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/pagemap.h>
21 #include <linux/cdev.h>
22 #include <linux/bootmem.h>
23 #include <linux/inotify.h>
24 #include <linux/mount.h>
27 * This is needed for the following functions:
29 * - invalidate_inode_buffers
32 * FIXME: remove all knowledge of the buffer layer from this file
34 #include <linux/buffer_head.h>
37 * New inode.c implementation.
39 * This implementation has the basic premise of trying
40 * to be extremely low-overhead and SMP-safe, yet be
41 * simple enough to be "obviously correct".
46 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
48 /* #define INODE_PARANOIA 1 */
49 /* #define INODE_DEBUG 1 */
52 * Inode lookup is no longer as critical as it used to be:
53 * most of the lookups are going to be through the dcache.
55 #define I_HASHBITS i_hash_shift
56 #define I_HASHMASK i_hash_mask
58 static unsigned int i_hash_mask __read_mostly;
59 static unsigned int i_hash_shift __read_mostly;
62 * Each inode can be on two separate lists. One is
63 * the hash list of the inode, used for lookups. The
64 * other linked list is the "type" list:
65 * "in_use" - valid inode, i_count > 0, i_nlink > 0
66 * "dirty" - as "in_use" but also dirty
67 * "unused" - valid inode, i_count = 0
69 * A "dirty" list is maintained for each super block,
70 * allowing for low-overhead inode sync() operations.
73 LIST_HEAD(inode_in_use);
74 LIST_HEAD(inode_unused);
75 static struct hlist_head *inode_hashtable __read_mostly;
78 * A simple spinlock to protect the list manipulations.
80 * NOTE! You also have to own the lock if you change
81 * the i_state of an inode while it is in use..
83 DEFINE_SPINLOCK(inode_lock);
86 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
87 * icache shrinking path, and the umount path. Without this exclusion,
88 * by the time prune_icache calls iput for the inode whose pages it has
89 * been invalidating, or by the time it calls clear_inode & destroy_inode
90 * from its final dispose_list, the struct super_block they refer to
91 * (for inode->i_sb->s_op) may already have been freed and reused.
93 static DEFINE_MUTEX(iprune_mutex);
96 * Statistics gathering..
98 struct inodes_stat_t inodes_stat;
100 static struct kmem_cache * inode_cachep __read_mostly;
102 static void wake_up_inode(struct inode *inode)
105 * Prevent speculative execution through spin_unlock(&inode_lock);
108 wake_up_bit(&inode->i_state, __I_LOCK);
112 * inode_init_always - perform inode structure intialisation
113 * @sb - superblock inode belongs to.
114 * @inode - inode to initialise
116 * These are initializations that need to be done on every inode
117 * allocation as the fields are not initialised by slab allocation.
119 struct inode *inode_init_always(struct super_block *sb, struct inode *inode)
121 static const struct address_space_operations empty_aops;
122 static struct inode_operations empty_iops;
123 static const struct file_operations empty_fops;
125 struct address_space * const mapping = &inode->i_data;
128 inode->i_blkbits = sb->s_blocksize_bits;
130 atomic_set(&inode->i_count, 1);
131 inode->i_op = &empty_iops;
132 inode->i_fop = &empty_fops;
134 atomic_set(&inode->i_writecount, 0);
138 inode->i_generation = 0;
140 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
142 inode->i_pipe = NULL;
143 inode->i_bdev = NULL;
144 inode->i_cdev = NULL;
146 inode->dirtied_when = 0;
147 if (security_inode_alloc(inode)) {
148 if (inode->i_sb->s_op->destroy_inode)
149 inode->i_sb->s_op->destroy_inode(inode);
151 kmem_cache_free(inode_cachep, (inode));
155 spin_lock_init(&inode->i_lock);
156 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
158 mutex_init(&inode->i_mutex);
159 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
161 init_rwsem(&inode->i_alloc_sem);
162 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
164 mapping->a_ops = &empty_aops;
165 mapping->host = inode;
167 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_PAGECACHE);
168 mapping->assoc_mapping = NULL;
169 mapping->backing_dev_info = &default_backing_dev_info;
170 mapping->writeback_index = 0;
173 * If the block_device provides a backing_dev_info for client
174 * inodes then use that. Otherwise the inode share the bdev's
178 struct backing_dev_info *bdi;
180 bdi = sb->s_bdev->bd_inode_backing_dev_info;
182 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
183 mapping->backing_dev_info = bdi;
185 inode->i_private = NULL;
186 inode->i_mapping = mapping;
190 EXPORT_SYMBOL(inode_init_always);
192 static struct inode *alloc_inode(struct super_block *sb)
196 if (sb->s_op->alloc_inode)
197 inode = sb->s_op->alloc_inode(sb);
199 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
202 return inode_init_always(sb, inode);
206 void destroy_inode(struct inode *inode)
208 BUG_ON(inode_has_buffers(inode));
209 security_inode_free(inode);
210 if (inode->i_sb->s_op->destroy_inode)
211 inode->i_sb->s_op->destroy_inode(inode);
213 kmem_cache_free(inode_cachep, (inode));
218 * These are initializations that only need to be done
219 * once, because the fields are idempotent across use
220 * of the inode, so let the slab aware of that.
222 void inode_init_once(struct inode *inode)
224 memset(inode, 0, sizeof(*inode));
225 INIT_HLIST_NODE(&inode->i_hash);
226 INIT_LIST_HEAD(&inode->i_dentry);
227 INIT_LIST_HEAD(&inode->i_devices);
228 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
229 spin_lock_init(&inode->i_data.tree_lock);
230 spin_lock_init(&inode->i_data.i_mmap_lock);
231 INIT_LIST_HEAD(&inode->i_data.private_list);
232 spin_lock_init(&inode->i_data.private_lock);
233 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
234 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
235 i_size_ordered_init(inode);
236 #ifdef CONFIG_INOTIFY
237 INIT_LIST_HEAD(&inode->inotify_watches);
238 mutex_init(&inode->inotify_mutex);
242 EXPORT_SYMBOL(inode_init_once);
244 static void init_once(void *foo)
246 struct inode * inode = (struct inode *) foo;
248 inode_init_once(inode);
252 * inode_lock must be held
254 void __iget(struct inode * inode)
256 if (atomic_read(&inode->i_count)) {
257 atomic_inc(&inode->i_count);
260 atomic_inc(&inode->i_count);
261 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
262 list_move(&inode->i_list, &inode_in_use);
263 inodes_stat.nr_unused--;
267 * clear_inode - clear an inode
268 * @inode: inode to clear
270 * This is called by the filesystem to tell us
271 * that the inode is no longer useful. We just
272 * terminate it with extreme prejudice.
274 void clear_inode(struct inode *inode)
277 invalidate_inode_buffers(inode);
279 BUG_ON(inode->i_data.nrpages);
280 BUG_ON(!(inode->i_state & I_FREEING));
281 BUG_ON(inode->i_state & I_CLEAR);
282 inode_sync_wait(inode);
284 if (inode->i_sb->s_op->clear_inode)
285 inode->i_sb->s_op->clear_inode(inode);
286 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
288 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
290 inode->i_state = I_CLEAR;
293 EXPORT_SYMBOL(clear_inode);
296 * dispose_list - dispose of the contents of a local list
297 * @head: the head of the list to free
299 * Dispose-list gets a local list with local inodes in it, so it doesn't
300 * need to worry about list corruption and SMP locks.
302 static void dispose_list(struct list_head *head)
306 while (!list_empty(head)) {
309 inode = list_first_entry(head, struct inode, i_list);
310 list_del(&inode->i_list);
312 if (inode->i_data.nrpages)
313 truncate_inode_pages(&inode->i_data, 0);
316 spin_lock(&inode_lock);
317 hlist_del_init(&inode->i_hash);
318 list_del_init(&inode->i_sb_list);
319 spin_unlock(&inode_lock);
321 wake_up_inode(inode);
322 destroy_inode(inode);
325 spin_lock(&inode_lock);
326 inodes_stat.nr_inodes -= nr_disposed;
327 spin_unlock(&inode_lock);
331 * Invalidate all inodes for a device.
333 static int invalidate_list(struct list_head *head, struct list_head *dispose)
335 struct list_head *next;
336 int busy = 0, count = 0;
340 struct list_head * tmp = next;
341 struct inode * inode;
344 * We can reschedule here without worrying about the list's
345 * consistency because the per-sb list of inodes must not
346 * change during umount anymore, and because iprune_mutex keeps
347 * shrink_icache_memory() away.
349 cond_resched_lock(&inode_lock);
354 inode = list_entry(tmp, struct inode, i_sb_list);
355 invalidate_inode_buffers(inode);
356 if (!atomic_read(&inode->i_count)) {
357 list_move(&inode->i_list, dispose);
358 inode->i_state |= I_FREEING;
364 /* only unused inodes may be cached with i_count zero */
365 inodes_stat.nr_unused -= count;
370 * invalidate_inodes - discard the inodes on a device
373 * Discard all of the inodes for a given superblock. If the discard
374 * fails because there are busy inodes then a non zero value is returned.
375 * If the discard is successful all the inodes have been discarded.
377 int invalidate_inodes(struct super_block * sb)
380 LIST_HEAD(throw_away);
382 mutex_lock(&iprune_mutex);
383 spin_lock(&inode_lock);
384 inotify_unmount_inodes(&sb->s_inodes);
385 busy = invalidate_list(&sb->s_inodes, &throw_away);
386 spin_unlock(&inode_lock);
388 dispose_list(&throw_away);
389 mutex_unlock(&iprune_mutex);
394 EXPORT_SYMBOL(invalidate_inodes);
396 static int can_unuse(struct inode *inode)
400 if (inode_has_buffers(inode))
402 if (atomic_read(&inode->i_count))
404 if (inode->i_data.nrpages)
410 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
411 * a temporary list and then are freed outside inode_lock by dispose_list().
413 * Any inodes which are pinned purely because of attached pagecache have their
414 * pagecache removed. We expect the final iput() on that inode to add it to
415 * the front of the inode_unused list. So look for it there and if the
416 * inode is still freeable, proceed. The right inode is found 99.9% of the
417 * time in testing on a 4-way.
419 * If the inode has metadata buffers attached to mapping->private_list then
420 * try to remove them.
422 static void prune_icache(int nr_to_scan)
427 unsigned long reap = 0;
429 mutex_lock(&iprune_mutex);
430 spin_lock(&inode_lock);
431 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
434 if (list_empty(&inode_unused))
437 inode = list_entry(inode_unused.prev, struct inode, i_list);
439 if (inode->i_state || atomic_read(&inode->i_count)) {
440 list_move(&inode->i_list, &inode_unused);
443 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
445 spin_unlock(&inode_lock);
446 if (remove_inode_buffers(inode))
447 reap += invalidate_mapping_pages(&inode->i_data,
450 spin_lock(&inode_lock);
452 if (inode != list_entry(inode_unused.next,
453 struct inode, i_list))
454 continue; /* wrong inode or list_empty */
455 if (!can_unuse(inode))
458 list_move(&inode->i_list, &freeable);
459 inode->i_state |= I_FREEING;
462 inodes_stat.nr_unused -= nr_pruned;
463 if (current_is_kswapd())
464 __count_vm_events(KSWAPD_INODESTEAL, reap);
466 __count_vm_events(PGINODESTEAL, reap);
467 spin_unlock(&inode_lock);
469 dispose_list(&freeable);
470 mutex_unlock(&iprune_mutex);
474 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
475 * "unused" means that no dentries are referring to the inodes: the files are
476 * not open and the dcache references to those inodes have already been
479 * This function is passed the number of inodes to scan, and it returns the
480 * total number of remaining possibly-reclaimable inodes.
482 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
486 * Nasty deadlock avoidance. We may hold various FS locks,
487 * and we don't want to recurse into the FS that called us
488 * in clear_inode() and friends..
490 if (!(gfp_mask & __GFP_FS))
494 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
497 static struct shrinker icache_shrinker = {
498 .shrink = shrink_icache_memory,
499 .seeks = DEFAULT_SEEKS,
502 static void __wait_on_freeing_inode(struct inode *inode);
504 * Called with the inode lock held.
505 * NOTE: we are not increasing the inode-refcount, you must call __iget()
506 * by hand after calling find_inode now! This simplifies iunique and won't
507 * add any additional branch in the common code.
509 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
511 struct hlist_node *node;
512 struct inode * inode = NULL;
515 hlist_for_each_entry(inode, node, head, i_hash) {
516 if (inode->i_sb != sb)
518 if (!test(inode, data))
520 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
521 __wait_on_freeing_inode(inode);
526 return node ? inode : NULL;
530 * find_inode_fast is the fast path version of find_inode, see the comment at
531 * iget_locked for details.
533 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
535 struct hlist_node *node;
536 struct inode * inode = NULL;
539 hlist_for_each_entry(inode, node, head, i_hash) {
540 if (inode->i_ino != ino)
542 if (inode->i_sb != sb)
544 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
545 __wait_on_freeing_inode(inode);
550 return node ? inode : NULL;
554 * new_inode - obtain an inode
557 * Allocates a new inode for given superblock. The default gfp_mask
558 * for allocations related to inode->i_mapping is GFP_HIGHUSER_PAGECACHE.
559 * If HIGHMEM pages are unsuitable or it is known that pages allocated
560 * for the page cache are not reclaimable or migratable,
561 * mapping_set_gfp_mask() must be called with suitable flags on the
562 * newly created inode's mapping
565 struct inode *new_inode(struct super_block *sb)
568 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
569 * error if st_ino won't fit in target struct field. Use 32bit counter
570 * here to attempt to avoid that.
572 static unsigned int last_ino;
573 struct inode * inode;
575 spin_lock_prefetch(&inode_lock);
577 inode = alloc_inode(sb);
579 spin_lock(&inode_lock);
580 inodes_stat.nr_inodes++;
581 list_add(&inode->i_list, &inode_in_use);
582 list_add(&inode->i_sb_list, &sb->s_inodes);
583 inode->i_ino = ++last_ino;
585 spin_unlock(&inode_lock);
590 EXPORT_SYMBOL(new_inode);
592 void unlock_new_inode(struct inode *inode)
594 #ifdef CONFIG_DEBUG_LOCK_ALLOC
595 if (inode->i_mode & S_IFDIR) {
596 struct file_system_type *type = inode->i_sb->s_type;
599 * ensure nobody is actually holding i_mutex
601 mutex_destroy(&inode->i_mutex);
602 mutex_init(&inode->i_mutex);
603 lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key);
607 * This is special! We do not need the spinlock
608 * when clearing I_LOCK, because we're guaranteed
609 * that nobody else tries to do anything about the
610 * state of the inode when it is locked, as we
611 * just created it (so there can be no old holders
612 * that haven't tested I_LOCK).
614 inode->i_state &= ~(I_LOCK|I_NEW);
615 wake_up_inode(inode);
618 EXPORT_SYMBOL(unlock_new_inode);
621 * This is called without the inode lock held.. Be careful.
623 * We no longer cache the sb_flags in i_flags - see fs.h
624 * -- rmk@arm.uk.linux.org
626 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)
628 struct inode * inode;
630 inode = alloc_inode(sb);
634 spin_lock(&inode_lock);
635 /* We released the lock, so.. */
636 old = find_inode(sb, head, test, data);
638 if (set(inode, data))
641 inodes_stat.nr_inodes++;
642 list_add(&inode->i_list, &inode_in_use);
643 list_add(&inode->i_sb_list, &sb->s_inodes);
644 hlist_add_head(&inode->i_hash, head);
645 inode->i_state = I_LOCK|I_NEW;
646 spin_unlock(&inode_lock);
648 /* Return the locked inode with I_NEW set, the
649 * caller is responsible for filling in the contents
655 * Uhhuh, somebody else created the same inode under
656 * us. Use the old inode instead of the one we just
660 spin_unlock(&inode_lock);
661 destroy_inode(inode);
663 wait_on_inode(inode);
668 spin_unlock(&inode_lock);
669 destroy_inode(inode);
674 * get_new_inode_fast is the fast path version of get_new_inode, see the
675 * comment at iget_locked for details.
677 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
679 struct inode * inode;
681 inode = alloc_inode(sb);
685 spin_lock(&inode_lock);
686 /* We released the lock, so.. */
687 old = find_inode_fast(sb, head, ino);
690 inodes_stat.nr_inodes++;
691 list_add(&inode->i_list, &inode_in_use);
692 list_add(&inode->i_sb_list, &sb->s_inodes);
693 hlist_add_head(&inode->i_hash, head);
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 static unsigned long hash(struct super_block *sb, unsigned long hashval)
721 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
723 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
724 return tmp & I_HASHMASK;
728 * iunique - get a unique inode number
730 * @max_reserved: highest reserved inode number
732 * Obtain an inode number that is unique on the system for a given
733 * superblock. This is used by file systems that have no natural
734 * permanent inode numbering system. An inode number is returned that
735 * is higher than the reserved limit but unique.
738 * With a large number of inodes live on the file system this function
739 * currently becomes quite slow.
741 ino_t iunique(struct super_block *sb, ino_t max_reserved)
744 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
745 * error if st_ino won't fit in target struct field. Use 32bit counter
746 * here to attempt to avoid that.
748 static unsigned int counter;
750 struct hlist_head *head;
753 spin_lock(&inode_lock);
755 if (counter <= max_reserved)
756 counter = max_reserved + 1;
758 head = inode_hashtable + hash(sb, res);
759 inode = find_inode_fast(sb, head, res);
760 } while (inode != NULL);
761 spin_unlock(&inode_lock);
765 EXPORT_SYMBOL(iunique);
767 struct inode *igrab(struct inode *inode)
769 spin_lock(&inode_lock);
770 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
774 * Handle the case where s_op->clear_inode is not been
775 * called yet, and somebody is calling igrab
776 * while the inode is getting freed.
779 spin_unlock(&inode_lock);
783 EXPORT_SYMBOL(igrab);
786 * ifind - internal function, you want ilookup5() or iget5().
787 * @sb: super block of file system to search
788 * @head: the head of the list to search
789 * @test: callback used for comparisons between inodes
790 * @data: opaque data pointer to pass to @test
791 * @wait: if true wait for the inode to be unlocked, if false do not
793 * ifind() searches for the inode specified by @data in the inode
794 * cache. This is a generalized version of ifind_fast() for file systems where
795 * the inode number is not sufficient for unique identification of an inode.
797 * If the inode is in the cache, the inode is returned with an incremented
800 * Otherwise NULL is returned.
802 * Note, @test is called with the inode_lock held, so can't sleep.
804 static struct inode *ifind(struct super_block *sb,
805 struct hlist_head *head, int (*test)(struct inode *, void *),
806 void *data, const int wait)
810 spin_lock(&inode_lock);
811 inode = find_inode(sb, head, test, data);
814 spin_unlock(&inode_lock);
816 wait_on_inode(inode);
819 spin_unlock(&inode_lock);
824 * ifind_fast - internal function, you want ilookup() or iget().
825 * @sb: super block of file system to search
826 * @head: head of the list to search
827 * @ino: inode number to search for
829 * ifind_fast() searches for the inode @ino in the inode cache. This is for
830 * file systems where the inode number is sufficient for unique identification
833 * If the inode is in the cache, the inode is returned with an incremented
836 * Otherwise NULL is returned.
838 static struct inode *ifind_fast(struct super_block *sb,
839 struct hlist_head *head, unsigned long ino)
843 spin_lock(&inode_lock);
844 inode = find_inode_fast(sb, head, ino);
847 spin_unlock(&inode_lock);
848 wait_on_inode(inode);
851 spin_unlock(&inode_lock);
856 * ilookup5_nowait - search for an inode in the inode cache
857 * @sb: super block of file system to search
858 * @hashval: hash value (usually inode number) to search for
859 * @test: callback used for comparisons between inodes
860 * @data: opaque data pointer to pass to @test
862 * ilookup5() uses ifind() to search for the inode specified by @hashval and
863 * @data in the inode cache. This is a generalized version of ilookup() for
864 * file systems where the inode number is not sufficient for unique
865 * identification of an inode.
867 * If the inode is in the cache, the inode is returned with an incremented
868 * reference count. Note, the inode lock is not waited upon so you have to be
869 * very careful what you do with the returned inode. You probably should be
870 * using ilookup5() instead.
872 * Otherwise NULL is returned.
874 * Note, @test is called with the inode_lock held, so can't sleep.
876 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
877 int (*test)(struct inode *, void *), void *data)
879 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
881 return ifind(sb, head, test, data, 0);
884 EXPORT_SYMBOL(ilookup5_nowait);
887 * ilookup5 - search for an inode in the inode cache
888 * @sb: super block of file system to search
889 * @hashval: hash value (usually inode number) to search for
890 * @test: callback used for comparisons between inodes
891 * @data: opaque data pointer to pass to @test
893 * ilookup5() uses ifind() to search for the inode specified by @hashval and
894 * @data in the inode cache. This is a generalized version of ilookup() for
895 * file systems where the inode number is not sufficient for unique
896 * identification of an inode.
898 * If the inode is in the cache, the inode lock is waited upon and the inode is
899 * returned with an incremented reference count.
901 * Otherwise NULL is returned.
903 * Note, @test is called with the inode_lock held, so can't sleep.
905 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
906 int (*test)(struct inode *, void *), void *data)
908 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
910 return ifind(sb, head, test, data, 1);
913 EXPORT_SYMBOL(ilookup5);
916 * ilookup - search for an inode in the inode cache
917 * @sb: super block of file system to search
918 * @ino: inode number to search for
920 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
921 * This is for file systems where the inode number is sufficient for unique
922 * identification of an inode.
924 * If the inode is in the cache, the inode is returned with an incremented
927 * Otherwise NULL is returned.
929 struct inode *ilookup(struct super_block *sb, unsigned long ino)
931 struct hlist_head *head = inode_hashtable + hash(sb, ino);
933 return ifind_fast(sb, head, ino);
936 EXPORT_SYMBOL(ilookup);
939 * iget5_locked - obtain an inode from a mounted file system
940 * @sb: super block of file system
941 * @hashval: hash value (usually inode number) to get
942 * @test: callback used for comparisons between inodes
943 * @set: callback used to initialize a new struct inode
944 * @data: opaque data pointer to pass to @test and @set
946 * iget5_locked() uses ifind() to search for the inode specified by @hashval
947 * and @data in the inode cache and if present it is returned with an increased
948 * reference count. This is a generalized version of iget_locked() for file
949 * systems where the inode number is not sufficient for unique identification
952 * If the inode is not in cache, get_new_inode() is called to allocate a new
953 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
954 * file system gets to fill it in before unlocking it via unlock_new_inode().
956 * Note both @test and @set are called with the inode_lock held, so can't sleep.
958 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
959 int (*test)(struct inode *, void *),
960 int (*set)(struct inode *, void *), void *data)
962 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
965 inode = ifind(sb, head, test, data, 1);
969 * get_new_inode() will do the right thing, re-trying the search
970 * in case it had to block at any point.
972 return get_new_inode(sb, head, test, set, data);
975 EXPORT_SYMBOL(iget5_locked);
978 * iget_locked - obtain an inode from a mounted file system
979 * @sb: super block of file system
980 * @ino: inode number to get
982 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
983 * the inode cache and if present it is returned with an increased reference
984 * count. This is for file systems where the inode number is sufficient for
985 * unique identification of an inode.
987 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
988 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
989 * The file system gets to fill it in before unlocking it via
990 * unlock_new_inode().
992 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
994 struct hlist_head *head = inode_hashtable + hash(sb, ino);
997 inode = ifind_fast(sb, head, ino);
1001 * get_new_inode_fast() will do the right thing, re-trying the search
1002 * in case it had to block at any point.
1004 return get_new_inode_fast(sb, head, ino);
1007 EXPORT_SYMBOL(iget_locked);
1010 * __insert_inode_hash - hash an inode
1011 * @inode: unhashed inode
1012 * @hashval: unsigned long value used to locate this object in the
1015 * Add an inode to the inode hash for this superblock.
1017 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1019 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1020 spin_lock(&inode_lock);
1021 hlist_add_head(&inode->i_hash, head);
1022 spin_unlock(&inode_lock);
1025 EXPORT_SYMBOL(__insert_inode_hash);
1028 * remove_inode_hash - remove an inode from the hash
1029 * @inode: inode to unhash
1031 * Remove an inode from the superblock.
1033 void remove_inode_hash(struct inode *inode)
1035 spin_lock(&inode_lock);
1036 hlist_del_init(&inode->i_hash);
1037 spin_unlock(&inode_lock);
1040 EXPORT_SYMBOL(remove_inode_hash);
1043 * Tell the filesystem that this inode is no longer of any interest and should
1044 * be completely destroyed.
1046 * We leave the inode in the inode hash table until *after* the filesystem's
1047 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1048 * instigate) will always find up-to-date information either in the hash or on
1051 * I_FREEING is set so that no-one will take a new reference to the inode while
1052 * it is being deleted.
1054 void generic_delete_inode(struct inode *inode)
1056 const struct super_operations *op = inode->i_sb->s_op;
1058 list_del_init(&inode->i_list);
1059 list_del_init(&inode->i_sb_list);
1060 inode->i_state |= I_FREEING;
1061 inodes_stat.nr_inodes--;
1062 spin_unlock(&inode_lock);
1064 security_inode_delete(inode);
1066 if (op->delete_inode) {
1067 void (*delete)(struct inode *) = op->delete_inode;
1068 if (!is_bad_inode(inode))
1070 /* Filesystems implementing their own
1071 * s_op->delete_inode are required to call
1072 * truncate_inode_pages and clear_inode()
1076 truncate_inode_pages(&inode->i_data, 0);
1079 spin_lock(&inode_lock);
1080 hlist_del_init(&inode->i_hash);
1081 spin_unlock(&inode_lock);
1082 wake_up_inode(inode);
1083 BUG_ON(inode->i_state != I_CLEAR);
1084 destroy_inode(inode);
1087 EXPORT_SYMBOL(generic_delete_inode);
1089 static void generic_forget_inode(struct inode *inode)
1091 struct super_block *sb = inode->i_sb;
1093 if (!hlist_unhashed(&inode->i_hash)) {
1094 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1095 list_move(&inode->i_list, &inode_unused);
1096 inodes_stat.nr_unused++;
1097 if (sb->s_flags & MS_ACTIVE) {
1098 spin_unlock(&inode_lock);
1101 inode->i_state |= I_WILL_FREE;
1102 spin_unlock(&inode_lock);
1103 write_inode_now(inode, 1);
1104 spin_lock(&inode_lock);
1105 inode->i_state &= ~I_WILL_FREE;
1106 inodes_stat.nr_unused--;
1107 hlist_del_init(&inode->i_hash);
1109 list_del_init(&inode->i_list);
1110 list_del_init(&inode->i_sb_list);
1111 inode->i_state |= I_FREEING;
1112 inodes_stat.nr_inodes--;
1113 spin_unlock(&inode_lock);
1114 if (inode->i_data.nrpages)
1115 truncate_inode_pages(&inode->i_data, 0);
1117 wake_up_inode(inode);
1118 destroy_inode(inode);
1122 * Normal UNIX filesystem behaviour: delete the
1123 * inode when the usage count drops to zero, and
1126 void generic_drop_inode(struct inode *inode)
1128 if (!inode->i_nlink)
1129 generic_delete_inode(inode);
1131 generic_forget_inode(inode);
1134 EXPORT_SYMBOL_GPL(generic_drop_inode);
1137 * Called when we're dropping the last reference
1140 * Call the FS "drop()" function, defaulting to
1141 * the legacy UNIX filesystem behaviour..
1143 * NOTE! NOTE! NOTE! We're called with the inode lock
1144 * held, and the drop function is supposed to release
1147 static inline void iput_final(struct inode *inode)
1149 const struct super_operations *op = inode->i_sb->s_op;
1150 void (*drop)(struct inode *) = generic_drop_inode;
1152 if (op && op->drop_inode)
1153 drop = op->drop_inode;
1158 * iput - put an inode
1159 * @inode: inode to put
1161 * Puts an inode, dropping its usage count. If the inode use count hits
1162 * zero, the inode is then freed and may also be destroyed.
1164 * Consequently, iput() can sleep.
1166 void iput(struct inode *inode)
1169 BUG_ON(inode->i_state == I_CLEAR);
1171 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1176 EXPORT_SYMBOL(iput);
1179 * bmap - find a block number in a file
1180 * @inode: inode of file
1181 * @block: block to find
1183 * Returns the block number on the device holding the inode that
1184 * is the disk block number for the block of the file requested.
1185 * That is, asked for block 4 of inode 1 the function will return the
1186 * disk block relative to the disk start that holds that block of the
1189 sector_t bmap(struct inode * inode, sector_t block)
1192 if (inode->i_mapping->a_ops->bmap)
1193 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1196 EXPORT_SYMBOL(bmap);
1199 * touch_atime - update the access time
1200 * @mnt: mount the inode is accessed on
1201 * @dentry: dentry accessed
1203 * Update the accessed time on an inode and mark it for writeback.
1204 * This function automatically handles read only file systems and media,
1205 * as well as the "noatime" flag and inode specific "noatime" markers.
1207 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1209 struct inode *inode = dentry->d_inode;
1210 struct timespec now;
1212 if (mnt_want_write(mnt))
1214 if (inode->i_flags & S_NOATIME)
1216 if (IS_NOATIME(inode))
1218 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1221 if (mnt->mnt_flags & MNT_NOATIME)
1223 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1225 if (mnt->mnt_flags & MNT_RELATIME) {
1227 * With relative atime, only update atime if the previous
1228 * atime is earlier than either the ctime or mtime.
1230 if (timespec_compare(&inode->i_mtime, &inode->i_atime) < 0 &&
1231 timespec_compare(&inode->i_ctime, &inode->i_atime) < 0)
1235 now = current_fs_time(inode->i_sb);
1236 if (timespec_equal(&inode->i_atime, &now))
1239 inode->i_atime = now;
1240 mark_inode_dirty_sync(inode);
1242 mnt_drop_write(mnt);
1244 EXPORT_SYMBOL(touch_atime);
1247 * file_update_time - update mtime and ctime time
1248 * @file: file accessed
1250 * Update the mtime and ctime members of an inode and mark the inode
1251 * for writeback. Note that this function is meant exclusively for
1252 * usage in the file write path of filesystems, and filesystems may
1253 * choose to explicitly ignore update via this function with the
1254 * S_NOCTIME inode flag, e.g. for network filesystem where these
1255 * timestamps are handled by the server.
1258 void file_update_time(struct file *file)
1260 struct inode *inode = file->f_path.dentry->d_inode;
1261 struct timespec now;
1265 if (IS_NOCMTIME(inode))
1268 err = mnt_want_write(file->f_path.mnt);
1272 now = current_fs_time(inode->i_sb);
1273 if (!timespec_equal(&inode->i_mtime, &now)) {
1274 inode->i_mtime = now;
1278 if (!timespec_equal(&inode->i_ctime, &now)) {
1279 inode->i_ctime = now;
1283 if (IS_I_VERSION(inode)) {
1284 inode_inc_iversion(inode);
1289 mark_inode_dirty_sync(inode);
1290 mnt_drop_write(file->f_path.mnt);
1293 EXPORT_SYMBOL(file_update_time);
1295 int inode_needs_sync(struct inode *inode)
1299 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1304 EXPORT_SYMBOL(inode_needs_sync);
1306 int inode_wait(void *word)
1313 * If we try to find an inode in the inode hash while it is being
1314 * deleted, we have to wait until the filesystem completes its
1315 * deletion before reporting that it isn't found. This function waits
1316 * until the deletion _might_ have completed. Callers are responsible
1317 * to recheck inode state.
1319 * It doesn't matter if I_LOCK is not set initially, a call to
1320 * wake_up_inode() after removing from the hash list will DTRT.
1322 * This is called with inode_lock held.
1324 static void __wait_on_freeing_inode(struct inode *inode)
1326 wait_queue_head_t *wq;
1327 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1328 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1329 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1330 spin_unlock(&inode_lock);
1332 finish_wait(wq, &wait.wait);
1333 spin_lock(&inode_lock);
1337 * We rarely want to lock two inodes that do not have a parent/child
1338 * relationship (such as directory, child inode) simultaneously. The
1339 * vast majority of file systems should be able to get along fine
1340 * without this. Do not use these functions except as a last resort.
1342 void inode_double_lock(struct inode *inode1, struct inode *inode2)
1344 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1346 mutex_lock(&inode1->i_mutex);
1348 mutex_lock(&inode2->i_mutex);
1352 if (inode1 < inode2) {
1353 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1354 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1356 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1357 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1360 EXPORT_SYMBOL(inode_double_lock);
1362 void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1365 mutex_unlock(&inode1->i_mutex);
1367 if (inode2 && inode2 != inode1)
1368 mutex_unlock(&inode2->i_mutex);
1370 EXPORT_SYMBOL(inode_double_unlock);
1372 static __initdata unsigned long ihash_entries;
1373 static int __init set_ihash_entries(char *str)
1377 ihash_entries = simple_strtoul(str, &str, 0);
1380 __setup("ihash_entries=", set_ihash_entries);
1383 * Initialize the waitqueues and inode hash table.
1385 void __init inode_init_early(void)
1389 /* If hashes are distributed across NUMA nodes, defer
1390 * hash allocation until vmalloc space is available.
1396 alloc_large_system_hash("Inode-cache",
1397 sizeof(struct hlist_head),
1405 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1406 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1409 void __init inode_init(void)
1413 /* inode slab cache */
1414 inode_cachep = kmem_cache_create("inode_cache",
1415 sizeof(struct inode),
1417 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1420 register_shrinker(&icache_shrinker);
1422 /* Hash may have been set up in inode_init_early */
1427 alloc_large_system_hash("Inode-cache",
1428 sizeof(struct hlist_head),
1436 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1437 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1440 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1442 inode->i_mode = mode;
1443 if (S_ISCHR(mode)) {
1444 inode->i_fop = &def_chr_fops;
1445 inode->i_rdev = rdev;
1446 } else if (S_ISBLK(mode)) {
1447 inode->i_fop = &def_blk_fops;
1448 inode->i_rdev = rdev;
1449 } else if (S_ISFIFO(mode))
1450 inode->i_fop = &def_fifo_fops;
1451 else if (S_ISSOCK(mode))
1452 inode->i_fop = &bad_sock_fops;
1454 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1457 EXPORT_SYMBOL(init_special_inode);