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
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;
119 /* essential because of inode slab reuse */
121 inode->i_blkbits = sb->s_blocksize_bits;
123 atomic_set(&inode->i_count, 1);
124 inode->i_op = &empty_iops;
125 inode->i_fop = &empty_fops;
127 atomic_set(&inode->i_writecount, 0);
131 inode->i_generation = 0;
133 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
135 inode->i_pipe = NULL;
136 inode->i_bdev = NULL;
137 inode->i_cdev = NULL;
139 inode->i_security = NULL;
140 inode->dirtied_when = 0;
141 if (security_inode_alloc(inode)) {
142 if (inode->i_sb->s_op->destroy_inode)
143 inode->i_sb->s_op->destroy_inode(inode);
145 kmem_cache_free(inode_cachep, (inode));
149 mapping->a_ops = &empty_aops;
150 mapping->host = inode;
152 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
153 mapping->assoc_mapping = NULL;
154 mapping->backing_dev_info = &default_backing_dev_info;
157 * If the block_device provides a backing_dev_info for client
158 * inodes then use that. Otherwise the inode share the bdev's
162 struct backing_dev_info *bdi;
164 bdi = sb->s_bdev->bd_inode_backing_dev_info;
166 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
167 mapping->backing_dev_info = bdi;
169 memset(&inode->u, 0, sizeof(inode->u));
170 inode->i_mapping = mapping;
175 void destroy_inode(struct inode *inode)
177 if (inode_has_buffers(inode))
179 security_inode_free(inode);
180 if (inode->i_sb->s_op->destroy_inode)
181 inode->i_sb->s_op->destroy_inode(inode);
183 kmem_cache_free(inode_cachep, (inode));
188 * These are initializations that only need to be done
189 * once, because the fields are idempotent across use
190 * of the inode, so let the slab aware of that.
192 void inode_init_once(struct inode *inode)
194 memset(inode, 0, sizeof(*inode));
195 INIT_HLIST_NODE(&inode->i_hash);
196 INIT_LIST_HEAD(&inode->i_dentry);
197 INIT_LIST_HEAD(&inode->i_devices);
198 sema_init(&inode->i_sem, 1);
199 init_rwsem(&inode->i_alloc_sem);
200 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
201 rwlock_init(&inode->i_data.tree_lock);
202 spin_lock_init(&inode->i_data.i_mmap_lock);
203 INIT_LIST_HEAD(&inode->i_data.private_list);
204 spin_lock_init(&inode->i_data.private_lock);
205 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
206 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
207 spin_lock_init(&inode->i_lock);
208 i_size_ordered_init(inode);
211 EXPORT_SYMBOL(inode_init_once);
213 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
215 struct inode * inode = (struct inode *) foo;
217 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
218 SLAB_CTOR_CONSTRUCTOR)
219 inode_init_once(inode);
223 * inode_lock must be held
225 void __iget(struct inode * inode)
227 if (atomic_read(&inode->i_count)) {
228 atomic_inc(&inode->i_count);
231 atomic_inc(&inode->i_count);
232 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
233 list_move(&inode->i_list, &inode_in_use);
234 inodes_stat.nr_unused--;
237 EXPORT_SYMBOL_GPL(__iget);
240 * clear_inode - clear an inode
241 * @inode: inode to clear
243 * This is called by the filesystem to tell us
244 * that the inode is no longer useful. We just
245 * terminate it with extreme prejudice.
247 void clear_inode(struct inode *inode)
250 invalidate_inode_buffers(inode);
252 if (inode->i_data.nrpages)
254 if (!(inode->i_state & I_FREEING))
256 if (inode->i_state & I_CLEAR)
258 wait_on_inode(inode);
260 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
261 inode->i_sb->s_op->clear_inode(inode);
266 inode->i_state = I_CLEAR;
269 EXPORT_SYMBOL(clear_inode);
272 * dispose_list - dispose of the contents of a local list
273 * @head: the head of the list to free
275 * Dispose-list gets a local list with local inodes in it, so it doesn't
276 * need to worry about list corruption and SMP locks.
278 static void dispose_list(struct list_head *head)
282 while (!list_empty(head)) {
285 inode = list_entry(head->next, struct inode, i_list);
286 list_del(&inode->i_list);
288 if (inode->i_data.nrpages)
289 truncate_inode_pages(&inode->i_data, 0);
291 destroy_inode(inode);
294 spin_lock(&inode_lock);
295 inodes_stat.nr_inodes -= nr_disposed;
296 spin_unlock(&inode_lock);
300 * Invalidate all inodes for a device.
302 static int invalidate_list(struct list_head *head, struct list_head *dispose)
304 struct list_head *next;
305 int busy = 0, count = 0;
309 struct list_head * tmp = next;
310 struct inode * inode;
313 * We can reschedule here without worrying about the list's
314 * consistency because the per-sb list of inodes must not
315 * change during umount anymore, and because iprune_sem keeps
316 * shrink_icache_memory() away.
318 cond_resched_lock(&inode_lock);
323 inode = list_entry(tmp, struct inode, i_sb_list);
324 invalidate_inode_buffers(inode);
325 if (!atomic_read(&inode->i_count)) {
326 hlist_del_init(&inode->i_hash);
327 list_del(&inode->i_sb_list);
328 list_move(&inode->i_list, dispose);
329 inode->i_state |= I_FREEING;
335 /* only unused inodes may be cached with i_count zero */
336 inodes_stat.nr_unused -= count;
341 * invalidate_inodes - discard the inodes on a device
344 * Discard all of the inodes for a given superblock. If the discard
345 * fails because there are busy inodes then a non zero value is returned.
346 * If the discard is successful all the inodes have been discarded.
348 int invalidate_inodes(struct super_block * sb)
351 LIST_HEAD(throw_away);
354 spin_lock(&inode_lock);
355 busy = invalidate_list(&sb->s_inodes, &throw_away);
356 spin_unlock(&inode_lock);
358 dispose_list(&throw_away);
364 EXPORT_SYMBOL(invalidate_inodes);
366 int __invalidate_device(struct block_device *bdev)
368 struct super_block *sb = get_super(bdev);
373 * no need to lock the super, get_super holds the
374 * read semaphore so the filesystem cannot go away
375 * under us (->put_super runs with the write lock
378 shrink_dcache_sb(sb);
379 res = invalidate_inodes(sb);
382 invalidate_bdev(bdev, 0);
385 EXPORT_SYMBOL(__invalidate_device);
387 static int can_unuse(struct inode *inode)
391 if (inode_has_buffers(inode))
393 if (atomic_read(&inode->i_count))
395 if (inode->i_data.nrpages)
401 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
402 * a temporary list and then are freed outside inode_lock by dispose_list().
404 * Any inodes which are pinned purely because of attached pagecache have their
405 * pagecache removed. We expect the final iput() on that inode to add it to
406 * the front of the inode_unused list. So look for it there and if the
407 * inode is still freeable, proceed. The right inode is found 99.9% of the
408 * time in testing on a 4-way.
410 * If the inode has metadata buffers attached to mapping->private_list then
411 * try to remove them.
413 static void prune_icache(int nr_to_scan)
418 unsigned long reap = 0;
421 spin_lock(&inode_lock);
422 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
425 if (list_empty(&inode_unused))
428 inode = list_entry(inode_unused.prev, struct inode, i_list);
430 if (inode->i_state || atomic_read(&inode->i_count)) {
431 list_move(&inode->i_list, &inode_unused);
434 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
436 spin_unlock(&inode_lock);
437 if (remove_inode_buffers(inode))
438 reap += invalidate_inode_pages(&inode->i_data);
440 spin_lock(&inode_lock);
442 if (inode != list_entry(inode_unused.next,
443 struct inode, i_list))
444 continue; /* wrong inode or list_empty */
445 if (!can_unuse(inode))
448 hlist_del_init(&inode->i_hash);
449 list_del_init(&inode->i_sb_list);
450 list_move(&inode->i_list, &freeable);
451 inode->i_state |= I_FREEING;
454 inodes_stat.nr_unused -= nr_pruned;
455 spin_unlock(&inode_lock);
457 dispose_list(&freeable);
460 if (current_is_kswapd())
461 mod_page_state(kswapd_inodesteal, reap);
463 mod_page_state(pginodesteal, reap);
467 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
468 * "unused" means that no dentries are referring to the inodes: the files are
469 * not open and the dcache references to those inodes have already been
472 * This function is passed the number of inodes to scan, and it returns the
473 * total number of remaining possibly-reclaimable inodes.
475 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
479 * Nasty deadlock avoidance. We may hold various FS locks,
480 * and we don't want to recurse into the FS that called us
481 * in clear_inode() and friends..
483 if (!(gfp_mask & __GFP_FS))
487 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
490 static void __wait_on_freeing_inode(struct inode *inode);
492 * Called with the inode lock held.
493 * NOTE: we are not increasing the inode-refcount, you must call __iget()
494 * by hand after calling find_inode now! This simplifies iunique and won't
495 * add any additional branch in the common code.
497 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
499 struct hlist_node *node;
500 struct inode * inode = NULL;
503 hlist_for_each (node, head) {
504 inode = hlist_entry(node, struct inode, i_hash);
505 if (inode->i_sb != sb)
507 if (!test(inode, data))
509 if (inode->i_state & (I_FREEING|I_CLEAR)) {
510 __wait_on_freeing_inode(inode);
515 return node ? inode : NULL;
519 * find_inode_fast is the fast path version of find_inode, see the comment at
520 * iget_locked for details.
522 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
524 struct hlist_node *node;
525 struct inode * inode = NULL;
528 hlist_for_each (node, head) {
529 inode = hlist_entry(node, struct inode, i_hash);
530 if (inode->i_ino != ino)
532 if (inode->i_sb != sb)
534 if (inode->i_state & (I_FREEING|I_CLEAR)) {
535 __wait_on_freeing_inode(inode);
540 return node ? inode : NULL;
544 * new_inode - obtain an inode
547 * Allocates a new inode for given superblock.
549 struct inode *new_inode(struct super_block *sb)
551 static unsigned long last_ino;
552 struct inode * inode;
554 spin_lock_prefetch(&inode_lock);
556 inode = alloc_inode(sb);
558 spin_lock(&inode_lock);
559 inodes_stat.nr_inodes++;
560 list_add(&inode->i_list, &inode_in_use);
561 list_add(&inode->i_sb_list, &sb->s_inodes);
562 inode->i_ino = ++last_ino;
564 spin_unlock(&inode_lock);
569 EXPORT_SYMBOL(new_inode);
571 void unlock_new_inode(struct inode *inode)
574 * This is special! We do not need the spinlock
575 * when clearing I_LOCK, because we're guaranteed
576 * that nobody else tries to do anything about the
577 * state of the inode when it is locked, as we
578 * just created it (so there can be no old holders
579 * that haven't tested I_LOCK).
581 inode->i_state &= ~(I_LOCK|I_NEW);
582 wake_up_inode(inode);
585 EXPORT_SYMBOL(unlock_new_inode);
588 * This is called without the inode lock held.. Be careful.
590 * We no longer cache the sb_flags in i_flags - see fs.h
591 * -- rmk@arm.uk.linux.org
593 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)
595 struct inode * inode;
597 inode = alloc_inode(sb);
601 spin_lock(&inode_lock);
602 /* We released the lock, so.. */
603 old = find_inode(sb, head, test, data);
605 if (set(inode, data))
608 inodes_stat.nr_inodes++;
609 list_add(&inode->i_list, &inode_in_use);
610 list_add(&inode->i_sb_list, &sb->s_inodes);
611 hlist_add_head(&inode->i_hash, head);
612 inode->i_state = I_LOCK|I_NEW;
613 spin_unlock(&inode_lock);
615 /* Return the locked inode with I_NEW set, the
616 * caller is responsible for filling in the contents
622 * Uhhuh, somebody else created the same inode under
623 * us. Use the old inode instead of the one we just
627 spin_unlock(&inode_lock);
628 destroy_inode(inode);
630 wait_on_inode(inode);
635 spin_unlock(&inode_lock);
636 destroy_inode(inode);
641 * get_new_inode_fast is the fast path version of get_new_inode, see the
642 * comment at iget_locked for details.
644 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
646 struct inode * inode;
648 inode = alloc_inode(sb);
652 spin_lock(&inode_lock);
653 /* We released the lock, so.. */
654 old = find_inode_fast(sb, head, ino);
657 inodes_stat.nr_inodes++;
658 list_add(&inode->i_list, &inode_in_use);
659 list_add(&inode->i_sb_list, &sb->s_inodes);
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 list_del_init(&inode->i_sb_list);
997 inode->i_state|=I_FREEING;
998 inodes_stat.nr_inodes--;
999 spin_unlock(&inode_lock);
1001 if (inode->i_data.nrpages)
1002 truncate_inode_pages(&inode->i_data, 0);
1004 security_inode_delete(inode);
1006 if (op->delete_inode) {
1007 void (*delete)(struct inode *) = op->delete_inode;
1008 if (!is_bad_inode(inode))
1010 /* s_op->delete_inode internally recalls clear_inode() */
1014 spin_lock(&inode_lock);
1015 hlist_del_init(&inode->i_hash);
1016 spin_unlock(&inode_lock);
1017 wake_up_inode(inode);
1018 if (inode->i_state != I_CLEAR)
1020 destroy_inode(inode);
1023 EXPORT_SYMBOL(generic_delete_inode);
1025 static void generic_forget_inode(struct inode *inode)
1027 struct super_block *sb = inode->i_sb;
1029 if (!hlist_unhashed(&inode->i_hash)) {
1030 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1031 list_move(&inode->i_list, &inode_unused);
1032 inodes_stat.nr_unused++;
1033 spin_unlock(&inode_lock);
1034 if (!sb || (sb->s_flags & MS_ACTIVE))
1036 write_inode_now(inode, 1);
1037 spin_lock(&inode_lock);
1038 inodes_stat.nr_unused--;
1039 hlist_del_init(&inode->i_hash);
1041 list_del_init(&inode->i_list);
1042 list_del_init(&inode->i_sb_list);
1043 inode->i_state|=I_FREEING;
1044 inodes_stat.nr_inodes--;
1045 spin_unlock(&inode_lock);
1046 if (inode->i_data.nrpages)
1047 truncate_inode_pages(&inode->i_data, 0);
1049 destroy_inode(inode);
1053 * Normal UNIX filesystem behaviour: delete the
1054 * inode when the usage count drops to zero, and
1057 static void generic_drop_inode(struct inode *inode)
1059 if (!inode->i_nlink)
1060 generic_delete_inode(inode);
1062 generic_forget_inode(inode);
1066 * Called when we're dropping the last reference
1069 * Call the FS "drop()" function, defaulting to
1070 * the legacy UNIX filesystem behaviour..
1072 * NOTE! NOTE! NOTE! We're called with the inode lock
1073 * held, and the drop function is supposed to release
1076 static inline void iput_final(struct inode *inode)
1078 struct super_operations *op = inode->i_sb->s_op;
1079 void (*drop)(struct inode *) = generic_drop_inode;
1081 if (op && op->drop_inode)
1082 drop = op->drop_inode;
1087 * iput - put an inode
1088 * @inode: inode to put
1090 * Puts an inode, dropping its usage count. If the inode use count hits
1091 * zero, the inode is then freed and may also be destroyed.
1093 * Consequently, iput() can sleep.
1095 void iput(struct inode *inode)
1098 struct super_operations *op = inode->i_sb->s_op;
1100 BUG_ON(inode->i_state == I_CLEAR);
1102 if (op && op->put_inode)
1103 op->put_inode(inode);
1105 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1110 EXPORT_SYMBOL(iput);
1113 * bmap - find a block number in a file
1114 * @inode: inode of file
1115 * @block: block to find
1117 * Returns the block number on the device holding the inode that
1118 * is the disk block number for the block of the file requested.
1119 * That is, asked for block 4 of inode 1 the function will return the
1120 * disk block relative to the disk start that holds that block of the
1123 sector_t bmap(struct inode * inode, sector_t block)
1126 if (inode->i_mapping->a_ops->bmap)
1127 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1131 EXPORT_SYMBOL(bmap);
1134 * update_atime - update the access time
1135 * @inode: inode accessed
1137 * Update the accessed time on an inode and mark it for writeback.
1138 * This function automatically handles read only file systems and media,
1139 * as well as the "noatime" flag and inode specific "noatime" markers.
1141 void update_atime(struct inode *inode)
1143 struct timespec now;
1145 if (IS_NOATIME(inode))
1147 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1149 if (IS_RDONLY(inode))
1152 now = current_fs_time(inode->i_sb);
1153 if (!timespec_equal(&inode->i_atime, &now)) {
1154 inode->i_atime = now;
1155 mark_inode_dirty_sync(inode);
1157 if (!timespec_equal(&inode->i_atime, &now))
1158 inode->i_atime = now;
1162 EXPORT_SYMBOL(update_atime);
1165 * inode_update_time - update mtime and ctime time
1166 * @inode: inode accessed
1167 * @ctime_too: update ctime too
1169 * Update the mtime time on an inode and mark it for writeback.
1170 * When ctime_too is specified update the ctime too.
1173 void inode_update_time(struct inode *inode, struct vfsmount *mnt, int ctime_too)
1175 struct timespec now;
1178 if (IS_NOCMTIME(inode))
1180 if (IS_RDONLY(inode) || MNT_IS_RDONLY(mnt))
1183 now = current_fs_time(inode->i_sb);
1184 if (!timespec_equal(&inode->i_mtime, &now))
1186 inode->i_mtime = now;
1189 if (!timespec_equal(&inode->i_ctime, &now))
1191 inode->i_ctime = now;
1194 mark_inode_dirty_sync(inode);
1197 EXPORT_SYMBOL(inode_update_time);
1199 int inode_needs_sync(struct inode *inode)
1203 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1208 EXPORT_SYMBOL(inode_needs_sync);
1211 * Quota functions that want to walk the inode lists..
1215 /* Function back in dquot.c */
1216 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1218 void remove_dquot_ref(struct super_block *sb, int type,
1219 struct list_head *tofree_head)
1221 struct inode *inode;
1224 return; /* nothing to do */
1225 spin_lock(&inode_lock); /* This lock is for inodes code */
1228 * We don't have to lock against quota code - test IS_QUOTAINIT is
1229 * just for speedup...
1231 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1232 if (!IS_NOQUOTA(inode))
1233 remove_inode_dquot_ref(inode, type, tofree_head);
1235 spin_unlock(&inode_lock);
1240 int inode_wait(void *word)
1247 * If we try to find an inode in the inode hash while it is being deleted, we
1248 * have to wait until the filesystem completes its deletion before reporting
1249 * that it isn't found. This is because iget will immediately call
1250 * ->read_inode, and we want to be sure that evidence of the deletion is found
1252 * This is called with inode_lock held.
1254 static void __wait_on_freeing_inode(struct inode *inode)
1256 wait_queue_head_t *wq;
1257 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1260 * I_FREEING and I_CLEAR are cleared in process context under
1261 * inode_lock, so we have to give the tasks who would clear them
1262 * a chance to run and acquire inode_lock.
1264 if (!(inode->i_state & I_LOCK)) {
1265 spin_unlock(&inode_lock);
1267 spin_lock(&inode_lock);
1270 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1271 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1272 spin_unlock(&inode_lock);
1274 finish_wait(wq, &wait.wait);
1275 spin_lock(&inode_lock);
1278 void wake_up_inode(struct inode *inode)
1281 * Prevent speculative execution through spin_unlock(&inode_lock);
1284 wake_up_bit(&inode->i_state, __I_LOCK);
1287 static __initdata unsigned long ihash_entries;
1288 static int __init set_ihash_entries(char *str)
1292 ihash_entries = simple_strtoul(str, &str, 0);
1295 __setup("ihash_entries=", set_ihash_entries);
1298 * Initialize the waitqueues and inode hash table.
1300 void __init inode_init_early(void)
1304 /* If hashes are distributed across NUMA nodes, defer
1305 * hash allocation until vmalloc space is available.
1311 alloc_large_system_hash("Inode-cache",
1312 sizeof(struct hlist_head),
1320 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1321 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1324 void __init inode_init(unsigned long mempages)
1328 /* inode slab cache */
1329 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1330 0, SLAB_RECLAIM_ACCOUNT|SLAB_PANIC, init_once, NULL);
1331 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1333 /* Hash may have been set up in inode_init_early */
1338 alloc_large_system_hash("Inode-cache",
1339 sizeof(struct hlist_head),
1347 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1348 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1351 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1353 inode->i_mode = mode;
1354 if (S_ISCHR(mode)) {
1355 inode->i_fop = &def_chr_fops;
1356 inode->i_rdev = rdev;
1357 } else if (S_ISBLK(mode)) {
1358 inode->i_fop = &def_blk_fops;
1359 inode->i_rdev = rdev;
1360 } else if (S_ISFIFO(mode))
1361 inode->i_fop = &def_fifo_fops;
1362 else if (S_ISSOCK(mode))
1363 inode->i_fop = &bad_sock_fops;
1365 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1368 EXPORT_SYMBOL(init_special_inode);