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
;
168 mapping_set_gfp_mask(mapping
, (GFP_HIGHUSER_MOVABLE
& ~__GFP_HIGHMEM
) | __GFP_SLOWHIGHMEM
);
169 //mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
171 mapping
->private_data
= NULL
;
172 mapping
->backing_dev_info
= &default_backing_dev_info
;
173 mapping
->writeback_index
= 0;
176 * If the block_device provides a backing_dev_info for client
177 * inodes then use that. Otherwise the inode share the bdev's
181 struct backing_dev_info
*bdi
;
183 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
184 mapping
->backing_dev_info
= bdi
;
186 inode
->i_private
= NULL
;
187 inode
->i_mapping
= mapping
;
188 INIT_HLIST_HEAD(&inode
->i_dentry
); /* buggered by rcu freeing */
189 #ifdef CONFIG_FS_POSIX_ACL
190 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
193 #ifdef CONFIG_FSNOTIFY
194 inode
->i_fsnotify_mask
= 0;
197 this_cpu_inc(nr_inodes
);
203 EXPORT_SYMBOL(inode_init_always
);
205 static struct inode
*alloc_inode(struct super_block
*sb
)
209 if (sb
->s_op
->alloc_inode
)
210 inode
= sb
->s_op
->alloc_inode(sb
);
212 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
217 if (unlikely(inode_init_always(sb
, inode
))) {
218 if (inode
->i_sb
->s_op
->destroy_inode
)
219 inode
->i_sb
->s_op
->destroy_inode(inode
);
221 kmem_cache_free(inode_cachep
, inode
);
228 void free_inode_nonrcu(struct inode
*inode
)
230 kmem_cache_free(inode_cachep
, inode
);
232 EXPORT_SYMBOL(free_inode_nonrcu
);
234 void __destroy_inode(struct inode
*inode
)
236 BUG_ON(inode_has_buffers(inode
));
237 security_inode_free(inode
);
238 fsnotify_inode_delete(inode
);
239 if (!inode
->i_nlink
) {
240 WARN_ON(atomic_long_read(&inode
->i_sb
->s_remove_count
) == 0);
241 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
244 #ifdef CONFIG_FS_POSIX_ACL
245 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
246 posix_acl_release(inode
->i_acl
);
247 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
248 posix_acl_release(inode
->i_default_acl
);
250 this_cpu_dec(nr_inodes
);
252 EXPORT_SYMBOL(__destroy_inode
);
254 static void i_callback(struct rcu_head
*head
)
256 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
257 kmem_cache_free(inode_cachep
, inode
);
260 static void destroy_inode(struct inode
*inode
)
262 BUG_ON(!list_empty(&inode
->i_lru
));
263 __destroy_inode(inode
);
264 if (inode
->i_sb
->s_op
->destroy_inode
)
265 inode
->i_sb
->s_op
->destroy_inode(inode
);
267 call_rcu(&inode
->i_rcu
, i_callback
);
271 * drop_nlink - directly drop an inode's link count
274 * This is a low-level filesystem helper to replace any
275 * direct filesystem manipulation of i_nlink. In cases
276 * where we are attempting to track writes to the
277 * filesystem, a decrement to zero means an imminent
278 * write when the file is truncated and actually unlinked
281 void drop_nlink(struct inode
*inode
)
283 WARN_ON(inode
->i_nlink
== 0);
286 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
288 EXPORT_SYMBOL(drop_nlink
);
291 * clear_nlink - directly zero an inode's link count
294 * This is a low-level filesystem helper to replace any
295 * direct filesystem manipulation of i_nlink. See
296 * drop_nlink() for why we care about i_nlink hitting zero.
298 void clear_nlink(struct inode
*inode
)
300 if (inode
->i_nlink
) {
301 inode
->__i_nlink
= 0;
302 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
305 EXPORT_SYMBOL(clear_nlink
);
308 * set_nlink - directly set an inode's link count
310 * @nlink: new nlink (should be non-zero)
312 * This is a low-level filesystem helper to replace any
313 * direct filesystem manipulation of i_nlink.
315 void set_nlink(struct inode
*inode
, unsigned int nlink
)
320 /* Yes, some filesystems do change nlink from zero to one */
321 if (inode
->i_nlink
== 0)
322 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
324 inode
->__i_nlink
= nlink
;
327 EXPORT_SYMBOL(set_nlink
);
330 * inc_nlink - directly increment an inode's link count
333 * This is a low-level filesystem helper to replace any
334 * direct filesystem manipulation of i_nlink. Currently,
335 * it is only here for parity with dec_nlink().
337 void inc_nlink(struct inode
*inode
)
339 if (WARN_ON(inode
->i_nlink
== 0))
340 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
344 EXPORT_SYMBOL(inc_nlink
);
346 void address_space_init_once(struct address_space
*mapping
)
348 memset(mapping
, 0, sizeof(*mapping
));
349 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
350 spin_lock_init(&mapping
->tree_lock
);
351 mutex_init(&mapping
->i_mmap_mutex
);
352 INIT_LIST_HEAD(&mapping
->private_list
);
353 spin_lock_init(&mapping
->private_lock
);
354 mapping
->i_mmap
= RB_ROOT
;
355 INIT_LIST_HEAD(&mapping
->i_mmap_nonlinear
);
357 EXPORT_SYMBOL(address_space_init_once
);
360 * These are initializations that only need to be done
361 * once, because the fields are idempotent across use
362 * of the inode, so let the slab aware of that.
364 void inode_init_once(struct inode
*inode
)
366 memset(inode
, 0, sizeof(*inode
));
367 INIT_HLIST_NODE(&inode
->i_hash
);
368 INIT_LIST_HEAD(&inode
->i_devices
);
369 INIT_LIST_HEAD(&inode
->i_wb_list
);
370 INIT_LIST_HEAD(&inode
->i_lru
);
371 address_space_init_once(&inode
->i_data
);
372 i_size_ordered_init(inode
);
373 #ifdef CONFIG_FSNOTIFY
374 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
377 EXPORT_SYMBOL(inode_init_once
);
379 static void init_once(void *foo
)
381 struct inode
*inode
= (struct inode
*) foo
;
383 inode_init_once(inode
);
387 * inode->i_lock must be held
389 void __iget(struct inode
*inode
)
391 atomic_inc(&inode
->i_count
);
395 * get additional reference to inode; caller must already hold one.
397 void ihold(struct inode
*inode
)
399 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
401 EXPORT_SYMBOL(ihold
);
403 static void inode_lru_list_add(struct inode
*inode
)
405 spin_lock(&inode
->i_sb
->s_inode_lru_lock
);
406 if (list_empty(&inode
->i_lru
)) {
407 list_add(&inode
->i_lru
, &inode
->i_sb
->s_inode_lru
);
408 inode
->i_sb
->s_nr_inodes_unused
++;
409 this_cpu_inc(nr_unused
);
411 spin_unlock(&inode
->i_sb
->s_inode_lru_lock
);
415 * Add inode to LRU if needed (inode is unused and clean).
417 * Needs inode->i_lock held.
419 void inode_add_lru(struct inode
*inode
)
421 if (!(inode
->i_state
& (I_DIRTY
| I_SYNC
| I_FREEING
| I_WILL_FREE
)) &&
422 !atomic_read(&inode
->i_count
) && inode
->i_sb
->s_flags
& MS_ACTIVE
)
423 inode_lru_list_add(inode
);
427 static void inode_lru_list_del(struct inode
*inode
)
429 spin_lock(&inode
->i_sb
->s_inode_lru_lock
);
430 if (!list_empty(&inode
->i_lru
)) {
431 list_del_init(&inode
->i_lru
);
432 inode
->i_sb
->s_nr_inodes_unused
--;
433 this_cpu_dec(nr_unused
);
435 spin_unlock(&inode
->i_sb
->s_inode_lru_lock
);
439 * inode_sb_list_add - add inode to the superblock list of inodes
440 * @inode: inode to add
442 void inode_sb_list_add(struct inode
*inode
)
444 spin_lock(&inode_sb_list_lock
);
445 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
446 spin_unlock(&inode_sb_list_lock
);
448 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
450 static inline void inode_sb_list_del(struct inode
*inode
)
452 if (!list_empty(&inode
->i_sb_list
)) {
453 spin_lock(&inode_sb_list_lock
);
454 list_del_init(&inode
->i_sb_list
);
455 spin_unlock(&inode_sb_list_lock
);
459 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
463 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
465 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> i_hash_shift
);
466 return tmp
& i_hash_mask
;
470 * __insert_inode_hash - hash an inode
471 * @inode: unhashed inode
472 * @hashval: unsigned long value used to locate this object in the
475 * Add an inode to the inode hash for this superblock.
477 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
479 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
481 spin_lock(&inode_hash_lock
);
482 spin_lock(&inode
->i_lock
);
483 hlist_add_head(&inode
->i_hash
, b
);
484 spin_unlock(&inode
->i_lock
);
485 spin_unlock(&inode_hash_lock
);
487 EXPORT_SYMBOL(__insert_inode_hash
);
490 * __remove_inode_hash - remove an inode from the hash
491 * @inode: inode to unhash
493 * Remove an inode from the superblock.
495 void __remove_inode_hash(struct inode
*inode
)
497 spin_lock(&inode_hash_lock
);
498 spin_lock(&inode
->i_lock
);
499 hlist_del_init(&inode
->i_hash
);
500 spin_unlock(&inode
->i_lock
);
501 spin_unlock(&inode_hash_lock
);
503 EXPORT_SYMBOL(__remove_inode_hash
);
505 void clear_inode(struct inode
*inode
)
509 * We have to cycle tree_lock here because reclaim can be still in the
510 * process of removing the last page (in __delete_from_page_cache())
511 * and we must not free mapping under it.
513 spin_lock_irq(&inode
->i_data
.tree_lock
);
514 BUG_ON(inode
->i_data
.nrpages
);
515 spin_unlock_irq(&inode
->i_data
.tree_lock
);
516 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
517 BUG_ON(!(inode
->i_state
& I_FREEING
));
518 BUG_ON(inode
->i_state
& I_CLEAR
);
519 /* don't need i_lock here, no concurrent mods to i_state */
520 inode
->i_state
= I_FREEING
| I_CLEAR
;
522 EXPORT_SYMBOL(clear_inode
);
525 * Free the inode passed in, removing it from the lists it is still connected
526 * to. We remove any pages still attached to the inode and wait for any IO that
527 * is still in progress before finally destroying the inode.
529 * An inode must already be marked I_FREEING so that we avoid the inode being
530 * moved back onto lists if we race with other code that manipulates the lists
531 * (e.g. writeback_single_inode). The caller is responsible for setting this.
533 * An inode must already be removed from the LRU list before being evicted from
534 * the cache. This should occur atomically with setting the I_FREEING state
535 * flag, so no inodes here should ever be on the LRU when being evicted.
537 static void evict(struct inode
*inode
)
539 const struct super_operations
*op
= inode
->i_sb
->s_op
;
541 BUG_ON(!(inode
->i_state
& I_FREEING
));
542 BUG_ON(!list_empty(&inode
->i_lru
));
544 if (!list_empty(&inode
->i_wb_list
))
545 inode_wb_list_del(inode
);
547 inode_sb_list_del(inode
);
550 * Wait for flusher thread to be done with the inode so that filesystem
551 * does not start destroying it while writeback is still running. Since
552 * the inode has I_FREEING set, flusher thread won't start new work on
553 * the inode. We just have to wait for running writeback to finish.
555 inode_wait_for_writeback(inode
);
557 if (op
->evict_inode
) {
558 op
->evict_inode(inode
);
560 if (inode
->i_data
.nrpages
)
561 truncate_inode_pages(&inode
->i_data
, 0);
564 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
566 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
569 remove_inode_hash(inode
);
571 spin_lock(&inode
->i_lock
);
572 wake_up_bit(&inode
->i_state
, __I_NEW
);
573 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
574 spin_unlock(&inode
->i_lock
);
576 destroy_inode(inode
);
580 * dispose_list - dispose of the contents of a local list
581 * @head: the head of the list to free
583 * Dispose-list gets a local list with local inodes in it, so it doesn't
584 * need to worry about list corruption and SMP locks.
586 static void dispose_list(struct list_head
*head
)
588 while (!list_empty(head
)) {
591 inode
= list_first_entry(head
, struct inode
, i_lru
);
592 list_del_init(&inode
->i_lru
);
599 * evict_inodes - evict all evictable inodes for a superblock
600 * @sb: superblock to operate on
602 * Make sure that no inodes with zero refcount are retained. This is
603 * called by superblock shutdown after having MS_ACTIVE flag removed,
604 * so any inode reaching zero refcount during or after that call will
605 * be immediately evicted.
607 void evict_inodes(struct super_block
*sb
)
609 struct inode
*inode
, *next
;
612 spin_lock(&inode_sb_list_lock
);
613 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
614 if (atomic_read(&inode
->i_count
))
617 spin_lock(&inode
->i_lock
);
618 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
619 spin_unlock(&inode
->i_lock
);
623 inode
->i_state
|= I_FREEING
;
624 inode_lru_list_del(inode
);
625 spin_unlock(&inode
->i_lock
);
626 list_add(&inode
->i_lru
, &dispose
);
628 spin_unlock(&inode_sb_list_lock
);
630 dispose_list(&dispose
);
634 * invalidate_inodes - attempt to free all inodes on a superblock
635 * @sb: superblock to operate on
636 * @kill_dirty: flag to guide handling of dirty inodes
638 * Attempts to free all inodes for a given superblock. If there were any
639 * busy inodes return a non-zero value, else zero.
640 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
643 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
646 struct inode
*inode
, *next
;
649 spin_lock(&inode_sb_list_lock
);
650 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
651 spin_lock(&inode
->i_lock
);
652 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
653 spin_unlock(&inode
->i_lock
);
656 if (inode
->i_state
& I_DIRTY
&& !kill_dirty
) {
657 spin_unlock(&inode
->i_lock
);
661 if (atomic_read(&inode
->i_count
)) {
662 spin_unlock(&inode
->i_lock
);
667 inode
->i_state
|= I_FREEING
;
668 inode_lru_list_del(inode
);
669 spin_unlock(&inode
->i_lock
);
670 list_add(&inode
->i_lru
, &dispose
);
672 spin_unlock(&inode_sb_list_lock
);
674 dispose_list(&dispose
);
679 static int can_unuse(struct inode
*inode
)
681 if (inode
->i_state
& ~I_REFERENCED
)
683 if (inode_has_buffers(inode
))
685 if (atomic_read(&inode
->i_count
))
687 if (inode
->i_data
.nrpages
)
693 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
694 * This is called from the superblock shrinker function with a number of inodes
695 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
696 * then are freed outside inode_lock by dispose_list().
698 * Any inodes which are pinned purely because of attached pagecache have their
699 * pagecache removed. If the inode has metadata buffers attached to
700 * mapping->private_list then try to remove them.
702 * If the inode has the I_REFERENCED flag set, then it means that it has been
703 * used recently - the flag is set in iput_final(). When we encounter such an
704 * inode, clear the flag and move it to the back of the LRU so it gets another
705 * pass through the LRU before it gets reclaimed. This is necessary because of
706 * the fact we are doing lazy LRU updates to minimise lock contention so the
707 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
708 * with this flag set because they are the inodes that are out of order.
710 void prune_icache_sb(struct super_block
*sb
, int nr_to_scan
)
714 unsigned long reap
= 0;
716 spin_lock(&sb
->s_inode_lru_lock
);
717 for (nr_scanned
= nr_to_scan
; nr_scanned
>= 0; nr_scanned
--) {
720 if (list_empty(&sb
->s_inode_lru
))
723 inode
= list_entry(sb
->s_inode_lru
.prev
, struct inode
, i_lru
);
726 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
727 * so use a trylock. If we fail to get the lock, just move the
728 * inode to the back of the list so we don't spin on it.
730 if (!spin_trylock(&inode
->i_lock
)) {
731 list_move(&inode
->i_lru
, &sb
->s_inode_lru
);
736 * Referenced or dirty inodes are still in use. Give them
737 * another pass through the LRU as we canot reclaim them now.
739 if (atomic_read(&inode
->i_count
) ||
740 (inode
->i_state
& ~I_REFERENCED
)) {
741 list_del_init(&inode
->i_lru
);
742 spin_unlock(&inode
->i_lock
);
743 sb
->s_nr_inodes_unused
--;
744 this_cpu_dec(nr_unused
);
748 /* recently referenced inodes get one more pass */
749 if (inode
->i_state
& I_REFERENCED
) {
750 inode
->i_state
&= ~I_REFERENCED
;
751 list_move(&inode
->i_lru
, &sb
->s_inode_lru
);
752 spin_unlock(&inode
->i_lock
);
755 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
757 spin_unlock(&inode
->i_lock
);
758 spin_unlock(&sb
->s_inode_lru_lock
);
759 if (remove_inode_buffers(inode
))
760 reap
+= invalidate_mapping_pages(&inode
->i_data
,
763 spin_lock(&sb
->s_inode_lru_lock
);
765 if (inode
!= list_entry(sb
->s_inode_lru
.next
,
766 struct inode
, i_lru
))
767 continue; /* wrong inode or list_empty */
768 /* avoid lock inversions with trylock */
769 if (!spin_trylock(&inode
->i_lock
))
771 if (!can_unuse(inode
)) {
772 spin_unlock(&inode
->i_lock
);
776 WARN_ON(inode
->i_state
& I_NEW
);
777 inode
->i_state
|= I_FREEING
;
778 spin_unlock(&inode
->i_lock
);
780 list_move(&inode
->i_lru
, &freeable
);
781 sb
->s_nr_inodes_unused
--;
782 this_cpu_dec(nr_unused
);
784 if (current_is_kswapd())
785 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
787 __count_vm_events(PGINODESTEAL
, reap
);
788 spin_unlock(&sb
->s_inode_lru_lock
);
789 if (current
->reclaim_state
)
790 current
->reclaim_state
->reclaimed_slab
+= reap
;
792 dispose_list(&freeable
);
795 static void __wait_on_freeing_inode(struct inode
*inode
);
797 * Called with the inode lock held.
799 static struct inode
*find_inode(struct super_block
*sb
,
800 struct hlist_head
*head
,
801 int (*test
)(struct inode
*, void *),
804 struct inode
*inode
= NULL
;
807 hlist_for_each_entry(inode
, head
, i_hash
) {
808 spin_lock(&inode
->i_lock
);
809 if (inode
->i_sb
!= sb
) {
810 spin_unlock(&inode
->i_lock
);
813 if (!test(inode
, data
)) {
814 spin_unlock(&inode
->i_lock
);
817 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
818 __wait_on_freeing_inode(inode
);
822 spin_unlock(&inode
->i_lock
);
829 * find_inode_fast is the fast path version of find_inode, see the comment at
830 * iget_locked for details.
832 static struct inode
*find_inode_fast(struct super_block
*sb
,
833 struct hlist_head
*head
, unsigned long ino
)
835 struct inode
*inode
= NULL
;
838 hlist_for_each_entry(inode
, head
, i_hash
) {
839 spin_lock(&inode
->i_lock
);
840 if (inode
->i_ino
!= ino
) {
841 spin_unlock(&inode
->i_lock
);
844 if (inode
->i_sb
!= sb
) {
845 spin_unlock(&inode
->i_lock
);
848 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
849 __wait_on_freeing_inode(inode
);
853 spin_unlock(&inode
->i_lock
);
860 * Each cpu owns a range of LAST_INO_BATCH numbers.
861 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
862 * to renew the exhausted range.
864 * This does not significantly increase overflow rate because every CPU can
865 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
866 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
867 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
868 * overflow rate by 2x, which does not seem too significant.
870 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
871 * error if st_ino won't fit in target struct field. Use 32bit counter
872 * here to attempt to avoid that.
874 #define LAST_INO_BATCH 1024
875 static DEFINE_PER_CPU(unsigned int, last_ino
);
877 unsigned int get_next_ino(void)
879 unsigned int *p
= &get_cpu_var(last_ino
);
880 unsigned int res
= *p
;
883 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
884 static atomic_t shared_last_ino
;
885 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
887 res
= next
- LAST_INO_BATCH
;
892 put_cpu_var(last_ino
);
895 EXPORT_SYMBOL(get_next_ino
);
898 * new_inode_pseudo - obtain an inode
901 * Allocates a new inode for given superblock.
902 * Inode wont be chained in superblock s_inodes list
904 * - fs can't be unmount
905 * - quotas, fsnotify, writeback can't work
907 struct inode
*new_inode_pseudo(struct super_block
*sb
)
909 struct inode
*inode
= alloc_inode(sb
);
912 spin_lock(&inode
->i_lock
);
914 spin_unlock(&inode
->i_lock
);
915 INIT_LIST_HEAD(&inode
->i_sb_list
);
921 * new_inode - obtain an inode
924 * Allocates a new inode for given superblock. The default gfp_mask
925 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
926 * If HIGHMEM pages are unsuitable or it is known that pages allocated
927 * for the page cache are not reclaimable or migratable,
928 * mapping_set_gfp_mask() must be called with suitable flags on the
929 * newly created inode's mapping
932 struct inode
*new_inode(struct super_block
*sb
)
936 spin_lock_prefetch(&inode_sb_list_lock
);
938 inode
= new_inode_pseudo(sb
);
940 inode_sb_list_add(inode
);
943 EXPORT_SYMBOL(new_inode
);
945 #ifdef CONFIG_DEBUG_LOCK_ALLOC
946 void lockdep_annotate_inode_mutex_key(struct inode
*inode
)
948 if (S_ISDIR(inode
->i_mode
)) {
949 struct file_system_type
*type
= inode
->i_sb
->s_type
;
951 /* Set new key only if filesystem hasn't already changed it */
952 if (lockdep_match_class(&inode
->i_mutex
, &type
->i_mutex_key
)) {
954 * ensure nobody is actually holding i_mutex
956 mutex_destroy(&inode
->i_mutex
);
957 mutex_init(&inode
->i_mutex
);
958 lockdep_set_class(&inode
->i_mutex
,
959 &type
->i_mutex_dir_key
);
963 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key
);
967 * unlock_new_inode - clear the I_NEW state and wake up any waiters
968 * @inode: new inode to unlock
970 * Called when the inode is fully initialised to clear the new state of the
971 * inode and wake up anyone waiting for the inode to finish initialisation.
973 void unlock_new_inode(struct inode
*inode
)
975 lockdep_annotate_inode_mutex_key(inode
);
976 spin_lock(&inode
->i_lock
);
977 WARN_ON(!(inode
->i_state
& I_NEW
));
978 inode
->i_state
&= ~I_NEW
;
980 wake_up_bit(&inode
->i_state
, __I_NEW
);
981 spin_unlock(&inode
->i_lock
);
983 EXPORT_SYMBOL(unlock_new_inode
);
986 * iget5_locked - obtain an inode from a mounted file system
987 * @sb: super block of file system
988 * @hashval: hash value (usually inode number) to get
989 * @test: callback used for comparisons between inodes
990 * @set: callback used to initialize a new struct inode
991 * @data: opaque data pointer to pass to @test and @set
993 * Search for the inode specified by @hashval and @data in the inode cache,
994 * and if present it is return it with an increased reference count. This is
995 * a generalized version of iget_locked() for file systems where the inode
996 * number is not sufficient for unique identification of an inode.
998 * If the inode is not in cache, allocate a new inode and return it locked,
999 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1000 * before unlocking it via unlock_new_inode().
1002 * Note both @test and @set are called with the inode_hash_lock held, so can't
1005 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1006 int (*test
)(struct inode
*, void *),
1007 int (*set
)(struct inode
*, void *), void *data
)
1009 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1010 struct inode
*inode
;
1012 spin_lock(&inode_hash_lock
);
1013 inode
= find_inode(sb
, head
, test
, data
);
1014 spin_unlock(&inode_hash_lock
);
1017 wait_on_inode(inode
);
1021 inode
= alloc_inode(sb
);
1025 spin_lock(&inode_hash_lock
);
1026 /* We released the lock, so.. */
1027 old
= find_inode(sb
, head
, test
, data
);
1029 if (set(inode
, data
))
1032 spin_lock(&inode
->i_lock
);
1033 inode
->i_state
= I_NEW
;
1034 hlist_add_head(&inode
->i_hash
, head
);
1035 spin_unlock(&inode
->i_lock
);
1036 inode_sb_list_add(inode
);
1037 spin_unlock(&inode_hash_lock
);
1039 /* Return the locked inode with I_NEW set, the
1040 * caller is responsible for filling in the contents
1046 * Uhhuh, somebody else created the same inode under
1047 * us. Use the old inode instead of the one we just
1050 spin_unlock(&inode_hash_lock
);
1051 destroy_inode(inode
);
1053 wait_on_inode(inode
);
1058 spin_unlock(&inode_hash_lock
);
1059 destroy_inode(inode
);
1062 EXPORT_SYMBOL(iget5_locked
);
1065 * iget_locked - obtain an inode from a mounted file system
1066 * @sb: super block of file system
1067 * @ino: inode number to get
1069 * Search for the inode specified by @ino in the inode cache and if present
1070 * return it with an increased reference count. This is for file systems
1071 * where the inode number is sufficient for unique identification of an inode.
1073 * If the inode is not in cache, allocate a new inode and return it locked,
1074 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1075 * before unlocking it via unlock_new_inode().
1077 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1079 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1080 struct inode
*inode
;
1082 spin_lock(&inode_hash_lock
);
1083 inode
= find_inode_fast(sb
, head
, ino
);
1084 spin_unlock(&inode_hash_lock
);
1086 wait_on_inode(inode
);
1090 inode
= alloc_inode(sb
);
1094 spin_lock(&inode_hash_lock
);
1095 /* We released the lock, so.. */
1096 old
= find_inode_fast(sb
, head
, ino
);
1099 spin_lock(&inode
->i_lock
);
1100 inode
->i_state
= I_NEW
;
1101 hlist_add_head(&inode
->i_hash
, head
);
1102 spin_unlock(&inode
->i_lock
);
1103 inode_sb_list_add(inode
);
1104 spin_unlock(&inode_hash_lock
);
1106 /* Return the locked inode with I_NEW set, the
1107 * caller is responsible for filling in the contents
1113 * Uhhuh, somebody else created the same inode under
1114 * us. Use the old inode instead of the one we just
1117 spin_unlock(&inode_hash_lock
);
1118 destroy_inode(inode
);
1120 wait_on_inode(inode
);
1124 EXPORT_SYMBOL(iget_locked
);
1127 * search the inode cache for a matching inode number.
1128 * If we find one, then the inode number we are trying to
1129 * allocate is not unique and so we should not use it.
1131 * Returns 1 if the inode number is unique, 0 if it is not.
1133 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1135 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1136 struct inode
*inode
;
1138 spin_lock(&inode_hash_lock
);
1139 hlist_for_each_entry(inode
, b
, i_hash
) {
1140 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
) {
1141 spin_unlock(&inode_hash_lock
);
1145 spin_unlock(&inode_hash_lock
);
1151 * iunique - get a unique inode number
1153 * @max_reserved: highest reserved inode number
1155 * Obtain an inode number that is unique on the system for a given
1156 * superblock. This is used by file systems that have no natural
1157 * permanent inode numbering system. An inode number is returned that
1158 * is higher than the reserved limit but unique.
1161 * With a large number of inodes live on the file system this function
1162 * currently becomes quite slow.
1164 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1167 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1168 * error if st_ino won't fit in target struct field. Use 32bit counter
1169 * here to attempt to avoid that.
1171 static DEFINE_SPINLOCK(iunique_lock
);
1172 static unsigned int counter
;
1175 spin_lock(&iunique_lock
);
1177 if (counter
<= max_reserved
)
1178 counter
= max_reserved
+ 1;
1180 } while (!test_inode_iunique(sb
, res
));
1181 spin_unlock(&iunique_lock
);
1185 EXPORT_SYMBOL(iunique
);
1187 struct inode
*igrab(struct inode
*inode
)
1189 spin_lock(&inode
->i_lock
);
1190 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1192 spin_unlock(&inode
->i_lock
);
1194 spin_unlock(&inode
->i_lock
);
1196 * Handle the case where s_op->clear_inode is not been
1197 * called yet, and somebody is calling igrab
1198 * while the inode is getting freed.
1204 EXPORT_SYMBOL(igrab
);
1207 * ilookup5_nowait - search for an inode in the inode cache
1208 * @sb: super block of file system to search
1209 * @hashval: hash value (usually inode number) to search for
1210 * @test: callback used for comparisons between inodes
1211 * @data: opaque data pointer to pass to @test
1213 * Search for the inode specified by @hashval and @data in the inode cache.
1214 * If the inode is in the cache, the inode is returned with an incremented
1217 * Note: I_NEW is not waited upon so you have to be very careful what you do
1218 * with the returned inode. You probably should be using ilookup5() instead.
1220 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1222 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1223 int (*test
)(struct inode
*, void *), void *data
)
1225 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1226 struct inode
*inode
;
1228 spin_lock(&inode_hash_lock
);
1229 inode
= find_inode(sb
, head
, test
, data
);
1230 spin_unlock(&inode_hash_lock
);
1234 EXPORT_SYMBOL(ilookup5_nowait
);
1237 * ilookup5 - search for an inode in the inode cache
1238 * @sb: super block of file system to search
1239 * @hashval: hash value (usually inode number) to search for
1240 * @test: callback used for comparisons between inodes
1241 * @data: opaque data pointer to pass to @test
1243 * Search for the inode specified by @hashval and @data in the inode cache,
1244 * and if the inode is in the cache, return the inode with an incremented
1245 * reference count. Waits on I_NEW before returning the inode.
1246 * returned with an incremented reference count.
1248 * This is a generalized version of ilookup() for file systems where the
1249 * inode number is not sufficient for unique identification of an inode.
1251 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1253 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1254 int (*test
)(struct inode
*, void *), void *data
)
1256 struct inode
*inode
= ilookup5_nowait(sb
, hashval
, test
, data
);
1259 wait_on_inode(inode
);
1262 EXPORT_SYMBOL(ilookup5
);
1265 * ilookup - search for an inode in the inode cache
1266 * @sb: super block of file system to search
1267 * @ino: inode number to search for
1269 * Search for the inode @ino in the inode cache, and if the inode is in the
1270 * cache, the inode is returned with an incremented reference count.
1272 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1274 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1275 struct inode
*inode
;
1277 spin_lock(&inode_hash_lock
);
1278 inode
= find_inode_fast(sb
, head
, ino
);
1279 spin_unlock(&inode_hash_lock
);
1282 wait_on_inode(inode
);
1285 EXPORT_SYMBOL(ilookup
);
1287 int insert_inode_locked(struct inode
*inode
)
1289 struct super_block
*sb
= inode
->i_sb
;
1290 ino_t ino
= inode
->i_ino
;
1291 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1294 struct inode
*old
= NULL
;
1295 spin_lock(&inode_hash_lock
);
1296 hlist_for_each_entry(old
, head
, i_hash
) {
1297 if (old
->i_ino
!= ino
)
1299 if (old
->i_sb
!= sb
)
1301 spin_lock(&old
->i_lock
);
1302 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1303 spin_unlock(&old
->i_lock
);
1309 spin_lock(&inode
->i_lock
);
1310 inode
->i_state
|= I_NEW
;
1311 hlist_add_head(&inode
->i_hash
, head
);
1312 spin_unlock(&inode
->i_lock
);
1313 spin_unlock(&inode_hash_lock
);
1317 spin_unlock(&old
->i_lock
);
1318 spin_unlock(&inode_hash_lock
);
1320 if (unlikely(!inode_unhashed(old
))) {
1327 EXPORT_SYMBOL(insert_inode_locked
);
1329 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1330 int (*test
)(struct inode
*, void *), void *data
)
1332 struct super_block
*sb
= inode
->i_sb
;
1333 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1336 struct inode
*old
= NULL
;
1338 spin_lock(&inode_hash_lock
);
1339 hlist_for_each_entry(old
, head
, i_hash
) {
1340 if (old
->i_sb
!= sb
)
1342 if (!test(old
, data
))
1344 spin_lock(&old
->i_lock
);
1345 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1346 spin_unlock(&old
->i_lock
);
1352 spin_lock(&inode
->i_lock
);
1353 inode
->i_state
|= I_NEW
;
1354 hlist_add_head(&inode
->i_hash
, head
);
1355 spin_unlock(&inode
->i_lock
);
1356 spin_unlock(&inode_hash_lock
);
1360 spin_unlock(&old
->i_lock
);
1361 spin_unlock(&inode_hash_lock
);
1363 if (unlikely(!inode_unhashed(old
))) {
1370 EXPORT_SYMBOL(insert_inode_locked4
);
1373 int generic_delete_inode(struct inode
*inode
)
1377 EXPORT_SYMBOL(generic_delete_inode
);
1380 * Called when we're dropping the last reference
1383 * Call the FS "drop_inode()" function, defaulting to
1384 * the legacy UNIX filesystem behaviour. If it tells
1385 * us to evict inode, do so. Otherwise, retain inode
1386 * in cache if fs is alive, sync and evict if fs is
1389 static void iput_final(struct inode
*inode
)
1391 struct super_block
*sb
= inode
->i_sb
;
1392 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1395 WARN_ON(inode
->i_state
& I_NEW
);
1398 drop
= op
->drop_inode(inode
);
1400 drop
= generic_drop_inode(inode
);
1402 if (!drop
&& (sb
->s_flags
& MS_ACTIVE
)) {
1403 inode
->i_state
|= I_REFERENCED
;
1404 inode_add_lru(inode
);
1405 spin_unlock(&inode
->i_lock
);
1410 inode
->i_state
|= I_WILL_FREE
;
1411 spin_unlock(&inode
->i_lock
);
1412 write_inode_now(inode
, 1);
1413 spin_lock(&inode
->i_lock
);
1414 WARN_ON(inode
->i_state
& I_NEW
);
1415 inode
->i_state
&= ~I_WILL_FREE
;
1418 inode
->i_state
|= I_FREEING
;
1419 if (!list_empty(&inode
->i_lru
))
1420 inode_lru_list_del(inode
);
1421 spin_unlock(&inode
->i_lock
);
1427 * iput - put an inode
1428 * @inode: inode to put
1430 * Puts an inode, dropping its usage count. If the inode use count hits
1431 * zero, the inode is then freed and may also be destroyed.
1433 * Consequently, iput() can sleep.
1435 void iput(struct inode
*inode
)
1438 BUG_ON(inode
->i_state
& I_CLEAR
);
1440 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
))
1444 EXPORT_SYMBOL(iput
);
1447 * bmap - find a block number in a file
1448 * @inode: inode of file
1449 * @block: block to find
1451 * Returns the block number on the device holding the inode that
1452 * is the disk block number for the block of the file requested.
1453 * That is, asked for block 4 of inode 1 the function will return the
1454 * disk block relative to the disk start that holds that block of the
1457 sector_t
bmap(struct inode
*inode
, sector_t block
)
1460 if (inode
->i_mapping
->a_ops
->bmap
)
1461 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1464 EXPORT_SYMBOL(bmap
);
1467 * With relative atime, only update atime if the previous atime is
1468 * earlier than either the ctime or mtime or if at least a day has
1469 * passed since the last atime update.
1471 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1472 struct timespec now
)
1475 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1478 * Is mtime younger than atime? If yes, update atime:
1480 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1483 * Is ctime younger than atime? If yes, update atime:
1485 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1489 * Is the previous atime value older than a day? If yes,
1492 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1495 * Good, we can skip the atime update:
1501 * This does the actual work of updating an inodes time or version. Must have
1502 * had called mnt_want_write() before calling this.
1504 static int update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1506 if (inode
->i_op
->update_time
)
1507 return inode
->i_op
->update_time(inode
, time
, flags
);
1509 if (flags
& S_ATIME
)
1510 inode
->i_atime
= *time
;
1511 if (flags
& S_VERSION
)
1512 inode_inc_iversion(inode
);
1513 if (flags
& S_CTIME
)
1514 inode
->i_ctime
= *time
;
1515 if (flags
& S_MTIME
)
1516 inode
->i_mtime
= *time
;
1517 mark_inode_dirty_sync(inode
);
1522 * touch_atime - update the access time
1523 * @path: the &struct path to update
1525 * Update the accessed time on an inode and mark it for writeback.
1526 * This function automatically handles read only file systems and media,
1527 * as well as the "noatime" flag and inode specific "noatime" markers.
1529 void touch_atime(struct path
*path
)
1531 struct vfsmount
*mnt
= path
->mnt
;
1532 struct inode
*inode
= path
->dentry
->d_inode
;
1533 struct timespec now
;
1535 if (inode
->i_flags
& S_NOATIME
)
1537 if (IS_NOATIME(inode
))
1539 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1542 if (mnt
->mnt_flags
& MNT_NOATIME
)
1544 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1547 now
= current_fs_time(inode
->i_sb
);
1549 if (!relatime_need_update(mnt
, inode
, now
))
1552 if (timespec_equal(&inode
->i_atime
, &now
))
1555 if (!sb_start_write_trylock(inode
->i_sb
))
1558 if (__mnt_want_write(mnt
))
1561 * File systems can error out when updating inodes if they need to
1562 * allocate new space to modify an inode (such is the case for
1563 * Btrfs), but since we touch atime while walking down the path we
1564 * really don't care if we failed to update the atime of the file,
1565 * so just ignore the return value.
1566 * We may also fail on filesystems that have the ability to make parts
1567 * of the fs read only, e.g. subvolumes in Btrfs.
1569 update_time(inode
, &now
, S_ATIME
);
1570 __mnt_drop_write(mnt
);
1572 sb_end_write(inode
->i_sb
);
1574 EXPORT_SYMBOL(touch_atime
);
1577 * The logic we want is
1579 * if suid or (sgid and xgrp)
1582 int should_remove_suid(struct dentry
*dentry
)
1584 umode_t mode
= dentry
->d_inode
->i_mode
;
1587 /* suid always must be killed */
1588 if (unlikely(mode
& S_ISUID
))
1589 kill
= ATTR_KILL_SUID
;
1592 * sgid without any exec bits is just a mandatory locking mark; leave
1593 * it alone. If some exec bits are set, it's a real sgid; kill it.
1595 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1596 kill
|= ATTR_KILL_SGID
;
1598 if (unlikely(kill
&& !capable(CAP_FSETID
) && S_ISREG(mode
)))
1603 EXPORT_SYMBOL(should_remove_suid
);
1605 static int __remove_suid(struct dentry
*dentry
, int kill
)
1607 struct iattr newattrs
;
1609 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1610 return notify_change(dentry
, &newattrs
);
1613 int file_remove_suid(struct file
*file
)
1615 struct dentry
*dentry
= file
->f_path
.dentry
;
1616 struct inode
*inode
= dentry
->d_inode
;
1621 /* Fast path for nothing security related */
1622 if (IS_NOSEC(inode
))
1625 killsuid
= should_remove_suid(dentry
);
1626 killpriv
= security_inode_need_killpriv(dentry
);
1631 error
= security_inode_killpriv(dentry
);
1632 if (!error
&& killsuid
)
1633 error
= __remove_suid(dentry
, killsuid
);
1634 if (!error
&& (inode
->i_sb
->s_flags
& MS_NOSEC
))
1635 inode
->i_flags
|= S_NOSEC
;
1639 EXPORT_SYMBOL(file_remove_suid
);
1642 * file_update_time - update mtime and ctime time
1643 * @file: file accessed
1645 * Update the mtime and ctime members of an inode and mark the inode
1646 * for writeback. Note that this function is meant exclusively for
1647 * usage in the file write path of filesystems, and filesystems may
1648 * choose to explicitly ignore update via this function with the
1649 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1650 * timestamps are handled by the server. This can return an error for
1651 * file systems who need to allocate space in order to update an inode.
1654 int file_update_time(struct file
*file
)
1656 struct inode
*inode
= file_inode(file
);
1657 struct timespec now
;
1661 /* First try to exhaust all avenues to not sync */
1662 if (IS_NOCMTIME(inode
))
1665 now
= current_fs_time(inode
->i_sb
);
1666 if (!timespec_equal(&inode
->i_mtime
, &now
))
1669 if (!timespec_equal(&inode
->i_ctime
, &now
))
1672 if (IS_I_VERSION(inode
))
1673 sync_it
|= S_VERSION
;
1678 /* Finally allowed to write? Takes lock. */
1679 if (__mnt_want_write_file(file
))
1682 ret
= update_time(inode
, &now
, sync_it
);
1683 __mnt_drop_write_file(file
);
1687 EXPORT_SYMBOL(file_update_time
);
1689 int inode_needs_sync(struct inode
*inode
)
1693 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1697 EXPORT_SYMBOL(inode_needs_sync
);
1699 int inode_wait(void *word
)
1704 EXPORT_SYMBOL(inode_wait
);
1707 * If we try to find an inode in the inode hash while it is being
1708 * deleted, we have to wait until the filesystem completes its
1709 * deletion before reporting that it isn't found. This function waits
1710 * until the deletion _might_ have completed. Callers are responsible
1711 * to recheck inode state.
1713 * It doesn't matter if I_NEW is not set initially, a call to
1714 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1717 static void __wait_on_freeing_inode(struct inode
*inode
)
1719 wait_queue_head_t
*wq
;
1720 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1721 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1722 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1723 spin_unlock(&inode
->i_lock
);
1724 spin_unlock(&inode_hash_lock
);
1726 finish_wait(wq
, &wait
.wait
);
1727 spin_lock(&inode_hash_lock
);
1730 static __initdata
unsigned long ihash_entries
;
1731 static int __init
set_ihash_entries(char *str
)
1735 ihash_entries
= simple_strtoul(str
, &str
, 0);
1738 __setup("ihash_entries=", set_ihash_entries
);
1741 * Initialize the waitqueues and inode hash table.
1743 void __init
inode_init_early(void)
1747 /* If hashes are distributed across NUMA nodes, defer
1748 * hash allocation until vmalloc space is available.
1754 alloc_large_system_hash("Inode-cache",
1755 sizeof(struct hlist_head
),
1764 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1765 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1768 void __init
inode_init(void)
1772 /* inode slab cache */
1773 inode_cachep
= kmem_cache_create("inode_cache",
1774 sizeof(struct inode
),
1776 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1780 /* Hash may have been set up in inode_init_early */
1785 alloc_large_system_hash("Inode-cache",
1786 sizeof(struct hlist_head
),
1795 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1796 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1799 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1801 inode
->i_mode
= mode
;
1802 if (S_ISCHR(mode
)) {
1803 inode
->i_fop
= &def_chr_fops
;
1804 inode
->i_rdev
= rdev
;
1805 } else if (S_ISBLK(mode
)) {
1806 inode
->i_fop
= &def_blk_fops
;
1807 inode
->i_rdev
= rdev
;
1808 } else if (S_ISFIFO(mode
))
1809 inode
->i_fop
= &pipefifo_fops
;
1810 else if (S_ISSOCK(mode
))
1811 inode
->i_fop
= &bad_sock_fops
;
1813 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1814 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1817 EXPORT_SYMBOL(init_special_inode
);
1820 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1822 * @dir: Directory inode
1823 * @mode: mode of the new inode
1825 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1828 inode
->i_uid
= current_fsuid();
1829 if (dir
&& dir
->i_mode
& S_ISGID
) {
1830 inode
->i_gid
= dir
->i_gid
;
1834 inode
->i_gid
= current_fsgid();
1835 inode
->i_mode
= mode
;
1837 EXPORT_SYMBOL(inode_init_owner
);
1840 * inode_owner_or_capable - check current task permissions to inode
1841 * @inode: inode being checked
1843 * Return true if current either has CAP_FOWNER in a namespace with the
1844 * inode owner uid mapped, or owns the file.
1846 bool inode_owner_or_capable(const struct inode
*inode
)
1848 struct user_namespace
*ns
;
1850 if (uid_eq(current_fsuid(), inode
->i_uid
))
1853 ns
= current_user_ns();
1854 if (ns_capable(ns
, CAP_FOWNER
) && kuid_has_mapping(ns
, inode
->i_uid
))
1858 EXPORT_SYMBOL(inode_owner_or_capable
);
1861 * Direct i/o helper functions
1863 static void __inode_dio_wait(struct inode
*inode
)
1865 wait_queue_head_t
*wq
= bit_waitqueue(&inode
->i_state
, __I_DIO_WAKEUP
);
1866 DEFINE_WAIT_BIT(q
, &inode
->i_state
, __I_DIO_WAKEUP
);
1869 prepare_to_wait(wq
, &q
.wait
, TASK_UNINTERRUPTIBLE
);
1870 if (atomic_read(&inode
->i_dio_count
))
1872 } while (atomic_read(&inode
->i_dio_count
));
1873 finish_wait(wq
, &q
.wait
);
1877 * inode_dio_wait - wait for outstanding DIO requests to finish
1878 * @inode: inode to wait for
1880 * Waits for all pending direct I/O requests to finish so that we can
1881 * proceed with a truncate or equivalent operation.
1883 * Must be called under a lock that serializes taking new references
1884 * to i_dio_count, usually by inode->i_mutex.
1886 void inode_dio_wait(struct inode
*inode
)
1888 if (atomic_read(&inode
->i_dio_count
))
1889 __inode_dio_wait(inode
);
1891 EXPORT_SYMBOL(inode_dio_wait
);
1894 * inode_dio_done - signal finish of a direct I/O requests
1895 * @inode: inode the direct I/O happens on
1897 * This is called once we've finished processing a direct I/O request,
1898 * and is used to wake up callers waiting for direct I/O to be quiesced.
1900 void inode_dio_done(struct inode
*inode
)
1902 if (atomic_dec_and_test(&inode
->i_dio_count
))
1903 wake_up_bit(&inode
->i_state
, __I_DIO_WAKEUP
);
1905 EXPORT_SYMBOL(inode_dio_done
);