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->i_sb->s_inode_list_lock protects:
32 * inode->i_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_io_list
35 * inode_hash_lock protects:
36 * inode_hashtable, inode->i_hash
40 * inode->i_sb->s_inode_list_lock
42 * Inode LRU list locks
48 * inode->i_sb->s_inode_list_lock
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
);
61 * Empty aops. Can be used for the cases where the user does not
62 * define any of the address_space operations.
64 const struct address_space_operations empty_aops
= {
66 EXPORT_SYMBOL(empty_aops
);
69 * Statistics gathering..
71 struct inodes_stat_t inodes_stat
;
73 static DEFINE_PER_CPU(unsigned long, nr_inodes
);
74 static DEFINE_PER_CPU(unsigned long, nr_unused
);
76 static struct kmem_cache
*inode_cachep __read_mostly
;
78 static long get_nr_inodes(void)
82 for_each_possible_cpu(i
)
83 sum
+= per_cpu(nr_inodes
, i
);
84 return sum
< 0 ? 0 : sum
;
87 static inline long get_nr_inodes_unused(void)
91 for_each_possible_cpu(i
)
92 sum
+= per_cpu(nr_unused
, i
);
93 return sum
< 0 ? 0 : sum
;
96 long get_nr_dirty_inodes(void)
98 /* not actually dirty inodes, but a wild approximation */
99 long nr_dirty
= get_nr_inodes() - get_nr_inodes_unused();
100 return nr_dirty
> 0 ? nr_dirty
: 0;
104 * Handle nr_inode sysctl
107 int proc_nr_inodes(struct ctl_table
*table
, int write
,
108 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
110 inodes_stat
.nr_inodes
= get_nr_inodes();
111 inodes_stat
.nr_unused
= get_nr_inodes_unused();
112 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
116 static int no_open(struct inode
*inode
, struct file
*file
)
122 * inode_init_always - perform inode structure intialisation
123 * @sb: superblock inode belongs to
124 * @inode: inode to initialise
126 * These are initializations that need to be done on every inode
127 * allocation as the fields are not initialised by slab allocation.
129 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
131 static const struct inode_operations empty_iops
;
132 static const struct file_operations no_open_fops
= {.open
= no_open
};
133 struct address_space
*const mapping
= &inode
->i_data
;
136 inode
->i_blkbits
= sb
->s_blocksize_bits
;
138 atomic_set(&inode
->i_count
, 1);
139 inode
->i_op
= &empty_iops
;
140 inode
->i_fop
= &no_open_fops
;
141 inode
->__i_nlink
= 1;
142 inode
->i_opflags
= 0;
143 i_uid_write(inode
, 0);
144 i_gid_write(inode
, 0);
145 atomic_set(&inode
->i_writecount
, 0);
149 inode
->i_generation
= 0;
150 inode
->i_pipe
= NULL
;
151 inode
->i_bdev
= NULL
;
152 inode
->i_cdev
= NULL
;
153 inode
->i_link
= NULL
;
155 inode
->dirtied_when
= 0;
157 if (security_inode_alloc(inode
))
159 spin_lock_init(&inode
->i_lock
);
160 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
162 mutex_init(&inode
->i_mutex
);
163 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
165 atomic_set(&inode
->i_dio_count
, 0);
167 mapping
->a_ops
= &empty_aops
;
168 mapping
->host
= inode
;
170 atomic_set(&mapping
->i_mmap_writable
, 0);
171 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
172 mapping
->private_data
= NULL
;
173 mapping
->writeback_index
= 0;
174 #if defined(CONFIG_FMP_ECRYPT_FS) || defined(CONFIG_FMP_EXT4CRYPT_FS)
176 memset(mapping
->key
, 0, KEY_MAX_SIZE
);
177 mapping
->key_length
= 0;
179 mapping
->sensitive_data_index
= 0;
180 mapping
->hash_tfm
= NULL
;
181 #ifdef CONFIG_CRYPTO_FIPS
182 mapping
->cc_enable
= 0;
184 mapping
->private_enc_mode
= 0;
185 mapping
->private_algo_mode
= 0;
186 mapping
->plain_text
= 0;
191 inode
->i_private
= NULL
;
192 inode
->i_mapping
= mapping
;
193 INIT_HLIST_HEAD(&inode
->i_dentry
); /* buggered by rcu freeing */
194 #ifdef CONFIG_FS_POSIX_ACL
195 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
198 #ifdef CONFIG_FSNOTIFY
199 inode
->i_fsnotify_mask
= 0;
201 inode
->i_flctx
= NULL
;
202 this_cpu_inc(nr_inodes
);
208 EXPORT_SYMBOL(inode_init_always
);
210 static struct inode
*alloc_inode(struct super_block
*sb
)
214 if (sb
->s_op
->alloc_inode
)
215 inode
= sb
->s_op
->alloc_inode(sb
);
217 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
222 if (unlikely(inode_init_always(sb
, inode
))) {
223 if (inode
->i_sb
->s_op
->destroy_inode
)
224 inode
->i_sb
->s_op
->destroy_inode(inode
);
226 kmem_cache_free(inode_cachep
, inode
);
233 void free_inode_nonrcu(struct inode
*inode
)
235 kmem_cache_free(inode_cachep
, inode
);
237 EXPORT_SYMBOL(free_inode_nonrcu
);
239 void __destroy_inode(struct inode
*inode
)
241 BUG_ON(inode_has_buffers(inode
));
242 inode_detach_wb(inode
);
243 security_inode_free(inode
);
244 fsnotify_inode_delete(inode
);
245 locks_free_lock_context(inode
->i_flctx
);
246 if (!inode
->i_nlink
) {
247 WARN_ON(atomic_long_read(&inode
->i_sb
->s_remove_count
) == 0);
248 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
251 #ifdef CONFIG_FS_POSIX_ACL
252 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
253 posix_acl_release(inode
->i_acl
);
254 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
255 posix_acl_release(inode
->i_default_acl
);
257 this_cpu_dec(nr_inodes
);
259 EXPORT_SYMBOL(__destroy_inode
);
261 static void i_callback(struct rcu_head
*head
)
263 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
264 kmem_cache_free(inode_cachep
, inode
);
267 static void destroy_inode(struct inode
*inode
)
269 BUG_ON(!list_empty(&inode
->i_lru
));
270 __destroy_inode(inode
);
271 if (inode
->i_sb
->s_op
->destroy_inode
)
272 inode
->i_sb
->s_op
->destroy_inode(inode
);
274 call_rcu(&inode
->i_rcu
, i_callback
);
278 * drop_nlink - directly drop an inode's link count
281 * This is a low-level filesystem helper to replace any
282 * direct filesystem manipulation of i_nlink. In cases
283 * where we are attempting to track writes to the
284 * filesystem, a decrement to zero means an imminent
285 * write when the file is truncated and actually unlinked
288 void drop_nlink(struct inode
*inode
)
290 WARN_ON(inode
->i_nlink
== 0);
293 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
295 EXPORT_SYMBOL(drop_nlink
);
298 * clear_nlink - directly zero an inode's link count
301 * This is a low-level filesystem helper to replace any
302 * direct filesystem manipulation of i_nlink. See
303 * drop_nlink() for why we care about i_nlink hitting zero.
305 void clear_nlink(struct inode
*inode
)
307 if (inode
->i_nlink
) {
308 inode
->__i_nlink
= 0;
309 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
312 EXPORT_SYMBOL(clear_nlink
);
315 * set_nlink - directly set an inode's link count
317 * @nlink: new nlink (should be non-zero)
319 * This is a low-level filesystem helper to replace any
320 * direct filesystem manipulation of i_nlink.
322 void set_nlink(struct inode
*inode
, unsigned int nlink
)
327 /* Yes, some filesystems do change nlink from zero to one */
328 if (inode
->i_nlink
== 0)
329 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
331 inode
->__i_nlink
= nlink
;
334 EXPORT_SYMBOL(set_nlink
);
337 * inc_nlink - directly increment an inode's link count
340 * This is a low-level filesystem helper to replace any
341 * direct filesystem manipulation of i_nlink. Currently,
342 * it is only here for parity with dec_nlink().
344 void inc_nlink(struct inode
*inode
)
346 if (unlikely(inode
->i_nlink
== 0)) {
347 WARN_ON(!(inode
->i_state
& I_LINKABLE
));
348 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
353 EXPORT_SYMBOL(inc_nlink
);
355 void address_space_init_once(struct address_space
*mapping
)
357 memset(mapping
, 0, sizeof(*mapping
));
358 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
359 spin_lock_init(&mapping
->tree_lock
);
360 init_rwsem(&mapping
->i_mmap_rwsem
);
361 INIT_LIST_HEAD(&mapping
->private_list
);
362 spin_lock_init(&mapping
->private_lock
);
363 mapping
->i_mmap
= RB_ROOT
;
365 EXPORT_SYMBOL(address_space_init_once
);
368 * These are initializations that only need to be done
369 * once, because the fields are idempotent across use
370 * of the inode, so let the slab aware of that.
372 void inode_init_once(struct inode
*inode
)
374 memset(inode
, 0, sizeof(*inode
));
375 INIT_HLIST_NODE(&inode
->i_hash
);
376 INIT_LIST_HEAD(&inode
->i_devices
);
377 INIT_LIST_HEAD(&inode
->i_io_list
);
378 INIT_LIST_HEAD(&inode
->i_lru
);
379 address_space_init_once(&inode
->i_data
);
380 i_size_ordered_init(inode
);
381 #ifdef CONFIG_FSNOTIFY
382 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
385 EXPORT_SYMBOL(inode_init_once
);
387 static void init_once(void *foo
)
389 struct inode
*inode
= (struct inode
*) foo
;
391 inode_init_once(inode
);
395 * inode->i_lock must be held
397 void __iget(struct inode
*inode
)
399 atomic_inc(&inode
->i_count
);
403 * get additional reference to inode; caller must already hold one.
405 void ihold(struct inode
*inode
)
407 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
409 EXPORT_SYMBOL(ihold
);
411 static void inode_lru_list_add(struct inode
*inode
)
413 if (list_lru_add(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
414 this_cpu_inc(nr_unused
);
418 * Add inode to LRU if needed (inode is unused and clean).
420 * Needs inode->i_lock held.
422 void inode_add_lru(struct inode
*inode
)
424 if (!(inode
->i_state
& (I_DIRTY_ALL
| I_SYNC
|
425 I_FREEING
| I_WILL_FREE
)) &&
426 !atomic_read(&inode
->i_count
) && inode
->i_sb
->s_flags
& MS_ACTIVE
)
427 inode_lru_list_add(inode
);
431 static void inode_lru_list_del(struct inode
*inode
)
434 if (list_lru_del(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
435 this_cpu_dec(nr_unused
);
439 * inode_sb_list_add - add inode to the superblock list of inodes
440 * @inode: inode to add
442 void inode_sb_list_add(struct inode
*inode
)
444 spin_lock(&inode
->i_sb
->s_inode_list_lock
);
445 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
446 spin_unlock(&inode
->i_sb
->s_inode_list_lock
);
448 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
450 static inline void inode_sb_list_del(struct inode
*inode
)
452 if (!list_empty(&inode
->i_sb_list
)) {
453 spin_lock(&inode
->i_sb
->s_inode_list_lock
);
454 list_del_init(&inode
->i_sb_list
);
455 spin_unlock(&inode
->i_sb
->s_inode_list_lock
);
459 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
463 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
465 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> i_hash_shift
);
466 return tmp
& i_hash_mask
;
470 * __insert_inode_hash - hash an inode
471 * @inode: unhashed inode
472 * @hashval: unsigned long value used to locate this object in the
475 * Add an inode to the inode hash for this superblock.
477 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
479 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
481 spin_lock(&inode_hash_lock
);
482 spin_lock(&inode
->i_lock
);
483 hlist_add_head(&inode
->i_hash
, b
);
484 spin_unlock(&inode
->i_lock
);
485 spin_unlock(&inode_hash_lock
);
487 EXPORT_SYMBOL(__insert_inode_hash
);
490 * __remove_inode_hash - remove an inode from the hash
491 * @inode: inode to unhash
493 * Remove an inode from the superblock.
495 void __remove_inode_hash(struct inode
*inode
)
497 spin_lock(&inode_hash_lock
);
498 spin_lock(&inode
->i_lock
);
499 hlist_del_init(&inode
->i_hash
);
500 spin_unlock(&inode
->i_lock
);
501 spin_unlock(&inode_hash_lock
);
503 EXPORT_SYMBOL(__remove_inode_hash
);
505 void clear_inode(struct inode
*inode
)
509 * We have to cycle tree_lock here because reclaim can be still in the
510 * process of removing the last page (in __delete_from_page_cache())
511 * and we must not free mapping under it.
513 spin_lock_irq(&inode
->i_data
.tree_lock
);
514 BUG_ON(inode
->i_data
.nrpages
);
515 BUG_ON(inode
->i_data
.nrshadows
);
516 spin_unlock_irq(&inode
->i_data
.tree_lock
);
517 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
518 BUG_ON(!(inode
->i_state
& I_FREEING
));
519 BUG_ON(inode
->i_state
& I_CLEAR
);
520 /* don't need i_lock here, no concurrent mods to i_state */
521 inode
->i_state
= I_FREEING
| I_CLEAR
;
523 EXPORT_SYMBOL(clear_inode
);
526 * Free the inode passed in, removing it from the lists it is still connected
527 * to. We remove any pages still attached to the inode and wait for any IO that
528 * is still in progress before finally destroying the inode.
530 * An inode must already be marked I_FREEING so that we avoid the inode being
531 * moved back onto lists if we race with other code that manipulates the lists
532 * (e.g. writeback_single_inode). The caller is responsible for setting this.
534 * An inode must already be removed from the LRU list before being evicted from
535 * the cache. This should occur atomically with setting the I_FREEING state
536 * flag, so no inodes here should ever be on the LRU when being evicted.
538 static void evict(struct inode
*inode
)
540 const struct super_operations
*op
= inode
->i_sb
->s_op
;
542 BUG_ON(!(inode
->i_state
& I_FREEING
));
543 BUG_ON(!list_empty(&inode
->i_lru
));
545 if (!list_empty(&inode
->i_io_list
))
546 inode_io_list_del(inode
);
548 inode_sb_list_del(inode
);
551 * Wait for flusher thread to be done with the inode so that filesystem
552 * does not start destroying it while writeback is still running. Since
553 * the inode has I_FREEING set, flusher thread won't start new work on
554 * the inode. We just have to wait for running writeback to finish.
556 inode_wait_for_writeback(inode
);
558 if (op
->evict_inode
) {
559 op
->evict_inode(inode
);
561 truncate_inode_pages_final(&inode
->i_data
);
564 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
566 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
569 remove_inode_hash(inode
);
571 spin_lock(&inode
->i_lock
);
572 wake_up_bit(&inode
->i_state
, __I_NEW
);
573 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
574 spin_unlock(&inode
->i_lock
);
576 destroy_inode(inode
);
580 * dispose_list - dispose of the contents of a local list
581 * @head: the head of the list to free
583 * Dispose-list gets a local list with local inodes in it, so it doesn't
584 * need to worry about list corruption and SMP locks.
586 static void dispose_list(struct list_head
*head
)
588 while (!list_empty(head
)) {
591 inode
= list_first_entry(head
, struct inode
, i_lru
);
592 list_del_init(&inode
->i_lru
);
600 * evict_inodes - evict all evictable inodes for a superblock
601 * @sb: superblock to operate on
603 * Make sure that no inodes with zero refcount are retained. This is
604 * called by superblock shutdown after having MS_ACTIVE flag removed,
605 * so any inode reaching zero refcount during or after that call will
606 * be immediately evicted.
608 void evict_inodes(struct super_block
*sb
)
610 struct inode
*inode
, *next
;
614 spin_lock(&sb
->s_inode_list_lock
);
615 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
616 if (atomic_read(&inode
->i_count
))
619 spin_lock(&inode
->i_lock
);
620 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
621 spin_unlock(&inode
->i_lock
);
625 inode
->i_state
|= I_FREEING
;
626 inode_lru_list_del(inode
);
627 spin_unlock(&inode
->i_lock
);
628 list_add(&inode
->i_lru
, &dispose
);
631 * We can have a ton of inodes to evict at unmount time given
632 * enough memory, check to see if we need to go to sleep for a
633 * bit so we don't livelock.
635 if (need_resched()) {
636 spin_unlock(&sb
->s_inode_list_lock
);
638 dispose_list(&dispose
);
642 spin_unlock(&sb
->s_inode_list_lock
);
644 dispose_list(&dispose
);
648 * invalidate_inodes - attempt to free all inodes on a superblock
649 * @sb: superblock to operate on
650 * @kill_dirty: flag to guide handling of dirty inodes
652 * Attempts to free all inodes for a given superblock. If there were any
653 * busy inodes return a non-zero value, else zero.
654 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
657 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
660 struct inode
*inode
, *next
;
663 spin_lock(&sb
->s_inode_list_lock
);
664 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
665 spin_lock(&inode
->i_lock
);
666 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
667 spin_unlock(&inode
->i_lock
);
670 if (inode
->i_state
& I_DIRTY_ALL
&& !kill_dirty
) {
671 spin_unlock(&inode
->i_lock
);
675 if (atomic_read(&inode
->i_count
)) {
676 spin_unlock(&inode
->i_lock
);
681 inode
->i_state
|= I_FREEING
;
682 inode_lru_list_del(inode
);
683 spin_unlock(&inode
->i_lock
);
684 list_add(&inode
->i_lru
, &dispose
);
686 spin_unlock(&sb
->s_inode_list_lock
);
688 dispose_list(&dispose
);
694 * Isolate the inode from the LRU in preparation for freeing it.
696 * Any inodes which are pinned purely because of attached pagecache have their
697 * pagecache removed. If the inode has metadata buffers attached to
698 * mapping->private_list then try to remove them.
700 * If the inode has the I_REFERENCED flag set, then it means that it has been
701 * used recently - the flag is set in iput_final(). When we encounter such an
702 * inode, clear the flag and move it to the back of the LRU so it gets another
703 * pass through the LRU before it gets reclaimed. This is necessary because of
704 * the fact we are doing lazy LRU updates to minimise lock contention so the
705 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
706 * with this flag set because they are the inodes that are out of order.
708 static enum lru_status
inode_lru_isolate(struct list_head
*item
,
709 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
711 struct list_head
*freeable
= arg
;
712 struct inode
*inode
= container_of(item
, struct inode
, i_lru
);
715 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
716 * If we fail to get the lock, just skip it.
718 if (!spin_trylock(&inode
->i_lock
))
722 * Referenced or dirty inodes are still in use. Give them another pass
723 * through the LRU as we canot reclaim them now.
725 if (atomic_read(&inode
->i_count
) ||
726 (inode
->i_state
& ~I_REFERENCED
)) {
727 list_lru_isolate(lru
, &inode
->i_lru
);
728 spin_unlock(&inode
->i_lock
);
729 this_cpu_dec(nr_unused
);
733 /* recently referenced inodes get one more pass */
734 if (inode
->i_state
& I_REFERENCED
) {
735 inode
->i_state
&= ~I_REFERENCED
;
736 spin_unlock(&inode
->i_lock
);
740 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
742 spin_unlock(&inode
->i_lock
);
743 spin_unlock(lru_lock
);
744 if (remove_inode_buffers(inode
)) {
746 reap
= invalidate_mapping_pages(&inode
->i_data
, 0, -1);
747 if (current_is_kswapd())
748 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
750 __count_vm_events(PGINODESTEAL
, reap
);
751 if (current
->reclaim_state
)
752 current
->reclaim_state
->reclaimed_slab
+= reap
;
759 WARN_ON(inode
->i_state
& I_NEW
);
760 inode
->i_state
|= I_FREEING
;
761 list_lru_isolate_move(lru
, &inode
->i_lru
, freeable
);
762 spin_unlock(&inode
->i_lock
);
764 this_cpu_dec(nr_unused
);
769 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
770 * This is called from the superblock shrinker function with a number of inodes
771 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
772 * then are freed outside inode_lock by dispose_list().
774 long prune_icache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
779 freed
= list_lru_shrink_walk(&sb
->s_inode_lru
, sc
,
780 inode_lru_isolate
, &freeable
);
781 dispose_list(&freeable
);
785 static void __wait_on_freeing_inode(struct inode
*inode
);
787 * Called with the inode lock held.
789 static struct inode
*find_inode(struct super_block
*sb
,
790 struct hlist_head
*head
,
791 int (*test
)(struct inode
*, void *),
794 struct inode
*inode
= NULL
;
797 hlist_for_each_entry(inode
, head
, i_hash
) {
798 if (inode
->i_sb
!= sb
)
800 if (!test(inode
, data
))
802 spin_lock(&inode
->i_lock
);
803 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
804 __wait_on_freeing_inode(inode
);
808 spin_unlock(&inode
->i_lock
);
815 * find_inode_fast is the fast path version of find_inode, see the comment at
816 * iget_locked for details.
818 static struct inode
*find_inode_fast(struct super_block
*sb
,
819 struct hlist_head
*head
, unsigned long ino
)
821 struct inode
*inode
= NULL
;
824 hlist_for_each_entry(inode
, head
, i_hash
) {
825 if (inode
->i_ino
!= ino
)
827 if (inode
->i_sb
!= sb
)
829 spin_lock(&inode
->i_lock
);
830 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
831 __wait_on_freeing_inode(inode
);
835 spin_unlock(&inode
->i_lock
);
842 * Each cpu owns a range of LAST_INO_BATCH numbers.
843 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
844 * to renew the exhausted range.
846 * This does not significantly increase overflow rate because every CPU can
847 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
848 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
849 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
850 * overflow rate by 2x, which does not seem too significant.
852 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
853 * error if st_ino won't fit in target struct field. Use 32bit counter
854 * here to attempt to avoid that.
856 #define LAST_INO_BATCH 1024
857 static DEFINE_PER_CPU(unsigned int, last_ino
);
859 unsigned int get_next_ino(void)
861 unsigned int *p
= &get_cpu_var(last_ino
);
862 unsigned int res
= *p
;
865 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
866 static atomic_t shared_last_ino
;
867 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
869 res
= next
- LAST_INO_BATCH
;
874 /* get_next_ino should not provide a 0 inode number */
878 put_cpu_var(last_ino
);
881 EXPORT_SYMBOL(get_next_ino
);
884 * new_inode_pseudo - obtain an inode
887 * Allocates a new inode for given superblock.
888 * Inode wont be chained in superblock s_inodes list
890 * - fs can't be unmount
891 * - quotas, fsnotify, writeback can't work
893 struct inode
*new_inode_pseudo(struct super_block
*sb
)
895 struct inode
*inode
= alloc_inode(sb
);
898 spin_lock(&inode
->i_lock
);
900 spin_unlock(&inode
->i_lock
);
901 INIT_LIST_HEAD(&inode
->i_sb_list
);
907 * new_inode - obtain an inode
910 * Allocates a new inode for given superblock. The default gfp_mask
911 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
912 * If HIGHMEM pages are unsuitable or it is known that pages allocated
913 * for the page cache are not reclaimable or migratable,
914 * mapping_set_gfp_mask() must be called with suitable flags on the
915 * newly created inode's mapping
918 struct inode
*new_inode(struct super_block
*sb
)
922 spin_lock_prefetch(&sb
->s_inode_list_lock
);
924 inode
= new_inode_pseudo(sb
);
926 inode_sb_list_add(inode
);
929 EXPORT_SYMBOL(new_inode
);
931 #ifdef CONFIG_DEBUG_LOCK_ALLOC
932 void lockdep_annotate_inode_mutex_key(struct inode
*inode
)
934 if (S_ISDIR(inode
->i_mode
)) {
935 struct file_system_type
*type
= inode
->i_sb
->s_type
;
937 /* Set new key only if filesystem hasn't already changed it */
938 if (lockdep_match_class(&inode
->i_mutex
, &type
->i_mutex_key
)) {
940 * ensure nobody is actually holding i_mutex
942 mutex_destroy(&inode
->i_mutex
);
943 mutex_init(&inode
->i_mutex
);
944 lockdep_set_class(&inode
->i_mutex
,
945 &type
->i_mutex_dir_key
);
949 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key
);
953 * unlock_new_inode - clear the I_NEW state and wake up any waiters
954 * @inode: new inode to unlock
956 * Called when the inode is fully initialised to clear the new state of the
957 * inode and wake up anyone waiting for the inode to finish initialisation.
959 void unlock_new_inode(struct inode
*inode
)
961 lockdep_annotate_inode_mutex_key(inode
);
962 spin_lock(&inode
->i_lock
);
963 WARN_ON(!(inode
->i_state
& I_NEW
));
964 inode
->i_state
&= ~I_NEW
;
966 wake_up_bit(&inode
->i_state
, __I_NEW
);
967 spin_unlock(&inode
->i_lock
);
969 EXPORT_SYMBOL(unlock_new_inode
);
972 * lock_two_nondirectories - take two i_mutexes on non-directory objects
974 * Lock any non-NULL argument that is not a directory.
975 * Zero, one or two objects may be locked by this function.
977 * @inode1: first inode to lock
978 * @inode2: second inode to lock
980 void lock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
983 swap(inode1
, inode2
);
985 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
986 mutex_lock(&inode1
->i_mutex
);
987 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
988 mutex_lock_nested(&inode2
->i_mutex
, I_MUTEX_NONDIR2
);
990 EXPORT_SYMBOL(lock_two_nondirectories
);
993 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
994 * @inode1: first inode to unlock
995 * @inode2: second inode to unlock
997 void unlock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
999 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
1000 mutex_unlock(&inode1
->i_mutex
);
1001 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
1002 mutex_unlock(&inode2
->i_mutex
);
1004 EXPORT_SYMBOL(unlock_two_nondirectories
);
1007 * iget5_locked - obtain an inode from a mounted file system
1008 * @sb: super block of file system
1009 * @hashval: hash value (usually inode number) to get
1010 * @test: callback used for comparisons between inodes
1011 * @set: callback used to initialize a new struct inode
1012 * @data: opaque data pointer to pass to @test and @set
1014 * Search for the inode specified by @hashval and @data in the inode cache,
1015 * and if present it is return it with an increased reference count. This is
1016 * a generalized version of iget_locked() for file systems where the inode
1017 * number is not sufficient for unique identification of an inode.
1019 * If the inode is not in cache, allocate a new inode and return it locked,
1020 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1021 * before unlocking it via unlock_new_inode().
1023 * Note both @test and @set are called with the inode_hash_lock held, so can't
1026 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1027 int (*test
)(struct inode
*, void *),
1028 int (*set
)(struct inode
*, void *), void *data
)
1030 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1031 struct inode
*inode
;
1033 spin_lock(&inode_hash_lock
);
1034 inode
= find_inode(sb
, head
, test
, data
);
1035 spin_unlock(&inode_hash_lock
);
1038 wait_on_inode(inode
);
1042 inode
= alloc_inode(sb
);
1046 spin_lock(&inode_hash_lock
);
1047 /* We released the lock, so.. */
1048 old
= find_inode(sb
, head
, test
, data
);
1050 if (set(inode
, data
))
1053 spin_lock(&inode
->i_lock
);
1054 inode
->i_state
= I_NEW
;
1055 hlist_add_head(&inode
->i_hash
, head
);
1056 spin_unlock(&inode
->i_lock
);
1057 inode_sb_list_add(inode
);
1058 spin_unlock(&inode_hash_lock
);
1060 /* Return the locked inode with I_NEW set, the
1061 * caller is responsible for filling in the contents
1067 * Uhhuh, somebody else created the same inode under
1068 * us. Use the old inode instead of the one we just
1071 spin_unlock(&inode_hash_lock
);
1072 destroy_inode(inode
);
1074 wait_on_inode(inode
);
1079 spin_unlock(&inode_hash_lock
);
1080 destroy_inode(inode
);
1083 EXPORT_SYMBOL(iget5_locked
);
1086 * iget_locked - obtain an inode from a mounted file system
1087 * @sb: super block of file system
1088 * @ino: inode number to get
1090 * Search for the inode specified by @ino in the inode cache and if present
1091 * return it with an increased reference count. This is for file systems
1092 * where the inode number is sufficient for unique identification of an inode.
1094 * If the inode is not in cache, allocate a new inode and return it locked,
1095 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1096 * before unlocking it via unlock_new_inode().
1098 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1100 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1101 struct inode
*inode
;
1103 spin_lock(&inode_hash_lock
);
1104 inode
= find_inode_fast(sb
, head
, ino
);
1105 spin_unlock(&inode_hash_lock
);
1107 wait_on_inode(inode
);
1111 inode
= alloc_inode(sb
);
1115 spin_lock(&inode_hash_lock
);
1116 /* We released the lock, so.. */
1117 old
= find_inode_fast(sb
, head
, ino
);
1120 spin_lock(&inode
->i_lock
);
1121 inode
->i_state
= I_NEW
;
1122 hlist_add_head(&inode
->i_hash
, head
);
1123 spin_unlock(&inode
->i_lock
);
1124 inode_sb_list_add(inode
);
1125 spin_unlock(&inode_hash_lock
);
1127 /* Return the locked inode with I_NEW set, the
1128 * caller is responsible for filling in the contents
1134 * Uhhuh, somebody else created the same inode under
1135 * us. Use the old inode instead of the one we just
1138 spin_unlock(&inode_hash_lock
);
1139 destroy_inode(inode
);
1141 wait_on_inode(inode
);
1145 EXPORT_SYMBOL(iget_locked
);
1148 * search the inode cache for a matching inode number.
1149 * If we find one, then the inode number we are trying to
1150 * allocate is not unique and so we should not use it.
1152 * Returns 1 if the inode number is unique, 0 if it is not.
1154 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1156 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1157 struct inode
*inode
;
1159 spin_lock(&inode_hash_lock
);
1160 hlist_for_each_entry(inode
, b
, i_hash
) {
1161 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
) {
1162 spin_unlock(&inode_hash_lock
);
1166 spin_unlock(&inode_hash_lock
);
1172 * iunique - get a unique inode number
1174 * @max_reserved: highest reserved inode number
1176 * Obtain an inode number that is unique on the system for a given
1177 * superblock. This is used by file systems that have no natural
1178 * permanent inode numbering system. An inode number is returned that
1179 * is higher than the reserved limit but unique.
1182 * With a large number of inodes live on the file system this function
1183 * currently becomes quite slow.
1185 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1188 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1189 * error if st_ino won't fit in target struct field. Use 32bit counter
1190 * here to attempt to avoid that.
1192 static DEFINE_SPINLOCK(iunique_lock
);
1193 static unsigned int counter
;
1196 spin_lock(&iunique_lock
);
1198 if (counter
<= max_reserved
)
1199 counter
= max_reserved
+ 1;
1201 } while (!test_inode_iunique(sb
, res
));
1202 spin_unlock(&iunique_lock
);
1206 EXPORT_SYMBOL(iunique
);
1208 struct inode
*igrab(struct inode
*inode
)
1210 spin_lock(&inode
->i_lock
);
1211 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1213 spin_unlock(&inode
->i_lock
);
1215 spin_unlock(&inode
->i_lock
);
1217 * Handle the case where s_op->clear_inode is not been
1218 * called yet, and somebody is calling igrab
1219 * while the inode is getting freed.
1225 EXPORT_SYMBOL(igrab
);
1228 * ilookup5_nowait - 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 * If the inode is in the cache, the inode is returned with an incremented
1238 * Note: I_NEW is not waited upon so you have to be very careful what you do
1239 * with the returned inode. You probably should be using ilookup5() instead.
1241 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1243 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1244 int (*test
)(struct inode
*, void *), void *data
)
1246 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1247 struct inode
*inode
;
1249 spin_lock(&inode_hash_lock
);
1250 inode
= find_inode(sb
, head
, test
, data
);
1251 spin_unlock(&inode_hash_lock
);
1255 EXPORT_SYMBOL(ilookup5_nowait
);
1258 * ilookup5 - search for an inode in the inode cache
1259 * @sb: super block of file system to search
1260 * @hashval: hash value (usually inode number) to search for
1261 * @test: callback used for comparisons between inodes
1262 * @data: opaque data pointer to pass to @test
1264 * Search for the inode specified by @hashval and @data in the inode cache,
1265 * and if the inode is in the cache, return the inode with an incremented
1266 * reference count. Waits on I_NEW before returning the inode.
1267 * returned with an incremented reference count.
1269 * This is a generalized version of ilookup() for file systems where the
1270 * inode number is not sufficient for unique identification of an inode.
1272 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1274 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1275 int (*test
)(struct inode
*, void *), void *data
)
1277 struct inode
*inode
= ilookup5_nowait(sb
, hashval
, test
, data
);
1280 wait_on_inode(inode
);
1283 EXPORT_SYMBOL(ilookup5
);
1286 * ilookup - search for an inode in the inode cache
1287 * @sb: super block of file system to search
1288 * @ino: inode number to search for
1290 * Search for the inode @ino in the inode cache, and if the inode is in the
1291 * cache, the inode is returned with an incremented reference count.
1293 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1295 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1296 struct inode
*inode
;
1298 spin_lock(&inode_hash_lock
);
1299 inode
= find_inode_fast(sb
, head
, ino
);
1300 spin_unlock(&inode_hash_lock
);
1303 wait_on_inode(inode
);
1306 EXPORT_SYMBOL(ilookup
);
1309 * find_inode_nowait - find an inode in the inode cache
1310 * @sb: super block of file system to search
1311 * @hashval: hash value (usually inode number) to search for
1312 * @match: callback used for comparisons between inodes
1313 * @data: opaque data pointer to pass to @match
1315 * Search for the inode specified by @hashval and @data in the inode
1316 * cache, where the helper function @match will return 0 if the inode
1317 * does not match, 1 if the inode does match, and -1 if the search
1318 * should be stopped. The @match function must be responsible for
1319 * taking the i_lock spin_lock and checking i_state for an inode being
1320 * freed or being initialized, and incrementing the reference count
1321 * before returning 1. It also must not sleep, since it is called with
1322 * the inode_hash_lock spinlock held.
1324 * This is a even more generalized version of ilookup5() when the
1325 * function must never block --- find_inode() can block in
1326 * __wait_on_freeing_inode() --- or when the caller can not increment
1327 * the reference count because the resulting iput() might cause an
1328 * inode eviction. The tradeoff is that the @match funtion must be
1329 * very carefully implemented.
1331 struct inode
*find_inode_nowait(struct super_block
*sb
,
1332 unsigned long hashval
,
1333 int (*match
)(struct inode
*, unsigned long,
1337 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1338 struct inode
*inode
, *ret_inode
= NULL
;
1341 spin_lock(&inode_hash_lock
);
1342 hlist_for_each_entry(inode
, head
, i_hash
) {
1343 if (inode
->i_sb
!= sb
)
1345 mval
= match(inode
, hashval
, data
);
1353 spin_unlock(&inode_hash_lock
);
1356 EXPORT_SYMBOL(find_inode_nowait
);
1358 int insert_inode_locked(struct inode
*inode
)
1360 struct super_block
*sb
= inode
->i_sb
;
1361 ino_t ino
= inode
->i_ino
;
1362 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1365 struct inode
*old
= NULL
;
1366 spin_lock(&inode_hash_lock
);
1367 hlist_for_each_entry(old
, head
, i_hash
) {
1368 if (old
->i_ino
!= ino
)
1370 if (old
->i_sb
!= sb
)
1372 spin_lock(&old
->i_lock
);
1373 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1374 spin_unlock(&old
->i_lock
);
1380 spin_lock(&inode
->i_lock
);
1381 inode
->i_state
|= I_NEW
;
1382 hlist_add_head(&inode
->i_hash
, head
);
1383 spin_unlock(&inode
->i_lock
);
1384 spin_unlock(&inode_hash_lock
);
1388 spin_unlock(&old
->i_lock
);
1389 spin_unlock(&inode_hash_lock
);
1391 if (unlikely(!inode_unhashed(old
))) {
1398 EXPORT_SYMBOL(insert_inode_locked
);
1400 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1401 int (*test
)(struct inode
*, void *), void *data
)
1403 struct super_block
*sb
= inode
->i_sb
;
1404 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1407 struct inode
*old
= NULL
;
1409 spin_lock(&inode_hash_lock
);
1410 hlist_for_each_entry(old
, head
, i_hash
) {
1411 if (old
->i_sb
!= sb
)
1413 if (!test(old
, data
))
1415 spin_lock(&old
->i_lock
);
1416 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1417 spin_unlock(&old
->i_lock
);
1423 spin_lock(&inode
->i_lock
);
1424 inode
->i_state
|= I_NEW
;
1425 hlist_add_head(&inode
->i_hash
, head
);
1426 spin_unlock(&inode
->i_lock
);
1427 spin_unlock(&inode_hash_lock
);
1431 spin_unlock(&old
->i_lock
);
1432 spin_unlock(&inode_hash_lock
);
1434 if (unlikely(!inode_unhashed(old
))) {
1441 EXPORT_SYMBOL(insert_inode_locked4
);
1444 int generic_delete_inode(struct inode
*inode
)
1448 EXPORT_SYMBOL(generic_delete_inode
);
1451 * Called when we're dropping the last reference
1454 * Call the FS "drop_inode()" function, defaulting to
1455 * the legacy UNIX filesystem behaviour. If it tells
1456 * us to evict inode, do so. Otherwise, retain inode
1457 * in cache if fs is alive, sync and evict if fs is
1460 static void iput_final(struct inode
*inode
)
1462 struct super_block
*sb
= inode
->i_sb
;
1463 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1466 WARN_ON(inode
->i_state
& I_NEW
);
1469 drop
= op
->drop_inode(inode
);
1471 drop
= generic_drop_inode(inode
);
1473 #if defined(CONFIG_FMP_ECRYPT_FS)
1474 if (!drop
&& (sb
->s_flags
& MS_ACTIVE
) && !inode
->i_mapping
->use_fmp
) {
1476 if (!drop
&& (sb
->s_flags
& MS_ACTIVE
)) {
1478 inode
->i_state
|= I_REFERENCED
;
1479 inode_add_lru(inode
);
1480 spin_unlock(&inode
->i_lock
);
1485 inode
->i_state
|= I_WILL_FREE
;
1486 spin_unlock(&inode
->i_lock
);
1487 write_inode_now(inode
, 1);
1488 spin_lock(&inode
->i_lock
);
1489 WARN_ON(inode
->i_state
& I_NEW
);
1490 inode
->i_state
&= ~I_WILL_FREE
;
1493 inode
->i_state
|= I_FREEING
;
1494 if (!list_empty(&inode
->i_lru
))
1495 inode_lru_list_del(inode
);
1496 spin_unlock(&inode
->i_lock
);
1502 * iput - put an inode
1503 * @inode: inode to put
1505 * Puts an inode, dropping its usage count. If the inode use count hits
1506 * zero, the inode is then freed and may also be destroyed.
1508 * Consequently, iput() can sleep.
1510 void iput(struct inode
*inode
)
1514 BUG_ON(inode
->i_state
& I_CLEAR
);
1516 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
)) {
1517 if (inode
->i_nlink
&& (inode
->i_state
& I_DIRTY_TIME
)) {
1518 atomic_inc(&inode
->i_count
);
1519 inode
->i_state
&= ~I_DIRTY_TIME
;
1520 spin_unlock(&inode
->i_lock
);
1521 trace_writeback_lazytime_iput(inode
);
1522 mark_inode_dirty_sync(inode
);
1528 EXPORT_SYMBOL(iput
);
1531 * bmap - find a block number in a file
1532 * @inode: inode of file
1533 * @block: block to find
1535 * Returns the block number on the device holding the inode that
1536 * is the disk block number for the block of the file requested.
1537 * That is, asked for block 4 of inode 1 the function will return the
1538 * disk block relative to the disk start that holds that block of the
1541 sector_t
bmap(struct inode
*inode
, sector_t block
)
1544 if (inode
->i_mapping
->a_ops
->bmap
)
1545 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1548 EXPORT_SYMBOL(bmap
);
1551 * With relative atime, only update atime if the previous atime is
1552 * earlier than either the ctime or mtime or if at least a day has
1553 * passed since the last atime update.
1555 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1556 struct timespec now
)
1559 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1562 * Is mtime younger than atime? If yes, update atime:
1564 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1567 * Is ctime younger than atime? If yes, update atime:
1569 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1573 * Is the previous atime value older than a day? If yes,
1576 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1579 * Good, we can skip the atime update:
1584 int generic_update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1586 int iflags
= I_DIRTY_TIME
;
1588 if (flags
& S_ATIME
)
1589 inode
->i_atime
= *time
;
1590 if (flags
& S_VERSION
)
1591 inode_inc_iversion(inode
);
1592 if (flags
& S_CTIME
)
1593 inode
->i_ctime
= *time
;
1594 if (flags
& S_MTIME
)
1595 inode
->i_mtime
= *time
;
1597 if (!(inode
->i_sb
->s_flags
& MS_LAZYTIME
) || (flags
& S_VERSION
))
1598 iflags
|= I_DIRTY_SYNC
;
1599 __mark_inode_dirty(inode
, iflags
);
1602 EXPORT_SYMBOL(generic_update_time
);
1605 * This does the actual work of updating an inodes time or version. Must have
1606 * had called mnt_want_write() before calling this.
1608 static int update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1610 int (*update_time
)(struct inode
*, struct timespec
*, int);
1612 update_time
= inode
->i_op
->update_time
? inode
->i_op
->update_time
:
1613 generic_update_time
;
1615 return update_time(inode
, time
, flags
);
1619 * touch_atime - update the access time
1620 * @path: the &struct path to update
1621 * @inode: inode to update
1623 * Update the accessed time on an inode and mark it for writeback.
1624 * This function automatically handles read only file systems and media,
1625 * as well as the "noatime" flag and inode specific "noatime" markers.
1627 bool atime_needs_update(const struct path
*path
, struct inode
*inode
)
1629 struct vfsmount
*mnt
= path
->mnt
;
1630 struct timespec now
;
1632 if (inode
->i_flags
& S_NOATIME
)
1634 if (IS_NOATIME(inode
))
1636 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1639 if (mnt
->mnt_flags
& MNT_NOATIME
)
1641 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1644 now
= current_fs_time(inode
->i_sb
);
1646 if (!relatime_need_update(mnt
, inode
, now
))
1649 if (timespec_equal(&inode
->i_atime
, &now
))
1655 void touch_atime(const struct path
*path
)
1657 struct vfsmount
*mnt
= path
->mnt
;
1658 struct inode
*inode
= d_inode(path
->dentry
);
1659 struct timespec now
;
1661 if (!atime_needs_update(path
, inode
))
1664 if (!sb_start_write_trylock(inode
->i_sb
))
1667 if (__mnt_want_write(mnt
) != 0)
1670 * File systems can error out when updating inodes if they need to
1671 * allocate new space to modify an inode (such is the case for
1672 * Btrfs), but since we touch atime while walking down the path we
1673 * really don't care if we failed to update the atime of the file,
1674 * so just ignore the return value.
1675 * We may also fail on filesystems that have the ability to make parts
1676 * of the fs read only, e.g. subvolumes in Btrfs.
1678 now
= current_fs_time(inode
->i_sb
);
1679 update_time(inode
, &now
, S_ATIME
);
1680 __mnt_drop_write(mnt
);
1682 sb_end_write(inode
->i_sb
);
1684 EXPORT_SYMBOL(touch_atime
);
1687 * The logic we want is
1689 * if suid or (sgid and xgrp)
1692 int should_remove_suid(struct dentry
*dentry
)
1694 umode_t mode
= d_inode(dentry
)->i_mode
;
1697 /* suid always must be killed */
1698 if (unlikely(mode
& S_ISUID
))
1699 kill
= ATTR_KILL_SUID
;
1702 * sgid without any exec bits is just a mandatory locking mark; leave
1703 * it alone. If some exec bits are set, it's a real sgid; kill it.
1705 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1706 kill
|= ATTR_KILL_SGID
;
1708 if (unlikely(kill
&& !capable(CAP_FSETID
) && S_ISREG(mode
)))
1713 EXPORT_SYMBOL(should_remove_suid
);
1716 * Return mask of changes for notify_change() that need to be done as a
1717 * response to write or truncate. Return 0 if nothing has to be changed.
1718 * Negative value on error (change should be denied).
1720 int dentry_needs_remove_privs(struct dentry
*dentry
)
1722 struct inode
*inode
= d_inode(dentry
);
1726 if (IS_NOSEC(inode
))
1729 mask
= should_remove_suid(dentry
);
1730 ret
= security_inode_need_killpriv(dentry
);
1734 mask
|= ATTR_KILL_PRIV
;
1737 EXPORT_SYMBOL(dentry_needs_remove_privs
);
1739 static int __remove_privs(struct vfsmount
*mnt
, struct dentry
*dentry
, int kill
)
1741 struct iattr newattrs
;
1743 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1745 * Note we call this on write, so notify_change will not
1746 * encounter any conflicting delegations:
1748 return notify_change2(mnt
, dentry
, &newattrs
, NULL
);
1752 * Remove special file priviledges (suid, capabilities) when file is written
1755 int file_remove_privs(struct file
*file
)
1757 struct dentry
*dentry
= file_dentry(file
);
1758 struct inode
*inode
= file_inode(file
);
1762 /* Fast path for nothing security related */
1763 if (IS_NOSEC(inode
))
1766 kill
= dentry_needs_remove_privs(dentry
);
1770 error
= __remove_privs(file
->f_path
.mnt
, dentry
, kill
);
1772 inode_has_no_xattr(inode
);
1776 EXPORT_SYMBOL(file_remove_privs
);
1779 * file_update_time - update mtime and ctime time
1780 * @file: file accessed
1782 * Update the mtime and ctime members of an inode and mark the inode
1783 * for writeback. Note that this function is meant exclusively for
1784 * usage in the file write path of filesystems, and filesystems may
1785 * choose to explicitly ignore update via this function with the
1786 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1787 * timestamps are handled by the server. This can return an error for
1788 * file systems who need to allocate space in order to update an inode.
1791 int file_update_time(struct file
*file
)
1793 struct inode
*inode
= file_inode(file
);
1794 struct timespec now
;
1798 /* First try to exhaust all avenues to not sync */
1799 if (IS_NOCMTIME(inode
))
1802 now
= current_fs_time(inode
->i_sb
);
1803 if (!timespec_equal(&inode
->i_mtime
, &now
))
1806 if (!timespec_equal(&inode
->i_ctime
, &now
))
1809 if (IS_I_VERSION(inode
))
1810 sync_it
|= S_VERSION
;
1815 /* Finally allowed to write? Takes lock. */
1816 if (__mnt_want_write_file(file
))
1819 ret
= update_time(inode
, &now
, sync_it
);
1820 __mnt_drop_write_file(file
);
1824 EXPORT_SYMBOL(file_update_time
);
1826 int inode_needs_sync(struct inode
*inode
)
1830 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1834 EXPORT_SYMBOL(inode_needs_sync
);
1837 * If we try to find an inode in the inode hash while it is being
1838 * deleted, we have to wait until the filesystem completes its
1839 * deletion before reporting that it isn't found. This function waits
1840 * until the deletion _might_ have completed. Callers are responsible
1841 * to recheck inode state.
1843 * It doesn't matter if I_NEW is not set initially, a call to
1844 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1847 static void __wait_on_freeing_inode(struct inode
*inode
)
1849 wait_queue_head_t
*wq
;
1850 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1851 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1852 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1853 spin_unlock(&inode
->i_lock
);
1854 spin_unlock(&inode_hash_lock
);
1856 finish_wait(wq
, &wait
.wait
);
1857 spin_lock(&inode_hash_lock
);
1860 static __initdata
unsigned long ihash_entries
;
1861 static int __init
set_ihash_entries(char *str
)
1865 ihash_entries
= simple_strtoul(str
, &str
, 0);
1868 __setup("ihash_entries=", set_ihash_entries
);
1871 * Initialize the waitqueues and inode hash table.
1873 void __init
inode_init_early(void)
1877 /* If hashes are distributed across NUMA nodes, defer
1878 * hash allocation until vmalloc space is available.
1884 alloc_large_system_hash("Inode-cache",
1885 sizeof(struct hlist_head
),
1894 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1895 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1898 void __init
inode_init(void)
1902 /* inode slab cache */
1903 inode_cachep
= kmem_cache_create("inode_cache",
1904 sizeof(struct inode
),
1906 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1910 /* Hash may have been set up in inode_init_early */
1915 alloc_large_system_hash("Inode-cache",
1916 sizeof(struct hlist_head
),
1925 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1926 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1929 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1931 inode
->i_mode
= mode
;
1932 if (S_ISCHR(mode
)) {
1933 inode
->i_fop
= &def_chr_fops
;
1934 inode
->i_rdev
= rdev
;
1935 } else if (S_ISBLK(mode
)) {
1936 inode
->i_fop
= &def_blk_fops
;
1937 inode
->i_rdev
= rdev
;
1938 } else if (S_ISFIFO(mode
))
1939 inode
->i_fop
= &pipefifo_fops
;
1940 else if (S_ISSOCK(mode
))
1941 ; /* leave it no_open_fops */
1943 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1944 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1947 EXPORT_SYMBOL(init_special_inode
);
1950 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1952 * @dir: Directory inode
1953 * @mode: mode of the new inode
1955 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1958 inode
->i_uid
= current_fsuid();
1959 if (dir
&& dir
->i_mode
& S_ISGID
) {
1960 inode
->i_gid
= dir
->i_gid
;
1962 /* Directories are special, and always inherit S_ISGID */
1965 else if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
) &&
1966 !in_group_p(inode
->i_gid
) &&
1967 !capable_wrt_inode_uidgid(dir
, CAP_FSETID
))
1970 inode
->i_gid
= current_fsgid();
1971 inode
->i_mode
= mode
;
1973 EXPORT_SYMBOL(inode_init_owner
);
1976 * inode_owner_or_capable - check current task permissions to inode
1977 * @inode: inode being checked
1979 * Return true if current either has CAP_FOWNER in a namespace with the
1980 * inode owner uid mapped, or owns the file.
1982 bool inode_owner_or_capable(const struct inode
*inode
)
1984 struct user_namespace
*ns
;
1986 if (uid_eq(current_fsuid(), inode
->i_uid
))
1989 ns
= current_user_ns();
1990 if (ns_capable(ns
, CAP_FOWNER
) && kuid_has_mapping(ns
, inode
->i_uid
))
1994 EXPORT_SYMBOL(inode_owner_or_capable
);
1997 * Direct i/o helper functions
1999 static void __inode_dio_wait(struct inode
*inode
)
2001 wait_queue_head_t
*wq
= bit_waitqueue(&inode
->i_state
, __I_DIO_WAKEUP
);
2002 DEFINE_WAIT_BIT(q
, &inode
->i_state
, __I_DIO_WAKEUP
);
2005 prepare_to_wait(wq
, &q
.wait
, TASK_UNINTERRUPTIBLE
);
2006 if (atomic_read(&inode
->i_dio_count
))
2008 } while (atomic_read(&inode
->i_dio_count
));
2009 finish_wait(wq
, &q
.wait
);
2013 * inode_dio_wait - wait for outstanding DIO requests to finish
2014 * @inode: inode to wait for
2016 * Waits for all pending direct I/O requests to finish so that we can
2017 * proceed with a truncate or equivalent operation.
2019 * Must be called under a lock that serializes taking new references
2020 * to i_dio_count, usually by inode->i_mutex.
2022 void inode_dio_wait(struct inode
*inode
)
2024 if (atomic_read(&inode
->i_dio_count
))
2025 __inode_dio_wait(inode
);
2027 EXPORT_SYMBOL(inode_dio_wait
);
2030 * inode_set_flags - atomically set some inode flags
2032 * Note: the caller should be holding i_mutex, or else be sure that
2033 * they have exclusive access to the inode structure (i.e., while the
2034 * inode is being instantiated). The reason for the cmpxchg() loop
2035 * --- which wouldn't be necessary if all code paths which modify
2036 * i_flags actually followed this rule, is that there is at least one
2037 * code path which doesn't today so we use cmpxchg() out of an abundance
2040 * In the long run, i_mutex is overkill, and we should probably look
2041 * at using the i_lock spinlock to protect i_flags, and then make sure
2042 * it is so documented in include/linux/fs.h and that all code follows
2043 * the locking convention!!
2045 void inode_set_flags(struct inode
*inode
, unsigned int flags
,
2048 unsigned int old_flags
, new_flags
;
2050 WARN_ON_ONCE(flags
& ~mask
);
2052 old_flags
= ACCESS_ONCE(inode
->i_flags
);
2053 new_flags
= (old_flags
& ~mask
) | flags
;
2054 } while (unlikely(cmpxchg(&inode
->i_flags
, old_flags
,
2055 new_flags
) != old_flags
));
2057 EXPORT_SYMBOL(inode_set_flags
);