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;
119 /* essential because of inode slab reuse */
121 inode->i_blkbits = sb->s_blocksize_bits;
123 atomic_set(&inode->i_count, 1);
124 inode->i_op = &empty_iops;
125 inode->i_fop = &empty_fops;
127 atomic_set(&inode->i_writecount, 0);
131 inode->i_generation = 0;
133 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
135 inode->i_pipe = NULL;
136 inode->i_bdev = NULL;
137 inode->i_cdev = NULL;
139 inode->dirtied_when = 0;
140 if (security_inode_alloc(inode)) {
141 if (inode->i_sb->s_op->destroy_inode)
142 inode->i_sb->s_op->destroy_inode(inode);
144 kmem_cache_free(inode_cachep, (inode));
148 mapping->a_ops = &empty_aops;
149 mapping->host = inode;
151 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
152 mapping->assoc_mapping = NULL;
153 mapping->backing_dev_info = &default_backing_dev_info;
156 * If the block_device provides a backing_dev_info for client
157 * inodes then use that. Otherwise the inode share the bdev's
161 struct backing_dev_info *bdi;
163 bdi = sb->s_bdev->bd_inode_backing_dev_info;
165 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
166 mapping->backing_dev_info = bdi;
168 inode->i_private = NULL;
169 inode->i_mapping = mapping;
174 void destroy_inode(struct inode *inode)
176 BUG_ON(inode_has_buffers(inode));
177 security_inode_free(inode);
178 if (inode->i_sb->s_op->destroy_inode)
179 inode->i_sb->s_op->destroy_inode(inode);
181 kmem_cache_free(inode_cachep, (inode));
186 * These are initializations that only need to be done
187 * once, because the fields are idempotent across use
188 * of the inode, so let the slab aware of that.
190 void inode_init_once(struct inode *inode)
192 memset(inode, 0, sizeof(*inode));
193 INIT_HLIST_NODE(&inode->i_hash);
194 INIT_LIST_HEAD(&inode->i_dentry);
195 INIT_LIST_HEAD(&inode->i_devices);
196 mutex_init(&inode->i_mutex);
197 init_rwsem(&inode->i_alloc_sem);
198 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
199 rwlock_init(&inode->i_data.tree_lock);
200 spin_lock_init(&inode->i_data.i_mmap_lock);
201 INIT_LIST_HEAD(&inode->i_data.private_list);
202 spin_lock_init(&inode->i_data.private_lock);
203 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
204 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
205 spin_lock_init(&inode->i_lock);
206 i_size_ordered_init(inode);
207 #ifdef CONFIG_INOTIFY
208 INIT_LIST_HEAD(&inode->inotify_watches);
209 mutex_init(&inode->inotify_mutex);
213 EXPORT_SYMBOL(inode_init_once);
215 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
217 struct inode * inode = (struct inode *) foo;
219 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
220 SLAB_CTOR_CONSTRUCTOR)
221 inode_init_once(inode);
225 * inode_lock must be held
227 void __iget(struct inode * inode)
229 if (atomic_read(&inode->i_count)) {
230 atomic_inc(&inode->i_count);
233 atomic_inc(&inode->i_count);
234 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
235 list_move(&inode->i_list, &inode_in_use);
236 inodes_stat.nr_unused--;
239 EXPORT_SYMBOL_GPL(__iget);
242 * clear_inode - clear an inode
243 * @inode: inode to clear
245 * This is called by the filesystem to tell us
246 * that the inode is no longer useful. We just
247 * terminate it with extreme prejudice.
249 void clear_inode(struct inode *inode)
252 invalidate_inode_buffers(inode);
254 BUG_ON(inode->i_data.nrpages);
255 BUG_ON(!(inode->i_state & I_FREEING));
256 BUG_ON(inode->i_state & I_CLEAR);
257 wait_on_inode(inode);
259 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
260 inode->i_sb->s_op->clear_inode(inode);
261 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
263 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
265 inode->i_state = I_CLEAR;
268 EXPORT_SYMBOL(clear_inode);
271 * dispose_list - dispose of the contents of a local list
272 * @head: the head of the list to free
274 * Dispose-list gets a local list with local inodes in it, so it doesn't
275 * need to worry about list corruption and SMP locks.
277 static void dispose_list(struct list_head *head)
281 while (!list_empty(head)) {
284 inode = list_entry(head->next, struct inode, i_list);
285 list_del(&inode->i_list);
287 if (inode->i_data.nrpages)
288 truncate_inode_pages(&inode->i_data, 0);
291 spin_lock(&inode_lock);
292 hlist_del_init(&inode->i_hash);
293 list_del_init(&inode->i_sb_list);
294 spin_unlock(&inode_lock);
296 wake_up_inode(inode);
297 destroy_inode(inode);
300 spin_lock(&inode_lock);
301 inodes_stat.nr_inodes -= nr_disposed;
302 spin_unlock(&inode_lock);
306 * Invalidate all inodes for a device.
308 static int invalidate_list(struct list_head *head, struct list_head *dispose)
310 struct list_head *next;
311 int busy = 0, count = 0;
315 struct list_head * tmp = next;
316 struct inode * inode;
319 * We can reschedule here without worrying about the list's
320 * consistency because the per-sb list of inodes must not
321 * change during umount anymore, and because iprune_mutex keeps
322 * shrink_icache_memory() away.
324 cond_resched_lock(&inode_lock);
329 inode = list_entry(tmp, struct inode, i_sb_list);
330 invalidate_inode_buffers(inode);
331 if (!atomic_read(&inode->i_count)) {
332 list_move(&inode->i_list, dispose);
333 inode->i_state |= I_FREEING;
339 /* only unused inodes may be cached with i_count zero */
340 inodes_stat.nr_unused -= count;
345 * invalidate_inodes - discard the inodes on a device
348 * Discard all of the inodes for a given superblock. If the discard
349 * fails because there are busy inodes then a non zero value is returned.
350 * If the discard is successful all the inodes have been discarded.
352 int invalidate_inodes(struct super_block * sb)
355 LIST_HEAD(throw_away);
357 mutex_lock(&iprune_mutex);
358 spin_lock(&inode_lock);
359 inotify_unmount_inodes(&sb->s_inodes);
360 busy = invalidate_list(&sb->s_inodes, &throw_away);
361 spin_unlock(&inode_lock);
363 dispose_list(&throw_away);
364 mutex_unlock(&iprune_mutex);
369 EXPORT_SYMBOL(invalidate_inodes);
371 static int can_unuse(struct inode *inode)
375 if (inode_has_buffers(inode))
377 if (atomic_read(&inode->i_count))
379 if (inode->i_data.nrpages)
385 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
386 * a temporary list and then are freed outside inode_lock by dispose_list().
388 * Any inodes which are pinned purely because of attached pagecache have their
389 * pagecache removed. We expect the final iput() on that inode to add it to
390 * the front of the inode_unused list. So look for it there and if the
391 * inode is still freeable, proceed. The right inode is found 99.9% of the
392 * time in testing on a 4-way.
394 * If the inode has metadata buffers attached to mapping->private_list then
395 * try to remove them.
397 static void prune_icache(int nr_to_scan)
402 unsigned long reap = 0;
404 mutex_lock(&iprune_mutex);
405 spin_lock(&inode_lock);
406 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
409 if (list_empty(&inode_unused))
412 inode = list_entry(inode_unused.prev, struct inode, i_list);
414 if (inode->i_state || atomic_read(&inode->i_count)) {
415 list_move(&inode->i_list, &inode_unused);
418 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
420 spin_unlock(&inode_lock);
421 if (remove_inode_buffers(inode))
422 reap += invalidate_inode_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 void __wait_on_freeing_inode(struct inode *inode);
473 * Called with the inode lock held.
474 * NOTE: we are not increasing the inode-refcount, you must call __iget()
475 * by hand after calling find_inode now! This simplifies iunique and won't
476 * add any additional branch in the common code.
478 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
480 struct hlist_node *node;
481 struct inode * inode = NULL;
484 hlist_for_each (node, head) {
485 inode = hlist_entry(node, struct inode, i_hash);
486 if (inode->i_sb != sb)
488 if (!test(inode, data))
490 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
491 __wait_on_freeing_inode(inode);
496 return node ? inode : NULL;
500 * find_inode_fast is the fast path version of find_inode, see the comment at
501 * iget_locked for details.
503 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
505 struct hlist_node *node;
506 struct inode * inode = NULL;
509 hlist_for_each (node, head) {
510 inode = hlist_entry(node, struct inode, i_hash);
511 if (inode->i_ino != ino)
513 if (inode->i_sb != sb)
515 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
516 __wait_on_freeing_inode(inode);
521 return node ? inode : NULL;
525 * new_inode - obtain an inode
528 * Allocates a new inode for given superblock.
530 struct inode *new_inode(struct super_block *sb)
532 static unsigned long last_ino;
533 struct inode * inode;
535 spin_lock_prefetch(&inode_lock);
537 inode = alloc_inode(sb);
539 spin_lock(&inode_lock);
540 inodes_stat.nr_inodes++;
541 list_add(&inode->i_list, &inode_in_use);
542 list_add(&inode->i_sb_list, &sb->s_inodes);
543 inode->i_ino = ++last_ino;
545 spin_unlock(&inode_lock);
550 EXPORT_SYMBOL(new_inode);
552 void unlock_new_inode(struct inode *inode)
555 * This is special! We do not need the spinlock
556 * when clearing I_LOCK, because we're guaranteed
557 * that nobody else tries to do anything about the
558 * state of the inode when it is locked, as we
559 * just created it (so there can be no old holders
560 * that haven't tested I_LOCK).
562 inode->i_state &= ~(I_LOCK|I_NEW);
563 wake_up_inode(inode);
566 EXPORT_SYMBOL(unlock_new_inode);
569 * This is called without the inode lock held.. Be careful.
571 * We no longer cache the sb_flags in i_flags - see fs.h
572 * -- rmk@arm.uk.linux.org
574 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)
576 struct inode * inode;
578 inode = alloc_inode(sb);
582 spin_lock(&inode_lock);
583 /* We released the lock, so.. */
584 old = find_inode(sb, head, test, data);
586 if (set(inode, data))
589 inodes_stat.nr_inodes++;
590 list_add(&inode->i_list, &inode_in_use);
591 list_add(&inode->i_sb_list, &sb->s_inodes);
592 hlist_add_head(&inode->i_hash, head);
593 inode->i_state = I_LOCK|I_NEW;
594 spin_unlock(&inode_lock);
596 /* Return the locked inode with I_NEW set, the
597 * caller is responsible for filling in the contents
603 * Uhhuh, somebody else created the same inode under
604 * us. Use the old inode instead of the one we just
608 spin_unlock(&inode_lock);
609 destroy_inode(inode);
611 wait_on_inode(inode);
616 spin_unlock(&inode_lock);
617 destroy_inode(inode);
622 * get_new_inode_fast is the fast path version of get_new_inode, see the
623 * comment at iget_locked for details.
625 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
627 struct inode * inode;
629 inode = alloc_inode(sb);
633 spin_lock(&inode_lock);
634 /* We released the lock, so.. */
635 old = find_inode_fast(sb, head, ino);
638 inodes_stat.nr_inodes++;
639 list_add(&inode->i_list, &inode_in_use);
640 list_add(&inode->i_sb_list, &sb->s_inodes);
641 hlist_add_head(&inode->i_hash, head);
642 inode->i_state = I_LOCK|I_NEW;
643 spin_unlock(&inode_lock);
645 /* Return the locked inode with I_NEW set, the
646 * caller is responsible for filling in the contents
652 * Uhhuh, somebody else created the same inode under
653 * us. Use the old inode instead of the one we just
657 spin_unlock(&inode_lock);
658 destroy_inode(inode);
660 wait_on_inode(inode);
665 static unsigned long hash(struct super_block *sb, unsigned long hashval)
669 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
671 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
672 return tmp & I_HASHMASK;
676 * iunique - get a unique inode number
678 * @max_reserved: highest reserved inode number
680 * Obtain an inode number that is unique on the system for a given
681 * superblock. This is used by file systems that have no natural
682 * permanent inode numbering system. An inode number is returned that
683 * is higher than the reserved limit but unique.
686 * With a large number of inodes live on the file system this function
687 * currently becomes quite slow.
689 ino_t iunique(struct super_block *sb, ino_t max_reserved)
691 static ino_t counter;
693 struct hlist_head * head;
695 spin_lock(&inode_lock);
697 if (counter > max_reserved) {
698 head = inode_hashtable + hash(sb,counter);
700 inode = find_inode_fast(sb, head, res);
702 spin_unlock(&inode_lock);
706 counter = max_reserved + 1;
712 EXPORT_SYMBOL(iunique);
714 struct inode *igrab(struct inode *inode)
716 spin_lock(&inode_lock);
717 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
721 * Handle the case where s_op->clear_inode is not been
722 * called yet, and somebody is calling igrab
723 * while the inode is getting freed.
726 spin_unlock(&inode_lock);
730 EXPORT_SYMBOL(igrab);
733 * ifind - internal function, you want ilookup5() or iget5().
734 * @sb: super block of file system to search
735 * @head: the head of the list to search
736 * @test: callback used for comparisons between inodes
737 * @data: opaque data pointer to pass to @test
738 * @wait: if true wait for the inode to be unlocked, if false do not
740 * ifind() searches for the inode specified by @data in the inode
741 * cache. This is a generalized version of ifind_fast() for file systems where
742 * the inode number is not sufficient for unique identification of an inode.
744 * If the inode is in the cache, the inode is returned with an incremented
747 * Otherwise NULL is returned.
749 * Note, @test is called with the inode_lock held, so can't sleep.
751 static struct inode *ifind(struct super_block *sb,
752 struct hlist_head *head, int (*test)(struct inode *, void *),
753 void *data, const int wait)
757 spin_lock(&inode_lock);
758 inode = find_inode(sb, head, test, data);
761 spin_unlock(&inode_lock);
763 wait_on_inode(inode);
766 spin_unlock(&inode_lock);
771 * ifind_fast - internal function, you want ilookup() or iget().
772 * @sb: super block of file system to search
773 * @head: head of the list to search
774 * @ino: inode number to search for
776 * ifind_fast() searches for the inode @ino in the inode cache. This is for
777 * file systems where the inode number is sufficient for unique identification
780 * If the inode is in the cache, the inode is returned with an incremented
783 * Otherwise NULL is returned.
785 static struct inode *ifind_fast(struct super_block *sb,
786 struct hlist_head *head, unsigned long ino)
790 spin_lock(&inode_lock);
791 inode = find_inode_fast(sb, head, ino);
794 spin_unlock(&inode_lock);
795 wait_on_inode(inode);
798 spin_unlock(&inode_lock);
803 * ilookup5_nowait - search for an inode in the inode cache
804 * @sb: super block of file system to search
805 * @hashval: hash value (usually inode number) to search for
806 * @test: callback used for comparisons between inodes
807 * @data: opaque data pointer to pass to @test
809 * ilookup5() uses ifind() to search for the inode specified by @hashval and
810 * @data in the inode cache. This is a generalized version of ilookup() for
811 * file systems where the inode number is not sufficient for unique
812 * identification of an inode.
814 * If the inode is in the cache, the inode is returned with an incremented
815 * reference count. Note, the inode lock is not waited upon so you have to be
816 * very careful what you do with the returned inode. You probably should be
817 * using ilookup5() instead.
819 * Otherwise NULL is returned.
821 * Note, @test is called with the inode_lock held, so can't sleep.
823 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
824 int (*test)(struct inode *, void *), void *data)
826 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
828 return ifind(sb, head, test, data, 0);
831 EXPORT_SYMBOL(ilookup5_nowait);
834 * ilookup5 - search for an inode in the inode cache
835 * @sb: super block of file system to search
836 * @hashval: hash value (usually inode number) to search for
837 * @test: callback used for comparisons between inodes
838 * @data: opaque data pointer to pass to @test
840 * ilookup5() uses ifind() to search for the inode specified by @hashval and
841 * @data in the inode cache. This is a generalized version of ilookup() for
842 * file systems where the inode number is not sufficient for unique
843 * identification of an inode.
845 * If the inode is in the cache, the inode lock is waited upon and the inode is
846 * returned with an incremented reference count.
848 * Otherwise NULL is returned.
850 * Note, @test is called with the inode_lock held, so can't sleep.
852 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
853 int (*test)(struct inode *, void *), void *data)
855 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
857 return ifind(sb, head, test, data, 1);
860 EXPORT_SYMBOL(ilookup5);
863 * ilookup - search for an inode in the inode cache
864 * @sb: super block of file system to search
865 * @ino: inode number to search for
867 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
868 * This is for file systems where the inode number is sufficient for unique
869 * identification of an inode.
871 * If the inode is in the cache, the inode is returned with an incremented
874 * Otherwise NULL is returned.
876 struct inode *ilookup(struct super_block *sb, unsigned long ino)
878 struct hlist_head *head = inode_hashtable + hash(sb, ino);
880 return ifind_fast(sb, head, ino);
883 EXPORT_SYMBOL(ilookup);
886 * iget5_locked - obtain an inode from a mounted file system
887 * @sb: super block of file system
888 * @hashval: hash value (usually inode number) to get
889 * @test: callback used for comparisons between inodes
890 * @set: callback used to initialize a new struct inode
891 * @data: opaque data pointer to pass to @test and @set
893 * This is iget() without the read_inode() portion of get_new_inode().
895 * iget5_locked() uses ifind() to search for the inode specified by @hashval
896 * and @data in the inode cache and if present it is returned with an increased
897 * reference count. This is a generalized version of iget_locked() for file
898 * systems where the inode number is not sufficient for unique identification
901 * If the inode is not in cache, get_new_inode() is called to allocate a new
902 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
903 * file system gets to fill it in before unlocking it via unlock_new_inode().
905 * Note both @test and @set are called with the inode_lock held, so can't sleep.
907 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
908 int (*test)(struct inode *, void *),
909 int (*set)(struct inode *, void *), void *data)
911 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
914 inode = ifind(sb, head, test, data, 1);
918 * get_new_inode() will do the right thing, re-trying the search
919 * in case it had to block at any point.
921 return get_new_inode(sb, head, test, set, data);
924 EXPORT_SYMBOL(iget5_locked);
927 * iget_locked - obtain an inode from a mounted file system
928 * @sb: super block of file system
929 * @ino: inode number to get
931 * This is iget() without the read_inode() portion of get_new_inode_fast().
933 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
934 * the inode cache and if present it is returned with an increased reference
935 * count. This is for file systems where the inode number is sufficient for
936 * unique identification of an inode.
938 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
939 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
940 * The file system gets to fill it in before unlocking it via
941 * unlock_new_inode().
943 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
945 struct hlist_head *head = inode_hashtable + hash(sb, ino);
948 inode = ifind_fast(sb, head, ino);
952 * get_new_inode_fast() will do the right thing, re-trying the search
953 * in case it had to block at any point.
955 return get_new_inode_fast(sb, head, ino);
958 EXPORT_SYMBOL(iget_locked);
961 * __insert_inode_hash - hash an inode
962 * @inode: unhashed inode
963 * @hashval: unsigned long value used to locate this object in the
966 * Add an inode to the inode hash for this superblock.
968 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
970 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
971 spin_lock(&inode_lock);
972 hlist_add_head(&inode->i_hash, head);
973 spin_unlock(&inode_lock);
976 EXPORT_SYMBOL(__insert_inode_hash);
979 * remove_inode_hash - remove an inode from the hash
980 * @inode: inode to unhash
982 * Remove an inode from the superblock.
984 void remove_inode_hash(struct inode *inode)
986 spin_lock(&inode_lock);
987 hlist_del_init(&inode->i_hash);
988 spin_unlock(&inode_lock);
991 EXPORT_SYMBOL(remove_inode_hash);
994 * Tell the filesystem that this inode is no longer of any interest and should
995 * be completely destroyed.
997 * We leave the inode in the inode hash table until *after* the filesystem's
998 * ->delete_inode completes. This ensures that an iget (such as nfsd might
999 * instigate) will always find up-to-date information either in the hash or on
1002 * I_FREEING is set so that no-one will take a new reference to the inode while
1003 * it is being deleted.
1005 void generic_delete_inode(struct inode *inode)
1007 struct super_operations *op = inode->i_sb->s_op;
1009 list_del_init(&inode->i_list);
1010 list_del_init(&inode->i_sb_list);
1011 inode->i_state |= I_FREEING;
1012 inodes_stat.nr_inodes--;
1013 spin_unlock(&inode_lock);
1015 security_inode_delete(inode);
1017 if (op->delete_inode) {
1018 void (*delete)(struct inode *) = op->delete_inode;
1019 if (!is_bad_inode(inode))
1021 /* Filesystems implementing their own
1022 * s_op->delete_inode are required to call
1023 * truncate_inode_pages and clear_inode()
1027 truncate_inode_pages(&inode->i_data, 0);
1030 spin_lock(&inode_lock);
1031 hlist_del_init(&inode->i_hash);
1032 spin_unlock(&inode_lock);
1033 wake_up_inode(inode);
1034 BUG_ON(inode->i_state != I_CLEAR);
1035 destroy_inode(inode);
1038 EXPORT_SYMBOL(generic_delete_inode);
1040 static void generic_forget_inode(struct inode *inode)
1042 struct super_block *sb = inode->i_sb;
1044 if (!hlist_unhashed(&inode->i_hash)) {
1045 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1046 list_move(&inode->i_list, &inode_unused);
1047 inodes_stat.nr_unused++;
1048 if (!sb || (sb->s_flags & MS_ACTIVE)) {
1049 spin_unlock(&inode_lock);
1052 inode->i_state |= I_WILL_FREE;
1053 spin_unlock(&inode_lock);
1054 write_inode_now(inode, 1);
1055 spin_lock(&inode_lock);
1056 inode->i_state &= ~I_WILL_FREE;
1057 inodes_stat.nr_unused--;
1058 hlist_del_init(&inode->i_hash);
1060 list_del_init(&inode->i_list);
1061 list_del_init(&inode->i_sb_list);
1062 inode->i_state |= I_FREEING;
1063 inodes_stat.nr_inodes--;
1064 spin_unlock(&inode_lock);
1065 if (inode->i_data.nrpages)
1066 truncate_inode_pages(&inode->i_data, 0);
1068 wake_up_inode(inode);
1069 destroy_inode(inode);
1073 * Normal UNIX filesystem behaviour: delete the
1074 * inode when the usage count drops to zero, and
1077 void generic_drop_inode(struct inode *inode)
1079 if (!inode->i_nlink)
1080 generic_delete_inode(inode);
1082 generic_forget_inode(inode);
1085 EXPORT_SYMBOL_GPL(generic_drop_inode);
1088 * Called when we're dropping the last reference
1091 * Call the FS "drop()" function, defaulting to
1092 * the legacy UNIX filesystem behaviour..
1094 * NOTE! NOTE! NOTE! We're called with the inode lock
1095 * held, and the drop function is supposed to release
1098 static inline void iput_final(struct inode *inode)
1100 struct super_operations *op = inode->i_sb->s_op;
1101 void (*drop)(struct inode *) = generic_drop_inode;
1103 if (op && op->drop_inode)
1104 drop = op->drop_inode;
1109 * iput - put an inode
1110 * @inode: inode to put
1112 * Puts an inode, dropping its usage count. If the inode use count hits
1113 * zero, the inode is then freed and may also be destroyed.
1115 * Consequently, iput() can sleep.
1117 void iput(struct inode *inode)
1120 struct super_operations *op = inode->i_sb->s_op;
1122 BUG_ON(inode->i_state == I_CLEAR);
1124 if (op && op->put_inode)
1125 op->put_inode(inode);
1127 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1132 EXPORT_SYMBOL(iput);
1135 * bmap - find a block number in a file
1136 * @inode: inode of file
1137 * @block: block to find
1139 * Returns the block number on the device holding the inode that
1140 * is the disk block number for the block of the file requested.
1141 * That is, asked for block 4 of inode 1 the function will return the
1142 * disk block relative to the disk start that holds that block of the
1145 sector_t bmap(struct inode * inode, sector_t block)
1148 if (inode->i_mapping->a_ops->bmap)
1149 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1152 EXPORT_SYMBOL(bmap);
1155 * touch_atime - update the access time
1156 * @mnt: mount the inode is accessed on
1157 * @dentry: dentry accessed
1159 * Update the accessed time on an inode and mark it for writeback.
1160 * This function automatically handles read only file systems and media,
1161 * as well as the "noatime" flag and inode specific "noatime" markers.
1163 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1165 struct inode *inode = dentry->d_inode;
1166 struct timespec now;
1168 if (IS_RDONLY(inode))
1170 if (inode->i_flags & S_NOATIME)
1172 if (inode->i_sb->s_flags & MS_NOATIME)
1174 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1178 * We may have a NULL vfsmount when coming from NFSD
1181 if (mnt->mnt_flags & MNT_NOATIME)
1183 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1186 if (mnt->mnt_flags & MNT_RELATIME) {
1188 * With relative atime, only update atime if the
1189 * previous atime is earlier than either the ctime or
1192 if (timespec_compare(&inode->i_mtime,
1193 &inode->i_atime) < 0 &&
1194 timespec_compare(&inode->i_ctime,
1195 &inode->i_atime) < 0)
1200 now = current_fs_time(inode->i_sb);
1201 if (timespec_equal(&inode->i_atime, &now))
1204 inode->i_atime = now;
1205 mark_inode_dirty_sync(inode);
1207 EXPORT_SYMBOL(touch_atime);
1210 * file_update_time - update mtime and ctime time
1211 * @file: file accessed
1213 * Update the mtime and ctime members of an inode and mark the inode
1214 * for writeback. Note that this function is meant exclusively for
1215 * usage in the file write path of filesystems, and filesystems may
1216 * choose to explicitly ignore update via this function with the
1217 * S_NOCTIME inode flag, e.g. for network filesystem where these
1218 * timestamps are handled by the server.
1221 void file_update_time(struct file *file)
1223 struct inode *inode = file->f_path.dentry->d_inode;
1224 struct timespec now;
1227 if (IS_NOCMTIME(inode))
1229 if (IS_RDONLY(inode))
1232 now = current_fs_time(inode->i_sb);
1233 if (!timespec_equal(&inode->i_mtime, &now)) {
1234 inode->i_mtime = now;
1238 if (!timespec_equal(&inode->i_ctime, &now)) {
1239 inode->i_ctime = now;
1244 mark_inode_dirty_sync(inode);
1247 EXPORT_SYMBOL(file_update_time);
1249 int inode_needs_sync(struct inode *inode)
1253 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1258 EXPORT_SYMBOL(inode_needs_sync);
1261 * Quota functions that want to walk the inode lists..
1265 void remove_dquot_ref(struct super_block *sb, int type,
1266 struct list_head *tofree_head)
1268 struct inode *inode;
1271 return; /* nothing to do */
1272 spin_lock(&inode_lock); /* This lock is for inodes code */
1275 * We don't have to lock against quota code - test IS_QUOTAINIT is
1276 * just for speedup...
1278 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1279 if (!IS_NOQUOTA(inode))
1280 remove_inode_dquot_ref(inode, type, tofree_head);
1282 spin_unlock(&inode_lock);
1287 int inode_wait(void *word)
1294 * If we try to find an inode in the inode hash while it is being
1295 * deleted, we have to wait until the filesystem completes its
1296 * deletion before reporting that it isn't found. This function waits
1297 * until the deletion _might_ have completed. Callers are responsible
1298 * to recheck inode state.
1300 * It doesn't matter if I_LOCK is not set initially, a call to
1301 * wake_up_inode() after removing from the hash list will DTRT.
1303 * This is called with inode_lock held.
1305 static void __wait_on_freeing_inode(struct inode *inode)
1307 wait_queue_head_t *wq;
1308 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1309 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1310 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1311 spin_unlock(&inode_lock);
1313 finish_wait(wq, &wait.wait);
1314 spin_lock(&inode_lock);
1317 void wake_up_inode(struct inode *inode)
1320 * Prevent speculative execution through spin_unlock(&inode_lock);
1323 wake_up_bit(&inode->i_state, __I_LOCK);
1327 * We rarely want to lock two inodes that do not have a parent/child
1328 * relationship (such as directory, child inode) simultaneously. The
1329 * vast majority of file systems should be able to get along fine
1330 * without this. Do not use these functions except as a last resort.
1332 void inode_double_lock(struct inode *inode1, struct inode *inode2)
1334 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1336 mutex_lock(&inode1->i_mutex);
1338 mutex_lock(&inode2->i_mutex);
1342 if (inode1 < inode2) {
1343 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1344 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1346 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1347 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1350 EXPORT_SYMBOL(inode_double_lock);
1352 void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1355 mutex_unlock(&inode1->i_mutex);
1357 if (inode2 && inode2 != inode1)
1358 mutex_unlock(&inode2->i_mutex);
1360 EXPORT_SYMBOL(inode_double_unlock);
1362 static __initdata unsigned long ihash_entries;
1363 static int __init set_ihash_entries(char *str)
1367 ihash_entries = simple_strtoul(str, &str, 0);
1370 __setup("ihash_entries=", set_ihash_entries);
1373 * Initialize the waitqueues and inode hash table.
1375 void __init inode_init_early(void)
1379 /* If hashes are distributed across NUMA nodes, defer
1380 * hash allocation until vmalloc space is available.
1386 alloc_large_system_hash("Inode-cache",
1387 sizeof(struct hlist_head),
1395 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1396 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1399 void __init inode_init(unsigned long mempages)
1403 /* inode slab cache */
1404 inode_cachep = kmem_cache_create("inode_cache",
1405 sizeof(struct inode),
1407 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1411 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1413 /* Hash may have been set up in inode_init_early */
1418 alloc_large_system_hash("Inode-cache",
1419 sizeof(struct hlist_head),
1427 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1428 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1431 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1433 inode->i_mode = mode;
1434 if (S_ISCHR(mode)) {
1435 inode->i_fop = &def_chr_fops;
1436 inode->i_rdev = rdev;
1437 } else if (S_ISBLK(mode)) {
1438 inode->i_fop = &def_blk_fops;
1439 inode->i_rdev = rdev;
1440 } else if (S_ISFIFO(mode))
1441 inode->i_fop = &def_fifo_fops;
1442 else if (S_ISSOCK(mode))
1443 inode->i_fop = &bad_sock_fops;
1445 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1448 EXPORT_SYMBOL(init_special_inode);