4 * (C) 1997 Linus Torvalds
7 #include <linux/config.h>
10 #include <linux/dcache.h>
11 #include <linux/init.h>
12 #include <linux/quotaops.h>
13 #include <linux/slab.h>
14 #include <linux/writeback.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/wait.h>
18 #include <linux/hash.h>
19 #include <linux/swap.h>
20 #include <linux/security.h>
21 #include <linux/pagemap.h>
22 #include <linux/cdev.h>
23 #include <linux/bootmem.h>
24 #include <linux/vs_base.h>
27 * This is needed for the following functions:
29 * - invalidate_inode_buffers
33 * FIXME: remove all knowledge of the buffer layer from this file
35 #include <linux/buffer_head.h>
38 * New inode.c implementation.
40 * This implementation has the basic premise of trying
41 * to be extremely low-overhead and SMP-safe, yet be
42 * simple enough to be "obviously correct".
47 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
49 /* #define INODE_PARANOIA 1 */
50 /* #define INODE_DEBUG 1 */
53 * Inode lookup is no longer as critical as it used to be:
54 * most of the lookups are going to be through the dcache.
56 #define I_HASHBITS i_hash_shift
57 #define I_HASHMASK i_hash_mask
59 static unsigned int i_hash_mask;
60 static unsigned int i_hash_shift;
63 * Each inode can be on two separate lists. One is
64 * the hash list of the inode, used for lookups. The
65 * other linked list is the "type" list:
66 * "in_use" - valid inode, i_count > 0, i_nlink > 0
67 * "dirty" - as "in_use" but also dirty
68 * "unused" - valid inode, i_count = 0
70 * A "dirty" list is maintained for each super block,
71 * allowing for low-overhead inode sync() operations.
74 LIST_HEAD(inode_in_use);
75 LIST_HEAD(inode_unused);
76 static struct hlist_head *inode_hashtable;
79 * A simple spinlock to protect the list manipulations.
81 * NOTE! You also have to own the lock if you change
82 * the i_state of an inode while it is in use..
84 spinlock_t inode_lock = SPIN_LOCK_UNLOCKED;
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;
101 static kmem_cache_t * inode_cachep;
103 static void prune_icache(int nr_to_scan);
106 #define INODE_UNUSED_THRESHOLD 15000
107 #define PRUNE_BATCH_COUNT 32
109 void try_to_clip_inodes(void)
111 unsigned long count = 0;
112 /* if there are a LOT of unused inodes in cache, better shrink a few first */
114 /* check lockless first to not take the lock always here; racing occasionally isn't a big deal */
115 if (inodes_stat.nr_unused > INODE_UNUSED_THRESHOLD) {
116 spin_lock(&inode_lock);
117 if (inodes_stat.nr_unused > INODE_UNUSED_THRESHOLD)
118 count = inodes_stat.nr_unused - INODE_UNUSED_THRESHOLD;
119 spin_unlock(&inode_lock);
126 static struct inode *alloc_inode(struct super_block *sb)
128 static struct address_space_operations empty_aops;
129 static struct inode_operations empty_iops;
130 static struct file_operations empty_fops;
133 if (sb->s_op->alloc_inode)
134 inode = sb->s_op->alloc_inode(sb);
136 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
139 struct address_space * const mapping = &inode->i_data;
142 // inode->i_dqh = dqhget(sb->s_dqh);
144 /* important because of inode slab reuse */
146 inode->i_blkbits = sb->s_blocksize_bits;
148 atomic_set(&inode->i_count, 1);
150 inode->i_op = &empty_iops;
151 inode->i_fop = &empty_fops;
153 atomic_set(&inode->i_writecount, 0);
157 inode->i_generation = 0;
159 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
161 inode->i_pipe = NULL;
162 inode->i_bdev = NULL;
163 inode->i_cdev = NULL;
165 inode->i_security = NULL;
166 inode->dirtied_when = 0;
167 if (security_inode_alloc(inode)) {
168 if (inode->i_sb->s_op->destroy_inode)
169 inode->i_sb->s_op->destroy_inode(inode);
171 kmem_cache_free(inode_cachep, (inode));
175 mapping->a_ops = &empty_aops;
176 mapping->host = inode;
178 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
179 mapping->assoc_mapping = NULL;
180 mapping->backing_dev_info = &default_backing_dev_info;
183 * If the block_device provides a backing_dev_info for client
184 * inodes then use that. Otherwise the inode share the bdev's
188 struct backing_dev_info *bdi;
190 bdi = sb->s_bdev->bd_inode_backing_dev_info;
192 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
193 mapping->backing_dev_info = bdi;
195 memset(&inode->u, 0, sizeof(inode->u));
196 inode->i_mapping = mapping;
201 void destroy_inode(struct inode *inode)
203 if (inode_has_buffers(inode))
205 security_inode_free(inode);
206 if (inode->i_sb->s_op->destroy_inode)
207 inode->i_sb->s_op->destroy_inode(inode);
209 kmem_cache_free(inode_cachep, (inode));
214 * These are initializations that only need to be done
215 * once, because the fields are idempotent across use
216 * of the inode, so let the slab aware of that.
218 void inode_init_once(struct inode *inode)
220 memset(inode, 0, sizeof(*inode));
221 INIT_HLIST_NODE(&inode->i_hash);
222 INIT_LIST_HEAD(&inode->i_dentry);
223 INIT_LIST_HEAD(&inode->i_devices);
224 sema_init(&inode->i_sem, 1);
225 init_rwsem(&inode->i_alloc_sem);
226 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
227 spin_lock_init(&inode->i_data.tree_lock);
228 spin_lock_init(&inode->i_data.i_mmap_lock);
229 atomic_set(&inode->i_data.truncate_count, 0);
230 INIT_LIST_HEAD(&inode->i_data.private_list);
231 spin_lock_init(&inode->i_data.private_lock);
232 INIT_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
233 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
234 spin_lock_init(&inode->i_lock);
235 i_size_ordered_init(inode);
238 EXPORT_SYMBOL(inode_init_once);
240 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
242 struct inode * inode = (struct inode *) foo;
244 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
245 SLAB_CTOR_CONSTRUCTOR)
246 inode_init_once(inode);
250 * inode_lock must be held
252 void __iget(struct inode * inode)
254 if (atomic_read(&inode->i_count)) {
255 atomic_inc(&inode->i_count);
258 atomic_inc(&inode->i_count);
259 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
260 list_move(&inode->i_list, &inode_in_use);
261 inodes_stat.nr_unused--;
265 * clear_inode - clear an inode
266 * @inode: inode to clear
268 * This is called by the filesystem to tell us
269 * that the inode is no longer useful. We just
270 * terminate it with extreme prejudice.
272 void clear_inode(struct inode *inode)
275 invalidate_inode_buffers(inode);
277 if (inode->i_data.nrpages)
279 if (!(inode->i_state & I_FREEING))
281 if (inode->i_state & I_CLEAR)
283 wait_on_inode(inode);
285 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
286 inode->i_sb->s_op->clear_inode(inode);
291 inode->i_state = I_CLEAR;
294 EXPORT_SYMBOL(clear_inode);
297 * dispose_list - dispose of the contents of a local list
298 * @head: the head of the list to free
300 * Dispose-list gets a local list with local inodes in it, so it doesn't
301 * need to worry about list corruption and SMP locks.
303 static void dispose_list(struct list_head *head)
307 while (!list_empty(head)) {
310 inode = list_entry(head->next, struct inode, i_list);
311 list_del(&inode->i_list);
313 if (inode->i_data.nrpages)
314 truncate_inode_pages(&inode->i_data, 0);
316 destroy_inode(inode);
319 spin_lock(&inode_lock);
320 inodes_stat.nr_inodes -= nr_disposed;
321 spin_unlock(&inode_lock);
325 * Invalidate all inodes for a device.
327 static int invalidate_list(struct list_head *head, struct super_block * sb, struct list_head * dispose)
329 struct list_head *next;
330 int busy = 0, count = 0;
334 struct list_head * tmp = next;
335 struct inode * inode;
340 inode = list_entry(tmp, struct inode, i_list);
341 if (inode->i_sb != sb)
343 invalidate_inode_buffers(inode);
344 if (!atomic_read(&inode->i_count)) {
345 hlist_del_init(&inode->i_hash);
346 list_move(&inode->i_list, dispose);
347 inode->i_state |= I_FREEING;
353 /* only unused inodes may be cached with i_count zero */
354 inodes_stat.nr_unused -= count;
359 * This is a two-stage process. First we collect all
360 * offending inodes onto the throw-away list, and in
361 * the second stage we actually dispose of them. This
362 * is because we don't want to sleep while messing
363 * with the global lists..
367 * invalidate_inodes - discard the inodes on a device
370 * Discard all of the inodes for a given superblock. If the discard
371 * fails because there are busy inodes then a non zero value is returned.
372 * If the discard is successful all the inodes have been discarded.
374 int invalidate_inodes(struct super_block * sb)
377 LIST_HEAD(throw_away);
380 spin_lock(&inode_lock);
381 busy = invalidate_list(&inode_in_use, sb, &throw_away);
382 busy |= invalidate_list(&inode_unused, sb, &throw_away);
383 busy |= invalidate_list(&sb->s_dirty, sb, &throw_away);
384 busy |= invalidate_list(&sb->s_io, sb, &throw_away);
385 spin_unlock(&inode_lock);
387 dispose_list(&throw_away);
393 EXPORT_SYMBOL(invalidate_inodes);
395 int __invalidate_device(struct block_device *bdev, int do_sync)
397 struct super_block *sb;
404 sb = get_super(bdev);
407 * no need to lock the super, get_super holds the
408 * read semaphore so the filesystem cannot go away
409 * under us (->put_super runs with the write lock
412 shrink_dcache_sb(sb);
413 res = invalidate_inodes(sb);
416 invalidate_bdev(bdev, 0);
420 EXPORT_SYMBOL(__invalidate_device);
422 static int can_unuse(struct inode *inode)
426 if (inode_has_buffers(inode))
428 if (atomic_read(&inode->i_count))
430 if (inode->i_data.nrpages)
436 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
437 * a temporary list and then are freed outside inode_lock by dispose_list().
439 * Any inodes which are pinned purely because of attached pagecache have their
440 * pagecache removed. We expect the final iput() on that inode to add it to
441 * the front of the inode_unused list. So look for it there and if the
442 * inode is still freeable, proceed. The right inode is found 99.9% of the
443 * time in testing on a 4-way.
445 * If the inode has metadata buffers attached to mapping->private_list then
446 * try to remove them.
448 static void prune_icache(int nr_to_scan)
453 unsigned long reap = 0;
456 spin_lock(&inode_lock);
457 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
460 if (list_empty(&inode_unused))
463 inode = list_entry(inode_unused.prev, struct inode, i_list);
465 if (inode->i_state || atomic_read(&inode->i_count)) {
466 list_move(&inode->i_list, &inode_unused);
469 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
471 spin_unlock(&inode_lock);
472 if (remove_inode_buffers(inode))
473 reap += invalidate_inode_pages(&inode->i_data);
475 spin_lock(&inode_lock);
477 if (inode != list_entry(inode_unused.next,
478 struct inode, i_list))
479 continue; /* wrong inode or list_empty */
480 if (!can_unuse(inode))
483 hlist_del_init(&inode->i_hash);
484 list_move(&inode->i_list, &freeable);
485 inode->i_state |= I_FREEING;
488 inodes_stat.nr_unused -= nr_pruned;
489 spin_unlock(&inode_lock);
491 dispose_list(&freeable);
494 if (current_is_kswapd())
495 mod_page_state(kswapd_inodesteal, reap);
497 mod_page_state(pginodesteal, reap);
501 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
502 * "unused" means that no dentries are referring to the inodes: the files are
503 * not open and the dcache references to those inodes have already been
506 * This function is passed the number of inodes to scan, and it returns the
507 * total number of remaining possibly-reclaimable inodes.
509 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
513 * Nasty deadlock avoidance. We may hold various FS locks,
514 * and we don't want to recurse into the FS that called us
515 * in clear_inode() and friends..
517 if (gfp_mask & __GFP_FS)
520 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
523 static void __wait_on_freeing_inode(struct inode *inode);
525 * Called with the inode lock held.
526 * NOTE: we are not increasing the inode-refcount, you must call __iget()
527 * by hand after calling find_inode now! This simplifies iunique and won't
528 * add any additional branch in the common code.
530 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
532 struct hlist_node *node;
533 struct inode * inode = NULL;
536 hlist_for_each (node, head) {
537 inode = hlist_entry(node, struct inode, i_hash);
538 if (inode->i_sb != sb)
540 if (!test(inode, data))
542 if (inode->i_state & (I_FREEING|I_CLEAR)) {
543 __wait_on_freeing_inode(inode);
548 return node ? inode : NULL;
552 * find_inode_fast is the fast path version of find_inode, see the comment at
553 * iget_locked for details.
555 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
557 struct hlist_node *node;
558 struct inode * inode = NULL;
561 hlist_for_each (node, head) {
562 inode = hlist_entry(node, struct inode, i_hash);
563 if (inode->i_ino != ino)
565 if (inode->i_sb != sb)
567 if (inode->i_state & (I_FREEING|I_CLEAR)) {
568 __wait_on_freeing_inode(inode);
573 return node ? inode : NULL;
577 * new_inode - obtain an inode
580 * Allocates a new inode for given superblock.
582 struct inode *new_inode(struct super_block *sb)
584 static unsigned long last_ino;
585 struct inode * inode;
587 spin_lock_prefetch(&inode_lock);
589 inode = alloc_inode(sb);
591 spin_lock(&inode_lock);
592 inodes_stat.nr_inodes++;
593 list_add(&inode->i_list, &inode_in_use);
594 inode->i_ino = ++last_ino;
596 spin_unlock(&inode_lock);
601 EXPORT_SYMBOL(new_inode);
603 void unlock_new_inode(struct inode *inode)
606 * This is special! We do not need the spinlock
607 * when clearing I_LOCK, because we're guaranteed
608 * that nobody else tries to do anything about the
609 * state of the inode when it is locked, as we
610 * just created it (so there can be no old holders
611 * that haven't tested I_LOCK).
613 inode->i_state &= ~(I_LOCK|I_NEW);
614 wake_up_inode(inode);
617 EXPORT_SYMBOL(unlock_new_inode);
620 * This is called without the inode lock held.. Be careful.
622 * We no longer cache the sb_flags in i_flags - see fs.h
623 * -- rmk@arm.uk.linux.org
625 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)
627 struct inode * inode;
629 inode = alloc_inode(sb);
633 spin_lock(&inode_lock);
634 /* We released the lock, so.. */
635 old = find_inode(sb, head, test, data);
637 if (set(inode, data))
640 inodes_stat.nr_inodes++;
641 list_add(&inode->i_list, &inode_in_use);
642 hlist_add_head(&inode->i_hash, head);
643 inode->i_state = I_LOCK|I_NEW;
644 spin_unlock(&inode_lock);
646 /* Return the locked inode with I_NEW set, the
647 * caller is responsible for filling in the contents
653 * Uhhuh, somebody else created the same inode under
654 * us. Use the old inode instead of the one we just
658 spin_unlock(&inode_lock);
659 destroy_inode(inode);
661 wait_on_inode(inode);
666 spin_unlock(&inode_lock);
667 destroy_inode(inode);
672 * get_new_inode_fast is the fast path version of get_new_inode, see the
673 * comment at iget_locked for details.
675 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
677 struct inode * inode;
679 inode = alloc_inode(sb);
683 spin_lock(&inode_lock);
684 /* We released the lock, so.. */
685 old = find_inode_fast(sb, head, ino);
688 inodes_stat.nr_inodes++;
689 list_add(&inode->i_list, &inode_in_use);
690 hlist_add_head(&inode->i_hash, head);
691 inode->i_state = I_LOCK|I_NEW;
692 spin_unlock(&inode_lock);
694 /* Return the locked inode with I_NEW set, the
695 * caller is responsible for filling in the contents
701 * Uhhuh, somebody else created the same inode under
702 * us. Use the old inode instead of the one we just
706 spin_unlock(&inode_lock);
707 destroy_inode(inode);
709 wait_on_inode(inode);
714 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
718 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
720 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
721 return tmp & I_HASHMASK;
725 * iunique - get a unique inode number
727 * @max_reserved: highest reserved inode number
729 * Obtain an inode number that is unique on the system for a given
730 * superblock. This is used by file systems that have no natural
731 * permanent inode numbering system. An inode number is returned that
732 * is higher than the reserved limit but unique.
735 * With a large number of inodes live on the file system this function
736 * currently becomes quite slow.
738 ino_t iunique(struct super_block *sb, ino_t max_reserved)
740 static ino_t counter;
742 struct hlist_head * head;
744 spin_lock(&inode_lock);
746 if (counter > max_reserved) {
747 head = inode_hashtable + hash(sb,counter);
749 inode = find_inode_fast(sb, head, res);
751 spin_unlock(&inode_lock);
755 counter = max_reserved + 1;
761 EXPORT_SYMBOL(iunique);
763 struct inode *igrab(struct inode *inode)
765 spin_lock(&inode_lock);
766 if (!(inode->i_state & I_FREEING))
770 * Handle the case where s_op->clear_inode is not been
771 * called yet, and somebody is calling igrab
772 * while the inode is getting freed.
775 spin_unlock(&inode_lock);
779 EXPORT_SYMBOL(igrab);
782 * ifind - internal function, you want ilookup5() or iget5().
783 * @sb: super block of file system to search
784 * @head: the head of the list to search
785 * @test: callback used for comparisons between inodes
786 * @data: opaque data pointer to pass to @test
788 * ifind() searches for the inode specified by @data in the inode
789 * cache. This is a generalized version of ifind_fast() for file systems where
790 * the inode number is not sufficient for unique identification of an inode.
792 * If the inode is in the cache, the inode is returned with an incremented
795 * Otherwise NULL is returned.
797 * Note, @test is called with the inode_lock held, so can't sleep.
799 static inline struct inode *ifind(struct super_block *sb,
800 struct hlist_head *head, int (*test)(struct inode *, void *),
805 spin_lock(&inode_lock);
806 inode = find_inode(sb, head, test, data);
809 spin_unlock(&inode_lock);
810 wait_on_inode(inode);
813 spin_unlock(&inode_lock);
818 * ifind_fast - internal function, you want ilookup() or iget().
819 * @sb: super block of file system to search
820 * @head: head of the list to search
821 * @ino: inode number to search for
823 * ifind_fast() searches for the inode @ino in the inode cache. This is for
824 * file systems where the inode number is sufficient for unique identification
827 * If the inode is in the cache, the inode is returned with an incremented
830 * Otherwise NULL is returned.
832 static inline struct inode *ifind_fast(struct super_block *sb,
833 struct hlist_head *head, unsigned long ino)
837 spin_lock(&inode_lock);
838 inode = find_inode_fast(sb, head, ino);
841 spin_unlock(&inode_lock);
842 wait_on_inode(inode);
845 spin_unlock(&inode_lock);
850 * ilookup5 - search for an inode in the inode cache
851 * @sb: super block of file system to search
852 * @hashval: hash value (usually inode number) to search for
853 * @test: callback used for comparisons between inodes
854 * @data: opaque data pointer to pass to @test
856 * ilookup5() uses ifind() to search for the inode specified by @hashval and
857 * @data in the inode cache. This is a generalized version of ilookup() for
858 * file systems where the inode number is not sufficient for unique
859 * identification of an inode.
861 * If the inode is in the cache, the inode is returned with an incremented
864 * Otherwise NULL is returned.
866 * Note, @test is called with the inode_lock held, so can't sleep.
868 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
869 int (*test)(struct inode *, void *), void *data)
871 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
873 return ifind(sb, head, test, data);
876 EXPORT_SYMBOL(ilookup5);
879 * ilookup - search for an inode in the inode cache
880 * @sb: super block of file system to search
881 * @ino: inode number to search for
883 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
884 * This is for file systems where the inode number is sufficient for unique
885 * identification of an inode.
887 * If the inode is in the cache, the inode is returned with an incremented
890 * Otherwise NULL is returned.
892 struct inode *ilookup(struct super_block *sb, unsigned long ino)
894 struct hlist_head *head = inode_hashtable + hash(sb, ino);
896 return ifind_fast(sb, head, ino);
899 EXPORT_SYMBOL(ilookup);
902 * iget5_locked - obtain an inode from a mounted file system
903 * @sb: super block of file system
904 * @hashval: hash value (usually inode number) to get
905 * @test: callback used for comparisons between inodes
906 * @set: callback used to initialize a new struct inode
907 * @data: opaque data pointer to pass to @test and @set
909 * This is iget() without the read_inode() portion of get_new_inode().
911 * iget5_locked() uses ifind() to search for the inode specified by @hashval
912 * and @data in the inode cache and if present it is returned with an increased
913 * reference count. This is a generalized version of iget_locked() for file
914 * systems where the inode number is not sufficient for unique identification
917 * If the inode is not in cache, get_new_inode() is called to allocate a new
918 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
919 * file system gets to fill it in before unlocking it via unlock_new_inode().
921 * Note both @test and @set are called with the inode_lock held, so can't sleep.
923 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
924 int (*test)(struct inode *, void *),
925 int (*set)(struct inode *, void *), void *data)
927 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
930 inode = ifind(sb, head, test, data);
934 * get_new_inode() will do the right thing, re-trying the search
935 * in case it had to block at any point.
937 return get_new_inode(sb, head, test, set, data);
940 EXPORT_SYMBOL(iget5_locked);
943 * iget_locked - obtain an inode from a mounted file system
944 * @sb: super block of file system
945 * @ino: inode number to get
947 * This is iget() without the read_inode() portion of get_new_inode_fast().
949 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
950 * the inode cache and if present it is returned with an increased reference
951 * count. This is for file systems where the inode number is sufficient for
952 * unique identification of an inode.
954 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
955 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
956 * The file system gets to fill it in before unlocking it via
957 * unlock_new_inode().
959 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
961 struct hlist_head *head = inode_hashtable + hash(sb, ino);
964 inode = ifind_fast(sb, head, ino);
968 * get_new_inode_fast() will do the right thing, re-trying the search
969 * in case it had to block at any point.
971 return get_new_inode_fast(sb, head, ino);
974 EXPORT_SYMBOL(iget_locked);
977 * __insert_inode_hash - hash an inode
978 * @inode: unhashed inode
979 * @hashval: unsigned long value used to locate this object in the
982 * Add an inode to the inode hash for this superblock.
984 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
986 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
987 spin_lock(&inode_lock);
988 hlist_add_head(&inode->i_hash, head);
989 spin_unlock(&inode_lock);
992 EXPORT_SYMBOL(__insert_inode_hash);
995 * remove_inode_hash - remove an inode from the hash
996 * @inode: inode to unhash
998 * Remove an inode from the superblock.
1000 void remove_inode_hash(struct inode *inode)
1002 spin_lock(&inode_lock);
1003 hlist_del_init(&inode->i_hash);
1004 spin_unlock(&inode_lock);
1007 EXPORT_SYMBOL(remove_inode_hash);
1010 * Tell the filesystem that this inode is no longer of any interest and should
1011 * be completely destroyed.
1013 * We leave the inode in the inode hash table until *after* the filesystem's
1014 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1015 * instigate) will always find up-to-date information either in the hash or on
1018 * I_FREEING is set so that no-one will take a new reference to the inode while
1019 * it is being deleted.
1021 void generic_delete_inode(struct inode *inode)
1023 struct super_operations *op = inode->i_sb->s_op;
1025 list_del_init(&inode->i_list);
1026 inode->i_state|=I_FREEING;
1027 inodes_stat.nr_inodes--;
1028 spin_unlock(&inode_lock);
1030 if (inode->i_data.nrpages)
1031 truncate_inode_pages(&inode->i_data, 0);
1033 security_inode_delete(inode);
1035 if (op->delete_inode) {
1036 void (*delete)(struct inode *) = op->delete_inode;
1037 if (!is_bad_inode(inode))
1039 /* s_op->delete_inode internally recalls clear_inode() */
1043 spin_lock(&inode_lock);
1044 hlist_del_init(&inode->i_hash);
1045 spin_unlock(&inode_lock);
1046 wake_up_inode(inode);
1047 if (inode->i_state != I_CLEAR)
1049 destroy_inode(inode);
1052 EXPORT_SYMBOL(generic_delete_inode);
1054 static void generic_forget_inode(struct inode *inode)
1056 struct super_block *sb = inode->i_sb;
1058 if (!hlist_unhashed(&inode->i_hash)) {
1059 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1060 list_move(&inode->i_list, &inode_unused);
1061 inodes_stat.nr_unused++;
1062 spin_unlock(&inode_lock);
1063 if (!sb || (sb->s_flags & MS_ACTIVE))
1065 write_inode_now(inode, 1);
1066 spin_lock(&inode_lock);
1067 inodes_stat.nr_unused--;
1068 hlist_del_init(&inode->i_hash);
1070 list_del_init(&inode->i_list);
1071 inode->i_state|=I_FREEING;
1072 inodes_stat.nr_inodes--;
1073 spin_unlock(&inode_lock);
1074 if (inode->i_data.nrpages)
1075 truncate_inode_pages(&inode->i_data, 0);
1077 destroy_inode(inode);
1081 * Normal UNIX filesystem behaviour: delete the
1082 * inode when the usage count drops to zero, and
1085 static void generic_drop_inode(struct inode *inode)
1087 if (!inode->i_nlink)
1088 generic_delete_inode(inode);
1090 generic_forget_inode(inode);
1094 * Called when we're dropping the last reference
1097 * Call the FS "drop()" function, defaulting to
1098 * the legacy UNIX filesystem behaviour..
1100 * NOTE! NOTE! NOTE! We're called with the inode lock
1101 * held, and the drop function is supposed to release
1104 static inline void iput_final(struct inode *inode)
1106 struct super_operations *op = inode->i_sb->s_op;
1107 void (*drop)(struct inode *) = generic_drop_inode;
1109 if (op && op->drop_inode)
1110 drop = op->drop_inode;
1115 * iput - put an inode
1116 * @inode: inode to put
1118 * Puts an inode, dropping its usage count. If the inode use count hits
1119 * zero the inode is also then freed and may be destroyed.
1121 void iput(struct inode *inode)
1124 struct super_operations *op = inode->i_sb->s_op;
1126 if (inode->i_state == I_CLEAR)
1129 if (op && op->put_inode)
1130 op->put_inode(inode);
1132 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1137 EXPORT_SYMBOL(iput);
1140 * bmap - find a block number in a file
1141 * @inode: inode of file
1142 * @block: block to find
1144 * Returns the block number on the device holding the inode that
1145 * is the disk block number for the block of the file requested.
1146 * That is, asked for block 4 of inode 1 the function will return the
1147 * disk block relative to the disk start that holds that block of the
1150 sector_t bmap(struct inode * inode, sector_t block)
1153 if (inode->i_mapping->a_ops->bmap)
1154 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1158 EXPORT_SYMBOL(bmap);
1161 * Return true if the filesystem which backs this inode considers the two
1162 * passed timespecs to be sufficiently different to warrant flushing the
1163 * altered time out to disk.
1165 static int inode_times_differ(struct inode *inode,
1166 struct timespec *old, struct timespec *new)
1168 if (IS_ONE_SECOND(inode))
1169 return old->tv_sec != new->tv_sec;
1170 return !timespec_equal(old, new);
1174 * update_atime - update the access time
1175 * @inode: inode accessed
1177 * Update the accessed time on an inode and mark it for writeback.
1178 * This function automatically handles read only file systems and media,
1179 * as well as the "noatime" flag and inode specific "noatime" markers.
1181 void update_atime(struct inode *inode)
1183 struct timespec now;
1185 if (IS_NOATIME(inode))
1187 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1189 if (IS_RDONLY(inode))
1192 now = current_kernel_time();
1193 if (inode_times_differ(inode, &inode->i_atime, &now)) {
1194 inode->i_atime = now;
1195 mark_inode_dirty_sync(inode);
1197 if (!timespec_equal(&inode->i_atime, &now))
1198 inode->i_atime = now;
1202 EXPORT_SYMBOL(update_atime);
1205 * inode_update_time - update mtime and ctime time
1206 * @inode: inode accessed
1207 * @ctime_too: update ctime too
1209 * Update the mtime time on an inode and mark it for writeback.
1210 * When ctime_too is specified update the ctime too.
1213 void inode_update_time(struct inode *inode, int ctime_too)
1215 struct timespec now;
1218 if (IS_NOCMTIME(inode))
1220 if (IS_RDONLY(inode))
1223 now = current_kernel_time();
1225 if (inode_times_differ(inode, &inode->i_mtime, &now))
1227 inode->i_mtime = now;
1230 if (inode_times_differ(inode, &inode->i_ctime, &now))
1232 inode->i_ctime = now;
1235 mark_inode_dirty_sync(inode);
1238 EXPORT_SYMBOL(inode_update_time);
1240 int inode_needs_sync(struct inode *inode)
1244 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1249 EXPORT_SYMBOL(inode_needs_sync);
1252 * Quota functions that want to walk the inode lists..
1256 /* Function back in dquot.c */
1257 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1259 void remove_dquot_ref(struct super_block *sb, int type, struct list_head *tofree_head)
1261 struct inode *inode;
1262 struct list_head *act_head;
1265 return; /* nothing to do */
1266 spin_lock(&inode_lock); /* This lock is for inodes code */
1268 /* We hold dqptr_sem so we are safe against the quota code */
1269 list_for_each(act_head, &inode_in_use) {
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, &inode_unused) {
1275 inode = list_entry(act_head, struct inode, i_list);
1276 if (inode->i_sb == sb && !IS_NOQUOTA(inode))
1277 remove_inode_dquot_ref(inode, type, tofree_head);
1279 list_for_each(act_head, &sb->s_dirty) {
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 list_for_each(act_head, &sb->s_io) {
1285 inode = list_entry(act_head, struct inode, i_list);
1286 if (!IS_NOQUOTA(inode))
1287 remove_inode_dquot_ref(inode, type, tofree_head);
1289 spin_unlock(&inode_lock);
1295 * Hashed waitqueues for wait_on_inode(). The table is pretty small - the
1296 * kernel doesn't lock many inodes at the same time.
1298 #define I_WAIT_TABLE_ORDER 3
1299 static struct i_wait_queue_head {
1300 wait_queue_head_t wqh;
1301 } ____cacheline_aligned_in_smp i_wait_queue_heads[1<<I_WAIT_TABLE_ORDER];
1304 * Return the address of the waitqueue_head to be used for this inode
1306 static wait_queue_head_t *i_waitq_head(struct inode *inode)
1308 return &i_wait_queue_heads[hash_ptr(inode, I_WAIT_TABLE_ORDER)].wqh;
1311 void __wait_on_inode(struct inode *inode)
1313 DECLARE_WAITQUEUE(wait, current);
1314 wait_queue_head_t *wq = i_waitq_head(inode);
1316 add_wait_queue(wq, &wait);
1318 set_current_state(TASK_UNINTERRUPTIBLE);
1319 if (inode->i_state & I_LOCK) {
1323 remove_wait_queue(wq, &wait);
1324 __set_current_state(TASK_RUNNING);
1328 * If we try to find an inode in the inode hash while it is being deleted, we
1329 * have to wait until the filesystem completes its deletion before reporting
1330 * that it isn't found. This is because iget will immediately call
1331 * ->read_inode, and we want to be sure that evidence of the deletion is found
1334 * This call might return early if an inode which shares the waitq is woken up.
1335 * This is most easily handled by the caller which will loop around again
1336 * looking for the inode.
1338 * This is called with inode_lock held.
1340 static void __wait_on_freeing_inode(struct inode *inode)
1342 DECLARE_WAITQUEUE(wait, current);
1343 wait_queue_head_t *wq = i_waitq_head(inode);
1345 add_wait_queue(wq, &wait);
1346 set_current_state(TASK_UNINTERRUPTIBLE);
1347 spin_unlock(&inode_lock);
1349 remove_wait_queue(wq, &wait);
1350 spin_lock(&inode_lock);
1353 void wake_up_inode(struct inode *inode)
1355 wait_queue_head_t *wq = i_waitq_head(inode);
1358 * Prevent speculative execution through spin_unlock(&inode_lock);
1361 if (waitqueue_active(wq))
1365 static __initdata unsigned long ihash_entries;
1366 static int __init set_ihash_entries(char *str)
1370 ihash_entries = simple_strtoul(str, &str, 0);
1373 __setup("ihash_entries=", set_ihash_entries);
1376 * Initialize the waitqueues and inode hash table.
1378 void __init inode_init_early(void)
1383 alloc_large_system_hash("Inode-cache",
1384 sizeof(struct hlist_head),
1391 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1392 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1395 void __init inode_init(unsigned long mempages)
1399 for (i = 0; i < ARRAY_SIZE(i_wait_queue_heads); i++)
1400 init_waitqueue_head(&i_wait_queue_heads[i].wqh);
1402 /* inode slab cache */
1403 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1404 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, init_once,
1406 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1409 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1411 inode->i_mode = mode;
1412 if (S_ISCHR(mode)) {
1413 inode->i_fop = &def_chr_fops;
1414 inode->i_rdev = rdev;
1415 } else if (S_ISBLK(mode)) {
1416 inode->i_fop = &def_blk_fops;
1417 inode->i_rdev = rdev;
1418 } else if (S_ISFIFO(mode))
1419 inode->i_fop = &def_fifo_fops;
1420 else if (S_ISSOCK(mode))
1421 inode->i_fop = &bad_sock_fops;
1423 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1426 EXPORT_SYMBOL(init_special_inode);