WILL_FREE
)) &&
417 !atomic_read(&inode
->i_count
) && inode
->i_sb
->s_flags
& MS_ACTIVE
)
418 inode_lru_list_add(inode
);
422 static void inode_lru_list_del(struct inode
*inode
)
425 if (list_lru_del(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
426 this_cpu_dec(nr_unused
);
430 * inode_sb_list_add - add inode to the superblock list of inodes
431 * @inode: inode to add
433 void inode_sb_list_add(struct inode
*inode
)
435 spin_lock(&inode_sb_list_lock
);
436 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
437 spin_unlock(&inode_sb_list_lock
);
439 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
441 static inline void inode_sb_list_del(struct inode
*inode
)
443 if (!list_empty(&inode
->i_sb_list
)) {
444 spin_lock(&inode_sb_list_lock
);
445 list_del_init(&inode
->i_sb_list
);
446 spin_unlock(&inode_sb_list_lock
);
450 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
454 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
456 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> i_hash_shift
);
457 return tmp
& i_hash_mask
;
461 * __insert_inode_hash - hash an inode
462 * @inode: unhashed inode
463 * @hashval: unsigned long value used to locate this object in the
466 * Add an inode to the inode hash for this superblock.
468 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
470 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
472 spin_lock(&inode_hash_lock
);
473 spin_lock(&inode
->i_lock
);
474 hlist_add_head(&inode
->i_hash
, b
);
475 spin_unlock(&inode
->i_lock
);
476 spin_unlock(&inode_hash_lock
);
478 EXPORT_SYMBOL(__insert_inode_hash
);
481 * __remove_inode_hash - remove an inode from the hash
482 * @inode: inode to unhash
484 * Remove an inode from the superblock.
486 void __remove_inode_hash(struct inode
*inode
)
488 spin_lock(&inode_hash_lock
);
489 spin_lock(&inode
->i_lock
);
490 hlist_del_init(&inode
->i_hash
);
491 spin_unlock(&inode
->i_lock
);
492 spin_unlock(&inode_hash_lock
);
494 EXPORT_SYMBOL(__remove_inode_hash
);
496 void clear_inode(struct inode
*inode
)
500 * We have to cycle tree_lock here because reclaim can be still in the
501 * process of removing the last page (in __delete_from_page_cache())
502 * and we must not free mapping under it.
504 spin_lock_irq(&inode
->i_data
.tree_lock
);
505 BUG_ON(inode
->i_data
.nrpages
);
506 spin_unlock_irq(&inode
->i_data
.tree_lock
);
507 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
508 BUG_ON(!(inode
->i_state
& I_FREEING
));
509 BUG_ON(inode
->i_state
& I_CLEAR
);
510 /* don't need i_lock here, no concurrent mods to i_state */
511 inode
->i_state
= I_FREEING
| I_CLEAR
;
513 EXPORT_SYMBOL(clear_inode
);
516 * Free the inode passed in, removing it from the lists it is still connected
517 * to. We remove any pages still attached to the inode and wait for any IO that
518 * is still in progress before finally destroying the inode.
520 * An inode must already be marked I_FREEING so that we avoid the inode being
521 * moved back onto lists if we race with other code that manipulates the lists
522 * (e.g. writeback_single_inode). The caller is responsible for setting this.
524 * An inode must already be removed from the LRU list before being evicted from
525 * the cache. This should occur atomically with setting the I_FREEING state
526 * flag, so no inodes here should ever be on the LRU when being evicted.
528 static void evict(struct inode
*inode
)
530 const struct super_operations
*op
= inode
->i_sb
->s_op
;
532 BUG_ON(!(inode
->i_state
& I_FREEING
));
533 BUG_ON(!list_empty(&inode
->i_lru
));
535 if (!list_empty(&inode
->i_wb_list
))
536 inode_wb_list_del(inode
);
538 inode_sb_list_del(inode
);
541 * Wait for flusher thread to be done with the inode so that filesystem
542 * does not start destroying it while writeback is still running. Since
543 * the inode has I_FREEING set, flusher thread won't start new work on
544 * the inode. We just have to wait for running writeback to finish.
546 inode_wait_for_writeback(inode
);
548 if (op
->evict_inode
) {
549 op
->evict_inode(inode
);
551 if (inode
->i_data
.nrpages
)
552 truncate_inode_pages(&inode
->i_data
, 0);
555 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
557 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
560 remove_inode_hash(inode
);
562 spin_lock(&inode
->i_lock
);
563 wake_up_bit(&inode
->i_state
, __I_NEW
);
564 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
565 spin_unlock(&inode
->i_lock
);
567 destroy_inode(inode
);
571 * dispose_list - dispose of the contents of a local list
572 * @head: the head of the list to free
574 * Dispose-list gets a local list with local inodes in it, so it doesn't
575 * need to worry about list corruption and SMP locks.
577 static void dispose_list(struct list_head
*head
)
579 while (!list_empty(head
)) {
582 inode
= list_first_entry(head
, struct inode
, i_lru
);
583 list_del_init(&inode
->i_lru
);
590 * evict_inodes - evict all evictable inodes for a superblock
591 * @sb: superblock to operate on
593 * Make sure that no inodes with zero refcount are retained. This is
594 * called by superblock shutdown after having MS_ACTIVE flag removed,
595 * so any inode reaching zero refcount during or after that call will
596 * be immediately evicted.
598 void evict_inodes(struct super_block
*sb
)
600 struct inode
*inode
, *next
;
603 spin_lock(&inode_sb_list_lock
);
604 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
605 if (atomic_read(&inode
->i_count
))
608 spin_lock(&inode
->i_lock
);
609 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
610 spin_unlock(&inode
->i_lock
);
614 inode
->i_state
|= I_FREEING
;
615 inode_lru_list_del(inode
);
616 spin_unlock(&inode
->i_lock
);
617 list_add(&inode
->i_lru
, &dispose
);
619 spin_unlock(&inode_sb_list_lock
);
621 dispose_list(&dispose
);
625 * invalidate_inodes - attempt to free all inodes on a superblock
626 * @sb: superblock to operate on
627 * @kill_dirty: flag to guide handling of dirty inodes
629 * Attempts to free all inodes for a given superblock. If there were any
630 * busy inodes return a non-zero value, else zero.
631 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
634 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
637 struct inode
*inode
, *next
;
640 spin_lock(&inode_sb_list_lock
);
641 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
642 spin_lock(&inode
->i_lock
);
643 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
644 spin_unlock(&inode
->i_lock
);
647 if (inode
->i_state
& I_DIRTY
&& !kill_dirty
) {
648 spin_unlock(&inode
->i_lock
);
652 if (atomic_read(&inode
->i_count
)) {
653 spin_unlock(&inode
->i_lock
);
658 inode
->i_state
|= I_FREEING
;
659 inode_lru_list_del(inode
);
660 spin_unlock(&inode
->i_lock
);
661 list_add(&inode
->i_lru
, &dispose
);
663 spin_unlock(&inode_sb_list_lock
);
665 dispose_list(&dispose
);
671 * Isolate the inode from the LRU in preparation for freeing it.
673 * Any inodes which are pinned purely because of attached pagecache have their
674 * pagecache removed. If the inode has metadata buffers attached to
675 * mapping->private_list then try to remove them.
677 * If the inode has the I_REFERENCED flag set, then it means that it has been
678 * used recently - the flag is set in iput_final(). When we encounter such an
679 * inode, clear the flag and move it to the back of the LRU so it gets another
680 * pass through the LRU before it gets reclaimed. This is necessary because of
681 * the fact we are doing lazy LRU updates to minimise lock contention so the
682 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
683 * with this flag set because they are the inodes that are out of order.
685 static enum lru_status
686 inode_lru_isolate(struct list_head
*item
, spinlock_t
*lru_lock
, void *arg
)
688 struct list_head
*freeable
= arg
;
689 struct inode
*inode
= container_of(item
, struct inode
, i_lru
);
692 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
693 * If we fail to get the lock, just skip it.
695 if (!spin_trylock(&inode
->i_lock
))
699 * Referenced or dirty inodes are still in use. Give them another pass
700 * through the LRU as we canot reclaim them now.
702 if (atomic_read(&inode
->i_count
) ||
703 (inode
->i_state
& ~I_REFERENCED
)) {
704 list_del_init(&inode
->i_lru
);
705 spin_unlock(&inode
->i_lock
);
706 this_cpu_dec(nr_unused
);
710 /* recently referenced inodes get one more pass */
711 if (inode
->i_state
& I_REFERENCED
) {
712 inode
->i_state
&= ~I_REFERENCED
;
713 spin_unlock(&inode
->i_lock
);
717 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
719 spin_unlock(&inode
->i_lock
);
720 spin_unlock(lru_lock
);
721 if (remove_inode_buffers(inode
)) {
723 reap
= invalidate_mapping_pages(&inode
->i_data
, 0, -1);
724 if (current_is_kswapd())
725 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
727 __count_vm_events(PGINODESTEAL
, reap
);
728 if (current
->reclaim_state
)
729 current
->reclaim_state
->reclaimed_slab
+= reap
;
736 WARN_ON(inode
->i_state
& I_NEW
);
737 inode
->i_state
|= I_FREEING
;
738 list_move(&inode
->i_lru
, freeable
);
739 spin_unlock(&inode
->i_lock
);
741 this_cpu_dec(nr_unused
);
746 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
747 * This is called from the superblock shrinker function with a number of inodes
748 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
749 * then are freed outside inode_lock by dispose_list().
751 long prune_icache_sb(struct super_block
*sb
, unsigned long nr_to_scan
,
757 freed
= list_lru_walk_node(&sb
->s_inode_lru
, nid
, inode_lru_isolate
,
758 &freeable
, &nr_to_scan
);
759 dispose_list(&freeable
);
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 inode
*inode
= NULL
;
775 hlist_for_each_entry(inode
, head
, i_hash
) {
776 if (inode
->i_sb
!= sb
)
778 if (!test(inode
, data
))
780 spin_lock(&inode
->i_lock
);
781 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
782 __wait_on_freeing_inode(inode
);
786 spin_unlock(&inode
->i_lock
);
793 * find_inode_fast is the fast path version of find_inode, see the comment at
794 * iget_locked for details.
796 static struct inode
*find_inode_fast(struct super_block
*sb
,
797 struct hlist_head
*head
, unsigned long ino
)
799 struct inode
*inode
= NULL
;
802 hlist_for_each_entry(inode
, head
, i_hash
) {
803 if (inode
->i_ino
!= ino
)
805 if (inode
->i_sb
!= sb
)
807 spin_lock(&inode
->i_lock
);
808 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
809 __wait_on_freeing_inode(inode
);
813 spin_unlock(&inode
->i_lock
);
820 * Each cpu owns a range of LAST_INO_BATCH numbers.
821 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
822 * to renew the exhausted range.
824 * This does not significantly increase overflow rate because every CPU can
825 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
826 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
827 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
828 * overflow rate by 2x, which does not seem too significant.
830 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
831 * error if st_ino won't fit in target struct field. Use 32bit counter
832 * here to attempt to avoid that.
834 #define LAST_INO_BATCH 1024
835 static DEFINE_PER_CPU(unsigned int, last_ino
);
837 unsigned int get_next_ino(void)
839 unsigned int *p
= &get_cpu_var(last_ino
);
840 unsigned int res
= *p
;
843 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
844 static atomic_t shared_last_ino
;
845 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
847 res
= next
- LAST_INO_BATCH
;
852 put_cpu_var(last_ino
);
855 EXPORT_SYMBOL(get_next_ino
);
858 * new_inode_pseudo - obtain an inode
861 * Allocates a new inode for given superblock.
862 * Inode wont be chained in superblock s_inodes list
864 * - fs can't be unmount
865 * - quotas, fsnotify, writeback can't work
867 struct inode
*new_inode_pseudo(struct super_block
*sb
)
869 struct inode
*inode
= alloc_inode(sb
);
872 spin_lock(&inode
->i_lock
);
874 spin_unlock(&inode
->i_lock
);
875 INIT_LIST_HEAD(&inode
->i_sb_list
);
881 * new_inode - obtain an inode
884 * Allocates a new inode for given superblock. The default gfp_mask
885 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
886 * If HIGHMEM pages are unsuitable or it is known that pages allocated
887 * for the page cache are not reclaimable or migratable,
888 * mapping_set_gfp_mask() must be called with suitable flags on the
889 * newly created inode's mapping
892 struct inode
*new_inode(struct super_block
*sb
)
896 spin_lock_prefetch(&inode_sb_list_lock
);
898 inode
= new_inode_pseudo(sb
);
900 inode_sb_list_add(inode
);
903 EXPORT_SYMBOL(new_inode
);
905 #ifdef CONFIG_DEBUG_LOCK_ALLOC
906 void lockdep_annotate_inode_mutex_key(struct inode
*inode
)
908 if (S_ISDIR(inode
->i_mode
)) {
909 struct file_system_type
*type
= inode
->i_sb
->s_type
;
911 /* Set new key only if filesystem hasn't already changed it */
912 if (lockdep_match_class(&inode
->i_mutex
, &type
->i_mutex_key
)) {
914 * ensure nobody is actually holding i_mutex
916 mutex_destroy(&inode
->i_mutex
);
917 mutex_init(&inode
->i_mutex
);
918 lockdep_set_class(&inode
->i_mutex
,
919 &type
->i_mutex_dir_key
);
923 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key
);
927 * unlock_new_inode - clear the I_NEW state and wake up any waiters
928 * @inode: new inode to unlock
930 * Called when the inode is fully initialised to clear the new state of the
931 * inode and wake up anyone waiting for the inode to finish initialisation.
933 void unlock_new_inode(struct inode
*inode
)
935 lockdep_annotate_inode_mutex_key(inode
);
936 spin_lock(&inode
->i_lock
);
937 WARN_ON(!(inode
->i_state
& I_NEW
));
938 inode
->i_state
&= ~I_NEW
;
940 wake_up_bit(&inode
->i_state
, __I_NEW
);
941 spin_unlock(&inode
->i_lock
);
943 EXPORT_SYMBOL(unlock_new_inode
);
946 * iget5_locked - obtain an inode from a mounted file system
947 * @sb: super block of file system
948 * @hashval: hash value (usually inode number) to get
949 * @test: callback used for comparisons between inodes
950 * @set: callback used to initialize a new struct inode
951 * @data: opaque data pointer to pass to @test and @set
953 * Search for the inode specified by @hashval and @data in the inode cache,
954 * and if present it is return it with an increased reference count. This is
955 * a generalized version of iget_locked() for file systems where the inode
956 * number is not sufficient for unique identification of an inode.
958 * If the inode is not in cache, allocate a new inode and return it locked,
959 * hashed, and with the I_NEW flag set. The file system gets to fill it in
960 * before unlocking it via unlock_new_inode().
962 * Note both @test and @set are called with the inode_hash_lock held, so can't
965 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
966 int (*test
)(struct inode
*, void *),
967 int (*set
)(struct inode
*, void *), void *data
)
969 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
972 spin_lock(&inode_hash_lock
);
973 inode
= find_inode(sb
, head
, test
, data
);
974 spin_unlock(&inode_hash_lock
);
977 wait_on_inode(inode
);
981 inode
= alloc_inode(sb
);
985 spin_lock(&inode_hash_lock
);
986 /* We released the lock, so.. */
987 old
= find_inode(sb
, head
, test
, data
);
989 if (set(inode
, data
))
992 spin_lock(&inode
->i_lock
);
993 inode
->i_state
= I_NEW
;
994 hlist_add_head(&inode
->i_hash
, head
);
995 spin_unlock(&inode
->i_lock
);
996 inode_sb_list_add(inode
);
997 spin_unlock(&inode_hash_lock
);
999 /* Return the locked inode with I_NEW set, the
1000 * caller is responsible for filling in the contents
1006 * Uhhuh, somebody else created the same inode under
1007 * us. Use the old inode instead of the one we just
1010 spin_unlock(&inode_hash_lock
);
1011 destroy_inode(inode
);
1013 wait_on_inode(inode
);
1018 spin_unlock(&inode_hash_lock
);
1019 destroy_inode(inode
);
1022 EXPORT_SYMBOL(iget5_locked
);
1025 * iget_locked - obtain an inode from a mounted file system
1026 * @sb: super block of file system
1027 * @ino: inode number to get
1029 * Search for the inode specified by @ino in the inode cache and if present
1030 * return it with an increased reference count. This is for file systems
1031 * where the inode number is sufficient for unique identification of an inode.
1033 * If the inode is not in cache, allocate a new inode and return it locked,
1034 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1035 * before unlocking it via unlock_new_inode().
1037 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1039 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1040 struct inode
*inode
;
1042 spin_lock(&inode_hash_lock
);
1043 inode
= find_inode_fast(sb
, head
, ino
);
1044 spin_unlock(&inode_hash_lock
);
1046 wait_on_inode(inode
);
1050 inode
= alloc_inode(sb
);
1054 spin_lock(&inode_hash_lock
);
1055 /* We released the lock, so.. */
1056 old
= find_inode_fast(sb
, head
, ino
);
1059 spin_lock(&inode
->i_lock
);
1060 inode
->i_state
= I_NEW
;
1061 hlist_add_head(&inode
->i_hash
, head
);
1062 spin_unlock(&inode
->i_lock
);
1063 inode_sb_list_add(inode
);
1064 spin_unlock(&inode_hash_lock
);
1066 /* Return the locked inode with I_NEW set, the
1067 * caller is responsible for filling in the contents
1073 * Uhhuh, somebody else created the same inode under
1074 * us. Use the old inode instead of the one we just
1077 spin_unlock(&inode_hash_lock
);
1078 destroy_inode(inode
);
1080 wait_on_inode(inode
);
1084 EXPORT_SYMBOL(iget_locked
);
1087 * search the inode cache for a matching inode number.
1088 * If we find one, then the inode number we are trying to
1089 * allocate is not unique and so we should not use it.
1091 * Returns 1 if the inode number is unique, 0 if it is not.
1093 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1095 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1096 struct inode
*inode
;
1098 spin_lock(&inode_hash_lock
);
1099 hlist_for_each_entry(inode
, b
, i_hash
) {
1100 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
) {
1101 spin_unlock(&inode_hash_lock
);
1105 spin_unlock(&inode_hash_lock
);
1111 * iunique - get a unique inode number
1113 * @max_reserved: highest reserved inode number
1115 * Obtain an inode number that is unique on the system for a given
1116 * superblock. This is used by file systems that have no natural
1117 * permanent inode numbering system. An inode number is returned that
1118 * is higher than the reserved limit but unique.
1121 * With a large number of inodes live on the file system this function
1122 * currently becomes quite slow.
1124 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1127 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1128 * error if st_ino won't fit in target struct field. Use 32bit counter
1129 * here to attempt to avoid that.
1131 static DEFINE_SPINLOCK(iunique_lock
);
1132 static unsigned int counter
;
1135 spin_lock(&iunique_lock
);
1137 if (counter
<= max_reserved
)
1138 counter
= max_reserved
+ 1;
1140 } while (!test_inode_iunique(sb
, res
));
1141 spin_unlock(&iunique_lock
);
1145 EXPORT_SYMBOL(iunique
);
1147 struct inode
*igrab(struct inode
*inode
)
1149 spin_lock(&inode
->i_lock
);
1150 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1152 spin_unlock(&inode
->i_lock
);
1154 spin_unlock(&inode
->i_lock
);
1156 * Handle the case where s_op->clear_inode is not been
1157 * called yet, and somebody is calling igrab
1158 * while the inode is getting freed.
1164 EXPORT_SYMBOL(igrab
);
1167 * ilookup5_nowait - search for an inode in the inode cache
1168 * @sb: super block of file system to search
1169 * @hashval: hash value (usually inode number) to search for
1170 * @test: callback used for comparisons between inodes
1171 * @data: opaque data pointer to pass to @test
1173 * Search for the inode specified by @hashval and @data in the inode cache.
1174 * If the inode is in the cache, the inode is returned with an incremented
1177 * Note: I_NEW is not waited upon so you have to be very careful what you do
1178 * with the returned inode. You probably should be using ilookup5() instead.
1180 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1182 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1183 int (*test
)(struct inode
*, void *), void *data
)
1185 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1186 struct inode
*inode
;
1188 spin_lock(&inode_hash_lock
);
1189 inode
= find_inode(sb
, head
, test
, data
);
1190 spin_unlock(&inode_hash_lock
);
1194 EXPORT_SYMBOL(ilookup5_nowait
);
1197 * ilookup5 - search for an inode in the inode cache
1198 * @sb: super block of file system to search
1199 * @hashval: hash value (usually inode number) to search for
1200 * @test: callback used for comparisons between inodes
1201 * @data: opaque data pointer to pass to @test
1203 * Search for the inode specified by @hashval and @data in the inode cache,
1204 * and if the inode is in the cache, return the inode with an incremented
1205 * reference count. Waits on I_NEW before returning the inode.
1206 * returned with an incremented reference count.
1208 * This is a generalized version of ilookup() for file systems where the
1209 * inode number is not sufficient for unique identification of an inode.
1211 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1213 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1214 int (*test
)(struct inode
*, void *), void *data
)
1216 struct inode
*inode
= ilookup5_nowait(sb
, hashval
, test
, data
);
1219 wait_on_inode(inode
);
1222 EXPORT_SYMBOL(ilookup5
);
1225 * ilookup - search for an inode in the inode cache
1226 * @sb: super block of file system to search
1227 * @ino: inode number to search for
1229 * Search for the inode @ino in the inode cache, and if the inode is in the
1230 * cache, the inode is returned with an incremented reference count.
1232 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1234 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1235 struct inode
*inode
;
1237 spin_lock(&inode_hash_lock
);
1238 inode
= find_inode_fast(sb
, head
, ino
);
1239 spin_unlock(&inode_hash_lock
);
1242 wait_on_inode(inode
);
1245 EXPORT_SYMBOL(ilookup
);
1247 int insert_inode_locked(struct inode
*inode
)
1249 struct super_block
*sb
= inode
->i_sb
;
1250 ino_t ino
= inode
->i_ino
;
1251 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1254 struct inode
*old
= NULL
;
1255 spin_lock(&inode_hash_lock
);
1256 hlist_for_each_entry(old
, head
, i_hash
) {
1257 if (old
->i_ino
!= ino
)
1259 if (old
->i_sb
!= sb
)
1261 spin_lock(&old
->i_lock
);
1262 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1263 spin_unlock(&old
->i_lock
);
1269 spin_lock(&inode
->i_lock
);
1270 inode
->i_state
|= I_NEW
;
1271 hlist_add_head(&inode
->i_hash
, head
);
1272 spin_unlock(&inode
->i_lock
);
1273 spin_unlock(&inode_hash_lock
);
1277 spin_unlock(&old
->i_lock
);
1278 spin_unlock(&inode_hash_lock
);
1280 if (unlikely(!inode_unhashed(old
))) {
1287 EXPORT_SYMBOL(insert_inode_locked
);
1289 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1290 int (*test
)(struct inode
*, void *), void *data
)
1292 struct super_block
*sb
= inode
->i_sb
;
1293 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1296 struct inode
*old
= NULL
;
1298 spin_lock(&inode_hash_lock
);
1299 hlist_for_each_entry(old
, head
, i_hash
) {
1300 if (old
->i_sb
!= sb
)
1302 if (!test(old
, data
))
1304 spin_lock(&old
->i_lock
);
1305 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1306 spin_unlock(&old
->i_lock
);
1312 spin_lock(&inode
->i_lock
);
1313 inode
->i_state
|= I_NEW
;
1314 hlist_add_head(&inode
->i_hash
, head
);
1315 spin_unlock(&inode
->i_lock
);
1316 spin_unlock(&inode_hash_lock
);
1320 spin_unlock(&old
->i_lock
);
1321 spin_unlock(&inode_hash_lock
);
1323 if (unlikely(!inode_unhashed(old
))) {
1330 EXPORT_SYMBOL(insert_inode_locked4
);
1333 int generic_delete_inode(struct inode
*inode
)
1337 EXPORT_SYMBOL(generic_delete_inode
);
1340 * Called when we're dropping the last reference
1343 * Call the FS "drop_inode()" function, defaulting to
1344 * the legacy UNIX filesystem behaviour. If it tells
1345 * us to evict inode, do so. Otherwise, retain inode
1346 * in cache if fs is alive, sync and evict if fs is
1349 static void iput_final(struct inode
*inode
)
1351 struct super_block
*sb
= inode
->i_sb
;
1352 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1355 WARN_ON(inode
->i_state
& I_NEW
);
1358 drop
= op
->drop_inode(inode
);
1360 drop
= generic_drop_inode(inode
);
1362 if (!drop
&& (sb
->s_flags
& MS_ACTIVE
)) {
1363 inode
->i_state
|= I_REFERENCED
;
1364 inode_add_lru(inode
);
1365 spin_unlock(&inode
->i_lock
);
1370 inode
->i_state
|= I_WILL_FREE
;
1371 spin_unlock(&inode
->i_lock
);
1372 write_inode_now(inode
, 1);
1373 spin_lock(&inode
->i_lock
);
1374 WARN_ON(inode
->i_state
& I_NEW
);
1375 inode
->i_state
&= ~I_WILL_FREE
;
1378 inode
->i_state
|= I_FREEING
;
1379 if (!list_empty(&inode
->i_lru
))
1380 inode_lru_list_del(inode
);
1381 spin_unlock(&inode
->i_lock
);
1387 * iput - put an inode
1388 * @inode: inode to put
1390 * Puts an inode, dropping its usage count. If the inode use count hits
1391 * zero, the inode is then freed and may also be destroyed.
1393 * Consequently, iput() can sleep.
1395 void iput(struct inode
*inode
)
1398 BUG_ON(inode
->i_state
& I_CLEAR
);
1400 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
))
1404 EXPORT_SYMBOL(iput
);
1407 * bmap - find a block number in a file
1408 * @inode: inode of file
1409 * @block: block to find
1411 * Returns the block number on the device holding the inode that
1412 * is the disk block number for the block of the file requested.
1413 * That is, asked for block 4 of inode 1 the function will return the
1414 * disk block relative to the disk start that holds that block of the
1417 sector_t
bmap(struct inode
*inode
, sector_t block
)
1420 if (inode
->i_mapping
->a_ops
->bmap
)
1421 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1424 EXPORT_SYMBOL(bmap
);
1427 * With relative atime, only update atime if the previous atime is
1428 * earlier than either the ctime or mtime or if at least a day has
1429 * passed since the last atime update.
1431 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1432 struct timespec now
)
1435 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1438 * Is mtime younger than atime? If yes, update atime:
1440 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1443 * Is ctime younger than atime? If yes, update atime:
1445 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1449 * Is the previous atime value older than a day? If yes,
1452 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1455 * Good, we can skip the atime update:
1461 * This does the actual work of updating an inodes time or version. Must have
1462 * had called mnt_want_write() before calling this.
1464 static int update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1466 if (inode
->i_op
->update_time
)
1467 return inode
->i_op
->update_time(inode
, time
, flags
);
1469 if (flags
& S_ATIME
)
1470 inode
->i_atime
= *time
;
1471 if (flags
& S_VERSION
)
1472 inode_inc_iversion(inode
);
1473 if (flags
& S_CTIME
)
1474 inode
->i_ctime
= *time
;
1475 if (flags
& S_MTIME
)
1476 inode
->i_mtime
= *time
;
1477 mark_inode_dirty_sync(inode
);
1482 * touch_atime - update the access time
1483 * @path: the &struct path to update
1485 * Update the accessed time on an inode and mark it for writeback.
1486 * This function automatically handles read only file systems and media,
1487 * as well as the "noatime" flag and inode specific "noatime" markers.
1489 void touch_atime(const struct path
*path
)
1491 struct vfsmount
*mnt
= path
->mnt
;
1492 struct inode
*inode
= path
->dentry
->d_inode
;
1493 struct timespec now
;
1495 if (inode
->i_flags
& S_NOATIME
)
1497 if (IS_NOATIME(inode
))
1499 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1502 if (mnt
->mnt_flags
& MNT_NOATIME
)
1504 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1507 now
= current_fs_time(inode
->i_sb
);
1509 if (!relatime_need_update(mnt
, inode
, now
))
1512 if (timespec_equal(&inode
->i_atime
, &now
))
1515 if (!sb_start_write_trylock(inode
->i_sb
))
1518 if (__mnt_want_write(mnt
))
1521 * File systems can error out when updating inodes if they need to
1522 * allocate new space to modify an inode (such is the case for
1523 * Btrfs), but since we touch atime while walking down the path we
1524 * really don't care if we failed to update the atime of the file,
1525 * so just ignore the return value.
1526 * We may also fail on filesystems that have the ability to make parts
1527 * of the fs read only, e.g. subvolumes in Btrfs.
1529 update_time(inode
, &now
, S_ATIME
);
1530 __mnt_drop_write(mnt
);
1532 sb_end_write(inode
->i_sb
);
1534 EXPORT_SYMBOL(touch_atime
);
1537 * The logic we want is
1539 * if suid or (sgid and xgrp)
1542 int should_remove_suid(struct dentry
*dentry
)
1544 umode_t mode
= dentry
->d_inode
->i_mode
;
1547 /* suid always must be killed */
1548 if (unlikely(mode
& S_ISUID
))
1549 kill
= ATTR_KILL_SUID
;
1552 * sgid without any exec bits is just a mandatory locking mark; leave
1553 * it alone. If some exec bits are set, it's a real sgid; kill it.
1555 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1556 kill
|= ATTR_KILL_SGID
;
1558 if (unlikely(kill
&& !capable(CAP_FSETID
) && S_ISREG(mode
)))
1563 EXPORT_SYMBOL(should_remove_suid
);
1565 static int __remove_suid(struct dentry
*dentry
, int kill
)
1567 struct iattr newattrs
;
1569 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1570 return notify_change(dentry
, &newattrs
);
1573 int file_remove_suid(struct file
*file
)
1575 struct dentry
*dentry
= file
->f_path
.dentry
;
1576 struct inode
*inode
= dentry
->d_inode
;
1581 /* Fast path for nothing security related */
1582 if (IS_NOSEC(inode
))
1585 killsuid
= should_remove_suid(dentry
);
1586 killpriv
= security_inode_need_killpriv(dentry
);
1591 error
= security_inode_killpriv(dentry
);
1592 if (!error
&& killsuid
)
1593 error
= __remove_suid(dentry
, killsuid
);
1594 if (!error
&& (inode
->i_sb
->s_flags
& MS_NOSEC
))
1595 inode
->i_flags
|= S_NOSEC
;
1599 EXPORT_SYMBOL(file_remove_suid
);
1602 * file_update_time - update mtime and ctime time
1603 * @file: file accessed
1605 * Update the mtime and ctime members of an inode and mark the inode
1606 * for writeback. Note that this function is meant exclusively for
1607 * usage in the file write path of filesystems, and filesystems may
1608 * choose to explicitly ignore update via this function with the
1609 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1610 * timestamps are handled by the server. This can return an error for
1611 * file systems who need to allocate space in order to update an inode.
1614 int file_update_time(struct file
*file
)
1616 struct inode
*inode
= file_inode(file
);
1617 struct timespec now
;
1621 /* First try to exhaust all avenues to not sync */
1622 if (IS_NOCMTIME(inode
))
1625 now
= current_fs_time(inode
->i_sb
);
1626 if (!timespec_equal(&inode
->i_mtime
, &now
))
1629 if (!timespec_equal(&inode
->i_ctime
, &now
))
1632 if (IS_I_VERSION(inode
))
1633 sync_it
|= S_VERSION
;
1638 /* Finally allowed to write? Takes lock. */
1639 if (__mnt_want_write_file(file
))
1642 ret
= update_time(inode
, &now
, sync_it
);
1643 __mnt_drop_write_file(file
);
1647 EXPORT_SYMBOL(file_update_time
);
1649 int inode_needs_sync(struct inode
*inode
)
1653 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1657 EXPORT_SYMBOL(inode_needs_sync
);
1659 int inode_wait(void *word
)
1664 EXPORT_SYMBOL(inode_wait
);
1667 * If we try to find an inode in the inode hash while it is being
1668 * deleted, we have to wait until the filesystem completes its
1669 * deletion before reporting that it isn't found. This function waits
1670 * until the deletion _might_ have completed. Callers are responsible
1671 * to recheck inode state.
1673 * It doesn't matter if I_NEW is not set initially, a call to
1674 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1677 static void __wait_on_freeing_inode(struct inode
*inode
)
1679 wait_queue_head_t
*wq
;
1680 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1681 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1682 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1683 spin_unlock(&inode
->i_lock
);
1684 spin_unlock(&inode_hash_lock
);
1686 finish_wait(wq
, &wait
.wait
);
1687 spin_lock(&inode_hash_lock
);
1690 static __initdata
unsigned long ihash_entries
;
1691 static int __init
set_ihash_entries(char *str
)
1695 ihash_entries
= simple_strtoul(str
, &str
, 0);
1698 __setup("ihash_entries=", set_ihash_entries
);
1701 * Initialize the waitqueues and inode hash table.
1703 void __init
inode_init_early(void)
1707 /* If hashes are distributed across NUMA nodes, defer
1708 * hash allocation until vmalloc space is available.
1714 alloc_large_system_hash("Inode-cache",
1715 sizeof(struct hlist_head
),
1724 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1725 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1728 void __init
inode_init(void)
1732 /* inode slab cache */
1733 inode_cachep
= kmem_cache_create("inode_cache",
1734 sizeof(struct inode
),
1736 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1740 /* Hash may have been set up in inode_init_early */
1745 alloc_large_system_hash("Inode-cache",
1746 sizeof(struct hlist_head
),
1755 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1756 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1759 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1761 inode
->i_mode
= mode
;
1762 if (S_ISCHR(mode
)) {
1763 inode
->i_fop
= &def_chr_fops
;
1764 inode
->i_rdev
= rdev
;
1765 } else if (S_ISBLK(mode
)) {
1766 inode
->i_fop
= &def_blk_fops
;
1767 inode
->i_rdev
= rdev
;
1768 } else if (S_ISFIFO(mode
))
1769 inode
->i_fop
= &pipefifo_fops
;
1770 else if (S_ISSOCK(mode
))
1771 inode
->i_fop
= &bad_sock_fops
;
1773 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1774 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1777 EXPORT_SYMBOL(init_special_inode
);
1780 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1782 * @dir: Directory inode
1783 * @mode: mode of the new inode
1785 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1788 inode
->i_uid
= current_fsuid();
1789 if (dir
&& dir
->i_mode
& S_ISGID
) {
1790 inode
->i_gid
= dir
->i_gid
;
1794 inode
->i_gid
= current_fsgid();
1795 inode
->i_mode
= mode
;
1797 EXPORT_SYMBOL(inode_init_owner
);
1800 * inode_owner_or_capable - check current task permissions to inode
1801 * @inode: inode being checked
1803 * Return true if current either has CAP_FOWNER to the inode, or
1806 bool inode_owner_or_capable(const struct inode
*inode
)
1808 if (uid_eq(current_fsuid(), inode
->i_uid
))
1810 if (inode_capable(inode
, CAP_FOWNER
))
1814 EXPORT_SYMBOL(inode_owner_or_capable
);
1817 * Direct i/o helper functions
1819 static void __inode_dio_wait(struct inode
*inode
)
1821 wait_queue_head_t
*wq
= bit_waitqueue(&inode
->i_state
, __I_DIO_WAKEUP
);
1822 DEFINE_WAIT_BIT(q
, &inode
->i_state
, __I_DIO_WAKEUP
);
1825 prepare_to_wait(wq
, &q
.wait
, TASK_UNINTERRUPTIBLE
);
1826 if (atomic_read(&inode
->i_dio_count
))
1828 } while (atomic_read(&inode
->i_dio_count
));
1829 finish_wait(wq
, &q
.wait
);
1833 * inode_dio_wait - wait for outstanding DIO requests to finish
1834 * @inode: inode to wait for
1836 * Waits for all pending direct I/O requests to finish so that we can
1837 * proceed with a truncate or equivalent operation.
1839 * Must be called under a lock that serializes taking new references
1840 * to i_dio_count, usually by inode->i_mutex.
1842 void inode_dio_wait(struct inode
*inode
)
1844 if (atomic_read(&inode
->i_dio_count
))
1845 __inode_dio_wait(inode
);
1847 EXPORT_SYMBOL(inode_dio_wait
);
1850 * inode_dio_done - signal finish of a direct I/O requests
1851 * @inode: inode the direct I/O happens on
1853 * This is called once we've finished processing a direct I/O request,
1854 * and is used to wake up callers waiting for direct I/O to be quiesced.
1856 void inode_dio_done(struct inode
*inode
)
1858 if (atomic_dec_and_test(&inode
->i_dio_count
))
1859 wake_up_bit(&inode
->i_state
, __I_DIO_WAKEUP
);
1861 EXPORT_SYMBOL(inode_dio_done
);