4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/quotaops.h>
12 #include <linux/slab.h>
13 #include <linux/writeback.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/wait.h>
17 #include <linux/rwsem.h>
18 #include <linux/hash.h>
19 #include <linux/swap.h>
20 #include <linux/security.h>
21 #include <linux/ima.h>
22 #include <linux/pagemap.h>
23 #include <linux/cdev.h>
24 #include <linux/bootmem.h>
25 #include <linux/inotify.h>
26 #include <linux/fsnotify.h>
27 #include <linux/mount.h>
28 #include <linux/async.h>
29 #include <linux/posix_acl.h>
32 * This is needed for the following functions:
34 * - invalidate_inode_buffers
37 * FIXME: remove all knowledge of the buffer layer from this file
39 #include <linux/buffer_head.h>
42 * New inode.c implementation.
44 * This implementation has the basic premise of trying
45 * to be extremely low-overhead and SMP-safe, yet be
46 * simple enough to be "obviously correct".
51 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
53 /* #define INODE_PARANOIA 1 */
54 /* #define INODE_DEBUG 1 */
57 * Inode lookup is no longer as critical as it used to be:
58 * most of the lookups are going to be through the dcache.
60 #define I_HASHBITS i_hash_shift
61 #define I_HASHMASK i_hash_mask
63 static unsigned int i_hash_mask __read_mostly
;
64 static unsigned int i_hash_shift __read_mostly
;
67 * Each inode can be on two separate lists. One is
68 * the hash list of the inode, used for lookups. The
69 * other linked list is the "type" list:
70 * "in_use" - valid inode, i_count > 0, i_nlink > 0
71 * "dirty" - as "in_use" but also dirty
72 * "unused" - valid inode, i_count = 0
74 * A "dirty" list is maintained for each super block,
75 * allowing for low-overhead inode sync() operations.
78 LIST_HEAD(inode_in_use
);
79 LIST_HEAD(inode_unused
);
80 static struct hlist_head
*inode_hashtable __read_mostly
;
83 * A simple spinlock to protect the list manipulations.
85 * NOTE! You also have to own the lock if you change
86 * the i_state of an inode while it is in use..
88 DEFINE_SPINLOCK(inode_lock
);
91 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
92 * icache shrinking path, and the umount path. Without this exclusion,
93 * by the time prune_icache calls iput for the inode whose pages it has
94 * been invalidating, or by the time it calls clear_inode & destroy_inode
95 * from its final dispose_list, the struct super_block they refer to
96 * (for inode->i_sb->s_op) may already have been freed and reused.
98 * We make this an rwsem because the fastpath is icache shrinking. In
99 * some cases a filesystem may be doing a significant amount of work in
100 * its inode reclaim code, so this should improve parallelism.
102 static DECLARE_RWSEM(iprune_sem
);
105 * Statistics gathering..
107 struct inodes_stat_t inodes_stat
;
109 static struct kmem_cache
*inode_cachep __read_mostly
;
111 static void wake_up_inode(struct inode
*inode
)
114 * Prevent speculative execution through spin_unlock(&inode_lock);
117 wake_up_bit(&inode
->i_state
, __I_LOCK
);
121 * inode_init_always - perform inode structure intialisation
122 * @sb: superblock inode belongs to
123 * @inode: inode to initialise
125 * These are initializations that need to be done on every inode
126 * allocation as the fields are not initialised by slab allocation.
128 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
130 static const struct address_space_operations empty_aops
;
131 static const struct inode_operations empty_iops
;
132 static const struct file_operations empty_fops
;
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
= &empty_fops
;
144 atomic_set(&inode
->i_writecount
, 0);
148 inode
->i_generation
= 0;
150 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
152 inode
->i_pipe
= NULL
;
153 inode
->i_bdev
= NULL
;
154 inode
->i_cdev
= NULL
;
156 inode
->dirtied_when
= 0;
158 if (security_inode_alloc(inode
))
161 /* allocate and initialize an i_integrity */
162 if (ima_inode_alloc(inode
))
163 goto out_free_security
;
165 spin_lock_init(&inode
->i_lock
);
166 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
168 mutex_init(&inode
->i_mutex
);
169 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
171 init_rwsem(&inode
->i_alloc_sem
);
172 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
174 mapping
->a_ops
= &empty_aops
;
175 mapping
->host
= inode
;
177 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
178 mapping
->assoc_mapping
= NULL
;
179 mapping
->backing_dev_info
= &default_backing_dev_info
;
180 mapping
->writeback_index
= 0;
183 * If the block_device provides a backing_dev_info for client
184 * inodes then use that. Otherwise the inode share the bdev's
188 struct backing_dev_info
*bdi
;
190 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
191 mapping
->backing_dev_info
= bdi
;
193 inode
->i_private
= NULL
;
194 inode
->i_mapping
= mapping
;
195 #ifdef CONFIG_FS_POSIX_ACL
196 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
199 #ifdef CONFIG_FSNOTIFY
200 inode
->i_fsnotify_mask
= 0;
206 security_inode_free(inode
);
210 EXPORT_SYMBOL(inode_init_always
);
212 static struct inode
*alloc_inode(struct super_block
*sb
)
216 if (sb
->s_op
->alloc_inode
)
217 inode
= sb
->s_op
->alloc_inode(sb
);
219 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
224 if (unlikely(inode_init_always(sb
, inode
))) {
225 if (inode
->i_sb
->s_op
->destroy_inode
)
226 inode
->i_sb
->s_op
->destroy_inode(inode
);
228 kmem_cache_free(inode_cachep
, inode
);
235 void __destroy_inode(struct inode
*inode
)
237 BUG_ON(inode_has_buffers(inode
));
238 ima_inode_free(inode
);
239 security_inode_free(inode
);
240 fsnotify_inode_delete(inode
);
241 #ifdef CONFIG_FS_POSIX_ACL
242 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
243 posix_acl_release(inode
->i_acl
);
244 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
245 posix_acl_release(inode
->i_default_acl
);
248 EXPORT_SYMBOL(__destroy_inode
);
250 void destroy_inode(struct inode
*inode
)
252 __destroy_inode(inode
);
253 if (inode
->i_sb
->s_op
->destroy_inode
)
254 inode
->i_sb
->s_op
->destroy_inode(inode
);
256 kmem_cache_free(inode_cachep
, (inode
));
260 * These are initializations that only need to be done
261 * once, because the fields are idempotent across use
262 * of the inode, so let the slab aware of that.
264 void inode_init_once(struct inode
*inode
)
266 memset(inode
, 0, sizeof(*inode
));
267 INIT_HLIST_NODE(&inode
->i_hash
);
268 INIT_LIST_HEAD(&inode
->i_dentry
);
269 INIT_LIST_HEAD(&inode
->i_devices
);
270 INIT_RADIX_TREE(&inode
->i_data
.page_tree
, GFP_ATOMIC
);
271 spin_lock_init(&inode
->i_data
.tree_lock
);
272 spin_lock_init(&inode
->i_data
.i_mmap_lock
);
273 INIT_LIST_HEAD(&inode
->i_data
.private_list
);
274 spin_lock_init(&inode
->i_data
.private_lock
);
275 INIT_RAW_PRIO_TREE_ROOT(&inode
->i_data
.i_mmap
);
276 INIT_LIST_HEAD(&inode
->i_data
.i_mmap_nonlinear
);
277 i_size_ordered_init(inode
);
278 #ifdef CONFIG_INOTIFY
279 INIT_LIST_HEAD(&inode
->inotify_watches
);
280 mutex_init(&inode
->inotify_mutex
);
282 #ifdef CONFIG_FSNOTIFY
283 INIT_HLIST_HEAD(&inode
->i_fsnotify_mark_entries
);
286 EXPORT_SYMBOL(inode_init_once
);
288 static void init_once(void *foo
)
290 struct inode
*inode
= (struct inode
*) foo
;
292 inode_init_once(inode
);
296 * inode_lock must be held
298 void __iget(struct inode
*inode
)
300 if (atomic_read(&inode
->i_count
)) {
301 atomic_inc(&inode
->i_count
);
304 atomic_inc(&inode
->i_count
);
305 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
306 list_move(&inode
->i_list
, &inode_in_use
);
307 inodes_stat
.nr_unused
--;
311 * clear_inode - clear an inode
312 * @inode: inode to clear
314 * This is called by the filesystem to tell us
315 * that the inode is no longer useful. We just
316 * terminate it with extreme prejudice.
318 void clear_inode(struct inode
*inode
)
321 invalidate_inode_buffers(inode
);
323 BUG_ON(inode
->i_data
.nrpages
);
324 BUG_ON(!(inode
->i_state
& I_FREEING
));
325 BUG_ON(inode
->i_state
& I_CLEAR
);
326 inode_sync_wait(inode
);
328 if (inode
->i_sb
->s_op
->clear_inode
)
329 inode
->i_sb
->s_op
->clear_inode(inode
);
330 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
332 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
334 inode
->i_state
= I_CLEAR
;
336 EXPORT_SYMBOL(clear_inode
);
339 * dispose_list - dispose of the contents of a local list
340 * @head: the head of the list to free
342 * Dispose-list gets a local list with local inodes in it, so it doesn't
343 * need to worry about list corruption and SMP locks.
345 static void dispose_list(struct list_head
*head
)
349 while (!list_empty(head
)) {
352 inode
= list_first_entry(head
, struct inode
, i_list
);
353 list_del(&inode
->i_list
);
355 if (inode
->i_data
.nrpages
)
356 truncate_inode_pages(&inode
->i_data
, 0);
359 spin_lock(&inode_lock
);
360 hlist_del_init(&inode
->i_hash
);
361 list_del_init(&inode
->i_sb_list
);
362 spin_unlock(&inode_lock
);
364 wake_up_inode(inode
);
365 destroy_inode(inode
);
368 spin_lock(&inode_lock
);
369 inodes_stat
.nr_inodes
-= nr_disposed
;
370 spin_unlock(&inode_lock
);
374 * Invalidate all inodes for a device.
376 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
378 struct list_head
*next
;
379 int busy
= 0, count
= 0;
383 struct list_head
*tmp
= next
;
387 * We can reschedule here without worrying about the list's
388 * consistency because the per-sb list of inodes must not
389 * change during umount anymore, and because iprune_sem keeps
390 * shrink_icache_memory() away.
392 cond_resched_lock(&inode_lock
);
397 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
398 if (inode
->i_state
& I_NEW
)
400 invalidate_inode_buffers(inode
);
401 if (!atomic_read(&inode
->i_count
)) {
402 list_move(&inode
->i_list
, dispose
);
403 WARN_ON(inode
->i_state
& I_NEW
);
404 inode
->i_state
|= I_FREEING
;
410 /* only unused inodes may be cached with i_count zero */
411 inodes_stat
.nr_unused
-= count
;
416 * invalidate_inodes - discard the inodes on a device
419 * Discard all of the inodes for a given superblock. If the discard
420 * fails because there are busy inodes then a non zero value is returned.
421 * If the discard is successful all the inodes have been discarded.
423 int invalidate_inodes(struct super_block
*sb
)
426 LIST_HEAD(throw_away
);
428 down_write(&iprune_sem
);
429 spin_lock(&inode_lock
);
430 inotify_unmount_inodes(&sb
->s_inodes
);
431 fsnotify_unmount_inodes(&sb
->s_inodes
);
432 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
433 spin_unlock(&inode_lock
);
435 dispose_list(&throw_away
);
436 up_write(&iprune_sem
);
440 EXPORT_SYMBOL(invalidate_inodes
);
442 static int can_unuse(struct inode
*inode
)
446 if (inode_has_buffers(inode
))
448 if (atomic_read(&inode
->i_count
))
450 if (inode
->i_data
.nrpages
)
456 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
457 * a temporary list and then are freed outside inode_lock by dispose_list().
459 * Any inodes which are pinned purely because of attached pagecache have their
460 * pagecache removed. We expect the final iput() on that inode to add it to
461 * the front of the inode_unused list. So look for it there and if the
462 * inode is still freeable, proceed. The right inode is found 99.9% of the
463 * time in testing on a 4-way.
465 * If the inode has metadata buffers attached to mapping->private_list then
466 * try to remove them.
468 static void prune_icache(int nr_to_scan
)
473 unsigned long reap
= 0;
475 down_read(&iprune_sem
);
476 spin_lock(&inode_lock
);
477 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
480 if (list_empty(&inode_unused
))
483 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
485 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
486 list_move(&inode
->i_list
, &inode_unused
);
489 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
491 spin_unlock(&inode_lock
);
492 if (remove_inode_buffers(inode
))
493 reap
+= invalidate_mapping_pages(&inode
->i_data
,
496 spin_lock(&inode_lock
);
498 if (inode
!= list_entry(inode_unused
.next
,
499 struct inode
, i_list
))
500 continue; /* wrong inode or list_empty */
501 if (!can_unuse(inode
))
504 list_move(&inode
->i_list
, &freeable
);
505 WARN_ON(inode
->i_state
& I_NEW
);
506 inode
->i_state
|= I_FREEING
;
509 inodes_stat
.nr_unused
-= nr_pruned
;
510 if (current_is_kswapd())
511 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
513 __count_vm_events(PGINODESTEAL
, reap
);
514 spin_unlock(&inode_lock
);
516 dispose_list(&freeable
);
517 up_read(&iprune_sem
);
521 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
522 * "unused" means that no dentries are referring to the inodes: the files are
523 * not open and the dcache references to those inodes have already been
526 * This function is passed the number of inodes to scan, and it returns the
527 * total number of remaining possibly-reclaimable inodes.
529 static int shrink_icache_memory(int nr
, gfp_t gfp_mask
)
533 * Nasty deadlock avoidance. We may hold various FS locks,
534 * and we don't want to recurse into the FS that called us
535 * in clear_inode() and friends..
537 if (!(gfp_mask
& __GFP_FS
))
541 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
544 static struct shrinker icache_shrinker
= {
545 .shrink
= shrink_icache_memory
,
546 .seeks
= DEFAULT_SEEKS
,
549 static void __wait_on_freeing_inode(struct inode
*inode
);
551 * Called with the inode lock held.
552 * NOTE: we are not increasing the inode-refcount, you must call __iget()
553 * by hand after calling find_inode now! This simplifies iunique and won't
554 * add any additional branch in the common code.
556 static struct inode
*find_inode(struct super_block
*sb
,
557 struct hlist_head
*head
,
558 int (*test
)(struct inode
*, void *),
561 struct hlist_node
*node
;
562 struct inode
*inode
= NULL
;
565 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
566 if (inode
->i_sb
!= sb
)
568 if (!test(inode
, data
))
570 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
571 __wait_on_freeing_inode(inode
);
576 return node
? inode
: NULL
;
580 * find_inode_fast is the fast path version of find_inode, see the comment at
581 * iget_locked for details.
583 static struct inode
*find_inode_fast(struct super_block
*sb
,
584 struct hlist_head
*head
, unsigned long ino
)
586 struct hlist_node
*node
;
587 struct inode
*inode
= NULL
;
590 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
591 if (inode
->i_ino
!= ino
)
593 if (inode
->i_sb
!= sb
)
595 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
596 __wait_on_freeing_inode(inode
);
601 return node
? inode
: NULL
;
604 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
608 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
610 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
611 return tmp
& I_HASHMASK
;
615 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
618 inodes_stat
.nr_inodes
++;
619 list_add(&inode
->i_list
, &inode_in_use
);
620 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
622 hlist_add_head(&inode
->i_hash
, head
);
626 * inode_add_to_lists - add a new inode to relevant lists
627 * @sb: superblock inode belongs to
628 * @inode: inode to mark in use
630 * When an inode is allocated it needs to be accounted for, added to the in use
631 * list, the owning superblock and the inode hash. This needs to be done under
632 * the inode_lock, so export a function to do this rather than the inode lock
633 * itself. We calculate the hash list to add to here so it is all internal
634 * which requires the caller to have already set up the inode number in the
637 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
639 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
641 spin_lock(&inode_lock
);
642 __inode_add_to_lists(sb
, head
, inode
);
643 spin_unlock(&inode_lock
);
645 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
648 * new_inode - obtain an inode
651 * Allocates a new inode for given superblock. The default gfp_mask
652 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
653 * If HIGHMEM pages are unsuitable or it is known that pages allocated
654 * for the page cache are not reclaimable or migratable,
655 * mapping_set_gfp_mask() must be called with suitable flags on the
656 * newly created inode's mapping
659 struct inode
*new_inode(struct super_block
*sb
)
662 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
663 * error if st_ino won't fit in target struct field. Use 32bit counter
664 * here to attempt to avoid that.
666 static unsigned int last_ino
;
669 spin_lock_prefetch(&inode_lock
);
671 inode
= alloc_inode(sb
);
673 spin_lock(&inode_lock
);
674 __inode_add_to_lists(sb
, NULL
, inode
);
675 inode
->i_ino
= ++last_ino
;
677 spin_unlock(&inode_lock
);
681 EXPORT_SYMBOL(new_inode
);
683 void unlock_new_inode(struct inode
*inode
)
685 #ifdef CONFIG_DEBUG_LOCK_ALLOC
686 if (inode
->i_mode
& S_IFDIR
) {
687 struct file_system_type
*type
= inode
->i_sb
->s_type
;
689 /* Set new key only if filesystem hasn't already changed it */
690 if (!lockdep_match_class(&inode
->i_mutex
,
691 &type
->i_mutex_key
)) {
693 * ensure nobody is actually holding i_mutex
695 mutex_destroy(&inode
->i_mutex
);
696 mutex_init(&inode
->i_mutex
);
697 lockdep_set_class(&inode
->i_mutex
,
698 &type
->i_mutex_dir_key
);
703 * This is special! We do not need the spinlock when clearing I_LOCK,
704 * because we're guaranteed that nobody else tries to do anything about
705 * the state of the inode when it is locked, as we just created it (so
706 * there can be no old holders that haven't tested I_LOCK).
707 * However we must emit the memory barrier so that other CPUs reliably
708 * see the clearing of I_LOCK after the other inode initialisation has
712 WARN_ON((inode
->i_state
& (I_LOCK
|I_NEW
)) != (I_LOCK
|I_NEW
));
713 inode
->i_state
&= ~(I_LOCK
|I_NEW
);
714 wake_up_inode(inode
);
716 EXPORT_SYMBOL(unlock_new_inode
);
719 * This is called without the inode lock held.. Be careful.
721 * We no longer cache the sb_flags in i_flags - see fs.h
722 * -- rmk@arm.uk.linux.org
724 static struct inode
*get_new_inode(struct super_block
*sb
,
725 struct hlist_head
*head
,
726 int (*test
)(struct inode
*, void *),
727 int (*set
)(struct inode
*, void *),
732 inode
= alloc_inode(sb
);
736 spin_lock(&inode_lock
);
737 /* We released the lock, so.. */
738 old
= find_inode(sb
, head
, test
, data
);
740 if (set(inode
, data
))
743 __inode_add_to_lists(sb
, head
, inode
);
744 inode
->i_state
= I_LOCK
|I_NEW
;
745 spin_unlock(&inode_lock
);
747 /* Return the locked inode with I_NEW set, the
748 * caller is responsible for filling in the contents
754 * Uhhuh, somebody else created the same inode under
755 * us. Use the old inode instead of the one we just
759 spin_unlock(&inode_lock
);
760 destroy_inode(inode
);
762 wait_on_inode(inode
);
767 spin_unlock(&inode_lock
);
768 destroy_inode(inode
);
773 * get_new_inode_fast is the fast path version of get_new_inode, see the
774 * comment at iget_locked for details.
776 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
777 struct hlist_head
*head
, unsigned long ino
)
781 inode
= alloc_inode(sb
);
785 spin_lock(&inode_lock
);
786 /* We released the lock, so.. */
787 old
= find_inode_fast(sb
, head
, ino
);
790 __inode_add_to_lists(sb
, head
, inode
);
791 inode
->i_state
= I_LOCK
|I_NEW
;
792 spin_unlock(&inode_lock
);
794 /* Return the locked inode with I_NEW set, the
795 * caller is responsible for filling in the contents
801 * Uhhuh, somebody else created the same inode under
802 * us. Use the old inode instead of the one we just
806 spin_unlock(&inode_lock
);
807 destroy_inode(inode
);
809 wait_on_inode(inode
);
815 * iunique - get a unique inode number
817 * @max_reserved: highest reserved inode number
819 * Obtain an inode number that is unique on the system for a given
820 * superblock. This is used by file systems that have no natural
821 * permanent inode numbering system. An inode number is returned that
822 * is higher than the reserved limit but unique.
825 * With a large number of inodes live on the file system this function
826 * currently becomes quite slow.
828 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
831 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
832 * error if st_ino won't fit in target struct field. Use 32bit counter
833 * here to attempt to avoid that.
835 static unsigned int counter
;
837 struct hlist_head
*head
;
840 spin_lock(&inode_lock
);
842 if (counter
<= max_reserved
)
843 counter
= max_reserved
+ 1;
845 head
= inode_hashtable
+ hash(sb
, res
);
846 inode
= find_inode_fast(sb
, head
, res
);
847 } while (inode
!= NULL
);
848 spin_unlock(&inode_lock
);
852 EXPORT_SYMBOL(iunique
);
854 struct inode
*igrab(struct inode
*inode
)
856 spin_lock(&inode_lock
);
857 if (!(inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)))
861 * Handle the case where s_op->clear_inode is not been
862 * called yet, and somebody is calling igrab
863 * while the inode is getting freed.
866 spin_unlock(&inode_lock
);
869 EXPORT_SYMBOL(igrab
);
872 * ifind - internal function, you want ilookup5() or iget5().
873 * @sb: super block of file system to search
874 * @head: the head of the list to search
875 * @test: callback used for comparisons between inodes
876 * @data: opaque data pointer to pass to @test
877 * @wait: if true wait for the inode to be unlocked, if false do not
879 * ifind() searches for the inode specified by @data in the inode
880 * cache. This is a generalized version of ifind_fast() for file systems where
881 * the inode number is not sufficient for unique identification of an inode.
883 * If the inode is in the cache, the inode is returned with an incremented
886 * Otherwise NULL is returned.
888 * Note, @test is called with the inode_lock held, so can't sleep.
890 static struct inode
*ifind(struct super_block
*sb
,
891 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
892 void *data
, const int wait
)
896 spin_lock(&inode_lock
);
897 inode
= find_inode(sb
, head
, test
, data
);
900 spin_unlock(&inode_lock
);
902 wait_on_inode(inode
);
905 spin_unlock(&inode_lock
);
910 * ifind_fast - internal function, you want ilookup() or iget().
911 * @sb: super block of file system to search
912 * @head: head of the list to search
913 * @ino: inode number to search for
915 * ifind_fast() searches for the inode @ino in the inode cache. This is for
916 * file systems where the inode number is sufficient for unique identification
919 * If the inode is in the cache, the inode is returned with an incremented
922 * Otherwise NULL is returned.
924 static struct inode
*ifind_fast(struct super_block
*sb
,
925 struct hlist_head
*head
, unsigned long ino
)
929 spin_lock(&inode_lock
);
930 inode
= find_inode_fast(sb
, head
, ino
);
933 spin_unlock(&inode_lock
);
934 wait_on_inode(inode
);
937 spin_unlock(&inode_lock
);
942 * ilookup5_nowait - search for an inode in the inode cache
943 * @sb: super block of file system to search
944 * @hashval: hash value (usually inode number) to search for
945 * @test: callback used for comparisons between inodes
946 * @data: opaque data pointer to pass to @test
948 * ilookup5() uses ifind() to search for the inode specified by @hashval and
949 * @data in the inode cache. This is a generalized version of ilookup() for
950 * file systems where the inode number is not sufficient for unique
951 * identification of an inode.
953 * If the inode is in the cache, the inode is returned with an incremented
954 * reference count. Note, the inode lock is not waited upon so you have to be
955 * very careful what you do with the returned inode. You probably should be
956 * using ilookup5() instead.
958 * Otherwise NULL is returned.
960 * Note, @test is called with the inode_lock held, so can't sleep.
962 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
963 int (*test
)(struct inode
*, void *), void *data
)
965 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
967 return ifind(sb
, head
, test
, data
, 0);
969 EXPORT_SYMBOL(ilookup5_nowait
);
972 * ilookup5 - search for an inode in the inode cache
973 * @sb: super block of file system to search
974 * @hashval: hash value (usually inode number) to search for
975 * @test: callback used for comparisons between inodes
976 * @data: opaque data pointer to pass to @test
978 * ilookup5() uses ifind() to search for the inode specified by @hashval and
979 * @data in the inode cache. This is a generalized version of ilookup() for
980 * file systems where the inode number is not sufficient for unique
981 * identification of an inode.
983 * If the inode is in the cache, the inode lock is waited upon and the inode is
984 * returned with an incremented reference count.
986 * Otherwise NULL is returned.
988 * Note, @test is called with the inode_lock held, so can't sleep.
990 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
991 int (*test
)(struct inode
*, void *), void *data
)
993 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
995 return ifind(sb
, head
, test
, data
, 1);
997 EXPORT_SYMBOL(ilookup5
);
1000 * ilookup - search for an inode in the inode cache
1001 * @sb: super block of file system to search
1002 * @ino: inode number to search for
1004 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1005 * This is for file systems where the inode number is sufficient for unique
1006 * identification of an inode.
1008 * If the inode is in the cache, the inode is returned with an incremented
1011 * Otherwise NULL is returned.
1013 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1015 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1017 return ifind_fast(sb
, head
, ino
);
1019 EXPORT_SYMBOL(ilookup
);
1022 * iget5_locked - obtain an inode from a mounted file system
1023 * @sb: super block of file system
1024 * @hashval: hash value (usually inode number) to get
1025 * @test: callback used for comparisons between inodes
1026 * @set: callback used to initialize a new struct inode
1027 * @data: opaque data pointer to pass to @test and @set
1029 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1030 * and @data in the inode cache and if present it is returned with an increased
1031 * reference count. This is a generalized version of iget_locked() for file
1032 * systems where the inode number is not sufficient for unique identification
1035 * If the inode is not in cache, get_new_inode() is called to allocate a new
1036 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1037 * file system gets to fill it in before unlocking it via unlock_new_inode().
1039 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1041 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1042 int (*test
)(struct inode
*, void *),
1043 int (*set
)(struct inode
*, void *), void *data
)
1045 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1046 struct inode
*inode
;
1048 inode
= ifind(sb
, head
, test
, data
, 1);
1052 * get_new_inode() will do the right thing, re-trying the search
1053 * in case it had to block at any point.
1055 return get_new_inode(sb
, head
, test
, set
, data
);
1057 EXPORT_SYMBOL(iget5_locked
);
1060 * iget_locked - obtain an inode from a mounted file system
1061 * @sb: super block of file system
1062 * @ino: inode number to get
1064 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1065 * the inode cache and if present it is returned with an increased reference
1066 * count. This is for file systems where the inode number is sufficient for
1067 * unique identification of an inode.
1069 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1070 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1071 * The file system gets to fill it in before unlocking it via
1072 * unlock_new_inode().
1074 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1076 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1077 struct inode
*inode
;
1079 inode
= ifind_fast(sb
, head
, ino
);
1083 * get_new_inode_fast() will do the right thing, re-trying the search
1084 * in case it had to block at any point.
1086 return get_new_inode_fast(sb
, head
, ino
);
1088 EXPORT_SYMBOL(iget_locked
);
1090 int insert_inode_locked(struct inode
*inode
)
1092 struct super_block
*sb
= inode
->i_sb
;
1093 ino_t ino
= inode
->i_ino
;
1094 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1096 inode
->i_state
|= I_LOCK
|I_NEW
;
1098 struct hlist_node
*node
;
1099 struct inode
*old
= NULL
;
1100 spin_lock(&inode_lock
);
1101 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1102 if (old
->i_ino
!= ino
)
1104 if (old
->i_sb
!= sb
)
1106 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1110 if (likely(!node
)) {
1111 hlist_add_head(&inode
->i_hash
, head
);
1112 spin_unlock(&inode_lock
);
1116 spin_unlock(&inode_lock
);
1118 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1125 EXPORT_SYMBOL(insert_inode_locked
);
1127 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1128 int (*test
)(struct inode
*, void *), void *data
)
1130 struct super_block
*sb
= inode
->i_sb
;
1131 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1133 inode
->i_state
|= I_LOCK
|I_NEW
;
1136 struct hlist_node
*node
;
1137 struct inode
*old
= NULL
;
1139 spin_lock(&inode_lock
);
1140 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1141 if (old
->i_sb
!= sb
)
1143 if (!test(old
, data
))
1145 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1149 if (likely(!node
)) {
1150 hlist_add_head(&inode
->i_hash
, head
);
1151 spin_unlock(&inode_lock
);
1155 spin_unlock(&inode_lock
);
1157 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1164 EXPORT_SYMBOL(insert_inode_locked4
);
1167 * __insert_inode_hash - hash an inode
1168 * @inode: unhashed inode
1169 * @hashval: unsigned long value used to locate this object in the
1172 * Add an inode to the inode hash for this superblock.
1174 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1176 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1177 spin_lock(&inode_lock
);
1178 hlist_add_head(&inode
->i_hash
, head
);
1179 spin_unlock(&inode_lock
);
1181 EXPORT_SYMBOL(__insert_inode_hash
);
1184 * remove_inode_hash - remove an inode from the hash
1185 * @inode: inode to unhash
1187 * Remove an inode from the superblock.
1189 void remove_inode_hash(struct inode
*inode
)
1191 spin_lock(&inode_lock
);
1192 hlist_del_init(&inode
->i_hash
);
1193 spin_unlock(&inode_lock
);
1195 EXPORT_SYMBOL(remove_inode_hash
);
1198 * Tell the filesystem that this inode is no longer of any interest and should
1199 * be completely destroyed.
1201 * We leave the inode in the inode hash table until *after* the filesystem's
1202 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1203 * instigate) will always find up-to-date information either in the hash or on
1206 * I_FREEING is set so that no-one will take a new reference to the inode while
1207 * it is being deleted.
1209 void generic_delete_inode(struct inode
*inode
)
1211 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1213 list_del_init(&inode
->i_list
);
1214 list_del_init(&inode
->i_sb_list
);
1215 WARN_ON(inode
->i_state
& I_NEW
);
1216 inode
->i_state
|= I_FREEING
;
1217 inodes_stat
.nr_inodes
--;
1218 spin_unlock(&inode_lock
);
1220 security_inode_delete(inode
);
1222 if (op
->delete_inode
) {
1223 void (*delete)(struct inode
*) = op
->delete_inode
;
1224 if (!is_bad_inode(inode
))
1226 /* Filesystems implementing their own
1227 * s_op->delete_inode are required to call
1228 * truncate_inode_pages and clear_inode()
1232 truncate_inode_pages(&inode
->i_data
, 0);
1235 spin_lock(&inode_lock
);
1236 hlist_del_init(&inode
->i_hash
);
1237 spin_unlock(&inode_lock
);
1238 wake_up_inode(inode
);
1239 BUG_ON(inode
->i_state
!= I_CLEAR
);
1240 destroy_inode(inode
);
1242 EXPORT_SYMBOL(generic_delete_inode
);
1245 * generic_detach_inode - remove inode from inode lists
1246 * @inode: inode to remove
1248 * Remove inode from inode lists, write it if it's dirty. This is just an
1249 * internal VFS helper exported for hugetlbfs. Do not use!
1251 * Returns 1 if inode should be completely destroyed.
1253 int generic_detach_inode(struct inode
*inode
)
1255 struct super_block
*sb
= inode
->i_sb
;
1257 if (!hlist_unhashed(&inode
->i_hash
)) {
1258 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1259 list_move(&inode
->i_list
, &inode_unused
);
1260 inodes_stat
.nr_unused
++;
1261 if (sb
->s_flags
& MS_ACTIVE
) {
1262 spin_unlock(&inode_lock
);
1265 WARN_ON(inode
->i_state
& I_NEW
);
1266 inode
->i_state
|= I_WILL_FREE
;
1267 spin_unlock(&inode_lock
);
1268 write_inode_now(inode
, 1);
1269 spin_lock(&inode_lock
);
1270 WARN_ON(inode
->i_state
& I_NEW
);
1271 inode
->i_state
&= ~I_WILL_FREE
;
1272 inodes_stat
.nr_unused
--;
1273 hlist_del_init(&inode
->i_hash
);
1275 list_del_init(&inode
->i_list
);
1276 list_del_init(&inode
->i_sb_list
);
1277 WARN_ON(inode
->i_state
& I_NEW
);
1278 inode
->i_state
|= I_FREEING
;
1279 inodes_stat
.nr_inodes
--;
1280 spin_unlock(&inode_lock
);
1283 EXPORT_SYMBOL_GPL(generic_detach_inode
);
1285 static void generic_forget_inode(struct inode
*inode
)
1287 if (!generic_detach_inode(inode
))
1289 if (inode
->i_data
.nrpages
)
1290 truncate_inode_pages(&inode
->i_data
, 0);
1292 wake_up_inode(inode
);
1293 destroy_inode(inode
);
1297 * Normal UNIX filesystem behaviour: delete the
1298 * inode when the usage count drops to zero, and
1301 void generic_drop_inode(struct inode
*inode
)
1303 if (!inode
->i_nlink
)
1304 generic_delete_inode(inode
);
1306 generic_forget_inode(inode
);
1308 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1311 * Called when we're dropping the last reference
1314 * Call the FS "drop()" function, defaulting to
1315 * the legacy UNIX filesystem behaviour..
1317 * NOTE! NOTE! NOTE! We're called with the inode lock
1318 * held, and the drop function is supposed to release
1321 static inline void iput_final(struct inode
*inode
)
1323 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1324 void (*drop
)(struct inode
*) = generic_drop_inode
;
1326 if (op
&& op
->drop_inode
)
1327 drop
= op
->drop_inode
;
1332 * iput - put an inode
1333 * @inode: inode to put
1335 * Puts an inode, dropping its usage count. If the inode use count hits
1336 * zero, the inode is then freed and may also be destroyed.
1338 * Consequently, iput() can sleep.
1340 void iput(struct inode
*inode
)
1343 BUG_ON(inode
->i_state
== I_CLEAR
);
1345 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1349 EXPORT_SYMBOL(iput
);
1352 * bmap - find a block number in a file
1353 * @inode: inode of file
1354 * @block: block to find
1356 * Returns the block number on the device holding the inode that
1357 * is the disk block number for the block of the file requested.
1358 * That is, asked for block 4 of inode 1 the function will return the
1359 * disk block relative to the disk start that holds that block of the
1362 sector_t
bmap(struct inode
*inode
, sector_t block
)
1365 if (inode
->i_mapping
->a_ops
->bmap
)
1366 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1369 EXPORT_SYMBOL(bmap
);
1372 * With relative atime, only update atime if the previous atime is
1373 * earlier than either the ctime or mtime or if at least a day has
1374 * passed since the last atime update.
1376 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1377 struct timespec now
)
1380 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1383 * Is mtime younger than atime? If yes, update atime:
1385 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1388 * Is ctime younger than atime? If yes, update atime:
1390 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1394 * Is the previous atime value older than a day? If yes,
1397 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1400 * Good, we can skip the atime update:
1406 * touch_atime - update the access time
1407 * @mnt: mount the inode is accessed on
1408 * @dentry: dentry accessed
1410 * Update the accessed time on an inode and mark it for writeback.
1411 * This function automatically handles read only file systems and media,
1412 * as well as the "noatime" flag and inode specific "noatime" markers.
1414 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1416 struct inode
*inode
= dentry
->d_inode
;
1417 struct timespec now
;
1419 if (inode
->i_flags
& S_NOATIME
)
1421 if (IS_NOATIME(inode
))
1423 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1426 if (mnt
->mnt_flags
& MNT_NOATIME
)
1428 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1431 now
= current_fs_time(inode
->i_sb
);
1433 if (!relatime_need_update(mnt
, inode
, now
))
1436 if (timespec_equal(&inode
->i_atime
, &now
))
1439 if (mnt_want_write(mnt
))
1442 inode
->i_atime
= now
;
1443 mark_inode_dirty_sync(inode
);
1444 mnt_drop_write(mnt
);
1446 EXPORT_SYMBOL(touch_atime
);
1449 * file_update_time - update mtime and ctime time
1450 * @file: file accessed
1452 * Update the mtime and ctime members of an inode and mark the inode
1453 * for writeback. Note that this function is meant exclusively for
1454 * usage in the file write path of filesystems, and filesystems may
1455 * choose to explicitly ignore update via this function with the
1456 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1457 * timestamps are handled by the server.
1460 void file_update_time(struct file
*file
)
1462 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1463 struct timespec now
;
1464 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1466 /* First try to exhaust all avenues to not sync */
1467 if (IS_NOCMTIME(inode
))
1470 now
= current_fs_time(inode
->i_sb
);
1471 if (!timespec_equal(&inode
->i_mtime
, &now
))
1474 if (!timespec_equal(&inode
->i_ctime
, &now
))
1477 if (IS_I_VERSION(inode
))
1478 sync_it
|= S_VERSION
;
1483 /* Finally allowed to write? Takes lock. */
1484 if (mnt_want_write_file(file
))
1487 /* Only change inode inside the lock region */
1488 if (sync_it
& S_VERSION
)
1489 inode_inc_iversion(inode
);
1490 if (sync_it
& S_CTIME
)
1491 inode
->i_ctime
= now
;
1492 if (sync_it
& S_MTIME
)
1493 inode
->i_mtime
= now
;
1494 mark_inode_dirty_sync(inode
);
1495 mnt_drop_write(file
->f_path
.mnt
);
1497 EXPORT_SYMBOL(file_update_time
);
1499 int inode_needs_sync(struct inode
*inode
)
1503 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1507 EXPORT_SYMBOL(inode_needs_sync
);
1509 int inode_wait(void *word
)
1514 EXPORT_SYMBOL(inode_wait
);
1517 * If we try to find an inode in the inode hash while it is being
1518 * deleted, we have to wait until the filesystem completes its
1519 * deletion before reporting that it isn't found. This function waits
1520 * until the deletion _might_ have completed. Callers are responsible
1521 * to recheck inode state.
1523 * It doesn't matter if I_LOCK is not set initially, a call to
1524 * wake_up_inode() after removing from the hash list will DTRT.
1526 * This is called with inode_lock held.
1528 static void __wait_on_freeing_inode(struct inode
*inode
)
1530 wait_queue_head_t
*wq
;
1531 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_LOCK
);
1532 wq
= bit_waitqueue(&inode
->i_state
, __I_LOCK
);
1533 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1534 spin_unlock(&inode_lock
);
1536 finish_wait(wq
, &wait
.wait
);
1537 spin_lock(&inode_lock
);
1540 static __initdata
unsigned long ihash_entries
;
1541 static int __init
set_ihash_entries(char *str
)
1545 ihash_entries
= simple_strtoul(str
, &str
, 0);
1548 __setup("ihash_entries=", set_ihash_entries
);
1551 * Initialize the waitqueues and inode hash table.
1553 void __init
inode_init_early(void)
1557 /* If hashes are distributed across NUMA nodes, defer
1558 * hash allocation until vmalloc space is available.
1564 alloc_large_system_hash("Inode-cache",
1565 sizeof(struct hlist_head
),
1573 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1574 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1577 void __init
inode_init(void)
1581 /* inode slab cache */
1582 inode_cachep
= kmem_cache_create("inode_cache",
1583 sizeof(struct inode
),
1585 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1588 register_shrinker(&icache_shrinker
);
1590 /* Hash may have been set up in inode_init_early */
1595 alloc_large_system_hash("Inode-cache",
1596 sizeof(struct hlist_head
),
1604 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1605 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1608 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1610 inode
->i_mode
= mode
;
1611 if (S_ISCHR(mode
)) {
1612 inode
->i_fop
= &def_chr_fops
;
1613 inode
->i_rdev
= rdev
;
1614 } else if (S_ISBLK(mode
)) {
1615 inode
->i_fop
= &def_blk_fops
;
1616 inode
->i_rdev
= rdev
;
1617 } else if (S_ISFIFO(mode
))
1618 inode
->i_fop
= &def_fifo_fops
;
1619 else if (S_ISSOCK(mode
))
1620 inode
->i_fop
= &bad_sock_fops
;
1622 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1623 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1626 EXPORT_SYMBOL(init_special_inode
);