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>
23 * Inode locking rules:
25 * inode->i_lock protects:
26 * inode->i_state, inode->i_hash, __iget()
27 * inode->i_sb->s_inode_lru_lock protects:
28 * inode->i_sb->s_inode_lru, inode->i_lru
29 * inode_sb_list_lock protects:
30 * sb->s_inodes, inode->i_sb_list
31 * bdi->wb.list_lock protects:
32 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
33 * inode_hash_lock protects:
34 * inode_hashtable, inode->i_hash
40 * inode->i_sb->s_inode_lru_lock
53 static unsigned int i_hash_mask __read_mostly
;
54 static unsigned int i_hash_shift __read_mostly
;
55 static struct hlist_head
*inode_hashtable __read_mostly
;
56 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_hash_lock
);
58 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_sb_list_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 int, nr_inodes
);
74 static DEFINE_PER_CPU(unsigned int, nr_unused
);
76 static struct kmem_cache
*inode_cachep __read_mostly
;
78 static int 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 int 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 int get_nr_dirty_inodes(void)
98 /* not actually dirty inodes, but a wild approximation */
99 int 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(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_dointvec(table
, write
, buffer
, lenp
, ppos
);
117 * inode_init_always - perform inode structure intialisation
118 * @sb: superblock inode belongs to
119 * @inode: inode to initialise
121 * These are initializations that need to be done on every inode
122 * allocation as the fields are not initialised by slab allocation.
124 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
126 static const struct inode_operations empty_iops
;
127 static const struct file_operations empty_fops
;
128 struct address_space
*const mapping
= &inode
->i_data
;
131 inode
->i_blkbits
= sb
->s_blocksize_bits
;
133 atomic_set(&inode
->i_count
, 1);
134 inode
->i_op
= &empty_iops
;
135 inode
->i_fop
= &empty_fops
;
136 inode
->__i_nlink
= 1;
137 inode
->i_opflags
= 0;
138 i_uid_write(inode
, 0);
139 i_gid_write(inode
, 0);
140 atomic_set(&inode
->i_writecount
, 0);
144 inode
->i_generation
= 0;
146 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
148 inode
->i_pipe
= NULL
;
149 inode
->i_bdev
= NULL
;
150 inode
->i_cdev
= NULL
;
152 inode
->dirtied_when
= 0;
154 if (security_inode_alloc(inode
))
156 spin_lock_init(&inode
->i_lock
);
157 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
159 mutex_init(&inode
->i_mutex
);
160 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
162 atomic_set(&inode
->i_dio_count
, 0);
164 mapping
->a_ops
= &empty_aops
;
165 mapping
->host
= inode
;
167 atomic_set(&mapping
->i_mmap_writable
, 0);
168 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
169 mapping
->private_data
= NULL
;
170 mapping
->backing_dev_info
= &default_backing_dev_info
;
171 mapping
->writeback_index
= 0;
172 #if defined(CONFIG_MMC_DW_FMP_ECRYPT_FS) || defined(CONFIG_UFS_FMP_ECRYPT_FS)
175 mapping
->key_length
= 0;
177 mapping
->sensitive_data_index
= 0;
178 mapping
->hash_tfm
= NULL
;
179 #ifdef CONFIG_CRYPTO_FIPS
180 mapping
->cc_enable
= 0;
188 * If the block_device provides a backing_dev_info for client
189 * inodes then use that. Otherwise the inode share the bdev's
193 struct backing_dev_info
*bdi
;
195 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
196 mapping
->backing_dev_info
= bdi
;
198 inode
->i_private
= NULL
;
199 inode
->i_mapping
= mapping
;
200 INIT_HLIST_HEAD(&inode
->i_dentry
); /* buggered by rcu freeing */
201 #ifdef CONFIG_FS_POSIX_ACL
202 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
205 #ifdef CONFIG_FSNOTIFY
206 inode
->i_fsnotify_mask
= 0;
209 this_cpu_inc(nr_inodes
);
215 EXPORT_SYMBOL(inode_init_always
);
217 static struct inode
*alloc_inode(struct super_block
*sb
)
221 if (sb
->s_op
->alloc_inode
)
222 inode
= sb
->s_op
->alloc_inode(sb
);
224 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
229 if (unlikely(inode_init_always(sb
, inode
))) {
230 if (inode
->i_sb
->s_op
->destroy_inode
)
231 inode
->i_sb
->s_op
->destroy_inode(inode
);
233 kmem_cache_free(inode_cachep
, inode
);
240 void free_inode_nonrcu(struct inode
*inode
)
242 kmem_cache_free(inode_cachep
, inode
);
244 EXPORT_SYMBOL(free_inode_nonrcu
);
246 void __destroy_inode(struct inode
*inode
)
248 BUG_ON(inode_has_buffers(inode
));
249 security_inode_free(inode
);
250 fsnotify_inode_delete(inode
);
251 if (!inode
->i_nlink
) {
252 WARN_ON(atomic_long_read(&inode
->i_sb
->s_remove_count
) == 0);
253 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
256 #ifdef CONFIG_FS_POSIX_ACL
257 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
258 posix_acl_release(inode
->i_acl
);
259 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
260 posix_acl_release(inode
->i_default_acl
);
262 this_cpu_dec(nr_inodes
);
264 EXPORT_SYMBOL(__destroy_inode
);
266 static void i_callback(struct rcu_head
*head
)
268 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
269 kmem_cache_free(inode_cachep
, inode
);
272 static void destroy_inode(struct inode
*inode
)
274 BUG_ON(!list_empty(&inode
->i_lru
));
275 __destroy_inode(inode
);
276 if (inode
->i_sb
->s_op
->destroy_inode
)
277 inode
->i_sb
->s_op
->destroy_inode(inode
);
279 call_rcu(&inode
->i_rcu
, i_callback
);
283 * drop_nlink - directly drop an inode's link count
286 * This is a low-level filesystem helper to replace any
287 * direct filesystem manipulation of i_nlink. In cases
288 * where we are attempting to track writes to the
289 * filesystem, a decrement to zero means an imminent
290 * write when the file is truncated and actually unlinked
293 void drop_nlink(struct inode
*inode
)
295 WARN_ON(inode
->i_nlink
== 0);
298 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
300 EXPORT_SYMBOL(drop_nlink
);
303 * clear_nlink - directly zero an inode's link count
306 * This is a low-level filesystem helper to replace any
307 * direct filesystem manipulation of i_nlink. See
308 * drop_nlink() for why we care about i_nlink hitting zero.
310 void clear_nlink(struct inode
*inode
)
312 if (inode
->i_nlink
) {
313 inode
->__i_nlink
= 0;
314 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
317 EXPORT_SYMBOL(clear_nlink
);
320 * set_nlink - directly set an inode's link count
322 * @nlink: new nlink (should be non-zero)
324 * This is a low-level filesystem helper to replace any
325 * direct filesystem manipulation of i_nlink.
327 void set_nlink(struct inode
*inode
, unsigned int nlink
)
332 /* Yes, some filesystems do change nlink from zero to one */
333 if (inode
->i_nlink
== 0)
334 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
336 inode
->__i_nlink
= nlink
;
339 EXPORT_SYMBOL(set_nlink
);
342 * inc_nlink - directly increment an inode's link count
345 * This is a low-level filesystem helper to replace any
346 * direct filesystem manipulation of i_nlink. Currently,
347 * it is only here for parity with dec_nlink().
349 void inc_nlink(struct inode
*inode
)
351 if (WARN_ON(inode
->i_nlink
== 0))
352 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
356 EXPORT_SYMBOL(inc_nlink
);
358 void address_space_init_once(struct address_space
*mapping
)
360 memset(mapping
, 0, sizeof(*mapping
));
361 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
362 spin_lock_init(&mapping
->tree_lock
);
363 mutex_init(&mapping
->i_mmap_mutex
);
364 INIT_LIST_HEAD(&mapping
->private_list
);
365 spin_lock_init(&mapping
->private_lock
);
366 mapping
->i_mmap
= RB_ROOT
;
367 INIT_LIST_HEAD(&mapping
->i_mmap_nonlinear
);
369 EXPORT_SYMBOL(address_space_init_once
);
372 * These are initializations that only need to be done
373 * once, because the fields are idempotent across use
374 * of the inode, so let the slab aware of that.
376 void inode_init_once(struct inode
*inode
)
378 memset(inode
, 0, sizeof(*inode
));
379 INIT_HLIST_NODE(&inode
->i_hash
);
380 INIT_LIST_HEAD(&inode
->i_devices
);
381 INIT_LIST_HEAD(&inode
->i_wb_list
);
382 INIT_LIST_HEAD(&inode
->i_lru
);
383 address_space_init_once(&inode
->i_data
);
384 i_size_ordered_init(inode
);
385 #ifdef CONFIG_FSNOTIFY
386 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
389 EXPORT_SYMBOL(inode_init_once
);
391 static void init_once(void *foo
)
393 struct inode
*inode
= (struct inode
*) foo
;
395 inode_init_once(inode
);
399 * inode->i_lock must be held
401 void __iget(struct inode
*inode
)
403 atomic_inc(&inode
->i_count
);
407 * get additional reference to inode; caller must already hold one.
409 void ihold(struct inode
*inode
)
411 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
413 EXPORT_SYMBOL(ihold
);
415 static void inode_lru_list_add(struct inode
*inode
)
417 spin_lock(&inode
->i_sb
->s_inode_lru_lock
);
418 if (list_empty(&inode
->i_lru
)) {
419 list_add(&inode
->i_lru
, &inode
->i_sb
->s_inode_lru
);
420 inode
->i_sb
->s_nr_inodes_unused
++;
421 this_cpu_inc(nr_unused
);
423 spin_unlock(&inode
->i_sb
->s_inode_lru_lock
);
427 * Add inode to LRU if needed (inode is unused and clean).
429 * Needs inode->i_lock held.
431 void inode_add_lru(struct inode
*inode
)
433 if (!(inode
->i_state
& (I_DIRTY
| I_SYNC
| I_FREEING
| I_WILL_FREE
)) &&
434 !atomic_read(&inode
->i_count
) && inode
->i_sb
->s_flags
& MS_ACTIVE
)
435 inode_lru_list_add(inode
);
439 static void inode_lru_list_del(struct inode
*inode
)
441 spin_lock(&inode
->i_sb
->s_inode_lru_lock
);
442 if (!list_empty(&inode
->i_lru
)) {
443 list_del_init(&inode
->i_lru
);
444 inode
->i_sb
->s_nr_inodes_unused
--;
445 this_cpu_dec(nr_unused
);
447 spin_unlock(&inode
->i_sb
->s_inode_lru_lock
);
451 * inode_sb_list_add - add inode to the superblock list of inodes
452 * @inode: inode to add
454 void inode_sb_list_add(struct inode
*inode
)
456 spin_lock(&inode_sb_list_lock
);
457 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
458 spin_unlock(&inode_sb_list_lock
);
460 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
462 static inline void inode_sb_list_del(struct inode
*inode
)
464 if (!list_empty(&inode
->i_sb_list
)) {
465 spin_lock(&inode_sb_list_lock
);
466 list_del_init(&inode
->i_sb_list
);
467 spin_unlock(&inode_sb_list_lock
);
471 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
475 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
477 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> i_hash_shift
);
478 return tmp
& i_hash_mask
;
482 * __insert_inode_hash - hash an inode
483 * @inode: unhashed inode
484 * @hashval: unsigned long value used to locate this object in the
487 * Add an inode to the inode hash for this superblock.
489 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
491 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
493 spin_lock(&inode_hash_lock
);
494 spin_lock(&inode
->i_lock
);
495 hlist_add_head(&inode
->i_hash
, b
);
496 spin_unlock(&inode
->i_lock
);
497 spin_unlock(&inode_hash_lock
);
499 EXPORT_SYMBOL(__insert_inode_hash
);
502 * __remove_inode_hash - remove an inode from the hash
503 * @inode: inode to unhash
505 * Remove an inode from the superblock.
507 void __remove_inode_hash(struct inode
*inode
)
509 spin_lock(&inode_hash_lock
);
510 spin_lock(&inode
->i_lock
);
511 hlist_del_init(&inode
->i_hash
);
512 spin_unlock(&inode
->i_lock
);
513 spin_unlock(&inode_hash_lock
);
515 EXPORT_SYMBOL(__remove_inode_hash
);
517 void clear_inode(struct inode
*inode
)
521 * We have to cycle tree_lock here because reclaim can be still in the
522 * process of removing the last page (in __delete_from_page_cache())
523 * and we must not free mapping under it.
525 spin_lock_irq(&inode
->i_data
.tree_lock
);
526 BUG_ON(inode
->i_data
.nrpages
);
527 BUG_ON(inode
->i_data
.nrshadows
);
528 spin_unlock_irq(&inode
->i_data
.tree_lock
);
529 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
530 BUG_ON(!(inode
->i_state
& I_FREEING
));
531 BUG_ON(inode
->i_state
& I_CLEAR
);
532 /* don't need i_lock here, no concurrent mods to i_state */
533 inode
->i_state
= I_FREEING
| I_CLEAR
;
535 EXPORT_SYMBOL(clear_inode
);
538 * Free the inode passed in, removing it from the lists it is still connected
539 * to. We remove any pages still attached to the inode and wait for any IO that
540 * is still in progress before finally destroying the inode.
542 * An inode must already be marked I_FREEING so that we avoid the inode being
543 * moved back onto lists if we race with other code that manipulates the lists
544 * (e.g. writeback_single_inode). The caller is responsible for setting this.
546 * An inode must already be removed from the LRU list before being evicted from
547 * the cache. This should occur atomically with setting the I_FREEING state
548 * flag, so no inodes here should ever be on the LRU when being evicted.
550 static void evict(struct inode
*inode
)
552 const struct super_operations
*op
= inode
->i_sb
->s_op
;
554 BUG_ON(!(inode
->i_state
& I_FREEING
));
555 BUG_ON(!list_empty(&inode
->i_lru
));
557 if (!list_empty(&inode
->i_wb_list
))
558 inode_wb_list_del(inode
);
560 inode_sb_list_del(inode
);
563 * Wait for flusher thread to be done with the inode so that filesystem
564 * does not start destroying it while writeback is still running. Since
565 * the inode has I_FREEING set, flusher thread won't start new work on
566 * the inode. We just have to wait for running writeback to finish.
568 inode_wait_for_writeback(inode
);
570 if (op
->evict_inode
) {
571 op
->evict_inode(inode
);
573 truncate_inode_pages_final(&inode
->i_data
);
576 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
578 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
581 remove_inode_hash(inode
);
583 spin_lock(&inode
->i_lock
);
584 wake_up_bit(&inode
->i_state
, __I_NEW
);
585 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
586 spin_unlock(&inode
->i_lock
);
588 destroy_inode(inode
);
592 * dispose_list - dispose of the contents of a local list
593 * @head: the head of the list to free
595 * Dispose-list gets a local list with local inodes in it, so it doesn't
596 * need to worry about list corruption and SMP locks.
598 static void dispose_list(struct list_head
*head
)
600 while (!list_empty(head
)) {
603 inode
= list_first_entry(head
, struct inode
, i_lru
);
604 list_del_init(&inode
->i_lru
);
611 * evict_inodes - evict all evictable inodes for a superblock
612 * @sb: superblock to operate on
614 * Make sure that no inodes with zero refcount are retained. This is
615 * called by superblock shutdown after having MS_ACTIVE flag removed,
616 * so any inode reaching zero refcount during or after that call will
617 * be immediately evicted.
619 void evict_inodes(struct super_block
*sb
)
621 struct inode
*inode
, *next
;
624 spin_lock(&inode_sb_list_lock
);
625 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
626 if (atomic_read(&inode
->i_count
))
629 spin_lock(&inode
->i_lock
);
630 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
631 spin_unlock(&inode
->i_lock
);
635 inode
->i_state
|= I_FREEING
;
636 inode_lru_list_del(inode
);
637 spin_unlock(&inode
->i_lock
);
638 list_add(&inode
->i_lru
, &dispose
);
640 spin_unlock(&inode_sb_list_lock
);
642 dispose_list(&dispose
);
646 * invalidate_inodes - attempt to free all inodes on a superblock
647 * @sb: superblock to operate on
648 * @kill_dirty: flag to guide handling of dirty inodes
650 * Attempts to free all inodes for a given superblock. If there were any
651 * busy inodes return a non-zero value, else zero.
652 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
655 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
658 struct inode
*inode
, *next
;
661 spin_lock(&inode_sb_list_lock
);
662 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
663 spin_lock(&inode
->i_lock
);
664 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
665 spin_unlock(&inode
->i_lock
);
668 if (inode
->i_state
& I_DIRTY
&& !kill_dirty
) {
669 spin_unlock(&inode
->i_lock
);
673 if (atomic_read(&inode
->i_count
)) {
674 spin_unlock(&inode
->i_lock
);
679 inode
->i_state
|= I_FREEING
;
680 inode_lru_list_del(inode
);
681 spin_unlock(&inode
->i_lock
);
682 list_add(&inode
->i_lru
, &dispose
);
684 spin_unlock(&inode_sb_list_lock
);
686 dispose_list(&dispose
);
691 static int can_unuse(struct inode
*inode
)
693 if (inode
->i_state
& ~I_REFERENCED
)
695 if (inode_has_buffers(inode
))
697 if (atomic_read(&inode
->i_count
))
699 if (inode
->i_data
.nrpages
)
705 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
706 * This is called from the superblock shrinker function with a number of inodes
707 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
708 * then are freed outside inode_lock by dispose_list().
710 * Any inodes which are pinned purely because of attached pagecache have their
711 * pagecache removed. If the inode has metadata buffers attached to
712 * mapping->private_list then try to remove them.
714 * If the inode has the I_REFERENCED flag set, then it means that it has been
715 * used recently - the flag is set in iput_final(). When we encounter such an
716 * inode, clear the flag and move it to the back of the LRU so it gets another
717 * pass through the LRU before it gets reclaimed. This is necessary because of
718 * the fact we are doing lazy LRU updates to minimise lock contention so the
719 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
720 * with this flag set because they are the inodes that are out of order.
722 void prune_icache_sb(struct super_block
*sb
, int nr_to_scan
)
726 unsigned long reap
= 0;
728 spin_lock(&sb
->s_inode_lru_lock
);
729 for (nr_scanned
= nr_to_scan
; nr_scanned
>= 0; nr_scanned
--) {
732 if (list_empty(&sb
->s_inode_lru
))
735 inode
= list_entry(sb
->s_inode_lru
.prev
, struct inode
, i_lru
);
738 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
739 * so use a trylock. If we fail to get the lock, just move the
740 * inode to the back of the list so we don't spin on it.
742 if (!spin_trylock(&inode
->i_lock
)) {
743 list_move(&inode
->i_lru
, &sb
->s_inode_lru
);
748 * Referenced or dirty inodes are still in use. Give them
749 * another pass through the LRU as we canot reclaim them now.
751 if (atomic_read(&inode
->i_count
) ||
752 (inode
->i_state
& ~I_REFERENCED
)) {
753 list_del_init(&inode
->i_lru
);
754 spin_unlock(&inode
->i_lock
);
755 sb
->s_nr_inodes_unused
--;
756 this_cpu_dec(nr_unused
);
760 /* recently referenced inodes get one more pass */
761 if (inode
->i_state
& I_REFERENCED
) {
762 inode
->i_state
&= ~I_REFERENCED
;
763 list_move(&inode
->i_lru
, &sb
->s_inode_lru
);
764 spin_unlock(&inode
->i_lock
);
767 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
769 spin_unlock(&inode
->i_lock
);
770 spin_unlock(&sb
->s_inode_lru_lock
);
771 if (remove_inode_buffers(inode
))
772 reap
+= invalidate_mapping_pages(&inode
->i_data
,
775 spin_lock(&sb
->s_inode_lru_lock
);
777 if (inode
!= list_entry(sb
->s_inode_lru
.next
,
778 struct inode
, i_lru
))
779 continue; /* wrong inode or list_empty */
780 /* avoid lock inversions with trylock */
781 if (!spin_trylock(&inode
->i_lock
))
783 if (!can_unuse(inode
)) {
784 spin_unlock(&inode
->i_lock
);
788 WARN_ON(inode
->i_state
& I_NEW
);
789 inode
->i_state
|= I_FREEING
;
790 spin_unlock(&inode
->i_lock
);
792 list_move(&inode
->i_lru
, &freeable
);
793 sb
->s_nr_inodes_unused
--;
794 this_cpu_dec(nr_unused
);
796 if (current_is_kswapd())
797 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
799 __count_vm_events(PGINODESTEAL
, reap
);
800 spin_unlock(&sb
->s_inode_lru_lock
);
801 if (current
->reclaim_state
)
802 current
->reclaim_state
->reclaimed_slab
+= reap
;
804 dispose_list(&freeable
);
807 static void __wait_on_freeing_inode(struct inode
*inode
);
809 * Called with the inode lock held.
811 static struct inode
*find_inode(struct super_block
*sb
,
812 struct hlist_head
*head
,
813 int (*test
)(struct inode
*, void *),
816 struct inode
*inode
= NULL
;
819 hlist_for_each_entry(inode
, head
, i_hash
) {
820 spin_lock(&inode
->i_lock
);
821 if (inode
->i_sb
!= sb
) {
822 spin_unlock(&inode
->i_lock
);
825 if (!test(inode
, data
)) {
826 spin_unlock(&inode
->i_lock
);
829 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
830 __wait_on_freeing_inode(inode
);
834 spin_unlock(&inode
->i_lock
);
841 * find_inode_fast is the fast path version of find_inode, see the comment at
842 * iget_locked for details.
844 static struct inode
*find_inode_fast(struct super_block
*sb
,
845 struct hlist_head
*head
, unsigned long ino
)
847 struct inode
*inode
= NULL
;
850 hlist_for_each_entry(inode
, head
, i_hash
) {
851 spin_lock(&inode
->i_lock
);
852 if (inode
->i_ino
!= ino
) {
853 spin_unlock(&inode
->i_lock
);
856 if (inode
->i_sb
!= sb
) {
857 spin_unlock(&inode
->i_lock
);
860 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
861 __wait_on_freeing_inode(inode
);
865 spin_unlock(&inode
->i_lock
);
872 * Each cpu owns a range of LAST_INO_BATCH numbers.
873 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
874 * to renew the exhausted range.
876 * This does not significantly increase overflow rate because every CPU can
877 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
878 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
879 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
880 * overflow rate by 2x, which does not seem too significant.
882 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
883 * error if st_ino won't fit in target struct field. Use 32bit counter
884 * here to attempt to avoid that.
886 #define LAST_INO_BATCH 1024
887 static DEFINE_PER_CPU(unsigned int, last_ino
);
889 unsigned int get_next_ino(void)
891 unsigned int *p
= &get_cpu_var(last_ino
);
892 unsigned int res
= *p
;
895 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
896 static atomic_t shared_last_ino
;
897 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
899 res
= next
- LAST_INO_BATCH
;
904 put_cpu_var(last_ino
);
907 EXPORT_SYMBOL(get_next_ino
);
910 * new_inode_pseudo - obtain an inode
913 * Allocates a new inode for given superblock.
914 * Inode wont be chained in superblock s_inodes list
916 * - fs can't be unmount
917 * - quotas, fsnotify, writeback can't work
919 struct inode
*new_inode_pseudo(struct super_block
*sb
)
921 struct inode
*inode
= alloc_inode(sb
);
924 spin_lock(&inode
->i_lock
);
926 spin_unlock(&inode
->i_lock
);
927 INIT_LIST_HEAD(&inode
->i_sb_list
);
933 * new_inode - obtain an inode
936 * Allocates a new inode for given superblock. The default gfp_mask
937 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
938 * If HIGHMEM pages are unsuitable or it is known that pages allocated
939 * for the page cache are not reclaimable or migratable,
940 * mapping_set_gfp_mask() must be called with suitable flags on the
941 * newly created inode's mapping
944 struct inode
*new_inode(struct super_block
*sb
)
948 spin_lock_prefetch(&inode_sb_list_lock
);
950 inode
= new_inode_pseudo(sb
);
952 inode_sb_list_add(inode
);
955 EXPORT_SYMBOL(new_inode
);
957 #ifdef CONFIG_DEBUG_LOCK_ALLOC
958 void lockdep_annotate_inode_mutex_key(struct inode
*inode
)
960 if (S_ISDIR(inode
->i_mode
)) {
961 struct file_system_type
*type
= inode
->i_sb
->s_type
;
963 /* Set new key only if filesystem hasn't already changed it */
964 if (lockdep_match_class(&inode
->i_mutex
, &type
->i_mutex_key
)) {
966 * ensure nobody is actually holding i_mutex
968 mutex_destroy(&inode
->i_mutex
);
969 mutex_init(&inode
->i_mutex
);
970 lockdep_set_class(&inode
->i_mutex
,
971 &type
->i_mutex_dir_key
);
975 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key
);
979 * unlock_new_inode - clear the I_NEW state and wake up any waiters
980 * @inode: new inode to unlock
982 * Called when the inode is fully initialised to clear the new state of the
983 * inode and wake up anyone waiting for the inode to finish initialisation.
985 void unlock_new_inode(struct inode
*inode
)
987 lockdep_annotate_inode_mutex_key(inode
);
988 spin_lock(&inode
->i_lock
);
989 WARN_ON(!(inode
->i_state
& I_NEW
));
990 inode
->i_state
&= ~I_NEW
;
992 wake_up_bit(&inode
->i_state
, __I_NEW
);
993 spin_unlock(&inode
->i_lock
);
995 EXPORT_SYMBOL(unlock_new_inode
);
998 * iget5_locked - obtain an inode from a mounted file system
999 * @sb: super block of file system
1000 * @hashval: hash value (usually inode number) to get
1001 * @test: callback used for comparisons between inodes
1002 * @set: callback used to initialize a new struct inode
1003 * @data: opaque data pointer to pass to @test and @set
1005 * Search for the inode specified by @hashval and @data in the inode cache,
1006 * and if present it is return it with an increased reference count. This is
1007 * a generalized version of iget_locked() for file systems where the inode
1008 * number is not sufficient for unique identification of an inode.
1010 * If the inode is not in cache, allocate a new inode and return it locked,
1011 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1012 * before unlocking it via unlock_new_inode().
1014 * Note both @test and @set are called with the inode_hash_lock held, so can't
1017 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1018 int (*test
)(struct inode
*, void *),
1019 int (*set
)(struct inode
*, void *), void *data
)
1021 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1022 struct inode
*inode
;
1024 spin_lock(&inode_hash_lock
);
1025 inode
= find_inode(sb
, head
, test
, data
);
1026 spin_unlock(&inode_hash_lock
);
1029 wait_on_inode(inode
);
1033 inode
= alloc_inode(sb
);
1037 spin_lock(&inode_hash_lock
);
1038 /* We released the lock, so.. */
1039 old
= find_inode(sb
, head
, test
, data
);
1041 if (set(inode
, data
))
1044 spin_lock(&inode
->i_lock
);
1045 inode
->i_state
= I_NEW
;
1046 hlist_add_head(&inode
->i_hash
, head
);
1047 spin_unlock(&inode
->i_lock
);
1048 inode_sb_list_add(inode
);
1049 spin_unlock(&inode_hash_lock
);
1051 /* Return the locked inode with I_NEW set, the
1052 * caller is responsible for filling in the contents
1058 * Uhhuh, somebody else created the same inode under
1059 * us. Use the old inode instead of the one we just
1062 spin_unlock(&inode_hash_lock
);
1063 destroy_inode(inode
);
1065 wait_on_inode(inode
);
1070 spin_unlock(&inode_hash_lock
);
1071 destroy_inode(inode
);
1074 EXPORT_SYMBOL(iget5_locked
);
1077 * iget_locked - obtain an inode from a mounted file system
1078 * @sb: super block of file system
1079 * @ino: inode number to get
1081 * Search for the inode specified by @ino in the inode cache and if present
1082 * return it with an increased reference count. This is for file systems
1083 * where the inode number is sufficient for unique identification of an inode.
1085 * If the inode is not in cache, allocate a new inode and return it locked,
1086 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1087 * before unlocking it via unlock_new_inode().
1089 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1091 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1092 struct inode
*inode
;
1094 spin_lock(&inode_hash_lock
);
1095 inode
= find_inode_fast(sb
, head
, ino
);
1096 spin_unlock(&inode_hash_lock
);
1098 wait_on_inode(inode
);
1102 inode
= alloc_inode(sb
);
1106 spin_lock(&inode_hash_lock
);
1107 /* We released the lock, so.. */
1108 old
= find_inode_fast(sb
, head
, ino
);
1111 spin_lock(&inode
->i_lock
);
1112 inode
->i_state
= I_NEW
;
1113 hlist_add_head(&inode
->i_hash
, head
);
1114 spin_unlock(&inode
->i_lock
);
1115 inode_sb_list_add(inode
);
1116 spin_unlock(&inode_hash_lock
);
1118 /* Return the locked inode with I_NEW set, the
1119 * caller is responsible for filling in the contents
1125 * Uhhuh, somebody else created the same inode under
1126 * us. Use the old inode instead of the one we just
1129 spin_unlock(&inode_hash_lock
);
1130 destroy_inode(inode
);
1132 wait_on_inode(inode
);
1136 EXPORT_SYMBOL(iget_locked
);
1139 * search the inode cache for a matching inode number.
1140 * If we find one, then the inode number we are trying to
1141 * allocate is not unique and so we should not use it.
1143 * Returns 1 if the inode number is unique, 0 if it is not.
1145 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1147 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1148 struct inode
*inode
;
1150 spin_lock(&inode_hash_lock
);
1151 hlist_for_each_entry(inode
, b
, i_hash
) {
1152 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
) {
1153 spin_unlock(&inode_hash_lock
);
1157 spin_unlock(&inode_hash_lock
);
1163 * iunique - get a unique inode number
1165 * @max_reserved: highest reserved inode number
1167 * Obtain an inode number that is unique on the system for a given
1168 * superblock. This is used by file systems that have no natural
1169 * permanent inode numbering system. An inode number is returned that
1170 * is higher than the reserved limit but unique.
1173 * With a large number of inodes live on the file system this function
1174 * currently becomes quite slow.
1176 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1179 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1180 * error if st_ino won't fit in target struct field. Use 32bit counter
1181 * here to attempt to avoid that.
1183 static DEFINE_SPINLOCK(iunique_lock
);
1184 static unsigned int counter
;
1187 spin_lock(&iunique_lock
);
1189 if (counter
<= max_reserved
)
1190 counter
= max_reserved
+ 1;
1192 } while (!test_inode_iunique(sb
, res
));
1193 spin_unlock(&iunique_lock
);
1197 EXPORT_SYMBOL(iunique
);
1199 struct inode
*igrab(struct inode
*inode
)
1201 spin_lock(&inode
->i_lock
);
1202 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1204 spin_unlock(&inode
->i_lock
);
1206 spin_unlock(&inode
->i_lock
);
1208 * Handle the case where s_op->clear_inode is not been
1209 * called yet, and somebody is calling igrab
1210 * while the inode is getting freed.
1216 EXPORT_SYMBOL(igrab
);
1219 * ilookup5_nowait - search for an inode in the inode cache
1220 * @sb: super block of file system to search
1221 * @hashval: hash value (usually inode number) to search for
1222 * @test: callback used for comparisons between inodes
1223 * @data: opaque data pointer to pass to @test
1225 * Search for the inode specified by @hashval and @data in the inode cache.
1226 * If the inode is in the cache, the inode is returned with an incremented
1229 * Note: I_NEW is not waited upon so you have to be very careful what you do
1230 * with the returned inode. You probably should be using ilookup5() instead.
1232 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1234 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1235 int (*test
)(struct inode
*, void *), void *data
)
1237 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1238 struct inode
*inode
;
1240 spin_lock(&inode_hash_lock
);
1241 inode
= find_inode(sb
, head
, test
, data
);
1242 spin_unlock(&inode_hash_lock
);
1246 EXPORT_SYMBOL(ilookup5_nowait
);
1249 * ilookup5 - search for an inode in the inode cache
1250 * @sb: super block of file system to search
1251 * @hashval: hash value (usually inode number) to search for
1252 * @test: callback used for comparisons between inodes
1253 * @data: opaque data pointer to pass to @test
1255 * Search for the inode specified by @hashval and @data in the inode cache,
1256 * and if the inode is in the cache, return the inode with an incremented
1257 * reference count. Waits on I_NEW before returning the inode.
1258 * returned with an incremented reference count.
1260 * This is a generalized version of ilookup() for file systems where the
1261 * inode number is not sufficient for unique identification of an inode.
1263 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1265 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1266 int (*test
)(struct inode
*, void *), void *data
)
1268 struct inode
*inode
= ilookup5_nowait(sb
, hashval
, test
, data
);
1271 wait_on_inode(inode
);
1274 EXPORT_SYMBOL(ilookup5
);
1277 * ilookup - search for an inode in the inode cache
1278 * @sb: super block of file system to search
1279 * @ino: inode number to search for
1281 * Search for the inode @ino in the inode cache, and if the inode is in the
1282 * cache, the inode is returned with an incremented reference count.
1284 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1286 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1287 struct inode
*inode
;
1289 spin_lock(&inode_hash_lock
);
1290 inode
= find_inode_fast(sb
, head
, ino
);
1291 spin_unlock(&inode_hash_lock
);
1294 wait_on_inode(inode
);
1297 EXPORT_SYMBOL(ilookup
);
1299 int insert_inode_locked(struct inode
*inode
)
1301 struct super_block
*sb
= inode
->i_sb
;
1302 ino_t ino
= inode
->i_ino
;
1303 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1306 struct inode
*old
= NULL
;
1307 spin_lock(&inode_hash_lock
);
1308 hlist_for_each_entry(old
, head
, i_hash
) {
1309 if (old
->i_ino
!= ino
)
1311 if (old
->i_sb
!= sb
)
1313 spin_lock(&old
->i_lock
);
1314 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1315 spin_unlock(&old
->i_lock
);
1321 spin_lock(&inode
->i_lock
);
1322 inode
->i_state
|= I_NEW
;
1323 hlist_add_head(&inode
->i_hash
, head
);
1324 spin_unlock(&inode
->i_lock
);
1325 spin_unlock(&inode_hash_lock
);
1329 spin_unlock(&old
->i_lock
);
1330 spin_unlock(&inode_hash_lock
);
1332 if (unlikely(!inode_unhashed(old
))) {
1339 EXPORT_SYMBOL(insert_inode_locked
);
1341 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1342 int (*test
)(struct inode
*, void *), void *data
)
1344 struct super_block
*sb
= inode
->i_sb
;
1345 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1348 struct inode
*old
= NULL
;
1350 spin_lock(&inode_hash_lock
);
1351 hlist_for_each_entry(old
, head
, i_hash
) {
1352 if (old
->i_sb
!= sb
)
1354 if (!test(old
, data
))
1356 spin_lock(&old
->i_lock
);
1357 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1358 spin_unlock(&old
->i_lock
);
1364 spin_lock(&inode
->i_lock
);
1365 inode
->i_state
|= I_NEW
;
1366 hlist_add_head(&inode
->i_hash
, head
);
1367 spin_unlock(&inode
->i_lock
);
1368 spin_unlock(&inode_hash_lock
);
1372 spin_unlock(&old
->i_lock
);
1373 spin_unlock(&inode_hash_lock
);
1375 if (unlikely(!inode_unhashed(old
))) {
1382 EXPORT_SYMBOL(insert_inode_locked4
);
1385 int generic_delete_inode(struct inode
*inode
)
1389 EXPORT_SYMBOL(generic_delete_inode
);
1392 * Called when we're dropping the last reference
1395 * Call the FS "drop_inode()" function, defaulting to
1396 * the legacy UNIX filesystem behaviour. If it tells
1397 * us to evict inode, do so. Otherwise, retain inode
1398 * in cache if fs is alive, sync and evict if fs is
1401 static void iput_final(struct inode
*inode
)
1403 struct super_block
*sb
= inode
->i_sb
;
1404 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1407 WARN_ON(inode
->i_state
& I_NEW
);
1410 drop
= op
->drop_inode(inode
);
1412 drop
= generic_drop_inode(inode
);
1414 if (!drop
&& (sb
->s_flags
& MS_ACTIVE
)) {
1415 inode
->i_state
|= I_REFERENCED
;
1416 inode_add_lru(inode
);
1417 spin_unlock(&inode
->i_lock
);
1422 inode
->i_state
|= I_WILL_FREE
;
1423 spin_unlock(&inode
->i_lock
);
1424 write_inode_now(inode
, 1);
1425 spin_lock(&inode
->i_lock
);
1426 WARN_ON(inode
->i_state
& I_NEW
);
1427 inode
->i_state
&= ~I_WILL_FREE
;
1430 inode
->i_state
|= I_FREEING
;
1431 if (!list_empty(&inode
->i_lru
))
1432 inode_lru_list_del(inode
);
1433 spin_unlock(&inode
->i_lock
);
1439 * iput - put an inode
1440 * @inode: inode to put
1442 * Puts an inode, dropping its usage count. If the inode use count hits
1443 * zero, the inode is then freed and may also be destroyed.
1445 * Consequently, iput() can sleep.
1447 void iput(struct inode
*inode
)
1450 BUG_ON(inode
->i_state
& I_CLEAR
);
1452 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
))
1456 EXPORT_SYMBOL(iput
);
1459 * bmap - find a block number in a file
1460 * @inode: inode of file
1461 * @block: block to find
1463 * Returns the block number on the device holding the inode that
1464 * is the disk block number for the block of the file requested.
1465 * That is, asked for block 4 of inode 1 the function will return the
1466 * disk block relative to the disk start that holds that block of the
1469 sector_t
bmap(struct inode
*inode
, sector_t block
)
1472 if (inode
->i_mapping
->a_ops
->bmap
)
1473 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1476 EXPORT_SYMBOL(bmap
);
1479 * With relative atime, only update atime if the previous atime is
1480 * earlier than either the ctime or mtime or if at least a day has
1481 * passed since the last atime update.
1483 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1484 struct timespec now
)
1487 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1490 * Is mtime younger than atime? If yes, update atime:
1492 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1495 * Is ctime younger than atime? If yes, update atime:
1497 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1501 * Is the previous atime value older than a day? If yes,
1504 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1507 * Good, we can skip the atime update:
1513 * This does the actual work of updating an inodes time or version. Must have
1514 * had called mnt_want_write() before calling this.
1516 static int update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1518 if (inode
->i_op
->update_time
)
1519 return inode
->i_op
->update_time(inode
, time
, flags
);
1521 if (flags
& S_ATIME
)
1522 inode
->i_atime
= *time
;
1523 if (flags
& S_VERSION
)
1524 inode_inc_iversion(inode
);
1525 if (flags
& S_CTIME
)
1526 inode
->i_ctime
= *time
;
1527 if (flags
& S_MTIME
)
1528 inode
->i_mtime
= *time
;
1529 mark_inode_dirty_sync(inode
);
1534 * touch_atime - update the access time
1535 * @path: the &struct path to update
1537 * Update the accessed time on an inode and mark it for writeback.
1538 * This function automatically handles read only file systems and media,
1539 * as well as the "noatime" flag and inode specific "noatime" markers.
1541 void touch_atime(struct path
*path
)
1543 struct vfsmount
*mnt
= path
->mnt
;
1544 struct inode
*inode
= path
->dentry
->d_inode
;
1545 struct timespec now
;
1547 if (inode
->i_flags
& S_NOATIME
)
1549 if (IS_NOATIME(inode
))
1551 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1554 if (mnt
->mnt_flags
& MNT_NOATIME
)
1556 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1559 now
= current_fs_time(inode
->i_sb
);
1561 if (!relatime_need_update(mnt
, inode
, now
))
1564 if (timespec_equal(&inode
->i_atime
, &now
))
1567 if (!sb_start_write_trylock(inode
->i_sb
))
1570 if (__mnt_want_write(mnt
))
1573 * File systems can error out when updating inodes if they need to
1574 * allocate new space to modify an inode (such is the case for
1575 * Btrfs), but since we touch atime while walking down the path we
1576 * really don't care if we failed to update the atime of the file,
1577 * so just ignore the return value.
1578 * We may also fail on filesystems that have the ability to make parts
1579 * of the fs read only, e.g. subvolumes in Btrfs.
1581 update_time(inode
, &now
, S_ATIME
);
1582 __mnt_drop_write(mnt
);
1584 sb_end_write(inode
->i_sb
);
1586 EXPORT_SYMBOL(touch_atime
);
1589 * The logic we want is
1591 * if suid or (sgid and xgrp)
1594 int should_remove_suid(struct dentry
*dentry
)
1596 umode_t mode
= dentry
->d_inode
->i_mode
;
1599 /* suid always must be killed */
1600 if (unlikely(mode
& S_ISUID
))
1601 kill
= ATTR_KILL_SUID
;
1604 * sgid without any exec bits is just a mandatory locking mark; leave
1605 * it alone. If some exec bits are set, it's a real sgid; kill it.
1607 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1608 kill
|= ATTR_KILL_SGID
;
1610 if (unlikely(kill
&& !capable(CAP_FSETID
) && S_ISREG(mode
)))
1615 EXPORT_SYMBOL(should_remove_suid
);
1617 static int __remove_suid(struct vfsmount
*mnt
, struct dentry
*dentry
, int kill
)
1619 struct iattr newattrs
;
1621 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1622 return notify_change2(mnt
, dentry
, &newattrs
);
1625 int file_remove_suid(struct file
*file
)
1627 struct dentry
*dentry
= file
->f_path
.dentry
;
1628 struct inode
*inode
= dentry
->d_inode
;
1633 /* Fast path for nothing security related */
1634 if (IS_NOSEC(inode
))
1637 killsuid
= should_remove_suid(dentry
);
1638 killpriv
= security_inode_need_killpriv(dentry
);
1643 error
= security_inode_killpriv(dentry
);
1644 if (!error
&& killsuid
)
1645 error
= __remove_suid(file
->f_path
.mnt
, dentry
, killsuid
);
1647 inode_has_no_xattr(inode
);
1651 EXPORT_SYMBOL(file_remove_suid
);
1654 * file_update_time - update mtime and ctime time
1655 * @file: file accessed
1657 * Update the mtime and ctime members of an inode and mark the inode
1658 * for writeback. Note that this function is meant exclusively for
1659 * usage in the file write path of filesystems, and filesystems may
1660 * choose to explicitly ignore update via this function with the
1661 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1662 * timestamps are handled by the server. This can return an error for
1663 * file systems who need to allocate space in order to update an inode.
1666 int file_update_time(struct file
*file
)
1668 struct inode
*inode
= file_inode(file
);
1669 struct timespec now
;
1673 /* First try to exhaust all avenues to not sync */
1674 if (IS_NOCMTIME(inode
))
1677 now
= current_fs_time(inode
->i_sb
);
1678 if (!timespec_equal(&inode
->i_mtime
, &now
))
1681 if (!timespec_equal(&inode
->i_ctime
, &now
))
1684 if (IS_I_VERSION(inode
))
1685 sync_it
|= S_VERSION
;
1690 /* Finally allowed to write? Takes lock. */
1691 if (__mnt_want_write_file(file
))
1694 ret
= update_time(inode
, &now
, sync_it
);
1695 __mnt_drop_write_file(file
);
1699 EXPORT_SYMBOL(file_update_time
);
1701 int inode_needs_sync(struct inode
*inode
)
1705 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1709 EXPORT_SYMBOL(inode_needs_sync
);
1711 int inode_wait(void *word
)
1716 EXPORT_SYMBOL(inode_wait
);
1719 * If we try to find an inode in the inode hash while it is being
1720 * deleted, we have to wait until the filesystem completes its
1721 * deletion before reporting that it isn't found. This function waits
1722 * until the deletion _might_ have completed. Callers are responsible
1723 * to recheck inode state.
1725 * It doesn't matter if I_NEW is not set initially, a call to
1726 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1729 static void __wait_on_freeing_inode(struct inode
*inode
)
1731 wait_queue_head_t
*wq
;
1732 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1733 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1734 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1735 spin_unlock(&inode
->i_lock
);
1736 spin_unlock(&inode_hash_lock
);
1738 finish_wait(wq
, &wait
.wait
);
1739 spin_lock(&inode_hash_lock
);
1742 static __initdata
unsigned long ihash_entries
;
1743 static int __init
set_ihash_entries(char *str
)
1747 ihash_entries
= simple_strtoul(str
, &str
, 0);
1750 __setup("ihash_entries=", set_ihash_entries
);
1753 * Initialize the waitqueues and inode hash table.
1755 void __init
inode_init_early(void)
1759 /* If hashes are distributed across NUMA nodes, defer
1760 * hash allocation until vmalloc space is available.
1766 alloc_large_system_hash("Inode-cache",
1767 sizeof(struct hlist_head
),
1776 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1777 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1780 void __init
inode_init(void)
1784 /* inode slab cache */
1785 inode_cachep
= kmem_cache_create("inode_cache",
1786 sizeof(struct inode
),
1788 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1792 /* Hash may have been set up in inode_init_early */
1797 alloc_large_system_hash("Inode-cache",
1798 sizeof(struct hlist_head
),
1807 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1808 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1811 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1813 inode
->i_mode
= mode
;
1814 if (S_ISCHR(mode
)) {
1815 inode
->i_fop
= &def_chr_fops
;
1816 inode
->i_rdev
= rdev
;
1817 } else if (S_ISBLK(mode
)) {
1818 inode
->i_fop
= &def_blk_fops
;
1819 inode
->i_rdev
= rdev
;
1820 } else if (S_ISFIFO(mode
))
1821 inode
->i_fop
= &pipefifo_fops
;
1822 else if (S_ISSOCK(mode
))
1823 inode
->i_fop
= &bad_sock_fops
;
1825 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1826 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1829 EXPORT_SYMBOL(init_special_inode
);
1832 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1834 * @dir: Directory inode
1835 * @mode: mode of the new inode
1837 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1840 inode
->i_uid
= current_fsuid();
1841 if (dir
&& dir
->i_mode
& S_ISGID
) {
1842 inode
->i_gid
= dir
->i_gid
;
1846 inode
->i_gid
= current_fsgid();
1847 inode
->i_mode
= mode
;
1849 EXPORT_SYMBOL(inode_init_owner
);
1852 * inode_owner_or_capable - check current task permissions to inode
1853 * @inode: inode being checked
1855 * Return true if current either has CAP_FOWNER in a namespace with the
1856 * inode owner uid mapped, or owns the file.
1858 bool inode_owner_or_capable(const struct inode
*inode
)
1860 struct user_namespace
*ns
;
1862 if (uid_eq(current_fsuid(), inode
->i_uid
))
1865 ns
= current_user_ns();
1866 if (ns_capable(ns
, CAP_FOWNER
) && kuid_has_mapping(ns
, inode
->i_uid
))
1870 EXPORT_SYMBOL(inode_owner_or_capable
);
1873 * Direct i/o helper functions
1875 static void __inode_dio_wait(struct inode
*inode
)
1877 wait_queue_head_t
*wq
= bit_waitqueue(&inode
->i_state
, __I_DIO_WAKEUP
);
1878 DEFINE_WAIT_BIT(q
, &inode
->i_state
, __I_DIO_WAKEUP
);
1881 prepare_to_wait(wq
, &q
.wait
, TASK_UNINTERRUPTIBLE
);
1882 if (atomic_read(&inode
->i_dio_count
))
1884 } while (atomic_read(&inode
->i_dio_count
));
1885 finish_wait(wq
, &q
.wait
);
1889 * inode_dio_wait - wait for outstanding DIO requests to finish
1890 * @inode: inode to wait for
1892 * Waits for all pending direct I/O requests to finish so that we can
1893 * proceed with a truncate or equivalent operation.
1895 * Must be called under a lock that serializes taking new references
1896 * to i_dio_count, usually by inode->i_mutex.
1898 void inode_dio_wait(struct inode
*inode
)
1900 if (atomic_read(&inode
->i_dio_count
))
1901 __inode_dio_wait(inode
);
1903 EXPORT_SYMBOL(inode_dio_wait
);
1906 * inode_dio_done - signal finish of a direct I/O requests
1907 * @inode: inode the direct I/O happens on
1909 * This is called once we've finished processing a direct I/O request,
1910 * and is used to wake up callers waiting for direct I/O to be quiesced.
1912 void inode_dio_done(struct inode
*inode
)
1914 if (atomic_dec_and_test(&inode
->i_dio_count
))
1915 wake_up_bit(&inode
->i_state
, __I_DIO_WAKEUP
);
1917 EXPORT_SYMBOL(inode_dio_done
);