4 * (C) 1997 Linus Torvalds
7 #include <linux/config.h>
10 #include <linux/dcache.h>
11 #include <linux/init.h>
12 #include <linux/quotaops.h>
13 #include <linux/slab.h>
14 #include <linux/writeback.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/wait.h>
18 #include <linux/hash.h>
19 #include <linux/swap.h>
20 #include <linux/security.h>
21 #include <linux/pagemap.h>
22 #include <linux/cdev.h>
23 #include <linux/bootmem.h>
24 #include <linux/inotify.h>
25 #include <linux/mount.h>
28 * This is needed for the following functions:
30 * - invalidate_inode_buffers
33 * FIXME: remove all knowledge of the buffer layer from this file
35 #include <linux/buffer_head.h>
38 * New inode.c implementation.
40 * This implementation has the basic premise of trying
41 * to be extremely low-overhead and SMP-safe, yet be
42 * simple enough to be "obviously correct".
47 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
49 /* #define INODE_PARANOIA 1 */
50 /* #define INODE_DEBUG 1 */
53 * Inode lookup is no longer as critical as it used to be:
54 * most of the lookups are going to be through the dcache.
56 #define I_HASHBITS i_hash_shift
57 #define I_HASHMASK i_hash_mask
59 static unsigned int i_hash_mask __read_mostly;
60 static unsigned int i_hash_shift __read_mostly;
63 * Each inode can be on two separate lists. One is
64 * the hash list of the inode, used for lookups. The
65 * other linked list is the "type" list:
66 * "in_use" - valid inode, i_count > 0, i_nlink > 0
67 * "dirty" - as "in_use" but also dirty
68 * "unused" - valid inode, i_count = 0
70 * A "dirty" list is maintained for each super block,
71 * allowing for low-overhead inode sync() operations.
74 LIST_HEAD(inode_in_use);
75 LIST_HEAD(inode_unused);
76 static struct hlist_head *inode_hashtable __read_mostly;
79 * A simple spinlock to protect the list manipulations.
81 * NOTE! You also have to own the lock if you change
82 * the i_state of an inode while it is in use..
84 DEFINE_SPINLOCK(inode_lock);
87 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
88 * icache shrinking path, and the umount path. Without this exclusion,
89 * by the time prune_icache calls iput for the inode whose pages it has
90 * been invalidating, or by the time it calls clear_inode & destroy_inode
91 * from its final dispose_list, the struct super_block they refer to
92 * (for inode->i_sb->s_op) may already have been freed and reused.
94 static DEFINE_MUTEX(iprune_mutex);
97 * Statistics gathering..
99 struct inodes_stat_t inodes_stat;
101 static kmem_cache_t * inode_cachep __read_mostly;
103 static struct inode *alloc_inode(struct super_block *sb)
105 static struct address_space_operations empty_aops;
106 static struct inode_operations empty_iops;
107 static const struct file_operations empty_fops;
110 if (sb->s_op->alloc_inode)
111 inode = sb->s_op->alloc_inode(sb);
113 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
116 struct address_space * const mapping = &inode->i_data;
120 /* essential because of inode slab reuse */
122 inode->i_blkbits = sb->s_blocksize_bits;
124 atomic_set(&inode->i_count, 1);
125 inode->i_op = &empty_iops;
126 inode->i_fop = &empty_fops;
128 atomic_set(&inode->i_writecount, 0);
132 inode->i_generation = 0;
134 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
136 inode->i_pipe = NULL;
137 inode->i_bdev = NULL;
138 inode->i_cdev = NULL;
140 inode->i_security = NULL;
141 inode->dirtied_when = 0;
142 if (security_inode_alloc(inode)) {
143 if (inode->i_sb->s_op->destroy_inode)
144 inode->i_sb->s_op->destroy_inode(inode);
146 kmem_cache_free(inode_cachep, (inode));
150 mapping->a_ops = &empty_aops;
151 mapping->host = inode;
153 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
154 mapping->assoc_mapping = NULL;
155 mapping->backing_dev_info = &default_backing_dev_info;
158 * If the block_device provides a backing_dev_info for client
159 * inodes then use that. Otherwise the inode share the bdev's
163 struct backing_dev_info *bdi;
165 bdi = sb->s_bdev->bd_inode_backing_dev_info;
167 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
168 mapping->backing_dev_info = bdi;
170 memset(&inode->u, 0, sizeof(inode->u));
171 inode->i_mapping = mapping;
176 void destroy_inode(struct inode *inode)
178 BUG_ON(inode_has_buffers(inode));
179 security_inode_free(inode);
180 if (inode->i_sb->s_op->destroy_inode)
181 inode->i_sb->s_op->destroy_inode(inode);
183 kmem_cache_free(inode_cachep, (inode));
188 * These are initializations that only need to be done
189 * once, because the fields are idempotent across use
190 * of the inode, so let the slab aware of that.
192 void inode_init_once(struct inode *inode)
194 memset(inode, 0, sizeof(*inode));
195 INIT_HLIST_NODE(&inode->i_hash);
196 INIT_LIST_HEAD(&inode->i_dentry);
197 INIT_LIST_HEAD(&inode->i_devices);
198 mutex_init(&inode->i_mutex);
199 init_rwsem(&inode->i_alloc_sem);
200 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
201 rwlock_init(&inode->i_data.tree_lock);
202 spin_lock_init(&inode->i_data.i_mmap_lock);
203 INIT_LIST_HEAD(&inode->i_data.private_list);
204 spin_lock_init(&inode->i_data.private_lock);
205 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
206 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
207 spin_lock_init(&inode->i_lock);
208 i_size_ordered_init(inode);
209 #ifdef CONFIG_INOTIFY
210 INIT_LIST_HEAD(&inode->inotify_watches);
211 mutex_init(&inode->inotify_mutex);
215 EXPORT_SYMBOL(inode_init_once);
217 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
219 struct inode * inode = (struct inode *) foo;
221 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
222 SLAB_CTOR_CONSTRUCTOR)
223 inode_init_once(inode);
227 * inode_lock must be held
229 void __iget(struct inode * inode)
231 if (atomic_read(&inode->i_count)) {
232 atomic_inc(&inode->i_count);
235 atomic_inc(&inode->i_count);
236 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
237 list_move(&inode->i_list, &inode_in_use);
238 inodes_stat.nr_unused--;
241 EXPORT_SYMBOL_GPL(__iget);
244 * clear_inode - clear an inode
245 * @inode: inode to clear
247 * This is called by the filesystem to tell us
248 * that the inode is no longer useful. We just
249 * terminate it with extreme prejudice.
251 void clear_inode(struct inode *inode)
254 invalidate_inode_buffers(inode);
256 BUG_ON(inode->i_data.nrpages);
257 BUG_ON(!(inode->i_state & I_FREEING));
258 BUG_ON(inode->i_state & I_CLEAR);
259 wait_on_inode(inode);
261 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
262 inode->i_sb->s_op->clear_inode(inode);
267 inode->i_state = I_CLEAR;
270 EXPORT_SYMBOL(clear_inode);
273 * dispose_list - dispose of the contents of a local list
274 * @head: the head of the list to free
276 * Dispose-list gets a local list with local inodes in it, so it doesn't
277 * need to worry about list corruption and SMP locks.
279 static void dispose_list(struct list_head *head)
283 while (!list_empty(head)) {
286 inode = list_entry(head->next, struct inode, i_list);
287 list_del(&inode->i_list);
289 if (inode->i_data.nrpages)
290 truncate_inode_pages(&inode->i_data, 0);
293 spin_lock(&inode_lock);
294 hlist_del_init(&inode->i_hash);
295 list_del_init(&inode->i_sb_list);
296 spin_unlock(&inode_lock);
298 wake_up_inode(inode);
299 destroy_inode(inode);
302 spin_lock(&inode_lock);
303 inodes_stat.nr_inodes -= nr_disposed;
304 spin_unlock(&inode_lock);
308 * Invalidate all inodes for a device.
310 static int invalidate_list(struct list_head *head, struct list_head *dispose)
312 struct list_head *next;
313 int busy = 0, count = 0;
317 struct list_head * tmp = next;
318 struct inode * inode;
321 * We can reschedule here without worrying about the list's
322 * consistency because the per-sb list of inodes must not
323 * change during umount anymore, and because iprune_mutex keeps
324 * shrink_icache_memory() away.
326 cond_resched_lock(&inode_lock);
331 inode = list_entry(tmp, struct inode, i_sb_list);
332 invalidate_inode_buffers(inode);
333 if (!atomic_read(&inode->i_count)) {
334 list_move(&inode->i_list, dispose);
335 inode->i_state |= I_FREEING;
341 /* only unused inodes may be cached with i_count zero */
342 inodes_stat.nr_unused -= count;
347 * invalidate_inodes - discard the inodes on a device
350 * Discard all of the inodes for a given superblock. If the discard
351 * fails because there are busy inodes then a non zero value is returned.
352 * If the discard is successful all the inodes have been discarded.
354 int invalidate_inodes(struct super_block * sb)
357 LIST_HEAD(throw_away);
359 mutex_lock(&iprune_mutex);
360 spin_lock(&inode_lock);
361 inotify_unmount_inodes(&sb->s_inodes);
362 busy = invalidate_list(&sb->s_inodes, &throw_away);
363 spin_unlock(&inode_lock);
365 dispose_list(&throw_away);
366 mutex_unlock(&iprune_mutex);
371 EXPORT_SYMBOL(invalidate_inodes);
373 int __invalidate_device(struct block_device *bdev)
375 struct super_block *sb = get_super(bdev);
380 * no need to lock the super, get_super holds the
381 * read mutex so the filesystem cannot go away
382 * under us (->put_super runs with the write lock
385 shrink_dcache_sb(sb);
386 res = invalidate_inodes(sb);
389 invalidate_bdev(bdev, 0);
392 EXPORT_SYMBOL(__invalidate_device);
394 static int can_unuse(struct inode *inode)
398 if (inode_has_buffers(inode))
400 if (atomic_read(&inode->i_count))
402 if (inode->i_data.nrpages)
408 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
409 * a temporary list and then are freed outside inode_lock by dispose_list().
411 * Any inodes which are pinned purely because of attached pagecache have their
412 * pagecache removed. We expect the final iput() on that inode to add it to
413 * the front of the inode_unused list. So look for it there and if the
414 * inode is still freeable, proceed. The right inode is found 99.9% of the
415 * time in testing on a 4-way.
417 * If the inode has metadata buffers attached to mapping->private_list then
418 * try to remove them.
420 static void prune_icache(int nr_to_scan)
425 unsigned long reap = 0;
427 mutex_lock(&iprune_mutex);
428 spin_lock(&inode_lock);
429 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
432 if (list_empty(&inode_unused))
435 inode = list_entry(inode_unused.prev, struct inode, i_list);
437 if (inode->i_state || atomic_read(&inode->i_count)) {
438 list_move(&inode->i_list, &inode_unused);
441 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
443 spin_unlock(&inode_lock);
444 if (remove_inode_buffers(inode))
445 reap += invalidate_inode_pages(&inode->i_data);
447 spin_lock(&inode_lock);
449 if (inode != list_entry(inode_unused.next,
450 struct inode, i_list))
451 continue; /* wrong inode or list_empty */
452 if (!can_unuse(inode))
455 list_move(&inode->i_list, &freeable);
456 inode->i_state |= I_FREEING;
459 inodes_stat.nr_unused -= nr_pruned;
460 spin_unlock(&inode_lock);
462 dispose_list(&freeable);
463 mutex_unlock(&iprune_mutex);
465 if (current_is_kswapd())
466 mod_page_state(kswapd_inodesteal, reap);
468 mod_page_state(pginodesteal, reap);
472 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
473 * "unused" means that no dentries are referring to the inodes: the files are
474 * not open and the dcache references to those inodes have already been
477 * This function is passed the number of inodes to scan, and it returns the
478 * total number of remaining possibly-reclaimable inodes.
480 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
484 * Nasty deadlock avoidance. We may hold various FS locks,
485 * and we don't want to recurse into the FS that called us
486 * in clear_inode() and friends..
488 if (!(gfp_mask & __GFP_FS))
492 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
495 static void __wait_on_freeing_inode(struct inode *inode);
497 * Called with the inode lock held.
498 * NOTE: we are not increasing the inode-refcount, you must call __iget()
499 * by hand after calling find_inode now! This simplifies iunique and won't
500 * add any additional branch in the common code.
502 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
504 struct hlist_node *node;
505 struct inode * inode = NULL;
508 hlist_for_each (node, head) {
509 inode = hlist_entry(node, struct inode, i_hash);
510 if (inode->i_sb != sb)
512 if (!test(inode, data))
514 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
515 __wait_on_freeing_inode(inode);
520 return node ? inode : NULL;
524 * find_inode_fast is the fast path version of find_inode, see the comment at
525 * iget_locked for details.
527 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
529 struct hlist_node *node;
530 struct inode * inode = NULL;
533 hlist_for_each (node, head) {
534 inode = hlist_entry(node, struct inode, i_hash);
535 if (inode->i_ino != ino)
537 if (inode->i_sb != sb)
539 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
540 __wait_on_freeing_inode(inode);
545 return node ? inode : NULL;
549 * new_inode - obtain an inode
552 * Allocates a new inode for given superblock.
554 struct inode *new_inode(struct super_block *sb)
556 static unsigned long last_ino;
557 struct inode * inode;
559 spin_lock_prefetch(&inode_lock);
561 inode = alloc_inode(sb);
563 spin_lock(&inode_lock);
564 inodes_stat.nr_inodes++;
565 list_add(&inode->i_list, &inode_in_use);
566 list_add(&inode->i_sb_list, &sb->s_inodes);
567 inode->i_ino = ++last_ino;
569 spin_unlock(&inode_lock);
574 EXPORT_SYMBOL(new_inode);
576 void unlock_new_inode(struct inode *inode)
579 * This is special! We do not need the spinlock
580 * when clearing I_LOCK, because we're guaranteed
581 * that nobody else tries to do anything about the
582 * state of the inode when it is locked, as we
583 * just created it (so there can be no old holders
584 * that haven't tested I_LOCK).
586 inode->i_state &= ~(I_LOCK|I_NEW);
587 wake_up_inode(inode);
590 EXPORT_SYMBOL(unlock_new_inode);
593 * This is called without the inode lock held.. Be careful.
595 * We no longer cache the sb_flags in i_flags - see fs.h
596 * -- rmk@arm.uk.linux.org
598 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)
600 struct inode * inode;
602 inode = alloc_inode(sb);
606 spin_lock(&inode_lock);
607 /* We released the lock, so.. */
608 old = find_inode(sb, head, test, data);
610 if (set(inode, data))
613 inodes_stat.nr_inodes++;
614 list_add(&inode->i_list, &inode_in_use);
615 list_add(&inode->i_sb_list, &sb->s_inodes);
616 hlist_add_head(&inode->i_hash, head);
617 inode->i_state = I_LOCK|I_NEW;
618 spin_unlock(&inode_lock);
620 /* Return the locked inode with I_NEW set, the
621 * caller is responsible for filling in the contents
627 * Uhhuh, somebody else created the same inode under
628 * us. Use the old inode instead of the one we just
632 spin_unlock(&inode_lock);
633 destroy_inode(inode);
635 wait_on_inode(inode);
640 spin_unlock(&inode_lock);
641 destroy_inode(inode);
646 * get_new_inode_fast is the fast path version of get_new_inode, see the
647 * comment at iget_locked for details.
649 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
651 struct inode * inode;
653 inode = alloc_inode(sb);
657 spin_lock(&inode_lock);
658 /* We released the lock, so.. */
659 old = find_inode_fast(sb, head, ino);
662 inodes_stat.nr_inodes++;
663 list_add(&inode->i_list, &inode_in_use);
664 list_add(&inode->i_sb_list, &sb->s_inodes);
665 hlist_add_head(&inode->i_hash, head);
666 inode->i_state = I_LOCK|I_NEW;
667 spin_unlock(&inode_lock);
669 /* Return the locked inode with I_NEW set, the
670 * caller is responsible for filling in the contents
676 * Uhhuh, somebody else created the same inode under
677 * us. Use the old inode instead of the one we just
681 spin_unlock(&inode_lock);
682 destroy_inode(inode);
684 wait_on_inode(inode);
689 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
693 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
695 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
696 return tmp & I_HASHMASK;
700 * iunique - get a unique inode number
702 * @max_reserved: highest reserved inode number
704 * Obtain an inode number that is unique on the system for a given
705 * superblock. This is used by file systems that have no natural
706 * permanent inode numbering system. An inode number is returned that
707 * is higher than the reserved limit but unique.
710 * With a large number of inodes live on the file system this function
711 * currently becomes quite slow.
713 ino_t iunique(struct super_block *sb, ino_t max_reserved)
715 static ino_t counter;
717 struct hlist_head * head;
719 spin_lock(&inode_lock);
721 if (counter > max_reserved) {
722 head = inode_hashtable + hash(sb,counter);
724 inode = find_inode_fast(sb, head, res);
726 spin_unlock(&inode_lock);
730 counter = max_reserved + 1;
736 EXPORT_SYMBOL(iunique);
738 struct inode *igrab(struct inode *inode)
740 spin_lock(&inode_lock);
741 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
745 * Handle the case where s_op->clear_inode is not been
746 * called yet, and somebody is calling igrab
747 * while the inode is getting freed.
750 spin_unlock(&inode_lock);
754 EXPORT_SYMBOL(igrab);
757 * ifind - internal function, you want ilookup5() or iget5().
758 * @sb: super block of file system to search
759 * @head: the head of the list to search
760 * @test: callback used for comparisons between inodes
761 * @data: opaque data pointer to pass to @test
762 * @wait: if true wait for the inode to be unlocked, if false do not
764 * ifind() searches for the inode specified by @data in the inode
765 * cache. This is a generalized version of ifind_fast() for file systems where
766 * the inode number is not sufficient for unique identification of an inode.
768 * If the inode is in the cache, the inode is returned with an incremented
771 * Otherwise NULL is returned.
773 * Note, @test is called with the inode_lock held, so can't sleep.
775 static struct inode *ifind(struct super_block *sb,
776 struct hlist_head *head, int (*test)(struct inode *, void *),
777 void *data, const int wait)
781 spin_lock(&inode_lock);
782 inode = find_inode(sb, head, test, data);
785 spin_unlock(&inode_lock);
787 wait_on_inode(inode);
790 spin_unlock(&inode_lock);
795 * ifind_fast - internal function, you want ilookup() or iget().
796 * @sb: super block of file system to search
797 * @head: head of the list to search
798 * @ino: inode number to search for
800 * ifind_fast() searches for the inode @ino in the inode cache. This is for
801 * file systems where the inode number is sufficient for unique identification
804 * If the inode is in the cache, the inode is returned with an incremented
807 * Otherwise NULL is returned.
809 static struct inode *ifind_fast(struct super_block *sb,
810 struct hlist_head *head, unsigned long ino)
814 spin_lock(&inode_lock);
815 inode = find_inode_fast(sb, head, ino);
818 spin_unlock(&inode_lock);
819 wait_on_inode(inode);
822 spin_unlock(&inode_lock);
827 * ilookup5_nowait - search for an inode in the inode cache
828 * @sb: super block of file system to search
829 * @hashval: hash value (usually inode number) to search for
830 * @test: callback used for comparisons between inodes
831 * @data: opaque data pointer to pass to @test
833 * ilookup5() uses ifind() to search for the inode specified by @hashval and
834 * @data in the inode cache. This is a generalized version of ilookup() for
835 * file systems where the inode number is not sufficient for unique
836 * identification of an inode.
838 * If the inode is in the cache, the inode is returned with an incremented
839 * reference count. Note, the inode lock is not waited upon so you have to be
840 * very careful what you do with the returned inode. You probably should be
841 * using ilookup5() instead.
843 * Otherwise NULL is returned.
845 * Note, @test is called with the inode_lock held, so can't sleep.
847 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
848 int (*test)(struct inode *, void *), void *data)
850 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
852 return ifind(sb, head, test, data, 0);
855 EXPORT_SYMBOL(ilookup5_nowait);
858 * ilookup5 - search for an inode in the inode cache
859 * @sb: super block of file system to search
860 * @hashval: hash value (usually inode number) to search for
861 * @test: callback used for comparisons between inodes
862 * @data: opaque data pointer to pass to @test
864 * ilookup5() uses ifind() to search for the inode specified by @hashval and
865 * @data in the inode cache. This is a generalized version of ilookup() for
866 * file systems where the inode number is not sufficient for unique
867 * identification of an inode.
869 * If the inode is in the cache, the inode lock is waited upon and the inode is
870 * returned with an incremented reference count.
872 * Otherwise NULL is returned.
874 * Note, @test is called with the inode_lock held, so can't sleep.
876 struct inode *ilookup5(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, 1);
884 EXPORT_SYMBOL(ilookup5);
887 * ilookup - search for an inode in the inode cache
888 * @sb: super block of file system to search
889 * @ino: inode number to search for
891 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
892 * This is for file systems where the inode number is sufficient for unique
893 * identification of an inode.
895 * If the inode is in the cache, the inode is returned with an incremented
898 * Otherwise NULL is returned.
900 struct inode *ilookup(struct super_block *sb, unsigned long ino)
902 struct hlist_head *head = inode_hashtable + hash(sb, ino);
904 return ifind_fast(sb, head, ino);
907 EXPORT_SYMBOL(ilookup);
910 * iget5_locked - obtain an inode from a mounted file system
911 * @sb: super block of file system
912 * @hashval: hash value (usually inode number) to get
913 * @test: callback used for comparisons between inodes
914 * @set: callback used to initialize a new struct inode
915 * @data: opaque data pointer to pass to @test and @set
917 * This is iget() without the read_inode() portion of get_new_inode().
919 * iget5_locked() uses ifind() to search for the inode specified by @hashval
920 * and @data in the inode cache and if present it is returned with an increased
921 * reference count. This is a generalized version of iget_locked() for file
922 * systems where the inode number is not sufficient for unique identification
925 * If the inode is not in cache, get_new_inode() is called to allocate a new
926 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
927 * file system gets to fill it in before unlocking it via unlock_new_inode().
929 * Note both @test and @set are called with the inode_lock held, so can't sleep.
931 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
932 int (*test)(struct inode *, void *),
933 int (*set)(struct inode *, void *), void *data)
935 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
938 inode = ifind(sb, head, test, data, 1);
942 * get_new_inode() will do the right thing, re-trying the search
943 * in case it had to block at any point.
945 return get_new_inode(sb, head, test, set, data);
948 EXPORT_SYMBOL(iget5_locked);
951 * iget_locked - obtain an inode from a mounted file system
952 * @sb: super block of file system
953 * @ino: inode number to get
955 * This is iget() without the read_inode() portion of get_new_inode_fast().
957 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
958 * the inode cache and if present it is returned with an increased reference
959 * count. This is for file systems where the inode number is sufficient for
960 * unique identification of an inode.
962 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
963 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
964 * The file system gets to fill it in before unlocking it via
965 * unlock_new_inode().
967 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
969 struct hlist_head *head = inode_hashtable + hash(sb, ino);
972 inode = ifind_fast(sb, head, ino);
976 * get_new_inode_fast() will do the right thing, re-trying the search
977 * in case it had to block at any point.
979 return get_new_inode_fast(sb, head, ino);
982 EXPORT_SYMBOL(iget_locked);
985 * __insert_inode_hash - hash an inode
986 * @inode: unhashed inode
987 * @hashval: unsigned long value used to locate this object in the
990 * Add an inode to the inode hash for this superblock.
992 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
994 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
995 spin_lock(&inode_lock);
996 hlist_add_head(&inode->i_hash, head);
997 spin_unlock(&inode_lock);
1000 EXPORT_SYMBOL(__insert_inode_hash);
1003 * remove_inode_hash - remove an inode from the hash
1004 * @inode: inode to unhash
1006 * Remove an inode from the superblock.
1008 void remove_inode_hash(struct inode *inode)
1010 spin_lock(&inode_lock);
1011 hlist_del_init(&inode->i_hash);
1012 spin_unlock(&inode_lock);
1015 EXPORT_SYMBOL(remove_inode_hash);
1018 * Tell the filesystem that this inode is no longer of any interest and should
1019 * be completely destroyed.
1021 * We leave the inode in the inode hash table until *after* the filesystem's
1022 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1023 * instigate) will always find up-to-date information either in the hash or on
1026 * I_FREEING is set so that no-one will take a new reference to the inode while
1027 * it is being deleted.
1029 void generic_delete_inode(struct inode *inode)
1031 struct super_operations *op = inode->i_sb->s_op;
1033 list_del_init(&inode->i_list);
1034 list_del_init(&inode->i_sb_list);
1035 inode->i_state|=I_FREEING;
1036 inodes_stat.nr_inodes--;
1037 spin_unlock(&inode_lock);
1039 security_inode_delete(inode);
1041 if (op->delete_inode) {
1042 void (*delete)(struct inode *) = op->delete_inode;
1043 if (!is_bad_inode(inode))
1045 /* Filesystems implementing their own
1046 * s_op->delete_inode are required to call
1047 * truncate_inode_pages and clear_inode()
1051 truncate_inode_pages(&inode->i_data, 0);
1054 spin_lock(&inode_lock);
1055 hlist_del_init(&inode->i_hash);
1056 spin_unlock(&inode_lock);
1057 wake_up_inode(inode);
1058 BUG_ON(inode->i_state != I_CLEAR);
1059 destroy_inode(inode);
1062 EXPORT_SYMBOL(generic_delete_inode);
1064 static void generic_forget_inode(struct inode *inode)
1066 struct super_block *sb = inode->i_sb;
1068 if (!hlist_unhashed(&inode->i_hash)) {
1069 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1070 list_move(&inode->i_list, &inode_unused);
1071 inodes_stat.nr_unused++;
1072 if (!sb || (sb->s_flags & MS_ACTIVE)) {
1073 spin_unlock(&inode_lock);
1076 inode->i_state |= I_WILL_FREE;
1077 spin_unlock(&inode_lock);
1078 write_inode_now(inode, 1);
1079 spin_lock(&inode_lock);
1080 inode->i_state &= ~I_WILL_FREE;
1081 inodes_stat.nr_unused--;
1082 hlist_del_init(&inode->i_hash);
1084 list_del_init(&inode->i_list);
1085 list_del_init(&inode->i_sb_list);
1086 inode->i_state |= I_FREEING;
1087 inodes_stat.nr_inodes--;
1088 spin_unlock(&inode_lock);
1089 if (inode->i_data.nrpages)
1090 truncate_inode_pages(&inode->i_data, 0);
1092 wake_up_inode(inode);
1093 destroy_inode(inode);
1097 * Normal UNIX filesystem behaviour: delete the
1098 * inode when the usage count drops to zero, and
1101 void generic_drop_inode(struct inode *inode)
1103 if (!inode->i_nlink)
1104 generic_delete_inode(inode);
1106 generic_forget_inode(inode);
1109 EXPORT_SYMBOL_GPL(generic_drop_inode);
1112 * Called when we're dropping the last reference
1115 * Call the FS "drop()" function, defaulting to
1116 * the legacy UNIX filesystem behaviour..
1118 * NOTE! NOTE! NOTE! We're called with the inode lock
1119 * held, and the drop function is supposed to release
1122 static inline void iput_final(struct inode *inode)
1124 struct super_operations *op = inode->i_sb->s_op;
1125 void (*drop)(struct inode *) = generic_drop_inode;
1127 if (op && op->drop_inode)
1128 drop = op->drop_inode;
1133 * iput - put an inode
1134 * @inode: inode to put
1136 * Puts an inode, dropping its usage count. If the inode use count hits
1137 * zero, the inode is then freed and may also be destroyed.
1139 * Consequently, iput() can sleep.
1141 void iput(struct inode *inode)
1144 struct super_operations *op = inode->i_sb->s_op;
1146 BUG_ON(inode->i_state == I_CLEAR);
1148 if (op && op->put_inode)
1149 op->put_inode(inode);
1151 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1156 EXPORT_SYMBOL(iput);
1159 * bmap - find a block number in a file
1160 * @inode: inode of file
1161 * @block: block to find
1163 * Returns the block number on the device holding the inode that
1164 * is the disk block number for the block of the file requested.
1165 * That is, asked for block 4 of inode 1 the function will return the
1166 * disk block relative to the disk start that holds that block of the
1169 sector_t bmap(struct inode * inode, sector_t block)
1172 if (inode->i_mapping->a_ops->bmap)
1173 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1177 EXPORT_SYMBOL(bmap);
1180 * touch_atime - update the access time
1181 * @mnt: mount the inode is accessed on
1182 * @dentry: dentry accessed
1184 * Update the accessed time on an inode and mark it for writeback.
1185 * This function automatically handles read only file systems and media,
1186 * as well as the "noatime" flag and inode specific "noatime" markers.
1188 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1190 struct inode *inode = dentry->d_inode;
1191 struct timespec now;
1193 if (IS_RDONLY(inode))
1196 if ((inode->i_flags & S_NOATIME) ||
1197 (inode->i_sb->s_flags & MS_NOATIME) ||
1198 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
1202 * We may have a NULL vfsmount when coming from NFSD
1205 ((mnt->mnt_flags & MNT_NOATIME) ||
1206 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))))
1209 now = current_fs_time(inode->i_sb);
1210 if (!timespec_equal(&inode->i_atime, &now)) {
1211 inode->i_atime = now;
1212 mark_inode_dirty_sync(inode);
1216 EXPORT_SYMBOL(touch_atime);
1219 * file_update_time - update mtime and ctime time
1220 * @file: file accessed
1222 * Update the mtime and ctime members of an inode and mark the inode
1223 * for writeback. Note that this function is meant exclusively for
1224 * usage in the file write path of filesystems, and filesystems may
1225 * choose to explicitly ignore update via this function with the
1226 * S_NOCTIME inode flag, e.g. for network filesystem where these
1227 * timestamps are handled by the server.
1230 void file_update_time(struct file *file)
1232 struct inode *inode = file->f_dentry->d_inode;
1233 struct timespec now;
1236 if (IS_NOCMTIME(inode))
1238 if (IS_RDONLY(inode))
1241 now = current_fs_time(inode->i_sb);
1242 if (!timespec_equal(&inode->i_mtime, &now))
1244 inode->i_mtime = now;
1246 if (!timespec_equal(&inode->i_ctime, &now))
1248 inode->i_ctime = now;
1251 mark_inode_dirty_sync(inode);
1254 EXPORT_SYMBOL(file_update_time);
1256 int inode_needs_sync(struct inode *inode)
1260 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1265 EXPORT_SYMBOL(inode_needs_sync);
1268 * Quota functions that want to walk the inode lists..
1272 /* Function back in dquot.c */
1273 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1275 void remove_dquot_ref(struct super_block *sb, int type,
1276 struct list_head *tofree_head)
1278 struct inode *inode;
1281 return; /* nothing to do */
1282 spin_lock(&inode_lock); /* This lock is for inodes code */
1285 * We don't have to lock against quota code - test IS_QUOTAINIT is
1286 * just for speedup...
1288 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1289 if (!IS_NOQUOTA(inode))
1290 remove_inode_dquot_ref(inode, type, tofree_head);
1292 spin_unlock(&inode_lock);
1297 int inode_wait(void *word)
1304 * If we try to find an inode in the inode hash while it is being
1305 * deleted, we have to wait until the filesystem completes its
1306 * deletion before reporting that it isn't found. This function waits
1307 * until the deletion _might_ have completed. Callers are responsible
1308 * to recheck inode state.
1310 * It doesn't matter if I_LOCK is not set initially, a call to
1311 * wake_up_inode() after removing from the hash list will DTRT.
1313 * This is called with inode_lock held.
1315 static void __wait_on_freeing_inode(struct inode *inode)
1317 wait_queue_head_t *wq;
1318 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1319 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1320 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1321 spin_unlock(&inode_lock);
1323 finish_wait(wq, &wait.wait);
1324 spin_lock(&inode_lock);
1327 void wake_up_inode(struct inode *inode)
1330 * Prevent speculative execution through spin_unlock(&inode_lock);
1333 wake_up_bit(&inode->i_state, __I_LOCK);
1336 static __initdata unsigned long ihash_entries;
1337 static int __init set_ihash_entries(char *str)
1341 ihash_entries = simple_strtoul(str, &str, 0);
1344 __setup("ihash_entries=", set_ihash_entries);
1347 * Initialize the waitqueues and inode hash table.
1349 void __init inode_init_early(void)
1353 /* If hashes are distributed across NUMA nodes, defer
1354 * hash allocation until vmalloc space is available.
1360 alloc_large_system_hash("Inode-cache",
1361 sizeof(struct hlist_head),
1369 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1370 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1373 void __init inode_init(unsigned long mempages)
1377 /* inode slab cache */
1378 inode_cachep = kmem_cache_create("inode_cache",
1379 sizeof(struct inode),
1381 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1385 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1387 /* Hash may have been set up in inode_init_early */
1392 alloc_large_system_hash("Inode-cache",
1393 sizeof(struct hlist_head),
1401 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1402 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1405 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1407 inode->i_mode = mode;
1408 if (S_ISCHR(mode)) {
1409 inode->i_fop = &def_chr_fops;
1410 inode->i_rdev = rdev;
1411 } else if (S_ISBLK(mode)) {
1412 inode->i_fop = &def_blk_fops;
1413 inode->i_rdev = rdev;
1414 } else if (S_ISFIFO(mode))
1415 inode->i_fop = &def_fifo_fops;
1416 else if (S_ISSOCK(mode))
1417 inode->i_fop = &bad_sock_fops;
1419 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1422 EXPORT_SYMBOL(init_special_inode);