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;
86 * iprune_sem 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 DECLARE_MUTEX(iprune_sem);
96 * Statistics gathering..
98 struct inodes_stat_t inodes_stat;
100 static kmem_cache_t * inode_cachep;
102 static void prune_icache(int nr_to_scan);
105 #define INODE_UNUSED_THRESHOLD 15000
106 #define PRUNE_BATCH_COUNT 32
108 void try_to_clip_inodes(void)
110 unsigned long count = 0;
111 /* if there are a LOT of unused inodes in cache, better shrink a few first */
113 /* check lockless first to not take the lock always here; racing occasionally isn't a big deal */
114 if (inodes_stat.nr_unused > INODE_UNUSED_THRESHOLD) {
115 spin_lock(&inode_lock);
116 if (inodes_stat.nr_unused > INODE_UNUSED_THRESHOLD)
117 count = inodes_stat.nr_unused - INODE_UNUSED_THRESHOLD;
118 spin_unlock(&inode_lock);
125 static struct inode *alloc_inode(struct super_block *sb)
127 static struct address_space_operations empty_aops;
128 static struct inode_operations empty_iops;
129 static struct file_operations empty_fops;
132 if (sb->s_op->alloc_inode)
133 inode = sb->s_op->alloc_inode(sb);
135 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
138 struct address_space * const mapping = &inode->i_data;
141 inode->i_blkbits = sb->s_blocksize_bits;
143 atomic_set(&inode->i_count, 1);
145 inode->i_op = &empty_iops;
146 inode->i_fop = &empty_fops;
148 atomic_set(&inode->i_writecount, 0);
152 inode->i_generation = 0;
154 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
156 inode->i_pipe = NULL;
157 inode->i_bdev = NULL;
158 inode->i_cdev = NULL;
160 inode->i_security = NULL;
161 inode->dirtied_when = 0;
162 if (security_inode_alloc(inode)) {
163 if (inode->i_sb->s_op->destroy_inode)
164 inode->i_sb->s_op->destroy_inode(inode);
166 kmem_cache_free(inode_cachep, (inode));
170 mapping->a_ops = &empty_aops;
171 mapping->host = inode;
173 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
174 mapping->assoc_mapping = NULL;
175 mapping->backing_dev_info = &default_backing_dev_info;
178 * If the block_device provides a backing_dev_info for client
179 * inodes then use that. Otherwise the inode share the bdev's
183 struct backing_dev_info *bdi;
185 bdi = sb->s_bdev->bd_inode_backing_dev_info;
187 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
188 mapping->backing_dev_info = bdi;
190 memset(&inode->u, 0, sizeof(inode->u));
191 inode->i_mapping = mapping;
196 void destroy_inode(struct inode *inode)
198 if (inode_has_buffers(inode))
200 security_inode_free(inode);
201 if (inode->i_sb->s_op->destroy_inode)
202 inode->i_sb->s_op->destroy_inode(inode);
204 kmem_cache_free(inode_cachep, (inode));
209 * These are initializations that only need to be done
210 * once, because the fields are idempotent across use
211 * of the inode, so let the slab aware of that.
213 void inode_init_once(struct inode *inode)
215 memset(inode, 0, sizeof(*inode));
216 INIT_HLIST_NODE(&inode->i_hash);
217 INIT_LIST_HEAD(&inode->i_dentry);
218 INIT_LIST_HEAD(&inode->i_devices);
219 sema_init(&inode->i_sem, 1);
220 init_rwsem(&inode->i_alloc_sem);
221 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
222 spin_lock_init(&inode->i_data.tree_lock);
223 spin_lock_init(&inode->i_data.i_mmap_lock);
224 atomic_set(&inode->i_data.truncate_count, 0);
225 INIT_LIST_HEAD(&inode->i_data.private_list);
226 spin_lock_init(&inode->i_data.private_lock);
227 INIT_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
228 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
229 spin_lock_init(&inode->i_lock);
230 i_size_ordered_init(inode);
233 EXPORT_SYMBOL(inode_init_once);
235 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
237 struct inode * inode = (struct inode *) foo;
239 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
240 SLAB_CTOR_CONSTRUCTOR)
241 inode_init_once(inode);
245 * inode_lock must be held
247 void __iget(struct inode * inode)
249 if (atomic_read(&inode->i_count)) {
250 atomic_inc(&inode->i_count);
253 atomic_inc(&inode->i_count);
254 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
255 list_move(&inode->i_list, &inode_in_use);
256 inodes_stat.nr_unused--;
260 * clear_inode - clear an inode
261 * @inode: inode to clear
263 * This is called by the filesystem to tell us
264 * that the inode is no longer useful. We just
265 * terminate it with extreme prejudice.
267 void clear_inode(struct inode *inode)
270 invalidate_inode_buffers(inode);
272 if (inode->i_data.nrpages)
274 if (!(inode->i_state & I_FREEING))
276 if (inode->i_state & I_CLEAR)
278 wait_on_inode(inode);
280 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
281 inode->i_sb->s_op->clear_inode(inode);
286 inode->i_state = I_CLEAR;
289 EXPORT_SYMBOL(clear_inode);
292 * dispose_list - dispose of the contents of a local list
293 * @head: the head of the list to free
295 * Dispose-list gets a local list with local inodes in it, so it doesn't
296 * need to worry about list corruption and SMP locks.
298 static void dispose_list(struct list_head *head)
302 while (!list_empty(head)) {
305 inode = list_entry(head->next, struct inode, i_list);
306 list_del(&inode->i_list);
308 if (inode->i_data.nrpages)
309 truncate_inode_pages(&inode->i_data, 0);
311 destroy_inode(inode);
314 spin_lock(&inode_lock);
315 inodes_stat.nr_inodes -= nr_disposed;
316 spin_unlock(&inode_lock);
320 * Invalidate all inodes for a device.
322 static int invalidate_list(struct list_head *head, struct super_block * sb, struct list_head * dispose)
324 struct list_head *next;
325 int busy = 0, count = 0;
329 struct list_head * tmp = next;
330 struct inode * inode;
335 inode = list_entry(tmp, struct inode, i_list);
336 if (inode->i_sb != sb)
338 invalidate_inode_buffers(inode);
339 if (!atomic_read(&inode->i_count)) {
340 hlist_del_init(&inode->i_hash);
341 list_move(&inode->i_list, dispose);
342 inode->i_state |= I_FREEING;
348 /* only unused inodes may be cached with i_count zero */
349 inodes_stat.nr_unused -= count;
354 * This is a two-stage process. First we collect all
355 * offending inodes onto the throw-away list, and in
356 * the second stage we actually dispose of them. This
357 * is because we don't want to sleep while messing
358 * with the global lists..
362 * invalidate_inodes - discard the inodes on a device
365 * Discard all of the inodes for a given superblock. If the discard
366 * fails because there are busy inodes then a non zero value is returned.
367 * If the discard is successful all the inodes have been discarded.
369 int invalidate_inodes(struct super_block * sb)
372 LIST_HEAD(throw_away);
375 spin_lock(&inode_lock);
376 busy = invalidate_list(&inode_in_use, sb, &throw_away);
377 busy |= invalidate_list(&inode_unused, sb, &throw_away);
378 busy |= invalidate_list(&sb->s_dirty, sb, &throw_away);
379 busy |= invalidate_list(&sb->s_io, sb, &throw_away);
380 spin_unlock(&inode_lock);
382 dispose_list(&throw_away);
388 EXPORT_SYMBOL(invalidate_inodes);
390 int __invalidate_device(struct block_device *bdev, int do_sync)
392 struct super_block *sb;
399 sb = get_super(bdev);
402 * no need to lock the super, get_super holds the
403 * read semaphore so the filesystem cannot go away
404 * under us (->put_super runs with the write lock
407 shrink_dcache_sb(sb);
408 res = invalidate_inodes(sb);
411 invalidate_bdev(bdev, 0);
415 EXPORT_SYMBOL(__invalidate_device);
417 static int can_unuse(struct inode *inode)
421 if (inode_has_buffers(inode))
423 if (atomic_read(&inode->i_count))
425 if (inode->i_data.nrpages)
431 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
432 * a temporary list and then are freed outside inode_lock by dispose_list().
434 * Any inodes which are pinned purely because of attached pagecache have their
435 * pagecache removed. We expect the final iput() on that inode to add it to
436 * the front of the inode_unused list. So look for it there and if the
437 * inode is still freeable, proceed. The right inode is found 99.9% of the
438 * time in testing on a 4-way.
440 * If the inode has metadata buffers attached to mapping->private_list then
441 * try to remove them.
443 static void prune_icache(int nr_to_scan)
448 unsigned long reap = 0;
451 spin_lock(&inode_lock);
452 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
455 if (list_empty(&inode_unused))
458 inode = list_entry(inode_unused.prev, struct inode, i_list);
460 if (inode->i_state || atomic_read(&inode->i_count)) {
461 list_move(&inode->i_list, &inode_unused);
464 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
466 spin_unlock(&inode_lock);
467 if (remove_inode_buffers(inode))
468 reap += invalidate_inode_pages(&inode->i_data);
470 spin_lock(&inode_lock);
472 if (inode != list_entry(inode_unused.next,
473 struct inode, i_list))
474 continue; /* wrong inode or list_empty */
475 if (!can_unuse(inode))
478 hlist_del_init(&inode->i_hash);
479 list_move(&inode->i_list, &freeable);
480 inode->i_state |= I_FREEING;
483 inodes_stat.nr_unused -= nr_pruned;
484 spin_unlock(&inode_lock);
486 dispose_list(&freeable);
489 if (current_is_kswapd())
490 mod_page_state(kswapd_inodesteal, reap);
492 mod_page_state(pginodesteal, reap);
496 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
497 * "unused" means that no dentries are referring to the inodes: the files are
498 * not open and the dcache references to those inodes have already been
501 * This function is passed the number of inodes to scan, and it returns the
502 * total number of remaining possibly-reclaimable inodes.
504 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
508 * Nasty deadlock avoidance. We may hold various FS locks,
509 * and we don't want to recurse into the FS that called us
510 * in clear_inode() and friends..
512 if (gfp_mask & __GFP_FS)
515 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
518 static void __wait_on_freeing_inode(struct inode *inode);
520 * Called with the inode lock held.
521 * NOTE: we are not increasing the inode-refcount, you must call __iget()
522 * by hand after calling find_inode now! This simplifies iunique and won't
523 * add any additional branch in the common code.
525 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
527 struct hlist_node *node;
528 struct inode * inode = NULL;
531 hlist_for_each (node, head) {
532 inode = hlist_entry(node, struct inode, i_hash);
533 if (inode->i_sb != sb)
535 if (!test(inode, data))
537 if (inode->i_state & (I_FREEING|I_CLEAR)) {
538 __wait_on_freeing_inode(inode);
543 return node ? inode : NULL;
547 * find_inode_fast is the fast path version of find_inode, see the comment at
548 * iget_locked for details.
550 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
552 struct hlist_node *node;
553 struct inode * inode = NULL;
556 hlist_for_each (node, head) {
557 inode = hlist_entry(node, struct inode, i_hash);
558 if (inode->i_ino != ino)
560 if (inode->i_sb != sb)
562 if (inode->i_state & (I_FREEING|I_CLEAR)) {
563 __wait_on_freeing_inode(inode);
568 return node ? inode : NULL;
572 * new_inode - obtain an inode
575 * Allocates a new inode for given superblock.
577 struct inode *new_inode(struct super_block *sb)
579 static unsigned long last_ino;
580 struct inode * inode;
582 spin_lock_prefetch(&inode_lock);
584 inode = alloc_inode(sb);
586 spin_lock(&inode_lock);
587 inodes_stat.nr_inodes++;
588 list_add(&inode->i_list, &inode_in_use);
589 inode->i_ino = ++last_ino;
591 spin_unlock(&inode_lock);
596 EXPORT_SYMBOL(new_inode);
598 void unlock_new_inode(struct inode *inode)
601 * This is special! We do not need the spinlock
602 * when clearing I_LOCK, because we're guaranteed
603 * that nobody else tries to do anything about the
604 * state of the inode when it is locked, as we
605 * just created it (so there can be no old holders
606 * that haven't tested I_LOCK).
608 inode->i_state &= ~(I_LOCK|I_NEW);
609 wake_up_inode(inode);
612 EXPORT_SYMBOL(unlock_new_inode);
615 * This is called without the inode lock held.. Be careful.
617 * We no longer cache the sb_flags in i_flags - see fs.h
618 * -- rmk@arm.uk.linux.org
620 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)
622 struct inode * inode;
624 inode = alloc_inode(sb);
628 spin_lock(&inode_lock);
629 /* We released the lock, so.. */
630 old = find_inode(sb, head, test, data);
632 if (set(inode, data))
635 inodes_stat.nr_inodes++;
636 list_add(&inode->i_list, &inode_in_use);
637 hlist_add_head(&inode->i_hash, head);
638 inode->i_state = I_LOCK|I_NEW;
639 spin_unlock(&inode_lock);
641 /* Return the locked inode with I_NEW set, the
642 * caller is responsible for filling in the contents
648 * Uhhuh, somebody else created the same inode under
649 * us. Use the old inode instead of the one we just
653 spin_unlock(&inode_lock);
654 destroy_inode(inode);
656 wait_on_inode(inode);
661 spin_unlock(&inode_lock);
662 destroy_inode(inode);
667 * get_new_inode_fast is the fast path version of get_new_inode, see the
668 * comment at iget_locked for details.
670 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
672 struct inode * inode;
674 inode = alloc_inode(sb);
678 spin_lock(&inode_lock);
679 /* We released the lock, so.. */
680 old = find_inode_fast(sb, head, ino);
683 inodes_stat.nr_inodes++;
684 list_add(&inode->i_list, &inode_in_use);
685 hlist_add_head(&inode->i_hash, head);
686 inode->i_state = I_LOCK|I_NEW;
687 spin_unlock(&inode_lock);
689 /* Return the locked inode with I_NEW set, the
690 * caller is responsible for filling in the contents
696 * Uhhuh, somebody else created the same inode under
697 * us. Use the old inode instead of the one we just
701 spin_unlock(&inode_lock);
702 destroy_inode(inode);
704 wait_on_inode(inode);
709 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
713 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
715 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
716 return tmp & I_HASHMASK;
720 * iunique - get a unique inode number
722 * @max_reserved: highest reserved inode number
724 * Obtain an inode number that is unique on the system for a given
725 * superblock. This is used by file systems that have no natural
726 * permanent inode numbering system. An inode number is returned that
727 * is higher than the reserved limit but unique.
730 * With a large number of inodes live on the file system this function
731 * currently becomes quite slow.
733 ino_t iunique(struct super_block *sb, ino_t max_reserved)
735 static ino_t counter;
737 struct hlist_head * head;
739 spin_lock(&inode_lock);
741 if (counter > max_reserved) {
742 head = inode_hashtable + hash(sb,counter);
744 inode = find_inode_fast(sb, head, res);
746 spin_unlock(&inode_lock);
750 counter = max_reserved + 1;
756 EXPORT_SYMBOL(iunique);
758 struct inode *igrab(struct inode *inode)
760 spin_lock(&inode_lock);
761 if (!(inode->i_state & I_FREEING))
765 * Handle the case where s_op->clear_inode is not been
766 * called yet, and somebody is calling igrab
767 * while the inode is getting freed.
770 spin_unlock(&inode_lock);
774 EXPORT_SYMBOL(igrab);
777 * ifind - internal function, you want ilookup5() or iget5().
778 * @sb: super block of file system to search
779 * @head: the head of the list to search
780 * @test: callback used for comparisons between inodes
781 * @data: opaque data pointer to pass to @test
783 * ifind() searches for the inode specified by @data in the inode
784 * cache. This is a generalized version of ifind_fast() for file systems where
785 * the inode number is not sufficient for unique identification of an inode.
787 * If the inode is in the cache, the inode is returned with an incremented
790 * Otherwise NULL is returned.
792 * Note, @test is called with the inode_lock held, so can't sleep.
794 static inline struct inode *ifind(struct super_block *sb,
795 struct hlist_head *head, int (*test)(struct inode *, void *),
800 spin_lock(&inode_lock);
801 inode = find_inode(sb, head, test, data);
804 spin_unlock(&inode_lock);
805 wait_on_inode(inode);
808 spin_unlock(&inode_lock);
813 * ifind_fast - internal function, you want ilookup() or iget().
814 * @sb: super block of file system to search
815 * @head: head of the list to search
816 * @ino: inode number to search for
818 * ifind_fast() searches for the inode @ino in the inode cache. This is for
819 * file systems where the inode number is sufficient for unique identification
822 * If the inode is in the cache, the inode is returned with an incremented
825 * Otherwise NULL is returned.
827 static inline struct inode *ifind_fast(struct super_block *sb,
828 struct hlist_head *head, unsigned long ino)
832 spin_lock(&inode_lock);
833 inode = find_inode_fast(sb, head, ino);
836 spin_unlock(&inode_lock);
837 wait_on_inode(inode);
840 spin_unlock(&inode_lock);
845 * ilookup5 - search for an inode in the inode cache
846 * @sb: super block of file system to search
847 * @hashval: hash value (usually inode number) to search for
848 * @test: callback used for comparisons between inodes
849 * @data: opaque data pointer to pass to @test
851 * ilookup5() uses ifind() to search for the inode specified by @hashval and
852 * @data in the inode cache. This is a generalized version of ilookup() for
853 * file systems where the inode number is not sufficient for unique
854 * identification of an inode.
856 * If the inode is in the cache, the inode is returned with an incremented
859 * Otherwise NULL is returned.
861 * Note, @test is called with the inode_lock held, so can't sleep.
863 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
864 int (*test)(struct inode *, void *), void *data)
866 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
868 return ifind(sb, head, test, data);
871 EXPORT_SYMBOL(ilookup5);
874 * ilookup - search for an inode in the inode cache
875 * @sb: super block of file system to search
876 * @ino: inode number to search for
878 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
879 * This is for file systems where the inode number is sufficient for unique
880 * identification of an inode.
882 * If the inode is in the cache, the inode is returned with an incremented
885 * Otherwise NULL is returned.
887 struct inode *ilookup(struct super_block *sb, unsigned long ino)
889 struct hlist_head *head = inode_hashtable + hash(sb, ino);
891 return ifind_fast(sb, head, ino);
894 EXPORT_SYMBOL(ilookup);
897 * iget5_locked - obtain an inode from a mounted file system
898 * @sb: super block of file system
899 * @hashval: hash value (usually inode number) to get
900 * @test: callback used for comparisons between inodes
901 * @set: callback used to initialize a new struct inode
902 * @data: opaque data pointer to pass to @test and @set
904 * This is iget() without the read_inode() portion of get_new_inode().
906 * iget5_locked() uses ifind() to search for the inode specified by @hashval
907 * and @data in the inode cache and if present it is returned with an increased
908 * reference count. This is a generalized version of iget_locked() for file
909 * systems where the inode number is not sufficient for unique identification
912 * If the inode is not in cache, get_new_inode() is called to allocate a new
913 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
914 * file system gets to fill it in before unlocking it via unlock_new_inode().
916 * Note both @test and @set are called with the inode_lock held, so can't sleep.
918 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
919 int (*test)(struct inode *, void *),
920 int (*set)(struct inode *, void *), void *data)
922 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
925 inode = ifind(sb, head, test, data);
929 * get_new_inode() will do the right thing, re-trying the search
930 * in case it had to block at any point.
932 return get_new_inode(sb, head, test, set, data);
935 EXPORT_SYMBOL(iget5_locked);
938 * iget_locked - obtain an inode from a mounted file system
939 * @sb: super block of file system
940 * @ino: inode number to get
942 * This is iget() without the read_inode() portion of get_new_inode_fast().
944 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
945 * the inode cache and if present it is returned with an increased reference
946 * count. This is for file systems where the inode number is sufficient for
947 * unique identification of an inode.
949 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
950 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
951 * The file system gets to fill it in before unlocking it via
952 * unlock_new_inode().
954 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
956 struct hlist_head *head = inode_hashtable + hash(sb, ino);
959 inode = ifind_fast(sb, head, ino);
963 * get_new_inode_fast() will do the right thing, re-trying the search
964 * in case it had to block at any point.
966 return get_new_inode_fast(sb, head, ino);
969 EXPORT_SYMBOL(iget_locked);
972 * __insert_inode_hash - hash an inode
973 * @inode: unhashed inode
974 * @hashval: unsigned long value used to locate this object in the
977 * Add an inode to the inode hash for this superblock.
979 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
981 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
982 spin_lock(&inode_lock);
983 hlist_add_head(&inode->i_hash, head);
984 spin_unlock(&inode_lock);
987 EXPORT_SYMBOL(__insert_inode_hash);
990 * remove_inode_hash - remove an inode from the hash
991 * @inode: inode to unhash
993 * Remove an inode from the superblock.
995 void remove_inode_hash(struct inode *inode)
997 spin_lock(&inode_lock);
998 hlist_del_init(&inode->i_hash);
999 spin_unlock(&inode_lock);
1002 EXPORT_SYMBOL(remove_inode_hash);
1005 * Tell the filesystem that this inode is no longer of any interest and should
1006 * be completely destroyed.
1008 * We leave the inode in the inode hash table until *after* the filesystem's
1009 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1010 * instigate) will always find up-to-date information either in the hash or on
1013 * I_FREEING is set so that no-one will take a new reference to the inode while
1014 * it is being deleted.
1016 void generic_delete_inode(struct inode *inode)
1018 struct super_operations *op = inode->i_sb->s_op;
1020 list_del_init(&inode->i_list);
1021 inode->i_state|=I_FREEING;
1022 inodes_stat.nr_inodes--;
1023 spin_unlock(&inode_lock);
1025 if (inode->i_data.nrpages)
1026 truncate_inode_pages(&inode->i_data, 0);
1028 security_inode_delete(inode);
1030 if (op->delete_inode) {
1031 void (*delete)(struct inode *) = op->delete_inode;
1032 if (!is_bad_inode(inode))
1034 /* s_op->delete_inode internally recalls clear_inode() */
1038 spin_lock(&inode_lock);
1039 hlist_del_init(&inode->i_hash);
1040 spin_unlock(&inode_lock);
1041 wake_up_inode(inode);
1042 if (inode->i_state != I_CLEAR)
1044 destroy_inode(inode);
1047 EXPORT_SYMBOL(generic_delete_inode);
1049 static void generic_forget_inode(struct inode *inode)
1051 struct super_block *sb = inode->i_sb;
1053 if (!hlist_unhashed(&inode->i_hash)) {
1054 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1055 list_move(&inode->i_list, &inode_unused);
1056 inodes_stat.nr_unused++;
1057 spin_unlock(&inode_lock);
1058 if (!sb || (sb->s_flags & MS_ACTIVE))
1060 write_inode_now(inode, 1);
1061 spin_lock(&inode_lock);
1062 inodes_stat.nr_unused--;
1063 hlist_del_init(&inode->i_hash);
1065 list_del_init(&inode->i_list);
1066 inode->i_state|=I_FREEING;
1067 inodes_stat.nr_inodes--;
1068 spin_unlock(&inode_lock);
1069 if (inode->i_data.nrpages)
1070 truncate_inode_pages(&inode->i_data, 0);
1072 destroy_inode(inode);
1076 * Normal UNIX filesystem behaviour: delete the
1077 * inode when the usage count drops to zero, and
1080 static void generic_drop_inode(struct inode *inode)
1082 if (!inode->i_nlink)
1083 generic_delete_inode(inode);
1085 generic_forget_inode(inode);
1089 * Called when we're dropping the last reference
1092 * Call the FS "drop()" function, defaulting to
1093 * the legacy UNIX filesystem behaviour..
1095 * NOTE! NOTE! NOTE! We're called with the inode lock
1096 * held, and the drop function is supposed to release
1099 static inline void iput_final(struct inode *inode)
1101 struct super_operations *op = inode->i_sb->s_op;
1102 void (*drop)(struct inode *) = generic_drop_inode;
1104 if (op && op->drop_inode)
1105 drop = op->drop_inode;
1110 * iput - put an inode
1111 * @inode: inode to put
1113 * Puts an inode, dropping its usage count. If the inode use count hits
1114 * zero the inode is also then freed and may be destroyed.
1116 void iput(struct inode *inode)
1119 struct super_operations *op = inode->i_sb->s_op;
1121 if (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);
1153 EXPORT_SYMBOL(bmap);
1156 * Return true if the filesystem which backs this inode considers the two
1157 * passed timespecs to be sufficiently different to warrant flushing the
1158 * altered time out to disk.
1160 static int inode_times_differ(struct inode *inode,
1161 struct timespec *old, struct timespec *new)
1163 if (IS_ONE_SECOND(inode))
1164 return old->tv_sec != new->tv_sec;
1165 return !timespec_equal(old, new);
1169 * update_atime - update the access time
1170 * @inode: inode accessed
1172 * Update the accessed time on an inode and mark it for writeback.
1173 * This function automatically handles read only file systems and media,
1174 * as well as the "noatime" flag and inode specific "noatime" markers.
1176 void update_atime(struct inode *inode)
1178 struct timespec now;
1180 if (IS_NOATIME(inode))
1182 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1184 if (IS_RDONLY(inode))
1187 now = current_kernel_time();
1188 if (inode_times_differ(inode, &inode->i_atime, &now)) {
1189 inode->i_atime = now;
1190 mark_inode_dirty_sync(inode);
1192 if (!timespec_equal(&inode->i_atime, &now))
1193 inode->i_atime = now;
1197 EXPORT_SYMBOL(update_atime);
1200 * inode_update_time - update mtime and ctime time
1201 * @inode: inode accessed
1202 * @ctime_too: update ctime too
1204 * Update the mtime time on an inode and mark it for writeback.
1205 * When ctime_too is specified update the ctime too.
1208 void inode_update_time(struct inode *inode, int ctime_too)
1210 struct timespec now;
1213 if (IS_NOCMTIME(inode))
1215 if (IS_RDONLY(inode))
1218 now = current_kernel_time();
1220 if (inode_times_differ(inode, &inode->i_mtime, &now))
1222 inode->i_mtime = now;
1225 if (inode_times_differ(inode, &inode->i_ctime, &now))
1227 inode->i_ctime = now;
1230 mark_inode_dirty_sync(inode);
1233 EXPORT_SYMBOL(inode_update_time);
1235 int inode_needs_sync(struct inode *inode)
1239 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1244 EXPORT_SYMBOL(inode_needs_sync);
1247 * Quota functions that want to walk the inode lists..
1251 /* Function back in dquot.c */
1252 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1254 void remove_dquot_ref(struct super_block *sb, int type, struct list_head *tofree_head)
1256 struct inode *inode;
1257 struct list_head *act_head;
1260 return; /* nothing to do */
1261 spin_lock(&inode_lock); /* This lock is for inodes code */
1263 /* We hold dqptr_sem so we are safe against the quota code */
1264 list_for_each(act_head, &inode_in_use) {
1265 inode = list_entry(act_head, struct inode, i_list);
1266 if (inode->i_sb == sb && !IS_NOQUOTA(inode))
1267 remove_inode_dquot_ref(inode, type, tofree_head);
1269 list_for_each(act_head, &inode_unused) {
1270 inode = list_entry(act_head, struct inode, i_list);
1271 if (inode->i_sb == sb && !IS_NOQUOTA(inode))
1272 remove_inode_dquot_ref(inode, type, tofree_head);
1274 list_for_each(act_head, &sb->s_dirty) {
1275 inode = list_entry(act_head, struct inode, i_list);
1276 if (!IS_NOQUOTA(inode))
1277 remove_inode_dquot_ref(inode, type, tofree_head);
1279 list_for_each(act_head, &sb->s_io) {
1280 inode = list_entry(act_head, struct inode, i_list);
1281 if (!IS_NOQUOTA(inode))
1282 remove_inode_dquot_ref(inode, type, tofree_head);
1284 spin_unlock(&inode_lock);
1290 * Hashed waitqueues for wait_on_inode(). The table is pretty small - the
1291 * kernel doesn't lock many inodes at the same time.
1293 #define I_WAIT_TABLE_ORDER 3
1294 static struct i_wait_queue_head {
1295 wait_queue_head_t wqh;
1296 } ____cacheline_aligned_in_smp i_wait_queue_heads[1<<I_WAIT_TABLE_ORDER];
1299 * Return the address of the waitqueue_head to be used for this inode
1301 static wait_queue_head_t *i_waitq_head(struct inode *inode)
1303 return &i_wait_queue_heads[hash_ptr(inode, I_WAIT_TABLE_ORDER)].wqh;
1306 void __wait_on_inode(struct inode *inode)
1308 DECLARE_WAITQUEUE(wait, current);
1309 wait_queue_head_t *wq = i_waitq_head(inode);
1311 add_wait_queue(wq, &wait);
1313 set_current_state(TASK_UNINTERRUPTIBLE);
1314 if (inode->i_state & I_LOCK) {
1318 remove_wait_queue(wq, &wait);
1319 __set_current_state(TASK_RUNNING);
1323 * If we try to find an inode in the inode hash while it is being deleted, we
1324 * have to wait until the filesystem completes its deletion before reporting
1325 * that it isn't found. This is because iget will immediately call
1326 * ->read_inode, and we want to be sure that evidence of the deletion is found
1329 * This call might return early if an inode which shares the waitq is woken up.
1330 * This is most easily handled by the caller which will loop around again
1331 * looking for the inode.
1333 * This is called with inode_lock held.
1335 static void __wait_on_freeing_inode(struct inode *inode)
1337 DECLARE_WAITQUEUE(wait, current);
1338 wait_queue_head_t *wq = i_waitq_head(inode);
1340 add_wait_queue(wq, &wait);
1341 set_current_state(TASK_UNINTERRUPTIBLE);
1342 spin_unlock(&inode_lock);
1344 remove_wait_queue(wq, &wait);
1345 spin_lock(&inode_lock);
1348 void wake_up_inode(struct inode *inode)
1350 wait_queue_head_t *wq = i_waitq_head(inode);
1353 * Prevent speculative execution through spin_unlock(&inode_lock);
1356 if (waitqueue_active(wq))
1360 static __initdata unsigned long ihash_entries;
1361 static int __init set_ihash_entries(char *str)
1365 ihash_entries = simple_strtoul(str, &str, 0);
1368 __setup("ihash_entries=", set_ihash_entries);
1371 * Initialize the waitqueues and inode hash table.
1373 void __init inode_init_early(void)
1378 alloc_large_system_hash("Inode-cache",
1379 sizeof(struct hlist_head),
1386 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1387 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1390 void __init inode_init(unsigned long mempages)
1394 for (i = 0; i < ARRAY_SIZE(i_wait_queue_heads); i++)
1395 init_waitqueue_head(&i_wait_queue_heads[i].wqh);
1397 /* inode slab cache */
1398 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1399 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, init_once,
1401 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1404 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1406 inode->i_mode = mode;
1407 if (S_ISCHR(mode)) {
1408 inode->i_fop = &def_chr_fops;
1409 inode->i_rdev = rdev;
1410 } else if (S_ISBLK(mode)) {
1411 inode->i_fop = &def_blk_fops;
1412 inode->i_rdev = rdev;
1413 } else if (S_ISFIFO(mode))
1414 inode->i_fop = &def_fifo_fops;
1415 else if (S_ISSOCK(mode))
1416 inode->i_fop = &bad_sock_fops;
1418 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1421 EXPORT_SYMBOL(init_special_inode);