2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
5 #include <linux/export.h>
8 #include <linux/backing-dev.h>
9 #include <linux/hash.h>
10 #include <linux/swap.h>
11 #include <linux/security.h>
12 #include <linux/cdev.h>
13 #include <linux/bootmem.h>
14 #include <linux/fsnotify.h>
15 #include <linux/mount.h>
16 #include <linux/posix_acl.h>
17 #include <linux/prefetch.h>
18 #include <linux/buffer_head.h> /* for inode_has_buffers */
19 #include <linux/ratelimit.h>
20 #include <linux/list_lru.h>
24 * Inode locking rules:
26 * inode->i_lock protects:
27 * inode->i_state, inode->i_hash, __iget()
28 * Inode LRU list locks protect:
29 * inode->i_sb->s_inode_lru, inode->i_lru
30 * inode_sb_list_lock protects:
31 * sb->s_inodes, inode->i_sb_list
32 * bdi->wb.list_lock protects:
33 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
34 * inode_hash_lock protects:
35 * inode_hashtable, inode->i_hash
41 * Inode LRU list locks
54 static unsigned int i_hash_mask __read_mostly
;
55 static unsigned int i_hash_shift __read_mostly
;
56 static struct hlist_head
*inode_hashtable __read_mostly
;
57 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_hash_lock
);
59 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_sb_list_lock
);
62 * Empty aops. Can be used for the cases where the user does not
63 * define any of the address_space operations.
65 const struct address_space_operations empty_aops
= {
67 EXPORT_SYMBOL(empty_aops
);
70 * Statistics gathering..
72 struct inodes_stat_t inodes_stat
;
74 static DEFINE_PER_CPU(unsigned long, nr_inodes
);
75 static DEFINE_PER_CPU(unsigned long, nr_unused
);
77 static struct kmem_cache
*inode_cachep __read_mostly
;
79 static long get_nr_inodes(void)
83 for_each_possible_cpu(i
)
84 sum
+= per_cpu(nr_inodes
, i
);
85 return sum
< 0 ? 0 : sum
;
88 static inline long get_nr_inodes_unused(void)
92 for_each_possible_cpu(i
)
93 sum
+= per_cpu(nr_unused
, i
);
94 return sum
< 0 ? 0 : sum
;
97 long get_nr_dirty_inodes(void)
99 /* not actually dirty inodes, but a wild approximation */
100 long nr_dirty
= get_nr_inodes() - get_nr_inodes_unused();
101 return nr_dirty
> 0 ? nr_dirty
: 0;
105 * Handle nr_inode sysctl
108 int proc_nr_inodes(struct ctl_table
*table
, int write
,
109 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
111 inodes_stat
.nr_inodes
= get_nr_inodes();
112 inodes_stat
.nr_unused
= get_nr_inodes_unused();
113 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
118 * inode_init_always - perform inode structure intialisation
119 * @sb: superblock inode belongs to
120 * @inode: inode to initialise
122 * These are initializations that need to be done on every inode
123 * allocation as the fields are not initialised by slab allocation.
125 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
127 static const struct inode_operations empty_iops
;
128 static const struct file_operations empty_fops
;
129 struct address_space
*const mapping
= &inode
->i_data
;
132 inode
->i_blkbits
= sb
->s_blocksize_bits
;
134 atomic_set(&inode
->i_count
, 1);
135 inode
->i_op
= &empty_iops
;
136 inode
->i_fop
= &empty_fops
;
137 inode
->__i_nlink
= 1;
138 inode
->i_opflags
= 0;
139 i_uid_write(inode
, 0);
140 i_gid_write(inode
, 0);
141 atomic_set(&inode
->i_writecount
, 0);
145 inode
->i_generation
= 0;
146 inode
->i_pipe
= NULL
;
147 inode
->i_bdev
= NULL
;
148 inode
->i_cdev
= NULL
;
150 inode
->dirtied_when
= 0;
152 if (security_inode_alloc(inode
))
154 spin_lock_init(&inode
->i_lock
);
155 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
157 mutex_init(&inode
->i_mutex
);
158 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
160 atomic_set(&inode
->i_dio_count
, 0);
162 mapping
->a_ops
= &empty_aops
;
163 mapping
->host
= inode
;
165 atomic_set(&mapping
->i_mmap_writable
, 0);
166 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
167 mapping
->private_data
= NULL
;
168 mapping
->backing_dev_info
= &default_backing_dev_info
;
169 mapping
->writeback_index
= 0;
172 * If the block_device provides a backing_dev_info for client
173 * inodes then use that. Otherwise the inode share the bdev's
177 struct backing_dev_info
*bdi
;
179 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
180 mapping
->backing_dev_info
= bdi
;
182 inode
->i_private
= NULL
;
183 inode
->i_mapping
= mapping
;
184 INIT_HLIST_HEAD(&inode
->i_dentry
); /* buggered by rcu freeing */
185 #ifdef CONFIG_FS_POSIX_ACL
186 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
189 #ifdef CONFIG_FSNOTIFY
190 inode
->i_fsnotify_mask
= 0;
193 this_cpu_inc(nr_inodes
);
199 EXPORT_SYMBOL(inode_init_always
);
201 static struct inode
*alloc_inode(struct super_block
*sb
)
205 if (sb
->s_op
->alloc_inode
)
206 inode
= sb
->s_op
->alloc_inode(sb
);
208 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
213 if (unlikely(inode_init_always(sb
, inode
))) {
214 if (inode
->i_sb
->s_op
->destroy_inode
)
215 inode
->i_sb
->s_op
->destroy_inode(inode
);
217 kmem_cache_free(inode_cachep
, inode
);
224 void free_inode_nonrcu(struct inode
*inode
)
226 kmem_cache_free(inode_cachep
, inode
);
228 EXPORT_SYMBOL(free_inode_nonrcu
);
230 void __destroy_inode(struct inode
*inode
)
232 BUG_ON(inode_has_buffers(inode
));
233 security_inode_free(inode
);
234 fsnotify_inode_delete(inode
);
235 if (!inode
->i_nlink
) {
236 WARN_ON(atomic_long_read(&inode
->i_sb
->s_remove_count
) == 0);
237 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
240 #ifdef CONFIG_FS_POSIX_ACL
241 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
242 posix_acl_release(inode
->i_acl
);
243 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
244 posix_acl_release(inode
->i_default_acl
);
246 this_cpu_dec(nr_inodes
);
248 EXPORT_SYMBOL(__destroy_inode
);
250 static void i_callback(struct rcu_head
*head
)
252 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
253 kmem_cache_free(inode_cachep
, inode
);
256 static void destroy_inode(struct inode
*inode
)
258 BUG_ON(!list_empty(&inode
->i_lru
));
259 __destroy_inode(inode
);
260 if (inode
->i_sb
->s_op
->destroy_inode
)
261 inode
->i_sb
->s_op
->destroy_inode(inode
);
263 call_rcu(&inode
->i_rcu
, i_callback
);
267 * drop_nlink - directly drop an inode's link count
270 * This is a low-level filesystem helper to replace any
271 * direct filesystem manipulation of i_nlink. In cases
272 * where we are attempting to track writes to the
273 * filesystem, a decrement to zero means an imminent
274 * write when the file is truncated and actually unlinked
277 void drop_nlink(struct inode
*inode
)
279 WARN_ON(inode
->i_nlink
== 0);
282 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
284 EXPORT_SYMBOL(drop_nlink
);
287 * clear_nlink - directly zero an inode's link count
290 * This is a low-level filesystem helper to replace any
291 * direct filesystem manipulation of i_nlink. See
292 * drop_nlink() for why we care about i_nlink hitting zero.
294 void clear_nlink(struct inode
*inode
)
296 if (inode
->i_nlink
) {
297 inode
->__i_nlink
= 0;
298 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
301 EXPORT_SYMBOL(clear_nlink
);
304 * set_nlink - directly set an inode's link count
306 * @nlink: new nlink (should be non-zero)
308 * This is a low-level filesystem helper to replace any
309 * direct filesystem manipulation of i_nlink.
311 void set_nlink(struct inode
*inode
, unsigned int nlink
)
316 /* Yes, some filesystems do change nlink from zero to one */
317 if (inode
->i_nlink
== 0)
318 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
320 inode
->__i_nlink
= nlink
;
323 EXPORT_SYMBOL(set_nlink
);
326 * inc_nlink - directly increment an inode's link count
329 * This is a low-level filesystem helper to replace any
330 * direct filesystem manipulation of i_nlink. Currently,
331 * it is only here for parity with dec_nlink().
333 void inc_nlink(struct inode
*inode
)
335 if (unlikely(inode
->i_nlink
== 0)) {
336 WARN_ON(!(inode
->i_state
& I_LINKABLE
));
337 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
342 EXPORT_SYMBOL(inc_nlink
);
344 void address_space_init_once(struct address_space
*mapping
)
346 memset(mapping
, 0, sizeof(*mapping
));
347 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
348 spin_lock_init(&mapping
->tree_lock
);
349 init_rwsem(&mapping
->i_mmap_rwsem
);
350 INIT_LIST_HEAD(&mapping
->private_list
);
351 spin_lock_init(&mapping
->private_lock
);
352 mapping
->i_mmap
= RB_ROOT
;
353 INIT_LIST_HEAD(&mapping
->i_mmap_nonlinear
);
355 EXPORT_SYMBOL(address_space_init_once
);
358 * These are initializations that only need to be done
359 * once, because the fields are idempotent across use
360 * of the inode, so let the slab aware of that.
362 void inode_init_once(struct inode
*inode
)
364 memset(inode
, 0, sizeof(*inode
));
365 INIT_HLIST_NODE(&inode
->i_hash
);
366 INIT_LIST_HEAD(&inode
->i_devices
);
367 INIT_LIST_HEAD(&inode
->i_wb_list
);
368 INIT_LIST_HEAD(&inode
->i_lru
);
369 address_space_init_once(&inode
->i_data
);
370 i_size_ordered_init(inode
);
371 #ifdef CONFIG_FSNOTIFY
372 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
375 EXPORT_SYMBOL(inode_init_once
);
377 static void init_once(void *foo
)
379 struct inode
*inode
= (struct inode
*) foo
;
381 inode_init_once(inode
);
385 * inode->i_lock must be held
387 void __iget(struct inode
*inode
)
389 atomic_inc(&inode
->i_count
);
393 * get additional reference to inode; caller must already hold one.
395 void ihold(struct inode
*inode
)
397 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
399 EXPORT_SYMBOL(ihold
);
401 static void inode_lru_list_add(struct inode
*inode
)
403 if (list_lru_add(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
404 this_cpu_inc(nr_unused
);
408 * Add inode to LRU if needed (inode is unused and clean).
410 * Needs inode->i_lock held.
412 void inode_add_lru(struct inode
*inode
)
414 if (!(inode
->i_state
& (I_DIRTY
| I_SYNC
| I_FREEING
| I_WILL_FREE
)) &&
415 !atomic_read(&inode
->i_count
) && inode
->i_sb
->s_flags
& MS_ACTIVE
)
416 inode_lru_list_add(inode
);
420 static void inode_lru_list_del(struct inode
*inode
)
423 if (list_lru_del(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
424 this_cpu_dec(nr_unused
);
428 * inode_sb_list_add - add inode to the superblock list of inodes
429 * @inode: inode to add
431 void inode_sb_list_add(struct inode
*inode
)
433 spin_lock(&inode_sb_list_lock
);
434 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
435 spin_unlock(&inode_sb_list_lock
);
437 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
439 static inline void inode_sb_list_del(struct inode
*inode
)
441 if (!list_empty(&inode
->i_sb_list
)) {
442 spin_lock(&inode_sb_list_lock
);
443 list_del_init(&inode
->i_sb_list
);
444 spin_unlock(&inode_sb_list_lock
);
448 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
452 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
454 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> i_hash_shift
);
455 return tmp
& i_hash_mask
;
459 * __insert_inode_hash - hash an inode
460 * @inode: unhashed inode
461 * @hashval: unsigned long value used to locate this object in the
464 * Add an inode to the inode hash for this superblock.
466 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
468 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
470 spin_lock(&inode_hash_lock
);
471 spin_lock(&inode
->i_lock
);
472 hlist_add_head(&inode
->i_hash
, b
);
473 spin_unlock(&inode
->i_lock
);
474 spin_unlock(&inode_hash_lock
);
476 EXPORT_SYMBOL(__insert_inode_hash
);
479 * __remove_inode_hash - remove an inode from the hash
480 * @inode: inode to unhash
482 * Remove an inode from the superblock.
484 void __remove_inode_hash(struct inode
*inode
)
486 spin_lock(&inode_hash_lock
);
487 spin_lock(&inode
->i_lock
);
488 hlist_del_init(&inode
->i_hash
);
489 spin_unlock(&inode
->i_lock
);
490 spin_unlock(&inode_hash_lock
);
492 EXPORT_SYMBOL(__remove_inode_hash
);
494 void clear_inode(struct inode
*inode
)
498 * We have to cycle tree_lock here because reclaim can be still in the
499 * process of removing the last page (in __delete_from_page_cache())
500 * and we must not free mapping under it.
502 spin_lock_irq(&inode
->i_data
.tree_lock
);
503 BUG_ON(inode
->i_data
.nrpages
);
504 BUG_ON(inode
->i_data
.nrshadows
);
505 spin_unlock_irq(&inode
->i_data
.tree_lock
);
506 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
507 BUG_ON(!(inode
->i_state
& I_FREEING
));
508 BUG_ON(inode
->i_state
& I_CLEAR
);
509 /* don't need i_lock here, no concurrent mods to i_state */
510 inode
->i_state
= I_FREEING
| I_CLEAR
;
512 EXPORT_SYMBOL(clear_inode
);
515 * Free the inode passed in, removing it from the lists it is still connected
516 * to. We remove any pages still attached to the inode and wait for any IO that
517 * is still in progress before finally destroying the inode.
519 * An inode must already be marked I_FREEING so that we avoid the inode being
520 * moved back onto lists if we race with other code that manipulates the lists
521 * (e.g. writeback_single_inode). The caller is responsible for setting this.
523 * An inode must already be removed from the LRU list before being evicted from
524 * the cache. This should occur atomically with setting the I_FREEING state
525 * flag, so no inodes here should ever be on the LRU when being evicted.
527 static void evict(struct inode
*inode
)
529 const struct super_operations
*op
= inode
->i_sb
->s_op
;
531 BUG_ON(!(inode
->i_state
& I_FREEING
));
532 BUG_ON(!list_empty(&inode
->i_lru
));
534 if (!list_empty(&inode
->i_wb_list
))
535 inode_wb_list_del(inode
);
537 inode_sb_list_del(inode
);
540 * Wait for flusher thread to be done with the inode so that filesystem
541 * does not start destroying it while writeback is still running. Since
542 * the inode has I_FREEING set, flusher thread won't start new work on
543 * the inode. We just have to wait for running writeback to finish.
545 inode_wait_for_writeback(inode
);
547 if (op
->evict_inode
) {
548 op
->evict_inode(inode
);
550 truncate_inode_pages_final(&inode
->i_data
);
553 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
555 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
558 remove_inode_hash(inode
);
560 spin_lock(&inode
->i_lock
);
561 wake_up_bit(&inode
->i_state
, __I_NEW
);
562 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
563 spin_unlock(&inode
->i_lock
);
565 destroy_inode(inode
);
569 * dispose_list - dispose of the contents of a local list
570 * @head: the head of the list to free
572 * Dispose-list gets a local list with local inodes in it, so it doesn't
573 * need to worry about list corruption and SMP locks.
575 static void dispose_list(struct list_head
*head
)
577 while (!list_empty(head
)) {
580 inode
= list_first_entry(head
, struct inode
, i_lru
);
581 list_del_init(&inode
->i_lru
);
588 * evict_inodes - evict all evictable inodes for a superblock
589 * @sb: superblock to operate on
591 * Make sure that no inodes with zero refcount are retained. This is
592 * called by superblock shutdown after having MS_ACTIVE flag removed,
593 * so any inode reaching zero refcount during or after that call will
594 * be immediately evicted.
596 void evict_inodes(struct super_block
*sb
)
598 struct inode
*inode
, *next
;
601 spin_lock(&inode_sb_list_lock
);
602 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
603 if (atomic_read(&inode
->i_count
))
606 spin_lock(&inode
->i_lock
);
607 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
608 spin_unlock(&inode
->i_lock
);
612 inode
->i_state
|= I_FREEING
;
613 inode_lru_list_del(inode
);
614 spin_unlock(&inode
->i_lock
);
615 list_add(&inode
->i_lru
, &dispose
);
617 spin_unlock(&inode_sb_list_lock
);
619 dispose_list(&dispose
);
623 * invalidate_inodes - attempt to free all inodes on a superblock
624 * @sb: superblock to operate on
625 * @kill_dirty: flag to guide handling of dirty inodes
627 * Attempts to free all inodes for a given superblock. If there were any
628 * busy inodes return a non-zero value, else zero.
629 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
632 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
635 struct inode
*inode
, *next
;
638 spin_lock(&inode_sb_list_lock
);
639 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
640 spin_lock(&inode
->i_lock
);
641 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
642 spin_unlock(&inode
->i_lock
);
645 if (inode
->i_state
& I_DIRTY
&& !kill_dirty
) {
646 spin_unlock(&inode
->i_lock
);
650 if (atomic_read(&inode
->i_count
)) {
651 spin_unlock(&inode
->i_lock
);
656 inode
->i_state
|= I_FREEING
;
657 inode_lru_list_del(inode
);
658 spin_unlock(&inode
->i_lock
);
659 list_add(&inode
->i_lru
, &dispose
);
661 spin_unlock(&inode_sb_list_lock
);
663 dispose_list(&dispose
);
669 * Isolate the inode from the LRU in preparation for freeing it.
671 * Any inodes which are pinned purely because of attached pagecache have their
672 * pagecache removed. If the inode has metadata buffers attached to
673 * mapping->private_list then try to remove them.
675 * If the inode has the I_REFERENCED flag set, then it means that it has been
676 * used recently - the flag is set in iput_final(). When we encounter such an
677 * inode, clear the flag and move it to the back of the LRU so it gets another
678 * pass through the LRU before it gets reclaimed. This is necessary because of
679 * the fact we are doing lazy LRU updates to minimise lock contention so the
680 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
681 * with this flag set because they are the inodes that are out of order.
683 static enum lru_status
684 inode_lru_isolate(struct list_head
*item
, spinlock_t
*lru_lock
, void *arg
)
686 struct list_head
*freeable
= arg
;
687 struct inode
*inode
= container_of(item
, struct inode
, i_lru
);
690 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
691 * If we fail to get the lock, just skip it.
693 if (!spin_trylock(&inode
->i_lock
))
697 * Referenced or dirty inodes are still in use. Give them another pass
698 * through the LRU as we canot reclaim them now.
700 if (atomic_read(&inode
->i_count
) ||
701 (inode
->i_state
& ~I_REFERENCED
)) {
702 list_del_init(&inode
->i_lru
);
703 spin_unlock(&inode
->i_lock
);
704 this_cpu_dec(nr_unused
);
708 /* recently referenced inodes get one more pass */
709 if (inode
->i_state
& I_REFERENCED
) {
710 inode
->i_state
&= ~I_REFERENCED
;
711 spin_unlock(&inode
->i_lock
);
715 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
717 spin_unlock(&inode
->i_lock
);
718 spin_unlock(lru_lock
);
719 if (remove_inode_buffers(inode
)) {
721 reap
= invalidate_mapping_pages(&inode
->i_data
, 0, -1);
722 if (current_is_kswapd())
723 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
725 __count_vm_events(PGINODESTEAL
, reap
);
726 if (current
->reclaim_state
)
727 current
->reclaim_state
->reclaimed_slab
+= reap
;
734 WARN_ON(inode
->i_state
& I_NEW
);
735 inode
->i_state
|= I_FREEING
;
736 list_move(&inode
->i_lru
, freeable
);
737 spin_unlock(&inode
->i_lock
);
739 this_cpu_dec(nr_unused
);
744 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
745 * This is called from the superblock shrinker function with a number of inodes
746 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
747 * then are freed outside inode_lock by dispose_list().
749 long prune_icache_sb(struct super_block
*sb
, unsigned long nr_to_scan
,
755 freed
= list_lru_walk_node(&sb
->s_inode_lru
, nid
, inode_lru_isolate
,
756 &freeable
, &nr_to_scan
);
757 dispose_list(&freeable
);
761 static void __wait_on_freeing_inode(struct inode
*inode
);
763 * Called with the inode lock held.
765 static struct inode
*find_inode(struct super_block
*sb
,
766 struct hlist_head
*head
,
767 int (*test
)(struct inode
*, void *),
770 struct inode
*inode
= NULL
;
773 hlist_for_each_entry(inode
, head
, i_hash
) {
774 if (inode
->i_sb
!= sb
)
776 if (!test(inode
, data
))
778 spin_lock(&inode
->i_lock
);
779 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
780 __wait_on_freeing_inode(inode
);
784 spin_unlock(&inode
->i_lock
);
791 * find_inode_fast is the fast path version of find_inode, see the comment at
792 * iget_locked for details.
794 static struct inode
*find_inode_fast(struct super_block
*sb
,
795 struct hlist_head
*head
, unsigned long ino
)
797 struct inode
*inode
= NULL
;
800 hlist_for_each_entry(inode
, head
, i_hash
) {
801 if (inode
->i_ino
!= ino
)
803 if (inode
->i_sb
!= sb
)
805 spin_lock(&inode
->i_lock
);
806 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
807 __wait_on_freeing_inode(inode
);
811 spin_unlock(&inode
->i_lock
);
818 * Each cpu owns a range of LAST_INO_BATCH numbers.
819 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
820 * to renew the exhausted range.
822 * This does not significantly increase overflow rate because every CPU can
823 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
824 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
825 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
826 * overflow rate by 2x, which does not seem too significant.
828 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
829 * error if st_ino won't fit in target struct field. Use 32bit counter
830 * here to attempt to avoid that.
832 #define LAST_INO_BATCH 1024
833 static DEFINE_PER_CPU(unsigned int, last_ino
);
835 unsigned int get_next_ino(void)
837 unsigned int *p
= &get_cpu_var(last_ino
);
838 unsigned int res
= *p
;
841 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
842 static atomic_t shared_last_ino
;
843 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
845 res
= next
- LAST_INO_BATCH
;
850 put_cpu_var(last_ino
);
853 EXPORT_SYMBOL(get_next_ino
);
856 * new_inode_pseudo - obtain an inode
859 * Allocates a new inode for given superblock.
860 * Inode wont be chained in superblock s_inodes list
862 * - fs can't be unmount
863 * - quotas, fsnotify, writeback can't work
865 struct inode
*new_inode_pseudo(struct super_block
*sb
)
867 struct inode
*inode
= alloc_inode(sb
);
870 spin_lock(&inode
->i_lock
);
872 spin_unlock(&inode
->i_lock
);
873 INIT_LIST_HEAD(&inode
->i_sb_list
);
879 * new_inode - obtain an inode
882 * Allocates a new inode for given superblock. The default gfp_mask
883 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
884 * If HIGHMEM pages are unsuitable or it is known that pages allocated
885 * for the page cache are not reclaimable or migratable,
886 * mapping_set_gfp_mask() must be called with suitable flags on the
887 * newly created inode's mapping
890 struct inode
*new_inode(struct super_block
*sb
)
894 spin_lock_prefetch(&inode_sb_list_lock
);
896 inode
= new_inode_pseudo(sb
);
898 inode_sb_list_add(inode
);
901 EXPORT_SYMBOL(new_inode
);
903 #ifdef CONFIG_DEBUG_LOCK_ALLOC
904 void lockdep_annotate_inode_mutex_key(struct inode
*inode
)
906 if (S_ISDIR(inode
->i_mode
)) {
907 struct file_system_type
*type
= inode
->i_sb
->s_type
;
909 /* Set new key only if filesystem hasn't already changed it */
910 if (lockdep_match_class(&inode
->i_mutex
, &type
->i_mutex_key
)) {
912 * ensure nobody is actually holding i_mutex
914 mutex_destroy(&inode
->i_mutex
);
915 mutex_init(&inode
->i_mutex
);
916 lockdep_set_class(&inode
->i_mutex
,
917 &type
->i_mutex_dir_key
);
921 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key
);
925 * unlock_new_inode - clear the I_NEW state and wake up any waiters
926 * @inode: new inode to unlock
928 * Called when the inode is fully initialised to clear the new state of the
929 * inode and wake up anyone waiting for the inode to finish initialisation.
931 void unlock_new_inode(struct inode
*inode
)
933 lockdep_annotate_inode_mutex_key(inode
);
934 spin_lock(&inode
->i_lock
);
935 WARN_ON(!(inode
->i_state
& I_NEW
));
936 inode
->i_state
&= ~I_NEW
;
938 wake_up_bit(&inode
->i_state
, __I_NEW
);
939 spin_unlock(&inode
->i_lock
);
941 EXPORT_SYMBOL(unlock_new_inode
);
944 * lock_two_nondirectories - take two i_mutexes on non-directory objects
946 * Lock any non-NULL argument that is not a directory.
947 * Zero, one or two objects may be locked by this function.
949 * @inode1: first inode to lock
950 * @inode2: second inode to lock
952 void lock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
955 swap(inode1
, inode2
);
957 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
958 mutex_lock(&inode1
->i_mutex
);
959 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
960 mutex_lock_nested(&inode2
->i_mutex
, I_MUTEX_NONDIR2
);
962 EXPORT_SYMBOL(lock_two_nondirectories
);
965 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
966 * @inode1: first inode to unlock
967 * @inode2: second inode to unlock
969 void unlock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
971 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
972 mutex_unlock(&inode1
->i_mutex
);
973 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
974 mutex_unlock(&inode2
->i_mutex
);
976 EXPORT_SYMBOL(unlock_two_nondirectories
);
979 * iget5_locked - obtain an inode from a mounted file system
980 * @sb: super block of file system
981 * @hashval: hash value (usually inode number) to get
982 * @test: callback used for comparisons between inodes
983 * @set: callback used to initialize a new struct inode
984 * @data: opaque data pointer to pass to @test and @set
986 * Search for the inode specified by @hashval and @data in the inode cache,
987 * and if present it is return it with an increased reference count. This is
988 * a generalized version of iget_locked() for file systems where the inode
989 * number is not sufficient for unique identification of an inode.
991 * If the inode is not in cache, allocate a new inode and return it locked,
992 * hashed, and with the I_NEW flag set. The file system gets to fill it in
993 * before unlocking it via unlock_new_inode().
995 * Note both @test and @set are called with the inode_hash_lock held, so can't
998 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
999 int (*test
)(struct inode
*, void *),
1000 int (*set
)(struct inode
*, void *), void *data
)
1002 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1003 struct inode
*inode
;
1005 spin_lock(&inode_hash_lock
);
1006 inode
= find_inode(sb
, head
, test
, data
);
1007 spin_unlock(&inode_hash_lock
);
1010 wait_on_inode(inode
);
1014 inode
= alloc_inode(sb
);
1018 spin_lock(&inode_hash_lock
);
1019 /* We released the lock, so.. */
1020 old
= find_inode(sb
, head
, test
, data
);
1022 if (set(inode
, data
))
1025 spin_lock(&inode
->i_lock
);
1026 inode
->i_state
= I_NEW
;
1027 hlist_add_head(&inode
->i_hash
, head
);
1028 spin_unlock(&inode
->i_lock
);
1029 inode_sb_list_add(inode
);
1030 spin_unlock(&inode_hash_lock
);
1032 /* Return the locked inode with I_NEW set, the
1033 * caller is responsible for filling in the contents
1039 * Uhhuh, somebody else created the same inode under
1040 * us. Use the old inode instead of the one we just
1043 spin_unlock(&inode_hash_lock
);
1044 destroy_inode(inode
);
1046 wait_on_inode(inode
);
1051 spin_unlock(&inode_hash_lock
);
1052 destroy_inode(inode
);
1055 EXPORT_SYMBOL(iget5_locked
);
1058 * iget_locked - obtain an inode from a mounted file system
1059 * @sb: super block of file system
1060 * @ino: inode number to get
1062 * Search for the inode specified by @ino in the inode cache and if present
1063 * return it with an increased reference count. This is for file systems
1064 * where the inode number is sufficient for unique identification of an inode.
1066 * If the inode is not in cache, allocate a new inode and return it locked,
1067 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1068 * before unlocking it via unlock_new_inode().
1070 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1072 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1073 struct inode
*inode
;
1075 spin_lock(&inode_hash_lock
);
1076 inode
= find_inode_fast(sb
, head
, ino
);
1077 spin_unlock(&inode_hash_lock
);
1079 wait_on_inode(inode
);
1083 inode
= alloc_inode(sb
);
1087 spin_lock(&inode_hash_lock
);
1088 /* We released the lock, so.. */
1089 old
= find_inode_fast(sb
, head
, ino
);
1092 spin_lock(&inode
->i_lock
);
1093 inode
->i_state
= I_NEW
;
1094 hlist_add_head(&inode
->i_hash
, head
);
1095 spin_unlock(&inode
->i_lock
);
1096 inode_sb_list_add(inode
);
1097 spin_unlock(&inode_hash_lock
);
1099 /* Return the locked inode with I_NEW set, the
1100 * caller is responsible for filling in the contents
1106 * Uhhuh, somebody else created the same inode under
1107 * us. Use the old inode instead of the one we just
1110 spin_unlock(&inode_hash_lock
);
1111 destroy_inode(inode
);
1113 wait_on_inode(inode
);
1117 EXPORT_SYMBOL(iget_locked
);
1120 * search the inode cache for a matching inode number.
1121 * If we find one, then the inode number we are trying to
1122 * allocate is not unique and so we should not use it.
1124 * Returns 1 if the inode number is unique, 0 if it is not.
1126 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1128 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1129 struct inode
*inode
;
1131 spin_lock(&inode_hash_lock
);
1132 hlist_for_each_entry(inode
, b
, i_hash
) {
1133 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
) {
1134 spin_unlock(&inode_hash_lock
);
1138 spin_unlock(&inode_hash_lock
);
1144 * iunique - get a unique inode number
1146 * @max_reserved: highest reserved inode number
1148 * Obtain an inode number that is unique on the system for a given
1149 * superblock. This is used by file systems that have no natural
1150 * permanent inode numbering system. An inode number is returned that
1151 * is higher than the reserved limit but unique.
1154 * With a large number of inodes live on the file system this function
1155 * currently becomes quite slow.
1157 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1160 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1161 * error if st_ino won't fit in target struct field. Use 32bit counter
1162 * here to attempt to avoid that.
1164 static DEFINE_SPINLOCK(iunique_lock
);
1165 static unsigned int counter
;
1168 spin_lock(&iunique_lock
);
1170 if (counter
<= max_reserved
)
1171 counter
= max_reserved
+ 1;
1173 } while (!test_inode_iunique(sb
, res
));
1174 spin_unlock(&iunique_lock
);
1178 EXPORT_SYMBOL(iunique
);
1180 struct inode
*igrab(struct inode
*inode
)
1182 spin_lock(&inode
->i_lock
);
1183 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1185 spin_unlock(&inode
->i_lock
);
1187 spin_unlock(&inode
->i_lock
);
1189 * Handle the case where s_op->clear_inode is not been
1190 * called yet, and somebody is calling igrab
1191 * while the inode is getting freed.
1197 EXPORT_SYMBOL(igrab
);
1200 * ilookup5_nowait - search for an inode in the inode cache
1201 * @sb: super block of file system to search
1202 * @hashval: hash value (usually inode number) to search for
1203 * @test: callback used for comparisons between inodes
1204 * @data: opaque data pointer to pass to @test
1206 * Search for the inode specified by @hashval and @data in the inode cache.
1207 * If the inode is in the cache, the inode is returned with an incremented
1210 * Note: I_NEW is not waited upon so you have to be very careful what you do
1211 * with the returned inode. You probably should be using ilookup5() instead.
1213 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1215 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1216 int (*test
)(struct inode
*, void *), void *data
)
1218 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1219 struct inode
*inode
;
1221 spin_lock(&inode_hash_lock
);
1222 inode
= find_inode(sb
, head
, test
, data
);
1223 spin_unlock(&inode_hash_lock
);
1227 EXPORT_SYMBOL(ilookup5_nowait
);
1230 * ilookup5 - search for an inode in the inode cache
1231 * @sb: super block of file system to search
1232 * @hashval: hash value (usually inode number) to search for
1233 * @test: callback used for comparisons between inodes
1234 * @data: opaque data pointer to pass to @test
1236 * Search for the inode specified by @hashval and @data in the inode cache,
1237 * and if the inode is in the cache, return the inode with an incremented
1238 * reference count. Waits on I_NEW before returning the inode.
1239 * returned with an incremented reference count.
1241 * This is a generalized version of ilookup() for file systems where the
1242 * inode number is not sufficient for unique identification of an inode.
1244 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1246 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1247 int (*test
)(struct inode
*, void *), void *data
)
1249 struct inode
*inode
= ilookup5_nowait(sb
, hashval
, test
, data
);
1252 wait_on_inode(inode
);
1255 EXPORT_SYMBOL(ilookup5
);
1258 * ilookup - search for an inode in the inode cache
1259 * @sb: super block of file system to search
1260 * @ino: inode number to search for
1262 * Search for the inode @ino in the inode cache, and if the inode is in the
1263 * cache, the inode is returned with an incremented reference count.
1265 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1267 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1268 struct inode
*inode
;
1270 spin_lock(&inode_hash_lock
);
1271 inode
= find_inode_fast(sb
, head
, ino
);
1272 spin_unlock(&inode_hash_lock
);
1275 wait_on_inode(inode
);
1278 EXPORT_SYMBOL(ilookup
);
1280 int insert_inode_locked(struct inode
*inode
)
1282 struct super_block
*sb
= inode
->i_sb
;
1283 ino_t ino
= inode
->i_ino
;
1284 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1287 struct inode
*old
= NULL
;
1288 spin_lock(&inode_hash_lock
);
1289 hlist_for_each_entry(old
, head
, i_hash
) {
1290 if (old
->i_ino
!= ino
)
1292 if (old
->i_sb
!= sb
)
1294 spin_lock(&old
->i_lock
);
1295 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1296 spin_unlock(&old
->i_lock
);
1302 spin_lock(&inode
->i_lock
);
1303 inode
->i_state
|= I_NEW
;
1304 hlist_add_head(&inode
->i_hash
, head
);
1305 spin_unlock(&inode
->i_lock
);
1306 spin_unlock(&inode_hash_lock
);
1310 spin_unlock(&old
->i_lock
);
1311 spin_unlock(&inode_hash_lock
);
1313 if (unlikely(!inode_unhashed(old
))) {
1320 EXPORT_SYMBOL(insert_inode_locked
);
1322 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1323 int (*test
)(struct inode
*, void *), void *data
)
1325 struct super_block
*sb
= inode
->i_sb
;
1326 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1329 struct inode
*old
= NULL
;
1331 spin_lock(&inode_hash_lock
);
1332 hlist_for_each_entry(old
, head
, i_hash
) {
1333 if (old
->i_sb
!= sb
)
1335 if (!test(old
, data
))
1337 spin_lock(&old
->i_lock
);
1338 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1339 spin_unlock(&old
->i_lock
);
1345 spin_lock(&inode
->i_lock
);
1346 inode
->i_state
|= I_NEW
;
1347 hlist_add_head(&inode
->i_hash
, head
);
1348 spin_unlock(&inode
->i_lock
);
1349 spin_unlock(&inode_hash_lock
);
1353 spin_unlock(&old
->i_lock
);
1354 spin_unlock(&inode_hash_lock
);
1356 if (unlikely(!inode_unhashed(old
))) {
1363 EXPORT_SYMBOL(insert_inode_locked4
);
1366 int generic_delete_inode(struct inode
*inode
)
1370 EXPORT_SYMBOL(generic_delete_inode
);
1373 * Called when we're dropping the last reference
1376 * Call the FS "drop_inode()" function, defaulting to
1377 * the legacy UNIX filesystem behaviour. If it tells
1378 * us to evict inode, do so. Otherwise, retain inode
1379 * in cache if fs is alive, sync and evict if fs is
1382 static void iput_final(struct inode
*inode
)
1384 struct super_block
*sb
= inode
->i_sb
;
1385 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1388 WARN_ON(inode
->i_state
& I_NEW
);
1391 drop
= op
->drop_inode(inode
);
1393 drop
= generic_drop_inode(inode
);
1395 if (!drop
&& (sb
->s_flags
& MS_ACTIVE
)) {
1396 inode
->i_state
|= I_REFERENCED
;
1397 inode_add_lru(inode
);
1398 spin_unlock(&inode
->i_lock
);
1403 inode
->i_state
|= I_WILL_FREE
;
1404 spin_unlock(&inode
->i_lock
);
1405 write_inode_now(inode
, 1);
1406 spin_lock(&inode
->i_lock
);
1407 WARN_ON(inode
->i_state
& I_NEW
);
1408 inode
->i_state
&= ~I_WILL_FREE
;
1411 inode
->i_state
|= I_FREEING
;
1412 if (!list_empty(&inode
->i_lru
))
1413 inode_lru_list_del(inode
);
1414 spin_unlock(&inode
->i_lock
);
1420 * iput - put an inode
1421 * @inode: inode to put
1423 * Puts an inode, dropping its usage count. If the inode use count hits
1424 * zero, the inode is then freed and may also be destroyed.
1426 * Consequently, iput() can sleep.
1428 void iput(struct inode
*inode
)
1431 BUG_ON(inode
->i_state
& I_CLEAR
);
1433 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
))
1437 EXPORT_SYMBOL(iput
);
1440 * bmap - find a block number in a file
1441 * @inode: inode of file
1442 * @block: block to find
1444 * Returns the block number on the device holding the inode that
1445 * is the disk block number for the block of the file requested.
1446 * That is, asked for block 4 of inode 1 the function will return the
1447 * disk block relative to the disk start that holds that block of the
1450 sector_t
bmap(struct inode
*inode
, sector_t block
)
1453 if (inode
->i_mapping
->a_ops
->bmap
)
1454 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1457 EXPORT_SYMBOL(bmap
);
1460 * With relative atime, only update atime if the previous atime is
1461 * earlier than either the ctime or mtime or if at least a day has
1462 * passed since the last atime update.
1464 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1465 struct timespec now
)
1468 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1471 * Is mtime younger than atime? If yes, update atime:
1473 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1476 * Is ctime younger than atime? If yes, update atime:
1478 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1482 * Is the previous atime value older than a day? If yes,
1485 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1488 * Good, we can skip the atime update:
1494 * This does the actual work of updating an inodes time or version. Must have
1495 * had called mnt_want_write() before calling this.
1497 static int update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1499 if (inode
->i_op
->update_time
)
1500 return inode
->i_op
->update_time(inode
, time
, flags
);
1502 if (flags
& S_ATIME
)
1503 inode
->i_atime
= *time
;
1504 if (flags
& S_VERSION
)
1505 inode_inc_iversion(inode
);
1506 if (flags
& S_CTIME
)
1507 inode
->i_ctime
= *time
;
1508 if (flags
& S_MTIME
)
1509 inode
->i_mtime
= *time
;
1510 mark_inode_dirty_sync(inode
);
1515 * touch_atime - update the access time
1516 * @path: the &struct path to update
1518 * Update the accessed time on an inode and mark it for writeback.
1519 * This function automatically handles read only file systems and media,
1520 * as well as the "noatime" flag and inode specific "noatime" markers.
1522 void touch_atime(const struct path
*path
)
1524 struct vfsmount
*mnt
= path
->mnt
;
1525 struct inode
*inode
= path
->dentry
->d_inode
;
1526 struct timespec now
;
1528 if (inode
->i_flags
& S_NOATIME
)
1530 if (IS_NOATIME(inode
))
1532 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1535 if (mnt
->mnt_flags
& MNT_NOATIME
)
1537 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1540 now
= current_fs_time(inode
->i_sb
);
1542 if (!relatime_need_update(mnt
, inode
, now
))
1545 if (timespec_equal(&inode
->i_atime
, &now
))
1548 if (!sb_start_write_trylock(inode
->i_sb
))
1551 if (__mnt_want_write(mnt
))
1554 * File systems can error out when updating inodes if they need to
1555 * allocate new space to modify an inode (such is the case for
1556 * Btrfs), but since we touch atime while walking down the path we
1557 * really don't care if we failed to update the atime of the file,
1558 * so just ignore the return value.
1559 * We may also fail on filesystems that have the ability to make parts
1560 * of the fs read only, e.g. subvolumes in Btrfs.
1562 update_time(inode
, &now
, S_ATIME
);
1563 __mnt_drop_write(mnt
);
1565 sb_end_write(inode
->i_sb
);
1567 EXPORT_SYMBOL(touch_atime
);
1570 * The logic we want is
1572 * if suid or (sgid and xgrp)
1575 int should_remove_suid(struct dentry
*dentry
)
1577 umode_t mode
= dentry
->d_inode
->i_mode
;
1580 /* suid always must be killed */
1581 if (unlikely(mode
& S_ISUID
))
1582 kill
= ATTR_KILL_SUID
;
1585 * sgid without any exec bits is just a mandatory locking mark; leave
1586 * it alone. If some exec bits are set, it's a real sgid; kill it.
1588 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1589 kill
|= ATTR_KILL_SGID
;
1591 if (unlikely(kill
&& !capable(CAP_FSETID
) && S_ISREG(mode
)))
1596 EXPORT_SYMBOL(should_remove_suid
);
1598 static int __remove_suid(struct dentry
*dentry
, int kill
)
1600 struct iattr newattrs
;
1602 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1604 * Note we call this on write, so notify_change will not
1605 * encounter any conflicting delegations:
1607 return notify_change(dentry
, &newattrs
, NULL
);
1610 int file_remove_suid(struct file
*file
)
1612 struct dentry
*dentry
= file
->f_path
.dentry
;
1613 struct inode
*inode
= dentry
->d_inode
;
1618 /* Fast path for nothing security related */
1619 if (IS_NOSEC(inode
))
1622 killsuid
= should_remove_suid(dentry
);
1623 killpriv
= security_inode_need_killpriv(dentry
);
1628 error
= security_inode_killpriv(dentry
);
1629 if (!error
&& killsuid
)
1630 error
= __remove_suid(dentry
, killsuid
);
1631 if (!error
&& (inode
->i_sb
->s_flags
& MS_NOSEC
))
1632 inode
->i_flags
|= S_NOSEC
;
1636 EXPORT_SYMBOL(file_remove_suid
);
1639 * file_update_time - update mtime and ctime time
1640 * @file: file accessed
1642 * Update the mtime and ctime members of an inode and mark the inode
1643 * for writeback. Note that this function is meant exclusively for
1644 * usage in the file write path of filesystems, and filesystems may
1645 * choose to explicitly ignore update via this function with the
1646 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1647 * timestamps are handled by the server. This can return an error for
1648 * file systems who need to allocate space in order to update an inode.
1651 int file_update_time(struct file
*file
)
1653 struct inode
*inode
= file_inode(file
);
1654 struct timespec now
;
1658 /* First try to exhaust all avenues to not sync */
1659 if (IS_NOCMTIME(inode
))
1662 now
= current_fs_time(inode
->i_sb
);
1663 if (!timespec_equal(&inode
->i_mtime
, &now
))
1666 if (!timespec_equal(&inode
->i_ctime
, &now
))
1669 if (IS_I_VERSION(inode
))
1670 sync_it
|= S_VERSION
;
1675 /* Finally allowed to write? Takes lock. */
1676 if (__mnt_want_write_file(file
))
1679 ret
= update_time(inode
, &now
, sync_it
);
1680 __mnt_drop_write_file(file
);
1684 EXPORT_SYMBOL(file_update_time
);
1686 int inode_needs_sync(struct inode
*inode
)
1690 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1694 EXPORT_SYMBOL(inode_needs_sync
);
1697 * If we try to find an inode in the inode hash while it is being
1698 * deleted, we have to wait until the filesystem completes its
1699 * deletion before reporting that it isn't found. This function waits
1700 * until the deletion _might_ have completed. Callers are responsible
1701 * to recheck inode state.
1703 * It doesn't matter if I_NEW is not set initially, a call to
1704 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1707 static void __wait_on_freeing_inode(struct inode
*inode
)
1709 wait_queue_head_t
*wq
;
1710 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1711 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1712 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1713 spin_unlock(&inode
->i_lock
);
1714 spin_unlock(&inode_hash_lock
);
1716 finish_wait(wq
, &wait
.wait
);
1717 spin_lock(&inode_hash_lock
);
1720 static __initdata
unsigned long ihash_entries
;
1721 static int __init
set_ihash_entries(char *str
)
1725 ihash_entries
= simple_strtoul(str
, &str
, 0);
1728 __setup("ihash_entries=", set_ihash_entries
);
1731 * Initialize the waitqueues and inode hash table.
1733 void __init
inode_init_early(void)
1737 /* If hashes are distributed across NUMA nodes, defer
1738 * hash allocation until vmalloc space is available.
1744 alloc_large_system_hash("Inode-cache",
1745 sizeof(struct hlist_head
),
1754 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1755 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1758 void __init
inode_init(void)
1762 /* inode slab cache */
1763 inode_cachep
= kmem_cache_create("inode_cache",
1764 sizeof(struct inode
),
1766 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1770 /* Hash may have been set up in inode_init_early */
1775 alloc_large_system_hash("Inode-cache",
1776 sizeof(struct hlist_head
),
1785 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1786 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1789 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1791 inode
->i_mode
= mode
;
1792 if (S_ISCHR(mode
)) {
1793 inode
->i_fop
= &def_chr_fops
;
1794 inode
->i_rdev
= rdev
;
1795 } else if (S_ISBLK(mode
)) {
1796 inode
->i_fop
= &def_blk_fops
;
1797 inode
->i_rdev
= rdev
;
1798 } else if (S_ISFIFO(mode
))
1799 inode
->i_fop
= &pipefifo_fops
;
1800 else if (S_ISSOCK(mode
))
1801 inode
->i_fop
= &bad_sock_fops
;
1803 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1804 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1807 EXPORT_SYMBOL(init_special_inode
);
1810 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1812 * @dir: Directory inode
1813 * @mode: mode of the new inode
1815 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1818 inode
->i_uid
= current_fsuid();
1819 if (dir
&& dir
->i_mode
& S_ISGID
) {
1820 inode
->i_gid
= dir
->i_gid
;
1824 inode
->i_gid
= current_fsgid();
1825 inode
->i_mode
= mode
;
1827 EXPORT_SYMBOL(inode_init_owner
);
1830 * inode_owner_or_capable - check current task permissions to inode
1831 * @inode: inode being checked
1833 * Return true if current either has CAP_FOWNER in a namespace with the
1834 * inode owner uid mapped, or owns the file.
1836 bool inode_owner_or_capable(const struct inode
*inode
)
1838 struct user_namespace
*ns
;
1840 if (uid_eq(current_fsuid(), inode
->i_uid
))
1843 ns
= current_user_ns();
1844 if (ns_capable(ns
, CAP_FOWNER
) && kuid_has_mapping(ns
, inode
->i_uid
))
1848 EXPORT_SYMBOL(inode_owner_or_capable
);
1851 * Direct i/o helper functions
1853 static void __inode_dio_wait(struct inode
*inode
)
1855 wait_queue_head_t
*wq
= bit_waitqueue(&inode
->i_state
, __I_DIO_WAKEUP
);
1856 DEFINE_WAIT_BIT(q
, &inode
->i_state
, __I_DIO_WAKEUP
);
1859 prepare_to_wait(wq
, &q
.wait
, TASK_UNINTERRUPTIBLE
);
1860 if (atomic_read(&inode
->i_dio_count
))
1862 } while (atomic_read(&inode
->i_dio_count
));
1863 finish_wait(wq
, &q
.wait
);
1867 * inode_dio_wait - wait for outstanding DIO requests to finish
1868 * @inode: inode to wait for
1870 * Waits for all pending direct I/O requests to finish so that we can
1871 * proceed with a truncate or equivalent operation.
1873 * Must be called under a lock that serializes taking new references
1874 * to i_dio_count, usually by inode->i_mutex.
1876 void inode_dio_wait(struct inode
*inode
)
1878 if (atomic_read(&inode
->i_dio_count
))
1879 __inode_dio_wait(inode
);
1881 EXPORT_SYMBOL(inode_dio_wait
);
1884 * inode_dio_done - signal finish of a direct I/O requests
1885 * @inode: inode the direct I/O happens on
1887 * This is called once we've finished processing a direct I/O request,
1888 * and is used to wake up callers waiting for direct I/O to be quiesced.
1890 void inode_dio_done(struct inode
*inode
)
1892 if (atomic_dec_and_test(&inode
->i_dio_count
))
1893 wake_up_bit(&inode
->i_state
, __I_DIO_WAKEUP
);
1895 EXPORT_SYMBOL(inode_dio_done
);
1898 * inode_set_flags - atomically set some inode flags
1900 * Note: the caller should be holding i_mutex, or else be sure that
1901 * they have exclusive access to the inode structure (i.e., while the
1902 * inode is being instantiated). The reason for the cmpxchg() loop
1903 * --- which wouldn't be necessary if all code paths which modify
1904 * i_flags actually followed this rule, is that there is at least one
1905 * code path which doesn't today --- for example,
1906 * __generic_file_aio_write() calls file_remove_suid() without holding
1907 * i_mutex --- so we use cmpxchg() out of an abundance of caution.
1909 * In the long run, i_mutex is overkill, and we should probably look
1910 * at using the i_lock spinlock to protect i_flags, and then make sure
1911 * it is so documented in include/linux/fs.h and that all code follows
1912 * the locking convention!!
1914 void inode_set_flags(struct inode
*inode
, unsigned int flags
,
1917 unsigned int old_flags
, new_flags
;
1919 WARN_ON_ONCE(flags
& ~mask
);
1921 old_flags
= ACCESS_ONCE(inode
->i_flags
);
1922 new_flags
= (old_flags
& ~mask
) | flags
;
1923 } while (unlikely(cmpxchg(&inode
->i_flags
, old_flags
,
1924 new_flags
) != old_flags
));
1926 EXPORT_SYMBOL(inode_set_flags
);