2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
5 #include <linux/export.h>
8 #include <linux/backing-dev.h>
9 #include <linux/hash.h>
10 #include <linux/swap.h>
11 #include <linux/security.h>
12 #include <linux/cdev.h>
13 #include <linux/bootmem.h>
14 #include <linux/fsnotify.h>
15 #include <linux/mount.h>
16 #include <linux/posix_acl.h>
17 #include <linux/prefetch.h>
18 #include <linux/buffer_head.h> /* for inode_has_buffers */
19 #include <linux/ratelimit.h>
20 #include <linux/list_lru.h>
21 #include <trace/events/writeback.h>
25 * Inode locking rules:
27 * inode->i_lock protects:
28 * inode->i_state, inode->i_hash, __iget()
29 * Inode LRU list locks protect:
30 * inode->i_sb->s_inode_lru, inode->i_lru
31 * inode->i_sb->s_inode_list_lock protects:
32 * inode->i_sb->s_inodes, inode->i_sb_list
33 * bdi->wb.list_lock protects:
34 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
35 * inode_hash_lock protects:
36 * inode_hashtable, inode->i_hash
40 * inode->i_sb->s_inode_list_lock
42 * Inode LRU list locks
48 * inode->i_sb->s_inode_list_lock
55 static unsigned int i_hash_mask __read_mostly
;
56 static unsigned int i_hash_shift __read_mostly
;
57 static struct hlist_head
*inode_hashtable __read_mostly
;
58 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_hash_lock
);
61 * Empty aops. Can be used for the cases where the user does not
62 * define any of the address_space operations.
64 const struct address_space_operations empty_aops
= {
66 EXPORT_SYMBOL(empty_aops
);
69 * Statistics gathering..
71 struct inodes_stat_t inodes_stat
;
73 static DEFINE_PER_CPU(unsigned long, nr_inodes
);
74 static DEFINE_PER_CPU(unsigned long, nr_unused
);
76 static struct kmem_cache
*inode_cachep __read_mostly
;
78 static long get_nr_inodes(void)
82 for_each_possible_cpu(i
)
83 sum
+= per_cpu(nr_inodes
, i
);
84 return sum
< 0 ? 0 : sum
;
87 static inline long get_nr_inodes_unused(void)
91 for_each_possible_cpu(i
)
92 sum
+= per_cpu(nr_unused
, i
);
93 return sum
< 0 ? 0 : sum
;
96 long get_nr_dirty_inodes(void)
98 /* not actually dirty inodes, but a wild approximation */
99 long nr_dirty
= get_nr_inodes() - get_nr_inodes_unused();
100 return nr_dirty
> 0 ? nr_dirty
: 0;
104 * Handle nr_inode sysctl
107 int proc_nr_inodes(struct ctl_table
*table
, int write
,
108 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
110 inodes_stat
.nr_inodes
= get_nr_inodes();
111 inodes_stat
.nr_unused
= get_nr_inodes_unused();
112 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
116 static int no_open(struct inode
*inode
, struct file
*file
)
122 * inode_init_always - perform inode structure initialisation
123 * @sb: superblock inode belongs to
124 * @inode: inode to initialise
126 * These are initializations that need to be done on every inode
127 * allocation as the fields are not initialised by slab allocation.
129 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
131 static const struct inode_operations empty_iops
;
132 static const struct file_operations no_open_fops
= {.open
= no_open
};
133 struct address_space
*const mapping
= &inode
->i_data
;
136 inode
->i_blkbits
= sb
->s_blocksize_bits
;
138 atomic_set(&inode
->i_count
, 1);
139 inode
->i_op
= &empty_iops
;
140 inode
->i_fop
= &no_open_fops
;
141 inode
->__i_nlink
= 1;
142 inode
->i_opflags
= 0;
144 inode
->i_opflags
|= IOP_XATTR
;
145 i_uid_write(inode
, 0);
146 i_gid_write(inode
, 0);
147 atomic_set(&inode
->i_writecount
, 0);
149 inode
->i_write_hint
= WRITE_LIFE_NOT_SET
;
152 inode
->i_generation
= 0;
153 inode
->i_pipe
= NULL
;
154 inode
->i_bdev
= NULL
;
155 inode
->i_cdev
= NULL
;
156 inode
->i_link
= NULL
;
157 inode
->i_dir_seq
= 0;
159 inode
->dirtied_when
= 0;
161 #ifdef CONFIG_CGROUP_WRITEBACK
162 inode
->i_wb_frn_winner
= 0;
163 inode
->i_wb_frn_avg_time
= 0;
164 inode
->i_wb_frn_history
= 0;
167 if (security_inode_alloc(inode
))
169 spin_lock_init(&inode
->i_lock
);
170 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
172 init_rwsem(&inode
->i_rwsem
);
173 lockdep_set_class(&inode
->i_rwsem
, &sb
->s_type
->i_mutex_key
);
175 atomic_set(&inode
->i_dio_count
, 0);
177 mapping
->a_ops
= &empty_aops
;
178 mapping
->host
= inode
;
181 atomic_set(&mapping
->i_mmap_writable
, 0);
182 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
183 mapping
->private_data
= NULL
;
184 mapping
->writeback_index
= 0;
185 inode
->i_private
= NULL
;
186 inode
->i_mapping
= mapping
;
187 INIT_HLIST_HEAD(&inode
->i_dentry
); /* buggered by rcu freeing */
188 #ifdef CONFIG_FS_POSIX_ACL
189 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
192 #ifdef CONFIG_FSNOTIFY
193 inode
->i_fsnotify_mask
= 0;
195 inode
->i_flctx
= NULL
;
196 this_cpu_inc(nr_inodes
);
202 EXPORT_SYMBOL(inode_init_always
);
204 static struct inode
*alloc_inode(struct super_block
*sb
)
208 if (sb
->s_op
->alloc_inode
)
209 inode
= sb
->s_op
->alloc_inode(sb
);
211 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
216 if (unlikely(inode_init_always(sb
, inode
))) {
217 if (inode
->i_sb
->s_op
->destroy_inode
)
218 inode
->i_sb
->s_op
->destroy_inode(inode
);
220 kmem_cache_free(inode_cachep
, inode
);
227 void free_inode_nonrcu(struct inode
*inode
)
229 kmem_cache_free(inode_cachep
, inode
);
231 EXPORT_SYMBOL(free_inode_nonrcu
);
233 void __destroy_inode(struct inode
*inode
)
235 BUG_ON(inode_has_buffers(inode
));
236 inode_detach_wb(inode
);
237 security_inode_free(inode
);
238 fsnotify_inode_delete(inode
);
239 locks_free_lock_context(inode
);
240 if (!inode
->i_nlink
) {
241 WARN_ON(atomic_long_read(&inode
->i_sb
->s_remove_count
) == 0);
242 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
245 #ifdef CONFIG_FS_POSIX_ACL
246 if (inode
->i_acl
&& !is_uncached_acl(inode
->i_acl
))
247 posix_acl_release(inode
->i_acl
);
248 if (inode
->i_default_acl
&& !is_uncached_acl(inode
->i_default_acl
))
249 posix_acl_release(inode
->i_default_acl
);
251 this_cpu_dec(nr_inodes
);
253 EXPORT_SYMBOL(__destroy_inode
);
255 static void i_callback(struct rcu_head
*head
)
257 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
258 kmem_cache_free(inode_cachep
, inode
);
261 static void destroy_inode(struct inode
*inode
)
263 BUG_ON(!list_empty(&inode
->i_lru
));
264 __destroy_inode(inode
);
265 if (inode
->i_sb
->s_op
->destroy_inode
)
266 inode
->i_sb
->s_op
->destroy_inode(inode
);
268 call_rcu(&inode
->i_rcu
, i_callback
);
272 * drop_nlink - directly drop an inode's link count
275 * This is a low-level filesystem helper to replace any
276 * direct filesystem manipulation of i_nlink. In cases
277 * where we are attempting to track writes to the
278 * filesystem, a decrement to zero means an imminent
279 * write when the file is truncated and actually unlinked
282 void drop_nlink(struct inode
*inode
)
284 WARN_ON(inode
->i_nlink
== 0);
287 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
289 EXPORT_SYMBOL(drop_nlink
);
292 * clear_nlink - directly zero an inode's link count
295 * This is a low-level filesystem helper to replace any
296 * direct filesystem manipulation of i_nlink. See
297 * drop_nlink() for why we care about i_nlink hitting zero.
299 void clear_nlink(struct inode
*inode
)
301 if (inode
->i_nlink
) {
302 inode
->__i_nlink
= 0;
303 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
306 EXPORT_SYMBOL(clear_nlink
);
309 * set_nlink - directly set an inode's link count
311 * @nlink: new nlink (should be non-zero)
313 * This is a low-level filesystem helper to replace any
314 * direct filesystem manipulation of i_nlink.
316 void set_nlink(struct inode
*inode
, unsigned int nlink
)
321 /* Yes, some filesystems do change nlink from zero to one */
322 if (inode
->i_nlink
== 0)
323 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
325 inode
->__i_nlink
= nlink
;
328 EXPORT_SYMBOL(set_nlink
);
331 * inc_nlink - directly increment an inode's link count
334 * This is a low-level filesystem helper to replace any
335 * direct filesystem manipulation of i_nlink. Currently,
336 * it is only here for parity with dec_nlink().
338 void inc_nlink(struct inode
*inode
)
340 if (unlikely(inode
->i_nlink
== 0)) {
341 WARN_ON(!(inode
->i_state
& I_LINKABLE
));
342 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
347 EXPORT_SYMBOL(inc_nlink
);
349 void address_space_init_once(struct address_space
*mapping
)
351 memset(mapping
, 0, sizeof(*mapping
));
352 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
| __GFP_ACCOUNT
);
353 spin_lock_init(&mapping
->tree_lock
);
354 init_rwsem(&mapping
->i_mmap_rwsem
);
355 INIT_LIST_HEAD(&mapping
->private_list
);
356 spin_lock_init(&mapping
->private_lock
);
357 mapping
->i_mmap
= RB_ROOT_CACHED
;
359 EXPORT_SYMBOL(address_space_init_once
);
362 * These are initializations that only need to be done
363 * once, because the fields are idempotent across use
364 * of the inode, so let the slab aware of that.
366 void inode_init_once(struct inode
*inode
)
368 memset(inode
, 0, sizeof(*inode
));
369 INIT_HLIST_NODE(&inode
->i_hash
);
370 INIT_LIST_HEAD(&inode
->i_devices
);
371 INIT_LIST_HEAD(&inode
->i_io_list
);
372 INIT_LIST_HEAD(&inode
->i_wb_list
);
373 INIT_LIST_HEAD(&inode
->i_lru
);
374 address_space_init_once(&inode
->i_data
);
375 i_size_ordered_init(inode
);
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 if (list_lru_add(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
406 this_cpu_inc(nr_unused
);
408 inode
->i_state
|= I_REFERENCED
;
412 * Add inode to LRU if needed (inode is unused and clean).
414 * Needs inode->i_lock held.
416 void inode_add_lru(struct inode
*inode
)
418 if (!(inode
->i_state
& (I_DIRTY_ALL
| I_SYNC
|
419 I_FREEING
| I_WILL_FREE
)) &&
420 !atomic_read(&inode
->i_count
) && inode
->i_sb
->s_flags
& SB_ACTIVE
)
421 inode_lru_list_add(inode
);
425 static void inode_lru_list_del(struct inode
*inode
)
428 if (list_lru_del(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
429 this_cpu_dec(nr_unused
);
433 * inode_sb_list_add - add inode to the superblock list of inodes
434 * @inode: inode to add
436 void inode_sb_list_add(struct inode
*inode
)
438 spin_lock(&inode
->i_sb
->s_inode_list_lock
);
439 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
440 spin_unlock(&inode
->i_sb
->s_inode_list_lock
);
442 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
444 static inline void inode_sb_list_del(struct inode
*inode
)
446 if (!list_empty(&inode
->i_sb_list
)) {
447 spin_lock(&inode
->i_sb
->s_inode_list_lock
);
448 list_del_init(&inode
->i_sb_list
);
449 spin_unlock(&inode
->i_sb
->s_inode_list_lock
);
453 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
457 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
459 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> i_hash_shift
);
460 return tmp
& i_hash_mask
;
464 * __insert_inode_hash - hash an inode
465 * @inode: unhashed inode
466 * @hashval: unsigned long value used to locate this object in the
469 * Add an inode to the inode hash for this superblock.
471 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
473 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
475 spin_lock(&inode_hash_lock
);
476 spin_lock(&inode
->i_lock
);
477 hlist_add_head(&inode
->i_hash
, b
);
478 spin_unlock(&inode
->i_lock
);
479 spin_unlock(&inode_hash_lock
);
481 EXPORT_SYMBOL(__insert_inode_hash
);
484 * __remove_inode_hash - remove an inode from the hash
485 * @inode: inode to unhash
487 * Remove an inode from the superblock.
489 void __remove_inode_hash(struct inode
*inode
)
491 spin_lock(&inode_hash_lock
);
492 spin_lock(&inode
->i_lock
);
493 hlist_del_init(&inode
->i_hash
);
494 spin_unlock(&inode
->i_lock
);
495 spin_unlock(&inode_hash_lock
);
497 EXPORT_SYMBOL(__remove_inode_hash
);
499 void clear_inode(struct inode
*inode
)
503 * We have to cycle tree_lock here because reclaim can be still in the
504 * process of removing the last page (in __delete_from_page_cache())
505 * and we must not free mapping under it.
507 spin_lock_irq(&inode
->i_data
.tree_lock
);
508 BUG_ON(inode
->i_data
.nrpages
);
509 BUG_ON(inode
->i_data
.nrexceptional
);
510 spin_unlock_irq(&inode
->i_data
.tree_lock
);
511 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
512 BUG_ON(!(inode
->i_state
& I_FREEING
));
513 BUG_ON(inode
->i_state
& I_CLEAR
);
514 BUG_ON(!list_empty(&inode
->i_wb_list
));
515 /* don't need i_lock here, no concurrent mods to i_state */
516 inode
->i_state
= I_FREEING
| I_CLEAR
;
518 EXPORT_SYMBOL(clear_inode
);
521 * Free the inode passed in, removing it from the lists it is still connected
522 * to. We remove any pages still attached to the inode and wait for any IO that
523 * is still in progress before finally destroying the inode.
525 * An inode must already be marked I_FREEING so that we avoid the inode being
526 * moved back onto lists if we race with other code that manipulates the lists
527 * (e.g. writeback_single_inode). The caller is responsible for setting this.
529 * An inode must already be removed from the LRU list before being evicted from
530 * the cache. This should occur atomically with setting the I_FREEING state
531 * flag, so no inodes here should ever be on the LRU when being evicted.
533 static void evict(struct inode
*inode
)
535 const struct super_operations
*op
= inode
->i_sb
->s_op
;
537 BUG_ON(!(inode
->i_state
& I_FREEING
));
538 BUG_ON(!list_empty(&inode
->i_lru
));
540 if (!list_empty(&inode
->i_io_list
))
541 inode_io_list_del(inode
);
543 inode_sb_list_del(inode
);
546 * Wait for flusher thread to be done with the inode so that filesystem
547 * does not start destroying it while writeback is still running. Since
548 * the inode has I_FREEING set, flusher thread won't start new work on
549 * the inode. We just have to wait for running writeback to finish.
551 inode_wait_for_writeback(inode
);
553 if (op
->evict_inode
) {
554 op
->evict_inode(inode
);
556 truncate_inode_pages_final(&inode
->i_data
);
559 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
561 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
564 remove_inode_hash(inode
);
566 spin_lock(&inode
->i_lock
);
567 wake_up_bit(&inode
->i_state
, __I_NEW
);
568 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
569 spin_unlock(&inode
->i_lock
);
571 destroy_inode(inode
);
575 * dispose_list - dispose of the contents of a local list
576 * @head: the head of the list to free
578 * Dispose-list gets a local list with local inodes in it, so it doesn't
579 * need to worry about list corruption and SMP locks.
581 static void dispose_list(struct list_head
*head
)
583 while (!list_empty(head
)) {
586 inode
= list_first_entry(head
, struct inode
, i_lru
);
587 list_del_init(&inode
->i_lru
);
595 * evict_inodes - evict all evictable inodes for a superblock
596 * @sb: superblock to operate on
598 * Make sure that no inodes with zero refcount are retained. This is
599 * called by superblock shutdown after having SB_ACTIVE flag removed,
600 * so any inode reaching zero refcount during or after that call will
601 * be immediately evicted.
603 void evict_inodes(struct super_block
*sb
)
605 struct inode
*inode
, *next
;
609 spin_lock(&sb
->s_inode_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
);
626 * We can have a ton of inodes to evict at unmount time given
627 * enough memory, check to see if we need to go to sleep for a
628 * bit so we don't livelock.
630 if (need_resched()) {
631 spin_unlock(&sb
->s_inode_list_lock
);
633 dispose_list(&dispose
);
637 spin_unlock(&sb
->s_inode_list_lock
);
639 dispose_list(&dispose
);
641 EXPORT_SYMBOL_GPL(evict_inodes
);
644 * invalidate_inodes - attempt to free all inodes on a superblock
645 * @sb: superblock to operate on
646 * @kill_dirty: flag to guide handling of dirty inodes
648 * Attempts to free all inodes for a given superblock. If there were any
649 * busy inodes return a non-zero value, else zero.
650 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
653 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
656 struct inode
*inode
, *next
;
659 spin_lock(&sb
->s_inode_list_lock
);
660 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
661 spin_lock(&inode
->i_lock
);
662 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
663 spin_unlock(&inode
->i_lock
);
666 if (inode
->i_state
& I_DIRTY_ALL
&& !kill_dirty
) {
667 spin_unlock(&inode
->i_lock
);
671 if (atomic_read(&inode
->i_count
)) {
672 spin_unlock(&inode
->i_lock
);
677 inode
->i_state
|= I_FREEING
;
678 inode_lru_list_del(inode
);
679 spin_unlock(&inode
->i_lock
);
680 list_add(&inode
->i_lru
, &dispose
);
682 spin_unlock(&sb
->s_inode_list_lock
);
684 dispose_list(&dispose
);
690 * Isolate the inode from the LRU in preparation for freeing it.
692 * Any inodes which are pinned purely because of attached pagecache have their
693 * pagecache removed. If the inode has metadata buffers attached to
694 * mapping->private_list then try to remove them.
696 * If the inode has the I_REFERENCED flag set, then it means that it has been
697 * used recently - the flag is set in iput_final(). When we encounter such an
698 * inode, clear the flag and move it to the back of the LRU so it gets another
699 * pass through the LRU before it gets reclaimed. This is necessary because of
700 * the fact we are doing lazy LRU updates to minimise lock contention so the
701 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
702 * with this flag set because they are the inodes that are out of order.
704 static enum lru_status
inode_lru_isolate(struct list_head
*item
,
705 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
707 struct list_head
*freeable
= arg
;
708 struct inode
*inode
= container_of(item
, struct inode
, i_lru
);
711 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
712 * If we fail to get the lock, just skip it.
714 if (!spin_trylock(&inode
->i_lock
))
718 * Referenced or dirty inodes are still in use. Give them another pass
719 * through the LRU as we canot reclaim them now.
721 if (atomic_read(&inode
->i_count
) ||
722 (inode
->i_state
& ~I_REFERENCED
)) {
723 list_lru_isolate(lru
, &inode
->i_lru
);
724 spin_unlock(&inode
->i_lock
);
725 this_cpu_dec(nr_unused
);
729 /* recently referenced inodes get one more pass */
730 if (inode
->i_state
& I_REFERENCED
) {
731 inode
->i_state
&= ~I_REFERENCED
;
732 spin_unlock(&inode
->i_lock
);
736 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
738 spin_unlock(&inode
->i_lock
);
739 spin_unlock(lru_lock
);
740 if (remove_inode_buffers(inode
)) {
742 reap
= invalidate_mapping_pages(&inode
->i_data
, 0, -1);
743 if (current_is_kswapd())
744 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
746 __count_vm_events(PGINODESTEAL
, reap
);
747 if (current
->reclaim_state
)
748 current
->reclaim_state
->reclaimed_slab
+= reap
;
755 WARN_ON(inode
->i_state
& I_NEW
);
756 inode
->i_state
|= I_FREEING
;
757 list_lru_isolate_move(lru
, &inode
->i_lru
, freeable
);
758 spin_unlock(&inode
->i_lock
);
760 this_cpu_dec(nr_unused
);
765 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
766 * This is called from the superblock shrinker function with a number of inodes
767 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
768 * then are freed outside inode_lock by dispose_list().
770 long prune_icache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
775 freed
= list_lru_shrink_walk(&sb
->s_inode_lru
, sc
,
776 inode_lru_isolate
, &freeable
);
777 dispose_list(&freeable
);
781 static void __wait_on_freeing_inode(struct inode
*inode
);
783 * Called with the inode lock held.
785 static struct inode
*find_inode(struct super_block
*sb
,
786 struct hlist_head
*head
,
787 int (*test
)(struct inode
*, void *),
790 struct inode
*inode
= NULL
;
793 hlist_for_each_entry(inode
, head
, i_hash
) {
794 if (inode
->i_sb
!= sb
)
796 if (!test(inode
, data
))
798 spin_lock(&inode
->i_lock
);
799 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
800 __wait_on_freeing_inode(inode
);
803 if (unlikely(inode
->i_state
& I_CREATING
)) {
804 spin_unlock(&inode
->i_lock
);
805 return ERR_PTR(-ESTALE
);
808 spin_unlock(&inode
->i_lock
);
815 * find_inode_fast is the fast path version of find_inode, see the comment at
816 * iget_locked for details.
818 static struct inode
*find_inode_fast(struct super_block
*sb
,
819 struct hlist_head
*head
, unsigned long ino
)
821 struct inode
*inode
= NULL
;
824 hlist_for_each_entry(inode
, head
, i_hash
) {
825 if (inode
->i_ino
!= ino
)
827 if (inode
->i_sb
!= sb
)
829 spin_lock(&inode
->i_lock
);
830 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
831 __wait_on_freeing_inode(inode
);
834 if (unlikely(inode
->i_state
& I_CREATING
)) {
835 spin_unlock(&inode
->i_lock
);
836 return ERR_PTR(-ESTALE
);
839 spin_unlock(&inode
->i_lock
);
846 * Each cpu owns a range of LAST_INO_BATCH numbers.
847 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
848 * to renew the exhausted range.
850 * This does not significantly increase overflow rate because every CPU can
851 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
852 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
853 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
854 * overflow rate by 2x, which does not seem too significant.
856 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
857 * error if st_ino won't fit in target struct field. Use 32bit counter
858 * here to attempt to avoid that.
860 #define LAST_INO_BATCH 1024
861 static DEFINE_PER_CPU(unsigned int, last_ino
);
863 unsigned int get_next_ino(void)
865 unsigned int *p
= &get_cpu_var(last_ino
);
866 unsigned int res
= *p
;
869 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
870 static atomic_t shared_last_ino
;
871 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
873 res
= next
- LAST_INO_BATCH
;
878 /* get_next_ino should not provide a 0 inode number */
882 put_cpu_var(last_ino
);
885 EXPORT_SYMBOL(get_next_ino
);
888 * new_inode_pseudo - obtain an inode
891 * Allocates a new inode for given superblock.
892 * Inode wont be chained in superblock s_inodes list
894 * - fs can't be unmount
895 * - quotas, fsnotify, writeback can't work
897 struct inode
*new_inode_pseudo(struct super_block
*sb
)
899 struct inode
*inode
= alloc_inode(sb
);
902 spin_lock(&inode
->i_lock
);
904 spin_unlock(&inode
->i_lock
);
905 INIT_LIST_HEAD(&inode
->i_sb_list
);
911 * new_inode - obtain an inode
914 * Allocates a new inode for given superblock. The default gfp_mask
915 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
916 * If HIGHMEM pages are unsuitable or it is known that pages allocated
917 * for the page cache are not reclaimable or migratable,
918 * mapping_set_gfp_mask() must be called with suitable flags on the
919 * newly created inode's mapping
922 struct inode
*new_inode(struct super_block
*sb
)
926 spin_lock_prefetch(&sb
->s_inode_list_lock
);
928 inode
= new_inode_pseudo(sb
);
930 inode_sb_list_add(inode
);
933 EXPORT_SYMBOL(new_inode
);
935 #ifdef CONFIG_DEBUG_LOCK_ALLOC
936 void lockdep_annotate_inode_mutex_key(struct inode
*inode
)
938 if (S_ISDIR(inode
->i_mode
)) {
939 struct file_system_type
*type
= inode
->i_sb
->s_type
;
941 /* Set new key only if filesystem hasn't already changed it */
942 if (lockdep_match_class(&inode
->i_rwsem
, &type
->i_mutex_key
)) {
944 * ensure nobody is actually holding i_mutex
946 // mutex_destroy(&inode->i_mutex);
947 init_rwsem(&inode
->i_rwsem
);
948 lockdep_set_class(&inode
->i_rwsem
,
949 &type
->i_mutex_dir_key
);
953 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key
);
957 * unlock_new_inode - clear the I_NEW state and wake up any waiters
958 * @inode: new inode to unlock
960 * Called when the inode is fully initialised to clear the new state of the
961 * inode and wake up anyone waiting for the inode to finish initialisation.
963 void unlock_new_inode(struct inode
*inode
)
965 lockdep_annotate_inode_mutex_key(inode
);
966 spin_lock(&inode
->i_lock
);
967 WARN_ON(!(inode
->i_state
& I_NEW
));
968 inode
->i_state
&= ~I_NEW
& ~I_CREATING
;
970 wake_up_bit(&inode
->i_state
, __I_NEW
);
971 spin_unlock(&inode
->i_lock
);
973 EXPORT_SYMBOL(unlock_new_inode
);
975 void discard_new_inode(struct inode
*inode
)
977 lockdep_annotate_inode_mutex_key(inode
);
978 spin_lock(&inode
->i_lock
);
979 WARN_ON(!(inode
->i_state
& I_NEW
));
980 inode
->i_state
&= ~I_NEW
;
982 wake_up_bit(&inode
->i_state
, __I_NEW
);
983 spin_unlock(&inode
->i_lock
);
986 EXPORT_SYMBOL(discard_new_inode
);
989 * lock_two_nondirectories - take two i_mutexes on non-directory objects
991 * Lock any non-NULL argument that is not a directory.
992 * Zero, one or two objects may be locked by this function.
994 * @inode1: first inode to lock
995 * @inode2: second inode to lock
997 void lock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
1000 swap(inode1
, inode2
);
1002 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
1004 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
1005 inode_lock_nested(inode2
, I_MUTEX_NONDIR2
);
1007 EXPORT_SYMBOL(lock_two_nondirectories
);
1010 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1011 * @inode1: first inode to unlock
1012 * @inode2: second inode to unlock
1014 void unlock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
1016 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
1017 inode_unlock(inode1
);
1018 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
1019 inode_unlock(inode2
);
1021 EXPORT_SYMBOL(unlock_two_nondirectories
);
1024 * iget5_locked - obtain an inode from a mounted file system
1025 * @sb: super block of file system
1026 * @hashval: hash value (usually inode number) to get
1027 * @test: callback used for comparisons between inodes
1028 * @set: callback used to initialize a new struct inode
1029 * @data: opaque data pointer to pass to @test and @set
1031 * Search for the inode specified by @hashval and @data in the inode cache,
1032 * and if present it is return it with an increased reference count. This is
1033 * a generalized version of iget_locked() for file systems where the inode
1034 * number is not sufficient for unique identification of an inode.
1036 * If the inode is not in cache, allocate a new inode and return it locked,
1037 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1038 * before unlocking it via unlock_new_inode().
1040 * Note both @test and @set are called with the inode_hash_lock held, so can't
1043 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1044 int (*test
)(struct inode
*, void *),
1045 int (*set
)(struct inode
*, void *), void *data
)
1047 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1048 struct inode
*inode
;
1050 spin_lock(&inode_hash_lock
);
1051 inode
= find_inode(sb
, head
, test
, data
);
1052 spin_unlock(&inode_hash_lock
);
1055 wait_on_inode(inode
);
1056 if (unlikely(inode_unhashed(inode
))) {
1063 inode
= alloc_inode(sb
);
1067 spin_lock(&inode_hash_lock
);
1068 /* We released the lock, so.. */
1069 old
= find_inode(sb
, head
, test
, data
);
1071 if (set(inode
, data
))
1074 spin_lock(&inode
->i_lock
);
1075 inode
->i_state
= I_NEW
;
1076 hlist_add_head(&inode
->i_hash
, head
);
1077 spin_unlock(&inode
->i_lock
);
1078 inode_sb_list_add(inode
);
1079 spin_unlock(&inode_hash_lock
);
1081 /* Return the locked inode with I_NEW set, the
1082 * caller is responsible for filling in the contents
1088 * Uhhuh, somebody else created the same inode under
1089 * us. Use the old inode instead of the one we just
1092 spin_unlock(&inode_hash_lock
);
1093 destroy_inode(inode
);
1095 wait_on_inode(inode
);
1096 if (unlikely(inode_unhashed(inode
))) {
1104 spin_unlock(&inode_hash_lock
);
1105 destroy_inode(inode
);
1108 EXPORT_SYMBOL(iget5_locked
);
1111 * iget_locked - obtain an inode from a mounted file system
1112 * @sb: super block of file system
1113 * @ino: inode number to get
1115 * Search for the inode specified by @ino in the inode cache and if present
1116 * return it with an increased reference count. This is for file systems
1117 * where the inode number is sufficient for unique identification of an inode.
1119 * If the inode is not in cache, allocate a new inode and return it locked,
1120 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1121 * before unlocking it via unlock_new_inode().
1123 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1125 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1126 struct inode
*inode
;
1128 spin_lock(&inode_hash_lock
);
1129 inode
= find_inode_fast(sb
, head
, ino
);
1130 spin_unlock(&inode_hash_lock
);
1134 wait_on_inode(inode
);
1135 if (unlikely(inode_unhashed(inode
))) {
1142 inode
= alloc_inode(sb
);
1146 spin_lock(&inode_hash_lock
);
1147 /* We released the lock, so.. */
1148 old
= find_inode_fast(sb
, head
, ino
);
1151 spin_lock(&inode
->i_lock
);
1152 inode
->i_state
= I_NEW
;
1153 hlist_add_head(&inode
->i_hash
, head
);
1154 spin_unlock(&inode
->i_lock
);
1155 inode_sb_list_add(inode
);
1156 spin_unlock(&inode_hash_lock
);
1158 /* Return the locked inode with I_NEW set, the
1159 * caller is responsible for filling in the contents
1165 * Uhhuh, somebody else created the same inode under
1166 * us. Use the old inode instead of the one we just
1169 spin_unlock(&inode_hash_lock
);
1170 destroy_inode(inode
);
1174 wait_on_inode(inode
);
1175 if (unlikely(inode_unhashed(inode
))) {
1182 EXPORT_SYMBOL(iget_locked
);
1185 * search the inode cache for a matching inode number.
1186 * If we find one, then the inode number we are trying to
1187 * allocate is not unique and so we should not use it.
1189 * Returns 1 if the inode number is unique, 0 if it is not.
1191 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1193 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1194 struct inode
*inode
;
1196 spin_lock(&inode_hash_lock
);
1197 hlist_for_each_entry(inode
, b
, i_hash
) {
1198 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
) {
1199 spin_unlock(&inode_hash_lock
);
1203 spin_unlock(&inode_hash_lock
);
1209 * iunique - get a unique inode number
1211 * @max_reserved: highest reserved inode number
1213 * Obtain an inode number that is unique on the system for a given
1214 * superblock. This is used by file systems that have no natural
1215 * permanent inode numbering system. An inode number is returned that
1216 * is higher than the reserved limit but unique.
1219 * With a large number of inodes live on the file system this function
1220 * currently becomes quite slow.
1222 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1225 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1226 * error if st_ino won't fit in target struct field. Use 32bit counter
1227 * here to attempt to avoid that.
1229 static DEFINE_SPINLOCK(iunique_lock
);
1230 static unsigned int counter
;
1233 spin_lock(&iunique_lock
);
1235 if (counter
<= max_reserved
)
1236 counter
= max_reserved
+ 1;
1238 } while (!test_inode_iunique(sb
, res
));
1239 spin_unlock(&iunique_lock
);
1243 EXPORT_SYMBOL(iunique
);
1245 struct inode
*igrab(struct inode
*inode
)
1247 spin_lock(&inode
->i_lock
);
1248 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1250 spin_unlock(&inode
->i_lock
);
1252 spin_unlock(&inode
->i_lock
);
1254 * Handle the case where s_op->clear_inode is not been
1255 * called yet, and somebody is calling igrab
1256 * while the inode is getting freed.
1262 EXPORT_SYMBOL(igrab
);
1265 * ilookup5_nowait - search for an inode in the inode cache
1266 * @sb: super block of file system to search
1267 * @hashval: hash value (usually inode number) to search for
1268 * @test: callback used for comparisons between inodes
1269 * @data: opaque data pointer to pass to @test
1271 * Search for the inode specified by @hashval and @data in the inode cache.
1272 * If the inode is in the cache, the inode is returned with an incremented
1275 * Note: I_NEW is not waited upon so you have to be very careful what you do
1276 * with the returned inode. You probably should be using ilookup5() instead.
1278 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1280 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1281 int (*test
)(struct inode
*, void *), void *data
)
1283 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1284 struct inode
*inode
;
1286 spin_lock(&inode_hash_lock
);
1287 inode
= find_inode(sb
, head
, test
, data
);
1288 spin_unlock(&inode_hash_lock
);
1290 return IS_ERR(inode
) ? NULL
: inode
;
1292 EXPORT_SYMBOL(ilookup5_nowait
);
1295 * ilookup5 - search for an inode in the inode cache
1296 * @sb: super block of file system to search
1297 * @hashval: hash value (usually inode number) to search for
1298 * @test: callback used for comparisons between inodes
1299 * @data: opaque data pointer to pass to @test
1301 * Search for the inode specified by @hashval and @data in the inode cache,
1302 * and if the inode is in the cache, return the inode with an incremented
1303 * reference count. Waits on I_NEW before returning the inode.
1304 * returned with an incremented reference count.
1306 * This is a generalized version of ilookup() for file systems where the
1307 * inode number is not sufficient for unique identification of an inode.
1309 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1311 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1312 int (*test
)(struct inode
*, void *), void *data
)
1314 struct inode
*inode
;
1316 inode
= ilookup5_nowait(sb
, hashval
, test
, data
);
1320 wait_on_inode(inode
);
1321 if (unlikely(inode_unhashed(inode
))) {
1328 EXPORT_SYMBOL(ilookup5
);
1331 * ilookup - search for an inode in the inode cache
1332 * @sb: super block of file system to search
1333 * @ino: inode number to search for
1335 * Search for the inode @ino in the inode cache, and if the inode is in the
1336 * cache, the inode is returned with an incremented reference count.
1338 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1340 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1341 struct inode
*inode
;
1343 spin_lock(&inode_hash_lock
);
1344 inode
= find_inode_fast(sb
, head
, ino
);
1345 spin_unlock(&inode_hash_lock
);
1348 wait_on_inode(inode
);
1349 if (unlikely(inode_unhashed(inode
))) {
1356 EXPORT_SYMBOL(ilookup
);
1359 * find_inode_nowait - find an inode in the inode cache
1360 * @sb: super block of file system to search
1361 * @hashval: hash value (usually inode number) to search for
1362 * @match: callback used for comparisons between inodes
1363 * @data: opaque data pointer to pass to @match
1365 * Search for the inode specified by @hashval and @data in the inode
1366 * cache, where the helper function @match will return 0 if the inode
1367 * does not match, 1 if the inode does match, and -1 if the search
1368 * should be stopped. The @match function must be responsible for
1369 * taking the i_lock spin_lock and checking i_state for an inode being
1370 * freed or being initialized, and incrementing the reference count
1371 * before returning 1. It also must not sleep, since it is called with
1372 * the inode_hash_lock spinlock held.
1374 * This is a even more generalized version of ilookup5() when the
1375 * function must never block --- find_inode() can block in
1376 * __wait_on_freeing_inode() --- or when the caller can not increment
1377 * the reference count because the resulting iput() might cause an
1378 * inode eviction. The tradeoff is that the @match funtion must be
1379 * very carefully implemented.
1381 struct inode
*find_inode_nowait(struct super_block
*sb
,
1382 unsigned long hashval
,
1383 int (*match
)(struct inode
*, unsigned long,
1387 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1388 struct inode
*inode
, *ret_inode
= NULL
;
1391 spin_lock(&inode_hash_lock
);
1392 hlist_for_each_entry(inode
, head
, i_hash
) {
1393 if (inode
->i_sb
!= sb
)
1395 mval
= match(inode
, hashval
, data
);
1403 spin_unlock(&inode_hash_lock
);
1406 EXPORT_SYMBOL(find_inode_nowait
);
1408 int insert_inode_locked(struct inode
*inode
)
1410 struct super_block
*sb
= inode
->i_sb
;
1411 ino_t ino
= inode
->i_ino
;
1412 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1415 struct inode
*old
= NULL
;
1416 spin_lock(&inode_hash_lock
);
1417 hlist_for_each_entry(old
, head
, i_hash
) {
1418 if (old
->i_ino
!= ino
)
1420 if (old
->i_sb
!= sb
)
1422 spin_lock(&old
->i_lock
);
1423 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1424 spin_unlock(&old
->i_lock
);
1430 spin_lock(&inode
->i_lock
);
1431 inode
->i_state
|= I_NEW
| I_CREATING
;
1432 hlist_add_head(&inode
->i_hash
, head
);
1433 spin_unlock(&inode
->i_lock
);
1434 spin_unlock(&inode_hash_lock
);
1437 if (unlikely(old
->i_state
& I_CREATING
)) {
1438 spin_unlock(&old
->i_lock
);
1439 spin_unlock(&inode_hash_lock
);
1443 spin_unlock(&old
->i_lock
);
1444 spin_unlock(&inode_hash_lock
);
1446 if (unlikely(!inode_unhashed(old
))) {
1453 EXPORT_SYMBOL(insert_inode_locked
);
1455 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1456 int (*test
)(struct inode
*, void *), void *data
)
1458 struct super_block
*sb
= inode
->i_sb
;
1459 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1461 inode
->i_state
|= I_CREATING
;
1464 struct inode
*old
= NULL
;
1466 spin_lock(&inode_hash_lock
);
1467 hlist_for_each_entry(old
, head
, i_hash
) {
1468 if (old
->i_sb
!= sb
)
1470 if (!test(old
, data
))
1472 spin_lock(&old
->i_lock
);
1473 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1474 spin_unlock(&old
->i_lock
);
1480 spin_lock(&inode
->i_lock
);
1481 inode
->i_state
|= I_NEW
| I_CREATING
;
1482 hlist_add_head(&inode
->i_hash
, head
);
1483 spin_unlock(&inode
->i_lock
);
1484 spin_unlock(&inode_hash_lock
);
1487 if (unlikely(old
->i_state
& I_CREATING
)) {
1488 spin_unlock(&old
->i_lock
);
1489 spin_unlock(&inode_hash_lock
);
1493 spin_unlock(&old
->i_lock
);
1494 spin_unlock(&inode_hash_lock
);
1496 if (unlikely(!inode_unhashed(old
))) {
1503 EXPORT_SYMBOL(insert_inode_locked4
);
1506 int generic_delete_inode(struct inode
*inode
)
1510 EXPORT_SYMBOL(generic_delete_inode
);
1513 * Called when we're dropping the last reference
1516 * Call the FS "drop_inode()" function, defaulting to
1517 * the legacy UNIX filesystem behaviour. If it tells
1518 * us to evict inode, do so. Otherwise, retain inode
1519 * in cache if fs is alive, sync and evict if fs is
1522 static void iput_final(struct inode
*inode
)
1524 struct super_block
*sb
= inode
->i_sb
;
1525 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1528 WARN_ON(inode
->i_state
& I_NEW
);
1531 drop
= op
->drop_inode(inode
);
1533 drop
= generic_drop_inode(inode
);
1535 if (!drop
&& (sb
->s_flags
& SB_ACTIVE
)) {
1536 inode_add_lru(inode
);
1537 spin_unlock(&inode
->i_lock
);
1542 inode
->i_state
|= I_WILL_FREE
;
1543 spin_unlock(&inode
->i_lock
);
1544 write_inode_now(inode
, 1);
1545 spin_lock(&inode
->i_lock
);
1546 WARN_ON(inode
->i_state
& I_NEW
);
1547 inode
->i_state
&= ~I_WILL_FREE
;
1550 inode
->i_state
|= I_FREEING
;
1551 if (!list_empty(&inode
->i_lru
))
1552 inode_lru_list_del(inode
);
1553 spin_unlock(&inode
->i_lock
);
1559 * iput - put an inode
1560 * @inode: inode to put
1562 * Puts an inode, dropping its usage count. If the inode use count hits
1563 * zero, the inode is then freed and may also be destroyed.
1565 * Consequently, iput() can sleep.
1567 void iput(struct inode
*inode
)
1571 BUG_ON(inode
->i_state
& I_CLEAR
);
1573 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
)) {
1574 if (inode
->i_nlink
&& (inode
->i_state
& I_DIRTY_TIME
)) {
1575 atomic_inc(&inode
->i_count
);
1576 inode
->i_state
&= ~I_DIRTY_TIME
;
1577 spin_unlock(&inode
->i_lock
);
1578 trace_writeback_lazytime_iput(inode
);
1579 mark_inode_dirty_sync(inode
);
1585 EXPORT_SYMBOL(iput
);
1588 * bmap - find a block number in a file
1589 * @inode: inode of file
1590 * @block: block to find
1592 * Returns the block number on the device holding the inode that
1593 * is the disk block number for the block of the file requested.
1594 * That is, asked for block 4 of inode 1 the function will return the
1595 * disk block relative to the disk start that holds that block of the
1598 sector_t
bmap(struct inode
*inode
, sector_t block
)
1601 if (inode
->i_mapping
->a_ops
->bmap
)
1602 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1605 EXPORT_SYMBOL(bmap
);
1608 * Update times in overlayed inode from underlying real inode
1610 static void update_ovl_inode_times(struct dentry
*dentry
, struct inode
*inode
,
1613 struct dentry
*upperdentry
;
1616 * Nothing to do if in rcu or if non-overlayfs
1618 if (rcu
|| likely(!(dentry
->d_flags
& DCACHE_OP_REAL
)))
1621 upperdentry
= d_real(dentry
, NULL
, 0, D_REAL_UPPER
);
1624 * If file is on lower then we can't update atime, so no worries about
1625 * stale mtime/ctime.
1628 struct inode
*realinode
= d_inode(upperdentry
);
1630 if ((!timespec_equal(&inode
->i_mtime
, &realinode
->i_mtime
) ||
1631 !timespec_equal(&inode
->i_ctime
, &realinode
->i_ctime
))) {
1632 inode
->i_mtime
= realinode
->i_mtime
;
1633 inode
->i_ctime
= realinode
->i_ctime
;
1639 * With relative atime, only update atime if the previous atime is
1640 * earlier than either the ctime or mtime or if at least a day has
1641 * passed since the last atime update.
1643 static int relatime_need_update(const struct path
*path
, struct inode
*inode
,
1644 struct timespec now
, bool rcu
)
1647 if (!(path
->mnt
->mnt_flags
& MNT_RELATIME
))
1650 update_ovl_inode_times(path
->dentry
, inode
, rcu
);
1652 * Is mtime younger than atime? If yes, update atime:
1654 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1657 * Is ctime younger than atime? If yes, update atime:
1659 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1663 * Is the previous atime value older than a day? If yes,
1666 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1669 * Good, we can skip the atime update:
1674 int generic_update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1676 int iflags
= I_DIRTY_TIME
;
1678 if (flags
& S_ATIME
)
1679 inode
->i_atime
= *time
;
1680 if (flags
& S_VERSION
)
1681 inode_inc_iversion(inode
);
1682 if (flags
& S_CTIME
)
1683 inode
->i_ctime
= *time
;
1684 if (flags
& S_MTIME
)
1685 inode
->i_mtime
= *time
;
1687 if (!(inode
->i_sb
->s_flags
& SB_LAZYTIME
) || (flags
& S_VERSION
))
1688 iflags
|= I_DIRTY_SYNC
;
1689 __mark_inode_dirty(inode
, iflags
);
1692 EXPORT_SYMBOL(generic_update_time
);
1695 * This does the actual work of updating an inodes time or version. Must have
1696 * had called mnt_want_write() before calling this.
1698 int update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1700 int (*update_time
)(struct inode
*, struct timespec
*, int);
1702 update_time
= inode
->i_op
->update_time
? inode
->i_op
->update_time
:
1703 generic_update_time
;
1705 return update_time(inode
, time
, flags
);
1707 EXPORT_SYMBOL_GPL(update_time
);
1710 * touch_atime - update the access time
1711 * @path: the &struct path to update
1712 * @inode: inode to update
1714 * Update the accessed time on an inode and mark it for writeback.
1715 * This function automatically handles read only file systems and media,
1716 * as well as the "noatime" flag and inode specific "noatime" markers.
1718 bool __atime_needs_update(const struct path
*path
, struct inode
*inode
,
1721 struct vfsmount
*mnt
= path
->mnt
;
1722 struct timespec now
;
1724 if (inode
->i_flags
& S_NOATIME
)
1727 /* Atime updates will likely cause i_uid and i_gid to be written
1728 * back improprely if their true value is unknown to the vfs.
1730 if (HAS_UNMAPPED_ID(inode
))
1733 if (IS_NOATIME(inode
))
1735 if ((inode
->i_sb
->s_flags
& SB_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1738 if (mnt
->mnt_flags
& MNT_NOATIME
)
1740 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1743 now
= current_time(inode
);
1745 if (!relatime_need_update(path
, inode
, now
, rcu
))
1748 if (timespec_equal(&inode
->i_atime
, &now
))
1754 void touch_atime(const struct path
*path
)
1756 struct vfsmount
*mnt
= path
->mnt
;
1757 struct inode
*inode
= d_inode(path
->dentry
);
1758 struct timespec now
;
1760 if (!__atime_needs_update(path
, inode
, false))
1763 if (!sb_start_write_trylock(inode
->i_sb
))
1766 if (__mnt_want_write(mnt
) != 0)
1769 * File systems can error out when updating inodes if they need to
1770 * allocate new space to modify an inode (such is the case for
1771 * Btrfs), but since we touch atime while walking down the path we
1772 * really don't care if we failed to update the atime of the file,
1773 * so just ignore the return value.
1774 * We may also fail on filesystems that have the ability to make parts
1775 * of the fs read only, e.g. subvolumes in Btrfs.
1777 now
= current_time(inode
);
1778 update_time(inode
, &now
, S_ATIME
);
1779 __mnt_drop_write(mnt
);
1781 sb_end_write(inode
->i_sb
);
1783 EXPORT_SYMBOL(touch_atime
);
1786 * The logic we want is
1788 * if suid or (sgid and xgrp)
1791 int should_remove_suid(struct dentry
*dentry
)
1793 struct inode
*inode
= d_inode(dentry
);
1794 umode_t mode
= inode
->i_mode
;
1797 /* suid always must be killed */
1798 if (unlikely(mode
& S_ISUID
))
1799 kill
= ATTR_KILL_SUID
;
1802 * sgid without any exec bits is just a mandatory locking mark; leave
1803 * it alone. If some exec bits are set, it's a real sgid; kill it.
1805 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1806 kill
|= ATTR_KILL_SGID
;
1808 if (unlikely(kill
&& !capable_wrt_inode_uidgid(inode
, CAP_FSETID
) &&
1814 EXPORT_SYMBOL(should_remove_suid
);
1817 * Return mask of changes for notify_change() that need to be done as a
1818 * response to write or truncate. Return 0 if nothing has to be changed.
1819 * Negative value on error (change should be denied).
1821 int dentry_needs_remove_privs(struct dentry
*dentry
)
1823 struct inode
*inode
= d_inode(dentry
);
1827 if (IS_NOSEC(inode
))
1830 mask
= should_remove_suid(dentry
);
1831 ret
= security_inode_need_killpriv(dentry
);
1835 mask
|= ATTR_KILL_PRIV
;
1839 static int __remove_privs(struct dentry
*dentry
, int kill
)
1841 struct iattr newattrs
;
1843 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1845 * Note we call this on write, so notify_change will not
1846 * encounter any conflicting delegations:
1848 return notify_change(dentry
, &newattrs
, NULL
);
1852 * Remove special file priviledges (suid, capabilities) when file is written
1855 int file_remove_privs(struct file
*file
)
1857 struct dentry
*dentry
= file_dentry(file
);
1858 struct inode
*inode
= file_inode(file
);
1862 /* Fast path for nothing security related */
1863 if (IS_NOSEC(inode
))
1866 kill
= dentry_needs_remove_privs(dentry
);
1870 error
= __remove_privs(dentry
, kill
);
1872 inode_has_no_xattr(inode
);
1876 EXPORT_SYMBOL(file_remove_privs
);
1879 * file_update_time - update mtime and ctime time
1880 * @file: file accessed
1882 * Update the mtime and ctime members of an inode and mark the inode
1883 * for writeback. Note that this function is meant exclusively for
1884 * usage in the file write path of filesystems, and filesystems may
1885 * choose to explicitly ignore update via this function with the
1886 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1887 * timestamps are handled by the server. This can return an error for
1888 * file systems who need to allocate space in order to update an inode.
1891 int file_update_time(struct file
*file
)
1893 struct inode
*inode
= file_inode(file
);
1894 struct timespec now
;
1898 /* First try to exhaust all avenues to not sync */
1899 if (IS_NOCMTIME(inode
))
1902 now
= current_time(inode
);
1903 if (!timespec_equal(&inode
->i_mtime
, &now
))
1906 if (!timespec_equal(&inode
->i_ctime
, &now
))
1909 if (IS_I_VERSION(inode
))
1910 sync_it
|= S_VERSION
;
1915 /* Finally allowed to write? Takes lock. */
1916 if (__mnt_want_write_file(file
))
1919 ret
= update_time(inode
, &now
, sync_it
);
1920 __mnt_drop_write_file(file
);
1924 EXPORT_SYMBOL(file_update_time
);
1926 int inode_needs_sync(struct inode
*inode
)
1930 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1934 EXPORT_SYMBOL(inode_needs_sync
);
1937 * If we try to find an inode in the inode hash while it is being
1938 * deleted, we have to wait until the filesystem completes its
1939 * deletion before reporting that it isn't found. This function waits
1940 * until the deletion _might_ have completed. Callers are responsible
1941 * to recheck inode state.
1943 * It doesn't matter if I_NEW is not set initially, a call to
1944 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1947 static void __wait_on_freeing_inode(struct inode
*inode
)
1949 wait_queue_head_t
*wq
;
1950 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1951 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1952 prepare_to_wait(wq
, &wait
.wq_entry
, TASK_UNINTERRUPTIBLE
);
1953 spin_unlock(&inode
->i_lock
);
1954 spin_unlock(&inode_hash_lock
);
1956 finish_wait(wq
, &wait
.wq_entry
);
1957 spin_lock(&inode_hash_lock
);
1960 static __initdata
unsigned long ihash_entries
;
1961 static int __init
set_ihash_entries(char *str
)
1965 ihash_entries
= simple_strtoul(str
, &str
, 0);
1968 __setup("ihash_entries=", set_ihash_entries
);
1971 * Initialize the waitqueues and inode hash table.
1973 void __init
inode_init_early(void)
1975 /* If hashes are distributed across NUMA nodes, defer
1976 * hash allocation until vmalloc space is available.
1982 alloc_large_system_hash("Inode-cache",
1983 sizeof(struct hlist_head
),
1986 HASH_EARLY
| HASH_ZERO
,
1993 void __init
inode_init(void)
1995 /* inode slab cache */
1996 inode_cachep
= kmem_cache_create("inode_cache",
1997 sizeof(struct inode
),
1999 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
2000 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
),
2003 /* Hash may have been set up in inode_init_early */
2008 alloc_large_system_hash("Inode-cache",
2009 sizeof(struct hlist_head
),
2019 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
2021 inode
->i_mode
= mode
;
2022 if (S_ISCHR(mode
)) {
2023 inode
->i_fop
= &def_chr_fops
;
2024 inode
->i_rdev
= rdev
;
2025 } else if (S_ISBLK(mode
)) {
2026 inode
->i_fop
= &def_blk_fops
;
2027 inode
->i_rdev
= rdev
;
2028 } else if (S_ISFIFO(mode
))
2029 inode
->i_fop
= &pipefifo_fops
;
2030 else if (S_ISSOCK(mode
))
2031 ; /* leave it no_open_fops */
2033 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
2034 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
2037 EXPORT_SYMBOL(init_special_inode
);
2040 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2042 * @dir: Directory inode
2043 * @mode: mode of the new inode
2045 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
2048 inode
->i_uid
= current_fsuid();
2049 if (dir
&& dir
->i_mode
& S_ISGID
) {
2050 inode
->i_gid
= dir
->i_gid
;
2052 /* Directories are special, and always inherit S_ISGID */
2055 else if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
) &&
2056 !in_group_p(inode
->i_gid
) &&
2057 !capable_wrt_inode_uidgid(dir
, CAP_FSETID
))
2060 inode
->i_gid
= current_fsgid();
2061 inode
->i_mode
= mode
;
2063 EXPORT_SYMBOL(inode_init_owner
);
2066 * inode_owner_or_capable - check current task permissions to inode
2067 * @inode: inode being checked
2069 * Return true if current either has CAP_FOWNER in a namespace with the
2070 * inode owner uid mapped, or owns the file.
2072 bool inode_owner_or_capable(const struct inode
*inode
)
2074 struct user_namespace
*ns
;
2076 if (uid_eq(current_fsuid(), inode
->i_uid
))
2079 ns
= current_user_ns();
2080 if (kuid_has_mapping(ns
, inode
->i_uid
) && ns_capable(ns
, CAP_FOWNER
))
2084 EXPORT_SYMBOL(inode_owner_or_capable
);
2087 * Direct i/o helper functions
2089 static void __inode_dio_wait(struct inode
*inode
)
2091 wait_queue_head_t
*wq
= bit_waitqueue(&inode
->i_state
, __I_DIO_WAKEUP
);
2092 DEFINE_WAIT_BIT(q
, &inode
->i_state
, __I_DIO_WAKEUP
);
2095 prepare_to_wait(wq
, &q
.wq_entry
, TASK_UNINTERRUPTIBLE
);
2096 if (atomic_read(&inode
->i_dio_count
))
2098 } while (atomic_read(&inode
->i_dio_count
));
2099 finish_wait(wq
, &q
.wq_entry
);
2103 * inode_dio_wait - wait for outstanding DIO requests to finish
2104 * @inode: inode to wait for
2106 * Waits for all pending direct I/O requests to finish so that we can
2107 * proceed with a truncate or equivalent operation.
2109 * Must be called under a lock that serializes taking new references
2110 * to i_dio_count, usually by inode->i_mutex.
2112 void inode_dio_wait(struct inode
*inode
)
2114 if (atomic_read(&inode
->i_dio_count
))
2115 __inode_dio_wait(inode
);
2117 EXPORT_SYMBOL(inode_dio_wait
);
2120 * inode_set_flags - atomically set some inode flags
2122 * Note: the caller should be holding i_mutex, or else be sure that
2123 * they have exclusive access to the inode structure (i.e., while the
2124 * inode is being instantiated). The reason for the cmpxchg() loop
2125 * --- which wouldn't be necessary if all code paths which modify
2126 * i_flags actually followed this rule, is that there is at least one
2127 * code path which doesn't today so we use cmpxchg() out of an abundance
2130 * In the long run, i_mutex is overkill, and we should probably look
2131 * at using the i_lock spinlock to protect i_flags, and then make sure
2132 * it is so documented in include/linux/fs.h and that all code follows
2133 * the locking convention!!
2135 void inode_set_flags(struct inode
*inode
, unsigned int flags
,
2138 unsigned int old_flags
, new_flags
;
2140 WARN_ON_ONCE(flags
& ~mask
);
2142 old_flags
= READ_ONCE(inode
->i_flags
);
2143 new_flags
= (old_flags
& ~mask
) | flags
;
2144 } while (unlikely(cmpxchg(&inode
->i_flags
, old_flags
,
2145 new_flags
) != old_flags
));
2147 EXPORT_SYMBOL(inode_set_flags
);
2149 void inode_nohighmem(struct inode
*inode
)
2151 mapping_set_gfp_mask(inode
->i_mapping
, GFP_USER
);
2153 EXPORT_SYMBOL(inode_nohighmem
);
2156 * current_time - Return FS time
2159 * Return the current time truncated to the time granularity supported by
2162 * Note that inode and inode->sb cannot be NULL.
2163 * Otherwise, the function warns and returns time without truncation.
2165 struct timespec
current_time(struct inode
*inode
)
2167 struct timespec now
= current_kernel_time();
2169 if (unlikely(!inode
->i_sb
)) {
2170 WARN(1, "current_time() called with uninitialized super_block in the inode");
2174 return timespec_trunc(now
, inode
->i_sb
->s_time_gran
);
2176 EXPORT_SYMBOL(current_time
);