4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/quotaops.h>
12 #include <linux/slab.h>
13 #include <linux/writeback.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/wait.h>
17 #include <linux/hash.h>
18 #include <linux/swap.h>
19 #include <linux/security.h>
20 #include <linux/pagemap.h>
21 #include <linux/cdev.h>
22 #include <linux/bootmem.h>
23 #include <linux/inotify.h>
24 #include <linux/mount.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 __read_mostly;
59 static unsigned int i_hash_shift __read_mostly;
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 __read_mostly;
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_mutex 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 static DEFINE_MUTEX(iprune_mutex);
96 * Statistics gathering..
98 struct inodes_stat_t inodes_stat;
100 static struct kmem_cache * inode_cachep __read_mostly;
102 static void wake_up_inode(struct inode *inode)
105 * Prevent speculative execution through spin_unlock(&inode_lock);
108 wake_up_bit(&inode->i_state, __I_LOCK);
111 static struct inode *alloc_inode(struct super_block *sb)
113 static const struct address_space_operations empty_aops;
114 static struct inode_operations empty_iops;
115 static const struct file_operations empty_fops;
118 if (sb->s_op->alloc_inode)
119 inode = sb->s_op->alloc_inode(sb);
121 inode = (struct inode *) kmem_cache_alloc(inode_cachep, GFP_KERNEL);
124 struct address_space * const mapping = &inode->i_data;
127 inode->i_blkbits = sb->s_blocksize_bits;
129 atomic_set(&inode->i_count, 1);
130 inode->i_op = &empty_iops;
131 inode->i_fop = &empty_fops;
133 atomic_set(&inode->i_writecount, 0);
137 inode->i_generation = 0;
139 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
141 inode->i_pipe = NULL;
142 inode->i_bdev = NULL;
143 inode->i_cdev = NULL;
145 inode->dirtied_when = 0;
146 if (security_inode_alloc(inode)) {
147 if (inode->i_sb->s_op->destroy_inode)
148 inode->i_sb->s_op->destroy_inode(inode);
150 kmem_cache_free(inode_cachep, (inode));
154 spin_lock_init(&inode->i_lock);
155 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
157 mutex_init(&inode->i_mutex);
158 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
160 init_rwsem(&inode->i_alloc_sem);
161 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
163 mapping->a_ops = &empty_aops;
164 mapping->host = inode;
166 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_PAGECACHE);
167 mapping->assoc_mapping = NULL;
168 mapping->backing_dev_info = &default_backing_dev_info;
169 mapping->writeback_index = 0;
172 * If the block_device provides a backing_dev_info for client
173 * inodes then use that. Otherwise the inode share the bdev's
177 struct backing_dev_info *bdi;
179 bdi = sb->s_bdev->bd_inode_backing_dev_info;
181 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
182 mapping->backing_dev_info = bdi;
184 inode->i_private = NULL;
185 inode->i_mapping = mapping;
190 void destroy_inode(struct inode *inode)
192 BUG_ON(inode_has_buffers(inode));
193 security_inode_free(inode);
194 if (inode->i_sb->s_op->destroy_inode)
195 inode->i_sb->s_op->destroy_inode(inode);
197 kmem_cache_free(inode_cachep, (inode));
202 * These are initializations that only need to be done
203 * once, because the fields are idempotent across use
204 * of the inode, so let the slab aware of that.
206 void inode_init_once(struct inode *inode)
208 memset(inode, 0, sizeof(*inode));
209 INIT_HLIST_NODE(&inode->i_hash);
210 INIT_LIST_HEAD(&inode->i_dentry);
211 INIT_LIST_HEAD(&inode->i_devices);
212 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
213 spin_lock_init(&inode->i_data.tree_lock);
214 spin_lock_init(&inode->i_data.i_mmap_lock);
215 INIT_LIST_HEAD(&inode->i_data.private_list);
216 spin_lock_init(&inode->i_data.private_lock);
217 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
218 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
219 i_size_ordered_init(inode);
220 #ifdef CONFIG_INOTIFY
221 INIT_LIST_HEAD(&inode->inotify_watches);
222 mutex_init(&inode->inotify_mutex);
226 EXPORT_SYMBOL(inode_init_once);
228 static void init_once(void *foo)
230 struct inode * inode = (struct inode *) foo;
232 inode_init_once(inode);
236 * inode_lock must be held
238 void __iget(struct inode * inode)
240 if (atomic_read(&inode->i_count)) {
241 atomic_inc(&inode->i_count);
244 atomic_inc(&inode->i_count);
245 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
246 list_move(&inode->i_list, &inode_in_use);
247 inodes_stat.nr_unused--;
251 * clear_inode - clear an inode
252 * @inode: inode to clear
254 * This is called by the filesystem to tell us
255 * that the inode is no longer useful. We just
256 * terminate it with extreme prejudice.
258 void clear_inode(struct inode *inode)
261 invalidate_inode_buffers(inode);
263 BUG_ON(inode->i_data.nrpages);
264 BUG_ON(!(inode->i_state & I_FREEING));
265 BUG_ON(inode->i_state & I_CLEAR);
266 inode_sync_wait(inode);
268 if (inode->i_sb->s_op->clear_inode)
269 inode->i_sb->s_op->clear_inode(inode);
270 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
272 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
274 inode->i_state = I_CLEAR;
277 EXPORT_SYMBOL(clear_inode);
280 * dispose_list - dispose of the contents of a local list
281 * @head: the head of the list to free
283 * Dispose-list gets a local list with local inodes in it, so it doesn't
284 * need to worry about list corruption and SMP locks.
286 static void dispose_list(struct list_head *head)
290 while (!list_empty(head)) {
293 inode = list_first_entry(head, struct inode, i_list);
294 list_del(&inode->i_list);
296 if (inode->i_data.nrpages)
297 truncate_inode_pages(&inode->i_data, 0);
300 spin_lock(&inode_lock);
301 hlist_del_init(&inode->i_hash);
302 list_del_init(&inode->i_sb_list);
303 spin_unlock(&inode_lock);
305 wake_up_inode(inode);
306 destroy_inode(inode);
309 spin_lock(&inode_lock);
310 inodes_stat.nr_inodes -= nr_disposed;
311 spin_unlock(&inode_lock);
315 * Invalidate all inodes for a device.
317 static int invalidate_list(struct list_head *head, struct list_head *dispose)
319 struct list_head *next;
320 int busy = 0, count = 0;
324 struct list_head * tmp = next;
325 struct inode * inode;
328 * We can reschedule here without worrying about the list's
329 * consistency because the per-sb list of inodes must not
330 * change during umount anymore, and because iprune_mutex keeps
331 * shrink_icache_memory() away.
333 cond_resched_lock(&inode_lock);
338 inode = list_entry(tmp, struct inode, i_sb_list);
339 invalidate_inode_buffers(inode);
340 if (!atomic_read(&inode->i_count)) {
341 list_move(&inode->i_list, dispose);
342 inode->i_state |= I_FREEING;
348 /* only unused inodes may be cached with i_count zero */
349 inodes_stat.nr_unused -= count;
354 * invalidate_inodes - discard the inodes on a device
357 * Discard all of the inodes for a given superblock. If the discard
358 * fails because there are busy inodes then a non zero value is returned.
359 * If the discard is successful all the inodes have been discarded.
361 int invalidate_inodes(struct super_block * sb)
364 LIST_HEAD(throw_away);
366 mutex_lock(&iprune_mutex);
367 spin_lock(&inode_lock);
368 inotify_unmount_inodes(&sb->s_inodes);
369 busy = invalidate_list(&sb->s_inodes, &throw_away);
370 spin_unlock(&inode_lock);
372 dispose_list(&throw_away);
373 mutex_unlock(&iprune_mutex);
378 EXPORT_SYMBOL(invalidate_inodes);
380 static int can_unuse(struct inode *inode)
384 if (inode_has_buffers(inode))
386 if (atomic_read(&inode->i_count))
388 if (inode->i_data.nrpages)
394 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
395 * a temporary list and then are freed outside inode_lock by dispose_list().
397 * Any inodes which are pinned purely because of attached pagecache have their
398 * pagecache removed. We expect the final iput() on that inode to add it to
399 * the front of the inode_unused list. So look for it there and if the
400 * inode is still freeable, proceed. The right inode is found 99.9% of the
401 * time in testing on a 4-way.
403 * If the inode has metadata buffers attached to mapping->private_list then
404 * try to remove them.
406 static void prune_icache(int nr_to_scan)
411 unsigned long reap = 0;
413 mutex_lock(&iprune_mutex);
414 spin_lock(&inode_lock);
415 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
418 if (list_empty(&inode_unused))
421 inode = list_entry(inode_unused.prev, struct inode, i_list);
423 if (inode->i_state || atomic_read(&inode->i_count)) {
424 list_move(&inode->i_list, &inode_unused);
427 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
429 spin_unlock(&inode_lock);
430 if (remove_inode_buffers(inode))
431 reap += invalidate_mapping_pages(&inode->i_data,
434 spin_lock(&inode_lock);
436 if (inode != list_entry(inode_unused.next,
437 struct inode, i_list))
438 continue; /* wrong inode or list_empty */
439 if (!can_unuse(inode))
442 list_move(&inode->i_list, &freeable);
443 inode->i_state |= I_FREEING;
446 inodes_stat.nr_unused -= nr_pruned;
447 if (current_is_kswapd())
448 __count_vm_events(KSWAPD_INODESTEAL, reap);
450 __count_vm_events(PGINODESTEAL, reap);
451 spin_unlock(&inode_lock);
453 dispose_list(&freeable);
454 mutex_unlock(&iprune_mutex);
458 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
459 * "unused" means that no dentries are referring to the inodes: the files are
460 * not open and the dcache references to those inodes have already been
463 * This function is passed the number of inodes to scan, and it returns the
464 * total number of remaining possibly-reclaimable inodes.
466 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
470 * Nasty deadlock avoidance. We may hold various FS locks,
471 * and we don't want to recurse into the FS that called us
472 * in clear_inode() and friends..
474 if (!(gfp_mask & __GFP_FS))
478 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
481 static struct shrinker icache_shrinker = {
482 .shrink = shrink_icache_memory,
483 .seeks = DEFAULT_SEEKS,
486 static void __wait_on_freeing_inode(struct inode *inode);
488 * Called with the inode lock held.
489 * NOTE: we are not increasing the inode-refcount, you must call __iget()
490 * by hand after calling find_inode now! This simplifies iunique and won't
491 * add any additional branch in the common code.
493 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
495 struct hlist_node *node;
496 struct inode * inode = NULL;
499 hlist_for_each_entry(inode, node, head, i_hash) {
500 if (inode->i_sb != sb)
502 if (!test(inode, data))
504 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
505 __wait_on_freeing_inode(inode);
510 return node ? inode : NULL;
514 * find_inode_fast is the fast path version of find_inode, see the comment at
515 * iget_locked for details.
517 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
519 struct hlist_node *node;
520 struct inode * inode = NULL;
523 hlist_for_each_entry(inode, node, head, i_hash) {
524 if (inode->i_ino != ino)
526 if (inode->i_sb != sb)
528 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
529 __wait_on_freeing_inode(inode);
534 return node ? inode : NULL;
538 * new_inode - obtain an inode
541 * Allocates a new inode for given superblock. The default gfp_mask
542 * for allocations related to inode->i_mapping is GFP_HIGHUSER_PAGECACHE.
543 * If HIGHMEM pages are unsuitable or it is known that pages allocated
544 * for the page cache are not reclaimable or migratable,
545 * mapping_set_gfp_mask() must be called with suitable flags on the
546 * newly created inode's mapping
549 struct inode *new_inode(struct super_block *sb)
552 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
553 * error if st_ino won't fit in target struct field. Use 32bit counter
554 * here to attempt to avoid that.
556 static unsigned int last_ino;
557 struct inode * inode;
559 spin_lock_prefetch(&inode_lock);
561 inode = alloc_inode(sb);
563 spin_lock(&inode_lock);
564 inodes_stat.nr_inodes++;
565 list_add(&inode->i_list, &inode_in_use);
566 list_add(&inode->i_sb_list, &sb->s_inodes);
567 inode->i_ino = ++last_ino;
569 spin_unlock(&inode_lock);
574 EXPORT_SYMBOL(new_inode);
576 void unlock_new_inode(struct inode *inode)
578 #ifdef CONFIG_DEBUG_LOCK_ALLOC
579 if (inode->i_mode & S_IFDIR) {
580 struct file_system_type *type = inode->i_sb->s_type;
583 * ensure nobody is actually holding i_mutex
585 mutex_destroy(&inode->i_mutex);
586 mutex_init(&inode->i_mutex);
587 lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key);
591 * This is special! We do not need the spinlock
592 * when clearing I_LOCK, because we're guaranteed
593 * that nobody else tries to do anything about the
594 * state of the inode when it is locked, as we
595 * just created it (so there can be no old holders
596 * that haven't tested I_LOCK).
598 inode->i_state &= ~(I_LOCK|I_NEW);
599 wake_up_inode(inode);
602 EXPORT_SYMBOL(unlock_new_inode);
605 * This is called without the inode lock held.. Be careful.
607 * We no longer cache the sb_flags in i_flags - see fs.h
608 * -- rmk@arm.uk.linux.org
610 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)
612 struct inode * inode;
614 inode = alloc_inode(sb);
618 spin_lock(&inode_lock);
619 /* We released the lock, so.. */
620 old = find_inode(sb, head, test, data);
622 if (set(inode, data))
625 inodes_stat.nr_inodes++;
626 list_add(&inode->i_list, &inode_in_use);
627 list_add(&inode->i_sb_list, &sb->s_inodes);
628 hlist_add_head(&inode->i_hash, head);
629 inode->i_state = I_LOCK|I_NEW;
630 spin_unlock(&inode_lock);
632 /* Return the locked inode with I_NEW set, the
633 * caller is responsible for filling in the contents
639 * Uhhuh, somebody else created the same inode under
640 * us. Use the old inode instead of the one we just
644 spin_unlock(&inode_lock);
645 destroy_inode(inode);
647 wait_on_inode(inode);
652 spin_unlock(&inode_lock);
653 destroy_inode(inode);
658 * get_new_inode_fast is the fast path version of get_new_inode, see the
659 * comment at iget_locked for details.
661 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
663 struct inode * inode;
665 inode = alloc_inode(sb);
669 spin_lock(&inode_lock);
670 /* We released the lock, so.. */
671 old = find_inode_fast(sb, head, ino);
674 inodes_stat.nr_inodes++;
675 list_add(&inode->i_list, &inode_in_use);
676 list_add(&inode->i_sb_list, &sb->s_inodes);
677 hlist_add_head(&inode->i_hash, head);
678 inode->i_state = I_LOCK|I_NEW;
679 spin_unlock(&inode_lock);
681 /* Return the locked inode with I_NEW set, the
682 * caller is responsible for filling in the contents
688 * Uhhuh, somebody else created the same inode under
689 * us. Use the old inode instead of the one we just
693 spin_unlock(&inode_lock);
694 destroy_inode(inode);
696 wait_on_inode(inode);
701 static unsigned long hash(struct super_block *sb, unsigned long hashval)
705 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
707 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
708 return tmp & I_HASHMASK;
712 * iunique - get a unique inode number
714 * @max_reserved: highest reserved inode number
716 * Obtain an inode number that is unique on the system for a given
717 * superblock. This is used by file systems that have no natural
718 * permanent inode numbering system. An inode number is returned that
719 * is higher than the reserved limit but unique.
722 * With a large number of inodes live on the file system this function
723 * currently becomes quite slow.
725 ino_t iunique(struct super_block *sb, ino_t max_reserved)
728 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
729 * error if st_ino won't fit in target struct field. Use 32bit counter
730 * here to attempt to avoid that.
732 static unsigned int counter;
734 struct hlist_head *head;
737 spin_lock(&inode_lock);
739 if (counter <= max_reserved)
740 counter = max_reserved + 1;
742 head = inode_hashtable + hash(sb, res);
743 inode = find_inode_fast(sb, head, res);
744 } while (inode != NULL);
745 spin_unlock(&inode_lock);
749 EXPORT_SYMBOL(iunique);
751 struct inode *igrab(struct inode *inode)
753 spin_lock(&inode_lock);
754 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
758 * Handle the case where s_op->clear_inode is not been
759 * called yet, and somebody is calling igrab
760 * while the inode is getting freed.
763 spin_unlock(&inode_lock);
767 EXPORT_SYMBOL(igrab);
770 * ifind - internal function, you want ilookup5() or iget5().
771 * @sb: super block of file system to search
772 * @head: the head of the list to search
773 * @test: callback used for comparisons between inodes
774 * @data: opaque data pointer to pass to @test
775 * @wait: if true wait for the inode to be unlocked, if false do not
777 * ifind() searches for the inode specified by @data in the inode
778 * cache. This is a generalized version of ifind_fast() for file systems where
779 * the inode number is not sufficient for unique identification of an inode.
781 * If the inode is in the cache, the inode is returned with an incremented
784 * Otherwise NULL is returned.
786 * Note, @test is called with the inode_lock held, so can't sleep.
788 static struct inode *ifind(struct super_block *sb,
789 struct hlist_head *head, int (*test)(struct inode *, void *),
790 void *data, const int wait)
794 spin_lock(&inode_lock);
795 inode = find_inode(sb, head, test, data);
798 spin_unlock(&inode_lock);
800 wait_on_inode(inode);
803 spin_unlock(&inode_lock);
808 * ifind_fast - internal function, you want ilookup() or iget().
809 * @sb: super block of file system to search
810 * @head: head of the list to search
811 * @ino: inode number to search for
813 * ifind_fast() searches for the inode @ino in the inode cache. This is for
814 * file systems where the inode number is sufficient for unique identification
817 * If the inode is in the cache, the inode is returned with an incremented
820 * Otherwise NULL is returned.
822 static struct inode *ifind_fast(struct super_block *sb,
823 struct hlist_head *head, unsigned long ino)
827 spin_lock(&inode_lock);
828 inode = find_inode_fast(sb, head, ino);
831 spin_unlock(&inode_lock);
832 wait_on_inode(inode);
835 spin_unlock(&inode_lock);
840 * ilookup5_nowait - search for an inode in the inode cache
841 * @sb: super block of file system to search
842 * @hashval: hash value (usually inode number) to search for
843 * @test: callback used for comparisons between inodes
844 * @data: opaque data pointer to pass to @test
846 * ilookup5() uses ifind() to search for the inode specified by @hashval and
847 * @data in the inode cache. This is a generalized version of ilookup() for
848 * file systems where the inode number is not sufficient for unique
849 * identification of an inode.
851 * If the inode is in the cache, the inode is returned with an incremented
852 * reference count. Note, the inode lock is not waited upon so you have to be
853 * very careful what you do with the returned inode. You probably should be
854 * using ilookup5() instead.
856 * Otherwise NULL is returned.
858 * Note, @test is called with the inode_lock held, so can't sleep.
860 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
861 int (*test)(struct inode *, void *), void *data)
863 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
865 return ifind(sb, head, test, data, 0);
868 EXPORT_SYMBOL(ilookup5_nowait);
871 * ilookup5 - search for an inode in the inode cache
872 * @sb: super block of file system to search
873 * @hashval: hash value (usually inode number) to search for
874 * @test: callback used for comparisons between inodes
875 * @data: opaque data pointer to pass to @test
877 * ilookup5() uses ifind() to search for the inode specified by @hashval and
878 * @data in the inode cache. This is a generalized version of ilookup() for
879 * file systems where the inode number is not sufficient for unique
880 * identification of an inode.
882 * If the inode is in the cache, the inode lock is waited upon and the inode is
883 * returned with an incremented reference count.
885 * Otherwise NULL is returned.
887 * Note, @test is called with the inode_lock held, so can't sleep.
889 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
890 int (*test)(struct inode *, void *), void *data)
892 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
894 return ifind(sb, head, test, data, 1);
897 EXPORT_SYMBOL(ilookup5);
900 * ilookup - search for an inode in the inode cache
901 * @sb: super block of file system to search
902 * @ino: inode number to search for
904 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
905 * This is for file systems where the inode number is sufficient for unique
906 * identification of an inode.
908 * If the inode is in the cache, the inode is returned with an incremented
911 * Otherwise NULL is returned.
913 struct inode *ilookup(struct super_block *sb, unsigned long ino)
915 struct hlist_head *head = inode_hashtable + hash(sb, ino);
917 return ifind_fast(sb, head, ino);
920 EXPORT_SYMBOL(ilookup);
923 * iget5_locked - obtain an inode from a mounted file system
924 * @sb: super block of file system
925 * @hashval: hash value (usually inode number) to get
926 * @test: callback used for comparisons between inodes
927 * @set: callback used to initialize a new struct inode
928 * @data: opaque data pointer to pass to @test and @set
930 * iget5_locked() uses ifind() to search for the inode specified by @hashval
931 * and @data in the inode cache and if present it is returned with an increased
932 * reference count. This is a generalized version of iget_locked() for file
933 * systems where the inode number is not sufficient for unique identification
936 * If the inode is not in cache, get_new_inode() is called to allocate a new
937 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
938 * file system gets to fill it in before unlocking it via unlock_new_inode().
940 * Note both @test and @set are called with the inode_lock held, so can't sleep.
942 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
943 int (*test)(struct inode *, void *),
944 int (*set)(struct inode *, void *), void *data)
946 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
949 inode = ifind(sb, head, test, data, 1);
953 * get_new_inode() will do the right thing, re-trying the search
954 * in case it had to block at any point.
956 return get_new_inode(sb, head, test, set, data);
959 EXPORT_SYMBOL(iget5_locked);
962 * iget_locked - obtain an inode from a mounted file system
963 * @sb: super block of file system
964 * @ino: inode number to get
966 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
967 * the inode cache and if present it is returned with an increased reference
968 * count. This is for file systems where the inode number is sufficient for
969 * unique identification of an inode.
971 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
972 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
973 * The file system gets to fill it in before unlocking it via
974 * unlock_new_inode().
976 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
978 struct hlist_head *head = inode_hashtable + hash(sb, ino);
981 inode = ifind_fast(sb, head, ino);
985 * get_new_inode_fast() will do the right thing, re-trying the search
986 * in case it had to block at any point.
988 return get_new_inode_fast(sb, head, ino);
991 EXPORT_SYMBOL(iget_locked);
994 * __insert_inode_hash - hash an inode
995 * @inode: unhashed inode
996 * @hashval: unsigned long value used to locate this object in the
999 * Add an inode to the inode hash for this superblock.
1001 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1003 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1004 spin_lock(&inode_lock);
1005 hlist_add_head(&inode->i_hash, head);
1006 spin_unlock(&inode_lock);
1009 EXPORT_SYMBOL(__insert_inode_hash);
1012 * remove_inode_hash - remove an inode from the hash
1013 * @inode: inode to unhash
1015 * Remove an inode from the superblock.
1017 void remove_inode_hash(struct inode *inode)
1019 spin_lock(&inode_lock);
1020 hlist_del_init(&inode->i_hash);
1021 spin_unlock(&inode_lock);
1024 EXPORT_SYMBOL(remove_inode_hash);
1027 * Tell the filesystem that this inode is no longer of any interest and should
1028 * be completely destroyed.
1030 * We leave the inode in the inode hash table until *after* the filesystem's
1031 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1032 * instigate) will always find up-to-date information either in the hash or on
1035 * I_FREEING is set so that no-one will take a new reference to the inode while
1036 * it is being deleted.
1038 void generic_delete_inode(struct inode *inode)
1040 const struct super_operations *op = inode->i_sb->s_op;
1042 list_del_init(&inode->i_list);
1043 list_del_init(&inode->i_sb_list);
1044 inode->i_state |= I_FREEING;
1045 inodes_stat.nr_inodes--;
1046 spin_unlock(&inode_lock);
1048 security_inode_delete(inode);
1050 if (op->delete_inode) {
1051 void (*delete)(struct inode *) = op->delete_inode;
1052 if (!is_bad_inode(inode))
1054 /* Filesystems implementing their own
1055 * s_op->delete_inode are required to call
1056 * truncate_inode_pages and clear_inode()
1060 truncate_inode_pages(&inode->i_data, 0);
1063 spin_lock(&inode_lock);
1064 hlist_del_init(&inode->i_hash);
1065 spin_unlock(&inode_lock);
1066 wake_up_inode(inode);
1067 BUG_ON(inode->i_state != I_CLEAR);
1068 destroy_inode(inode);
1071 EXPORT_SYMBOL(generic_delete_inode);
1073 static void generic_forget_inode(struct inode *inode)
1075 struct super_block *sb = inode->i_sb;
1077 if (!hlist_unhashed(&inode->i_hash)) {
1078 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1079 list_move(&inode->i_list, &inode_unused);
1080 inodes_stat.nr_unused++;
1081 if (sb->s_flags & MS_ACTIVE) {
1082 spin_unlock(&inode_lock);
1085 inode->i_state |= I_WILL_FREE;
1086 spin_unlock(&inode_lock);
1087 write_inode_now(inode, 1);
1088 spin_lock(&inode_lock);
1089 inode->i_state &= ~I_WILL_FREE;
1090 inodes_stat.nr_unused--;
1091 hlist_del_init(&inode->i_hash);
1093 list_del_init(&inode->i_list);
1094 list_del_init(&inode->i_sb_list);
1095 inode->i_state |= I_FREEING;
1096 inodes_stat.nr_inodes--;
1097 spin_unlock(&inode_lock);
1098 if (inode->i_data.nrpages)
1099 truncate_inode_pages(&inode->i_data, 0);
1101 wake_up_inode(inode);
1102 destroy_inode(inode);
1106 * Normal UNIX filesystem behaviour: delete the
1107 * inode when the usage count drops to zero, and
1110 void generic_drop_inode(struct inode *inode)
1112 if (!inode->i_nlink)
1113 generic_delete_inode(inode);
1115 generic_forget_inode(inode);
1118 EXPORT_SYMBOL_GPL(generic_drop_inode);
1121 * Called when we're dropping the last reference
1124 * Call the FS "drop()" function, defaulting to
1125 * the legacy UNIX filesystem behaviour..
1127 * NOTE! NOTE! NOTE! We're called with the inode lock
1128 * held, and the drop function is supposed to release
1131 static inline void iput_final(struct inode *inode)
1133 const struct super_operations *op = inode->i_sb->s_op;
1134 void (*drop)(struct inode *) = generic_drop_inode;
1136 if (op && op->drop_inode)
1137 drop = op->drop_inode;
1142 * iput - put an inode
1143 * @inode: inode to put
1145 * Puts an inode, dropping its usage count. If the inode use count hits
1146 * zero, the inode is then freed and may also be destroyed.
1148 * Consequently, iput() can sleep.
1150 void iput(struct inode *inode)
1153 BUG_ON(inode->i_state == I_CLEAR);
1155 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1160 EXPORT_SYMBOL(iput);
1163 * bmap - find a block number in a file
1164 * @inode: inode of file
1165 * @block: block to find
1167 * Returns the block number on the device holding the inode that
1168 * is the disk block number for the block of the file requested.
1169 * That is, asked for block 4 of inode 1 the function will return the
1170 * disk block relative to the disk start that holds that block of the
1173 sector_t bmap(struct inode * inode, sector_t block)
1176 if (inode->i_mapping->a_ops->bmap)
1177 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1180 EXPORT_SYMBOL(bmap);
1183 * touch_atime - update the access time
1184 * @mnt: mount the inode is accessed on
1185 * @dentry: dentry accessed
1187 * Update the accessed time on an inode and mark it for writeback.
1188 * This function automatically handles read only file systems and media,
1189 * as well as the "noatime" flag and inode specific "noatime" markers.
1191 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1193 struct inode *inode = dentry->d_inode;
1194 struct timespec now;
1196 if (mnt_want_write(mnt))
1198 if (inode->i_flags & S_NOATIME)
1200 if (IS_NOATIME(inode))
1202 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1205 if (mnt->mnt_flags & MNT_NOATIME)
1207 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1209 if (mnt->mnt_flags & MNT_RELATIME) {
1211 * With relative atime, only update atime if the previous
1212 * atime is earlier than either the ctime or mtime.
1214 if (timespec_compare(&inode->i_mtime, &inode->i_atime) < 0 &&
1215 timespec_compare(&inode->i_ctime, &inode->i_atime) < 0)
1219 now = current_fs_time(inode->i_sb);
1220 if (timespec_equal(&inode->i_atime, &now))
1223 inode->i_atime = now;
1224 mark_inode_dirty_sync(inode);
1226 mnt_drop_write(mnt);
1228 EXPORT_SYMBOL(touch_atime);
1231 * file_update_time - update mtime and ctime time
1232 * @file: file accessed
1234 * Update the mtime and ctime members of an inode and mark the inode
1235 * for writeback. Note that this function is meant exclusively for
1236 * usage in the file write path of filesystems, and filesystems may
1237 * choose to explicitly ignore update via this function with the
1238 * S_NOCTIME inode flag, e.g. for network filesystem where these
1239 * timestamps are handled by the server.
1242 void file_update_time(struct file *file)
1244 struct inode *inode = file->f_path.dentry->d_inode;
1245 struct timespec now;
1249 if (IS_NOCMTIME(inode))
1252 err = mnt_want_write(file->f_path.mnt);
1256 now = current_fs_time(inode->i_sb);
1257 if (!timespec_equal(&inode->i_mtime, &now)) {
1258 inode->i_mtime = now;
1262 if (!timespec_equal(&inode->i_ctime, &now)) {
1263 inode->i_ctime = now;
1267 if (IS_I_VERSION(inode)) {
1268 inode_inc_iversion(inode);
1273 mark_inode_dirty_sync(inode);
1274 mnt_drop_write(file->f_path.mnt);
1277 EXPORT_SYMBOL(file_update_time);
1279 int inode_needs_sync(struct inode *inode)
1283 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1288 EXPORT_SYMBOL(inode_needs_sync);
1290 int inode_wait(void *word)
1297 * If we try to find an inode in the inode hash while it is being
1298 * deleted, we have to wait until the filesystem completes its
1299 * deletion before reporting that it isn't found. This function waits
1300 * until the deletion _might_ have completed. Callers are responsible
1301 * to recheck inode state.
1303 * It doesn't matter if I_LOCK is not set initially, a call to
1304 * wake_up_inode() after removing from the hash list will DTRT.
1306 * This is called with inode_lock held.
1308 static void __wait_on_freeing_inode(struct inode *inode)
1310 wait_queue_head_t *wq;
1311 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1312 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1313 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1314 spin_unlock(&inode_lock);
1316 finish_wait(wq, &wait.wait);
1317 spin_lock(&inode_lock);
1321 * We rarely want to lock two inodes that do not have a parent/child
1322 * relationship (such as directory, child inode) simultaneously. The
1323 * vast majority of file systems should be able to get along fine
1324 * without this. Do not use these functions except as a last resort.
1326 void inode_double_lock(struct inode *inode1, struct inode *inode2)
1328 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1330 mutex_lock(&inode1->i_mutex);
1332 mutex_lock(&inode2->i_mutex);
1336 if (inode1 < inode2) {
1337 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1338 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1340 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1341 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1344 EXPORT_SYMBOL(inode_double_lock);
1346 void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1349 mutex_unlock(&inode1->i_mutex);
1351 if (inode2 && inode2 != inode1)
1352 mutex_unlock(&inode2->i_mutex);
1354 EXPORT_SYMBOL(inode_double_unlock);
1356 static __initdata unsigned long ihash_entries;
1357 static int __init set_ihash_entries(char *str)
1361 ihash_entries = simple_strtoul(str, &str, 0);
1364 __setup("ihash_entries=", set_ihash_entries);
1367 * Initialize the waitqueues and inode hash table.
1369 void __init inode_init_early(void)
1373 /* If hashes are distributed across NUMA nodes, defer
1374 * hash allocation until vmalloc space is available.
1380 alloc_large_system_hash("Inode-cache",
1381 sizeof(struct hlist_head),
1389 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1390 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1393 void __init inode_init(void)
1397 /* inode slab cache */
1398 inode_cachep = kmem_cache_create("inode_cache",
1399 sizeof(struct inode),
1401 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1404 register_shrinker(&icache_shrinker);
1406 /* Hash may have been set up in inode_init_early */
1411 alloc_large_system_hash("Inode-cache",
1412 sizeof(struct hlist_head),
1420 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1421 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1424 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1426 inode->i_mode = mode;
1427 if (S_ISCHR(mode)) {
1428 inode->i_fop = &def_chr_fops;
1429 inode->i_rdev = rdev;
1430 } else if (S_ISBLK(mode)) {
1431 inode->i_fop = &def_blk_fops;
1432 inode->i_rdev = rdev;
1433 } else if (S_ISFIFO(mode))
1434 inode->i_fop = &def_fifo_fops;
1435 else if (S_ISSOCK(mode))
1436 inode->i_fop = &bad_sock_fops;
1438 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1441 EXPORT_SYMBOL(init_special_inode);