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 intialisation
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);
151 inode
->i_generation
= 0;
152 inode
->i_pipe
= NULL
;
153 inode
->i_bdev
= NULL
;
154 inode
->i_cdev
= NULL
;
155 inode
->i_link
= NULL
;
156 inode
->i_dir_seq
= 0;
158 inode
->dirtied_when
= 0;
160 #ifdef CONFIG_CGROUP_WRITEBACK
161 inode
->i_wb_frn_winner
= 0;
162 inode
->i_wb_frn_avg_time
= 0;
163 inode
->i_wb_frn_history
= 0;
166 if (security_inode_alloc(inode
))
168 spin_lock_init(&inode
->i_lock
);
169 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
171 init_rwsem(&inode
->i_rwsem
);
172 lockdep_set_class(&inode
->i_rwsem
, &sb
->s_type
->i_mutex_key
);
174 atomic_set(&inode
->i_dio_count
, 0);
176 mapping
->a_ops
= &empty_aops
;
177 mapping
->host
= inode
;
179 atomic_set(&mapping
->i_mmap_writable
, 0);
180 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
181 mapping
->private_data
= NULL
;
182 mapping
->writeback_index
= 0;
183 inode
->i_private
= NULL
;
184 inode
->i_mapping
= mapping
;
185 INIT_HLIST_HEAD(&inode
->i_dentry
); /* buggered by rcu freeing */
186 #ifdef CONFIG_FS_POSIX_ACL
187 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
190 #ifdef CONFIG_FSNOTIFY
191 inode
->i_fsnotify_mask
= 0;
193 inode
->i_flctx
= NULL
;
194 this_cpu_inc(nr_inodes
);
200 EXPORT_SYMBOL(inode_init_always
);
202 static struct inode
*alloc_inode(struct super_block
*sb
)
206 if (sb
->s_op
->alloc_inode
)
207 inode
= sb
->s_op
->alloc_inode(sb
);
209 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
214 if (unlikely(inode_init_always(sb
, inode
))) {
215 if (inode
->i_sb
->s_op
->destroy_inode
)
216 inode
->i_sb
->s_op
->destroy_inode(inode
);
218 kmem_cache_free(inode_cachep
, inode
);
225 void free_inode_nonrcu(struct inode
*inode
)
227 kmem_cache_free(inode_cachep
, inode
);
229 EXPORT_SYMBOL(free_inode_nonrcu
);
231 void __destroy_inode(struct inode
*inode
)
233 BUG_ON(inode_has_buffers(inode
));
234 inode_detach_wb(inode
);
235 security_inode_free(inode
);
236 fsnotify_inode_delete(inode
);
237 locks_free_lock_context(inode
);
238 if (!inode
->i_nlink
) {
239 WARN_ON(atomic_long_read(&inode
->i_sb
->s_remove_count
) == 0);
240 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
243 #ifdef CONFIG_FS_POSIX_ACL
244 if (inode
->i_acl
&& !is_uncached_acl(inode
->i_acl
))
245 posix_acl_release(inode
->i_acl
);
246 if (inode
->i_default_acl
&& !is_uncached_acl(inode
->i_default_acl
))
247 posix_acl_release(inode
->i_default_acl
);
249 this_cpu_dec(nr_inodes
);
251 EXPORT_SYMBOL(__destroy_inode
);
253 static void i_callback(struct rcu_head
*head
)
255 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
256 kmem_cache_free(inode_cachep
, inode
);
259 static void destroy_inode(struct inode
*inode
)
261 BUG_ON(!list_empty(&inode
->i_lru
));
262 __destroy_inode(inode
);
263 if (inode
->i_sb
->s_op
->destroy_inode
)
264 inode
->i_sb
->s_op
->destroy_inode(inode
);
266 call_rcu(&inode
->i_rcu
, i_callback
);
270 * drop_nlink - directly drop an inode's link count
273 * This is a low-level filesystem helper to replace any
274 * direct filesystem manipulation of i_nlink. In cases
275 * where we are attempting to track writes to the
276 * filesystem, a decrement to zero means an imminent
277 * write when the file is truncated and actually unlinked
280 void drop_nlink(struct inode
*inode
)
282 WARN_ON(inode
->i_nlink
== 0);
285 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
287 EXPORT_SYMBOL(drop_nlink
);
290 * clear_nlink - directly zero an inode's link count
293 * This is a low-level filesystem helper to replace any
294 * direct filesystem manipulation of i_nlink. See
295 * drop_nlink() for why we care about i_nlink hitting zero.
297 void clear_nlink(struct inode
*inode
)
299 if (inode
->i_nlink
) {
300 inode
->__i_nlink
= 0;
301 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
304 EXPORT_SYMBOL(clear_nlink
);
307 * set_nlink - directly set an inode's link count
309 * @nlink: new nlink (should be non-zero)
311 * This is a low-level filesystem helper to replace any
312 * direct filesystem manipulation of i_nlink.
314 void set_nlink(struct inode
*inode
, unsigned int nlink
)
319 /* Yes, some filesystems do change nlink from zero to one */
320 if (inode
->i_nlink
== 0)
321 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
323 inode
->__i_nlink
= nlink
;
326 EXPORT_SYMBOL(set_nlink
);
329 * inc_nlink - directly increment an inode's link count
332 * This is a low-level filesystem helper to replace any
333 * direct filesystem manipulation of i_nlink. Currently,
334 * it is only here for parity with dec_nlink().
336 void inc_nlink(struct inode
*inode
)
338 if (unlikely(inode
->i_nlink
== 0)) {
339 WARN_ON(!(inode
->i_state
& I_LINKABLE
));
340 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
345 EXPORT_SYMBOL(inc_nlink
);
347 void address_space_init_once(struct address_space
*mapping
)
349 memset(mapping
, 0, sizeof(*mapping
));
350 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
| __GFP_ACCOUNT
);
351 spin_lock_init(&mapping
->tree_lock
);
352 init_rwsem(&mapping
->i_mmap_rwsem
);
353 INIT_LIST_HEAD(&mapping
->private_list
);
354 spin_lock_init(&mapping
->private_lock
);
355 mapping
->i_mmap
= RB_ROOT
;
357 EXPORT_SYMBOL(address_space_init_once
);
360 * These are initializations that only need to be done
361 * once, because the fields are idempotent across use
362 * of the inode, so let the slab aware of that.
364 void inode_init_once(struct inode
*inode
)
366 memset(inode
, 0, sizeof(*inode
));
367 INIT_HLIST_NODE(&inode
->i_hash
);
368 INIT_LIST_HEAD(&inode
->i_devices
);
369 INIT_LIST_HEAD(&inode
->i_io_list
);
370 INIT_LIST_HEAD(&inode
->i_wb_list
);
371 INIT_LIST_HEAD(&inode
->i_lru
);
372 address_space_init_once(&inode
->i_data
);
373 i_size_ordered_init(inode
);
375 EXPORT_SYMBOL(inode_init_once
);
377 static void init_once(void *foo
)
379 struct inode
*inode
= (struct inode
*) foo
;
381 inode_init_once(inode
);
385 * inode->i_lock must be held
387 void __iget(struct inode
*inode
)
389 atomic_inc(&inode
->i_count
);
393 * get additional reference to inode; caller must already hold one.
395 void ihold(struct inode
*inode
)
397 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
399 EXPORT_SYMBOL(ihold
);
401 static void inode_lru_list_add(struct inode
*inode
)
403 if (list_lru_add(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
404 this_cpu_inc(nr_unused
);
408 * Add inode to LRU if needed (inode is unused and clean).
410 * Needs inode->i_lock held.
412 void inode_add_lru(struct inode
*inode
)
414 if (!(inode
->i_state
& (I_DIRTY_ALL
| I_SYNC
|
415 I_FREEING
| I_WILL_FREE
)) &&
416 !atomic_read(&inode
->i_count
) && inode
->i_sb
->s_flags
& MS_ACTIVE
)
417 inode_lru_list_add(inode
);
421 static void inode_lru_list_del(struct inode
*inode
)
424 if (list_lru_del(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
425 this_cpu_dec(nr_unused
);
429 * inode_sb_list_add - add inode to the superblock list of inodes
430 * @inode: inode to add
432 void inode_sb_list_add(struct inode
*inode
)
434 spin_lock(&inode
->i_sb
->s_inode_list_lock
);
435 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
436 spin_unlock(&inode
->i_sb
->s_inode_list_lock
);
438 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
440 static inline void inode_sb_list_del(struct inode
*inode
)
442 if (!list_empty(&inode
->i_sb_list
)) {
443 spin_lock(&inode
->i_sb
->s_inode_list_lock
);
444 list_del_init(&inode
->i_sb_list
);
445 spin_unlock(&inode
->i_sb
->s_inode_list_lock
);
449 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
453 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
455 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> i_hash_shift
);
456 return tmp
& i_hash_mask
;
460 * __insert_inode_hash - hash an inode
461 * @inode: unhashed inode
462 * @hashval: unsigned long value used to locate this object in the
465 * Add an inode to the inode hash for this superblock.
467 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
469 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
471 spin_lock(&inode_hash_lock
);
472 spin_lock(&inode
->i_lock
);
473 hlist_add_head(&inode
->i_hash
, b
);
474 spin_unlock(&inode
->i_lock
);
475 spin_unlock(&inode_hash_lock
);
477 EXPORT_SYMBOL(__insert_inode_hash
);
480 * __remove_inode_hash - remove an inode from the hash
481 * @inode: inode to unhash
483 * Remove an inode from the superblock.
485 void __remove_inode_hash(struct inode
*inode
)
487 spin_lock(&inode_hash_lock
);
488 spin_lock(&inode
->i_lock
);
489 hlist_del_init(&inode
->i_hash
);
490 spin_unlock(&inode
->i_lock
);
491 spin_unlock(&inode_hash_lock
);
493 EXPORT_SYMBOL(__remove_inode_hash
);
495 void clear_inode(struct inode
*inode
)
499 * We have to cycle tree_lock here because reclaim can be still in the
500 * process of removing the last page (in __delete_from_page_cache())
501 * and we must not free mapping under it.
503 spin_lock_irq(&inode
->i_data
.tree_lock
);
504 BUG_ON(inode
->i_data
.nrpages
);
505 BUG_ON(inode
->i_data
.nrexceptional
);
506 spin_unlock_irq(&inode
->i_data
.tree_lock
);
507 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
508 BUG_ON(!(inode
->i_state
& I_FREEING
));
509 BUG_ON(inode
->i_state
& I_CLEAR
);
510 BUG_ON(!list_empty(&inode
->i_wb_list
));
511 /* don't need i_lock here, no concurrent mods to i_state */
512 inode
->i_state
= I_FREEING
| I_CLEAR
;
514 EXPORT_SYMBOL(clear_inode
);
517 * Free the inode passed in, removing it from the lists it is still connected
518 * to. We remove any pages still attached to the inode and wait for any IO that
519 * is still in progress before finally destroying the inode.
521 * An inode must already be marked I_FREEING so that we avoid the inode being
522 * moved back onto lists if we race with other code that manipulates the lists
523 * (e.g. writeback_single_inode). The caller is responsible for setting this.
525 * An inode must already be removed from the LRU list before being evicted from
526 * the cache. This should occur atomically with setting the I_FREEING state
527 * flag, so no inodes here should ever be on the LRU when being evicted.
529 static void evict(struct inode
*inode
)
531 const struct super_operations
*op
= inode
->i_sb
->s_op
;
533 BUG_ON(!(inode
->i_state
& I_FREEING
));
534 BUG_ON(!list_empty(&inode
->i_lru
));
536 if (!list_empty(&inode
->i_io_list
))
537 inode_io_list_del(inode
);
539 inode_sb_list_del(inode
);
542 * Wait for flusher thread to be done with the inode so that filesystem
543 * does not start destroying it while writeback is still running. Since
544 * the inode has I_FREEING set, flusher thread won't start new work on
545 * the inode. We just have to wait for running writeback to finish.
547 inode_wait_for_writeback(inode
);
549 if (op
->evict_inode
) {
550 op
->evict_inode(inode
);
552 truncate_inode_pages_final(&inode
->i_data
);
555 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
557 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
560 remove_inode_hash(inode
);
562 spin_lock(&inode
->i_lock
);
563 wake_up_bit(&inode
->i_state
, __I_NEW
);
564 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
565 spin_unlock(&inode
->i_lock
);
567 destroy_inode(inode
);
571 * dispose_list - dispose of the contents of a local list
572 * @head: the head of the list to free
574 * Dispose-list gets a local list with local inodes in it, so it doesn't
575 * need to worry about list corruption and SMP locks.
577 static void dispose_list(struct list_head
*head
)
579 while (!list_empty(head
)) {
582 inode
= list_first_entry(head
, struct inode
, i_lru
);
583 list_del_init(&inode
->i_lru
);
591 * evict_inodes - evict all evictable inodes for a superblock
592 * @sb: superblock to operate on
594 * Make sure that no inodes with zero refcount are retained. This is
595 * called by superblock shutdown after having MS_ACTIVE flag removed,
596 * so any inode reaching zero refcount during or after that call will
597 * be immediately evicted.
599 void evict_inodes(struct super_block
*sb
)
601 struct inode
*inode
, *next
;
605 spin_lock(&sb
->s_inode_list_lock
);
606 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
607 if (atomic_read(&inode
->i_count
))
610 spin_lock(&inode
->i_lock
);
611 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
612 spin_unlock(&inode
->i_lock
);
616 inode
->i_state
|= I_FREEING
;
617 inode_lru_list_del(inode
);
618 spin_unlock(&inode
->i_lock
);
619 list_add(&inode
->i_lru
, &dispose
);
622 * We can have a ton of inodes to evict at unmount time given
623 * enough memory, check to see if we need to go to sleep for a
624 * bit so we don't livelock.
626 if (need_resched()) {
627 spin_unlock(&sb
->s_inode_list_lock
);
629 dispose_list(&dispose
);
633 spin_unlock(&sb
->s_inode_list_lock
);
635 dispose_list(&dispose
);
639 * invalidate_inodes - attempt to free all inodes on a superblock
640 * @sb: superblock to operate on
641 * @kill_dirty: flag to guide handling of dirty inodes
643 * Attempts to free all inodes for a given superblock. If there were any
644 * busy inodes return a non-zero value, else zero.
645 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
648 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
651 struct inode
*inode
, *next
;
654 spin_lock(&sb
->s_inode_list_lock
);
655 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
656 spin_lock(&inode
->i_lock
);
657 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
658 spin_unlock(&inode
->i_lock
);
661 if (inode
->i_state
& I_DIRTY_ALL
&& !kill_dirty
) {
662 spin_unlock(&inode
->i_lock
);
666 if (atomic_read(&inode
->i_count
)) {
667 spin_unlock(&inode
->i_lock
);
672 inode
->i_state
|= I_FREEING
;
673 inode_lru_list_del(inode
);
674 spin_unlock(&inode
->i_lock
);
675 list_add(&inode
->i_lru
, &dispose
);
677 spin_unlock(&sb
->s_inode_list_lock
);
679 dispose_list(&dispose
);
685 * Isolate the inode from the LRU in preparation for freeing it.
687 * Any inodes which are pinned purely because of attached pagecache have their
688 * pagecache removed. If the inode has metadata buffers attached to
689 * mapping->private_list then try to remove them.
691 * If the inode has the I_REFERENCED flag set, then it means that it has been
692 * used recently - the flag is set in iput_final(). When we encounter such an
693 * inode, clear the flag and move it to the back of the LRU so it gets another
694 * pass through the LRU before it gets reclaimed. This is necessary because of
695 * the fact we are doing lazy LRU updates to minimise lock contention so the
696 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
697 * with this flag set because they are the inodes that are out of order.
699 static enum lru_status
inode_lru_isolate(struct list_head
*item
,
700 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
702 struct list_head
*freeable
= arg
;
703 struct inode
*inode
= container_of(item
, struct inode
, i_lru
);
706 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
707 * If we fail to get the lock, just skip it.
709 if (!spin_trylock(&inode
->i_lock
))
713 * Referenced or dirty inodes are still in use. Give them another pass
714 * through the LRU as we canot reclaim them now.
716 if (atomic_read(&inode
->i_count
) ||
717 (inode
->i_state
& ~I_REFERENCED
)) {
718 list_lru_isolate(lru
, &inode
->i_lru
);
719 spin_unlock(&inode
->i_lock
);
720 this_cpu_dec(nr_unused
);
724 /* recently referenced inodes get one more pass */
725 if (inode
->i_state
& I_REFERENCED
) {
726 inode
->i_state
&= ~I_REFERENCED
;
727 spin_unlock(&inode
->i_lock
);
731 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
733 spin_unlock(&inode
->i_lock
);
734 spin_unlock(lru_lock
);
735 if (remove_inode_buffers(inode
)) {
737 reap
= invalidate_mapping_pages(&inode
->i_data
, 0, -1);
738 if (current_is_kswapd())
739 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
741 __count_vm_events(PGINODESTEAL
, reap
);
742 if (current
->reclaim_state
)
743 current
->reclaim_state
->reclaimed_slab
+= reap
;
750 WARN_ON(inode
->i_state
& I_NEW
);
751 inode
->i_state
|= I_FREEING
;
752 list_lru_isolate_move(lru
, &inode
->i_lru
, freeable
);
753 spin_unlock(&inode
->i_lock
);
755 this_cpu_dec(nr_unused
);
760 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
761 * This is called from the superblock shrinker function with a number of inodes
762 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
763 * then are freed outside inode_lock by dispose_list().
765 long prune_icache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
770 freed
= list_lru_shrink_walk(&sb
->s_inode_lru
, sc
,
771 inode_lru_isolate
, &freeable
);
772 dispose_list(&freeable
);
776 static void __wait_on_freeing_inode(struct inode
*inode
);
778 * Called with the inode lock held.
780 static struct inode
*find_inode(struct super_block
*sb
,
781 struct hlist_head
*head
,
782 int (*test
)(struct inode
*, void *),
785 struct inode
*inode
= NULL
;
788 hlist_for_each_entry(inode
, head
, i_hash
) {
789 if (inode
->i_sb
!= sb
)
791 if (!test(inode
, data
))
793 spin_lock(&inode
->i_lock
);
794 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
795 __wait_on_freeing_inode(inode
);
799 spin_unlock(&inode
->i_lock
);
806 * find_inode_fast is the fast path version of find_inode, see the comment at
807 * iget_locked for details.
809 static struct inode
*find_inode_fast(struct super_block
*sb
,
810 struct hlist_head
*head
, unsigned long ino
)
812 struct inode
*inode
= NULL
;
815 hlist_for_each_entry(inode
, head
, i_hash
) {
816 if (inode
->i_ino
!= ino
)
818 if (inode
->i_sb
!= sb
)
820 spin_lock(&inode
->i_lock
);
821 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
822 __wait_on_freeing_inode(inode
);
826 spin_unlock(&inode
->i_lock
);
833 * Each cpu owns a range of LAST_INO_BATCH numbers.
834 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
835 * to renew the exhausted range.
837 * This does not significantly increase overflow rate because every CPU can
838 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
839 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
840 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
841 * overflow rate by 2x, which does not seem too significant.
843 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
844 * error if st_ino won't fit in target struct field. Use 32bit counter
845 * here to attempt to avoid that.
847 #define LAST_INO_BATCH 1024
848 static DEFINE_PER_CPU(unsigned int, last_ino
);
850 unsigned int get_next_ino(void)
852 unsigned int *p
= &get_cpu_var(last_ino
);
853 unsigned int res
= *p
;
856 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
857 static atomic_t shared_last_ino
;
858 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
860 res
= next
- LAST_INO_BATCH
;
865 /* get_next_ino should not provide a 0 inode number */
869 put_cpu_var(last_ino
);
872 EXPORT_SYMBOL(get_next_ino
);
875 * new_inode_pseudo - obtain an inode
878 * Allocates a new inode for given superblock.
879 * Inode wont be chained in superblock s_inodes list
881 * - fs can't be unmount
882 * - quotas, fsnotify, writeback can't work
884 struct inode
*new_inode_pseudo(struct super_block
*sb
)
886 struct inode
*inode
= alloc_inode(sb
);
889 spin_lock(&inode
->i_lock
);
891 spin_unlock(&inode
->i_lock
);
892 INIT_LIST_HEAD(&inode
->i_sb_list
);
898 * new_inode - obtain an inode
901 * Allocates a new inode for given superblock. The default gfp_mask
902 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
903 * If HIGHMEM pages are unsuitable or it is known that pages allocated
904 * for the page cache are not reclaimable or migratable,
905 * mapping_set_gfp_mask() must be called with suitable flags on the
906 * newly created inode's mapping
909 struct inode
*new_inode(struct super_block
*sb
)
913 spin_lock_prefetch(&sb
->s_inode_list_lock
);
915 inode
= new_inode_pseudo(sb
);
917 inode_sb_list_add(inode
);
920 EXPORT_SYMBOL(new_inode
);
922 #ifdef CONFIG_DEBUG_LOCK_ALLOC
923 void lockdep_annotate_inode_mutex_key(struct inode
*inode
)
925 if (S_ISDIR(inode
->i_mode
)) {
926 struct file_system_type
*type
= inode
->i_sb
->s_type
;
928 /* Set new key only if filesystem hasn't already changed it */
929 if (lockdep_match_class(&inode
->i_rwsem
, &type
->i_mutex_key
)) {
931 * ensure nobody is actually holding i_mutex
933 // mutex_destroy(&inode->i_mutex);
934 init_rwsem(&inode
->i_rwsem
);
935 lockdep_set_class(&inode
->i_rwsem
,
936 &type
->i_mutex_dir_key
);
940 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key
);
944 * unlock_new_inode - clear the I_NEW state and wake up any waiters
945 * @inode: new inode to unlock
947 * Called when the inode is fully initialised to clear the new state of the
948 * inode and wake up anyone waiting for the inode to finish initialisation.
950 void unlock_new_inode(struct inode
*inode
)
952 lockdep_annotate_inode_mutex_key(inode
);
953 spin_lock(&inode
->i_lock
);
954 WARN_ON(!(inode
->i_state
& I_NEW
));
955 inode
->i_state
&= ~I_NEW
;
957 wake_up_bit(&inode
->i_state
, __I_NEW
);
958 spin_unlock(&inode
->i_lock
);
960 EXPORT_SYMBOL(unlock_new_inode
);
963 * lock_two_nondirectories - take two i_mutexes on non-directory objects
965 * Lock any non-NULL argument that is not a directory.
966 * Zero, one or two objects may be locked by this function.
968 * @inode1: first inode to lock
969 * @inode2: second inode to lock
971 void lock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
974 swap(inode1
, inode2
);
976 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
978 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
979 inode_lock_nested(inode2
, I_MUTEX_NONDIR2
);
981 EXPORT_SYMBOL(lock_two_nondirectories
);
984 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
985 * @inode1: first inode to unlock
986 * @inode2: second inode to unlock
988 void unlock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
990 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
991 inode_unlock(inode1
);
992 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
993 inode_unlock(inode2
);
995 EXPORT_SYMBOL(unlock_two_nondirectories
);
998 * iget5_locked - obtain an inode from a mounted file system
999 * @sb: super block of file system
1000 * @hashval: hash value (usually inode number) to get
1001 * @test: callback used for comparisons between inodes
1002 * @set: callback used to initialize a new struct inode
1003 * @data: opaque data pointer to pass to @test and @set
1005 * Search for the inode specified by @hashval and @data in the inode cache,
1006 * and if present it is return it with an increased reference count. This is
1007 * a generalized version of iget_locked() for file systems where the inode
1008 * number is not sufficient for unique identification of an inode.
1010 * If the inode is not in cache, allocate a new inode and return it locked,
1011 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1012 * before unlocking it via unlock_new_inode().
1014 * Note both @test and @set are called with the inode_hash_lock held, so can't
1017 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1018 int (*test
)(struct inode
*, void *),
1019 int (*set
)(struct inode
*, void *), void *data
)
1021 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1022 struct inode
*inode
;
1024 spin_lock(&inode_hash_lock
);
1025 inode
= find_inode(sb
, head
, test
, data
);
1026 spin_unlock(&inode_hash_lock
);
1029 wait_on_inode(inode
);
1030 if (unlikely(inode_unhashed(inode
))) {
1037 inode
= alloc_inode(sb
);
1041 spin_lock(&inode_hash_lock
);
1042 /* We released the lock, so.. */
1043 old
= find_inode(sb
, head
, test
, data
);
1045 if (set(inode
, data
))
1048 spin_lock(&inode
->i_lock
);
1049 inode
->i_state
= I_NEW
;
1050 hlist_add_head(&inode
->i_hash
, head
);
1051 spin_unlock(&inode
->i_lock
);
1052 inode_sb_list_add(inode
);
1053 spin_unlock(&inode_hash_lock
);
1055 /* Return the locked inode with I_NEW set, the
1056 * caller is responsible for filling in the contents
1062 * Uhhuh, somebody else created the same inode under
1063 * us. Use the old inode instead of the one we just
1066 spin_unlock(&inode_hash_lock
);
1067 destroy_inode(inode
);
1069 wait_on_inode(inode
);
1070 if (unlikely(inode_unhashed(inode
))) {
1078 spin_unlock(&inode_hash_lock
);
1079 destroy_inode(inode
);
1082 EXPORT_SYMBOL(iget5_locked
);
1085 * iget_locked - obtain an inode from a mounted file system
1086 * @sb: super block of file system
1087 * @ino: inode number to get
1089 * Search for the inode specified by @ino in the inode cache and if present
1090 * return it with an increased reference count. This is for file systems
1091 * where the inode number is sufficient for unique identification of an inode.
1093 * If the inode is not in cache, allocate a new inode and return it locked,
1094 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1095 * before unlocking it via unlock_new_inode().
1097 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1099 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1100 struct inode
*inode
;
1102 spin_lock(&inode_hash_lock
);
1103 inode
= find_inode_fast(sb
, head
, ino
);
1104 spin_unlock(&inode_hash_lock
);
1106 wait_on_inode(inode
);
1107 if (unlikely(inode_unhashed(inode
))) {
1114 inode
= alloc_inode(sb
);
1118 spin_lock(&inode_hash_lock
);
1119 /* We released the lock, so.. */
1120 old
= find_inode_fast(sb
, head
, ino
);
1123 spin_lock(&inode
->i_lock
);
1124 inode
->i_state
= I_NEW
;
1125 hlist_add_head(&inode
->i_hash
, head
);
1126 spin_unlock(&inode
->i_lock
);
1127 inode_sb_list_add(inode
);
1128 spin_unlock(&inode_hash_lock
);
1130 /* Return the locked inode with I_NEW set, the
1131 * caller is responsible for filling in the contents
1137 * Uhhuh, somebody else created the same inode under
1138 * us. Use the old inode instead of the one we just
1141 spin_unlock(&inode_hash_lock
);
1142 destroy_inode(inode
);
1144 wait_on_inode(inode
);
1145 if (unlikely(inode_unhashed(inode
))) {
1152 EXPORT_SYMBOL(iget_locked
);
1155 * search the inode cache for a matching inode number.
1156 * If we find one, then the inode number we are trying to
1157 * allocate is not unique and so we should not use it.
1159 * Returns 1 if the inode number is unique, 0 if it is not.
1161 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1163 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1164 struct inode
*inode
;
1166 spin_lock(&inode_hash_lock
);
1167 hlist_for_each_entry(inode
, b
, i_hash
) {
1168 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
) {
1169 spin_unlock(&inode_hash_lock
);
1173 spin_unlock(&inode_hash_lock
);
1179 * iunique - get a unique inode number
1181 * @max_reserved: highest reserved inode number
1183 * Obtain an inode number that is unique on the system for a given
1184 * superblock. This is used by file systems that have no natural
1185 * permanent inode numbering system. An inode number is returned that
1186 * is higher than the reserved limit but unique.
1189 * With a large number of inodes live on the file system this function
1190 * currently becomes quite slow.
1192 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1195 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1196 * error if st_ino won't fit in target struct field. Use 32bit counter
1197 * here to attempt to avoid that.
1199 static DEFINE_SPINLOCK(iunique_lock
);
1200 static unsigned int counter
;
1203 spin_lock(&iunique_lock
);
1205 if (counter
<= max_reserved
)
1206 counter
= max_reserved
+ 1;
1208 } while (!test_inode_iunique(sb
, res
));
1209 spin_unlock(&iunique_lock
);
1213 EXPORT_SYMBOL(iunique
);
1215 struct inode
*igrab(struct inode
*inode
)
1217 spin_lock(&inode
->i_lock
);
1218 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1220 spin_unlock(&inode
->i_lock
);
1222 spin_unlock(&inode
->i_lock
);
1224 * Handle the case where s_op->clear_inode is not been
1225 * called yet, and somebody is calling igrab
1226 * while the inode is getting freed.
1232 EXPORT_SYMBOL(igrab
);
1235 * ilookup5_nowait - search for an inode in the inode cache
1236 * @sb: super block of file system to search
1237 * @hashval: hash value (usually inode number) to search for
1238 * @test: callback used for comparisons between inodes
1239 * @data: opaque data pointer to pass to @test
1241 * Search for the inode specified by @hashval and @data in the inode cache.
1242 * If the inode is in the cache, the inode is returned with an incremented
1245 * Note: I_NEW is not waited upon so you have to be very careful what you do
1246 * with the returned inode. You probably should be using ilookup5() instead.
1248 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1250 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1251 int (*test
)(struct inode
*, void *), void *data
)
1253 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1254 struct inode
*inode
;
1256 spin_lock(&inode_hash_lock
);
1257 inode
= find_inode(sb
, head
, test
, data
);
1258 spin_unlock(&inode_hash_lock
);
1262 EXPORT_SYMBOL(ilookup5_nowait
);
1265 * ilookup5 - 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 * and if the inode is in the cache, return the inode with an incremented
1273 * reference count. Waits on I_NEW before returning the inode.
1274 * returned with an incremented reference count.
1276 * This is a generalized version of ilookup() for file systems where the
1277 * inode number is not sufficient for unique identification of an inode.
1279 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1281 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1282 int (*test
)(struct inode
*, void *), void *data
)
1284 struct inode
*inode
;
1286 inode
= ilookup5_nowait(sb
, hashval
, test
, data
);
1288 wait_on_inode(inode
);
1289 if (unlikely(inode_unhashed(inode
))) {
1296 EXPORT_SYMBOL(ilookup5
);
1299 * ilookup - search for an inode in the inode cache
1300 * @sb: super block of file system to search
1301 * @ino: inode number to search for
1303 * Search for the inode @ino in the inode cache, and if the inode is in the
1304 * cache, the inode is returned with an incremented reference count.
1306 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1308 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1309 struct inode
*inode
;
1311 spin_lock(&inode_hash_lock
);
1312 inode
= find_inode_fast(sb
, head
, ino
);
1313 spin_unlock(&inode_hash_lock
);
1316 wait_on_inode(inode
);
1317 if (unlikely(inode_unhashed(inode
))) {
1324 EXPORT_SYMBOL(ilookup
);
1327 * find_inode_nowait - find an inode in the inode cache
1328 * @sb: super block of file system to search
1329 * @hashval: hash value (usually inode number) to search for
1330 * @match: callback used for comparisons between inodes
1331 * @data: opaque data pointer to pass to @match
1333 * Search for the inode specified by @hashval and @data in the inode
1334 * cache, where the helper function @match will return 0 if the inode
1335 * does not match, 1 if the inode does match, and -1 if the search
1336 * should be stopped. The @match function must be responsible for
1337 * taking the i_lock spin_lock and checking i_state for an inode being
1338 * freed or being initialized, and incrementing the reference count
1339 * before returning 1. It also must not sleep, since it is called with
1340 * the inode_hash_lock spinlock held.
1342 * This is a even more generalized version of ilookup5() when the
1343 * function must never block --- find_inode() can block in
1344 * __wait_on_freeing_inode() --- or when the caller can not increment
1345 * the reference count because the resulting iput() might cause an
1346 * inode eviction. The tradeoff is that the @match funtion must be
1347 * very carefully implemented.
1349 struct inode
*find_inode_nowait(struct super_block
*sb
,
1350 unsigned long hashval
,
1351 int (*match
)(struct inode
*, unsigned long,
1355 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1356 struct inode
*inode
, *ret_inode
= NULL
;
1359 spin_lock(&inode_hash_lock
);
1360 hlist_for_each_entry(inode
, head
, i_hash
) {
1361 if (inode
->i_sb
!= sb
)
1363 mval
= match(inode
, hashval
, data
);
1371 spin_unlock(&inode_hash_lock
);
1374 EXPORT_SYMBOL(find_inode_nowait
);
1376 int insert_inode_locked(struct inode
*inode
)
1378 struct super_block
*sb
= inode
->i_sb
;
1379 ino_t ino
= inode
->i_ino
;
1380 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1383 struct inode
*old
= NULL
;
1384 spin_lock(&inode_hash_lock
);
1385 hlist_for_each_entry(old
, head
, i_hash
) {
1386 if (old
->i_ino
!= ino
)
1388 if (old
->i_sb
!= sb
)
1390 spin_lock(&old
->i_lock
);
1391 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1392 spin_unlock(&old
->i_lock
);
1398 spin_lock(&inode
->i_lock
);
1399 inode
->i_state
|= I_NEW
;
1400 hlist_add_head(&inode
->i_hash
, head
);
1401 spin_unlock(&inode
->i_lock
);
1402 spin_unlock(&inode_hash_lock
);
1406 spin_unlock(&old
->i_lock
);
1407 spin_unlock(&inode_hash_lock
);
1409 if (unlikely(!inode_unhashed(old
))) {
1416 EXPORT_SYMBOL(insert_inode_locked
);
1418 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1419 int (*test
)(struct inode
*, void *), void *data
)
1421 struct super_block
*sb
= inode
->i_sb
;
1422 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1425 struct inode
*old
= NULL
;
1427 spin_lock(&inode_hash_lock
);
1428 hlist_for_each_entry(old
, head
, i_hash
) {
1429 if (old
->i_sb
!= sb
)
1431 if (!test(old
, data
))
1433 spin_lock(&old
->i_lock
);
1434 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1435 spin_unlock(&old
->i_lock
);
1441 spin_lock(&inode
->i_lock
);
1442 inode
->i_state
|= I_NEW
;
1443 hlist_add_head(&inode
->i_hash
, head
);
1444 spin_unlock(&inode
->i_lock
);
1445 spin_unlock(&inode_hash_lock
);
1449 spin_unlock(&old
->i_lock
);
1450 spin_unlock(&inode_hash_lock
);
1452 if (unlikely(!inode_unhashed(old
))) {
1459 EXPORT_SYMBOL(insert_inode_locked4
);
1462 int generic_delete_inode(struct inode
*inode
)
1466 EXPORT_SYMBOL(generic_delete_inode
);
1469 * Called when we're dropping the last reference
1472 * Call the FS "drop_inode()" function, defaulting to
1473 * the legacy UNIX filesystem behaviour. If it tells
1474 * us to evict inode, do so. Otherwise, retain inode
1475 * in cache if fs is alive, sync and evict if fs is
1478 static void iput_final(struct inode
*inode
)
1480 struct super_block
*sb
= inode
->i_sb
;
1481 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1484 WARN_ON(inode
->i_state
& I_NEW
);
1487 drop
= op
->drop_inode(inode
);
1489 drop
= generic_drop_inode(inode
);
1491 if (!drop
&& (sb
->s_flags
& MS_ACTIVE
)) {
1492 inode
->i_state
|= I_REFERENCED
;
1493 inode_add_lru(inode
);
1494 spin_unlock(&inode
->i_lock
);
1499 inode
->i_state
|= I_WILL_FREE
;
1500 spin_unlock(&inode
->i_lock
);
1501 write_inode_now(inode
, 1);
1502 spin_lock(&inode
->i_lock
);
1503 WARN_ON(inode
->i_state
& I_NEW
);
1504 inode
->i_state
&= ~I_WILL_FREE
;
1507 inode
->i_state
|= I_FREEING
;
1508 if (!list_empty(&inode
->i_lru
))
1509 inode_lru_list_del(inode
);
1510 spin_unlock(&inode
->i_lock
);
1516 * iput - put an inode
1517 * @inode: inode to put
1519 * Puts an inode, dropping its usage count. If the inode use count hits
1520 * zero, the inode is then freed and may also be destroyed.
1522 * Consequently, iput() can sleep.
1524 void iput(struct inode
*inode
)
1528 BUG_ON(inode
->i_state
& I_CLEAR
);
1530 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
)) {
1531 if (inode
->i_nlink
&& (inode
->i_state
& I_DIRTY_TIME
)) {
1532 atomic_inc(&inode
->i_count
);
1533 inode
->i_state
&= ~I_DIRTY_TIME
;
1534 spin_unlock(&inode
->i_lock
);
1535 trace_writeback_lazytime_iput(inode
);
1536 mark_inode_dirty_sync(inode
);
1542 EXPORT_SYMBOL(iput
);
1545 * bmap - find a block number in a file
1546 * @inode: inode of file
1547 * @block: block to find
1549 * Returns the block number on the device holding the inode that
1550 * is the disk block number for the block of the file requested.
1551 * That is, asked for block 4 of inode 1 the function will return the
1552 * disk block relative to the disk start that holds that block of the
1555 sector_t
bmap(struct inode
*inode
, sector_t block
)
1558 if (inode
->i_mapping
->a_ops
->bmap
)
1559 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1562 EXPORT_SYMBOL(bmap
);
1565 * Update times in overlayed inode from underlying real inode
1567 static void update_ovl_inode_times(struct dentry
*dentry
, struct inode
*inode
,
1571 struct inode
*realinode
= d_real_inode(dentry
);
1573 if (unlikely(inode
!= realinode
) &&
1574 (!timespec_equal(&inode
->i_mtime
, &realinode
->i_mtime
) ||
1575 !timespec_equal(&inode
->i_ctime
, &realinode
->i_ctime
))) {
1576 inode
->i_mtime
= realinode
->i_mtime
;
1577 inode
->i_ctime
= realinode
->i_ctime
;
1583 * With relative atime, only update atime if the previous atime is
1584 * earlier than either the ctime or mtime or if at least a day has
1585 * passed since the last atime update.
1587 static int relatime_need_update(const struct path
*path
, struct inode
*inode
,
1588 struct timespec now
, bool rcu
)
1591 if (!(path
->mnt
->mnt_flags
& MNT_RELATIME
))
1594 update_ovl_inode_times(path
->dentry
, inode
, rcu
);
1596 * Is mtime younger than atime? If yes, update atime:
1598 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1601 * Is ctime younger than atime? If yes, update atime:
1603 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1607 * Is the previous atime value older than a day? If yes,
1610 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1613 * Good, we can skip the atime update:
1618 int generic_update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1620 int iflags
= I_DIRTY_TIME
;
1622 if (flags
& S_ATIME
)
1623 inode
->i_atime
= *time
;
1624 if (flags
& S_VERSION
)
1625 inode_inc_iversion(inode
);
1626 if (flags
& S_CTIME
)
1627 inode
->i_ctime
= *time
;
1628 if (flags
& S_MTIME
)
1629 inode
->i_mtime
= *time
;
1631 if (!(inode
->i_sb
->s_flags
& MS_LAZYTIME
) || (flags
& S_VERSION
))
1632 iflags
|= I_DIRTY_SYNC
;
1633 __mark_inode_dirty(inode
, iflags
);
1636 EXPORT_SYMBOL(generic_update_time
);
1639 * This does the actual work of updating an inodes time or version. Must have
1640 * had called mnt_want_write() before calling this.
1642 static int update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1644 int (*update_time
)(struct inode
*, struct timespec
*, int);
1646 update_time
= inode
->i_op
->update_time
? inode
->i_op
->update_time
:
1647 generic_update_time
;
1649 return update_time(inode
, time
, flags
);
1653 * touch_atime - update the access time
1654 * @path: the &struct path to update
1655 * @inode: inode to update
1657 * Update the accessed time on an inode and mark it for writeback.
1658 * This function automatically handles read only file systems and media,
1659 * as well as the "noatime" flag and inode specific "noatime" markers.
1661 bool __atime_needs_update(const struct path
*path
, struct inode
*inode
,
1664 struct vfsmount
*mnt
= path
->mnt
;
1665 struct timespec now
;
1667 if (inode
->i_flags
& S_NOATIME
)
1670 /* Atime updates will likely cause i_uid and i_gid to be written
1671 * back improprely if their true value is unknown to the vfs.
1673 if (HAS_UNMAPPED_ID(inode
))
1676 if (IS_NOATIME(inode
))
1678 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1681 if (mnt
->mnt_flags
& MNT_NOATIME
)
1683 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1686 now
= current_time(inode
);
1688 if (!relatime_need_update(path
, inode
, now
, rcu
))
1691 if (timespec_equal(&inode
->i_atime
, &now
))
1697 void touch_atime(const struct path
*path
)
1699 struct vfsmount
*mnt
= path
->mnt
;
1700 struct inode
*inode
= d_inode(path
->dentry
);
1701 struct timespec now
;
1703 if (!__atime_needs_update(path
, inode
, false))
1706 if (!sb_start_write_trylock(inode
->i_sb
))
1709 if (__mnt_want_write(mnt
) != 0)
1712 * File systems can error out when updating inodes if they need to
1713 * allocate new space to modify an inode (such is the case for
1714 * Btrfs), but since we touch atime while walking down the path we
1715 * really don't care if we failed to update the atime of the file,
1716 * so just ignore the return value.
1717 * We may also fail on filesystems that have the ability to make parts
1718 * of the fs read only, e.g. subvolumes in Btrfs.
1720 now
= current_time(inode
);
1721 update_time(inode
, &now
, S_ATIME
);
1722 __mnt_drop_write(mnt
);
1724 sb_end_write(inode
->i_sb
);
1726 EXPORT_SYMBOL(touch_atime
);
1729 * The logic we want is
1731 * if suid or (sgid and xgrp)
1734 int should_remove_suid(struct dentry
*dentry
)
1736 umode_t mode
= d_inode(dentry
)->i_mode
;
1739 /* suid always must be killed */
1740 if (unlikely(mode
& S_ISUID
))
1741 kill
= ATTR_KILL_SUID
;
1744 * sgid without any exec bits is just a mandatory locking mark; leave
1745 * it alone. If some exec bits are set, it's a real sgid; kill it.
1747 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1748 kill
|= ATTR_KILL_SGID
;
1750 if (unlikely(kill
&& !capable(CAP_FSETID
) && S_ISREG(mode
)))
1755 EXPORT_SYMBOL(should_remove_suid
);
1758 * Return mask of changes for notify_change() that need to be done as a
1759 * response to write or truncate. Return 0 if nothing has to be changed.
1760 * Negative value on error (change should be denied).
1762 int dentry_needs_remove_privs(struct dentry
*dentry
)
1764 struct inode
*inode
= d_inode(dentry
);
1768 if (IS_NOSEC(inode
))
1771 mask
= should_remove_suid(dentry
);
1772 ret
= security_inode_need_killpriv(dentry
);
1776 mask
|= ATTR_KILL_PRIV
;
1780 static int __remove_privs(struct dentry
*dentry
, int kill
)
1782 struct iattr newattrs
;
1784 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1786 * Note we call this on write, so notify_change will not
1787 * encounter any conflicting delegations:
1789 return notify_change(dentry
, &newattrs
, NULL
);
1793 * Remove special file priviledges (suid, capabilities) when file is written
1796 int file_remove_privs(struct file
*file
)
1798 struct dentry
*dentry
= file_dentry(file
);
1799 struct inode
*inode
= file_inode(file
);
1803 /* Fast path for nothing security related */
1804 if (IS_NOSEC(inode
))
1807 kill
= dentry_needs_remove_privs(dentry
);
1811 error
= __remove_privs(dentry
, kill
);
1813 inode_has_no_xattr(inode
);
1817 EXPORT_SYMBOL(file_remove_privs
);
1820 * file_update_time - update mtime and ctime time
1821 * @file: file accessed
1823 * Update the mtime and ctime members of an inode and mark the inode
1824 * for writeback. Note that this function is meant exclusively for
1825 * usage in the file write path of filesystems, and filesystems may
1826 * choose to explicitly ignore update via this function with the
1827 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1828 * timestamps are handled by the server. This can return an error for
1829 * file systems who need to allocate space in order to update an inode.
1832 int file_update_time(struct file
*file
)
1834 struct inode
*inode
= file_inode(file
);
1835 struct timespec now
;
1839 /* First try to exhaust all avenues to not sync */
1840 if (IS_NOCMTIME(inode
))
1843 now
= current_time(inode
);
1844 if (!timespec_equal(&inode
->i_mtime
, &now
))
1847 if (!timespec_equal(&inode
->i_ctime
, &now
))
1850 if (IS_I_VERSION(inode
))
1851 sync_it
|= S_VERSION
;
1856 /* Finally allowed to write? Takes lock. */
1857 if (__mnt_want_write_file(file
))
1860 ret
= update_time(inode
, &now
, sync_it
);
1861 __mnt_drop_write_file(file
);
1865 EXPORT_SYMBOL(file_update_time
);
1867 int inode_needs_sync(struct inode
*inode
)
1871 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1875 EXPORT_SYMBOL(inode_needs_sync
);
1878 * If we try to find an inode in the inode hash while it is being
1879 * deleted, we have to wait until the filesystem completes its
1880 * deletion before reporting that it isn't found. This function waits
1881 * until the deletion _might_ have completed. Callers are responsible
1882 * to recheck inode state.
1884 * It doesn't matter if I_NEW is not set initially, a call to
1885 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1888 static void __wait_on_freeing_inode(struct inode
*inode
)
1890 wait_queue_head_t
*wq
;
1891 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1892 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1893 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1894 spin_unlock(&inode
->i_lock
);
1895 spin_unlock(&inode_hash_lock
);
1897 finish_wait(wq
, &wait
.wait
);
1898 spin_lock(&inode_hash_lock
);
1901 static __initdata
unsigned long ihash_entries
;
1902 static int __init
set_ihash_entries(char *str
)
1906 ihash_entries
= simple_strtoul(str
, &str
, 0);
1909 __setup("ihash_entries=", set_ihash_entries
);
1912 * Initialize the waitqueues and inode hash table.
1914 void __init
inode_init_early(void)
1918 /* If hashes are distributed across NUMA nodes, defer
1919 * hash allocation until vmalloc space is available.
1925 alloc_large_system_hash("Inode-cache",
1926 sizeof(struct hlist_head
),
1935 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1936 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1939 void __init
inode_init(void)
1943 /* inode slab cache */
1944 inode_cachep
= kmem_cache_create("inode_cache",
1945 sizeof(struct inode
),
1947 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1948 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
),
1951 /* Hash may have been set up in inode_init_early */
1956 alloc_large_system_hash("Inode-cache",
1957 sizeof(struct hlist_head
),
1966 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1967 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1970 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1972 inode
->i_mode
= mode
;
1973 if (S_ISCHR(mode
)) {
1974 inode
->i_fop
= &def_chr_fops
;
1975 inode
->i_rdev
= rdev
;
1976 } else if (S_ISBLK(mode
)) {
1977 inode
->i_fop
= &def_blk_fops
;
1978 inode
->i_rdev
= rdev
;
1979 } else if (S_ISFIFO(mode
))
1980 inode
->i_fop
= &pipefifo_fops
;
1981 else if (S_ISSOCK(mode
))
1982 ; /* leave it no_open_fops */
1984 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1985 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1988 EXPORT_SYMBOL(init_special_inode
);
1991 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1993 * @dir: Directory inode
1994 * @mode: mode of the new inode
1996 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1999 inode
->i_uid
= current_fsuid();
2000 if (dir
&& dir
->i_mode
& S_ISGID
) {
2001 inode
->i_gid
= dir
->i_gid
;
2005 inode
->i_gid
= current_fsgid();
2006 inode
->i_mode
= mode
;
2008 EXPORT_SYMBOL(inode_init_owner
);
2011 * inode_owner_or_capable - check current task permissions to inode
2012 * @inode: inode being checked
2014 * Return true if current either has CAP_FOWNER in a namespace with the
2015 * inode owner uid mapped, or owns the file.
2017 bool inode_owner_or_capable(const struct inode
*inode
)
2019 struct user_namespace
*ns
;
2021 if (uid_eq(current_fsuid(), inode
->i_uid
))
2024 ns
= current_user_ns();
2025 if (ns_capable(ns
, CAP_FOWNER
) && kuid_has_mapping(ns
, inode
->i_uid
))
2029 EXPORT_SYMBOL(inode_owner_or_capable
);
2032 * Direct i/o helper functions
2034 static void __inode_dio_wait(struct inode
*inode
)
2036 wait_queue_head_t
*wq
= bit_waitqueue(&inode
->i_state
, __I_DIO_WAKEUP
);
2037 DEFINE_WAIT_BIT(q
, &inode
->i_state
, __I_DIO_WAKEUP
);
2040 prepare_to_wait(wq
, &q
.wait
, TASK_UNINTERRUPTIBLE
);
2041 if (atomic_read(&inode
->i_dio_count
))
2043 } while (atomic_read(&inode
->i_dio_count
));
2044 finish_wait(wq
, &q
.wait
);
2048 * inode_dio_wait - wait for outstanding DIO requests to finish
2049 * @inode: inode to wait for
2051 * Waits for all pending direct I/O requests to finish so that we can
2052 * proceed with a truncate or equivalent operation.
2054 * Must be called under a lock that serializes taking new references
2055 * to i_dio_count, usually by inode->i_mutex.
2057 void inode_dio_wait(struct inode
*inode
)
2059 if (atomic_read(&inode
->i_dio_count
))
2060 __inode_dio_wait(inode
);
2062 EXPORT_SYMBOL(inode_dio_wait
);
2065 * inode_set_flags - atomically set some inode flags
2067 * Note: the caller should be holding i_mutex, or else be sure that
2068 * they have exclusive access to the inode structure (i.e., while the
2069 * inode is being instantiated). The reason for the cmpxchg() loop
2070 * --- which wouldn't be necessary if all code paths which modify
2071 * i_flags actually followed this rule, is that there is at least one
2072 * code path which doesn't today so we use cmpxchg() out of an abundance
2075 * In the long run, i_mutex is overkill, and we should probably look
2076 * at using the i_lock spinlock to protect i_flags, and then make sure
2077 * it is so documented in include/linux/fs.h and that all code follows
2078 * the locking convention!!
2080 void inode_set_flags(struct inode
*inode
, unsigned int flags
,
2083 unsigned int old_flags
, new_flags
;
2085 WARN_ON_ONCE(flags
& ~mask
);
2087 old_flags
= ACCESS_ONCE(inode
->i_flags
);
2088 new_flags
= (old_flags
& ~mask
) | flags
;
2089 } while (unlikely(cmpxchg(&inode
->i_flags
, old_flags
,
2090 new_flags
) != old_flags
));
2092 EXPORT_SYMBOL(inode_set_flags
);
2094 void inode_nohighmem(struct inode
*inode
)
2096 mapping_set_gfp_mask(inode
->i_mapping
, GFP_USER
);
2098 EXPORT_SYMBOL(inode_nohighmem
);
2101 * current_time - Return FS time
2104 * Return the current time truncated to the time granularity supported by
2107 * Note that inode and inode->sb cannot be NULL.
2108 * Otherwise, the function warns and returns time without truncation.
2110 struct timespec
current_time(struct inode
*inode
)
2112 struct timespec now
= current_kernel_time();
2114 if (unlikely(!inode
->i_sb
)) {
2115 WARN(1, "current_time() called with uninitialized super_block in the inode");
2119 return timespec_trunc(now
, inode
->i_sb
->s_time_gran
);
2121 EXPORT_SYMBOL(current_time
);