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>
25 * This is needed for the following functions:
27 * - invalidate_inode_buffers
31 * FIXME: remove all knowledge of the buffer layer from this file
33 #include <linux/buffer_head.h>
36 * New inode.c implementation.
38 * This implementation has the basic premise of trying
39 * to be extremely low-overhead and SMP-safe, yet be
40 * simple enough to be "obviously correct".
45 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
47 /* #define INODE_PARANOIA 1 */
48 /* #define INODE_DEBUG 1 */
51 * Inode lookup is no longer as critical as it used to be:
52 * most of the lookups are going to be through the dcache.
54 #define I_HASHBITS i_hash_shift
55 #define I_HASHMASK i_hash_mask
57 static unsigned int i_hash_mask;
58 static unsigned int i_hash_shift;
61 * Each inode can be on two separate lists. One is
62 * the hash list of the inode, used for lookups. The
63 * other linked list is the "type" list:
64 * "in_use" - valid inode, i_count > 0, i_nlink > 0
65 * "dirty" - as "in_use" but also dirty
66 * "unused" - valid inode, i_count = 0
68 * A "dirty" list is maintained for each super block,
69 * allowing for low-overhead inode sync() operations.
72 LIST_HEAD(inode_in_use);
73 LIST_HEAD(inode_unused);
74 static struct hlist_head *inode_hashtable;
77 * A simple spinlock to protect the list manipulations.
79 * NOTE! You also have to own the lock if you change
80 * the i_state of an inode while it is in use..
82 spinlock_t inode_lock = SPIN_LOCK_UNLOCKED;
85 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
86 * icache shrinking path, and the umount path. Without this exclusion,
87 * by the time prune_icache calls iput for the inode whose pages it has
88 * been invalidating, or by the time it calls clear_inode & destroy_inode
89 * from its final dispose_list, the struct super_block they refer to
90 * (for inode->i_sb->s_op) may already have been freed and reused.
92 static DECLARE_MUTEX(iprune_sem);
95 * Statistics gathering..
97 struct inodes_stat_t inodes_stat;
99 static kmem_cache_t * inode_cachep;
101 static struct inode *alloc_inode(struct super_block *sb)
103 static struct address_space_operations empty_aops;
104 static struct inode_operations empty_iops;
105 static struct file_operations empty_fops;
108 if (sb->s_op->alloc_inode)
109 inode = sb->s_op->alloc_inode(sb);
111 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
114 struct address_space * const mapping = &inode->i_data;
117 inode->i_blkbits = sb->s_blocksize_bits;
119 atomic_set(&inode->i_count, 1);
121 inode->i_op = &empty_iops;
122 inode->i_fop = &empty_fops;
124 atomic_set(&inode->i_writecount, 0);
128 inode->i_generation = 0;
130 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
132 inode->i_pipe = NULL;
133 inode->i_bdev = NULL;
134 inode->i_cdev = NULL;
136 inode->i_security = NULL;
137 inode->dirtied_when = 0;
138 if (security_inode_alloc(inode)) {
139 if (inode->i_sb->s_op->destroy_inode)
140 inode->i_sb->s_op->destroy_inode(inode);
142 kmem_cache_free(inode_cachep, (inode));
146 mapping->a_ops = &empty_aops;
147 mapping->host = inode;
149 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
150 mapping->assoc_mapping = NULL;
151 mapping->backing_dev_info = &default_backing_dev_info;
154 * If the block_device provides a backing_dev_info for client
155 * inodes then use that. Otherwise the inode share the bdev's
159 struct backing_dev_info *bdi;
161 bdi = sb->s_bdev->bd_inode_backing_dev_info;
163 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
164 mapping->backing_dev_info = bdi;
166 memset(&inode->u, 0, sizeof(inode->u));
167 inode->i_mapping = mapping;
172 void destroy_inode(struct inode *inode)
174 if (inode_has_buffers(inode))
176 security_inode_free(inode);
177 if (inode->i_sb->s_op->destroy_inode)
178 inode->i_sb->s_op->destroy_inode(inode);
180 kmem_cache_free(inode_cachep, (inode));
185 * These are initializations that only need to be done
186 * once, because the fields are idempotent across use
187 * of the inode, so let the slab aware of that.
189 void inode_init_once(struct inode *inode)
191 memset(inode, 0, sizeof(*inode));
192 INIT_HLIST_NODE(&inode->i_hash);
193 INIT_LIST_HEAD(&inode->i_dentry);
194 INIT_LIST_HEAD(&inode->i_devices);
195 sema_init(&inode->i_sem, 1);
196 init_rwsem(&inode->i_alloc_sem);
197 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
198 spin_lock_init(&inode->i_data.tree_lock);
199 spin_lock_init(&inode->i_data.i_mmap_lock);
200 atomic_set(&inode->i_data.truncate_count, 0);
201 INIT_LIST_HEAD(&inode->i_data.private_list);
202 spin_lock_init(&inode->i_data.private_lock);
203 INIT_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
204 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
205 spin_lock_init(&inode->i_lock);
206 i_size_ordered_init(inode);
209 EXPORT_SYMBOL(inode_init_once);
211 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
213 struct inode * inode = (struct inode *) foo;
215 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
216 SLAB_CTOR_CONSTRUCTOR)
217 inode_init_once(inode);
221 * inode_lock must be held
223 void __iget(struct inode * inode)
225 if (atomic_read(&inode->i_count)) {
226 atomic_inc(&inode->i_count);
229 atomic_inc(&inode->i_count);
230 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
231 list_move(&inode->i_list, &inode_in_use);
232 inodes_stat.nr_unused--;
236 * clear_inode - clear an inode
237 * @inode: inode to clear
239 * This is called by the filesystem to tell us
240 * that the inode is no longer useful. We just
241 * terminate it with extreme prejudice.
243 void clear_inode(struct inode *inode)
245 invalidate_inode_buffers(inode);
247 if (inode->i_data.nrpages)
249 if (!(inode->i_state & I_FREEING))
251 if (inode->i_state & I_CLEAR)
253 wait_on_inode(inode);
255 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
256 inode->i_sb->s_op->clear_inode(inode);
261 inode->i_state = I_CLEAR;
264 EXPORT_SYMBOL(clear_inode);
267 * dispose_list - dispose of the contents of a local list
268 * @head: the head of the list to free
270 * Dispose-list gets a local list with local inodes in it, so it doesn't
271 * need to worry about list corruption and SMP locks.
273 static void dispose_list(struct list_head *head)
277 while (!list_empty(head)) {
280 inode = list_entry(head->next, struct inode, i_list);
281 list_del(&inode->i_list);
283 if (inode->i_data.nrpages)
284 truncate_inode_pages(&inode->i_data, 0);
286 destroy_inode(inode);
289 spin_lock(&inode_lock);
290 inodes_stat.nr_inodes -= nr_disposed;
291 spin_unlock(&inode_lock);
295 * Invalidate all inodes for a device.
297 static int invalidate_list(struct list_head *head, struct super_block * sb, struct list_head * dispose)
299 struct list_head *next;
300 int busy = 0, count = 0;
304 struct list_head * tmp = next;
305 struct inode * inode;
310 inode = list_entry(tmp, struct inode, i_list);
311 if (inode->i_sb != sb)
313 invalidate_inode_buffers(inode);
314 if (!atomic_read(&inode->i_count)) {
315 hlist_del_init(&inode->i_hash);
316 list_move(&inode->i_list, dispose);
317 inode->i_state |= I_FREEING;
323 /* only unused inodes may be cached with i_count zero */
324 inodes_stat.nr_unused -= count;
329 * This is a two-stage process. First we collect all
330 * offending inodes onto the throw-away list, and in
331 * the second stage we actually dispose of them. This
332 * is because we don't want to sleep while messing
333 * with the global lists..
337 * invalidate_inodes - discard the inodes on a device
340 * Discard all of the inodes for a given superblock. If the discard
341 * fails because there are busy inodes then a non zero value is returned.
342 * If the discard is successful all the inodes have been discarded.
344 int invalidate_inodes(struct super_block * sb)
347 LIST_HEAD(throw_away);
350 spin_lock(&inode_lock);
351 busy = invalidate_list(&inode_in_use, sb, &throw_away);
352 busy |= invalidate_list(&inode_unused, sb, &throw_away);
353 busy |= invalidate_list(&sb->s_dirty, sb, &throw_away);
354 busy |= invalidate_list(&sb->s_io, sb, &throw_away);
355 spin_unlock(&inode_lock);
357 dispose_list(&throw_away);
363 EXPORT_SYMBOL(invalidate_inodes);
365 int __invalidate_device(struct block_device *bdev, int do_sync)
367 struct super_block *sb;
374 sb = get_super(bdev);
377 * no need to lock the super, get_super holds the
378 * read semaphore so the filesystem cannot go away
379 * under us (->put_super runs with the write lock
382 shrink_dcache_sb(sb);
383 res = invalidate_inodes(sb);
386 invalidate_bdev(bdev, 0);
390 EXPORT_SYMBOL(__invalidate_device);
392 static int can_unuse(struct inode *inode)
396 if (inode_has_buffers(inode))
398 if (atomic_read(&inode->i_count))
400 if (inode->i_data.nrpages)
406 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
407 * a temporary list and then are freed outside inode_lock by dispose_list().
409 * Any inodes which are pinned purely because of attached pagecache have their
410 * pagecache removed. We expect the final iput() on that inode to add it to
411 * the front of the inode_unused list. So look for it there and if the
412 * inode is still freeable, proceed. The right inode is found 99.9% of the
413 * time in testing on a 4-way.
415 * If the inode has metadata buffers attached to mapping->private_list then
416 * try to remove them.
418 static void prune_icache(int nr_to_scan)
423 unsigned long reap = 0;
426 spin_lock(&inode_lock);
427 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
430 if (list_empty(&inode_unused))
433 inode = list_entry(inode_unused.prev, struct inode, i_list);
435 if (inode->i_state || atomic_read(&inode->i_count)) {
436 list_move(&inode->i_list, &inode_unused);
439 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
441 spin_unlock(&inode_lock);
442 if (remove_inode_buffers(inode))
443 reap += invalidate_inode_pages(&inode->i_data);
445 spin_lock(&inode_lock);
447 if (inode != list_entry(inode_unused.next,
448 struct inode, i_list))
449 continue; /* wrong inode or list_empty */
450 if (!can_unuse(inode))
453 hlist_del_init(&inode->i_hash);
454 list_move(&inode->i_list, &freeable);
455 inode->i_state |= I_FREEING;
458 inodes_stat.nr_unused -= nr_pruned;
459 spin_unlock(&inode_lock);
461 dispose_list(&freeable);
464 if (current_is_kswapd())
465 mod_page_state(kswapd_inodesteal, reap);
467 mod_page_state(pginodesteal, reap);
471 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
472 * "unused" means that no dentries are referring to the inodes: the files are
473 * not open and the dcache references to those inodes have already been
476 * This function is passed the number of inodes to scan, and it returns the
477 * total number of remaining possibly-reclaimable inodes.
479 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
483 * Nasty deadlock avoidance. We may hold various FS locks,
484 * and we don't want to recurse into the FS that called us
485 * in clear_inode() and friends..
487 if (gfp_mask & __GFP_FS)
490 return inodes_stat.nr_unused;
493 static void __wait_on_freeing_inode(struct inode *inode);
495 * Called with the inode lock held.
496 * NOTE: we are not increasing the inode-refcount, you must call __iget()
497 * by hand after calling find_inode now! This simplifies iunique and won't
498 * add any additional branch in the common code.
500 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
502 struct hlist_node *node;
503 struct inode * inode = NULL;
506 hlist_for_each (node, head) {
507 inode = hlist_entry(node, struct inode, i_hash);
508 if (inode->i_sb != sb)
510 if (!test(inode, data))
512 if (inode->i_state & (I_FREEING|I_CLEAR)) {
513 __wait_on_freeing_inode(inode);
518 return node ? inode : NULL;
522 * find_inode_fast is the fast path version of find_inode, see the comment at
523 * iget_locked for details.
525 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
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_ino != ino)
535 if (inode->i_sb != sb)
537 if (inode->i_state & (I_FREEING|I_CLEAR)) {
538 __wait_on_freeing_inode(inode);
543 return node ? inode : NULL;
547 * new_inode - obtain an inode
550 * Allocates a new inode for given superblock.
552 struct inode *new_inode(struct super_block *sb)
554 static unsigned long last_ino;
555 struct inode * inode;
557 spin_lock_prefetch(&inode_lock);
559 inode = alloc_inode(sb);
561 spin_lock(&inode_lock);
562 inodes_stat.nr_inodes++;
563 list_add(&inode->i_list, &inode_in_use);
564 inode->i_ino = ++last_ino;
566 spin_unlock(&inode_lock);
571 EXPORT_SYMBOL(new_inode);
573 void unlock_new_inode(struct inode *inode)
576 * This is special! We do not need the spinlock
577 * when clearing I_LOCK, because we're guaranteed
578 * that nobody else tries to do anything about the
579 * state of the inode when it is locked, as we
580 * just created it (so there can be no old holders
581 * that haven't tested I_LOCK).
583 inode->i_state &= ~(I_LOCK|I_NEW);
584 wake_up_inode(inode);
587 EXPORT_SYMBOL(unlock_new_inode);
590 * This is called without the inode lock held.. Be careful.
592 * We no longer cache the sb_flags in i_flags - see fs.h
593 * -- rmk@arm.uk.linux.org
595 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)
597 struct inode * inode;
599 inode = alloc_inode(sb);
603 spin_lock(&inode_lock);
604 /* We released the lock, so.. */
605 old = find_inode(sb, head, test, data);
607 if (set(inode, data))
610 inodes_stat.nr_inodes++;
611 list_add(&inode->i_list, &inode_in_use);
612 hlist_add_head(&inode->i_hash, head);
613 inode->i_state = I_LOCK|I_NEW;
614 spin_unlock(&inode_lock);
616 /* Return the locked inode with I_NEW set, the
617 * caller is responsible for filling in the contents
623 * Uhhuh, somebody else created the same inode under
624 * us. Use the old inode instead of the one we just
628 spin_unlock(&inode_lock);
629 destroy_inode(inode);
631 wait_on_inode(inode);
636 spin_unlock(&inode_lock);
637 destroy_inode(inode);
642 * get_new_inode_fast is the fast path version of get_new_inode, see the
643 * comment at iget_locked for details.
645 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
647 struct inode * inode;
649 inode = alloc_inode(sb);
653 spin_lock(&inode_lock);
654 /* We released the lock, so.. */
655 old = find_inode_fast(sb, head, ino);
658 inodes_stat.nr_inodes++;
659 list_add(&inode->i_list, &inode_in_use);
660 hlist_add_head(&inode->i_hash, head);
661 inode->i_state = I_LOCK|I_NEW;
662 spin_unlock(&inode_lock);
664 /* Return the locked inode with I_NEW set, the
665 * caller is responsible for filling in the contents
671 * Uhhuh, somebody else created the same inode under
672 * us. Use the old inode instead of the one we just
676 spin_unlock(&inode_lock);
677 destroy_inode(inode);
679 wait_on_inode(inode);
684 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
688 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
690 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
691 return tmp & I_HASHMASK;
695 * iunique - get a unique inode number
697 * @max_reserved: highest reserved inode number
699 * Obtain an inode number that is unique on the system for a given
700 * superblock. This is used by file systems that have no natural
701 * permanent inode numbering system. An inode number is returned that
702 * is higher than the reserved limit but unique.
705 * With a large number of inodes live on the file system this function
706 * currently becomes quite slow.
708 ino_t iunique(struct super_block *sb, ino_t max_reserved)
710 static ino_t counter;
712 struct hlist_head * head;
714 spin_lock(&inode_lock);
716 if (counter > max_reserved) {
717 head = inode_hashtable + hash(sb,counter);
719 inode = find_inode_fast(sb, head, res);
721 spin_unlock(&inode_lock);
725 counter = max_reserved + 1;
731 EXPORT_SYMBOL(iunique);
733 struct inode *igrab(struct inode *inode)
735 spin_lock(&inode_lock);
736 if (!(inode->i_state & I_FREEING))
740 * Handle the case where s_op->clear_inode is not been
741 * called yet, and somebody is calling igrab
742 * while the inode is getting freed.
745 spin_unlock(&inode_lock);
749 EXPORT_SYMBOL(igrab);
752 * ifind - internal function, you want ilookup5() or iget5().
753 * @sb: super block of file system to search
754 * @head: the head of the list to search
755 * @test: callback used for comparisons between inodes
756 * @data: opaque data pointer to pass to @test
758 * ifind() searches for the inode specified by @data in the inode
759 * cache. This is a generalized version of ifind_fast() for file systems where
760 * the inode number is not sufficient for unique identification of an inode.
762 * If the inode is in the cache, the inode is returned with an incremented
765 * Otherwise NULL is returned.
767 * Note, @test is called with the inode_lock held, so can't sleep.
769 static inline struct inode *ifind(struct super_block *sb,
770 struct hlist_head *head, int (*test)(struct inode *, void *),
775 spin_lock(&inode_lock);
776 inode = find_inode(sb, head, test, data);
779 spin_unlock(&inode_lock);
780 wait_on_inode(inode);
783 spin_unlock(&inode_lock);
788 * ifind_fast - internal function, you want ilookup() or iget().
789 * @sb: super block of file system to search
790 * @head: head of the list to search
791 * @ino: inode number to search for
793 * ifind_fast() searches for the inode @ino in the inode cache. This is for
794 * file systems where the inode number is sufficient for unique identification
797 * If the inode is in the cache, the inode is returned with an incremented
800 * Otherwise NULL is returned.
802 static inline struct inode *ifind_fast(struct super_block *sb,
803 struct hlist_head *head, unsigned long ino)
807 spin_lock(&inode_lock);
808 inode = find_inode_fast(sb, head, ino);
811 spin_unlock(&inode_lock);
812 wait_on_inode(inode);
815 spin_unlock(&inode_lock);
820 * ilookup5 - search for an inode in the inode cache
821 * @sb: super block of file system to search
822 * @hashval: hash value (usually inode number) to search for
823 * @test: callback used for comparisons between inodes
824 * @data: opaque data pointer to pass to @test
826 * ilookup5() uses ifind() to search for the inode specified by @hashval and
827 * @data in the inode cache. This is a generalized version of ilookup() for
828 * file systems where the inode number is not sufficient for unique
829 * identification of an inode.
831 * If the inode is in the cache, the inode is returned with an incremented
834 * Otherwise NULL is returned.
836 * Note, @test is called with the inode_lock held, so can't sleep.
838 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
839 int (*test)(struct inode *, void *), void *data)
841 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
843 return ifind(sb, head, test, data);
846 EXPORT_SYMBOL(ilookup5);
849 * ilookup - search for an inode in the inode cache
850 * @sb: super block of file system to search
851 * @ino: inode number to search for
853 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
854 * This is for file systems where the inode number is sufficient for unique
855 * identification of an inode.
857 * If the inode is in the cache, the inode is returned with an incremented
860 * Otherwise NULL is returned.
862 struct inode *ilookup(struct super_block *sb, unsigned long ino)
864 struct hlist_head *head = inode_hashtable + hash(sb, ino);
866 return ifind_fast(sb, head, ino);
869 EXPORT_SYMBOL(ilookup);
872 * iget5_locked - obtain an inode from a mounted file system
873 * @sb: super block of file system
874 * @hashval: hash value (usually inode number) to get
875 * @test: callback used for comparisons between inodes
876 * @set: callback used to initialize a new struct inode
877 * @data: opaque data pointer to pass to @test and @set
879 * This is iget() without the read_inode() portion of get_new_inode().
881 * iget5_locked() uses ifind() to search for the inode specified by @hashval
882 * and @data in the inode cache and if present it is returned with an increased
883 * reference count. This is a generalized version of iget_locked() for file
884 * systems where the inode number is not sufficient for unique identification
887 * If the inode is not in cache, get_new_inode() is called to allocate a new
888 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
889 * file system gets to fill it in before unlocking it via unlock_new_inode().
891 * Note both @test and @set are called with the inode_lock held, so can't sleep.
893 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
894 int (*test)(struct inode *, void *),
895 int (*set)(struct inode *, void *), void *data)
897 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
900 inode = ifind(sb, head, test, data);
904 * get_new_inode() will do the right thing, re-trying the search
905 * in case it had to block at any point.
907 return get_new_inode(sb, head, test, set, data);
910 EXPORT_SYMBOL(iget5_locked);
913 * iget_locked - obtain an inode from a mounted file system
914 * @sb: super block of file system
915 * @ino: inode number to get
917 * This is iget() without the read_inode() portion of get_new_inode_fast().
919 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
920 * the inode cache and if present it is returned with an increased reference
921 * count. This is for file systems where the inode number is sufficient for
922 * unique identification of an inode.
924 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
925 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
926 * The file system gets to fill it in before unlocking it via
927 * unlock_new_inode().
929 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
931 struct hlist_head *head = inode_hashtable + hash(sb, ino);
934 inode = ifind_fast(sb, head, ino);
938 * get_new_inode_fast() will do the right thing, re-trying the search
939 * in case it had to block at any point.
941 return get_new_inode_fast(sb, head, ino);
944 EXPORT_SYMBOL(iget_locked);
947 * __insert_inode_hash - hash an inode
948 * @inode: unhashed inode
949 * @hashval: unsigned long value used to locate this object in the
952 * Add an inode to the inode hash for this superblock.
954 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
956 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
957 spin_lock(&inode_lock);
958 hlist_add_head(&inode->i_hash, head);
959 spin_unlock(&inode_lock);
962 EXPORT_SYMBOL(__insert_inode_hash);
965 * remove_inode_hash - remove an inode from the hash
966 * @inode: inode to unhash
968 * Remove an inode from the superblock.
970 void remove_inode_hash(struct inode *inode)
972 spin_lock(&inode_lock);
973 hlist_del_init(&inode->i_hash);
974 spin_unlock(&inode_lock);
977 EXPORT_SYMBOL(remove_inode_hash);
980 * Tell the filesystem that this inode is no longer of any interest and should
981 * be completely destroyed.
983 * We leave the inode in the inode hash table until *after* the filesystem's
984 * ->delete_inode completes. This ensures that an iget (such as nfsd might
985 * instigate) will always find up-to-date information either in the hash or on
988 * I_FREEING is set so that no-one will take a new reference to the inode while
989 * it is being deleted.
991 void generic_delete_inode(struct inode *inode)
993 struct super_operations *op = inode->i_sb->s_op;
995 list_del_init(&inode->i_list);
996 inode->i_state|=I_FREEING;
997 inodes_stat.nr_inodes--;
998 spin_unlock(&inode_lock);
1000 if (inode->i_data.nrpages)
1001 truncate_inode_pages(&inode->i_data, 0);
1003 security_inode_delete(inode);
1005 if (op->delete_inode) {
1006 void (*delete)(struct inode *) = op->delete_inode;
1007 if (!is_bad_inode(inode))
1009 /* s_op->delete_inode internally recalls clear_inode() */
1013 spin_lock(&inode_lock);
1014 hlist_del_init(&inode->i_hash);
1015 spin_unlock(&inode_lock);
1016 wake_up_inode(inode);
1017 if (inode->i_state != I_CLEAR)
1019 destroy_inode(inode);
1022 EXPORT_SYMBOL(generic_delete_inode);
1024 static void generic_forget_inode(struct inode *inode)
1026 struct super_block *sb = inode->i_sb;
1028 if (!hlist_unhashed(&inode->i_hash)) {
1029 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1030 list_move(&inode->i_list, &inode_unused);
1031 inodes_stat.nr_unused++;
1032 spin_unlock(&inode_lock);
1033 if (!sb || (sb->s_flags & MS_ACTIVE))
1035 write_inode_now(inode, 1);
1036 spin_lock(&inode_lock);
1037 inodes_stat.nr_unused--;
1038 hlist_del_init(&inode->i_hash);
1040 list_del_init(&inode->i_list);
1041 inode->i_state|=I_FREEING;
1042 inodes_stat.nr_inodes--;
1043 spin_unlock(&inode_lock);
1044 if (inode->i_data.nrpages)
1045 truncate_inode_pages(&inode->i_data, 0);
1047 destroy_inode(inode);
1051 * Normal UNIX filesystem behaviour: delete the
1052 * inode when the usage count drops to zero, and
1055 static void generic_drop_inode(struct inode *inode)
1057 if (!inode->i_nlink)
1058 generic_delete_inode(inode);
1060 generic_forget_inode(inode);
1064 * Called when we're dropping the last reference
1067 * Call the FS "drop()" function, defaulting to
1068 * the legacy UNIX filesystem behaviour..
1070 * NOTE! NOTE! NOTE! We're called with the inode lock
1071 * held, and the drop function is supposed to release
1074 static inline void iput_final(struct inode *inode)
1076 struct super_operations *op = inode->i_sb->s_op;
1077 void (*drop)(struct inode *) = generic_drop_inode;
1079 if (op && op->drop_inode)
1080 drop = op->drop_inode;
1085 * iput - put an inode
1086 * @inode: inode to put
1088 * Puts an inode, dropping its usage count. If the inode use count hits
1089 * zero the inode is also then freed and may be destroyed.
1091 void iput(struct inode *inode)
1094 struct super_operations *op = inode->i_sb->s_op;
1096 if (inode->i_state == I_CLEAR)
1099 if (op && op->put_inode)
1100 op->put_inode(inode);
1102 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1107 EXPORT_SYMBOL(iput);
1110 * bmap - find a block number in a file
1111 * @inode: inode of file
1112 * @block: block to find
1114 * Returns the block number on the device holding the inode that
1115 * is the disk block number for the block of the file requested.
1116 * That is, asked for block 4 of inode 1 the function will return the
1117 * disk block relative to the disk start that holds that block of the
1120 sector_t bmap(struct inode * inode, sector_t block)
1123 if (inode->i_mapping->a_ops->bmap)
1124 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1128 EXPORT_SYMBOL(bmap);
1131 * Return true if the filesystem which backs this inode considers the two
1132 * passed timespecs to be sufficiently different to warrant flushing the
1133 * altered time out to disk.
1135 static int inode_times_differ(struct inode *inode,
1136 struct timespec *old, struct timespec *new)
1138 if (IS_ONE_SECOND(inode))
1139 return old->tv_sec != new->tv_sec;
1140 return !timespec_equal(old, new);
1144 * update_atime - update the access time
1145 * @inode: inode accessed
1147 * Update the accessed time on an inode and mark it for writeback.
1148 * This function automatically handles read only file systems and media,
1149 * as well as the "noatime" flag and inode specific "noatime" markers.
1151 void update_atime(struct inode *inode)
1153 struct timespec now;
1155 if (IS_NOATIME(inode))
1157 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1159 if (IS_RDONLY(inode))
1162 now = current_kernel_time();
1163 if (inode_times_differ(inode, &inode->i_atime, &now)) {
1164 inode->i_atime = now;
1165 mark_inode_dirty_sync(inode);
1167 if (!timespec_equal(&inode->i_atime, &now))
1168 inode->i_atime = now;
1172 EXPORT_SYMBOL(update_atime);
1175 * inode_update_time - update mtime and ctime time
1176 * @inode: inode accessed
1177 * @ctime_too: update ctime too
1179 * Update the mtime time on an inode and mark it for writeback.
1180 * When ctime_too is specified update the ctime too.
1183 void inode_update_time(struct inode *inode, int ctime_too)
1185 struct timespec now;
1188 if (IS_NOCMTIME(inode))
1190 if (IS_RDONLY(inode))
1193 now = current_kernel_time();
1195 if (inode_times_differ(inode, &inode->i_mtime, &now))
1197 inode->i_mtime = now;
1200 if (inode_times_differ(inode, &inode->i_ctime, &now))
1202 inode->i_ctime = now;
1205 mark_inode_dirty_sync(inode);
1208 EXPORT_SYMBOL(inode_update_time);
1210 int inode_needs_sync(struct inode *inode)
1214 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1219 EXPORT_SYMBOL(inode_needs_sync);
1222 * Quota functions that want to walk the inode lists..
1226 /* Function back in dquot.c */
1227 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1229 void remove_dquot_ref(struct super_block *sb, int type, struct list_head *tofree_head)
1231 struct inode *inode;
1232 struct list_head *act_head;
1235 return; /* nothing to do */
1236 spin_lock(&inode_lock); /* This lock is for inodes code */
1237 /* We don't have to lock against quota code - test IS_QUOTAINIT is just for speedup... */
1239 list_for_each(act_head, &inode_in_use) {
1240 inode = list_entry(act_head, struct inode, i_list);
1241 if (inode->i_sb == sb && IS_QUOTAINIT(inode))
1242 remove_inode_dquot_ref(inode, type, tofree_head);
1244 list_for_each(act_head, &inode_unused) {
1245 inode = list_entry(act_head, struct inode, i_list);
1246 if (inode->i_sb == sb && IS_QUOTAINIT(inode))
1247 remove_inode_dquot_ref(inode, type, tofree_head);
1249 list_for_each(act_head, &sb->s_dirty) {
1250 inode = list_entry(act_head, struct inode, i_list);
1251 if (IS_QUOTAINIT(inode))
1252 remove_inode_dquot_ref(inode, type, tofree_head);
1254 list_for_each(act_head, &sb->s_io) {
1255 inode = list_entry(act_head, struct inode, i_list);
1256 if (IS_QUOTAINIT(inode))
1257 remove_inode_dquot_ref(inode, type, tofree_head);
1259 spin_unlock(&inode_lock);
1265 * Hashed waitqueues for wait_on_inode(). The table is pretty small - the
1266 * kernel doesn't lock many inodes at the same time.
1268 #define I_WAIT_TABLE_ORDER 3
1269 static struct i_wait_queue_head {
1270 wait_queue_head_t wqh;
1271 } ____cacheline_aligned_in_smp i_wait_queue_heads[1<<I_WAIT_TABLE_ORDER];
1274 * Return the address of the waitqueue_head to be used for this inode
1276 static wait_queue_head_t *i_waitq_head(struct inode *inode)
1278 return &i_wait_queue_heads[hash_ptr(inode, I_WAIT_TABLE_ORDER)].wqh;
1281 void __wait_on_inode(struct inode *inode)
1283 DECLARE_WAITQUEUE(wait, current);
1284 wait_queue_head_t *wq = i_waitq_head(inode);
1286 add_wait_queue(wq, &wait);
1288 set_current_state(TASK_UNINTERRUPTIBLE);
1289 if (inode->i_state & I_LOCK) {
1293 remove_wait_queue(wq, &wait);
1294 __set_current_state(TASK_RUNNING);
1298 * If we try to find an inode in the inode hash while it is being deleted, we
1299 * have to wait until the filesystem completes its deletion before reporting
1300 * that it isn't found. This is because iget will immediately call
1301 * ->read_inode, and we want to be sure that evidence of the deletion is found
1304 * This call might return early if an inode which shares the waitq is woken up.
1305 * This is most easily handled by the caller which will loop around again
1306 * looking for the inode.
1308 * This is called with inode_lock held.
1310 static void __wait_on_freeing_inode(struct inode *inode)
1312 DECLARE_WAITQUEUE(wait, current);
1313 wait_queue_head_t *wq = i_waitq_head(inode);
1315 add_wait_queue(wq, &wait);
1316 set_current_state(TASK_UNINTERRUPTIBLE);
1317 spin_unlock(&inode_lock);
1319 remove_wait_queue(wq, &wait);
1320 spin_lock(&inode_lock);
1323 void wake_up_inode(struct inode *inode)
1325 wait_queue_head_t *wq = i_waitq_head(inode);
1328 * Prevent speculative execution through spin_unlock(&inode_lock);
1331 if (waitqueue_active(wq))
1335 static __initdata unsigned long ihash_entries;
1336 static int __init set_ihash_entries(char *str)
1340 ihash_entries = simple_strtoul(str, &str, 0);
1343 __setup("ihash_entries=", set_ihash_entries);
1346 * Initialize the waitqueues and inode hash table.
1348 void __init inode_init(unsigned long mempages)
1350 struct hlist_head *head;
1351 unsigned long order;
1352 unsigned int nr_hash;
1355 for (i = 0; i < ARRAY_SIZE(i_wait_queue_heads); i++)
1356 init_waitqueue_head(&i_wait_queue_heads[i].wqh);
1359 ihash_entries = PAGE_SHIFT < 14 ?
1360 mempages >> (14 - PAGE_SHIFT) :
1361 mempages << (PAGE_SHIFT - 14);
1363 ihash_entries *= sizeof(struct hlist_head);
1364 for (order = 0; ((1UL << order) << PAGE_SHIFT) < ihash_entries; order++)
1373 nr_hash = (1UL << order) * PAGE_SIZE /
1374 sizeof(struct hlist_head);
1375 i_hash_mask = (nr_hash - 1);
1379 while ((tmp >>= 1UL) != 0UL)
1382 inode_hashtable = (struct hlist_head *)
1383 __get_free_pages(GFP_ATOMIC, order);
1384 } while (inode_hashtable == NULL && --order >= 0);
1386 printk("Inode-cache hash table entries: %d (order: %ld, %ld bytes)\n",
1387 nr_hash, order, (PAGE_SIZE << order));
1389 if (!inode_hashtable)
1390 panic("Failed to allocate inode hash table\n");
1392 head = inode_hashtable;
1395 INIT_HLIST_HEAD(head);
1400 /* inode slab cache */
1401 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1402 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, init_once,
1404 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1407 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1409 inode->i_mode = mode;
1410 if (S_ISCHR(mode)) {
1411 inode->i_fop = &def_chr_fops;
1412 inode->i_rdev = rdev;
1413 } else if (S_ISBLK(mode)) {
1414 inode->i_fop = &def_blk_fops;
1415 inode->i_rdev = rdev;
1416 } else if (S_ISFIFO(mode))
1417 inode->i_fop = &def_fifo_fops;
1418 else if (S_ISSOCK(mode))
1419 inode->i_fop = &bad_sock_fops;
1421 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1424 EXPORT_SYMBOL(init_special_inode);