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
);
117 static int no_open(struct inode
*inode
, struct file
*file
)
123 * inode_init_always - perform inode structure intialisation
124 * @sb: superblock inode belongs to
125 * @inode: inode to initialise
127 * These are initializations that need to be done on every inode
128 * allocation as the fields are not initialised by slab allocation.
130 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
132 static const struct inode_operations empty_iops
;
133 static const struct file_operations no_open_fops
= {.open
= no_open
};
134 struct address_space
*const mapping
= &inode
->i_data
;
137 inode
->i_blkbits
= sb
->s_blocksize_bits
;
139 atomic_set(&inode
->i_count
, 1);
140 inode
->i_op
= &empty_iops
;
141 inode
->i_fop
= &no_open_fops
;
142 inode
->__i_nlink
= 1;
143 inode
->i_opflags
= 0;
144 i_uid_write(inode
, 0);
145 i_gid_write(inode
, 0);
146 atomic_set(&inode
->i_writecount
, 0);
150 inode
->i_generation
= 0;
152 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
154 inode
->i_pipe
= NULL
;
155 inode
->i_bdev
= NULL
;
156 inode
->i_cdev
= NULL
;
158 inode
->dirtied_when
= 0;
160 if (security_inode_alloc(inode
))
162 spin_lock_init(&inode
->i_lock
);
163 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
165 mutex_init(&inode
->i_mutex
);
166 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
168 atomic_set(&inode
->i_dio_count
, 0);
170 mapping
->a_ops
= &empty_aops
;
171 mapping
->host
= inode
;
173 atomic_set(&mapping
->i_mmap_writable
, 0);
174 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
175 mapping
->private_data
= NULL
;
176 mapping
->backing_dev_info
= &default_backing_dev_info
;
177 mapping
->writeback_index
= 0;
180 * If the block_device provides a backing_dev_info for client
181 * inodes then use that. Otherwise the inode share the bdev's
185 struct backing_dev_info
*bdi
;
187 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
188 mapping
->backing_dev_info
= bdi
;
190 inode
->i_private
= NULL
;
191 inode
->i_mapping
= mapping
;
192 INIT_HLIST_HEAD(&inode
->i_dentry
); /* buggered by rcu freeing */
193 #ifdef CONFIG_FS_POSIX_ACL
194 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
197 #ifdef CONFIG_FSNOTIFY
198 inode
->i_fsnotify_mask
= 0;
201 this_cpu_inc(nr_inodes
);
207 EXPORT_SYMBOL(inode_init_always
);
209 static struct inode
*alloc_inode(struct super_block
*sb
)
213 if (sb
->s_op
->alloc_inode
)
214 inode
= sb
->s_op
->alloc_inode(sb
);
216 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
221 if (unlikely(inode_init_always(sb
, inode
))) {
222 if (inode
->i_sb
->s_op
->destroy_inode
)
223 inode
->i_sb
->s_op
->destroy_inode(inode
);
225 kmem_cache_free(inode_cachep
, inode
);
232 void free_inode_nonrcu(struct inode
*inode
)
234 kmem_cache_free(inode_cachep
, inode
);
236 EXPORT_SYMBOL(free_inode_nonrcu
);
238 void __destroy_inode(struct inode
*inode
)
240 BUG_ON(inode_has_buffers(inode
));
241 security_inode_free(inode
);
242 fsnotify_inode_delete(inode
);
243 if (!inode
->i_nlink
) {
244 WARN_ON(atomic_long_read(&inode
->i_sb
->s_remove_count
) == 0);
245 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
248 #ifdef CONFIG_FS_POSIX_ACL
249 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
250 posix_acl_release(inode
->i_acl
);
251 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
252 posix_acl_release(inode
->i_default_acl
);
254 this_cpu_dec(nr_inodes
);
256 EXPORT_SYMBOL(__destroy_inode
);
258 static void i_callback(struct rcu_head
*head
)
260 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
261 kmem_cache_free(inode_cachep
, inode
);
264 static void destroy_inode(struct inode
*inode
)
266 BUG_ON(!list_empty(&inode
->i_lru
));
267 __destroy_inode(inode
);
268 if (inode
->i_sb
->s_op
->destroy_inode
)
269 inode
->i_sb
->s_op
->destroy_inode(inode
);
271 call_rcu(&inode
->i_rcu
, i_callback
);
275 * drop_nlink - directly drop an inode's link count
278 * This is a low-level filesystem helper to replace any
279 * direct filesystem manipulation of i_nlink. In cases
280 * where we are attempting to track writes to the
281 * filesystem, a decrement to zero means an imminent
282 * write when the file is truncated and actually unlinked
285 void drop_nlink(struct inode
*inode
)
287 WARN_ON(inode
->i_nlink
== 0);
290 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
292 EXPORT_SYMBOL(drop_nlink
);
295 * clear_nlink - directly zero an inode's link count
298 * This is a low-level filesystem helper to replace any
299 * direct filesystem manipulation of i_nlink. See
300 * drop_nlink() for why we care about i_nlink hitting zero.
302 void clear_nlink(struct inode
*inode
)
304 if (inode
->i_nlink
) {
305 inode
->__i_nlink
= 0;
306 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
309 EXPORT_SYMBOL(clear_nlink
);
312 * set_nlink - directly set an inode's link count
314 * @nlink: new nlink (should be non-zero)
316 * This is a low-level filesystem helper to replace any
317 * direct filesystem manipulation of i_nlink.
319 void set_nlink(struct inode
*inode
, unsigned int nlink
)
324 /* Yes, some filesystems do change nlink from zero to one */
325 if (inode
->i_nlink
== 0)
326 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
328 inode
->__i_nlink
= nlink
;
331 EXPORT_SYMBOL(set_nlink
);
334 * inc_nlink - directly increment an inode's link count
337 * This is a low-level filesystem helper to replace any
338 * direct filesystem manipulation of i_nlink. Currently,
339 * it is only here for parity with dec_nlink().
341 void inc_nlink(struct inode
*inode
)
343 if (unlikely(inode
->i_nlink
== 0)) {
344 WARN_ON(!(inode
->i_state
& I_LINKABLE
));
345 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
350 EXPORT_SYMBOL(inc_nlink
);
352 void address_space_init_once(struct address_space
*mapping
)
354 memset(mapping
, 0, sizeof(*mapping
));
355 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
356 spin_lock_init(&mapping
->tree_lock
);
357 mutex_init(&mapping
->i_mmap_mutex
);
358 INIT_LIST_HEAD(&mapping
->private_list
);
359 spin_lock_init(&mapping
->private_lock
);
360 mapping
->i_mmap
= RB_ROOT
;
361 INIT_LIST_HEAD(&mapping
->i_mmap_nonlinear
);
363 EXPORT_SYMBOL(address_space_init_once
);
366 * These are initializations that only need to be done
367 * once, because the fields are idempotent across use
368 * of the inode, so let the slab aware of that.
370 void inode_init_once(struct inode
*inode
)
372 memset(inode
, 0, sizeof(*inode
));
373 INIT_HLIST_NODE(&inode
->i_hash
);
374 INIT_LIST_HEAD(&inode
->i_devices
);
375 INIT_LIST_HEAD(&inode
->i_wb_list
);
376 INIT_LIST_HEAD(&inode
->i_lru
);
377 address_space_init_once(&inode
->i_data
);
378 i_size_ordered_init(inode
);
379 #ifdef CONFIG_FSNOTIFY
380 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
383 EXPORT_SYMBOL(inode_init_once
);
385 static void init_once(void *foo
)
387 struct inode
*inode
= (struct inode
*) foo
;
389 inode_init_once(inode
);
393 * inode->i_lock must be held
395 void __iget(struct inode
*inode
)
397 atomic_inc(&inode
->i_count
);
401 * get additional reference to inode; caller must already hold one.
403 void ihold(struct inode
*inode
)
405 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
407 EXPORT_SYMBOL(ihold
);
409 static void inode_lru_list_add(struct inode
*inode
)
411 if (list_lru_add(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
412 this_cpu_inc(nr_unused
);
416 * Add inode to LRU if needed (inode is unused and clean).
418 * Needs inode->i_lock held.
420 void inode_add_lru(struct inode
*inode
)
422 if (!(inode
->i_state
& (I_DIRTY
| I_SYNC
| I_FREEING
| I_WILL_FREE
)) &&
423 !atomic_read(&inode
->i_count
) && inode
->i_sb
->s_flags
& MS_ACTIVE
)
424 inode_lru_list_add(inode
);
428 static void inode_lru_list_del(struct inode
*inode
)
431 if (list_lru_del(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
432 this_cpu_dec(nr_unused
);
436 * inode_sb_list_add - add inode to the superblock list of inodes
437 * @inode: inode to add
439 void inode_sb_list_add(struct inode
*inode
)
441 spin_lock(&inode_sb_list_lock
);
442 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
443 spin_unlock(&inode_sb_list_lock
);
445 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
447 static inline void inode_sb_list_del(struct inode
*inode
)
449 if (!list_empty(&inode
->i_sb_list
)) {
450 spin_lock(&inode_sb_list_lock
);
451 list_del_init(&inode
->i_sb_list
);
452 spin_unlock(&inode_sb_list_lock
);
456 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
460 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
462 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> i_hash_shift
);
463 return tmp
& i_hash_mask
;
467 * __insert_inode_hash - hash an inode
468 * @inode: unhashed inode
469 * @hashval: unsigned long value used to locate this object in the
472 * Add an inode to the inode hash for this superblock.
474 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
476 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
478 spin_lock(&inode_hash_lock
);
479 spin_lock(&inode
->i_lock
);
480 hlist_add_head(&inode
->i_hash
, b
);
481 spin_unlock(&inode
->i_lock
);
482 spin_unlock(&inode_hash_lock
);
484 EXPORT_SYMBOL(__insert_inode_hash
);
487 * __remove_inode_hash - remove an inode from the hash
488 * @inode: inode to unhash
490 * Remove an inode from the superblock.
492 void __remove_inode_hash(struct inode
*inode
)
494 spin_lock(&inode_hash_lock
);
495 spin_lock(&inode
->i_lock
);
496 hlist_del_init(&inode
->i_hash
);
497 spin_unlock(&inode
->i_lock
);
498 spin_unlock(&inode_hash_lock
);
500 EXPORT_SYMBOL(__remove_inode_hash
);
502 void clear_inode(struct inode
*inode
)
506 * We have to cycle tree_lock here because reclaim can be still in the
507 * process of removing the last page (in __delete_from_page_cache())
508 * and we must not free mapping under it.
510 spin_lock_irq(&inode
->i_data
.tree_lock
);
511 BUG_ON(inode
->i_data
.nrpages
);
512 BUG_ON(inode
->i_data
.nrshadows
);
513 spin_unlock_irq(&inode
->i_data
.tree_lock
);
514 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
515 BUG_ON(!(inode
->i_state
& I_FREEING
));
516 BUG_ON(inode
->i_state
& I_CLEAR
);
517 /* don't need i_lock here, no concurrent mods to i_state */
518 inode
->i_state
= I_FREEING
| I_CLEAR
;
520 EXPORT_SYMBOL(clear_inode
);
523 * Free the inode passed in, removing it from the lists it is still connected
524 * to. We remove any pages still attached to the inode and wait for any IO that
525 * is still in progress before finally destroying the inode.
527 * An inode must already be marked I_FREEING so that we avoid the inode being
528 * moved back onto lists if we race with other code that manipulates the lists
529 * (e.g. writeback_single_inode). The caller is responsible for setting this.
531 * An inode must already be removed from the LRU list before being evicted from
532 * the cache. This should occur atomically with setting the I_FREEING state
533 * flag, so no inodes here should ever be on the LRU when being evicted.
535 static void evict(struct inode
*inode
)
537 const struct super_operations
*op
= inode
->i_sb
->s_op
;
539 BUG_ON(!(inode
->i_state
& I_FREEING
));
540 BUG_ON(!list_empty(&inode
->i_lru
));
542 if (!list_empty(&inode
->i_wb_list
))
543 inode_wb_list_del(inode
);
545 inode_sb_list_del(inode
);
548 * Wait for flusher thread to be done with the inode so that filesystem
549 * does not start destroying it while writeback is still running. Since
550 * the inode has I_FREEING set, flusher thread won't start new work on
551 * the inode. We just have to wait for running writeback to finish.
553 inode_wait_for_writeback(inode
);
555 if (op
->evict_inode
) {
556 op
->evict_inode(inode
);
558 truncate_inode_pages_final(&inode
->i_data
);
561 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
563 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
566 remove_inode_hash(inode
);
568 spin_lock(&inode
->i_lock
);
569 wake_up_bit(&inode
->i_state
, __I_NEW
);
570 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
571 spin_unlock(&inode
->i_lock
);
573 destroy_inode(inode
);
577 * dispose_list - dispose of the contents of a local list
578 * @head: the head of the list to free
580 * Dispose-list gets a local list with local inodes in it, so it doesn't
581 * need to worry about list corruption and SMP locks.
583 static void dispose_list(struct list_head
*head
)
585 while (!list_empty(head
)) {
588 inode
= list_first_entry(head
, struct inode
, i_lru
);
589 list_del_init(&inode
->i_lru
);
596 * evict_inodes - evict all evictable inodes for a superblock
597 * @sb: superblock to operate on
599 * Make sure that no inodes with zero refcount are retained. This is
600 * called by superblock shutdown after having MS_ACTIVE flag removed,
601 * so any inode reaching zero refcount during or after that call will
602 * be immediately evicted.
604 void evict_inodes(struct super_block
*sb
)
606 struct inode
*inode
, *next
;
609 spin_lock(&inode_sb_list_lock
);
610 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
611 if (atomic_read(&inode
->i_count
))
614 spin_lock(&inode
->i_lock
);
615 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
616 spin_unlock(&inode
->i_lock
);
620 inode
->i_state
|= I_FREEING
;
621 inode_lru_list_del(inode
);
622 spin_unlock(&inode
->i_lock
);
623 list_add(&inode
->i_lru
, &dispose
);
625 spin_unlock(&inode_sb_list_lock
);
627 dispose_list(&dispose
);
631 * invalidate_inodes - attempt to free all inodes on a superblock
632 * @sb: superblock to operate on
633 * @kill_dirty: flag to guide handling of dirty inodes
635 * Attempts to free all inodes for a given superblock. If there were any
636 * busy inodes return a non-zero value, else zero.
637 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
640 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
643 struct inode
*inode
, *next
;
646 spin_lock(&inode_sb_list_lock
);
647 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
648 spin_lock(&inode
->i_lock
);
649 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
650 spin_unlock(&inode
->i_lock
);
653 if (inode
->i_state
& I_DIRTY
&& !kill_dirty
) {
654 spin_unlock(&inode
->i_lock
);
658 if (atomic_read(&inode
->i_count
)) {
659 spin_unlock(&inode
->i_lock
);
664 inode
->i_state
|= I_FREEING
;
665 inode_lru_list_del(inode
);
666 spin_unlock(&inode
->i_lock
);
667 list_add(&inode
->i_lru
, &dispose
);
669 spin_unlock(&inode_sb_list_lock
);
671 dispose_list(&dispose
);
677 * Isolate the inode from the LRU in preparation for freeing it.
679 * Any inodes which are pinned purely because of attached pagecache have their
680 * pagecache removed. If the inode has metadata buffers attached to
681 * mapping->private_list then try to remove them.
683 * If the inode has the I_REFERENCED flag set, then it means that it has been
684 * used recently - the flag is set in iput_final(). When we encounter such an
685 * inode, clear the flag and move it to the back of the LRU so it gets another
686 * pass through the LRU before it gets reclaimed. This is necessary because of
687 * the fact we are doing lazy LRU updates to minimise lock contention so the
688 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
689 * with this flag set because they are the inodes that are out of order.
691 static enum lru_status
692 inode_lru_isolate(struct list_head
*item
, spinlock_t
*lru_lock
, void *arg
)
694 struct list_head
*freeable
= arg
;
695 struct inode
*inode
= container_of(item
, struct inode
, i_lru
);
698 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
699 * If we fail to get the lock, just skip it.
701 if (!spin_trylock(&inode
->i_lock
))
705 * Referenced or dirty inodes are still in use. Give them another pass
706 * through the LRU as we canot reclaim them now.
708 if (atomic_read(&inode
->i_count
) ||
709 (inode
->i_state
& ~I_REFERENCED
)) {
710 list_del_init(&inode
->i_lru
);
711 spin_unlock(&inode
->i_lock
);
712 this_cpu_dec(nr_unused
);
716 /* recently referenced inodes get one more pass */
717 if (inode
->i_state
& I_REFERENCED
) {
718 inode
->i_state
&= ~I_REFERENCED
;
719 spin_unlock(&inode
->i_lock
);
723 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
725 spin_unlock(&inode
->i_lock
);
726 spin_unlock(lru_lock
);
727 if (remove_inode_buffers(inode
)) {
729 reap
= invalidate_mapping_pages(&inode
->i_data
, 0, -1);
730 if (current_is_kswapd())
731 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
733 __count_vm_events(PGINODESTEAL
, reap
);
734 if (current
->reclaim_state
)
735 current
->reclaim_state
->reclaimed_slab
+= reap
;
742 WARN_ON(inode
->i_state
& I_NEW
);
743 inode
->i_state
|= I_FREEING
;
744 list_move(&inode
->i_lru
, freeable
);
745 spin_unlock(&inode
->i_lock
);
747 this_cpu_dec(nr_unused
);
752 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
753 * This is called from the superblock shrinker function with a number of inodes
754 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
755 * then are freed outside inode_lock by dispose_list().
757 long prune_icache_sb(struct super_block
*sb
, unsigned long nr_to_scan
,
763 freed
= list_lru_walk_node(&sb
->s_inode_lru
, nid
, inode_lru_isolate
,
764 &freeable
, &nr_to_scan
);
765 dispose_list(&freeable
);
769 static void __wait_on_freeing_inode(struct inode
*inode
);
771 * Called with the inode lock held.
773 static struct inode
*find_inode(struct super_block
*sb
,
774 struct hlist_head
*head
,
775 int (*test
)(struct inode
*, void *),
778 struct inode
*inode
= NULL
;
781 hlist_for_each_entry(inode
, head
, i_hash
) {
782 if (inode
->i_sb
!= sb
)
784 if (!test(inode
, data
))
786 spin_lock(&inode
->i_lock
);
787 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
788 __wait_on_freeing_inode(inode
);
792 spin_unlock(&inode
->i_lock
);
799 * find_inode_fast is the fast path version of find_inode, see the comment at
800 * iget_locked for details.
802 static struct inode
*find_inode_fast(struct super_block
*sb
,
803 struct hlist_head
*head
, unsigned long ino
)
805 struct inode
*inode
= NULL
;
808 hlist_for_each_entry(inode
, head
, i_hash
) {
809 if (inode
->i_ino
!= ino
)
811 if (inode
->i_sb
!= sb
)
813 spin_lock(&inode
->i_lock
);
814 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
815 __wait_on_freeing_inode(inode
);
819 spin_unlock(&inode
->i_lock
);
826 * Each cpu owns a range of LAST_INO_BATCH numbers.
827 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
828 * to renew the exhausted range.
830 * This does not significantly increase overflow rate because every CPU can
831 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
832 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
833 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
834 * overflow rate by 2x, which does not seem too significant.
836 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
837 * error if st_ino won't fit in target struct field. Use 32bit counter
838 * here to attempt to avoid that.
840 #define LAST_INO_BATCH 1024
841 static DEFINE_PER_CPU(unsigned int, last_ino
);
843 unsigned int get_next_ino(void)
845 unsigned int *p
= &get_cpu_var(last_ino
);
846 unsigned int res
= *p
;
849 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
850 static atomic_t shared_last_ino
;
851 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
853 res
= next
- LAST_INO_BATCH
;
858 put_cpu_var(last_ino
);
861 EXPORT_SYMBOL(get_next_ino
);
864 * new_inode_pseudo - obtain an inode
867 * Allocates a new inode for given superblock.
868 * Inode wont be chained in superblock s_inodes list
870 * - fs can't be unmount
871 * - quotas, fsnotify, writeback can't work
873 struct inode
*new_inode_pseudo(struct super_block
*sb
)
875 struct inode
*inode
= alloc_inode(sb
);
878 spin_lock(&inode
->i_lock
);
880 spin_unlock(&inode
->i_lock
);
881 INIT_LIST_HEAD(&inode
->i_sb_list
);
887 * new_inode - obtain an inode
890 * Allocates a new inode for given superblock. The default gfp_mask
891 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
892 * If HIGHMEM pages are unsuitable or it is known that pages allocated
893 * for the page cache are not reclaimable or migratable,
894 * mapping_set_gfp_mask() must be called with suitable flags on the
895 * newly created inode's mapping
898 struct inode
*new_inode(struct super_block
*sb
)
902 spin_lock_prefetch(&inode_sb_list_lock
);
904 inode
= new_inode_pseudo(sb
);
906 inode_sb_list_add(inode
);
909 EXPORT_SYMBOL(new_inode
);
911 #ifdef CONFIG_DEBUG_LOCK_ALLOC
912 void lockdep_annotate_inode_mutex_key(struct inode
*inode
)
914 if (S_ISDIR(inode
->i_mode
)) {
915 struct file_system_type
*type
= inode
->i_sb
->s_type
;
917 /* Set new key only if filesystem hasn't already changed it */
918 if (lockdep_match_class(&inode
->i_mutex
, &type
->i_mutex_key
)) {
920 * ensure nobody is actually holding i_mutex
922 mutex_destroy(&inode
->i_mutex
);
923 mutex_init(&inode
->i_mutex
);
924 lockdep_set_class(&inode
->i_mutex
,
925 &type
->i_mutex_dir_key
);
929 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key
);
933 * unlock_new_inode - clear the I_NEW state and wake up any waiters
934 * @inode: new inode to unlock
936 * Called when the inode is fully initialised to clear the new state of the
937 * inode and wake up anyone waiting for the inode to finish initialisation.
939 void unlock_new_inode(struct inode
*inode
)
941 lockdep_annotate_inode_mutex_key(inode
);
942 spin_lock(&inode
->i_lock
);
943 WARN_ON(!(inode
->i_state
& I_NEW
));
944 inode
->i_state
&= ~I_NEW
;
946 wake_up_bit(&inode
->i_state
, __I_NEW
);
947 spin_unlock(&inode
->i_lock
);
949 EXPORT_SYMBOL(unlock_new_inode
);
952 * lock_two_nondirectories - take two i_mutexes on non-directory objects
954 * Lock any non-NULL argument that is not a directory.
955 * Zero, one or two objects may be locked by this function.
957 * @inode1: first inode to lock
958 * @inode2: second inode to lock
960 void lock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
963 swap(inode1
, inode2
);
965 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
966 mutex_lock(&inode1
->i_mutex
);
967 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
968 mutex_lock_nested(&inode2
->i_mutex
, I_MUTEX_NONDIR2
);
970 EXPORT_SYMBOL(lock_two_nondirectories
);
973 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
974 * @inode1: first inode to unlock
975 * @inode2: second inode to unlock
977 void unlock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
979 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
980 mutex_unlock(&inode1
->i_mutex
);
981 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
982 mutex_unlock(&inode2
->i_mutex
);
984 EXPORT_SYMBOL(unlock_two_nondirectories
);
987 * iget5_locked - obtain an inode from a mounted file system
988 * @sb: super block of file system
989 * @hashval: hash value (usually inode number) to get
990 * @test: callback used for comparisons between inodes
991 * @set: callback used to initialize a new struct inode
992 * @data: opaque data pointer to pass to @test and @set
994 * Search for the inode specified by @hashval and @data in the inode cache,
995 * and if present it is return it with an increased reference count. This is
996 * a generalized version of iget_locked() for file systems where the inode
997 * number is not sufficient for unique identification of an inode.
999 * If the inode is not in cache, allocate a new inode and return it locked,
1000 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1001 * before unlocking it via unlock_new_inode().
1003 * Note both @test and @set are called with the inode_hash_lock held, so can't
1006 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1007 int (*test
)(struct inode
*, void *),
1008 int (*set
)(struct inode
*, void *), void *data
)
1010 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1011 struct inode
*inode
;
1013 spin_lock(&inode_hash_lock
);
1014 inode
= find_inode(sb
, head
, test
, data
);
1015 spin_unlock(&inode_hash_lock
);
1018 wait_on_inode(inode
);
1022 inode
= alloc_inode(sb
);
1026 spin_lock(&inode_hash_lock
);
1027 /* We released the lock, so.. */
1028 old
= find_inode(sb
, head
, test
, data
);
1030 if (set(inode
, data
))
1033 spin_lock(&inode
->i_lock
);
1034 inode
->i_state
= I_NEW
;
1035 hlist_add_head(&inode
->i_hash
, head
);
1036 spin_unlock(&inode
->i_lock
);
1037 inode_sb_list_add(inode
);
1038 spin_unlock(&inode_hash_lock
);
1040 /* Return the locked inode with I_NEW set, the
1041 * caller is responsible for filling in the contents
1047 * Uhhuh, somebody else created the same inode under
1048 * us. Use the old inode instead of the one we just
1051 spin_unlock(&inode_hash_lock
);
1052 destroy_inode(inode
);
1054 wait_on_inode(inode
);
1059 spin_unlock(&inode_hash_lock
);
1060 destroy_inode(inode
);
1063 EXPORT_SYMBOL(iget5_locked
);
1066 * iget_locked - obtain an inode from a mounted file system
1067 * @sb: super block of file system
1068 * @ino: inode number to get
1070 * Search for the inode specified by @ino in the inode cache and if present
1071 * return it with an increased reference count. This is for file systems
1072 * where the inode number is sufficient for unique identification of an inode.
1074 * If the inode is not in cache, allocate a new inode and return it locked,
1075 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1076 * before unlocking it via unlock_new_inode().
1078 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1080 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1081 struct inode
*inode
;
1083 spin_lock(&inode_hash_lock
);
1084 inode
= find_inode_fast(sb
, head
, ino
);
1085 spin_unlock(&inode_hash_lock
);
1087 wait_on_inode(inode
);
1091 inode
= alloc_inode(sb
);
1095 spin_lock(&inode_hash_lock
);
1096 /* We released the lock, so.. */
1097 old
= find_inode_fast(sb
, head
, ino
);
1100 spin_lock(&inode
->i_lock
);
1101 inode
->i_state
= I_NEW
;
1102 hlist_add_head(&inode
->i_hash
, head
);
1103 spin_unlock(&inode
->i_lock
);
1104 inode_sb_list_add(inode
);
1105 spin_unlock(&inode_hash_lock
);
1107 /* Return the locked inode with I_NEW set, the
1108 * caller is responsible for filling in the contents
1114 * Uhhuh, somebody else created the same inode under
1115 * us. Use the old inode instead of the one we just
1118 spin_unlock(&inode_hash_lock
);
1119 destroy_inode(inode
);
1121 wait_on_inode(inode
);
1125 EXPORT_SYMBOL(iget_locked
);
1128 * search the inode cache for a matching inode number.
1129 * If we find one, then the inode number we are trying to
1130 * allocate is not unique and so we should not use it.
1132 * Returns 1 if the inode number is unique, 0 if it is not.
1134 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1136 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1137 struct inode
*inode
;
1139 spin_lock(&inode_hash_lock
);
1140 hlist_for_each_entry(inode
, b
, i_hash
) {
1141 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
) {
1142 spin_unlock(&inode_hash_lock
);
1146 spin_unlock(&inode_hash_lock
);
1152 * iunique - get a unique inode number
1154 * @max_reserved: highest reserved inode number
1156 * Obtain an inode number that is unique on the system for a given
1157 * superblock. This is used by file systems that have no natural
1158 * permanent inode numbering system. An inode number is returned that
1159 * is higher than the reserved limit but unique.
1162 * With a large number of inodes live on the file system this function
1163 * currently becomes quite slow.
1165 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1168 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1169 * error if st_ino won't fit in target struct field. Use 32bit counter
1170 * here to attempt to avoid that.
1172 static DEFINE_SPINLOCK(iunique_lock
);
1173 static unsigned int counter
;
1176 spin_lock(&iunique_lock
);
1178 if (counter
<= max_reserved
)
1179 counter
= max_reserved
+ 1;
1181 } while (!test_inode_iunique(sb
, res
));
1182 spin_unlock(&iunique_lock
);
1186 EXPORT_SYMBOL(iunique
);
1188 struct inode
*igrab(struct inode
*inode
)
1190 spin_lock(&inode
->i_lock
);
1191 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1193 spin_unlock(&inode
->i_lock
);
1195 spin_unlock(&inode
->i_lock
);
1197 * Handle the case where s_op->clear_inode is not been
1198 * called yet, and somebody is calling igrab
1199 * while the inode is getting freed.
1205 EXPORT_SYMBOL(igrab
);
1208 * ilookup5_nowait - search for an inode in the inode cache
1209 * @sb: super block of file system to search
1210 * @hashval: hash value (usually inode number) to search for
1211 * @test: callback used for comparisons between inodes
1212 * @data: opaque data pointer to pass to @test
1214 * Search for the inode specified by @hashval and @data in the inode cache.
1215 * If the inode is in the cache, the inode is returned with an incremented
1218 * Note: I_NEW is not waited upon so you have to be very careful what you do
1219 * with the returned inode. You probably should be using ilookup5() instead.
1221 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1223 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1224 int (*test
)(struct inode
*, void *), void *data
)
1226 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1227 struct inode
*inode
;
1229 spin_lock(&inode_hash_lock
);
1230 inode
= find_inode(sb
, head
, test
, data
);
1231 spin_unlock(&inode_hash_lock
);
1235 EXPORT_SYMBOL(ilookup5_nowait
);
1238 * ilookup5 - search for an inode in the inode cache
1239 * @sb: super block of file system to search
1240 * @hashval: hash value (usually inode number) to search for
1241 * @test: callback used for comparisons between inodes
1242 * @data: opaque data pointer to pass to @test
1244 * Search for the inode specified by @hashval and @data in the inode cache,
1245 * and if the inode is in the cache, return the inode with an incremented
1246 * reference count. Waits on I_NEW before returning the inode.
1247 * returned with an incremented reference count.
1249 * This is a generalized version of ilookup() for file systems where the
1250 * inode number is not sufficient for unique identification of an inode.
1252 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1254 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1255 int (*test
)(struct inode
*, void *), void *data
)
1257 struct inode
*inode
= ilookup5_nowait(sb
, hashval
, test
, data
);
1260 wait_on_inode(inode
);
1263 EXPORT_SYMBOL(ilookup5
);
1266 * ilookup - search for an inode in the inode cache
1267 * @sb: super block of file system to search
1268 * @ino: inode number to search for
1270 * Search for the inode @ino in the inode cache, and if the inode is in the
1271 * cache, the inode is returned with an incremented reference count.
1273 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1275 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1276 struct inode
*inode
;
1278 spin_lock(&inode_hash_lock
);
1279 inode
= find_inode_fast(sb
, head
, ino
);
1280 spin_unlock(&inode_hash_lock
);
1283 wait_on_inode(inode
);
1286 EXPORT_SYMBOL(ilookup
);
1288 int insert_inode_locked(struct inode
*inode
)
1290 struct super_block
*sb
= inode
->i_sb
;
1291 ino_t ino
= inode
->i_ino
;
1292 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1295 struct inode
*old
= NULL
;
1296 spin_lock(&inode_hash_lock
);
1297 hlist_for_each_entry(old
, head
, i_hash
) {
1298 if (old
->i_ino
!= ino
)
1300 if (old
->i_sb
!= sb
)
1302 spin_lock(&old
->i_lock
);
1303 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1304 spin_unlock(&old
->i_lock
);
1310 spin_lock(&inode
->i_lock
);
1311 inode
->i_state
|= I_NEW
;
1312 hlist_add_head(&inode
->i_hash
, head
);
1313 spin_unlock(&inode
->i_lock
);
1314 spin_unlock(&inode_hash_lock
);
1318 spin_unlock(&old
->i_lock
);
1319 spin_unlock(&inode_hash_lock
);
1321 if (unlikely(!inode_unhashed(old
))) {
1328 EXPORT_SYMBOL(insert_inode_locked
);
1330 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1331 int (*test
)(struct inode
*, void *), void *data
)
1333 struct super_block
*sb
= inode
->i_sb
;
1334 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1337 struct inode
*old
= NULL
;
1339 spin_lock(&inode_hash_lock
);
1340 hlist_for_each_entry(old
, head
, i_hash
) {
1341 if (old
->i_sb
!= sb
)
1343 if (!test(old
, data
))
1345 spin_lock(&old
->i_lock
);
1346 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1347 spin_unlock(&old
->i_lock
);
1353 spin_lock(&inode
->i_lock
);
1354 inode
->i_state
|= I_NEW
;
1355 hlist_add_head(&inode
->i_hash
, head
);
1356 spin_unlock(&inode
->i_lock
);
1357 spin_unlock(&inode_hash_lock
);
1361 spin_unlock(&old
->i_lock
);
1362 spin_unlock(&inode_hash_lock
);
1364 if (unlikely(!inode_unhashed(old
))) {
1371 EXPORT_SYMBOL(insert_inode_locked4
);
1374 int generic_delete_inode(struct inode
*inode
)
1378 EXPORT_SYMBOL(generic_delete_inode
);
1381 * Called when we're dropping the last reference
1384 * Call the FS "drop_inode()" function, defaulting to
1385 * the legacy UNIX filesystem behaviour. If it tells
1386 * us to evict inode, do so. Otherwise, retain inode
1387 * in cache if fs is alive, sync and evict if fs is
1390 static void iput_final(struct inode
*inode
)
1392 struct super_block
*sb
= inode
->i_sb
;
1393 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1396 WARN_ON(inode
->i_state
& I_NEW
);
1399 drop
= op
->drop_inode(inode
);
1401 drop
= generic_drop_inode(inode
);
1403 if (!drop
&& (sb
->s_flags
& MS_ACTIVE
)) {
1404 inode
->i_state
|= I_REFERENCED
;
1405 inode_add_lru(inode
);
1406 spin_unlock(&inode
->i_lock
);
1411 inode
->i_state
|= I_WILL_FREE
;
1412 spin_unlock(&inode
->i_lock
);
1413 write_inode_now(inode
, 1);
1414 spin_lock(&inode
->i_lock
);
1415 WARN_ON(inode
->i_state
& I_NEW
);
1416 inode
->i_state
&= ~I_WILL_FREE
;
1419 inode
->i_state
|= I_FREEING
;
1420 if (!list_empty(&inode
->i_lru
))
1421 inode_lru_list_del(inode
);
1422 spin_unlock(&inode
->i_lock
);
1428 * iput - put an inode
1429 * @inode: inode to put
1431 * Puts an inode, dropping its usage count. If the inode use count hits
1432 * zero, the inode is then freed and may also be destroyed.
1434 * Consequently, iput() can sleep.
1436 void iput(struct inode
*inode
)
1439 BUG_ON(inode
->i_state
& I_CLEAR
);
1441 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
))
1445 EXPORT_SYMBOL(iput
);
1448 * bmap - find a block number in a file
1449 * @inode: inode of file
1450 * @block: block to find
1452 * Returns the block number on the device holding the inode that
1453 * is the disk block number for the block of the file requested.
1454 * That is, asked for block 4 of inode 1 the function will return the
1455 * disk block relative to the disk start that holds that block of the
1458 sector_t
bmap(struct inode
*inode
, sector_t block
)
1461 if (inode
->i_mapping
->a_ops
->bmap
)
1462 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1465 EXPORT_SYMBOL(bmap
);
1468 * With relative atime, only update atime if the previous atime is
1469 * earlier than either the ctime or mtime or if at least a day has
1470 * passed since the last atime update.
1472 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1473 struct timespec now
)
1476 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1479 * Is mtime younger than atime? If yes, update atime:
1481 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1484 * Is ctime younger than atime? If yes, update atime:
1486 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1490 * Is the previous atime value older than a day? If yes,
1493 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1496 * Good, we can skip the atime update:
1502 * This does the actual work of updating an inodes time or version. Must have
1503 * had called mnt_want_write() before calling this.
1505 static int update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1507 if (inode
->i_op
->update_time
)
1508 return inode
->i_op
->update_time(inode
, time
, flags
);
1510 if (flags
& S_ATIME
)
1511 inode
->i_atime
= *time
;
1512 if (flags
& S_VERSION
)
1513 inode_inc_iversion(inode
);
1514 if (flags
& S_CTIME
)
1515 inode
->i_ctime
= *time
;
1516 if (flags
& S_MTIME
)
1517 inode
->i_mtime
= *time
;
1518 mark_inode_dirty_sync(inode
);
1523 * touch_atime - update the access time
1524 * @path: the &struct path to update
1526 * Update the accessed time on an inode and mark it for writeback.
1527 * This function automatically handles read only file systems and media,
1528 * as well as the "noatime" flag and inode specific "noatime" markers.
1530 void touch_atime(const struct path
*path
)
1532 struct vfsmount
*mnt
= path
->mnt
;
1533 struct inode
*inode
= path
->dentry
->d_inode
;
1534 struct timespec now
;
1536 if (inode
->i_flags
& S_NOATIME
)
1538 if (IS_NOATIME(inode
))
1540 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1543 if (mnt
->mnt_flags
& MNT_NOATIME
)
1545 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1548 now
= current_fs_time(inode
->i_sb
);
1550 if (!relatime_need_update(mnt
, inode
, now
))
1553 if (timespec_equal(&inode
->i_atime
, &now
))
1556 if (!sb_start_write_trylock(inode
->i_sb
))
1559 if (__mnt_want_write(mnt
))
1562 * File systems can error out when updating inodes if they need to
1563 * allocate new space to modify an inode (such is the case for
1564 * Btrfs), but since we touch atime while walking down the path we
1565 * really don't care if we failed to update the atime of the file,
1566 * so just ignore the return value.
1567 * We may also fail on filesystems that have the ability to make parts
1568 * of the fs read only, e.g. subvolumes in Btrfs.
1570 update_time(inode
, &now
, S_ATIME
);
1571 __mnt_drop_write(mnt
);
1573 sb_end_write(inode
->i_sb
);
1575 EXPORT_SYMBOL(touch_atime
);
1578 * The logic we want is
1580 * if suid or (sgid and xgrp)
1583 int should_remove_suid(struct dentry
*dentry
)
1585 umode_t mode
= dentry
->d_inode
->i_mode
;
1588 /* suid always must be killed */
1589 if (unlikely(mode
& S_ISUID
))
1590 kill
= ATTR_KILL_SUID
;
1593 * sgid without any exec bits is just a mandatory locking mark; leave
1594 * it alone. If some exec bits are set, it's a real sgid; kill it.
1596 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1597 kill
|= ATTR_KILL_SGID
;
1599 if (unlikely(kill
&& !capable(CAP_FSETID
) && S_ISREG(mode
)))
1604 EXPORT_SYMBOL(should_remove_suid
);
1606 static int __remove_suid(struct dentry
*dentry
, int kill
)
1608 struct iattr newattrs
;
1610 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1612 * Note we call this on write, so notify_change will not
1613 * encounter any conflicting delegations:
1615 return notify_change(dentry
, &newattrs
, NULL
);
1618 int file_remove_suid(struct file
*file
)
1620 struct dentry
*dentry
= file
->f_path
.dentry
;
1621 struct inode
*inode
= dentry
->d_inode
;
1626 /* Fast path for nothing security related */
1627 if (IS_NOSEC(inode
))
1630 killsuid
= should_remove_suid(dentry
);
1631 killpriv
= security_inode_need_killpriv(dentry
);
1636 error
= security_inode_killpriv(dentry
);
1637 if (!error
&& killsuid
)
1638 error
= __remove_suid(dentry
, killsuid
);
1639 if (!error
&& (inode
->i_sb
->s_flags
& MS_NOSEC
))
1640 inode
->i_flags
|= S_NOSEC
;
1644 EXPORT_SYMBOL(file_remove_suid
);
1647 * file_update_time - update mtime and ctime time
1648 * @file: file accessed
1650 * Update the mtime and ctime members of an inode and mark the inode
1651 * for writeback. Note that this function is meant exclusively for
1652 * usage in the file write path of filesystems, and filesystems may
1653 * choose to explicitly ignore update via this function with the
1654 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1655 * timestamps are handled by the server. This can return an error for
1656 * file systems who need to allocate space in order to update an inode.
1659 int file_update_time(struct file
*file
)
1661 struct inode
*inode
= file_inode(file
);
1662 struct timespec now
;
1666 /* First try to exhaust all avenues to not sync */
1667 if (IS_NOCMTIME(inode
))
1670 now
= current_fs_time(inode
->i_sb
);
1671 if (!timespec_equal(&inode
->i_mtime
, &now
))
1674 if (!timespec_equal(&inode
->i_ctime
, &now
))
1677 if (IS_I_VERSION(inode
))
1678 sync_it
|= S_VERSION
;
1683 /* Finally allowed to write? Takes lock. */
1684 if (__mnt_want_write_file(file
))
1687 ret
= update_time(inode
, &now
, sync_it
);
1688 __mnt_drop_write_file(file
);
1692 EXPORT_SYMBOL(file_update_time
);
1694 int inode_needs_sync(struct inode
*inode
)
1698 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1702 EXPORT_SYMBOL(inode_needs_sync
);
1705 * If we try to find an inode in the inode hash while it is being
1706 * deleted, we have to wait until the filesystem completes its
1707 * deletion before reporting that it isn't found. This function waits
1708 * until the deletion _might_ have completed. Callers are responsible
1709 * to recheck inode state.
1711 * It doesn't matter if I_NEW is not set initially, a call to
1712 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1715 static void __wait_on_freeing_inode(struct inode
*inode
)
1717 wait_queue_head_t
*wq
;
1718 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1719 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1720 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1721 spin_unlock(&inode
->i_lock
);
1722 spin_unlock(&inode_hash_lock
);
1724 finish_wait(wq
, &wait
.wait
);
1725 spin_lock(&inode_hash_lock
);
1728 static __initdata
unsigned long ihash_entries
;
1729 static int __init
set_ihash_entries(char *str
)
1733 ihash_entries
= simple_strtoul(str
, &str
, 0);
1736 __setup("ihash_entries=", set_ihash_entries
);
1739 * Initialize the waitqueues and inode hash table.
1741 void __init
inode_init_early(void)
1745 /* If hashes are distributed across NUMA nodes, defer
1746 * hash allocation until vmalloc space is available.
1752 alloc_large_system_hash("Inode-cache",
1753 sizeof(struct hlist_head
),
1762 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1763 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1766 void __init
inode_init(void)
1770 /* inode slab cache */
1771 inode_cachep
= kmem_cache_create("inode_cache",
1772 sizeof(struct inode
),
1774 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1778 /* Hash may have been set up in inode_init_early */
1783 alloc_large_system_hash("Inode-cache",
1784 sizeof(struct hlist_head
),
1793 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1794 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1797 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1799 inode
->i_mode
= mode
;
1800 if (S_ISCHR(mode
)) {
1801 inode
->i_fop
= &def_chr_fops
;
1802 inode
->i_rdev
= rdev
;
1803 } else if (S_ISBLK(mode
)) {
1804 inode
->i_fop
= &def_blk_fops
;
1805 inode
->i_rdev
= rdev
;
1806 } else if (S_ISFIFO(mode
))
1807 inode
->i_fop
= &pipefifo_fops
;
1808 else if (S_ISSOCK(mode
))
1809 ; /* leave it no_open_fops */
1811 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1812 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1815 EXPORT_SYMBOL(init_special_inode
);
1818 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1820 * @dir: Directory inode
1821 * @mode: mode of the new inode
1823 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1826 inode
->i_uid
= current_fsuid();
1827 if (dir
&& dir
->i_mode
& S_ISGID
) {
1828 inode
->i_gid
= dir
->i_gid
;
1832 inode
->i_gid
= current_fsgid();
1833 inode
->i_mode
= mode
;
1835 EXPORT_SYMBOL(inode_init_owner
);
1838 * inode_owner_or_capable - check current task permissions to inode
1839 * @inode: inode being checked
1841 * Return true if current either has CAP_FOWNER in a namespace with the
1842 * inode owner uid mapped, or owns the file.
1844 bool inode_owner_or_capable(const struct inode
*inode
)
1846 struct user_namespace
*ns
;
1848 if (uid_eq(current_fsuid(), inode
->i_uid
))
1851 ns
= current_user_ns();
1852 if (ns_capable(ns
, CAP_FOWNER
) && kuid_has_mapping(ns
, inode
->i_uid
))
1856 EXPORT_SYMBOL(inode_owner_or_capable
);
1859 * Direct i/o helper functions
1861 static void __inode_dio_wait(struct inode
*inode
)
1863 wait_queue_head_t
*wq
= bit_waitqueue(&inode
->i_state
, __I_DIO_WAKEUP
);
1864 DEFINE_WAIT_BIT(q
, &inode
->i_state
, __I_DIO_WAKEUP
);
1867 prepare_to_wait(wq
, &q
.wait
, TASK_UNINTERRUPTIBLE
);
1868 if (atomic_read(&inode
->i_dio_count
))
1870 } while (atomic_read(&inode
->i_dio_count
));
1871 finish_wait(wq
, &q
.wait
);
1875 * inode_dio_wait - wait for outstanding DIO requests to finish
1876 * @inode: inode to wait for
1878 * Waits for all pending direct I/O requests to finish so that we can
1879 * proceed with a truncate or equivalent operation.
1881 * Must be called under a lock that serializes taking new references
1882 * to i_dio_count, usually by inode->i_mutex.
1884 void inode_dio_wait(struct inode
*inode
)
1886 if (atomic_read(&inode
->i_dio_count
))
1887 __inode_dio_wait(inode
);
1889 EXPORT_SYMBOL(inode_dio_wait
);
1892 * inode_dio_done - signal finish of a direct I/O requests
1893 * @inode: inode the direct I/O happens on
1895 * This is called once we've finished processing a direct I/O request,
1896 * and is used to wake up callers waiting for direct I/O to be quiesced.
1898 void inode_dio_done(struct inode
*inode
)
1900 if (atomic_dec_and_test(&inode
->i_dio_count
))
1901 wake_up_bit(&inode
->i_state
, __I_DIO_WAKEUP
);
1903 EXPORT_SYMBOL(inode_dio_done
);
1906 * inode_set_flags - atomically set some inode flags
1908 * Note: the caller should be holding i_mutex, or else be sure that
1909 * they have exclusive access to the inode structure (i.e., while the
1910 * inode is being instantiated). The reason for the cmpxchg() loop
1911 * --- which wouldn't be necessary if all code paths which modify
1912 * i_flags actually followed this rule, is that there is at least one
1913 * code path which doesn't today --- for example,
1914 * __generic_file_aio_write() calls file_remove_suid() without holding
1915 * i_mutex --- so we use cmpxchg() out of an abundance of caution.
1917 * In the long run, i_mutex is overkill, and we should probably look
1918 * at using the i_lock spinlock to protect i_flags, and then make sure
1919 * it is so documented in include/linux/fs.h and that all code follows
1920 * the locking convention!!
1922 void inode_set_flags(struct inode
*inode
, unsigned int flags
,
1925 unsigned int old_flags
, new_flags
;
1927 WARN_ON_ONCE(flags
& ~mask
);
1929 old_flags
= ACCESS_ONCE(inode
->i_flags
);
1930 new_flags
= (old_flags
& ~mask
) | flags
;
1931 } while (unlikely(cmpxchg(&inode
->i_flags
, old_flags
,
1932 new_flags
) != old_flags
));
1934 EXPORT_SYMBOL(inode_set_flags
);