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>
21 #include <trace/events/writeback.h>
25 * Inode locking rules:
27 * inode->i_lock protects:
28 * inode->i_state, inode->i_hash, __iget()
29 * Inode LRU list locks protect:
30 * inode->i_sb->s_inode_lru, inode->i_lru
31 * inode_sb_list_lock protects:
32 * sb->s_inodes, inode->i_sb_list
33 * bdi->wb.list_lock protects:
34 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_wb_list
35 * inode_hash_lock protects:
36 * inode_hashtable, inode->i_hash
42 * Inode LRU list locks
55 static unsigned int i_hash_mask __read_mostly
;
56 static unsigned int i_hash_shift __read_mostly
;
57 static struct hlist_head
*inode_hashtable __read_mostly
;
58 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_hash_lock
);
60 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_sb_list_lock
);
63 * Empty aops. Can be used for the cases where the user does not
64 * define any of the address_space operations.
66 const struct address_space_operations empty_aops
= {
68 EXPORT_SYMBOL(empty_aops
);
71 * Statistics gathering..
73 struct inodes_stat_t inodes_stat
;
75 static DEFINE_PER_CPU(unsigned long, nr_inodes
);
76 static DEFINE_PER_CPU(unsigned long, nr_unused
);
78 static struct kmem_cache
*inode_cachep __read_mostly
;
80 static long get_nr_inodes(void)
84 for_each_possible_cpu(i
)
85 sum
+= per_cpu(nr_inodes
, i
);
86 return sum
< 0 ? 0 : sum
;
89 static inline long get_nr_inodes_unused(void)
93 for_each_possible_cpu(i
)
94 sum
+= per_cpu(nr_unused
, i
);
95 return sum
< 0 ? 0 : sum
;
98 long get_nr_dirty_inodes(void)
100 /* not actually dirty inodes, but a wild approximation */
101 long nr_dirty
= get_nr_inodes() - get_nr_inodes_unused();
102 return nr_dirty
> 0 ? nr_dirty
: 0;
106 * Handle nr_inode sysctl
109 int proc_nr_inodes(struct ctl_table
*table
, int write
,
110 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
112 inodes_stat
.nr_inodes
= get_nr_inodes();
113 inodes_stat
.nr_unused
= get_nr_inodes_unused();
114 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
118 static int no_open(struct inode
*inode
, struct file
*file
)
124 * inode_init_always - perform inode structure intialisation
125 * @sb: superblock inode belongs to
126 * @inode: inode to initialise
128 * These are initializations that need to be done on every inode
129 * allocation as the fields are not initialised by slab allocation.
131 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
133 static const struct inode_operations empty_iops
;
134 static const struct file_operations no_open_fops
= {.open
= no_open
};
135 struct address_space
*const mapping
= &inode
->i_data
;
138 inode
->i_blkbits
= sb
->s_blocksize_bits
;
140 atomic_set(&inode
->i_count
, 1);
141 inode
->i_op
= &empty_iops
;
142 inode
->i_fop
= &no_open_fops
;
143 inode
->__i_nlink
= 1;
144 inode
->i_opflags
= 0;
145 i_uid_write(inode
, 0);
146 i_gid_write(inode
, 0);
147 atomic_set(&inode
->i_writecount
, 0);
151 inode
->i_generation
= 0;
152 inode
->i_pipe
= NULL
;
153 inode
->i_bdev
= NULL
;
154 inode
->i_cdev
= NULL
;
155 inode
->i_link
= NULL
;
157 inode
->dirtied_when
= 0;
159 if (security_inode_alloc(inode
))
161 spin_lock_init(&inode
->i_lock
);
162 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
164 mutex_init(&inode
->i_mutex
);
165 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
167 atomic_set(&inode
->i_dio_count
, 0);
169 mapping
->a_ops
= &empty_aops
;
170 mapping
->host
= inode
;
172 atomic_set(&mapping
->i_mmap_writable
, 0);
173 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
174 mapping
->private_data
= NULL
;
175 mapping
->writeback_index
= 0;
176 inode
->i_private
= NULL
;
177 inode
->i_mapping
= mapping
;
178 INIT_HLIST_HEAD(&inode
->i_dentry
); /* buggered by rcu freeing */
179 #ifdef CONFIG_FS_POSIX_ACL
180 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
183 #ifdef CONFIG_FSNOTIFY
184 inode
->i_fsnotify_mask
= 0;
186 inode
->i_flctx
= NULL
;
187 this_cpu_inc(nr_inodes
);
193 EXPORT_SYMBOL(inode_init_always
);
195 static struct inode
*alloc_inode(struct super_block
*sb
)
199 if (sb
->s_op
->alloc_inode
)
200 inode
= sb
->s_op
->alloc_inode(sb
);
202 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
207 if (unlikely(inode_init_always(sb
, inode
))) {
208 if (inode
->i_sb
->s_op
->destroy_inode
)
209 inode
->i_sb
->s_op
->destroy_inode(inode
);
211 kmem_cache_free(inode_cachep
, inode
);
218 void free_inode_nonrcu(struct inode
*inode
)
220 kmem_cache_free(inode_cachep
, inode
);
222 EXPORT_SYMBOL(free_inode_nonrcu
);
224 void __destroy_inode(struct inode
*inode
)
226 BUG_ON(inode_has_buffers(inode
));
227 security_inode_free(inode
);
228 fsnotify_inode_delete(inode
);
229 locks_free_lock_context(inode
->i_flctx
);
230 if (!inode
->i_nlink
) {
231 WARN_ON(atomic_long_read(&inode
->i_sb
->s_remove_count
) == 0);
232 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
235 #ifdef CONFIG_FS_POSIX_ACL
236 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
237 posix_acl_release(inode
->i_acl
);
238 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
239 posix_acl_release(inode
->i_default_acl
);
241 this_cpu_dec(nr_inodes
);
243 EXPORT_SYMBOL(__destroy_inode
);
245 static void i_callback(struct rcu_head
*head
)
247 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
248 kmem_cache_free(inode_cachep
, inode
);
251 static void destroy_inode(struct inode
*inode
)
253 BUG_ON(!list_empty(&inode
->i_lru
));
254 __destroy_inode(inode
);
255 if (inode
->i_sb
->s_op
->destroy_inode
)
256 inode
->i_sb
->s_op
->destroy_inode(inode
);
258 call_rcu(&inode
->i_rcu
, i_callback
);
262 * drop_nlink - directly drop an inode's link count
265 * This is a low-level filesystem helper to replace any
266 * direct filesystem manipulation of i_nlink. In cases
267 * where we are attempting to track writes to the
268 * filesystem, a decrement to zero means an imminent
269 * write when the file is truncated and actually unlinked
272 void drop_nlink(struct inode
*inode
)
274 WARN_ON(inode
->i_nlink
== 0);
277 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
279 EXPORT_SYMBOL(drop_nlink
);
282 * clear_nlink - directly zero an inode's link count
285 * This is a low-level filesystem helper to replace any
286 * direct filesystem manipulation of i_nlink. See
287 * drop_nlink() for why we care about i_nlink hitting zero.
289 void clear_nlink(struct inode
*inode
)
291 if (inode
->i_nlink
) {
292 inode
->__i_nlink
= 0;
293 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
296 EXPORT_SYMBOL(clear_nlink
);
299 * set_nlink - directly set an inode's link count
301 * @nlink: new nlink (should be non-zero)
303 * This is a low-level filesystem helper to replace any
304 * direct filesystem manipulation of i_nlink.
306 void set_nlink(struct inode
*inode
, unsigned int nlink
)
311 /* Yes, some filesystems do change nlink from zero to one */
312 if (inode
->i_nlink
== 0)
313 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
315 inode
->__i_nlink
= nlink
;
318 EXPORT_SYMBOL(set_nlink
);
321 * inc_nlink - directly increment an inode's link count
324 * This is a low-level filesystem helper to replace any
325 * direct filesystem manipulation of i_nlink. Currently,
326 * it is only here for parity with dec_nlink().
328 void inc_nlink(struct inode
*inode
)
330 if (unlikely(inode
->i_nlink
== 0)) {
331 WARN_ON(!(inode
->i_state
& I_LINKABLE
));
332 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
337 EXPORT_SYMBOL(inc_nlink
);
339 void address_space_init_once(struct address_space
*mapping
)
341 memset(mapping
, 0, sizeof(*mapping
));
342 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
343 spin_lock_init(&mapping
->tree_lock
);
344 init_rwsem(&mapping
->i_mmap_rwsem
);
345 INIT_LIST_HEAD(&mapping
->private_list
);
346 spin_lock_init(&mapping
->private_lock
);
347 mapping
->i_mmap
= RB_ROOT
;
349 EXPORT_SYMBOL(address_space_init_once
);
352 * These are initializations that only need to be done
353 * once, because the fields are idempotent across use
354 * of the inode, so let the slab aware of that.
356 void inode_init_once(struct inode
*inode
)
358 memset(inode
, 0, sizeof(*inode
));
359 INIT_HLIST_NODE(&inode
->i_hash
);
360 INIT_LIST_HEAD(&inode
->i_devices
);
361 INIT_LIST_HEAD(&inode
->i_wb_list
);
362 INIT_LIST_HEAD(&inode
->i_lru
);
363 address_space_init_once(&inode
->i_data
);
364 i_size_ordered_init(inode
);
365 #ifdef CONFIG_FSNOTIFY
366 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
369 EXPORT_SYMBOL(inode_init_once
);
371 static void init_once(void *foo
)
373 struct inode
*inode
= (struct inode
*) foo
;
375 inode_init_once(inode
);
379 * inode->i_lock must be held
381 void __iget(struct inode
*inode
)
383 atomic_inc(&inode
->i_count
);
387 * get additional reference to inode; caller must already hold one.
389 void ihold(struct inode
*inode
)
391 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
393 EXPORT_SYMBOL(ihold
);
395 static void inode_lru_list_add(struct inode
*inode
)
397 if (list_lru_add(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
398 this_cpu_inc(nr_unused
);
402 * Add inode to LRU if needed (inode is unused and clean).
404 * Needs inode->i_lock held.
406 void inode_add_lru(struct inode
*inode
)
408 if (!(inode
->i_state
& (I_DIRTY_ALL
| I_SYNC
|
409 I_FREEING
| I_WILL_FREE
)) &&
410 !atomic_read(&inode
->i_count
) && inode
->i_sb
->s_flags
& MS_ACTIVE
)
411 inode_lru_list_add(inode
);
415 static void inode_lru_list_del(struct inode
*inode
)
418 if (list_lru_del(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
419 this_cpu_dec(nr_unused
);
423 * inode_sb_list_add - add inode to the superblock list of inodes
424 * @inode: inode to add
426 void inode_sb_list_add(struct inode
*inode
)
428 spin_lock(&inode_sb_list_lock
);
429 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
430 spin_unlock(&inode_sb_list_lock
);
432 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
434 static inline void inode_sb_list_del(struct inode
*inode
)
436 if (!list_empty(&inode
->i_sb_list
)) {
437 spin_lock(&inode_sb_list_lock
);
438 list_del_init(&inode
->i_sb_list
);
439 spin_unlock(&inode_sb_list_lock
);
443 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
447 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
449 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> i_hash_shift
);
450 return tmp
& i_hash_mask
;
454 * __insert_inode_hash - hash an inode
455 * @inode: unhashed inode
456 * @hashval: unsigned long value used to locate this object in the
459 * Add an inode to the inode hash for this superblock.
461 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
463 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
465 spin_lock(&inode_hash_lock
);
466 spin_lock(&inode
->i_lock
);
467 hlist_add_head(&inode
->i_hash
, b
);
468 spin_unlock(&inode
->i_lock
);
469 spin_unlock(&inode_hash_lock
);
471 EXPORT_SYMBOL(__insert_inode_hash
);
474 * __remove_inode_hash - remove an inode from the hash
475 * @inode: inode to unhash
477 * Remove an inode from the superblock.
479 void __remove_inode_hash(struct inode
*inode
)
481 spin_lock(&inode_hash_lock
);
482 spin_lock(&inode
->i_lock
);
483 hlist_del_init(&inode
->i_hash
);
484 spin_unlock(&inode
->i_lock
);
485 spin_unlock(&inode_hash_lock
);
487 EXPORT_SYMBOL(__remove_inode_hash
);
489 void clear_inode(struct inode
*inode
)
493 * We have to cycle tree_lock here because reclaim can be still in the
494 * process of removing the last page (in __delete_from_page_cache())
495 * and we must not free mapping under it.
497 spin_lock_irq(&inode
->i_data
.tree_lock
);
498 BUG_ON(inode
->i_data
.nrpages
);
499 BUG_ON(inode
->i_data
.nrshadows
);
500 spin_unlock_irq(&inode
->i_data
.tree_lock
);
501 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
502 BUG_ON(!(inode
->i_state
& I_FREEING
));
503 BUG_ON(inode
->i_state
& I_CLEAR
);
504 /* don't need i_lock here, no concurrent mods to i_state */
505 inode
->i_state
= I_FREEING
| I_CLEAR
;
507 EXPORT_SYMBOL(clear_inode
);
510 * Free the inode passed in, removing it from the lists it is still connected
511 * to. We remove any pages still attached to the inode and wait for any IO that
512 * is still in progress before finally destroying the inode.
514 * An inode must already be marked I_FREEING so that we avoid the inode being
515 * moved back onto lists if we race with other code that manipulates the lists
516 * (e.g. writeback_single_inode). The caller is responsible for setting this.
518 * An inode must already be removed from the LRU list before being evicted from
519 * the cache. This should occur atomically with setting the I_FREEING state
520 * flag, so no inodes here should ever be on the LRU when being evicted.
522 static void evict(struct inode
*inode
)
524 const struct super_operations
*op
= inode
->i_sb
->s_op
;
526 BUG_ON(!(inode
->i_state
& I_FREEING
));
527 BUG_ON(!list_empty(&inode
->i_lru
));
529 if (!list_empty(&inode
->i_wb_list
))
530 inode_wb_list_del(inode
);
532 inode_sb_list_del(inode
);
535 * Wait for flusher thread to be done with the inode so that filesystem
536 * does not start destroying it while writeback is still running. Since
537 * the inode has I_FREEING set, flusher thread won't start new work on
538 * the inode. We just have to wait for running writeback to finish.
540 inode_wait_for_writeback(inode
);
542 if (op
->evict_inode
) {
543 op
->evict_inode(inode
);
545 truncate_inode_pages_final(&inode
->i_data
);
548 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
550 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
553 remove_inode_hash(inode
);
555 spin_lock(&inode
->i_lock
);
556 wake_up_bit(&inode
->i_state
, __I_NEW
);
557 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
558 spin_unlock(&inode
->i_lock
);
560 destroy_inode(inode
);
564 * dispose_list - dispose of the contents of a local list
565 * @head: the head of the list to free
567 * Dispose-list gets a local list with local inodes in it, so it doesn't
568 * need to worry about list corruption and SMP locks.
570 static void dispose_list(struct list_head
*head
)
572 while (!list_empty(head
)) {
575 inode
= list_first_entry(head
, struct inode
, i_lru
);
576 list_del_init(&inode
->i_lru
);
583 * evict_inodes - evict all evictable inodes for a superblock
584 * @sb: superblock to operate on
586 * Make sure that no inodes with zero refcount are retained. This is
587 * called by superblock shutdown after having MS_ACTIVE flag removed,
588 * so any inode reaching zero refcount during or after that call will
589 * be immediately evicted.
591 void evict_inodes(struct super_block
*sb
)
593 struct inode
*inode
, *next
;
596 spin_lock(&inode_sb_list_lock
);
597 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
598 if (atomic_read(&inode
->i_count
))
601 spin_lock(&inode
->i_lock
);
602 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
603 spin_unlock(&inode
->i_lock
);
607 inode
->i_state
|= I_FREEING
;
608 inode_lru_list_del(inode
);
609 spin_unlock(&inode
->i_lock
);
610 list_add(&inode
->i_lru
, &dispose
);
612 spin_unlock(&inode_sb_list_lock
);
614 dispose_list(&dispose
);
618 * invalidate_inodes - attempt to free all inodes on a superblock
619 * @sb: superblock to operate on
620 * @kill_dirty: flag to guide handling of dirty inodes
622 * Attempts to free all inodes for a given superblock. If there were any
623 * busy inodes return a non-zero value, else zero.
624 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
627 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
630 struct inode
*inode
, *next
;
633 spin_lock(&inode_sb_list_lock
);
634 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
635 spin_lock(&inode
->i_lock
);
636 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
637 spin_unlock(&inode
->i_lock
);
640 if (inode
->i_state
& I_DIRTY_ALL
&& !kill_dirty
) {
641 spin_unlock(&inode
->i_lock
);
645 if (atomic_read(&inode
->i_count
)) {
646 spin_unlock(&inode
->i_lock
);
651 inode
->i_state
|= I_FREEING
;
652 inode_lru_list_del(inode
);
653 spin_unlock(&inode
->i_lock
);
654 list_add(&inode
->i_lru
, &dispose
);
656 spin_unlock(&inode_sb_list_lock
);
658 dispose_list(&dispose
);
664 * Isolate the inode from the LRU in preparation for freeing it.
666 * Any inodes which are pinned purely because of attached pagecache have their
667 * pagecache removed. If the inode has metadata buffers attached to
668 * mapping->private_list then try to remove them.
670 * If the inode has the I_REFERENCED flag set, then it means that it has been
671 * used recently - the flag is set in iput_final(). When we encounter such an
672 * inode, clear the flag and move it to the back of the LRU so it gets another
673 * pass through the LRU before it gets reclaimed. This is necessary because of
674 * the fact we are doing lazy LRU updates to minimise lock contention so the
675 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
676 * with this flag set because they are the inodes that are out of order.
678 static enum lru_status
inode_lru_isolate(struct list_head
*item
,
679 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
681 struct list_head
*freeable
= arg
;
682 struct inode
*inode
= container_of(item
, struct inode
, i_lru
);
685 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
686 * If we fail to get the lock, just skip it.
688 if (!spin_trylock(&inode
->i_lock
))
692 * Referenced or dirty inodes are still in use. Give them another pass
693 * through the LRU as we canot reclaim them now.
695 if (atomic_read(&inode
->i_count
) ||
696 (inode
->i_state
& ~I_REFERENCED
)) {
697 list_lru_isolate(lru
, &inode
->i_lru
);
698 spin_unlock(&inode
->i_lock
);
699 this_cpu_dec(nr_unused
);
703 /* recently referenced inodes get one more pass */
704 if (inode
->i_state
& I_REFERENCED
) {
705 inode
->i_state
&= ~I_REFERENCED
;
706 spin_unlock(&inode
->i_lock
);
710 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
712 spin_unlock(&inode
->i_lock
);
713 spin_unlock(lru_lock
);
714 if (remove_inode_buffers(inode
)) {
716 reap
= invalidate_mapping_pages(&inode
->i_data
, 0, -1);
717 if (current_is_kswapd())
718 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
720 __count_vm_events(PGINODESTEAL
, reap
);
721 if (current
->reclaim_state
)
722 current
->reclaim_state
->reclaimed_slab
+= reap
;
729 WARN_ON(inode
->i_state
& I_NEW
);
730 inode
->i_state
|= I_FREEING
;
731 list_lru_isolate_move(lru
, &inode
->i_lru
, freeable
);
732 spin_unlock(&inode
->i_lock
);
734 this_cpu_dec(nr_unused
);
739 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
740 * This is called from the superblock shrinker function with a number of inodes
741 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
742 * then are freed outside inode_lock by dispose_list().
744 long prune_icache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
749 freed
= list_lru_shrink_walk(&sb
->s_inode_lru
, sc
,
750 inode_lru_isolate
, &freeable
);
751 dispose_list(&freeable
);
755 static void __wait_on_freeing_inode(struct inode
*inode
);
757 * Called with the inode lock held.
759 static struct inode
*find_inode(struct super_block
*sb
,
760 struct hlist_head
*head
,
761 int (*test
)(struct inode
*, void *),
764 struct inode
*inode
= NULL
;
767 hlist_for_each_entry(inode
, head
, i_hash
) {
768 if (inode
->i_sb
!= sb
)
770 if (!test(inode
, data
))
772 spin_lock(&inode
->i_lock
);
773 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
774 __wait_on_freeing_inode(inode
);
778 spin_unlock(&inode
->i_lock
);
785 * find_inode_fast is the fast path version of find_inode, see the comment at
786 * iget_locked for details.
788 static struct inode
*find_inode_fast(struct super_block
*sb
,
789 struct hlist_head
*head
, unsigned long ino
)
791 struct inode
*inode
= NULL
;
794 hlist_for_each_entry(inode
, head
, i_hash
) {
795 if (inode
->i_ino
!= ino
)
797 if (inode
->i_sb
!= sb
)
799 spin_lock(&inode
->i_lock
);
800 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
801 __wait_on_freeing_inode(inode
);
805 spin_unlock(&inode
->i_lock
);
812 * Each cpu owns a range of LAST_INO_BATCH numbers.
813 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
814 * to renew the exhausted range.
816 * This does not significantly increase overflow rate because every CPU can
817 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
818 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
819 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
820 * overflow rate by 2x, which does not seem too significant.
822 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
823 * error if st_ino won't fit in target struct field. Use 32bit counter
824 * here to attempt to avoid that.
826 #define LAST_INO_BATCH 1024
827 static DEFINE_PER_CPU(unsigned int, last_ino
);
829 unsigned int get_next_ino(void)
831 unsigned int *p
= &get_cpu_var(last_ino
);
832 unsigned int res
= *p
;
835 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
836 static atomic_t shared_last_ino
;
837 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
839 res
= next
- LAST_INO_BATCH
;
844 /* get_next_ino should not provide a 0 inode number */
848 put_cpu_var(last_ino
);
851 EXPORT_SYMBOL(get_next_ino
);
854 * new_inode_pseudo - obtain an inode
857 * Allocates a new inode for given superblock.
858 * Inode wont be chained in superblock s_inodes list
860 * - fs can't be unmount
861 * - quotas, fsnotify, writeback can't work
863 struct inode
*new_inode_pseudo(struct super_block
*sb
)
865 struct inode
*inode
= alloc_inode(sb
);
868 spin_lock(&inode
->i_lock
);
870 spin_unlock(&inode
->i_lock
);
871 INIT_LIST_HEAD(&inode
->i_sb_list
);
877 * new_inode - obtain an inode
880 * Allocates a new inode for given superblock. The default gfp_mask
881 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
882 * If HIGHMEM pages are unsuitable or it is known that pages allocated
883 * for the page cache are not reclaimable or migratable,
884 * mapping_set_gfp_mask() must be called with suitable flags on the
885 * newly created inode's mapping
888 struct inode
*new_inode(struct super_block
*sb
)
892 spin_lock_prefetch(&inode_sb_list_lock
);
894 inode
= new_inode_pseudo(sb
);
896 inode_sb_list_add(inode
);
899 EXPORT_SYMBOL(new_inode
);
901 #ifdef CONFIG_DEBUG_LOCK_ALLOC
902 void lockdep_annotate_inode_mutex_key(struct inode
*inode
)
904 if (S_ISDIR(inode
->i_mode
)) {
905 struct file_system_type
*type
= inode
->i_sb
->s_type
;
907 /* Set new key only if filesystem hasn't already changed it */
908 if (lockdep_match_class(&inode
->i_mutex
, &type
->i_mutex_key
)) {
910 * ensure nobody is actually holding i_mutex
912 mutex_destroy(&inode
->i_mutex
);
913 mutex_init(&inode
->i_mutex
);
914 lockdep_set_class(&inode
->i_mutex
,
915 &type
->i_mutex_dir_key
);
919 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key
);
923 * unlock_new_inode - clear the I_NEW state and wake up any waiters
924 * @inode: new inode to unlock
926 * Called when the inode is fully initialised to clear the new state of the
927 * inode and wake up anyone waiting for the inode to finish initialisation.
929 void unlock_new_inode(struct inode
*inode
)
931 lockdep_annotate_inode_mutex_key(inode
);
932 spin_lock(&inode
->i_lock
);
933 WARN_ON(!(inode
->i_state
& I_NEW
));
934 inode
->i_state
&= ~I_NEW
;
936 wake_up_bit(&inode
->i_state
, __I_NEW
);
937 spin_unlock(&inode
->i_lock
);
939 EXPORT_SYMBOL(unlock_new_inode
);
942 * lock_two_nondirectories - take two i_mutexes on non-directory objects
944 * Lock any non-NULL argument that is not a directory.
945 * Zero, one or two objects may be locked by this function.
947 * @inode1: first inode to lock
948 * @inode2: second inode to lock
950 void lock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
953 swap(inode1
, inode2
);
955 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
956 mutex_lock(&inode1
->i_mutex
);
957 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
958 mutex_lock_nested(&inode2
->i_mutex
, I_MUTEX_NONDIR2
);
960 EXPORT_SYMBOL(lock_two_nondirectories
);
963 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
964 * @inode1: first inode to unlock
965 * @inode2: second inode to unlock
967 void unlock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
969 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
970 mutex_unlock(&inode1
->i_mutex
);
971 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
972 mutex_unlock(&inode2
->i_mutex
);
974 EXPORT_SYMBOL(unlock_two_nondirectories
);
977 * iget5_locked - obtain an inode from a mounted file system
978 * @sb: super block of file system
979 * @hashval: hash value (usually inode number) to get
980 * @test: callback used for comparisons between inodes
981 * @set: callback used to initialize a new struct inode
982 * @data: opaque data pointer to pass to @test and @set
984 * Search for the inode specified by @hashval and @data in the inode cache,
985 * and if present it is return it with an increased reference count. This is
986 * a generalized version of iget_locked() for file systems where the inode
987 * number is not sufficient for unique identification of an inode.
989 * If the inode is not in cache, allocate a new inode and return it locked,
990 * hashed, and with the I_NEW flag set. The file system gets to fill it in
991 * before unlocking it via unlock_new_inode().
993 * Note both @test and @set are called with the inode_hash_lock held, so can't
996 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
997 int (*test
)(struct inode
*, void *),
998 int (*set
)(struct inode
*, void *), void *data
)
1000 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1001 struct inode
*inode
;
1003 spin_lock(&inode_hash_lock
);
1004 inode
= find_inode(sb
, head
, test
, data
);
1005 spin_unlock(&inode_hash_lock
);
1008 wait_on_inode(inode
);
1012 inode
= alloc_inode(sb
);
1016 spin_lock(&inode_hash_lock
);
1017 /* We released the lock, so.. */
1018 old
= find_inode(sb
, head
, test
, data
);
1020 if (set(inode
, data
))
1023 spin_lock(&inode
->i_lock
);
1024 inode
->i_state
= I_NEW
;
1025 hlist_add_head(&inode
->i_hash
, head
);
1026 spin_unlock(&inode
->i_lock
);
1027 inode_sb_list_add(inode
);
1028 spin_unlock(&inode_hash_lock
);
1030 /* Return the locked inode with I_NEW set, the
1031 * caller is responsible for filling in the contents
1037 * Uhhuh, somebody else created the same inode under
1038 * us. Use the old inode instead of the one we just
1041 spin_unlock(&inode_hash_lock
);
1042 destroy_inode(inode
);
1044 wait_on_inode(inode
);
1049 spin_unlock(&inode_hash_lock
);
1050 destroy_inode(inode
);
1053 EXPORT_SYMBOL(iget5_locked
);
1056 * iget_locked - obtain an inode from a mounted file system
1057 * @sb: super block of file system
1058 * @ino: inode number to get
1060 * Search for the inode specified by @ino in the inode cache and if present
1061 * return it with an increased reference count. This is for file systems
1062 * where the inode number is sufficient for unique identification of an inode.
1064 * If the inode is not in cache, allocate a new inode and return it locked,
1065 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1066 * before unlocking it via unlock_new_inode().
1068 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1070 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1071 struct inode
*inode
;
1073 spin_lock(&inode_hash_lock
);
1074 inode
= find_inode_fast(sb
, head
, ino
);
1075 spin_unlock(&inode_hash_lock
);
1077 wait_on_inode(inode
);
1081 inode
= alloc_inode(sb
);
1085 spin_lock(&inode_hash_lock
);
1086 /* We released the lock, so.. */
1087 old
= find_inode_fast(sb
, head
, ino
);
1090 spin_lock(&inode
->i_lock
);
1091 inode
->i_state
= I_NEW
;
1092 hlist_add_head(&inode
->i_hash
, head
);
1093 spin_unlock(&inode
->i_lock
);
1094 inode_sb_list_add(inode
);
1095 spin_unlock(&inode_hash_lock
);
1097 /* Return the locked inode with I_NEW set, the
1098 * caller is responsible for filling in the contents
1104 * Uhhuh, somebody else created the same inode under
1105 * us. Use the old inode instead of the one we just
1108 spin_unlock(&inode_hash_lock
);
1109 destroy_inode(inode
);
1111 wait_on_inode(inode
);
1115 EXPORT_SYMBOL(iget_locked
);
1118 * search the inode cache for a matching inode number.
1119 * If we find one, then the inode number we are trying to
1120 * allocate is not unique and so we should not use it.
1122 * Returns 1 if the inode number is unique, 0 if it is not.
1124 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1126 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1127 struct inode
*inode
;
1129 spin_lock(&inode_hash_lock
);
1130 hlist_for_each_entry(inode
, b
, i_hash
) {
1131 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
) {
1132 spin_unlock(&inode_hash_lock
);
1136 spin_unlock(&inode_hash_lock
);
1142 * iunique - get a unique inode number
1144 * @max_reserved: highest reserved inode number
1146 * Obtain an inode number that is unique on the system for a given
1147 * superblock. This is used by file systems that have no natural
1148 * permanent inode numbering system. An inode number is returned that
1149 * is higher than the reserved limit but unique.
1152 * With a large number of inodes live on the file system this function
1153 * currently becomes quite slow.
1155 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1158 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1159 * error if st_ino won't fit in target struct field. Use 32bit counter
1160 * here to attempt to avoid that.
1162 static DEFINE_SPINLOCK(iunique_lock
);
1163 static unsigned int counter
;
1166 spin_lock(&iunique_lock
);
1168 if (counter
<= max_reserved
)
1169 counter
= max_reserved
+ 1;
1171 } while (!test_inode_iunique(sb
, res
));
1172 spin_unlock(&iunique_lock
);
1176 EXPORT_SYMBOL(iunique
);
1178 struct inode
*igrab(struct inode
*inode
)
1180 spin_lock(&inode
->i_lock
);
1181 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1183 spin_unlock(&inode
->i_lock
);
1185 spin_unlock(&inode
->i_lock
);
1187 * Handle the case where s_op->clear_inode is not been
1188 * called yet, and somebody is calling igrab
1189 * while the inode is getting freed.
1195 EXPORT_SYMBOL(igrab
);
1198 * ilookup5_nowait - search for an inode in the inode cache
1199 * @sb: super block of file system to search
1200 * @hashval: hash value (usually inode number) to search for
1201 * @test: callback used for comparisons between inodes
1202 * @data: opaque data pointer to pass to @test
1204 * Search for the inode specified by @hashval and @data in the inode cache.
1205 * If the inode is in the cache, the inode is returned with an incremented
1208 * Note: I_NEW is not waited upon so you have to be very careful what you do
1209 * with the returned inode. You probably should be using ilookup5() instead.
1211 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1213 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1214 int (*test
)(struct inode
*, void *), void *data
)
1216 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1217 struct inode
*inode
;
1219 spin_lock(&inode_hash_lock
);
1220 inode
= find_inode(sb
, head
, test
, data
);
1221 spin_unlock(&inode_hash_lock
);
1225 EXPORT_SYMBOL(ilookup5_nowait
);
1228 * ilookup5 - search for an inode in the inode cache
1229 * @sb: super block of file system to search
1230 * @hashval: hash value (usually inode number) to search for
1231 * @test: callback used for comparisons between inodes
1232 * @data: opaque data pointer to pass to @test
1234 * Search for the inode specified by @hashval and @data in the inode cache,
1235 * and if the inode is in the cache, return the inode with an incremented
1236 * reference count. Waits on I_NEW before returning the inode.
1237 * returned with an incremented reference count.
1239 * This is a generalized version of ilookup() for file systems where the
1240 * inode number is not sufficient for unique identification of an inode.
1242 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1244 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1245 int (*test
)(struct inode
*, void *), void *data
)
1247 struct inode
*inode
= ilookup5_nowait(sb
, hashval
, test
, data
);
1250 wait_on_inode(inode
);
1253 EXPORT_SYMBOL(ilookup5
);
1256 * ilookup - search for an inode in the inode cache
1257 * @sb: super block of file system to search
1258 * @ino: inode number to search for
1260 * Search for the inode @ino in the inode cache, and if the inode is in the
1261 * cache, the inode is returned with an incremented reference count.
1263 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1265 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1266 struct inode
*inode
;
1268 spin_lock(&inode_hash_lock
);
1269 inode
= find_inode_fast(sb
, head
, ino
);
1270 spin_unlock(&inode_hash_lock
);
1273 wait_on_inode(inode
);
1276 EXPORT_SYMBOL(ilookup
);
1279 * find_inode_nowait - find an inode in the inode cache
1280 * @sb: super block of file system to search
1281 * @hashval: hash value (usually inode number) to search for
1282 * @match: callback used for comparisons between inodes
1283 * @data: opaque data pointer to pass to @match
1285 * Search for the inode specified by @hashval and @data in the inode
1286 * cache, where the helper function @match will return 0 if the inode
1287 * does not match, 1 if the inode does match, and -1 if the search
1288 * should be stopped. The @match function must be responsible for
1289 * taking the i_lock spin_lock and checking i_state for an inode being
1290 * freed or being initialized, and incrementing the reference count
1291 * before returning 1. It also must not sleep, since it is called with
1292 * the inode_hash_lock spinlock held.
1294 * This is a even more generalized version of ilookup5() when the
1295 * function must never block --- find_inode() can block in
1296 * __wait_on_freeing_inode() --- or when the caller can not increment
1297 * the reference count because the resulting iput() might cause an
1298 * inode eviction. The tradeoff is that the @match funtion must be
1299 * very carefully implemented.
1301 struct inode
*find_inode_nowait(struct super_block
*sb
,
1302 unsigned long hashval
,
1303 int (*match
)(struct inode
*, unsigned long,
1307 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1308 struct inode
*inode
, *ret_inode
= NULL
;
1311 spin_lock(&inode_hash_lock
);
1312 hlist_for_each_entry(inode
, head
, i_hash
) {
1313 if (inode
->i_sb
!= sb
)
1315 mval
= match(inode
, hashval
, data
);
1323 spin_unlock(&inode_hash_lock
);
1326 EXPORT_SYMBOL(find_inode_nowait
);
1328 int insert_inode_locked(struct inode
*inode
)
1330 struct super_block
*sb
= inode
->i_sb
;
1331 ino_t ino
= inode
->i_ino
;
1332 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1335 struct inode
*old
= NULL
;
1336 spin_lock(&inode_hash_lock
);
1337 hlist_for_each_entry(old
, head
, i_hash
) {
1338 if (old
->i_ino
!= ino
)
1340 if (old
->i_sb
!= sb
)
1342 spin_lock(&old
->i_lock
);
1343 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1344 spin_unlock(&old
->i_lock
);
1350 spin_lock(&inode
->i_lock
);
1351 inode
->i_state
|= I_NEW
;
1352 hlist_add_head(&inode
->i_hash
, head
);
1353 spin_unlock(&inode
->i_lock
);
1354 spin_unlock(&inode_hash_lock
);
1358 spin_unlock(&old
->i_lock
);
1359 spin_unlock(&inode_hash_lock
);
1361 if (unlikely(!inode_unhashed(old
))) {
1368 EXPORT_SYMBOL(insert_inode_locked
);
1370 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1371 int (*test
)(struct inode
*, void *), void *data
)
1373 struct super_block
*sb
= inode
->i_sb
;
1374 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1377 struct inode
*old
= NULL
;
1379 spin_lock(&inode_hash_lock
);
1380 hlist_for_each_entry(old
, head
, i_hash
) {
1381 if (old
->i_sb
!= sb
)
1383 if (!test(old
, data
))
1385 spin_lock(&old
->i_lock
);
1386 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1387 spin_unlock(&old
->i_lock
);
1393 spin_lock(&inode
->i_lock
);
1394 inode
->i_state
|= I_NEW
;
1395 hlist_add_head(&inode
->i_hash
, head
);
1396 spin_unlock(&inode
->i_lock
);
1397 spin_unlock(&inode_hash_lock
);
1401 spin_unlock(&old
->i_lock
);
1402 spin_unlock(&inode_hash_lock
);
1404 if (unlikely(!inode_unhashed(old
))) {
1411 EXPORT_SYMBOL(insert_inode_locked4
);
1414 int generic_delete_inode(struct inode
*inode
)
1418 EXPORT_SYMBOL(generic_delete_inode
);
1421 * Called when we're dropping the last reference
1424 * Call the FS "drop_inode()" function, defaulting to
1425 * the legacy UNIX filesystem behaviour. If it tells
1426 * us to evict inode, do so. Otherwise, retain inode
1427 * in cache if fs is alive, sync and evict if fs is
1430 static void iput_final(struct inode
*inode
)
1432 struct super_block
*sb
= inode
->i_sb
;
1433 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1436 WARN_ON(inode
->i_state
& I_NEW
);
1439 drop
= op
->drop_inode(inode
);
1441 drop
= generic_drop_inode(inode
);
1443 if (!drop
&& (sb
->s_flags
& MS_ACTIVE
)) {
1444 inode
->i_state
|= I_REFERENCED
;
1445 inode_add_lru(inode
);
1446 spin_unlock(&inode
->i_lock
);
1451 inode
->i_state
|= I_WILL_FREE
;
1452 spin_unlock(&inode
->i_lock
);
1453 write_inode_now(inode
, 1);
1454 spin_lock(&inode
->i_lock
);
1455 WARN_ON(inode
->i_state
& I_NEW
);
1456 inode
->i_state
&= ~I_WILL_FREE
;
1459 inode
->i_state
|= I_FREEING
;
1460 if (!list_empty(&inode
->i_lru
))
1461 inode_lru_list_del(inode
);
1462 spin_unlock(&inode
->i_lock
);
1468 * iput - put an inode
1469 * @inode: inode to put
1471 * Puts an inode, dropping its usage count. If the inode use count hits
1472 * zero, the inode is then freed and may also be destroyed.
1474 * Consequently, iput() can sleep.
1476 void iput(struct inode
*inode
)
1480 BUG_ON(inode
->i_state
& I_CLEAR
);
1482 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
)) {
1483 if (inode
->i_nlink
&& (inode
->i_state
& I_DIRTY_TIME
)) {
1484 atomic_inc(&inode
->i_count
);
1485 inode
->i_state
&= ~I_DIRTY_TIME
;
1486 spin_unlock(&inode
->i_lock
);
1487 trace_writeback_lazytime_iput(inode
);
1488 mark_inode_dirty_sync(inode
);
1494 EXPORT_SYMBOL(iput
);
1497 * bmap - find a block number in a file
1498 * @inode: inode of file
1499 * @block: block to find
1501 * Returns the block number on the device holding the inode that
1502 * is the disk block number for the block of the file requested.
1503 * That is, asked for block 4 of inode 1 the function will return the
1504 * disk block relative to the disk start that holds that block of the
1507 sector_t
bmap(struct inode
*inode
, sector_t block
)
1510 if (inode
->i_mapping
->a_ops
->bmap
)
1511 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1514 EXPORT_SYMBOL(bmap
);
1517 * With relative atime, only update atime if the previous atime is
1518 * earlier than either the ctime or mtime or if at least a day has
1519 * passed since the last atime update.
1521 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1522 struct timespec now
)
1525 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1528 * Is mtime younger than atime? If yes, update atime:
1530 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1533 * Is ctime younger than atime? If yes, update atime:
1535 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1539 * Is the previous atime value older than a day? If yes,
1542 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1545 * Good, we can skip the atime update:
1550 int generic_update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1552 int iflags
= I_DIRTY_TIME
;
1554 if (flags
& S_ATIME
)
1555 inode
->i_atime
= *time
;
1556 if (flags
& S_VERSION
)
1557 inode_inc_iversion(inode
);
1558 if (flags
& S_CTIME
)
1559 inode
->i_ctime
= *time
;
1560 if (flags
& S_MTIME
)
1561 inode
->i_mtime
= *time
;
1563 if (!(inode
->i_sb
->s_flags
& MS_LAZYTIME
) || (flags
& S_VERSION
))
1564 iflags
|= I_DIRTY_SYNC
;
1565 __mark_inode_dirty(inode
, iflags
);
1568 EXPORT_SYMBOL(generic_update_time
);
1571 * This does the actual work of updating an inodes time or version. Must have
1572 * had called mnt_want_write() before calling this.
1574 static int update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1576 int (*update_time
)(struct inode
*, struct timespec
*, int);
1578 update_time
= inode
->i_op
->update_time
? inode
->i_op
->update_time
:
1579 generic_update_time
;
1581 return update_time(inode
, time
, flags
);
1585 * touch_atime - update the access time
1586 * @path: the &struct path to update
1588 * Update the accessed time on an inode and mark it for writeback.
1589 * This function automatically handles read only file systems and media,
1590 * as well as the "noatime" flag and inode specific "noatime" markers.
1592 bool atime_needs_update(const struct path
*path
, struct inode
*inode
)
1594 struct vfsmount
*mnt
= path
->mnt
;
1595 struct timespec now
;
1597 if (inode
->i_flags
& S_NOATIME
)
1599 if (IS_NOATIME(inode
))
1601 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1604 if (mnt
->mnt_flags
& MNT_NOATIME
)
1606 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1609 now
= current_fs_time(inode
->i_sb
);
1611 if (!relatime_need_update(mnt
, inode
, now
))
1614 if (timespec_equal(&inode
->i_atime
, &now
))
1620 void touch_atime(const struct path
*path
)
1622 struct vfsmount
*mnt
= path
->mnt
;
1623 struct inode
*inode
= d_inode(path
->dentry
);
1624 struct timespec now
;
1626 if (!atime_needs_update(path
, inode
))
1629 if (!sb_start_write_trylock(inode
->i_sb
))
1632 if (__mnt_want_write(mnt
) != 0)
1635 * File systems can error out when updating inodes if they need to
1636 * allocate new space to modify an inode (such is the case for
1637 * Btrfs), but since we touch atime while walking down the path we
1638 * really don't care if we failed to update the atime of the file,
1639 * so just ignore the return value.
1640 * We may also fail on filesystems that have the ability to make parts
1641 * of the fs read only, e.g. subvolumes in Btrfs.
1643 now
= current_fs_time(inode
->i_sb
);
1644 update_time(inode
, &now
, S_ATIME
);
1645 __mnt_drop_write(mnt
);
1647 sb_end_write(inode
->i_sb
);
1649 EXPORT_SYMBOL(touch_atime
);
1652 * The logic we want is
1654 * if suid or (sgid and xgrp)
1657 int should_remove_suid(struct dentry
*dentry
)
1659 umode_t mode
= d_inode(dentry
)->i_mode
;
1662 /* suid always must be killed */
1663 if (unlikely(mode
& S_ISUID
))
1664 kill
= ATTR_KILL_SUID
;
1667 * sgid without any exec bits is just a mandatory locking mark; leave
1668 * it alone. If some exec bits are set, it's a real sgid; kill it.
1670 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1671 kill
|= ATTR_KILL_SGID
;
1673 if (unlikely(kill
&& !capable(CAP_FSETID
) && S_ISREG(mode
)))
1678 EXPORT_SYMBOL(should_remove_suid
);
1681 * Return mask of changes for notify_change() that need to be done as a
1682 * response to write or truncate. Return 0 if nothing has to be changed.
1683 * Negative value on error (change should be denied).
1685 int dentry_needs_remove_privs(struct dentry
*dentry
)
1687 struct inode
*inode
= d_inode(dentry
);
1691 if (IS_NOSEC(inode
))
1694 mask
= should_remove_suid(dentry
);
1695 ret
= security_inode_need_killpriv(dentry
);
1699 mask
|= ATTR_KILL_PRIV
;
1702 EXPORT_SYMBOL(dentry_needs_remove_privs
);
1704 static int __remove_privs(struct dentry
*dentry
, int kill
)
1706 struct iattr newattrs
;
1708 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1710 * Note we call this on write, so notify_change will not
1711 * encounter any conflicting delegations:
1713 return notify_change(dentry
, &newattrs
, NULL
);
1717 * Remove special file priviledges (suid, capabilities) when file is written
1720 int file_remove_privs(struct file
*file
)
1722 struct dentry
*dentry
= file
->f_path
.dentry
;
1723 struct inode
*inode
= d_inode(dentry
);
1727 /* Fast path for nothing security related */
1728 if (IS_NOSEC(inode
))
1731 kill
= file_needs_remove_privs(file
);
1735 error
= __remove_privs(dentry
, kill
);
1737 inode_has_no_xattr(inode
);
1741 EXPORT_SYMBOL(file_remove_privs
);
1744 * file_update_time - update mtime and ctime time
1745 * @file: file accessed
1747 * Update the mtime and ctime members of an inode and mark the inode
1748 * for writeback. Note that this function is meant exclusively for
1749 * usage in the file write path of filesystems, and filesystems may
1750 * choose to explicitly ignore update via this function with the
1751 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1752 * timestamps are handled by the server. This can return an error for
1753 * file systems who need to allocate space in order to update an inode.
1756 int file_update_time(struct file
*file
)
1758 struct inode
*inode
= file_inode(file
);
1759 struct timespec now
;
1763 /* First try to exhaust all avenues to not sync */
1764 if (IS_NOCMTIME(inode
))
1767 now
= current_fs_time(inode
->i_sb
);
1768 if (!timespec_equal(&inode
->i_mtime
, &now
))
1771 if (!timespec_equal(&inode
->i_ctime
, &now
))
1774 if (IS_I_VERSION(inode
))
1775 sync_it
|= S_VERSION
;
1780 /* Finally allowed to write? Takes lock. */
1781 if (__mnt_want_write_file(file
))
1784 ret
= update_time(inode
, &now
, sync_it
);
1785 __mnt_drop_write_file(file
);
1789 EXPORT_SYMBOL(file_update_time
);
1791 int inode_needs_sync(struct inode
*inode
)
1795 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1799 EXPORT_SYMBOL(inode_needs_sync
);
1802 * If we try to find an inode in the inode hash while it is being
1803 * deleted, we have to wait until the filesystem completes its
1804 * deletion before reporting that it isn't found. This function waits
1805 * until the deletion _might_ have completed. Callers are responsible
1806 * to recheck inode state.
1808 * It doesn't matter if I_NEW is not set initially, a call to
1809 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1812 static void __wait_on_freeing_inode(struct inode
*inode
)
1814 wait_queue_head_t
*wq
;
1815 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1816 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1817 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1818 spin_unlock(&inode
->i_lock
);
1819 spin_unlock(&inode_hash_lock
);
1821 finish_wait(wq
, &wait
.wait
);
1822 spin_lock(&inode_hash_lock
);
1825 static __initdata
unsigned long ihash_entries
;
1826 static int __init
set_ihash_entries(char *str
)
1830 ihash_entries
= simple_strtoul(str
, &str
, 0);
1833 __setup("ihash_entries=", set_ihash_entries
);
1836 * Initialize the waitqueues and inode hash table.
1838 void __init
inode_init_early(void)
1842 /* If hashes are distributed across NUMA nodes, defer
1843 * hash allocation until vmalloc space is available.
1849 alloc_large_system_hash("Inode-cache",
1850 sizeof(struct hlist_head
),
1859 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1860 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1863 void __init
inode_init(void)
1867 /* inode slab cache */
1868 inode_cachep
= kmem_cache_create("inode_cache",
1869 sizeof(struct inode
),
1871 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1875 /* Hash may have been set up in inode_init_early */
1880 alloc_large_system_hash("Inode-cache",
1881 sizeof(struct hlist_head
),
1890 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1891 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1894 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1896 inode
->i_mode
= mode
;
1897 if (S_ISCHR(mode
)) {
1898 inode
->i_fop
= &def_chr_fops
;
1899 inode
->i_rdev
= rdev
;
1900 } else if (S_ISBLK(mode
)) {
1901 inode
->i_fop
= &def_blk_fops
;
1902 inode
->i_rdev
= rdev
;
1903 } else if (S_ISFIFO(mode
))
1904 inode
->i_fop
= &pipefifo_fops
;
1905 else if (S_ISSOCK(mode
))
1906 ; /* leave it no_open_fops */
1908 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1909 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1912 EXPORT_SYMBOL(init_special_inode
);
1915 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1917 * @dir: Directory inode
1918 * @mode: mode of the new inode
1920 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1923 inode
->i_uid
= current_fsuid();
1924 if (dir
&& dir
->i_mode
& S_ISGID
) {
1925 inode
->i_gid
= dir
->i_gid
;
1929 inode
->i_gid
= current_fsgid();
1930 inode
->i_mode
= mode
;
1932 EXPORT_SYMBOL(inode_init_owner
);
1935 * inode_owner_or_capable - check current task permissions to inode
1936 * @inode: inode being checked
1938 * Return true if current either has CAP_FOWNER in a namespace with the
1939 * inode owner uid mapped, or owns the file.
1941 bool inode_owner_or_capable(const struct inode
*inode
)
1943 struct user_namespace
*ns
;
1945 if (uid_eq(current_fsuid(), inode
->i_uid
))
1948 ns
= current_user_ns();
1949 if (ns_capable(ns
, CAP_FOWNER
) && kuid_has_mapping(ns
, inode
->i_uid
))
1953 EXPORT_SYMBOL(inode_owner_or_capable
);
1956 * Direct i/o helper functions
1958 static void __inode_dio_wait(struct inode
*inode
)
1960 wait_queue_head_t
*wq
= bit_waitqueue(&inode
->i_state
, __I_DIO_WAKEUP
);
1961 DEFINE_WAIT_BIT(q
, &inode
->i_state
, __I_DIO_WAKEUP
);
1964 prepare_to_wait(wq
, &q
.wait
, TASK_UNINTERRUPTIBLE
);
1965 if (atomic_read(&inode
->i_dio_count
))
1967 } while (atomic_read(&inode
->i_dio_count
));
1968 finish_wait(wq
, &q
.wait
);
1972 * inode_dio_wait - wait for outstanding DIO requests to finish
1973 * @inode: inode to wait for
1975 * Waits for all pending direct I/O requests to finish so that we can
1976 * proceed with a truncate or equivalent operation.
1978 * Must be called under a lock that serializes taking new references
1979 * to i_dio_count, usually by inode->i_mutex.
1981 void inode_dio_wait(struct inode
*inode
)
1983 if (atomic_read(&inode
->i_dio_count
))
1984 __inode_dio_wait(inode
);
1986 EXPORT_SYMBOL(inode_dio_wait
);
1989 * inode_set_flags - atomically set some inode flags
1991 * Note: the caller should be holding i_mutex, or else be sure that
1992 * they have exclusive access to the inode structure (i.e., while the
1993 * inode is being instantiated). The reason for the cmpxchg() loop
1994 * --- which wouldn't be necessary if all code paths which modify
1995 * i_flags actually followed this rule, is that there is at least one
1996 * code path which doesn't today so we use cmpxchg() out of an abundance
1999 * In the long run, i_mutex is overkill, and we should probably look
2000 * at using the i_lock spinlock to protect i_flags, and then make sure
2001 * it is so documented in include/linux/fs.h and that all code follows
2002 * the locking convention!!
2004 void inode_set_flags(struct inode
*inode
, unsigned int flags
,
2007 unsigned int old_flags
, new_flags
;
2009 WARN_ON_ONCE(flags
& ~mask
);
2011 old_flags
= ACCESS_ONCE(inode
->i_flags
);
2012 new_flags
= (old_flags
& ~mask
) | flags
;
2013 } while (unlikely(cmpxchg(&inode
->i_flags
, old_flags
,
2014 new_flags
) != old_flags
));
2016 EXPORT_SYMBOL(inode_set_flags
);