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>
26 * This is needed for the following functions:
28 * - 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;
59 static unsigned int i_hash_shift;
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;
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 spinlock_t inode_lock = SPIN_LOCK_UNLOCKED;
84 EXPORT_SYMBOL(inode_lock);
87 * iprune_sem 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 DECLARE_MUTEX(iprune_sem);
97 * Statistics gathering..
99 struct inodes_stat_t inodes_stat;
100 EXPORT_SYMBOL(inodes_stat);
102 static kmem_cache_t * inode_cachep;
104 static struct inode *alloc_inode(struct super_block *sb)
106 static struct address_space_operations empty_aops;
107 static struct inode_operations empty_iops;
108 static struct file_operations empty_fops;
111 if (sb->s_op->alloc_inode)
112 inode = sb->s_op->alloc_inode(sb);
114 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
117 struct address_space * const mapping = &inode->i_data;
120 inode->i_blkbits = sb->s_blocksize_bits;
122 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->i_security = NULL;
140 inode->dirtied_when = 0;
141 if (security_inode_alloc(inode)) {
142 if (inode->i_sb->s_op->destroy_inode)
143 inode->i_sb->s_op->destroy_inode(inode);
145 kmem_cache_free(inode_cachep, (inode));
149 mapping->a_ops = &empty_aops;
150 mapping->host = inode;
152 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
153 mapping->assoc_mapping = NULL;
154 mapping->backing_dev_info = &default_backing_dev_info;
157 * If the block_device provides a backing_dev_info for client
158 * inodes then use that. Otherwise the inode share the bdev's
162 struct backing_dev_info *bdi;
164 bdi = sb->s_bdev->bd_inode_backing_dev_info;
166 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
167 mapping->backing_dev_info = bdi;
169 memset(&inode->u, 0, sizeof(inode->u));
170 inode->i_mapping = mapping;
175 void destroy_inode(struct inode *inode)
177 if (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));
185 EXPORT_SYMBOL(destroy_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 sema_init(&inode->i_sem, 1);
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 atomic_set(&inode->i_data.truncate_count, 0);
204 INIT_LIST_HEAD(&inode->i_data.private_list);
205 spin_lock_init(&inode->i_data.private_lock);
206 INIT_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
207 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
208 spin_lock_init(&inode->i_lock);
209 i_size_ordered_init(inode);
212 EXPORT_SYMBOL(inode_init_once);
214 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
216 struct inode * inode = (struct inode *) foo;
218 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
219 SLAB_CTOR_CONSTRUCTOR)
220 inode_init_once(inode);
224 * inode_lock must be held
226 void __iget(struct inode * inode)
228 if (atomic_read(&inode->i_count)) {
229 atomic_inc(&inode->i_count);
232 atomic_inc(&inode->i_count);
233 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
234 list_move(&inode->i_list, &inode_in_use);
235 inodes_stat.nr_unused--;
237 EXPORT_SYMBOL(__iget);
240 * clear_inode - clear an inode
241 * @inode: inode to clear
243 * This is called by the filesystem to tell us
244 * that the inode is no longer useful. We just
245 * terminate it with extreme prejudice.
247 void clear_inode(struct inode *inode)
249 invalidate_inode_buffers(inode);
251 if (inode->i_data.nrpages)
253 if (!(inode->i_state & I_FREEING))
255 if (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);
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);
290 destroy_inode(inode);
293 spin_lock(&inode_lock);
294 inodes_stat.nr_inodes -= nr_disposed;
295 spin_unlock(&inode_lock);
299 * Invalidate all inodes for a device.
301 static int invalidate_list(struct list_head *head, struct list_head *dispose)
303 struct list_head *next;
304 int busy = 0, count = 0;
308 struct list_head * tmp = next;
309 struct inode * inode;
314 inode = list_entry(tmp, struct inode, i_sb_list);
315 invalidate_inode_buffers(inode);
316 if (!atomic_read(&inode->i_count)) {
317 hlist_del_init(&inode->i_hash);
318 list_del(&inode->i_sb_list);
319 list_move(&inode->i_list, dispose);
320 inode->i_state |= I_FREEING;
326 /* only unused inodes may be cached with i_count zero */
327 inodes_stat.nr_unused -= count;
332 * This is a two-stage process. First we collect all
333 * offending inodes onto the throw-away list, and in
334 * the second stage we actually dispose of them. This
335 * is because we don't want to sleep while messing
336 * with the global lists..
340 * invalidate_inodes - discard the inodes on a device
343 * Discard all of the inodes for a given superblock. If the discard
344 * fails because there are busy inodes then a non zero value is returned.
345 * If the discard is successful all the inodes have been discarded.
347 int invalidate_inodes(struct super_block * sb)
350 LIST_HEAD(throw_away);
353 spin_lock(&inode_lock);
354 busy = invalidate_list(&sb->s_inodes, &throw_away);
355 spin_unlock(&inode_lock);
357 dispose_list(&throw_away);
363 EXPORT_SYMBOL(invalidate_inodes);
365 int __invalidate_device(struct block_device *bdev, int do_sync)
367 struct super_block *sb;
374 sb = get_super(bdev);
377 * no need to lock the super, get_super holds the
378 * read semaphore so the filesystem cannot go away
379 * under us (->put_super runs with the write lock
382 shrink_dcache_sb(sb);
383 res = invalidate_inodes(sb);
386 invalidate_bdev(bdev, 0);
390 EXPORT_SYMBOL(__invalidate_device);
392 static int can_unuse(struct inode *inode)
396 if (inode_has_buffers(inode))
398 if (atomic_read(&inode->i_count))
400 if (inode->i_data.nrpages)
406 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
407 * a temporary list and then are freed outside inode_lock by dispose_list().
409 * Any inodes which are pinned purely because of attached pagecache have their
410 * pagecache removed. We expect the final iput() on that inode to add it to
411 * the front of the inode_unused list. So look for it there and if the
412 * inode is still freeable, proceed. The right inode is found 99.9% of the
413 * time in testing on a 4-way.
415 * If the inode has metadata buffers attached to mapping->private_list then
416 * try to remove them.
418 static void prune_icache(int nr_to_scan)
423 unsigned long reap = 0;
426 spin_lock(&inode_lock);
427 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
430 if (list_empty(&inode_unused))
433 inode = list_entry(inode_unused.prev, struct inode, i_list);
435 if (inode->i_state || atomic_read(&inode->i_count)) {
436 list_move(&inode->i_list, &inode_unused);
439 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
441 spin_unlock(&inode_lock);
442 if (remove_inode_buffers(inode))
443 reap += invalidate_inode_pages(&inode->i_data);
445 spin_lock(&inode_lock);
447 if (inode != list_entry(inode_unused.next,
448 struct inode, i_list))
449 continue; /* wrong inode or list_empty */
450 if (!can_unuse(inode))
453 hlist_del_init(&inode->i_hash);
454 list_del_init(&inode->i_sb_list);
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);
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, unsigned int 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)) {
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)) {
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))
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
763 * ifind() searches for the inode specified by @data in the inode
764 * cache. This is a generalized version of ifind_fast() for file systems where
765 * the inode number is not sufficient for unique identification of an inode.
767 * If the inode is in the cache, the inode is returned with an incremented
770 * Otherwise NULL is returned.
772 * Note, @test is called with the inode_lock held, so can't sleep.
774 static inline struct inode *ifind(struct super_block *sb,
775 struct hlist_head *head, int (*test)(struct inode *, void *),
780 spin_lock(&inode_lock);
781 inode = find_inode(sb, head, test, data);
784 spin_unlock(&inode_lock);
785 wait_on_inode(inode);
788 spin_unlock(&inode_lock);
793 * ifind_fast - internal function, you want ilookup() or iget().
794 * @sb: super block of file system to search
795 * @head: head of the list to search
796 * @ino: inode number to search for
798 * ifind_fast() searches for the inode @ino in the inode cache. This is for
799 * file systems where the inode number is sufficient for unique identification
802 * If the inode is in the cache, the inode is returned with an incremented
805 * Otherwise NULL is returned.
807 static inline struct inode *ifind_fast(struct super_block *sb,
808 struct hlist_head *head, unsigned long ino)
812 spin_lock(&inode_lock);
813 inode = find_inode_fast(sb, head, ino);
816 spin_unlock(&inode_lock);
817 wait_on_inode(inode);
820 spin_unlock(&inode_lock);
825 * ilookup5 - search for an inode in the inode cache
826 * @sb: super block of file system to search
827 * @hashval: hash value (usually inode number) to search for
828 * @test: callback used for comparisons between inodes
829 * @data: opaque data pointer to pass to @test
831 * ilookup5() uses ifind() to search for the inode specified by @hashval and
832 * @data in the inode cache. This is a generalized version of ilookup() for
833 * file systems where the inode number is not sufficient for unique
834 * identification of an inode.
836 * If the inode is in the cache, the inode is returned with an incremented
839 * Otherwise NULL is returned.
841 * Note, @test is called with the inode_lock held, so can't sleep.
843 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
844 int (*test)(struct inode *, void *), void *data)
846 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
848 return ifind(sb, head, test, data);
851 EXPORT_SYMBOL(ilookup5);
854 * ilookup - search for an inode in the inode cache
855 * @sb: super block of file system to search
856 * @ino: inode number to search for
858 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
859 * This is for file systems where the inode number is sufficient for unique
860 * identification of an inode.
862 * If the inode is in the cache, the inode is returned with an incremented
865 * Otherwise NULL is returned.
867 struct inode *ilookup(struct super_block *sb, unsigned long ino)
869 struct hlist_head *head = inode_hashtable + hash(sb, ino);
871 return ifind_fast(sb, head, ino);
874 EXPORT_SYMBOL(ilookup);
877 * iget5_locked - obtain an inode from a mounted file system
878 * @sb: super block of file system
879 * @hashval: hash value (usually inode number) to get
880 * @test: callback used for comparisons between inodes
881 * @set: callback used to initialize a new struct inode
882 * @data: opaque data pointer to pass to @test and @set
884 * This is iget() without the read_inode() portion of get_new_inode().
886 * iget5_locked() uses ifind() to search for the inode specified by @hashval
887 * and @data in the inode cache and if present it is returned with an increased
888 * reference count. This is a generalized version of iget_locked() for file
889 * systems where the inode number is not sufficient for unique identification
892 * If the inode is not in cache, get_new_inode() is called to allocate a new
893 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
894 * file system gets to fill it in before unlocking it via unlock_new_inode().
896 * Note both @test and @set are called with the inode_lock held, so can't sleep.
898 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
899 int (*test)(struct inode *, void *),
900 int (*set)(struct inode *, void *), void *data)
902 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
905 inode = ifind(sb, head, test, data);
909 * get_new_inode() will do the right thing, re-trying the search
910 * in case it had to block at any point.
912 return get_new_inode(sb, head, test, set, data);
915 EXPORT_SYMBOL(iget5_locked);
918 * iget_locked - obtain an inode from a mounted file system
919 * @sb: super block of file system
920 * @ino: inode number to get
922 * This is iget() without the read_inode() portion of get_new_inode_fast().
924 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
925 * the inode cache and if present it is returned with an increased reference
926 * count. This is for file systems where the inode number is sufficient for
927 * unique identification of an inode.
929 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
930 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
931 * The file system gets to fill it in before unlocking it via
932 * unlock_new_inode().
934 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
936 struct hlist_head *head = inode_hashtable + hash(sb, ino);
939 inode = ifind_fast(sb, head, ino);
943 * get_new_inode_fast() will do the right thing, re-trying the search
944 * in case it had to block at any point.
946 return get_new_inode_fast(sb, head, ino);
949 EXPORT_SYMBOL(iget_locked);
952 * __insert_inode_hash - hash an inode
953 * @inode: unhashed inode
954 * @hashval: unsigned long value used to locate this object in the
957 * Add an inode to the inode hash for this superblock.
959 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
961 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
962 spin_lock(&inode_lock);
963 hlist_add_head(&inode->i_hash, head);
964 spin_unlock(&inode_lock);
967 EXPORT_SYMBOL(__insert_inode_hash);
970 * remove_inode_hash - remove an inode from the hash
971 * @inode: inode to unhash
973 * Remove an inode from the superblock.
975 void remove_inode_hash(struct inode *inode)
977 spin_lock(&inode_lock);
978 hlist_del_init(&inode->i_hash);
979 spin_unlock(&inode_lock);
982 EXPORT_SYMBOL(remove_inode_hash);
985 * Tell the filesystem that this inode is no longer of any interest and should
986 * be completely destroyed.
988 * We leave the inode in the inode hash table until *after* the filesystem's
989 * ->delete_inode completes. This ensures that an iget (such as nfsd might
990 * instigate) will always find up-to-date information either in the hash or on
993 * I_FREEING is set so that no-one will take a new reference to the inode while
994 * it is being deleted.
996 void generic_delete_inode(struct inode *inode)
998 struct super_operations *op = inode->i_sb->s_op;
1000 list_del_init(&inode->i_list);
1001 list_del_init(&inode->i_sb_list);
1002 inode->i_state|=I_FREEING;
1003 inodes_stat.nr_inodes--;
1004 spin_unlock(&inode_lock);
1006 if (inode->i_data.nrpages)
1007 truncate_inode_pages(&inode->i_data, 0);
1009 security_inode_delete(inode);
1011 if (op->delete_inode) {
1012 void (*delete)(struct inode *) = op->delete_inode;
1013 if (!is_bad_inode(inode))
1015 /* s_op->delete_inode internally recalls clear_inode() */
1019 spin_lock(&inode_lock);
1020 hlist_del_init(&inode->i_hash);
1021 spin_unlock(&inode_lock);
1022 wake_up_inode(inode);
1023 if (inode->i_state != I_CLEAR)
1025 destroy_inode(inode);
1028 EXPORT_SYMBOL(generic_delete_inode);
1030 void generic_forget_inode(struct inode *inode)
1032 struct super_block *sb = inode->i_sb;
1034 if (!hlist_unhashed(&inode->i_hash)) {
1035 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1036 list_move(&inode->i_list, &inode_unused);
1037 inodes_stat.nr_unused++;
1038 spin_unlock(&inode_lock);
1039 if (!sb || (sb->s_flags & MS_ACTIVE))
1041 write_inode_now(inode, 1);
1042 spin_lock(&inode_lock);
1043 inodes_stat.nr_unused--;
1044 hlist_del_init(&inode->i_hash);
1046 list_del_init(&inode->i_list);
1047 list_del_init(&inode->i_sb_list);
1048 inode->i_state|=I_FREEING;
1049 inodes_stat.nr_inodes--;
1050 spin_unlock(&inode_lock);
1051 if (inode->i_data.nrpages)
1052 truncate_inode_pages(&inode->i_data, 0);
1054 destroy_inode(inode);
1056 EXPORT_SYMBOL(generic_forget_inode);
1059 * Normal UNIX filesystem behaviour: delete the
1060 * inode when the usage count drops to zero, and
1063 static void generic_drop_inode(struct inode *inode)
1065 if (!inode->i_nlink)
1066 generic_delete_inode(inode);
1068 generic_forget_inode(inode);
1072 * Called when we're dropping the last reference
1075 * Call the FS "drop()" function, defaulting to
1076 * the legacy UNIX filesystem behaviour..
1078 * NOTE! NOTE! NOTE! We're called with the inode lock
1079 * held, and the drop function is supposed to release
1082 static inline void iput_final(struct inode *inode)
1084 struct super_operations *op = inode->i_sb->s_op;
1085 void (*drop)(struct inode *) = generic_drop_inode;
1087 if (op && op->drop_inode)
1088 drop = op->drop_inode;
1093 * iput - put an inode
1094 * @inode: inode to put
1096 * Puts an inode, dropping its usage count. If the inode use count hits
1097 * zero the inode is also then freed and may be destroyed.
1099 void iput(struct inode *inode)
1102 struct super_operations *op = inode->i_sb->s_op;
1104 if (inode->i_state == I_CLEAR)
1107 if (op && op->put_inode)
1108 op->put_inode(inode);
1110 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1115 EXPORT_SYMBOL(iput);
1118 * bmap - find a block number in a file
1119 * @inode: inode of file
1120 * @block: block to find
1122 * Returns the block number on the device holding the inode that
1123 * is the disk block number for the block of the file requested.
1124 * That is, asked for block 4 of inode 1 the function will return the
1125 * disk block relative to the disk start that holds that block of the
1128 sector_t bmap(struct inode * inode, sector_t block)
1131 if (inode->i_mapping->a_ops->bmap)
1132 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1136 EXPORT_SYMBOL(bmap);
1139 * Return true if the filesystem which backs this inode considers the two
1140 * passed timespecs to be sufficiently different to warrant flushing the
1141 * altered time out to disk.
1143 static int inode_times_differ(struct inode *inode,
1144 struct timespec *old, struct timespec *new)
1146 if (IS_ONE_SECOND(inode))
1147 return old->tv_sec != new->tv_sec;
1148 return !timespec_equal(old, new);
1152 * update_atime - update the access time
1153 * @inode: inode accessed
1155 * Update the accessed time on an inode and mark it for writeback.
1156 * This function automatically handles read only file systems and media,
1157 * as well as the "noatime" flag and inode specific "noatime" markers.
1159 void update_atime(struct inode *inode)
1161 struct timespec now;
1163 if (IS_NOATIME(inode))
1165 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1167 if (IS_RDONLY(inode))
1170 now = current_kernel_time();
1171 if (inode_times_differ(inode, &inode->i_atime, &now)) {
1172 inode->i_atime = now;
1173 mark_inode_dirty_sync(inode);
1175 if (!timespec_equal(&inode->i_atime, &now))
1176 inode->i_atime = now;
1180 EXPORT_SYMBOL(update_atime);
1183 * inode_update_time - update mtime and ctime time
1184 * @inode: inode accessed
1185 * @ctime_too: update ctime too
1187 * Update the mtime time on an inode and mark it for writeback.
1188 * When ctime_too is specified update the ctime too.
1191 void inode_update_time(struct inode *inode, int ctime_too)
1193 struct timespec now;
1196 if (IS_NOCMTIME(inode))
1198 if (IS_RDONLY(inode))
1201 now = current_kernel_time();
1203 if (inode_times_differ(inode, &inode->i_mtime, &now))
1205 inode->i_mtime = now;
1208 if (inode_times_differ(inode, &inode->i_ctime, &now))
1210 inode->i_ctime = now;
1213 mark_inode_dirty_sync(inode);
1216 EXPORT_SYMBOL(inode_update_time);
1218 int inode_needs_sync(struct inode *inode)
1222 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1227 EXPORT_SYMBOL(inode_needs_sync);
1230 * Quota functions that want to walk the inode lists..
1234 /* Function back in dquot.c */
1235 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1237 void remove_dquot_ref(struct super_block *sb, int type,
1238 struct list_head *tofree_head)
1240 struct inode *inode;
1243 return; /* nothing to do */
1244 spin_lock(&inode_lock); /* This lock is for inodes code */
1247 * We don't have to lock against quota code - test IS_QUOTAINIT is
1248 * just for speedup...
1250 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1251 if (!IS_NOQUOTA(inode))
1252 remove_inode_dquot_ref(inode, type, tofree_head);
1254 spin_unlock(&inode_lock);
1260 * Hashed waitqueues for wait_on_inode(). The table is pretty small - the
1261 * kernel doesn't lock many inodes at the same time.
1263 #define I_WAIT_TABLE_ORDER 3
1264 static struct i_wait_queue_head {
1265 wait_queue_head_t wqh;
1266 } ____cacheline_aligned_in_smp i_wait_queue_heads[1<<I_WAIT_TABLE_ORDER];
1269 * Return the address of the waitqueue_head to be used for this inode
1271 static wait_queue_head_t *i_waitq_head(struct inode *inode)
1273 return &i_wait_queue_heads[hash_ptr(inode, I_WAIT_TABLE_ORDER)].wqh;
1276 void __wait_on_inode(struct inode *inode)
1278 DECLARE_WAITQUEUE(wait, current);
1279 wait_queue_head_t *wq = i_waitq_head(inode);
1281 add_wait_queue(wq, &wait);
1283 set_current_state(TASK_UNINTERRUPTIBLE);
1284 if (inode->i_state & I_LOCK) {
1288 remove_wait_queue(wq, &wait);
1289 __set_current_state(TASK_RUNNING);
1293 * If we try to find an inode in the inode hash while it is being deleted, we
1294 * have to wait until the filesystem completes its deletion before reporting
1295 * that it isn't found. This is because iget will immediately call
1296 * ->read_inode, and we want to be sure that evidence of the deletion is found
1299 * This call might return early if an inode which shares the waitq is woken up.
1300 * This is most easily handled by the caller which will loop around again
1301 * looking for the inode.
1303 * This is called with inode_lock held.
1305 static void __wait_on_freeing_inode(struct inode *inode)
1307 DECLARE_WAITQUEUE(wait, current);
1308 wait_queue_head_t *wq = i_waitq_head(inode);
1310 add_wait_queue(wq, &wait);
1311 set_current_state(TASK_UNINTERRUPTIBLE);
1312 spin_unlock(&inode_lock);
1314 remove_wait_queue(wq, &wait);
1315 spin_lock(&inode_lock);
1318 void wake_up_inode(struct inode *inode)
1320 wait_queue_head_t *wq = i_waitq_head(inode);
1323 * Prevent speculative execution through spin_unlock(&inode_lock);
1326 if (waitqueue_active(wq))
1329 EXPORT_SYMBOL(wake_up_inode);
1331 static __initdata unsigned long ihash_entries;
1332 static int __init set_ihash_entries(char *str)
1336 ihash_entries = simple_strtoul(str, &str, 0);
1339 __setup("ihash_entries=", set_ihash_entries);
1342 * Initialize the waitqueues and inode hash table.
1344 void __init inode_init_early(void)
1349 alloc_large_system_hash("Inode-cache",
1350 sizeof(struct hlist_head),
1357 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1358 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1361 void __init inode_init(unsigned long mempages)
1365 for (i = 0; i < ARRAY_SIZE(i_wait_queue_heads); i++)
1366 init_waitqueue_head(&i_wait_queue_heads[i].wqh);
1368 /* inode slab cache */
1369 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1370 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, init_once,
1372 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1375 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1377 inode->i_mode = mode;
1378 if (S_ISCHR(mode)) {
1379 inode->i_fop = &def_chr_fops;
1380 inode->i_rdev = rdev;
1381 } else if (S_ISBLK(mode)) {
1382 inode->i_fop = &def_blk_fops;
1383 inode->i_rdev = rdev;
1384 } else if (S_ISFIFO(mode))
1385 inode->i_fop = &def_fifo_fops;
1386 else if (S_ISSOCK(mode))
1387 inode->i_fop = &bad_sock_fops;
1389 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1392 EXPORT_SYMBOL(init_special_inode);