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 struct inode *alloc_inode(struct super_block *sb)
104 static const struct address_space_operations empty_aops;
105 static struct inode_operations empty_iops;
106 static const struct file_operations empty_fops;
109 if (sb->s_op->alloc_inode)
110 inode = sb->s_op->alloc_inode(sb);
112 inode = (struct inode *) kmem_cache_alloc(inode_cachep, GFP_KERNEL);
115 struct address_space * const mapping = &inode->i_data;
118 inode->i_blkbits = sb->s_blocksize_bits;
120 atomic_set(&inode->i_count, 1);
121 inode->i_op = &empty_iops;
122 inode->i_fop = &empty_fops;
124 atomic_set(&inode->i_writecount, 0);
128 inode->i_generation = 0;
130 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
132 inode->i_pipe = NULL;
133 inode->i_bdev = NULL;
134 inode->i_cdev = NULL;
136 inode->dirtied_when = 0;
137 if (security_inode_alloc(inode)) {
138 if (inode->i_sb->s_op->destroy_inode)
139 inode->i_sb->s_op->destroy_inode(inode);
141 kmem_cache_free(inode_cachep, (inode));
145 spin_lock_init(&inode->i_lock);
146 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
148 mutex_init(&inode->i_mutex);
149 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
151 init_rwsem(&inode->i_alloc_sem);
152 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
154 mapping->a_ops = &empty_aops;
155 mapping->host = inode;
157 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_PAGECACHE);
158 mapping->assoc_mapping = NULL;
159 mapping->backing_dev_info = &default_backing_dev_info;
162 * If the block_device provides a backing_dev_info for client
163 * inodes then use that. Otherwise the inode share the bdev's
167 struct backing_dev_info *bdi;
169 bdi = sb->s_bdev->bd_inode_backing_dev_info;
171 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
172 mapping->backing_dev_info = bdi;
174 inode->i_private = NULL;
175 inode->i_mapping = mapping;
180 void destroy_inode(struct inode *inode)
182 BUG_ON(inode_has_buffers(inode));
183 security_inode_free(inode);
184 if (inode->i_sb->s_op->destroy_inode)
185 inode->i_sb->s_op->destroy_inode(inode);
187 kmem_cache_free(inode_cachep, (inode));
192 * These are initializations that only need to be done
193 * once, because the fields are idempotent across use
194 * of the inode, so let the slab aware of that.
196 void inode_init_once(struct inode *inode)
198 memset(inode, 0, sizeof(*inode));
199 INIT_HLIST_NODE(&inode->i_hash);
200 INIT_LIST_HEAD(&inode->i_dentry);
201 INIT_LIST_HEAD(&inode->i_devices);
202 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
203 rwlock_init(&inode->i_data.tree_lock);
204 spin_lock_init(&inode->i_data.i_mmap_lock);
205 INIT_LIST_HEAD(&inode->i_data.private_list);
206 spin_lock_init(&inode->i_data.private_lock);
207 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
208 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
209 i_size_ordered_init(inode);
210 #ifdef CONFIG_INOTIFY
211 INIT_LIST_HEAD(&inode->inotify_watches);
212 mutex_init(&inode->inotify_mutex);
216 EXPORT_SYMBOL(inode_init_once);
218 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
220 struct inode * inode = (struct inode *) foo;
222 inode_init_once(inode);
226 * inode_lock must be held
228 void __iget(struct inode * inode)
230 if (atomic_read(&inode->i_count)) {
231 atomic_inc(&inode->i_count);
234 atomic_inc(&inode->i_count);
235 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
236 list_move(&inode->i_list, &inode_in_use);
237 inodes_stat.nr_unused--;
241 * clear_inode - clear an inode
242 * @inode: inode to clear
244 * This is called by the filesystem to tell us
245 * that the inode is no longer useful. We just
246 * terminate it with extreme prejudice.
248 void clear_inode(struct inode *inode)
251 invalidate_inode_buffers(inode);
253 BUG_ON(inode->i_data.nrpages);
254 BUG_ON(!(inode->i_state & I_FREEING));
255 BUG_ON(inode->i_state & I_CLEAR);
256 wait_on_inode(inode);
258 if (inode->i_sb->s_op->clear_inode)
259 inode->i_sb->s_op->clear_inode(inode);
260 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
262 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
264 inode->i_state = I_CLEAR;
267 EXPORT_SYMBOL(clear_inode);
270 * dispose_list - dispose of the contents of a local list
271 * @head: the head of the list to free
273 * Dispose-list gets a local list with local inodes in it, so it doesn't
274 * need to worry about list corruption and SMP locks.
276 static void dispose_list(struct list_head *head)
280 while (!list_empty(head)) {
283 inode = list_first_entry(head, struct inode, i_list);
284 list_del(&inode->i_list);
286 if (inode->i_data.nrpages)
287 truncate_inode_pages(&inode->i_data, 0);
290 spin_lock(&inode_lock);
291 hlist_del_init(&inode->i_hash);
292 list_del_init(&inode->i_sb_list);
293 spin_unlock(&inode_lock);
295 wake_up_inode(inode);
296 destroy_inode(inode);
299 spin_lock(&inode_lock);
300 inodes_stat.nr_inodes -= nr_disposed;
301 spin_unlock(&inode_lock);
305 * Invalidate all inodes for a device.
307 static int invalidate_list(struct list_head *head, struct list_head *dispose)
309 struct list_head *next;
310 int busy = 0, count = 0;
314 struct list_head * tmp = next;
315 struct inode * inode;
318 * We can reschedule here without worrying about the list's
319 * consistency because the per-sb list of inodes must not
320 * change during umount anymore, and because iprune_mutex keeps
321 * shrink_icache_memory() away.
323 cond_resched_lock(&inode_lock);
328 inode = list_entry(tmp, struct inode, i_sb_list);
329 invalidate_inode_buffers(inode);
330 if (!atomic_read(&inode->i_count)) {
331 list_move(&inode->i_list, dispose);
332 inode->i_state |= I_FREEING;
338 /* only unused inodes may be cached with i_count zero */
339 inodes_stat.nr_unused -= count;
344 * invalidate_inodes - discard the inodes on a device
347 * Discard all of the inodes for a given superblock. If the discard
348 * fails because there are busy inodes then a non zero value is returned.
349 * If the discard is successful all the inodes have been discarded.
351 int invalidate_inodes(struct super_block * sb)
354 LIST_HEAD(throw_away);
356 mutex_lock(&iprune_mutex);
357 spin_lock(&inode_lock);
358 inotify_unmount_inodes(&sb->s_inodes);
359 busy = invalidate_list(&sb->s_inodes, &throw_away);
360 spin_unlock(&inode_lock);
362 dispose_list(&throw_away);
363 mutex_unlock(&iprune_mutex);
368 EXPORT_SYMBOL(invalidate_inodes);
370 static int can_unuse(struct inode *inode)
374 if (inode_has_buffers(inode))
376 if (atomic_read(&inode->i_count))
378 if (inode->i_data.nrpages)
384 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
385 * a temporary list and then are freed outside inode_lock by dispose_list().
387 * Any inodes which are pinned purely because of attached pagecache have their
388 * pagecache removed. We expect the final iput() on that inode to add it to
389 * the front of the inode_unused list. So look for it there and if the
390 * inode is still freeable, proceed. The right inode is found 99.9% of the
391 * time in testing on a 4-way.
393 * If the inode has metadata buffers attached to mapping->private_list then
394 * try to remove them.
396 static void prune_icache(int nr_to_scan)
401 unsigned long reap = 0;
403 mutex_lock(&iprune_mutex);
404 spin_lock(&inode_lock);
405 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
408 if (list_empty(&inode_unused))
411 inode = list_entry(inode_unused.prev, struct inode, i_list);
413 if (inode->i_state || atomic_read(&inode->i_count)) {
414 list_move(&inode->i_list, &inode_unused);
417 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
419 spin_unlock(&inode_lock);
420 if (remove_inode_buffers(inode))
421 reap += invalidate_mapping_pages(&inode->i_data,
424 spin_lock(&inode_lock);
426 if (inode != list_entry(inode_unused.next,
427 struct inode, i_list))
428 continue; /* wrong inode or list_empty */
429 if (!can_unuse(inode))
432 list_move(&inode->i_list, &freeable);
433 inode->i_state |= I_FREEING;
436 inodes_stat.nr_unused -= nr_pruned;
437 if (current_is_kswapd())
438 __count_vm_events(KSWAPD_INODESTEAL, reap);
440 __count_vm_events(PGINODESTEAL, reap);
441 spin_unlock(&inode_lock);
443 dispose_list(&freeable);
444 mutex_unlock(&iprune_mutex);
448 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
449 * "unused" means that no dentries are referring to the inodes: the files are
450 * not open and the dcache references to those inodes have already been
453 * This function is passed the number of inodes to scan, and it returns the
454 * total number of remaining possibly-reclaimable inodes.
456 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
460 * Nasty deadlock avoidance. We may hold various FS locks,
461 * and we don't want to recurse into the FS that called us
462 * in clear_inode() and friends..
464 if (!(gfp_mask & __GFP_FS))
468 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
471 static struct shrinker icache_shrinker = {
472 .shrink = shrink_icache_memory,
473 .seeks = DEFAULT_SEEKS,
476 static void __wait_on_freeing_inode(struct inode *inode);
478 * Called with the inode lock held.
479 * NOTE: we are not increasing the inode-refcount, you must call __iget()
480 * by hand after calling find_inode now! This simplifies iunique and won't
481 * add any additional branch in the common code.
483 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
485 struct hlist_node *node;
486 struct inode * inode = NULL;
489 hlist_for_each (node, head) {
490 inode = hlist_entry(node, struct inode, i_hash);
491 if (inode->i_sb != sb)
493 if (!test(inode, data))
495 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
496 __wait_on_freeing_inode(inode);
501 return node ? inode : NULL;
505 * find_inode_fast is the fast path version of find_inode, see the comment at
506 * iget_locked for details.
508 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
510 struct hlist_node *node;
511 struct inode * inode = NULL;
514 hlist_for_each (node, head) {
515 inode = hlist_entry(node, struct inode, i_hash);
516 if (inode->i_ino != ino)
518 if (inode->i_sb != sb)
520 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
521 __wait_on_freeing_inode(inode);
526 return node ? inode : NULL;
530 * new_inode - obtain an inode
533 * Allocates a new inode for given superblock. The default gfp_mask
534 * for allocations related to inode->i_mapping is GFP_HIGHUSER_PAGECACHE.
535 * If HIGHMEM pages are unsuitable or it is known that pages allocated
536 * for the page cache are not reclaimable or migratable,
537 * mapping_set_gfp_mask() must be called with suitable flags on the
538 * newly created inode's mapping
541 struct inode *new_inode(struct super_block *sb)
544 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
545 * error if st_ino won't fit in target struct field. Use 32bit counter
546 * here to attempt to avoid that.
548 static unsigned int last_ino;
549 struct inode * inode;
551 spin_lock_prefetch(&inode_lock);
553 inode = alloc_inode(sb);
555 spin_lock(&inode_lock);
556 inodes_stat.nr_inodes++;
557 list_add(&inode->i_list, &inode_in_use);
558 list_add(&inode->i_sb_list, &sb->s_inodes);
559 inode->i_ino = ++last_ino;
561 spin_unlock(&inode_lock);
566 EXPORT_SYMBOL(new_inode);
568 void unlock_new_inode(struct inode *inode)
571 * This is special! We do not need the spinlock
572 * when clearing I_LOCK, because we're guaranteed
573 * that nobody else tries to do anything about the
574 * state of the inode when it is locked, as we
575 * just created it (so there can be no old holders
576 * that haven't tested I_LOCK).
578 inode->i_state &= ~(I_LOCK|I_NEW);
579 wake_up_inode(inode);
582 EXPORT_SYMBOL(unlock_new_inode);
585 * This is called without the inode lock held.. Be careful.
587 * We no longer cache the sb_flags in i_flags - see fs.h
588 * -- rmk@arm.uk.linux.org
590 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)
592 struct inode * inode;
594 inode = alloc_inode(sb);
598 spin_lock(&inode_lock);
599 /* We released the lock, so.. */
600 old = find_inode(sb, head, test, data);
602 if (set(inode, data))
605 inodes_stat.nr_inodes++;
606 list_add(&inode->i_list, &inode_in_use);
607 list_add(&inode->i_sb_list, &sb->s_inodes);
608 hlist_add_head(&inode->i_hash, head);
609 inode->i_state = I_LOCK|I_NEW;
610 spin_unlock(&inode_lock);
612 /* Return the locked inode with I_NEW set, the
613 * caller is responsible for filling in the contents
619 * Uhhuh, somebody else created the same inode under
620 * us. Use the old inode instead of the one we just
624 spin_unlock(&inode_lock);
625 destroy_inode(inode);
627 wait_on_inode(inode);
632 spin_unlock(&inode_lock);
633 destroy_inode(inode);
638 * get_new_inode_fast is the fast path version of get_new_inode, see the
639 * comment at iget_locked for details.
641 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
643 struct inode * inode;
645 inode = alloc_inode(sb);
649 spin_lock(&inode_lock);
650 /* We released the lock, so.. */
651 old = find_inode_fast(sb, head, ino);
654 inodes_stat.nr_inodes++;
655 list_add(&inode->i_list, &inode_in_use);
656 list_add(&inode->i_sb_list, &sb->s_inodes);
657 hlist_add_head(&inode->i_hash, head);
658 inode->i_state = I_LOCK|I_NEW;
659 spin_unlock(&inode_lock);
661 /* Return the locked inode with I_NEW set, the
662 * caller is responsible for filling in the contents
668 * Uhhuh, somebody else created the same inode under
669 * us. Use the old inode instead of the one we just
673 spin_unlock(&inode_lock);
674 destroy_inode(inode);
676 wait_on_inode(inode);
681 static unsigned long hash(struct super_block *sb, unsigned long hashval)
685 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
687 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
688 return tmp & I_HASHMASK;
692 * iunique - get a unique inode number
694 * @max_reserved: highest reserved inode number
696 * Obtain an inode number that is unique on the system for a given
697 * superblock. This is used by file systems that have no natural
698 * permanent inode numbering system. An inode number is returned that
699 * is higher than the reserved limit but unique.
702 * With a large number of inodes live on the file system this function
703 * currently becomes quite slow.
705 ino_t iunique(struct super_block *sb, ino_t max_reserved)
708 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
709 * error if st_ino won't fit in target struct field. Use 32bit counter
710 * here to attempt to avoid that.
712 static unsigned int counter;
714 struct hlist_head *head;
717 spin_lock(&inode_lock);
719 if (counter <= max_reserved)
720 counter = max_reserved + 1;
722 head = inode_hashtable + hash(sb, res);
723 inode = find_inode_fast(sb, head, res);
724 } while (inode != NULL);
725 spin_unlock(&inode_lock);
729 EXPORT_SYMBOL(iunique);
731 struct inode *igrab(struct inode *inode)
733 spin_lock(&inode_lock);
734 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
738 * Handle the case where s_op->clear_inode is not been
739 * called yet, and somebody is calling igrab
740 * while the inode is getting freed.
743 spin_unlock(&inode_lock);
747 EXPORT_SYMBOL(igrab);
750 * ifind - internal function, you want ilookup5() or iget5().
751 * @sb: super block of file system to search
752 * @head: the head of the list to search
753 * @test: callback used for comparisons between inodes
754 * @data: opaque data pointer to pass to @test
755 * @wait: if true wait for the inode to be unlocked, if false do not
757 * ifind() searches for the inode specified by @data in the inode
758 * cache. This is a generalized version of ifind_fast() for file systems where
759 * the inode number is not sufficient for unique identification of an inode.
761 * If the inode is in the cache, the inode is returned with an incremented
764 * Otherwise NULL is returned.
766 * Note, @test is called with the inode_lock held, so can't sleep.
768 static struct inode *ifind(struct super_block *sb,
769 struct hlist_head *head, int (*test)(struct inode *, void *),
770 void *data, const int wait)
774 spin_lock(&inode_lock);
775 inode = find_inode(sb, head, test, data);
778 spin_unlock(&inode_lock);
780 wait_on_inode(inode);
783 spin_unlock(&inode_lock);
788 * ifind_fast - internal function, you want ilookup() or iget().
789 * @sb: super block of file system to search
790 * @head: head of the list to search
791 * @ino: inode number to search for
793 * ifind_fast() searches for the inode @ino in the inode cache. This is for
794 * file systems where the inode number is sufficient for unique identification
797 * If the inode is in the cache, the inode is returned with an incremented
800 * Otherwise NULL is returned.
802 static struct inode *ifind_fast(struct super_block *sb,
803 struct hlist_head *head, unsigned long ino)
807 spin_lock(&inode_lock);
808 inode = find_inode_fast(sb, head, ino);
811 spin_unlock(&inode_lock);
812 wait_on_inode(inode);
815 spin_unlock(&inode_lock);
820 * ilookup5_nowait - search for an inode in the inode cache
821 * @sb: super block of file system to search
822 * @hashval: hash value (usually inode number) to search for
823 * @test: callback used for comparisons between inodes
824 * @data: opaque data pointer to pass to @test
826 * ilookup5() uses ifind() to search for the inode specified by @hashval and
827 * @data in the inode cache. This is a generalized version of ilookup() for
828 * file systems where the inode number is not sufficient for unique
829 * identification of an inode.
831 * If the inode is in the cache, the inode is returned with an incremented
832 * reference count. Note, the inode lock is not waited upon so you have to be
833 * very careful what you do with the returned inode. You probably should be
834 * using ilookup5() instead.
836 * Otherwise NULL is returned.
838 * Note, @test is called with the inode_lock held, so can't sleep.
840 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
841 int (*test)(struct inode *, void *), void *data)
843 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
845 return ifind(sb, head, test, data, 0);
848 EXPORT_SYMBOL(ilookup5_nowait);
851 * ilookup5 - search for an inode in the inode cache
852 * @sb: super block of file system to search
853 * @hashval: hash value (usually inode number) to search for
854 * @test: callback used for comparisons between inodes
855 * @data: opaque data pointer to pass to @test
857 * ilookup5() uses ifind() to search for the inode specified by @hashval and
858 * @data in the inode cache. This is a generalized version of ilookup() for
859 * file systems where the inode number is not sufficient for unique
860 * identification of an inode.
862 * If the inode is in the cache, the inode lock is waited upon and the inode is
863 * returned with an incremented reference count.
865 * Otherwise NULL is returned.
867 * Note, @test is called with the inode_lock held, so can't sleep.
869 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
870 int (*test)(struct inode *, void *), void *data)
872 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
874 return ifind(sb, head, test, data, 1);
877 EXPORT_SYMBOL(ilookup5);
880 * ilookup - search for an inode in the inode cache
881 * @sb: super block of file system to search
882 * @ino: inode number to search for
884 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
885 * This is for file systems where the inode number is sufficient for unique
886 * identification of an inode.
888 * If the inode is in the cache, the inode is returned with an incremented
891 * Otherwise NULL is returned.
893 struct inode *ilookup(struct super_block *sb, unsigned long ino)
895 struct hlist_head *head = inode_hashtable + hash(sb, ino);
897 return ifind_fast(sb, head, ino);
900 EXPORT_SYMBOL(ilookup);
903 * iget5_locked - obtain an inode from a mounted file system
904 * @sb: super block of file system
905 * @hashval: hash value (usually inode number) to get
906 * @test: callback used for comparisons between inodes
907 * @set: callback used to initialize a new struct inode
908 * @data: opaque data pointer to pass to @test and @set
910 * This is iget() without the read_inode() portion of get_new_inode().
912 * iget5_locked() uses ifind() to search for the inode specified by @hashval
913 * and @data in the inode cache and if present it is returned with an increased
914 * reference count. This is a generalized version of iget_locked() for file
915 * systems where the inode number is not sufficient for unique identification
918 * If the inode is not in cache, get_new_inode() is called to allocate a new
919 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
920 * file system gets to fill it in before unlocking it via unlock_new_inode().
922 * Note both @test and @set are called with the inode_lock held, so can't sleep.
924 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
925 int (*test)(struct inode *, void *),
926 int (*set)(struct inode *, void *), void *data)
928 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
931 inode = ifind(sb, head, test, data, 1);
935 * get_new_inode() will do the right thing, re-trying the search
936 * in case it had to block at any point.
938 return get_new_inode(sb, head, test, set, data);
941 EXPORT_SYMBOL(iget5_locked);
944 * iget_locked - obtain an inode from a mounted file system
945 * @sb: super block of file system
946 * @ino: inode number to get
948 * This is iget() without the read_inode() portion of get_new_inode_fast().
950 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
951 * the inode cache and if present it is returned with an increased reference
952 * count. This is for file systems where the inode number is sufficient for
953 * unique identification of an inode.
955 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
956 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
957 * The file system gets to fill it in before unlocking it via
958 * unlock_new_inode().
960 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
962 struct hlist_head *head = inode_hashtable + hash(sb, ino);
965 inode = ifind_fast(sb, head, ino);
969 * get_new_inode_fast() will do the right thing, re-trying the search
970 * in case it had to block at any point.
972 return get_new_inode_fast(sb, head, ino);
975 EXPORT_SYMBOL(iget_locked);
978 * __insert_inode_hash - hash an inode
979 * @inode: unhashed inode
980 * @hashval: unsigned long value used to locate this object in the
983 * Add an inode to the inode hash for this superblock.
985 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
987 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
988 spin_lock(&inode_lock);
989 hlist_add_head(&inode->i_hash, head);
990 spin_unlock(&inode_lock);
993 EXPORT_SYMBOL(__insert_inode_hash);
996 * remove_inode_hash - remove an inode from the hash
997 * @inode: inode to unhash
999 * Remove an inode from the superblock.
1001 void remove_inode_hash(struct inode *inode)
1003 spin_lock(&inode_lock);
1004 hlist_del_init(&inode->i_hash);
1005 spin_unlock(&inode_lock);
1008 EXPORT_SYMBOL(remove_inode_hash);
1011 * Tell the filesystem that this inode is no longer of any interest and should
1012 * be completely destroyed.
1014 * We leave the inode in the inode hash table until *after* the filesystem's
1015 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1016 * instigate) will always find up-to-date information either in the hash or on
1019 * I_FREEING is set so that no-one will take a new reference to the inode while
1020 * it is being deleted.
1022 void generic_delete_inode(struct inode *inode)
1024 const struct super_operations *op = inode->i_sb->s_op;
1026 list_del_init(&inode->i_list);
1027 list_del_init(&inode->i_sb_list);
1028 inode->i_state |= I_FREEING;
1029 inodes_stat.nr_inodes--;
1030 spin_unlock(&inode_lock);
1032 security_inode_delete(inode);
1034 if (op->delete_inode) {
1035 void (*delete)(struct inode *) = op->delete_inode;
1036 if (!is_bad_inode(inode))
1038 /* Filesystems implementing their own
1039 * s_op->delete_inode are required to call
1040 * truncate_inode_pages and clear_inode()
1044 truncate_inode_pages(&inode->i_data, 0);
1047 spin_lock(&inode_lock);
1048 hlist_del_init(&inode->i_hash);
1049 spin_unlock(&inode_lock);
1050 wake_up_inode(inode);
1051 BUG_ON(inode->i_state != I_CLEAR);
1052 destroy_inode(inode);
1055 EXPORT_SYMBOL(generic_delete_inode);
1057 static void generic_forget_inode(struct inode *inode)
1059 struct super_block *sb = inode->i_sb;
1061 if (!hlist_unhashed(&inode->i_hash)) {
1062 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1063 list_move(&inode->i_list, &inode_unused);
1064 inodes_stat.nr_unused++;
1065 if (sb->s_flags & MS_ACTIVE) {
1066 spin_unlock(&inode_lock);
1069 inode->i_state |= I_WILL_FREE;
1070 spin_unlock(&inode_lock);
1071 write_inode_now(inode, 1);
1072 spin_lock(&inode_lock);
1073 inode->i_state &= ~I_WILL_FREE;
1074 inodes_stat.nr_unused--;
1075 hlist_del_init(&inode->i_hash);
1077 list_del_init(&inode->i_list);
1078 list_del_init(&inode->i_sb_list);
1079 inode->i_state |= I_FREEING;
1080 inodes_stat.nr_inodes--;
1081 spin_unlock(&inode_lock);
1082 if (inode->i_data.nrpages)
1083 truncate_inode_pages(&inode->i_data, 0);
1085 wake_up_inode(inode);
1086 destroy_inode(inode);
1090 * Normal UNIX filesystem behaviour: delete the
1091 * inode when the usage count drops to zero, and
1094 void generic_drop_inode(struct inode *inode)
1096 if (!inode->i_nlink)
1097 generic_delete_inode(inode);
1099 generic_forget_inode(inode);
1102 EXPORT_SYMBOL_GPL(generic_drop_inode);
1105 * Called when we're dropping the last reference
1108 * Call the FS "drop()" function, defaulting to
1109 * the legacy UNIX filesystem behaviour..
1111 * NOTE! NOTE! NOTE! We're called with the inode lock
1112 * held, and the drop function is supposed to release
1115 static inline void iput_final(struct inode *inode)
1117 const struct super_operations *op = inode->i_sb->s_op;
1118 void (*drop)(struct inode *) = generic_drop_inode;
1120 if (op && op->drop_inode)
1121 drop = op->drop_inode;
1126 * iput - put an inode
1127 * @inode: inode to put
1129 * Puts an inode, dropping its usage count. If the inode use count hits
1130 * zero, the inode is then freed and may also be destroyed.
1132 * Consequently, iput() can sleep.
1134 void iput(struct inode *inode)
1137 const struct super_operations *op = inode->i_sb->s_op;
1139 BUG_ON(inode->i_state == I_CLEAR);
1141 if (op && op->put_inode)
1142 op->put_inode(inode);
1144 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1149 EXPORT_SYMBOL(iput);
1152 * bmap - find a block number in a file
1153 * @inode: inode of file
1154 * @block: block to find
1156 * Returns the block number on the device holding the inode that
1157 * is the disk block number for the block of the file requested.
1158 * That is, asked for block 4 of inode 1 the function will return the
1159 * disk block relative to the disk start that holds that block of the
1162 sector_t bmap(struct inode * inode, sector_t block)
1165 if (inode->i_mapping->a_ops->bmap)
1166 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1169 EXPORT_SYMBOL(bmap);
1172 * touch_atime - update the access time
1173 * @mnt: mount the inode is accessed on
1174 * @dentry: dentry accessed
1176 * Update the accessed time on an inode and mark it for writeback.
1177 * This function automatically handles read only file systems and media,
1178 * as well as the "noatime" flag and inode specific "noatime" markers.
1180 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1182 struct inode *inode = dentry->d_inode;
1183 struct timespec now;
1185 if (inode->i_flags & S_NOATIME)
1187 if (IS_NOATIME(inode))
1189 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1193 * We may have a NULL vfsmount when coming from NFSD
1196 if (mnt->mnt_flags & MNT_NOATIME)
1198 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1201 if (mnt->mnt_flags & MNT_RELATIME) {
1203 * With relative atime, only update atime if the
1204 * previous atime is earlier than either the ctime or
1207 if (timespec_compare(&inode->i_mtime,
1208 &inode->i_atime) < 0 &&
1209 timespec_compare(&inode->i_ctime,
1210 &inode->i_atime) < 0)
1215 now = current_fs_time(inode->i_sb);
1216 if (timespec_equal(&inode->i_atime, &now))
1219 inode->i_atime = now;
1220 mark_inode_dirty_sync(inode);
1222 EXPORT_SYMBOL(touch_atime);
1225 * file_update_time - update mtime and ctime time
1226 * @file: file accessed
1228 * Update the mtime and ctime members of an inode and mark the inode
1229 * for writeback. Note that this function is meant exclusively for
1230 * usage in the file write path of filesystems, and filesystems may
1231 * choose to explicitly ignore update via this function with the
1232 * S_NOCTIME inode flag, e.g. for network filesystem where these
1233 * timestamps are handled by the server.
1236 void file_update_time(struct file *file)
1238 struct inode *inode = file->f_path.dentry->d_inode;
1239 struct timespec now;
1242 if (IS_NOCMTIME(inode))
1244 if (IS_RDONLY(inode))
1247 now = current_fs_time(inode->i_sb);
1248 if (!timespec_equal(&inode->i_mtime, &now)) {
1249 inode->i_mtime = now;
1253 if (!timespec_equal(&inode->i_ctime, &now)) {
1254 inode->i_ctime = now;
1259 mark_inode_dirty_sync(inode);
1262 EXPORT_SYMBOL(file_update_time);
1264 int inode_needs_sync(struct inode *inode)
1268 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1273 EXPORT_SYMBOL(inode_needs_sync);
1275 int inode_wait(void *word)
1282 * If we try to find an inode in the inode hash while it is being
1283 * deleted, we have to wait until the filesystem completes its
1284 * deletion before reporting that it isn't found. This function waits
1285 * until the deletion _might_ have completed. Callers are responsible
1286 * to recheck inode state.
1288 * It doesn't matter if I_LOCK is not set initially, a call to
1289 * wake_up_inode() after removing from the hash list will DTRT.
1291 * This is called with inode_lock held.
1293 static void __wait_on_freeing_inode(struct inode *inode)
1295 wait_queue_head_t *wq;
1296 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1297 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1298 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1299 spin_unlock(&inode_lock);
1301 finish_wait(wq, &wait.wait);
1302 spin_lock(&inode_lock);
1305 void wake_up_inode(struct inode *inode)
1308 * Prevent speculative execution through spin_unlock(&inode_lock);
1311 wake_up_bit(&inode->i_state, __I_LOCK);
1315 * We rarely want to lock two inodes that do not have a parent/child
1316 * relationship (such as directory, child inode) simultaneously. The
1317 * vast majority of file systems should be able to get along fine
1318 * without this. Do not use these functions except as a last resort.
1320 void inode_double_lock(struct inode *inode1, struct inode *inode2)
1322 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1324 mutex_lock(&inode1->i_mutex);
1326 mutex_lock(&inode2->i_mutex);
1330 if (inode1 < inode2) {
1331 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1332 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1334 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1335 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1338 EXPORT_SYMBOL(inode_double_lock);
1340 void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1343 mutex_unlock(&inode1->i_mutex);
1345 if (inode2 && inode2 != inode1)
1346 mutex_unlock(&inode2->i_mutex);
1348 EXPORT_SYMBOL(inode_double_unlock);
1350 static __initdata unsigned long ihash_entries;
1351 static int __init set_ihash_entries(char *str)
1355 ihash_entries = simple_strtoul(str, &str, 0);
1358 __setup("ihash_entries=", set_ihash_entries);
1361 * Initialize the waitqueues and inode hash table.
1363 void __init inode_init_early(void)
1367 /* If hashes are distributed across NUMA nodes, defer
1368 * hash allocation until vmalloc space is available.
1374 alloc_large_system_hash("Inode-cache",
1375 sizeof(struct hlist_head),
1383 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1384 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1387 void __init inode_init(unsigned long mempages)
1391 /* inode slab cache */
1392 inode_cachep = kmem_cache_create("inode_cache",
1393 sizeof(struct inode),
1395 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1398 register_shrinker(&icache_shrinker);
1400 /* Hash may have been set up in inode_init_early */
1405 alloc_large_system_hash("Inode-cache",
1406 sizeof(struct hlist_head),
1414 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1415 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1418 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1420 inode->i_mode = mode;
1421 if (S_ISCHR(mode)) {
1422 inode->i_fop = &def_chr_fops;
1423 inode->i_rdev = rdev;
1424 } else if (S_ISBLK(mode)) {
1425 inode->i_fop = &def_blk_fops;
1426 inode->i_rdev = rdev;
1427 } else if (S_ISFIFO(mode))
1428 inode->i_fop = &def_fifo_fops;
1429 else if (S_ISSOCK(mode))
1430 inode->i_fop = &bad_sock_fops;
1432 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1435 EXPORT_SYMBOL(init_special_inode);