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 DEFINE_SPINLOCK(inode_lock);
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 struct inode *alloc_inode(struct super_block *sb)
104 static struct address_space_operations empty_aops;
105 static struct inode_operations empty_iops;
106 static struct file_operations empty_fops;
109 if (sb->s_op->alloc_inode)
110 inode = sb->s_op->alloc_inode(sb);
112 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
115 struct address_space * const mapping = &inode->i_data;
118 // inode->i_dqh = dqhget(sb->s_dqh);
120 /* essential because of inode slab reuse */
122 inode->i_blkbits = sb->s_blocksize_bits;
124 atomic_set(&inode->i_count, 1);
126 inode->i_op = &empty_iops;
127 inode->i_fop = &empty_fops;
129 atomic_set(&inode->i_writecount, 0);
133 inode->i_generation = 0;
135 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
137 inode->i_pipe = NULL;
138 inode->i_bdev = NULL;
139 inode->i_cdev = NULL;
141 inode->i_security = NULL;
142 inode->dirtied_when = 0;
143 if (security_inode_alloc(inode)) {
144 if (inode->i_sb->s_op->destroy_inode)
145 inode->i_sb->s_op->destroy_inode(inode);
147 kmem_cache_free(inode_cachep, (inode));
151 mapping->a_ops = &empty_aops;
152 mapping->host = inode;
154 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
155 mapping->assoc_mapping = NULL;
156 mapping->backing_dev_info = &default_backing_dev_info;
159 * If the block_device provides a backing_dev_info for client
160 * inodes then use that. Otherwise the inode share the bdev's
164 struct backing_dev_info *bdi;
166 bdi = sb->s_bdev->bd_inode_backing_dev_info;
168 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
169 mapping->backing_dev_info = bdi;
171 memset(&inode->u, 0, sizeof(inode->u));
172 inode->i_mapping = mapping;
177 void destroy_inode(struct inode *inode)
179 if (inode_has_buffers(inode))
181 security_inode_free(inode);
182 if (inode->i_sb->s_op->destroy_inode)
183 inode->i_sb->s_op->destroy_inode(inode);
185 kmem_cache_free(inode_cachep, (inode));
190 * These are initializations that only need to be done
191 * once, because the fields are idempotent across use
192 * of the inode, so let the slab aware of that.
194 void inode_init_once(struct inode *inode)
196 memset(inode, 0, sizeof(*inode));
197 INIT_HLIST_NODE(&inode->i_hash);
198 INIT_LIST_HEAD(&inode->i_dentry);
199 INIT_LIST_HEAD(&inode->i_devices);
200 sema_init(&inode->i_sem, 1);
201 init_rwsem(&inode->i_alloc_sem);
202 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
203 spin_lock_init(&inode->i_data.tree_lock);
204 spin_lock_init(&inode->i_data.i_mmap_lock);
205 INIT_LIST_HEAD(&inode->i_data.private_list);
206 spin_lock_init(&inode->i_data.private_lock);
207 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
208 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
209 spin_lock_init(&inode->i_lock);
210 i_size_ordered_init(inode);
213 EXPORT_SYMBOL(inode_init_once);
215 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
217 struct inode * inode = (struct inode *) foo;
219 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
220 SLAB_CTOR_CONSTRUCTOR)
221 inode_init_once(inode);
225 * inode_lock must be held
227 void __iget(struct inode * inode)
229 if (atomic_read(&inode->i_count)) {
230 atomic_inc(&inode->i_count);
233 atomic_inc(&inode->i_count);
234 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
235 list_move(&inode->i_list, &inode_in_use);
236 inodes_stat.nr_unused--;
239 EXPORT_SYMBOL_GPL(__iget);
242 * clear_inode - clear an inode
243 * @inode: inode to clear
245 * This is called by the filesystem to tell us
246 * that the inode is no longer useful. We just
247 * terminate it with extreme prejudice.
249 void clear_inode(struct inode *inode)
252 invalidate_inode_buffers(inode);
254 if (inode->i_data.nrpages)
256 if (!(inode->i_state & I_FREEING))
258 if (inode->i_state & I_CLEAR)
260 wait_on_inode(inode);
262 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
263 inode->i_sb->s_op->clear_inode(inode);
268 inode->i_state = I_CLEAR;
271 EXPORT_SYMBOL(clear_inode);
274 * dispose_list - dispose of the contents of a local list
275 * @head: the head of the list to free
277 * Dispose-list gets a local list with local inodes in it, so it doesn't
278 * need to worry about list corruption and SMP locks.
280 static void dispose_list(struct list_head *head)
284 while (!list_empty(head)) {
287 inode = list_entry(head->next, struct inode, i_list);
288 list_del(&inode->i_list);
290 if (inode->i_data.nrpages)
291 truncate_inode_pages(&inode->i_data, 0);
293 destroy_inode(inode);
296 spin_lock(&inode_lock);
297 inodes_stat.nr_inodes -= nr_disposed;
298 spin_unlock(&inode_lock);
302 * Invalidate all inodes for a device.
304 static int invalidate_list(struct list_head *head, struct list_head *dispose)
306 struct list_head *next;
307 int busy = 0, count = 0;
311 struct list_head * tmp = next;
312 struct inode * inode;
315 * We can reschedule here without worrying about the list's
316 * consistency because the per-sb list of inodes must not
317 * change during umount anymore, and because iprune_sem keeps
318 * shrink_icache_memory() away.
320 cond_resched_lock(&inode_lock);
325 inode = list_entry(tmp, struct inode, i_sb_list);
326 invalidate_inode_buffers(inode);
327 if (!atomic_read(&inode->i_count)) {
328 hlist_del_init(&inode->i_hash);
329 list_del(&inode->i_sb_list);
330 list_move(&inode->i_list, dispose);
331 inode->i_state |= I_FREEING;
337 /* only unused inodes may be cached with i_count zero */
338 inodes_stat.nr_unused -= count;
343 * This is a two-stage process. First we collect all
344 * offending inodes onto the throw-away list, and in
345 * the second stage we actually dispose of them. This
346 * is because we don't want to sleep while messing
347 * with the global lists..
351 * invalidate_inodes - discard the inodes on a device
354 * Discard all of the inodes for a given superblock. If the discard
355 * fails because there are busy inodes then a non zero value is returned.
356 * If the discard is successful all the inodes have been discarded.
358 int invalidate_inodes(struct super_block * sb)
361 LIST_HEAD(throw_away);
364 spin_lock(&inode_lock);
365 busy = invalidate_list(&sb->s_inodes, &throw_away);
366 spin_unlock(&inode_lock);
368 dispose_list(&throw_away);
374 EXPORT_SYMBOL(invalidate_inodes);
376 int __invalidate_device(struct block_device *bdev, int do_sync)
378 struct super_block *sb;
385 sb = get_super(bdev);
388 * no need to lock the super, get_super holds the
389 * read semaphore so the filesystem cannot go away
390 * under us (->put_super runs with the write lock
393 shrink_dcache_sb(sb);
394 res = invalidate_inodes(sb);
397 invalidate_bdev(bdev, 0);
401 EXPORT_SYMBOL(__invalidate_device);
403 static int can_unuse(struct inode *inode)
407 if (inode_has_buffers(inode))
409 if (atomic_read(&inode->i_count))
411 if (inode->i_data.nrpages)
417 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
418 * a temporary list and then are freed outside inode_lock by dispose_list().
420 * Any inodes which are pinned purely because of attached pagecache have their
421 * pagecache removed. We expect the final iput() on that inode to add it to
422 * the front of the inode_unused list. So look for it there and if the
423 * inode is still freeable, proceed. The right inode is found 99.9% of the
424 * time in testing on a 4-way.
426 * If the inode has metadata buffers attached to mapping->private_list then
427 * try to remove them.
429 static void prune_icache(int nr_to_scan)
434 unsigned long reap = 0;
437 spin_lock(&inode_lock);
438 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
441 if (list_empty(&inode_unused))
444 inode = list_entry(inode_unused.prev, struct inode, i_list);
446 if (inode->i_state || atomic_read(&inode->i_count)) {
447 list_move(&inode->i_list, &inode_unused);
450 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
452 spin_unlock(&inode_lock);
453 if (remove_inode_buffers(inode))
454 reap += invalidate_inode_pages(&inode->i_data);
456 spin_lock(&inode_lock);
458 if (inode != list_entry(inode_unused.next,
459 struct inode, i_list))
460 continue; /* wrong inode or list_empty */
461 if (!can_unuse(inode))
464 hlist_del_init(&inode->i_hash);
465 list_del_init(&inode->i_sb_list);
466 list_move(&inode->i_list, &freeable);
467 inode->i_state |= I_FREEING;
470 inodes_stat.nr_unused -= nr_pruned;
471 spin_unlock(&inode_lock);
473 dispose_list(&freeable);
476 if (current_is_kswapd())
477 mod_page_state(kswapd_inodesteal, reap);
479 mod_page_state(pginodesteal, reap);
483 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
484 * "unused" means that no dentries are referring to the inodes: the files are
485 * not open and the dcache references to those inodes have already been
488 * This function is passed the number of inodes to scan, and it returns the
489 * total number of remaining possibly-reclaimable inodes.
491 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
495 * Nasty deadlock avoidance. We may hold various FS locks,
496 * and we don't want to recurse into the FS that called us
497 * in clear_inode() and friends..
499 if (!(gfp_mask & __GFP_FS))
503 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
506 static void __wait_on_freeing_inode(struct inode *inode);
508 * Called with the inode lock held.
509 * NOTE: we are not increasing the inode-refcount, you must call __iget()
510 * by hand after calling find_inode now! This simplifies iunique and won't
511 * add any additional branch in the common code.
513 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
515 struct hlist_node *node;
516 struct inode * inode = NULL;
519 hlist_for_each (node, head) {
520 inode = hlist_entry(node, struct inode, i_hash);
521 if (inode->i_sb != sb)
523 if (!test(inode, data))
525 if (inode->i_state & (I_FREEING|I_CLEAR)) {
526 __wait_on_freeing_inode(inode);
531 return node ? inode : NULL;
535 * find_inode_fast is the fast path version of find_inode, see the comment at
536 * iget_locked for details.
538 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
540 struct hlist_node *node;
541 struct inode * inode = NULL;
544 hlist_for_each (node, head) {
545 inode = hlist_entry(node, struct inode, i_hash);
546 if (inode->i_ino != ino)
548 if (inode->i_sb != sb)
550 if (inode->i_state & (I_FREEING|I_CLEAR)) {
551 __wait_on_freeing_inode(inode);
556 return node ? inode : NULL;
560 * new_inode - obtain an inode
563 * Allocates a new inode for given superblock.
565 struct inode *new_inode(struct super_block *sb)
567 static unsigned long last_ino;
568 struct inode * inode;
570 spin_lock_prefetch(&inode_lock);
572 inode = alloc_inode(sb);
574 spin_lock(&inode_lock);
575 inodes_stat.nr_inodes++;
576 list_add(&inode->i_list, &inode_in_use);
577 list_add(&inode->i_sb_list, &sb->s_inodes);
578 inode->i_ino = ++last_ino;
580 spin_unlock(&inode_lock);
585 EXPORT_SYMBOL(new_inode);
587 void unlock_new_inode(struct inode *inode)
590 * This is special! We do not need the spinlock
591 * when clearing I_LOCK, because we're guaranteed
592 * that nobody else tries to do anything about the
593 * state of the inode when it is locked, as we
594 * just created it (so there can be no old holders
595 * that haven't tested I_LOCK).
597 inode->i_state &= ~(I_LOCK|I_NEW);
598 wake_up_inode(inode);
601 EXPORT_SYMBOL(unlock_new_inode);
604 * This is called without the inode lock held.. Be careful.
606 * We no longer cache the sb_flags in i_flags - see fs.h
607 * -- rmk@arm.uk.linux.org
609 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)
611 struct inode * inode;
613 inode = alloc_inode(sb);
617 spin_lock(&inode_lock);
618 /* We released the lock, so.. */
619 old = find_inode(sb, head, test, data);
621 if (set(inode, data))
624 inodes_stat.nr_inodes++;
625 list_add(&inode->i_list, &inode_in_use);
626 list_add(&inode->i_sb_list, &sb->s_inodes);
627 hlist_add_head(&inode->i_hash, head);
628 inode->i_state = I_LOCK|I_NEW;
629 spin_unlock(&inode_lock);
631 /* Return the locked inode with I_NEW set, the
632 * caller is responsible for filling in the contents
638 * Uhhuh, somebody else created the same inode under
639 * us. Use the old inode instead of the one we just
643 spin_unlock(&inode_lock);
644 destroy_inode(inode);
646 wait_on_inode(inode);
651 spin_unlock(&inode_lock);
652 destroy_inode(inode);
657 * get_new_inode_fast is the fast path version of get_new_inode, see the
658 * comment at iget_locked for details.
660 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
662 struct inode * inode;
664 inode = alloc_inode(sb);
668 spin_lock(&inode_lock);
669 /* We released the lock, so.. */
670 old = find_inode_fast(sb, head, ino);
673 inodes_stat.nr_inodes++;
674 list_add(&inode->i_list, &inode_in_use);
675 list_add(&inode->i_sb_list, &sb->s_inodes);
676 hlist_add_head(&inode->i_hash, head);
677 inode->i_state = I_LOCK|I_NEW;
678 spin_unlock(&inode_lock);
680 /* Return the locked inode with I_NEW set, the
681 * caller is responsible for filling in the contents
687 * Uhhuh, somebody else created the same inode under
688 * us. Use the old inode instead of the one we just
692 spin_unlock(&inode_lock);
693 destroy_inode(inode);
695 wait_on_inode(inode);
700 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
704 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
706 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
707 return tmp & I_HASHMASK;
711 * iunique - get a unique inode number
713 * @max_reserved: highest reserved inode number
715 * Obtain an inode number that is unique on the system for a given
716 * superblock. This is used by file systems that have no natural
717 * permanent inode numbering system. An inode number is returned that
718 * is higher than the reserved limit but unique.
721 * With a large number of inodes live on the file system this function
722 * currently becomes quite slow.
724 ino_t iunique(struct super_block *sb, ino_t max_reserved)
726 static ino_t counter;
728 struct hlist_head * head;
730 spin_lock(&inode_lock);
732 if (counter > max_reserved) {
733 head = inode_hashtable + hash(sb,counter);
735 inode = find_inode_fast(sb, head, res);
737 spin_unlock(&inode_lock);
741 counter = max_reserved + 1;
747 EXPORT_SYMBOL(iunique);
749 struct inode *igrab(struct inode *inode)
751 spin_lock(&inode_lock);
752 if (!(inode->i_state & I_FREEING))
756 * Handle the case where s_op->clear_inode is not been
757 * called yet, and somebody is calling igrab
758 * while the inode is getting freed.
761 spin_unlock(&inode_lock);
765 EXPORT_SYMBOL(igrab);
768 * ifind - internal function, you want ilookup5() or iget5().
769 * @sb: super block of file system to search
770 * @head: the head of the list to search
771 * @test: callback used for comparisons between inodes
772 * @data: opaque data pointer to pass to @test
774 * ifind() searches for the inode specified by @data in the inode
775 * cache. This is a generalized version of ifind_fast() for file systems where
776 * the inode number is not sufficient for unique identification of an inode.
778 * If the inode is in the cache, the inode is returned with an incremented
781 * Otherwise NULL is returned.
783 * Note, @test is called with the inode_lock held, so can't sleep.
785 static inline struct inode *ifind(struct super_block *sb,
786 struct hlist_head *head, int (*test)(struct inode *, void *),
791 spin_lock(&inode_lock);
792 inode = find_inode(sb, head, test, data);
795 spin_unlock(&inode_lock);
796 wait_on_inode(inode);
799 spin_unlock(&inode_lock);
804 * ifind_fast - internal function, you want ilookup() or iget().
805 * @sb: super block of file system to search
806 * @head: head of the list to search
807 * @ino: inode number to search for
809 * ifind_fast() searches for the inode @ino in the inode cache. This is for
810 * file systems where the inode number is sufficient for unique identification
813 * If the inode is in the cache, the inode is returned with an incremented
816 * Otherwise NULL is returned.
818 static inline struct inode *ifind_fast(struct super_block *sb,
819 struct hlist_head *head, unsigned long ino)
823 spin_lock(&inode_lock);
824 inode = find_inode_fast(sb, head, ino);
827 spin_unlock(&inode_lock);
828 wait_on_inode(inode);
831 spin_unlock(&inode_lock);
836 * ilookup5 - search for an inode in the inode cache
837 * @sb: super block of file system to search
838 * @hashval: hash value (usually inode number) to search for
839 * @test: callback used for comparisons between inodes
840 * @data: opaque data pointer to pass to @test
842 * ilookup5() uses ifind() to search for the inode specified by @hashval and
843 * @data in the inode cache. This is a generalized version of ilookup() for
844 * file systems where the inode number is not sufficient for unique
845 * identification of an inode.
847 * If the inode is in the cache, the inode is returned with an incremented
850 * Otherwise NULL is returned.
852 * Note, @test is called with the inode_lock held, so can't sleep.
854 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
855 int (*test)(struct inode *, void *), void *data)
857 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
859 return ifind(sb, head, test, data);
862 EXPORT_SYMBOL(ilookup5);
865 * ilookup - search for an inode in the inode cache
866 * @sb: super block of file system to search
867 * @ino: inode number to search for
869 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
870 * This is for file systems where the inode number is sufficient for unique
871 * identification of an inode.
873 * If the inode is in the cache, the inode is returned with an incremented
876 * Otherwise NULL is returned.
878 struct inode *ilookup(struct super_block *sb, unsigned long ino)
880 struct hlist_head *head = inode_hashtable + hash(sb, ino);
882 return ifind_fast(sb, head, ino);
885 EXPORT_SYMBOL(ilookup);
888 * iget5_locked - obtain an inode from a mounted file system
889 * @sb: super block of file system
890 * @hashval: hash value (usually inode number) to get
891 * @test: callback used for comparisons between inodes
892 * @set: callback used to initialize a new struct inode
893 * @data: opaque data pointer to pass to @test and @set
895 * This is iget() without the read_inode() portion of get_new_inode().
897 * iget5_locked() uses ifind() to search for the inode specified by @hashval
898 * and @data in the inode cache and if present it is returned with an increased
899 * reference count. This is a generalized version of iget_locked() for file
900 * systems where the inode number is not sufficient for unique identification
903 * If the inode is not in cache, get_new_inode() is called to allocate a new
904 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
905 * file system gets to fill it in before unlocking it via unlock_new_inode().
907 * Note both @test and @set are called with the inode_lock held, so can't sleep.
909 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
910 int (*test)(struct inode *, void *),
911 int (*set)(struct inode *, void *), void *data)
913 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
916 inode = ifind(sb, head, test, data);
920 * get_new_inode() will do the right thing, re-trying the search
921 * in case it had to block at any point.
923 return get_new_inode(sb, head, test, set, data);
926 EXPORT_SYMBOL(iget5_locked);
929 * iget_locked - obtain an inode from a mounted file system
930 * @sb: super block of file system
931 * @ino: inode number to get
933 * This is iget() without the read_inode() portion of get_new_inode_fast().
935 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
936 * the inode cache and if present it is returned with an increased reference
937 * count. This is for file systems where the inode number is sufficient for
938 * unique identification of an inode.
940 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
941 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
942 * The file system gets to fill it in before unlocking it via
943 * unlock_new_inode().
945 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
947 struct hlist_head *head = inode_hashtable + hash(sb, ino);
950 inode = ifind_fast(sb, head, ino);
954 * get_new_inode_fast() will do the right thing, re-trying the search
955 * in case it had to block at any point.
957 return get_new_inode_fast(sb, head, ino);
960 EXPORT_SYMBOL(iget_locked);
963 * __insert_inode_hash - hash an inode
964 * @inode: unhashed inode
965 * @hashval: unsigned long value used to locate this object in the
968 * Add an inode to the inode hash for this superblock.
970 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
972 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
973 spin_lock(&inode_lock);
974 hlist_add_head(&inode->i_hash, head);
975 spin_unlock(&inode_lock);
978 EXPORT_SYMBOL(__insert_inode_hash);
981 * remove_inode_hash - remove an inode from the hash
982 * @inode: inode to unhash
984 * Remove an inode from the superblock.
986 void remove_inode_hash(struct inode *inode)
988 spin_lock(&inode_lock);
989 hlist_del_init(&inode->i_hash);
990 spin_unlock(&inode_lock);
993 EXPORT_SYMBOL(remove_inode_hash);
996 * Tell the filesystem that this inode is no longer of any interest and should
997 * be completely destroyed.
999 * We leave the inode in the inode hash table until *after* the filesystem's
1000 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1001 * instigate) will always find up-to-date information either in the hash or on
1004 * I_FREEING is set so that no-one will take a new reference to the inode while
1005 * it is being deleted.
1007 void generic_delete_inode(struct inode *inode)
1009 struct super_operations *op = inode->i_sb->s_op;
1011 list_del_init(&inode->i_list);
1012 list_del_init(&inode->i_sb_list);
1013 inode->i_state|=I_FREEING;
1014 inodes_stat.nr_inodes--;
1015 spin_unlock(&inode_lock);
1017 if (inode->i_data.nrpages)
1018 truncate_inode_pages(&inode->i_data, 0);
1020 security_inode_delete(inode);
1022 if (op->delete_inode) {
1023 void (*delete)(struct inode *) = op->delete_inode;
1024 if (!is_bad_inode(inode))
1026 /* s_op->delete_inode internally recalls clear_inode() */
1030 spin_lock(&inode_lock);
1031 hlist_del_init(&inode->i_hash);
1032 spin_unlock(&inode_lock);
1033 wake_up_inode(inode);
1034 if (inode->i_state != I_CLEAR)
1036 destroy_inode(inode);
1039 EXPORT_SYMBOL(generic_delete_inode);
1041 static void generic_forget_inode(struct inode *inode)
1043 struct super_block *sb = inode->i_sb;
1045 if (!hlist_unhashed(&inode->i_hash)) {
1046 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1047 list_move(&inode->i_list, &inode_unused);
1048 inodes_stat.nr_unused++;
1049 spin_unlock(&inode_lock);
1050 if (!sb || (sb->s_flags & MS_ACTIVE))
1052 write_inode_now(inode, 1);
1053 spin_lock(&inode_lock);
1054 inodes_stat.nr_unused--;
1055 hlist_del_init(&inode->i_hash);
1057 list_del_init(&inode->i_list);
1058 list_del_init(&inode->i_sb_list);
1059 inode->i_state|=I_FREEING;
1060 inodes_stat.nr_inodes--;
1061 spin_unlock(&inode_lock);
1062 if (inode->i_data.nrpages)
1063 truncate_inode_pages(&inode->i_data, 0);
1065 destroy_inode(inode);
1069 * Normal UNIX filesystem behaviour: delete the
1070 * inode when the usage count drops to zero, and
1073 static void generic_drop_inode(struct inode *inode)
1075 if (!inode->i_nlink)
1076 generic_delete_inode(inode);
1078 generic_forget_inode(inode);
1082 * Called when we're dropping the last reference
1085 * Call the FS "drop()" function, defaulting to
1086 * the legacy UNIX filesystem behaviour..
1088 * NOTE! NOTE! NOTE! We're called with the inode lock
1089 * held, and the drop function is supposed to release
1092 static inline void iput_final(struct inode *inode)
1094 struct super_operations *op = inode->i_sb->s_op;
1095 void (*drop)(struct inode *) = generic_drop_inode;
1097 if (op && op->drop_inode)
1098 drop = op->drop_inode;
1103 * iput - put an inode
1104 * @inode: inode to put
1106 * Puts an inode, dropping its usage count. If the inode use count hits
1107 * zero the inode is also then freed and may be destroyed.
1109 void iput(struct inode *inode)
1112 struct super_operations *op = inode->i_sb->s_op;
1114 if (inode->i_state == I_CLEAR)
1117 if (op && op->put_inode)
1118 op->put_inode(inode);
1120 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1125 EXPORT_SYMBOL(iput);
1128 * bmap - find a block number in a file
1129 * @inode: inode of file
1130 * @block: block to find
1132 * Returns the block number on the device holding the inode that
1133 * is the disk block number for the block of the file requested.
1134 * That is, asked for block 4 of inode 1 the function will return the
1135 * disk block relative to the disk start that holds that block of the
1138 sector_t bmap(struct inode * inode, sector_t block)
1141 if (inode->i_mapping->a_ops->bmap)
1142 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1146 EXPORT_SYMBOL(bmap);
1149 * update_atime - update the access time
1150 * @inode: inode accessed
1152 * Update the accessed time on an inode and mark it for writeback.
1153 * This function automatically handles read only file systems and media,
1154 * as well as the "noatime" flag and inode specific "noatime" markers.
1156 void update_atime(struct inode *inode)
1158 struct timespec now;
1160 if (IS_NOATIME(inode))
1162 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1164 if (IS_RDONLY(inode))
1167 now = current_fs_time(inode->i_sb);
1168 if (!timespec_equal(&inode->i_atime, &now)) {
1169 inode->i_atime = now;
1170 mark_inode_dirty_sync(inode);
1172 if (!timespec_equal(&inode->i_atime, &now))
1173 inode->i_atime = now;
1177 EXPORT_SYMBOL(update_atime);
1180 * inode_update_time - update mtime and ctime time
1181 * @inode: inode accessed
1182 * @ctime_too: update ctime too
1184 * Update the mtime time on an inode and mark it for writeback.
1185 * When ctime_too is specified update the ctime too.
1188 void inode_update_time(struct inode *inode, int ctime_too)
1190 struct timespec now;
1193 if (IS_NOCMTIME(inode))
1195 if (IS_RDONLY(inode))
1198 now = current_fs_time(inode->i_sb);
1199 if (!timespec_equal(&inode->i_mtime, &now))
1201 inode->i_mtime = now;
1204 if (!timespec_equal(&inode->i_ctime, &now))
1206 inode->i_ctime = now;
1209 mark_inode_dirty_sync(inode);
1212 EXPORT_SYMBOL(inode_update_time);
1214 int inode_needs_sync(struct inode *inode)
1218 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1223 EXPORT_SYMBOL(inode_needs_sync);
1226 * Quota functions that want to walk the inode lists..
1230 /* Function back in dquot.c */
1231 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1233 void remove_dquot_ref(struct super_block *sb, int type,
1234 struct list_head *tofree_head)
1236 struct inode *inode;
1239 return; /* nothing to do */
1240 spin_lock(&inode_lock); /* This lock is for inodes code */
1243 * We don't have to lock against quota code - test IS_QUOTAINIT is
1244 * just for speedup...
1246 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1247 if (!IS_NOQUOTA(inode))
1248 remove_inode_dquot_ref(inode, type, tofree_head);
1250 spin_unlock(&inode_lock);
1255 int inode_wait(void *word)
1262 * If we try to find an inode in the inode hash while it is being deleted, we
1263 * have to wait until the filesystem completes its deletion before reporting
1264 * that it isn't found. This is because iget will immediately call
1265 * ->read_inode, and we want to be sure that evidence of the deletion is found
1267 * This is called with inode_lock held.
1269 static void __wait_on_freeing_inode(struct inode *inode)
1271 wait_queue_head_t *wq;
1272 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1275 * I_FREEING and I_CLEAR are cleared in process context under
1276 * inode_lock, so we have to give the tasks who would clear them
1277 * a chance to run and acquire inode_lock.
1279 if (!(inode->i_state & I_LOCK)) {
1280 spin_unlock(&inode_lock);
1282 spin_lock(&inode_lock);
1285 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1286 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1287 spin_unlock(&inode_lock);
1289 finish_wait(wq, &wait.wait);
1290 spin_lock(&inode_lock);
1293 void wake_up_inode(struct inode *inode)
1296 * Prevent speculative execution through spin_unlock(&inode_lock);
1299 wake_up_bit(&inode->i_state, __I_LOCK);
1302 static __initdata unsigned long ihash_entries;
1303 static int __init set_ihash_entries(char *str)
1307 ihash_entries = simple_strtoul(str, &str, 0);
1310 __setup("ihash_entries=", set_ihash_entries);
1313 * Initialize the waitqueues and inode hash table.
1315 void __init inode_init_early(void)
1319 /* If hashes are distributed across NUMA nodes, defer
1320 * hash allocation until vmalloc space is available.
1326 alloc_large_system_hash("Inode-cache",
1327 sizeof(struct hlist_head),
1335 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1336 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1339 void __init inode_init(unsigned long mempages)
1343 /* inode slab cache */
1344 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1345 0, SLAB_PANIC, init_once, NULL);
1346 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1348 /* Hash may have been set up in inode_init_early */
1353 alloc_large_system_hash("Inode-cache",
1354 sizeof(struct hlist_head),
1362 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1363 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1366 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1368 inode->i_mode = mode;
1369 if (S_ISCHR(mode)) {
1370 inode->i_fop = &def_chr_fops;
1371 inode->i_rdev = rdev;
1372 } else if (S_ISBLK(mode)) {
1373 inode->i_fop = &def_blk_fops;
1374 inode->i_rdev = rdev;
1375 } else if (S_ISFIFO(mode))
1376 inode->i_fop = &def_fifo_fops;
1377 else if (S_ISSOCK(mode))
1378 inode->i_fop = &bad_sock_fops;
1380 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1383 EXPORT_SYMBOL(init_special_inode);