4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/slab.h>
12 #include <linux/writeback.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/wait.h>
16 #include <linux/rwsem.h>
17 #include <linux/hash.h>
18 #include <linux/swap.h>
19 #include <linux/security.h>
20 #include <linux/pagemap.h>
21 #include <linux/cdev.h>
22 #include <linux/bootmem.h>
23 #include <linux/inotify.h>
24 #include <linux/fsnotify.h>
25 #include <linux/mount.h>
26 #include <linux/async.h>
27 #include <linux/posix_acl.h>
30 * This is needed for the following functions:
32 * - invalidate_inode_buffers
35 * FIXME: remove all knowledge of the buffer layer from this file
37 #include <linux/buffer_head.h>
40 * New inode.c implementation.
42 * This implementation has the basic premise of trying
43 * to be extremely low-overhead and SMP-safe, yet be
44 * simple enough to be "obviously correct".
49 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
51 /* #define INODE_PARANOIA 1 */
52 /* #define INODE_DEBUG 1 */
55 * Inode lookup is no longer as critical as it used to be:
56 * most of the lookups are going to be through the dcache.
58 #define I_HASHBITS i_hash_shift
59 #define I_HASHMASK i_hash_mask
61 static unsigned int i_hash_mask __read_mostly;
62 static unsigned int i_hash_shift __read_mostly;
65 * Each inode can be on two separate lists. One is
66 * the hash list of the inode, used for lookups. The
67 * other linked list is the "type" list:
68 * "in_use" - valid inode, i_count > 0, i_nlink > 0
69 * "dirty" - as "in_use" but also dirty
70 * "unused" - valid inode, i_count = 0
72 * A "dirty" list is maintained for each super block,
73 * allowing for low-overhead inode sync() operations.
76 LIST_HEAD(inode_in_use);
77 LIST_HEAD(inode_unused);
78 static struct hlist_head *inode_hashtable __read_mostly;
81 * A simple spinlock to protect the list manipulations.
83 * NOTE! You also have to own the lock if you change
84 * the i_state of an inode while it is in use..
86 DEFINE_SPINLOCK(inode_lock);
89 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
90 * icache shrinking path, and the umount path. Without this exclusion,
91 * by the time prune_icache calls iput for the inode whose pages it has
92 * been invalidating, or by the time it calls clear_inode & destroy_inode
93 * from its final dispose_list, the struct super_block they refer to
94 * (for inode->i_sb->s_op) may already have been freed and reused.
96 * We make this an rwsem because the fastpath is icache shrinking. In
97 * some cases a filesystem may be doing a significant amount of work in
98 * its inode reclaim code, so this should improve parallelism.
100 static DECLARE_RWSEM(iprune_sem);
103 * Statistics gathering..
105 struct inodes_stat_t inodes_stat;
107 static struct kmem_cache *inode_cachep __read_mostly;
109 static void wake_up_inode(struct inode *inode)
112 * Prevent speculative execution through spin_unlock(&inode_lock);
115 wake_up_bit(&inode->i_state, __I_NEW);
119 * inode_init_always - perform inode structure intialisation
120 * @sb: superblock inode belongs to
121 * @inode: inode to initialise
123 * These are initializations that need to be done on every inode
124 * allocation as the fields are not initialised by slab allocation.
126 int inode_init_always(struct super_block *sb, struct inode *inode)
128 static const struct address_space_operations empty_aops;
129 static const struct inode_operations empty_iops;
130 static const struct file_operations empty_fops;
131 struct address_space *const mapping = &inode->i_data;
134 inode->i_blkbits = sb->s_blocksize_bits;
136 atomic_set(&inode->i_count, 1);
137 inode->i_op = &empty_iops;
138 inode->i_fop = &empty_fops;
142 atomic_set(&inode->i_writecount, 0);
146 inode->i_generation = 0;
148 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
150 inode->i_pipe = NULL;
151 inode->i_bdev = NULL;
152 inode->i_cdev = NULL;
154 inode->dirtied_when = 0;
156 if (security_inode_alloc(inode))
158 spin_lock_init(&inode->i_lock);
159 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
161 mutex_init(&inode->i_mutex);
162 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
164 init_rwsem(&inode->i_alloc_sem);
165 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
167 mapping->a_ops = &empty_aops;
168 mapping->host = inode;
170 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
171 mapping->assoc_mapping = NULL;
172 mapping->backing_dev_info = &default_backing_dev_info;
173 mapping->writeback_index = 0;
176 * If the block_device provides a backing_dev_info for client
177 * inodes then use that. Otherwise the inode share the bdev's
181 struct backing_dev_info *bdi;
183 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
184 mapping->backing_dev_info = bdi;
186 inode->i_private = NULL;
187 inode->i_mapping = mapping;
188 #ifdef CONFIG_FS_POSIX_ACL
189 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
192 #ifdef CONFIG_FSNOTIFY
193 inode->i_fsnotify_mask = 0;
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 __destroy_inode(struct inode *inode)
227 BUG_ON(inode_has_buffers(inode));
228 security_inode_free(inode);
229 fsnotify_inode_delete(inode);
230 #ifdef CONFIG_FS_POSIX_ACL
231 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
232 posix_acl_release(inode->i_acl);
233 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
234 posix_acl_release(inode->i_default_acl);
237 EXPORT_SYMBOL(__destroy_inode);
239 void destroy_inode(struct inode *inode)
241 __destroy_inode(inode);
242 if (inode->i_sb->s_op->destroy_inode)
243 inode->i_sb->s_op->destroy_inode(inode);
245 kmem_cache_free(inode_cachep, (inode));
249 * These are initializations that only need to be done
250 * once, because the fields are idempotent across use
251 * of the inode, so let the slab aware of that.
253 void inode_init_once(struct inode *inode)
255 memset(inode, 0, sizeof(*inode));
256 INIT_HLIST_NODE(&inode->i_hash);
257 INIT_LIST_HEAD(&inode->i_dentry);
258 INIT_LIST_HEAD(&inode->i_devices);
259 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
260 spin_lock_init(&inode->i_data.tree_lock);
261 spin_lock_init(&inode->i_data.i_mmap_lock);
262 INIT_LIST_HEAD(&inode->i_data.private_list);
263 spin_lock_init(&inode->i_data.private_lock);
264 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
265 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
266 i_size_ordered_init(inode);
267 #ifdef CONFIG_INOTIFY
268 INIT_LIST_HEAD(&inode->inotify_watches);
269 mutex_init(&inode->inotify_mutex);
271 #ifdef CONFIG_FSNOTIFY
272 INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries);
275 EXPORT_SYMBOL(inode_init_once);
277 static void init_once(void *foo)
279 struct inode *inode = (struct inode *) foo;
281 inode_init_once(inode);
285 * inode_lock must be held
287 void __iget(struct inode *inode)
289 if (atomic_read(&inode->i_count)) {
290 atomic_inc(&inode->i_count);
293 atomic_inc(&inode->i_count);
294 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
295 list_move(&inode->i_list, &inode_in_use);
296 inodes_stat.nr_unused--;
300 * clear_inode - clear an inode
301 * @inode: inode to clear
303 * This is called by the filesystem to tell us
304 * that the inode is no longer useful. We just
305 * terminate it with extreme prejudice.
307 void clear_inode(struct inode *inode)
310 invalidate_inode_buffers(inode);
312 BUG_ON(inode->i_data.nrpages);
313 BUG_ON(!(inode->i_state & I_FREEING));
314 BUG_ON(inode->i_state & I_CLEAR);
315 inode_sync_wait(inode);
316 if (inode->i_sb->s_op->clear_inode)
317 inode->i_sb->s_op->clear_inode(inode);
318 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
320 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
322 inode->i_state = I_CLEAR;
324 EXPORT_SYMBOL(clear_inode);
327 * dispose_list - dispose of the contents of a local list
328 * @head: the head of the list to free
330 * Dispose-list gets a local list with local inodes in it, so it doesn't
331 * need to worry about list corruption and SMP locks.
333 static void dispose_list(struct list_head *head)
337 while (!list_empty(head)) {
340 inode = list_first_entry(head, struct inode, i_list);
341 list_del(&inode->i_list);
343 if (inode->i_data.nrpages)
344 truncate_inode_pages(&inode->i_data, 0);
347 spin_lock(&inode_lock);
348 hlist_del_init(&inode->i_hash);
349 list_del_init(&inode->i_sb_list);
350 spin_unlock(&inode_lock);
352 wake_up_inode(inode);
353 destroy_inode(inode);
356 spin_lock(&inode_lock);
357 inodes_stat.nr_inodes -= nr_disposed;
358 spin_unlock(&inode_lock);
362 * Invalidate all inodes for a device.
364 static int invalidate_list(struct list_head *head, struct list_head *dispose)
366 struct list_head *next;
367 int busy = 0, count = 0;
371 struct list_head *tmp = next;
375 * We can reschedule here without worrying about the list's
376 * consistency because the per-sb list of inodes must not
377 * change during umount anymore, and because iprune_sem keeps
378 * shrink_icache_memory() away.
380 cond_resched_lock(&inode_lock);
385 inode = list_entry(tmp, struct inode, i_sb_list);
386 if (inode->i_state & I_NEW)
388 invalidate_inode_buffers(inode);
389 if (!atomic_read(&inode->i_count)) {
390 list_move(&inode->i_list, dispose);
391 WARN_ON(inode->i_state & I_NEW);
392 inode->i_state |= I_FREEING;
398 /* only unused inodes may be cached with i_count zero */
399 inodes_stat.nr_unused -= count;
404 * invalidate_inodes - discard the inodes on a device
407 * Discard all of the inodes for a given superblock. If the discard
408 * fails because there are busy inodes then a non zero value is returned.
409 * If the discard is successful all the inodes have been discarded.
411 int invalidate_inodes(struct super_block *sb)
414 LIST_HEAD(throw_away);
416 down_write(&iprune_sem);
417 spin_lock(&inode_lock);
418 inotify_unmount_inodes(&sb->s_inodes);
419 fsnotify_unmount_inodes(&sb->s_inodes);
420 busy = invalidate_list(&sb->s_inodes, &throw_away);
421 spin_unlock(&inode_lock);
423 dispose_list(&throw_away);
424 up_write(&iprune_sem);
428 EXPORT_SYMBOL(invalidate_inodes);
430 static int can_unuse(struct inode *inode)
434 if (inode_has_buffers(inode))
436 if (atomic_read(&inode->i_count))
438 if (inode->i_data.nrpages)
444 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
445 * a temporary list and then are freed outside inode_lock by dispose_list().
447 * Any inodes which are pinned purely because of attached pagecache have their
448 * pagecache removed. We expect the final iput() on that inode to add it to
449 * the front of the inode_unused list. So look for it there and if the
450 * inode is still freeable, proceed. The right inode is found 99.9% of the
451 * time in testing on a 4-way.
453 * If the inode has metadata buffers attached to mapping->private_list then
454 * try to remove them.
456 static void prune_icache(int nr_to_scan)
461 unsigned long reap = 0;
463 down_read(&iprune_sem);
464 spin_lock(&inode_lock);
465 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
468 if (list_empty(&inode_unused))
471 inode = list_entry(inode_unused.prev, struct inode, i_list);
473 if (inode->i_state || atomic_read(&inode->i_count)) {
474 list_move(&inode->i_list, &inode_unused);
477 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
479 spin_unlock(&inode_lock);
480 if (remove_inode_buffers(inode))
481 reap += invalidate_mapping_pages(&inode->i_data,
484 spin_lock(&inode_lock);
486 if (inode != list_entry(inode_unused.next,
487 struct inode, i_list))
488 continue; /* wrong inode or list_empty */
489 if (!can_unuse(inode))
492 list_move(&inode->i_list, &freeable);
493 WARN_ON(inode->i_state & I_NEW);
494 inode->i_state |= I_FREEING;
497 inodes_stat.nr_unused -= nr_pruned;
498 if (current_is_kswapd())
499 __count_vm_events(KSWAPD_INODESTEAL, reap);
501 __count_vm_events(PGINODESTEAL, reap);
502 spin_unlock(&inode_lock);
504 dispose_list(&freeable);
505 up_read(&iprune_sem);
509 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
510 * "unused" means that no dentries are referring to the inodes: the files are
511 * not open and the dcache references to those inodes have already been
514 * This function is passed the number of inodes to scan, and it returns the
515 * total number of remaining possibly-reclaimable inodes.
517 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
521 * Nasty deadlock avoidance. We may hold various FS locks,
522 * and we don't want to recurse into the FS that called us
523 * in clear_inode() and friends..
525 if (!(gfp_mask & __GFP_FS))
529 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
532 static struct shrinker icache_shrinker = {
533 .shrink = shrink_icache_memory,
534 .seeks = DEFAULT_SEEKS,
537 static void __wait_on_freeing_inode(struct inode *inode);
539 * Called with the inode lock held.
540 * NOTE: we are not increasing the inode-refcount, you must call __iget()
541 * by hand after calling find_inode now! This simplifies iunique and won't
542 * add any additional branch in the common code.
544 static struct inode *find_inode(struct super_block *sb,
545 struct hlist_head *head,
546 int (*test)(struct inode *, void *),
549 struct hlist_node *node;
550 struct inode *inode = NULL;
553 hlist_for_each_entry(inode, node, head, i_hash) {
554 if (inode->i_sb != sb)
556 if (!test(inode, data))
558 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
559 __wait_on_freeing_inode(inode);
564 return node ? inode : NULL;
568 * find_inode_fast is the fast path version of find_inode, see the comment at
569 * iget_locked for details.
571 static struct inode *find_inode_fast(struct super_block *sb,
572 struct hlist_head *head, unsigned long ino)
574 struct hlist_node *node;
575 struct inode *inode = NULL;
578 hlist_for_each_entry(inode, node, head, i_hash) {
579 if (inode->i_ino != ino)
581 if (inode->i_sb != sb)
583 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
584 __wait_on_freeing_inode(inode);
589 return node ? inode : NULL;
592 static unsigned long hash(struct super_block *sb, unsigned long hashval)
596 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
598 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
599 return tmp & I_HASHMASK;
603 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
606 inodes_stat.nr_inodes++;
607 list_add(&inode->i_list, &inode_in_use);
608 list_add(&inode->i_sb_list, &sb->s_inodes);
610 hlist_add_head(&inode->i_hash, head);
614 * inode_add_to_lists - add a new inode to relevant lists
615 * @sb: superblock inode belongs to
616 * @inode: inode to mark in use
618 * When an inode is allocated it needs to be accounted for, added to the in use
619 * list, the owning superblock and the inode hash. This needs to be done under
620 * the inode_lock, so export a function to do this rather than the inode lock
621 * itself. We calculate the hash list to add to here so it is all internal
622 * which requires the caller to have already set up the inode number in the
625 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
627 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
629 spin_lock(&inode_lock);
630 __inode_add_to_lists(sb, head, inode);
631 spin_unlock(&inode_lock);
633 EXPORT_SYMBOL_GPL(inode_add_to_lists);
636 * new_inode - obtain an inode
639 * Allocates a new inode for given superblock. The default gfp_mask
640 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
641 * If HIGHMEM pages are unsuitable or it is known that pages allocated
642 * for the page cache are not reclaimable or migratable,
643 * mapping_set_gfp_mask() must be called with suitable flags on the
644 * newly created inode's mapping
647 struct inode *new_inode(struct super_block *sb)
650 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
651 * error if st_ino won't fit in target struct field. Use 32bit counter
652 * here to attempt to avoid that.
654 static unsigned int last_ino;
657 spin_lock_prefetch(&inode_lock);
659 inode = alloc_inode(sb);
661 spin_lock(&inode_lock);
662 __inode_add_to_lists(sb, NULL, inode);
663 inode->i_ino = ++last_ino;
665 spin_unlock(&inode_lock);
669 EXPORT_SYMBOL(new_inode);
671 void unlock_new_inode(struct inode *inode)
673 #ifdef CONFIG_DEBUG_LOCK_ALLOC
674 if (inode->i_mode & S_IFDIR) {
675 struct file_system_type *type = inode->i_sb->s_type;
677 /* Set new key only if filesystem hasn't already changed it */
678 if (!lockdep_match_class(&inode->i_mutex,
679 &type->i_mutex_key)) {
681 * ensure nobody is actually holding i_mutex
683 mutex_destroy(&inode->i_mutex);
684 mutex_init(&inode->i_mutex);
685 lockdep_set_class(&inode->i_mutex,
686 &type->i_mutex_dir_key);
691 * This is special! We do not need the spinlock when clearing I_NEW,
692 * because we're guaranteed that nobody else tries to do anything about
693 * the state of the inode when it is locked, as we just created it (so
694 * there can be no old holders that haven't tested I_NEW).
695 * However we must emit the memory barrier so that other CPUs reliably
696 * see the clearing of I_NEW after the other inode initialisation has
700 WARN_ON(!(inode->i_state & I_NEW));
701 inode->i_state &= ~I_NEW;
702 wake_up_inode(inode);
704 EXPORT_SYMBOL(unlock_new_inode);
707 * This is called without the inode lock held.. Be careful.
709 * We no longer cache the sb_flags in i_flags - see fs.h
710 * -- rmk@arm.uk.linux.org
712 static struct inode *get_new_inode(struct super_block *sb,
713 struct hlist_head *head,
714 int (*test)(struct inode *, void *),
715 int (*set)(struct inode *, void *),
720 inode = alloc_inode(sb);
724 spin_lock(&inode_lock);
725 /* We released the lock, so.. */
726 old = find_inode(sb, head, test, data);
728 if (set(inode, data))
731 __inode_add_to_lists(sb, head, inode);
732 inode->i_state = I_NEW;
733 spin_unlock(&inode_lock);
735 /* Return the locked inode with I_NEW set, the
736 * caller is responsible for filling in the contents
742 * Uhhuh, somebody else created the same inode under
743 * us. Use the old inode instead of the one we just
747 spin_unlock(&inode_lock);
748 destroy_inode(inode);
750 wait_on_inode(inode);
755 spin_unlock(&inode_lock);
756 destroy_inode(inode);
761 * get_new_inode_fast is the fast path version of get_new_inode, see the
762 * comment at iget_locked for details.
764 static struct inode *get_new_inode_fast(struct super_block *sb,
765 struct hlist_head *head, unsigned long ino)
769 inode = alloc_inode(sb);
773 spin_lock(&inode_lock);
774 /* We released the lock, so.. */
775 old = find_inode_fast(sb, head, ino);
778 __inode_add_to_lists(sb, head, inode);
779 inode->i_state = I_NEW;
780 spin_unlock(&inode_lock);
782 /* Return the locked inode with I_NEW set, the
783 * caller is responsible for filling in the contents
789 * Uhhuh, somebody else created the same inode under
790 * us. Use the old inode instead of the one we just
794 spin_unlock(&inode_lock);
795 destroy_inode(inode);
797 wait_on_inode(inode);
803 * iunique - get a unique inode number
805 * @max_reserved: highest reserved inode number
807 * Obtain an inode number that is unique on the system for a given
808 * superblock. This is used by file systems that have no natural
809 * permanent inode numbering system. An inode number is returned that
810 * is higher than the reserved limit but unique.
813 * With a large number of inodes live on the file system this function
814 * currently becomes quite slow.
816 ino_t iunique(struct super_block *sb, ino_t max_reserved)
819 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
820 * error if st_ino won't fit in target struct field. Use 32bit counter
821 * here to attempt to avoid that.
823 static unsigned int counter;
825 struct hlist_head *head;
828 spin_lock(&inode_lock);
830 if (counter <= max_reserved)
831 counter = max_reserved + 1;
833 head = inode_hashtable + hash(sb, res);
834 inode = find_inode_fast(sb, head, res);
835 } while (inode != NULL);
836 spin_unlock(&inode_lock);
840 EXPORT_SYMBOL(iunique);
842 struct inode *igrab(struct inode *inode)
844 spin_lock(&inode_lock);
845 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
849 * Handle the case where s_op->clear_inode is not been
850 * called yet, and somebody is calling igrab
851 * while the inode is getting freed.
854 spin_unlock(&inode_lock);
857 EXPORT_SYMBOL(igrab);
860 * ifind - internal function, you want ilookup5() or iget5().
861 * @sb: super block of file system to search
862 * @head: the head of the list to search
863 * @test: callback used for comparisons between inodes
864 * @data: opaque data pointer to pass to @test
865 * @wait: if true wait for the inode to be unlocked, if false do not
867 * ifind() searches for the inode specified by @data in the inode
868 * cache. This is a generalized version of ifind_fast() for file systems where
869 * the inode number is not sufficient for unique identification of an inode.
871 * If the inode is in the cache, the inode is returned with an incremented
874 * Otherwise NULL is returned.
876 * Note, @test is called with the inode_lock held, so can't sleep.
878 static struct inode *ifind(struct super_block *sb,
879 struct hlist_head *head, int (*test)(struct inode *, void *),
880 void *data, const int wait)
884 spin_lock(&inode_lock);
885 inode = find_inode(sb, head, test, data);
888 spin_unlock(&inode_lock);
890 wait_on_inode(inode);
893 spin_unlock(&inode_lock);
898 * ifind_fast - internal function, you want ilookup() or iget().
899 * @sb: super block of file system to search
900 * @head: head of the list to search
901 * @ino: inode number to search for
903 * ifind_fast() searches for the inode @ino in the inode cache. This is for
904 * file systems where the inode number is sufficient for unique identification
907 * If the inode is in the cache, the inode is returned with an incremented
910 * Otherwise NULL is returned.
912 static struct inode *ifind_fast(struct super_block *sb,
913 struct hlist_head *head, unsigned long ino)
917 spin_lock(&inode_lock);
918 inode = find_inode_fast(sb, head, ino);
921 spin_unlock(&inode_lock);
922 wait_on_inode(inode);
925 spin_unlock(&inode_lock);
930 * ilookup5_nowait - search for an inode in the inode cache
931 * @sb: super block of file system to search
932 * @hashval: hash value (usually inode number) to search for
933 * @test: callback used for comparisons between inodes
934 * @data: opaque data pointer to pass to @test
936 * ilookup5() uses ifind() to search for the inode specified by @hashval and
937 * @data in the inode cache. This is a generalized version of ilookup() for
938 * file systems where the inode number is not sufficient for unique
939 * identification of an inode.
941 * If the inode is in the cache, the inode is returned with an incremented
942 * reference count. Note, the inode lock is not waited upon so you have to be
943 * very careful what you do with the returned inode. You probably should be
944 * using ilookup5() instead.
946 * Otherwise NULL is returned.
948 * Note, @test is called with the inode_lock held, so can't sleep.
950 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
951 int (*test)(struct inode *, void *), void *data)
953 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
955 return ifind(sb, head, test, data, 0);
957 EXPORT_SYMBOL(ilookup5_nowait);
960 * ilookup5 - search for an inode in the inode cache
961 * @sb: super block of file system to search
962 * @hashval: hash value (usually inode number) to search for
963 * @test: callback used for comparisons between inodes
964 * @data: opaque data pointer to pass to @test
966 * ilookup5() uses ifind() to search for the inode specified by @hashval and
967 * @data in the inode cache. This is a generalized version of ilookup() for
968 * file systems where the inode number is not sufficient for unique
969 * identification of an inode.
971 * If the inode is in the cache, the inode lock is waited upon and the inode is
972 * returned with an incremented reference count.
974 * Otherwise NULL is returned.
976 * Note, @test is called with the inode_lock held, so can't sleep.
978 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
979 int (*test)(struct inode *, void *), void *data)
981 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
983 return ifind(sb, head, test, data, 1);
985 EXPORT_SYMBOL(ilookup5);
988 * ilookup - search for an inode in the inode cache
989 * @sb: super block of file system to search
990 * @ino: inode number to search for
992 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
993 * This is for file systems where the inode number is sufficient for unique
994 * identification of an inode.
996 * If the inode is in the cache, the inode is returned with an incremented
999 * Otherwise NULL is returned.
1001 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1003 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1005 return ifind_fast(sb, head, ino);
1007 EXPORT_SYMBOL(ilookup);
1010 * iget5_locked - obtain an inode from a mounted file system
1011 * @sb: super block of file system
1012 * @hashval: hash value (usually inode number) to get
1013 * @test: callback used for comparisons between inodes
1014 * @set: callback used to initialize a new struct inode
1015 * @data: opaque data pointer to pass to @test and @set
1017 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1018 * and @data in the inode cache and if present it is returned with an increased
1019 * reference count. This is a generalized version of iget_locked() for file
1020 * systems where the inode number is not sufficient for unique identification
1023 * If the inode is not in cache, get_new_inode() is called to allocate a new
1024 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1025 * file system gets to fill it in before unlocking it via unlock_new_inode().
1027 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1029 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1030 int (*test)(struct inode *, void *),
1031 int (*set)(struct inode *, void *), void *data)
1033 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1034 struct inode *inode;
1036 inode = ifind(sb, head, test, data, 1);
1040 * get_new_inode() will do the right thing, re-trying the search
1041 * in case it had to block at any point.
1043 return get_new_inode(sb, head, test, set, data);
1045 EXPORT_SYMBOL(iget5_locked);
1048 * iget_locked - obtain an inode from a mounted file system
1049 * @sb: super block of file system
1050 * @ino: inode number to get
1052 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1053 * the inode cache and if present it is returned with an increased reference
1054 * count. This is for file systems where the inode number is sufficient for
1055 * unique identification of an inode.
1057 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1058 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1059 * The file system gets to fill it in before unlocking it via
1060 * unlock_new_inode().
1062 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1064 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1065 struct inode *inode;
1067 inode = ifind_fast(sb, head, ino);
1071 * get_new_inode_fast() will do the right thing, re-trying the search
1072 * in case it had to block at any point.
1074 return get_new_inode_fast(sb, head, ino);
1076 EXPORT_SYMBOL(iget_locked);
1078 int insert_inode_locked(struct inode *inode)
1080 struct super_block *sb = inode->i_sb;
1081 ino_t ino = inode->i_ino;
1082 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1084 inode->i_state |= I_NEW;
1086 struct hlist_node *node;
1087 struct inode *old = NULL;
1088 spin_lock(&inode_lock);
1089 hlist_for_each_entry(old, node, head, i_hash) {
1090 if (old->i_ino != ino)
1092 if (old->i_sb != sb)
1094 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1098 if (likely(!node)) {
1099 hlist_add_head(&inode->i_hash, head);
1100 spin_unlock(&inode_lock);
1104 spin_unlock(&inode_lock);
1106 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1113 EXPORT_SYMBOL(insert_inode_locked);
1115 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1116 int (*test)(struct inode *, void *), void *data)
1118 struct super_block *sb = inode->i_sb;
1119 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1121 inode->i_state |= I_NEW;
1124 struct hlist_node *node;
1125 struct inode *old = NULL;
1127 spin_lock(&inode_lock);
1128 hlist_for_each_entry(old, node, head, i_hash) {
1129 if (old->i_sb != sb)
1131 if (!test(old, data))
1133 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1137 if (likely(!node)) {
1138 hlist_add_head(&inode->i_hash, head);
1139 spin_unlock(&inode_lock);
1143 spin_unlock(&inode_lock);
1145 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1152 EXPORT_SYMBOL(insert_inode_locked4);
1155 * __insert_inode_hash - hash an inode
1156 * @inode: unhashed inode
1157 * @hashval: unsigned long value used to locate this object in the
1160 * Add an inode to the inode hash for this superblock.
1162 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1164 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1165 spin_lock(&inode_lock);
1166 hlist_add_head(&inode->i_hash, head);
1167 spin_unlock(&inode_lock);
1169 EXPORT_SYMBOL(__insert_inode_hash);
1172 * remove_inode_hash - remove an inode from the hash
1173 * @inode: inode to unhash
1175 * Remove an inode from the superblock.
1177 void remove_inode_hash(struct inode *inode)
1179 spin_lock(&inode_lock);
1180 hlist_del_init(&inode->i_hash);
1181 spin_unlock(&inode_lock);
1183 EXPORT_SYMBOL(remove_inode_hash);
1186 * Tell the filesystem that this inode is no longer of any interest and should
1187 * be completely destroyed.
1189 * We leave the inode in the inode hash table until *after* the filesystem's
1190 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1191 * instigate) will always find up-to-date information either in the hash or on
1194 * I_FREEING is set so that no-one will take a new reference to the inode while
1195 * it is being deleted.
1197 void generic_delete_inode(struct inode *inode)
1199 const struct super_operations *op = inode->i_sb->s_op;
1201 list_del_init(&inode->i_list);
1202 list_del_init(&inode->i_sb_list);
1203 WARN_ON(inode->i_state & I_NEW);
1204 inode->i_state |= I_FREEING;
1205 inodes_stat.nr_inodes--;
1206 spin_unlock(&inode_lock);
1208 security_inode_delete(inode);
1210 if (op->delete_inode) {
1211 void (*delete)(struct inode *) = op->delete_inode;
1212 /* Filesystems implementing their own
1213 * s_op->delete_inode are required to call
1214 * truncate_inode_pages and clear_inode()
1218 truncate_inode_pages(&inode->i_data, 0);
1221 spin_lock(&inode_lock);
1222 hlist_del_init(&inode->i_hash);
1223 spin_unlock(&inode_lock);
1224 wake_up_inode(inode);
1225 BUG_ON(inode->i_state != I_CLEAR);
1226 destroy_inode(inode);
1228 EXPORT_SYMBOL(generic_delete_inode);
1231 * generic_detach_inode - remove inode from inode lists
1232 * @inode: inode to remove
1234 * Remove inode from inode lists, write it if it's dirty. This is just an
1235 * internal VFS helper exported for hugetlbfs. Do not use!
1237 * Returns 1 if inode should be completely destroyed.
1239 int generic_detach_inode(struct inode *inode)
1241 struct super_block *sb = inode->i_sb;
1243 if (!hlist_unhashed(&inode->i_hash)) {
1244 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1245 list_move(&inode->i_list, &inode_unused);
1246 inodes_stat.nr_unused++;
1247 if (sb->s_flags & MS_ACTIVE) {
1248 spin_unlock(&inode_lock);
1251 WARN_ON(inode->i_state & I_NEW);
1252 inode->i_state |= I_WILL_FREE;
1253 spin_unlock(&inode_lock);
1254 write_inode_now(inode, 1);
1255 spin_lock(&inode_lock);
1256 WARN_ON(inode->i_state & I_NEW);
1257 inode->i_state &= ~I_WILL_FREE;
1258 inodes_stat.nr_unused--;
1259 hlist_del_init(&inode->i_hash);
1261 list_del_init(&inode->i_list);
1262 list_del_init(&inode->i_sb_list);
1263 WARN_ON(inode->i_state & I_NEW);
1264 inode->i_state |= I_FREEING;
1265 inodes_stat.nr_inodes--;
1266 spin_unlock(&inode_lock);
1269 EXPORT_SYMBOL_GPL(generic_detach_inode);
1271 static void generic_forget_inode(struct inode *inode)
1273 if (!generic_detach_inode(inode))
1275 if (inode->i_data.nrpages)
1276 truncate_inode_pages(&inode->i_data, 0);
1278 wake_up_inode(inode);
1279 destroy_inode(inode);
1283 * Normal UNIX filesystem behaviour: delete the
1284 * inode when the usage count drops to zero, and
1287 void generic_drop_inode(struct inode *inode)
1289 if (!inode->i_nlink)
1290 generic_delete_inode(inode);
1292 generic_forget_inode(inode);
1294 EXPORT_SYMBOL_GPL(generic_drop_inode);
1297 * Called when we're dropping the last reference
1300 * Call the FS "drop()" function, defaulting to
1301 * the legacy UNIX filesystem behaviour..
1303 * NOTE! NOTE! NOTE! We're called with the inode lock
1304 * held, and the drop function is supposed to release
1307 static inline void iput_final(struct inode *inode)
1309 const struct super_operations *op = inode->i_sb->s_op;
1310 void (*drop)(struct inode *) = generic_drop_inode;
1312 if (op && op->drop_inode)
1313 drop = op->drop_inode;
1318 * iput - put an inode
1319 * @inode: inode to put
1321 * Puts an inode, dropping its usage count. If the inode use count hits
1322 * zero, the inode is then freed and may also be destroyed.
1324 * Consequently, iput() can sleep.
1326 void iput(struct inode *inode)
1329 BUG_ON(inode->i_state == I_CLEAR);
1331 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1335 EXPORT_SYMBOL(iput);
1338 * bmap - find a block number in a file
1339 * @inode: inode of file
1340 * @block: block to find
1342 * Returns the block number on the device holding the inode that
1343 * is the disk block number for the block of the file requested.
1344 * That is, asked for block 4 of inode 1 the function will return the
1345 * disk block relative to the disk start that holds that block of the
1348 sector_t bmap(struct inode *inode, sector_t block)
1351 if (inode->i_mapping->a_ops->bmap)
1352 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1355 EXPORT_SYMBOL(bmap);
1358 * With relative atime, only update atime if the previous atime is
1359 * earlier than either the ctime or mtime or if at least a day has
1360 * passed since the last atime update.
1362 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1363 struct timespec now)
1366 if (!(mnt->mnt_flags & MNT_RELATIME))
1369 * Is mtime younger than atime? If yes, update atime:
1371 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1374 * Is ctime younger than atime? If yes, update atime:
1376 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1380 * Is the previous atime value older than a day? If yes,
1383 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1386 * Good, we can skip the atime update:
1392 * touch_atime - update the access time
1393 * @mnt: mount the inode is accessed on
1394 * @dentry: dentry accessed
1396 * Update the accessed time on an inode and mark it for writeback.
1397 * This function automatically handles read only file systems and media,
1398 * as well as the "noatime" flag and inode specific "noatime" markers.
1400 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1402 struct inode *inode = dentry->d_inode;
1403 struct timespec now;
1405 if (inode->i_flags & S_NOATIME)
1407 if (IS_NOATIME(inode))
1409 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1412 if (mnt->mnt_flags & MNT_NOATIME)
1414 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1417 now = current_fs_time(inode->i_sb);
1419 if (!relatime_need_update(mnt, inode, now))
1422 if (timespec_equal(&inode->i_atime, &now))
1425 if (mnt_want_write(mnt))
1428 inode->i_atime = now;
1429 mark_inode_dirty_sync(inode);
1430 mnt_drop_write(mnt);
1432 EXPORT_SYMBOL(touch_atime);
1435 * file_update_time - update mtime and ctime time
1436 * @file: file accessed
1438 * Update the mtime and ctime members of an inode and mark the inode
1439 * for writeback. Note that this function is meant exclusively for
1440 * usage in the file write path of filesystems, and filesystems may
1441 * choose to explicitly ignore update via this function with the
1442 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1443 * timestamps are handled by the server.
1446 void file_update_time(struct file *file)
1448 struct inode *inode = file->f_path.dentry->d_inode;
1449 struct timespec now;
1450 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1452 /* First try to exhaust all avenues to not sync */
1453 if (IS_NOCMTIME(inode))
1456 now = current_fs_time(inode->i_sb);
1457 if (!timespec_equal(&inode->i_mtime, &now))
1460 if (!timespec_equal(&inode->i_ctime, &now))
1463 if (IS_I_VERSION(inode))
1464 sync_it |= S_VERSION;
1469 /* Finally allowed to write? Takes lock. */
1470 if (mnt_want_write_file(file))
1473 /* Only change inode inside the lock region */
1474 if (sync_it & S_VERSION)
1475 inode_inc_iversion(inode);
1476 if (sync_it & S_CTIME)
1477 inode->i_ctime = now;
1478 if (sync_it & S_MTIME)
1479 inode->i_mtime = now;
1480 mark_inode_dirty_sync(inode);
1481 mnt_drop_write(file->f_path.mnt);
1483 EXPORT_SYMBOL(file_update_time);
1485 int inode_needs_sync(struct inode *inode)
1489 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1493 EXPORT_SYMBOL(inode_needs_sync);
1495 int inode_wait(void *word)
1500 EXPORT_SYMBOL(inode_wait);
1503 * If we try to find an inode in the inode hash while it is being
1504 * deleted, we have to wait until the filesystem completes its
1505 * deletion before reporting that it isn't found. This function waits
1506 * until the deletion _might_ have completed. Callers are responsible
1507 * to recheck inode state.
1509 * It doesn't matter if I_NEW is not set initially, a call to
1510 * wake_up_inode() after removing from the hash list will DTRT.
1512 * This is called with inode_lock held.
1514 static void __wait_on_freeing_inode(struct inode *inode)
1516 wait_queue_head_t *wq;
1517 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1518 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1519 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1520 spin_unlock(&inode_lock);
1522 finish_wait(wq, &wait.wait);
1523 spin_lock(&inode_lock);
1526 static __initdata unsigned long ihash_entries;
1527 static int __init set_ihash_entries(char *str)
1531 ihash_entries = simple_strtoul(str, &str, 0);
1534 __setup("ihash_entries=", set_ihash_entries);
1537 * Initialize the waitqueues and inode hash table.
1539 void __init inode_init_early(void)
1543 /* If hashes are distributed across NUMA nodes, defer
1544 * hash allocation until vmalloc space is available.
1550 alloc_large_system_hash("Inode-cache",
1551 sizeof(struct hlist_head),
1559 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1560 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1563 void __init inode_init(void)
1567 /* inode slab cache */
1568 inode_cachep = kmem_cache_create("inode_cache",
1569 sizeof(struct inode),
1571 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1574 register_shrinker(&icache_shrinker);
1576 /* Hash may have been set up in inode_init_early */
1581 alloc_large_system_hash("Inode-cache",
1582 sizeof(struct hlist_head),
1590 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1591 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1594 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1596 inode->i_mode = mode;
1597 if (S_ISCHR(mode)) {
1598 inode->i_fop = &def_chr_fops;
1599 inode->i_rdev = rdev;
1600 } else if (S_ISBLK(mode)) {
1601 inode->i_fop = &def_blk_fops;
1602 inode->i_rdev = rdev;
1603 } else if (S_ISFIFO(mode))
1604 inode->i_fop = &def_fifo_fops;
1605 else if (S_ISSOCK(mode))
1606 inode->i_fop = &bad_sock_fops;
1608 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1609 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1612 EXPORT_SYMBOL(init_special_inode);