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/hash.h>
18 #include <linux/swap.h>
19 #include <linux/security.h>
20 #include <linux/ima.h>
21 #include <linux/pagemap.h>
22 #include <linux/cdev.h>
23 #include <linux/bootmem.h>
24 #include <linux/inotify.h>
25 #include <linux/mount.h>
28 * This is needed for the following functions:
30 * - invalidate_inode_buffers
33 * FIXME: remove all knowledge of the buffer layer from this file
35 #include <linux/buffer_head.h>
38 * New inode.c implementation.
40 * This implementation has the basic premise of trying
41 * to be extremely low-overhead and SMP-safe, yet be
42 * simple enough to be "obviously correct".
47 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
49 /* #define INODE_PARANOIA 1 */
50 /* #define INODE_DEBUG 1 */
53 * Inode lookup is no longer as critical as it used to be:
54 * most of the lookups are going to be through the dcache.
56 #define I_HASHBITS i_hash_shift
57 #define I_HASHMASK i_hash_mask
59 static unsigned int i_hash_mask __read_mostly;
60 static unsigned int i_hash_shift __read_mostly;
63 * Each inode can be on two separate lists. One is
64 * the hash list of the inode, used for lookups. The
65 * other linked list is the "type" list:
66 * "in_use" - valid inode, i_count > 0, i_nlink > 0
67 * "dirty" - as "in_use" but also dirty
68 * "unused" - valid inode, i_count = 0
70 * A "dirty" list is maintained for each super block,
71 * allowing for low-overhead inode sync() operations.
74 LIST_HEAD(inode_in_use);
75 LIST_HEAD(inode_unused);
76 static struct hlist_head *inode_hashtable __read_mostly;
79 * A simple spinlock to protect the list manipulations.
81 * NOTE! You also have to own the lock if you change
82 * the i_state of an inode while it is in use..
84 DEFINE_SPINLOCK(inode_lock);
87 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
88 * icache shrinking path, and the umount path. Without this exclusion,
89 * by the time prune_icache calls iput for the inode whose pages it has
90 * been invalidating, or by the time it calls clear_inode & destroy_inode
91 * from its final dispose_list, the struct super_block they refer to
92 * (for inode->i_sb->s_op) may already have been freed and reused.
94 static DEFINE_MUTEX(iprune_mutex);
97 * Statistics gathering..
99 struct inodes_stat_t inodes_stat;
101 static struct kmem_cache * inode_cachep __read_mostly;
103 static void wake_up_inode(struct inode *inode)
106 * Prevent speculative execution through spin_unlock(&inode_lock);
109 wake_up_bit(&inode->i_state, __I_LOCK);
113 * inode_init_always - perform inode structure intialisation
114 * @sb - superblock inode belongs to.
115 * @inode - inode to initialise
117 * These are initializations that need to be done on every inode
118 * allocation as the fields are not initialised by slab allocation.
120 struct inode *inode_init_always(struct super_block *sb, struct inode *inode)
122 static const struct address_space_operations empty_aops;
123 static struct inode_operations empty_iops;
124 static const struct file_operations empty_fops;
126 struct address_space * const mapping = &inode->i_data;
129 inode->i_blkbits = sb->s_blocksize_bits;
131 atomic_set(&inode->i_count, 1);
132 inode->i_op = &empty_iops;
133 inode->i_fop = &empty_fops;
135 atomic_set(&inode->i_writecount, 0);
139 inode->i_generation = 0;
141 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
143 inode->i_pipe = NULL;
144 inode->i_bdev = NULL;
145 inode->i_cdev = NULL;
147 inode->dirtied_when = 0;
149 if (security_inode_alloc(inode))
152 /* allocate and initialize an i_integrity */
153 if (ima_inode_alloc(inode))
154 goto out_free_security;
156 spin_lock_init(&inode->i_lock);
157 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
159 mutex_init(&inode->i_mutex);
160 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
162 init_rwsem(&inode->i_alloc_sem);
163 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
165 mapping->a_ops = &empty_aops;
166 mapping->host = inode;
168 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_PAGECACHE);
169 mapping->assoc_mapping = NULL;
170 mapping->backing_dev_info = &default_backing_dev_info;
171 mapping->writeback_index = 0;
174 * If the block_device provides a backing_dev_info for client
175 * inodes then use that. Otherwise the inode share the bdev's
179 struct backing_dev_info *bdi;
181 bdi = sb->s_bdev->bd_inode_backing_dev_info;
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;
192 security_inode_free(inode);
194 if (inode->i_sb->s_op->destroy_inode)
195 inode->i_sb->s_op->destroy_inode(inode);
197 kmem_cache_free(inode_cachep, (inode));
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);
212 return inode_init_always(sb, inode);
216 void destroy_inode(struct inode *inode)
218 BUG_ON(inode_has_buffers(inode));
219 security_inode_free(inode);
220 if (inode->i_sb->s_op->destroy_inode)
221 inode->i_sb->s_op->destroy_inode(inode);
223 kmem_cache_free(inode_cachep, (inode));
225 EXPORT_SYMBOL(destroy_inode);
229 * These are initializations that only need to be done
230 * once, because the fields are idempotent across use
231 * of the inode, so let the slab aware of that.
233 void inode_init_once(struct inode *inode)
235 memset(inode, 0, sizeof(*inode));
236 INIT_HLIST_NODE(&inode->i_hash);
237 INIT_LIST_HEAD(&inode->i_dentry);
238 INIT_LIST_HEAD(&inode->i_devices);
239 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
240 spin_lock_init(&inode->i_data.tree_lock);
241 spin_lock_init(&inode->i_data.i_mmap_lock);
242 INIT_LIST_HEAD(&inode->i_data.private_list);
243 spin_lock_init(&inode->i_data.private_lock);
244 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
245 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
246 i_size_ordered_init(inode);
247 #ifdef CONFIG_INOTIFY
248 INIT_LIST_HEAD(&inode->inotify_watches);
249 mutex_init(&inode->inotify_mutex);
253 EXPORT_SYMBOL(inode_init_once);
255 static void init_once(void *foo)
257 struct inode * inode = (struct inode *) foo;
259 inode_init_once(inode);
263 * inode_lock must be held
265 void __iget(struct inode * inode)
267 if (atomic_read(&inode->i_count)) {
268 atomic_inc(&inode->i_count);
271 atomic_inc(&inode->i_count);
272 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
273 list_move(&inode->i_list, &inode_in_use);
274 inodes_stat.nr_unused--;
278 * clear_inode - clear an inode
279 * @inode: inode to clear
281 * This is called by the filesystem to tell us
282 * that the inode is no longer useful. We just
283 * terminate it with extreme prejudice.
285 void clear_inode(struct inode *inode)
288 invalidate_inode_buffers(inode);
290 BUG_ON(inode->i_data.nrpages);
291 BUG_ON(!(inode->i_state & I_FREEING));
292 BUG_ON(inode->i_state & I_CLEAR);
293 inode_sync_wait(inode);
295 if (inode->i_sb->s_op->clear_inode)
296 inode->i_sb->s_op->clear_inode(inode);
297 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
299 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
301 inode->i_state = I_CLEAR;
304 EXPORT_SYMBOL(clear_inode);
307 * dispose_list - dispose of the contents of a local list
308 * @head: the head of the list to free
310 * Dispose-list gets a local list with local inodes in it, so it doesn't
311 * need to worry about list corruption and SMP locks.
313 static void dispose_list(struct list_head *head)
317 while (!list_empty(head)) {
320 inode = list_first_entry(head, struct inode, i_list);
321 list_del(&inode->i_list);
323 if (inode->i_data.nrpages)
324 truncate_inode_pages(&inode->i_data, 0);
327 spin_lock(&inode_lock);
328 hlist_del_init(&inode->i_hash);
329 list_del_init(&inode->i_sb_list);
330 spin_unlock(&inode_lock);
332 wake_up_inode(inode);
333 destroy_inode(inode);
336 spin_lock(&inode_lock);
337 inodes_stat.nr_inodes -= nr_disposed;
338 spin_unlock(&inode_lock);
342 * Invalidate all inodes for a device.
344 static int invalidate_list(struct list_head *head, struct list_head *dispose)
346 struct list_head *next;
347 int busy = 0, count = 0;
351 struct list_head * tmp = next;
352 struct inode * inode;
355 * We can reschedule here without worrying about the list's
356 * consistency because the per-sb list of inodes must not
357 * change during umount anymore, and because iprune_mutex keeps
358 * shrink_icache_memory() away.
360 cond_resched_lock(&inode_lock);
365 inode = list_entry(tmp, struct inode, i_sb_list);
366 invalidate_inode_buffers(inode);
367 if (!atomic_read(&inode->i_count)) {
368 list_move(&inode->i_list, dispose);
369 inode->i_state |= I_FREEING;
375 /* only unused inodes may be cached with i_count zero */
376 inodes_stat.nr_unused -= count;
381 * invalidate_inodes - discard the inodes on a device
384 * Discard all of the inodes for a given superblock. If the discard
385 * fails because there are busy inodes then a non zero value is returned.
386 * If the discard is successful all the inodes have been discarded.
388 int invalidate_inodes(struct super_block * sb)
391 LIST_HEAD(throw_away);
393 mutex_lock(&iprune_mutex);
394 spin_lock(&inode_lock);
395 inotify_unmount_inodes(&sb->s_inodes);
396 busy = invalidate_list(&sb->s_inodes, &throw_away);
397 spin_unlock(&inode_lock);
399 dispose_list(&throw_away);
400 mutex_unlock(&iprune_mutex);
405 EXPORT_SYMBOL(invalidate_inodes);
407 static int can_unuse(struct inode *inode)
411 if (inode_has_buffers(inode))
413 if (atomic_read(&inode->i_count))
415 if (inode->i_data.nrpages)
421 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
422 * a temporary list and then are freed outside inode_lock by dispose_list().
424 * Any inodes which are pinned purely because of attached pagecache have their
425 * pagecache removed. We expect the final iput() on that inode to add it to
426 * the front of the inode_unused list. So look for it there and if the
427 * inode is still freeable, proceed. The right inode is found 99.9% of the
428 * time in testing on a 4-way.
430 * If the inode has metadata buffers attached to mapping->private_list then
431 * try to remove them.
433 static void prune_icache(int nr_to_scan)
438 unsigned long reap = 0;
440 mutex_lock(&iprune_mutex);
441 spin_lock(&inode_lock);
442 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
445 if (list_empty(&inode_unused))
448 inode = list_entry(inode_unused.prev, struct inode, i_list);
450 if (inode->i_state || atomic_read(&inode->i_count)) {
451 list_move(&inode->i_list, &inode_unused);
454 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
456 spin_unlock(&inode_lock);
457 if (remove_inode_buffers(inode))
458 reap += invalidate_mapping_pages(&inode->i_data,
461 spin_lock(&inode_lock);
463 if (inode != list_entry(inode_unused.next,
464 struct inode, i_list))
465 continue; /* wrong inode or list_empty */
466 if (!can_unuse(inode))
469 list_move(&inode->i_list, &freeable);
470 inode->i_state |= I_FREEING;
473 inodes_stat.nr_unused -= nr_pruned;
474 if (current_is_kswapd())
475 __count_vm_events(KSWAPD_INODESTEAL, reap);
477 __count_vm_events(PGINODESTEAL, reap);
478 spin_unlock(&inode_lock);
480 dispose_list(&freeable);
481 mutex_unlock(&iprune_mutex);
485 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
486 * "unused" means that no dentries are referring to the inodes: the files are
487 * not open and the dcache references to those inodes have already been
490 * This function is passed the number of inodes to scan, and it returns the
491 * total number of remaining possibly-reclaimable inodes.
493 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
497 * Nasty deadlock avoidance. We may hold various FS locks,
498 * and we don't want to recurse into the FS that called us
499 * in clear_inode() and friends..
501 if (!(gfp_mask & __GFP_FS))
505 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
508 static struct shrinker icache_shrinker = {
509 .shrink = shrink_icache_memory,
510 .seeks = DEFAULT_SEEKS,
513 static void __wait_on_freeing_inode(struct inode *inode);
515 * Called with the inode lock held.
516 * NOTE: we are not increasing the inode-refcount, you must call __iget()
517 * by hand after calling find_inode now! This simplifies iunique and won't
518 * add any additional branch in the common code.
520 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
522 struct hlist_node *node;
523 struct inode * inode = NULL;
526 hlist_for_each_entry(inode, node, head, i_hash) {
527 if (inode->i_sb != sb)
529 if (!test(inode, data))
531 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
532 __wait_on_freeing_inode(inode);
537 return node ? inode : NULL;
541 * find_inode_fast is the fast path version of find_inode, see the comment at
542 * iget_locked for details.
544 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
546 struct hlist_node *node;
547 struct inode * inode = NULL;
550 hlist_for_each_entry(inode, node, head, i_hash) {
551 if (inode->i_ino != ino)
553 if (inode->i_sb != sb)
555 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
556 __wait_on_freeing_inode(inode);
561 return node ? inode : NULL;
564 static unsigned long hash(struct super_block *sb, unsigned long hashval)
568 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
570 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
571 return tmp & I_HASHMASK;
575 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
578 inodes_stat.nr_inodes++;
579 list_add(&inode->i_list, &inode_in_use);
580 list_add(&inode->i_sb_list, &sb->s_inodes);
582 hlist_add_head(&inode->i_hash, head);
586 * inode_add_to_lists - add a new inode to relevant lists
587 * @sb - superblock inode belongs to.
588 * @inode - inode to mark in use
590 * When an inode is allocated it needs to be accounted for, added to the in use
591 * list, the owning superblock and the inode hash. This needs to be done under
592 * the inode_lock, so export a function to do this rather than the inode lock
593 * itself. We calculate the hash list to add to here so it is all internal
594 * which requires the caller to have already set up the inode number in the
597 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
599 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
601 spin_lock(&inode_lock);
602 __inode_add_to_lists(sb, head, inode);
603 spin_unlock(&inode_lock);
605 EXPORT_SYMBOL_GPL(inode_add_to_lists);
608 * new_inode - obtain an inode
611 * Allocates a new inode for given superblock. The default gfp_mask
612 * for allocations related to inode->i_mapping is GFP_HIGHUSER_PAGECACHE.
613 * If HIGHMEM pages are unsuitable or it is known that pages allocated
614 * for the page cache are not reclaimable or migratable,
615 * mapping_set_gfp_mask() must be called with suitable flags on the
616 * newly created inode's mapping
619 struct inode *new_inode(struct super_block *sb)
622 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
623 * error if st_ino won't fit in target struct field. Use 32bit counter
624 * here to attempt to avoid that.
626 static unsigned int last_ino;
627 struct inode * inode;
629 spin_lock_prefetch(&inode_lock);
631 inode = alloc_inode(sb);
633 spin_lock(&inode_lock);
634 __inode_add_to_lists(sb, NULL, inode);
635 inode->i_ino = ++last_ino;
637 spin_unlock(&inode_lock);
642 EXPORT_SYMBOL(new_inode);
644 void unlock_new_inode(struct inode *inode)
646 #ifdef CONFIG_DEBUG_LOCK_ALLOC
647 if (inode->i_mode & S_IFDIR) {
648 struct file_system_type *type = inode->i_sb->s_type;
651 * ensure nobody is actually holding i_mutex
653 mutex_destroy(&inode->i_mutex);
654 mutex_init(&inode->i_mutex);
655 lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key);
659 * This is special! We do not need the spinlock
660 * when clearing I_LOCK, because we're guaranteed
661 * that nobody else tries to do anything about the
662 * state of the inode when it is locked, as we
663 * just created it (so there can be no old holders
664 * that haven't tested I_LOCK).
666 inode->i_state &= ~(I_LOCK|I_NEW);
667 wake_up_inode(inode);
670 EXPORT_SYMBOL(unlock_new_inode);
673 * This is called without the inode lock held.. Be careful.
675 * We no longer cache the sb_flags in i_flags - see fs.h
676 * -- rmk@arm.uk.linux.org
678 static struct inode * get_new_inode(struct super_block *sb, struct hlist_head *head, int (*test)(struct inode *, void *), int (*set)(struct inode *, void *), void *data)
680 struct inode * inode;
682 inode = alloc_inode(sb);
686 spin_lock(&inode_lock);
687 /* We released the lock, so.. */
688 old = find_inode(sb, head, test, data);
690 if (set(inode, data))
693 __inode_add_to_lists(sb, head, inode);
694 inode->i_state = I_LOCK|I_NEW;
695 spin_unlock(&inode_lock);
697 /* Return the locked inode with I_NEW set, the
698 * caller is responsible for filling in the contents
704 * Uhhuh, somebody else created the same inode under
705 * us. Use the old inode instead of the one we just
709 spin_unlock(&inode_lock);
710 destroy_inode(inode);
712 wait_on_inode(inode);
717 spin_unlock(&inode_lock);
718 destroy_inode(inode);
723 * get_new_inode_fast is the fast path version of get_new_inode, see the
724 * comment at iget_locked for details.
726 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
728 struct inode * inode;
730 inode = alloc_inode(sb);
734 spin_lock(&inode_lock);
735 /* We released the lock, so.. */
736 old = find_inode_fast(sb, head, ino);
739 __inode_add_to_lists(sb, head, inode);
740 inode->i_state = I_LOCK|I_NEW;
741 spin_unlock(&inode_lock);
743 /* Return the locked inode with I_NEW set, the
744 * caller is responsible for filling in the contents
750 * Uhhuh, somebody else created the same inode under
751 * us. Use the old inode instead of the one we just
755 spin_unlock(&inode_lock);
756 destroy_inode(inode);
758 wait_on_inode(inode);
764 * iunique - get a unique inode number
766 * @max_reserved: highest reserved inode number
768 * Obtain an inode number that is unique on the system for a given
769 * superblock. This is used by file systems that have no natural
770 * permanent inode numbering system. An inode number is returned that
771 * is higher than the reserved limit but unique.
774 * With a large number of inodes live on the file system this function
775 * currently becomes quite slow.
777 ino_t iunique(struct super_block *sb, ino_t max_reserved)
780 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
781 * error if st_ino won't fit in target struct field. Use 32bit counter
782 * here to attempt to avoid that.
784 static unsigned int counter;
786 struct hlist_head *head;
789 spin_lock(&inode_lock);
791 if (counter <= max_reserved)
792 counter = max_reserved + 1;
794 head = inode_hashtable + hash(sb, res);
795 inode = find_inode_fast(sb, head, res);
796 } while (inode != NULL);
797 spin_unlock(&inode_lock);
801 EXPORT_SYMBOL(iunique);
803 struct inode *igrab(struct inode *inode)
805 spin_lock(&inode_lock);
806 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
810 * Handle the case where s_op->clear_inode is not been
811 * called yet, and somebody is calling igrab
812 * while the inode is getting freed.
815 spin_unlock(&inode_lock);
819 EXPORT_SYMBOL(igrab);
822 * ifind - internal function, you want ilookup5() or iget5().
823 * @sb: super block of file system to search
824 * @head: the head of the list to search
825 * @test: callback used for comparisons between inodes
826 * @data: opaque data pointer to pass to @test
827 * @wait: if true wait for the inode to be unlocked, if false do not
829 * ifind() searches for the inode specified by @data in the inode
830 * cache. This is a generalized version of ifind_fast() for file systems where
831 * the inode number is not sufficient for unique identification of an inode.
833 * If the inode is in the cache, the inode is returned with an incremented
836 * Otherwise NULL is returned.
838 * Note, @test is called with the inode_lock held, so can't sleep.
840 static struct inode *ifind(struct super_block *sb,
841 struct hlist_head *head, int (*test)(struct inode *, void *),
842 void *data, const int wait)
846 spin_lock(&inode_lock);
847 inode = find_inode(sb, head, test, data);
850 spin_unlock(&inode_lock);
852 wait_on_inode(inode);
855 spin_unlock(&inode_lock);
860 * ifind_fast - internal function, you want ilookup() or iget().
861 * @sb: super block of file system to search
862 * @head: head of the list to search
863 * @ino: inode number to search for
865 * ifind_fast() searches for the inode @ino in the inode cache. This is for
866 * file systems where the inode number is sufficient for unique identification
869 * If the inode is in the cache, the inode is returned with an incremented
872 * Otherwise NULL is returned.
874 static struct inode *ifind_fast(struct super_block *sb,
875 struct hlist_head *head, unsigned long ino)
879 spin_lock(&inode_lock);
880 inode = find_inode_fast(sb, head, ino);
883 spin_unlock(&inode_lock);
884 wait_on_inode(inode);
887 spin_unlock(&inode_lock);
892 * ilookup5_nowait - search for an inode in the inode cache
893 * @sb: super block of file system to search
894 * @hashval: hash value (usually inode number) to search for
895 * @test: callback used for comparisons between inodes
896 * @data: opaque data pointer to pass to @test
898 * ilookup5() uses ifind() to search for the inode specified by @hashval and
899 * @data in the inode cache. This is a generalized version of ilookup() for
900 * file systems where the inode number is not sufficient for unique
901 * identification of an inode.
903 * If the inode is in the cache, the inode is returned with an incremented
904 * reference count. Note, the inode lock is not waited upon so you have to be
905 * very careful what you do with the returned inode. You probably should be
906 * using ilookup5() instead.
908 * Otherwise NULL is returned.
910 * Note, @test is called with the inode_lock held, so can't sleep.
912 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
913 int (*test)(struct inode *, void *), void *data)
915 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
917 return ifind(sb, head, test, data, 0);
920 EXPORT_SYMBOL(ilookup5_nowait);
923 * ilookup5 - search for an inode in the inode cache
924 * @sb: super block of file system to search
925 * @hashval: hash value (usually inode number) to search for
926 * @test: callback used for comparisons between inodes
927 * @data: opaque data pointer to pass to @test
929 * ilookup5() uses ifind() to search for the inode specified by @hashval and
930 * @data in the inode cache. This is a generalized version of ilookup() for
931 * file systems where the inode number is not sufficient for unique
932 * identification of an inode.
934 * If the inode is in the cache, the inode lock is waited upon and the inode is
935 * returned with an incremented reference count.
937 * Otherwise NULL is returned.
939 * Note, @test is called with the inode_lock held, so can't sleep.
941 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
942 int (*test)(struct inode *, void *), void *data)
944 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
946 return ifind(sb, head, test, data, 1);
949 EXPORT_SYMBOL(ilookup5);
952 * ilookup - search for an inode in the inode cache
953 * @sb: super block of file system to search
954 * @ino: inode number to search for
956 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
957 * This is for file systems where the inode number is sufficient for unique
958 * identification of an inode.
960 * If the inode is in the cache, the inode is returned with an incremented
963 * Otherwise NULL is returned.
965 struct inode *ilookup(struct super_block *sb, unsigned long ino)
967 struct hlist_head *head = inode_hashtable + hash(sb, ino);
969 return ifind_fast(sb, head, ino);
972 EXPORT_SYMBOL(ilookup);
975 * iget5_locked - obtain an inode from a mounted file system
976 * @sb: super block of file system
977 * @hashval: hash value (usually inode number) to get
978 * @test: callback used for comparisons between inodes
979 * @set: callback used to initialize a new struct inode
980 * @data: opaque data pointer to pass to @test and @set
982 * iget5_locked() uses ifind() to search for the inode specified by @hashval
983 * and @data in the inode cache and if present it is returned with an increased
984 * reference count. This is a generalized version of iget_locked() for file
985 * systems where the inode number is not sufficient for unique identification
988 * If the inode is not in cache, get_new_inode() is called to allocate a new
989 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
990 * file system gets to fill it in before unlocking it via unlock_new_inode().
992 * Note both @test and @set are called with the inode_lock held, so can't sleep.
994 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
995 int (*test)(struct inode *, void *),
996 int (*set)(struct inode *, void *), void *data)
998 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1001 inode = ifind(sb, head, test, data, 1);
1005 * get_new_inode() will do the right thing, re-trying the search
1006 * in case it had to block at any point.
1008 return get_new_inode(sb, head, test, set, data);
1011 EXPORT_SYMBOL(iget5_locked);
1014 * iget_locked - obtain an inode from a mounted file system
1015 * @sb: super block of file system
1016 * @ino: inode number to get
1018 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1019 * the inode cache and if present it is returned with an increased reference
1020 * count. This is for file systems where the inode number is sufficient for
1021 * unique identification of an inode.
1023 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1024 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1025 * The file system gets to fill it in before unlocking it via
1026 * unlock_new_inode().
1028 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1030 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1031 struct inode *inode;
1033 inode = ifind_fast(sb, head, ino);
1037 * get_new_inode_fast() will do the right thing, re-trying the search
1038 * in case it had to block at any point.
1040 return get_new_inode_fast(sb, head, ino);
1043 EXPORT_SYMBOL(iget_locked);
1046 * __insert_inode_hash - hash an inode
1047 * @inode: unhashed inode
1048 * @hashval: unsigned long value used to locate this object in the
1051 * Add an inode to the inode hash for this superblock.
1053 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1055 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1056 spin_lock(&inode_lock);
1057 hlist_add_head(&inode->i_hash, head);
1058 spin_unlock(&inode_lock);
1061 EXPORT_SYMBOL(__insert_inode_hash);
1064 * remove_inode_hash - remove an inode from the hash
1065 * @inode: inode to unhash
1067 * Remove an inode from the superblock.
1069 void remove_inode_hash(struct inode *inode)
1071 spin_lock(&inode_lock);
1072 hlist_del_init(&inode->i_hash);
1073 spin_unlock(&inode_lock);
1076 EXPORT_SYMBOL(remove_inode_hash);
1079 * Tell the filesystem that this inode is no longer of any interest and should
1080 * be completely destroyed.
1082 * We leave the inode in the inode hash table until *after* the filesystem's
1083 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1084 * instigate) will always find up-to-date information either in the hash or on
1087 * I_FREEING is set so that no-one will take a new reference to the inode while
1088 * it is being deleted.
1090 void generic_delete_inode(struct inode *inode)
1092 const struct super_operations *op = inode->i_sb->s_op;
1094 list_del_init(&inode->i_list);
1095 list_del_init(&inode->i_sb_list);
1096 inode->i_state |= I_FREEING;
1097 inodes_stat.nr_inodes--;
1098 spin_unlock(&inode_lock);
1100 security_inode_delete(inode);
1102 if (op->delete_inode) {
1103 void (*delete)(struct inode *) = op->delete_inode;
1104 if (!is_bad_inode(inode))
1106 /* Filesystems implementing their own
1107 * s_op->delete_inode are required to call
1108 * truncate_inode_pages and clear_inode()
1112 truncate_inode_pages(&inode->i_data, 0);
1115 spin_lock(&inode_lock);
1116 hlist_del_init(&inode->i_hash);
1117 spin_unlock(&inode_lock);
1118 wake_up_inode(inode);
1119 BUG_ON(inode->i_state != I_CLEAR);
1120 destroy_inode(inode);
1123 EXPORT_SYMBOL(generic_delete_inode);
1125 static void generic_forget_inode(struct inode *inode)
1127 struct super_block *sb = inode->i_sb;
1129 if (!hlist_unhashed(&inode->i_hash)) {
1130 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1131 list_move(&inode->i_list, &inode_unused);
1132 inodes_stat.nr_unused++;
1133 if (sb->s_flags & MS_ACTIVE) {
1134 spin_unlock(&inode_lock);
1137 inode->i_state |= I_WILL_FREE;
1138 spin_unlock(&inode_lock);
1139 write_inode_now(inode, 1);
1140 spin_lock(&inode_lock);
1141 inode->i_state &= ~I_WILL_FREE;
1142 inodes_stat.nr_unused--;
1143 hlist_del_init(&inode->i_hash);
1145 list_del_init(&inode->i_list);
1146 list_del_init(&inode->i_sb_list);
1147 inode->i_state |= I_FREEING;
1148 inodes_stat.nr_inodes--;
1149 spin_unlock(&inode_lock);
1150 if (inode->i_data.nrpages)
1151 truncate_inode_pages(&inode->i_data, 0);
1153 wake_up_inode(inode);
1154 destroy_inode(inode);
1158 * Normal UNIX filesystem behaviour: delete the
1159 * inode when the usage count drops to zero, and
1162 void generic_drop_inode(struct inode *inode)
1164 if (!inode->i_nlink)
1165 generic_delete_inode(inode);
1167 generic_forget_inode(inode);
1170 EXPORT_SYMBOL_GPL(generic_drop_inode);
1173 * Called when we're dropping the last reference
1176 * Call the FS "drop()" function, defaulting to
1177 * the legacy UNIX filesystem behaviour..
1179 * NOTE! NOTE! NOTE! We're called with the inode lock
1180 * held, and the drop function is supposed to release
1183 static inline void iput_final(struct inode *inode)
1185 const struct super_operations *op = inode->i_sb->s_op;
1186 void (*drop)(struct inode *) = generic_drop_inode;
1188 if (op && op->drop_inode)
1189 drop = op->drop_inode;
1194 * iput - put an inode
1195 * @inode: inode to put
1197 * Puts an inode, dropping its usage count. If the inode use count hits
1198 * zero, the inode is then freed and may also be destroyed.
1200 * Consequently, iput() can sleep.
1202 void iput(struct inode *inode)
1205 BUG_ON(inode->i_state == I_CLEAR);
1207 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1212 EXPORT_SYMBOL(iput);
1215 * bmap - find a block number in a file
1216 * @inode: inode of file
1217 * @block: block to find
1219 * Returns the block number on the device holding the inode that
1220 * is the disk block number for the block of the file requested.
1221 * That is, asked for block 4 of inode 1 the function will return the
1222 * disk block relative to the disk start that holds that block of the
1225 sector_t bmap(struct inode * inode, sector_t block)
1228 if (inode->i_mapping->a_ops->bmap)
1229 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1232 EXPORT_SYMBOL(bmap);
1235 * touch_atime - update the access time
1236 * @mnt: mount the inode is accessed on
1237 * @dentry: dentry accessed
1239 * Update the accessed time on an inode and mark it for writeback.
1240 * This function automatically handles read only file systems and media,
1241 * as well as the "noatime" flag and inode specific "noatime" markers.
1243 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1245 struct inode *inode = dentry->d_inode;
1246 struct timespec now;
1248 if (mnt_want_write(mnt))
1250 if (inode->i_flags & S_NOATIME)
1252 if (IS_NOATIME(inode))
1254 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1257 if (mnt->mnt_flags & MNT_NOATIME)
1259 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1261 if (mnt->mnt_flags & MNT_RELATIME) {
1263 * With relative atime, only update atime if the previous
1264 * atime is earlier than either the ctime or mtime.
1266 if (timespec_compare(&inode->i_mtime, &inode->i_atime) < 0 &&
1267 timespec_compare(&inode->i_ctime, &inode->i_atime) < 0)
1271 now = current_fs_time(inode->i_sb);
1272 if (timespec_equal(&inode->i_atime, &now))
1275 inode->i_atime = now;
1276 mark_inode_dirty_sync(inode);
1278 mnt_drop_write(mnt);
1280 EXPORT_SYMBOL(touch_atime);
1283 * file_update_time - update mtime and ctime time
1284 * @file: file accessed
1286 * Update the mtime and ctime members of an inode and mark the inode
1287 * for writeback. Note that this function is meant exclusively for
1288 * usage in the file write path of filesystems, and filesystems may
1289 * choose to explicitly ignore update via this function with the
1290 * S_NOCTIME inode flag, e.g. for network filesystem where these
1291 * timestamps are handled by the server.
1294 void file_update_time(struct file *file)
1296 struct inode *inode = file->f_path.dentry->d_inode;
1297 struct timespec now;
1301 if (IS_NOCMTIME(inode))
1304 err = mnt_want_write(file->f_path.mnt);
1308 now = current_fs_time(inode->i_sb);
1309 if (!timespec_equal(&inode->i_mtime, &now)) {
1310 inode->i_mtime = now;
1314 if (!timespec_equal(&inode->i_ctime, &now)) {
1315 inode->i_ctime = now;
1319 if (IS_I_VERSION(inode)) {
1320 inode_inc_iversion(inode);
1325 mark_inode_dirty_sync(inode);
1326 mnt_drop_write(file->f_path.mnt);
1329 EXPORT_SYMBOL(file_update_time);
1331 int inode_needs_sync(struct inode *inode)
1335 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1340 EXPORT_SYMBOL(inode_needs_sync);
1342 int inode_wait(void *word)
1347 EXPORT_SYMBOL(inode_wait);
1350 * If we try to find an inode in the inode hash while it is being
1351 * deleted, we have to wait until the filesystem completes its
1352 * deletion before reporting that it isn't found. This function waits
1353 * until the deletion _might_ have completed. Callers are responsible
1354 * to recheck inode state.
1356 * It doesn't matter if I_LOCK is not set initially, a call to
1357 * wake_up_inode() after removing from the hash list will DTRT.
1359 * This is called with inode_lock held.
1361 static void __wait_on_freeing_inode(struct inode *inode)
1363 wait_queue_head_t *wq;
1364 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1365 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1366 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1367 spin_unlock(&inode_lock);
1369 finish_wait(wq, &wait.wait);
1370 spin_lock(&inode_lock);
1374 * We rarely want to lock two inodes that do not have a parent/child
1375 * relationship (such as directory, child inode) simultaneously. The
1376 * vast majority of file systems should be able to get along fine
1377 * without this. Do not use these functions except as a last resort.
1379 void inode_double_lock(struct inode *inode1, struct inode *inode2)
1381 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1383 mutex_lock(&inode1->i_mutex);
1385 mutex_lock(&inode2->i_mutex);
1389 if (inode1 < inode2) {
1390 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1391 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1393 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1394 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1397 EXPORT_SYMBOL(inode_double_lock);
1399 void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1402 mutex_unlock(&inode1->i_mutex);
1404 if (inode2 && inode2 != inode1)
1405 mutex_unlock(&inode2->i_mutex);
1407 EXPORT_SYMBOL(inode_double_unlock);
1409 static __initdata unsigned long ihash_entries;
1410 static int __init set_ihash_entries(char *str)
1414 ihash_entries = simple_strtoul(str, &str, 0);
1417 __setup("ihash_entries=", set_ihash_entries);
1420 * Initialize the waitqueues and inode hash table.
1422 void __init inode_init_early(void)
1426 /* If hashes are distributed across NUMA nodes, defer
1427 * hash allocation until vmalloc space is available.
1433 alloc_large_system_hash("Inode-cache",
1434 sizeof(struct hlist_head),
1442 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1443 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1446 void __init inode_init(void)
1450 /* inode slab cache */
1451 inode_cachep = kmem_cache_create("inode_cache",
1452 sizeof(struct inode),
1454 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1457 register_shrinker(&icache_shrinker);
1459 /* Hash may have been set up in inode_init_early */
1464 alloc_large_system_hash("Inode-cache",
1465 sizeof(struct hlist_head),
1473 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1474 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1477 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1479 inode->i_mode = mode;
1480 if (S_ISCHR(mode)) {
1481 inode->i_fop = &def_chr_fops;
1482 inode->i_rdev = rdev;
1483 } else if (S_ISBLK(mode)) {
1484 inode->i_fop = &def_blk_fops;
1485 inode->i_rdev = rdev;
1486 } else if (S_ISFIFO(mode))
1487 inode->i_fop = &def_fifo_fops;
1488 else if (S_ISSOCK(mode))
1489 inode->i_fop = &bad_sock_fops;
1491 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1494 EXPORT_SYMBOL(init_special_inode);