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/ima.h>
21 #include <linux/pagemap.h>
22 #include <linux/cdev.h>
23 #include <linux/bootmem.h>
24 #include <linux/inotify.h>
25 #include <linux/fsnotify.h>
26 #include <linux/mount.h>
27 #include <linux/async.h>
28 #include <linux/posix_acl.h>
31 * This is needed for the following functions:
33 * - invalidate_inode_buffers
36 * FIXME: remove all knowledge of the buffer layer from this file
38 #include <linux/buffer_head.h>
41 * New inode.c implementation.
43 * This implementation has the basic premise of trying
44 * to be extremely low-overhead and SMP-safe, yet be
45 * simple enough to be "obviously correct".
50 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
52 /* #define INODE_PARANOIA 1 */
53 /* #define INODE_DEBUG 1 */
56 * Inode lookup is no longer as critical as it used to be:
57 * most of the lookups are going to be through the dcache.
59 #define I_HASHBITS i_hash_shift
60 #define I_HASHMASK i_hash_mask
62 static unsigned int i_hash_mask __read_mostly;
63 static unsigned int i_hash_shift __read_mostly;
66 * Each inode can be on two separate lists. One is
67 * the hash list of the inode, used for lookups. The
68 * other linked list is the "type" list:
69 * "in_use" - valid inode, i_count > 0, i_nlink > 0
70 * "dirty" - as "in_use" but also dirty
71 * "unused" - valid inode, i_count = 0
73 * A "dirty" list is maintained for each super block,
74 * allowing for low-overhead inode sync() operations.
77 LIST_HEAD(inode_in_use);
78 LIST_HEAD(inode_unused);
79 static struct hlist_head *inode_hashtable __read_mostly;
82 * A simple spinlock to protect the list manipulations.
84 * NOTE! You also have to own the lock if you change
85 * the i_state of an inode while it is in use..
87 DEFINE_SPINLOCK(inode_lock);
90 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
91 * icache shrinking path, and the umount path. Without this exclusion,
92 * by the time prune_icache calls iput for the inode whose pages it has
93 * been invalidating, or by the time it calls clear_inode & destroy_inode
94 * from its final dispose_list, the struct super_block they refer to
95 * (for inode->i_sb->s_op) may already have been freed and reused.
97 static DEFINE_MUTEX(iprune_mutex);
100 * Statistics gathering..
102 struct inodes_stat_t inodes_stat;
104 static struct kmem_cache *inode_cachep __read_mostly;
106 static void wake_up_inode(struct inode *inode)
109 * Prevent speculative execution through spin_unlock(&inode_lock);
112 wake_up_bit(&inode->i_state, __I_LOCK);
116 * inode_init_always - perform inode structure intialisation
117 * @sb: superblock inode belongs to
118 * @inode: inode to initialise
120 * These are initializations that need to be done on every inode
121 * allocation as the fields are not initialised by slab allocation.
123 int inode_init_always(struct super_block *sb, struct inode *inode)
125 static const struct address_space_operations empty_aops;
126 static struct inode_operations empty_iops;
127 static const struct file_operations empty_fops;
128 struct address_space *const mapping = &inode->i_data;
131 inode->i_blkbits = sb->s_blocksize_bits;
133 atomic_set(&inode->i_count, 1);
134 inode->i_op = &empty_iops;
135 inode->i_fop = &empty_fops;
139 atomic_set(&inode->i_writecount, 0);
143 inode->i_generation = 0;
145 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
147 inode->i_pipe = NULL;
148 inode->i_bdev = NULL;
149 inode->i_cdev = NULL;
151 inode->dirtied_when = 0;
153 if (security_inode_alloc(inode))
156 /* allocate and initialize an i_integrity */
157 if (ima_inode_alloc(inode))
158 goto out_free_security;
160 spin_lock_init(&inode->i_lock);
161 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
163 mutex_init(&inode->i_mutex);
164 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
166 init_rwsem(&inode->i_alloc_sem);
167 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
169 mapping->a_ops = &empty_aops;
170 mapping->host = inode;
172 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
173 mapping->assoc_mapping = NULL;
174 mapping->backing_dev_info = &default_backing_dev_info;
175 mapping->writeback_index = 0;
178 * If the block_device provides a backing_dev_info for client
179 * inodes then use that. Otherwise the inode share the bdev's
183 struct backing_dev_info *bdi;
185 bdi = sb->s_bdev->bd_inode_backing_dev_info;
187 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
188 mapping->backing_dev_info = bdi;
190 inode->i_private = NULL;
191 inode->i_mapping = mapping;
192 #ifdef CONFIG_FS_POSIX_ACL
193 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
196 #ifdef CONFIG_FSNOTIFY
197 inode->i_fsnotify_mask = 0;
203 security_inode_free(inode);
207 EXPORT_SYMBOL(inode_init_always);
209 static struct inode *alloc_inode(struct super_block *sb)
213 if (sb->s_op->alloc_inode)
214 inode = sb->s_op->alloc_inode(sb);
216 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
221 if (unlikely(inode_init_always(sb, inode))) {
222 if (inode->i_sb->s_op->destroy_inode)
223 inode->i_sb->s_op->destroy_inode(inode);
225 kmem_cache_free(inode_cachep, inode);
232 void destroy_inode(struct inode *inode)
234 BUG_ON(inode_has_buffers(inode));
235 ima_inode_free(inode);
236 security_inode_free(inode);
237 fsnotify_inode_delete(inode);
238 #ifdef CONFIG_FS_POSIX_ACL
239 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
240 posix_acl_release(inode->i_acl);
241 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
242 posix_acl_release(inode->i_default_acl);
244 if (inode->i_sb->s_op->destroy_inode)
245 inode->i_sb->s_op->destroy_inode(inode);
247 kmem_cache_free(inode_cachep, (inode));
249 EXPORT_SYMBOL(destroy_inode);
253 * These are initializations that only need to be done
254 * once, because the fields are idempotent across use
255 * of the inode, so let the slab aware of that.
257 void inode_init_once(struct inode *inode)
259 memset(inode, 0, sizeof(*inode));
260 INIT_HLIST_NODE(&inode->i_hash);
261 INIT_LIST_HEAD(&inode->i_dentry);
262 INIT_LIST_HEAD(&inode->i_devices);
263 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
264 spin_lock_init(&inode->i_data.tree_lock);
265 spin_lock_init(&inode->i_data.i_mmap_lock);
266 INIT_LIST_HEAD(&inode->i_data.private_list);
267 spin_lock_init(&inode->i_data.private_lock);
268 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
269 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
270 i_size_ordered_init(inode);
271 #ifdef CONFIG_INOTIFY
272 INIT_LIST_HEAD(&inode->inotify_watches);
273 mutex_init(&inode->inotify_mutex);
275 #ifdef CONFIG_FSNOTIFY
276 INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries);
279 EXPORT_SYMBOL(inode_init_once);
281 static void init_once(void *foo)
283 struct inode *inode = (struct inode *) foo;
285 inode_init_once(inode);
289 * inode_lock must be held
291 void __iget(struct inode *inode)
293 if (atomic_read(&inode->i_count)) {
294 atomic_inc(&inode->i_count);
297 atomic_inc(&inode->i_count);
298 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
299 list_move(&inode->i_list, &inode_in_use);
300 inodes_stat.nr_unused--;
304 * clear_inode - clear an inode
305 * @inode: inode to clear
307 * This is called by the filesystem to tell us
308 * that the inode is no longer useful. We just
309 * terminate it with extreme prejudice.
311 void clear_inode(struct inode *inode)
314 invalidate_inode_buffers(inode);
316 BUG_ON(inode->i_data.nrpages);
317 BUG_ON(!(inode->i_state & I_FREEING));
318 BUG_ON(inode->i_state & I_CLEAR);
319 inode_sync_wait(inode);
321 if (inode->i_sb->s_op->clear_inode)
322 inode->i_sb->s_op->clear_inode(inode);
323 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
325 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
327 inode->i_state = I_CLEAR;
329 EXPORT_SYMBOL(clear_inode);
332 * dispose_list - dispose of the contents of a local list
333 * @head: the head of the list to free
335 * Dispose-list gets a local list with local inodes in it, so it doesn't
336 * need to worry about list corruption and SMP locks.
338 static void dispose_list(struct list_head *head)
342 while (!list_empty(head)) {
345 inode = list_first_entry(head, struct inode, i_list);
346 list_del(&inode->i_list);
348 if (inode->i_data.nrpages)
349 truncate_inode_pages(&inode->i_data, 0);
352 spin_lock(&inode_lock);
353 hlist_del_init(&inode->i_hash);
354 list_del_init(&inode->i_sb_list);
355 spin_unlock(&inode_lock);
357 wake_up_inode(inode);
358 destroy_inode(inode);
361 spin_lock(&inode_lock);
362 inodes_stat.nr_inodes -= nr_disposed;
363 spin_unlock(&inode_lock);
367 * Invalidate all inodes for a device.
369 static int invalidate_list(struct list_head *head, struct list_head *dispose)
371 struct list_head *next;
372 int busy = 0, count = 0;
376 struct list_head *tmp = next;
380 * We can reschedule here without worrying about the list's
381 * consistency because the per-sb list of inodes must not
382 * change during umount anymore, and because iprune_mutex keeps
383 * shrink_icache_memory() away.
385 cond_resched_lock(&inode_lock);
390 inode = list_entry(tmp, struct inode, i_sb_list);
391 if (inode->i_state & I_NEW)
393 invalidate_inode_buffers(inode);
394 if (!atomic_read(&inode->i_count)) {
395 list_move(&inode->i_list, dispose);
396 WARN_ON(inode->i_state & I_NEW);
397 inode->i_state |= I_FREEING;
403 /* only unused inodes may be cached with i_count zero */
404 inodes_stat.nr_unused -= count;
409 * invalidate_inodes - discard the inodes on a device
412 * Discard all of the inodes for a given superblock. If the discard
413 * fails because there are busy inodes then a non zero value is returned.
414 * If the discard is successful all the inodes have been discarded.
416 int invalidate_inodes(struct super_block *sb)
419 LIST_HEAD(throw_away);
421 mutex_lock(&iprune_mutex);
422 spin_lock(&inode_lock);
423 inotify_unmount_inodes(&sb->s_inodes);
424 fsnotify_unmount_inodes(&sb->s_inodes);
425 busy = invalidate_list(&sb->s_inodes, &throw_away);
426 spin_unlock(&inode_lock);
428 dispose_list(&throw_away);
429 mutex_unlock(&iprune_mutex);
433 EXPORT_SYMBOL(invalidate_inodes);
435 static int can_unuse(struct inode *inode)
439 if (inode_has_buffers(inode))
441 if (atomic_read(&inode->i_count))
443 if (inode->i_data.nrpages)
449 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
450 * a temporary list and then are freed outside inode_lock by dispose_list().
452 * Any inodes which are pinned purely because of attached pagecache have their
453 * pagecache removed. We expect the final iput() on that inode to add it to
454 * the front of the inode_unused list. So look for it there and if the
455 * inode is still freeable, proceed. The right inode is found 99.9% of the
456 * time in testing on a 4-way.
458 * If the inode has metadata buffers attached to mapping->private_list then
459 * try to remove them.
461 static void prune_icache(int nr_to_scan)
466 unsigned long reap = 0;
468 mutex_lock(&iprune_mutex);
469 spin_lock(&inode_lock);
470 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
473 if (list_empty(&inode_unused))
476 inode = list_entry(inode_unused.prev, struct inode, i_list);
478 if (inode->i_state || atomic_read(&inode->i_count)) {
479 list_move(&inode->i_list, &inode_unused);
482 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
484 spin_unlock(&inode_lock);
485 if (remove_inode_buffers(inode))
486 reap += invalidate_mapping_pages(&inode->i_data,
489 spin_lock(&inode_lock);
491 if (inode != list_entry(inode_unused.next,
492 struct inode, i_list))
493 continue; /* wrong inode or list_empty */
494 if (!can_unuse(inode))
497 list_move(&inode->i_list, &freeable);
498 WARN_ON(inode->i_state & I_NEW);
499 inode->i_state |= I_FREEING;
502 inodes_stat.nr_unused -= nr_pruned;
503 if (current_is_kswapd())
504 __count_vm_events(KSWAPD_INODESTEAL, reap);
506 __count_vm_events(PGINODESTEAL, reap);
507 spin_unlock(&inode_lock);
509 dispose_list(&freeable);
510 mutex_unlock(&iprune_mutex);
514 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
515 * "unused" means that no dentries are referring to the inodes: the files are
516 * not open and the dcache references to those inodes have already been
519 * This function is passed the number of inodes to scan, and it returns the
520 * total number of remaining possibly-reclaimable inodes.
522 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
526 * Nasty deadlock avoidance. We may hold various FS locks,
527 * and we don't want to recurse into the FS that called us
528 * in clear_inode() and friends..
530 if (!(gfp_mask & __GFP_FS))
534 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
537 static struct shrinker icache_shrinker = {
538 .shrink = shrink_icache_memory,
539 .seeks = DEFAULT_SEEKS,
542 static void __wait_on_freeing_inode(struct inode *inode);
544 * Called with the inode lock held.
545 * NOTE: we are not increasing the inode-refcount, you must call __iget()
546 * by hand after calling find_inode now! This simplifies iunique and won't
547 * add any additional branch in the common code.
549 static struct inode *find_inode(struct super_block *sb,
550 struct hlist_head *head,
551 int (*test)(struct inode *, void *),
554 struct hlist_node *node;
555 struct inode *inode = NULL;
558 hlist_for_each_entry(inode, node, head, i_hash) {
559 if (inode->i_sb != sb)
561 if (!test(inode, data))
563 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
564 __wait_on_freeing_inode(inode);
569 return node ? inode : NULL;
573 * find_inode_fast is the fast path version of find_inode, see the comment at
574 * iget_locked for details.
576 static struct inode *find_inode_fast(struct super_block *sb,
577 struct hlist_head *head, unsigned long ino)
579 struct hlist_node *node;
580 struct inode *inode = NULL;
583 hlist_for_each_entry(inode, node, head, i_hash) {
584 if (inode->i_ino != ino)
586 if (inode->i_sb != sb)
588 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
589 __wait_on_freeing_inode(inode);
594 return node ? inode : NULL;
597 static unsigned long hash(struct super_block *sb, unsigned long hashval)
601 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
603 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
604 return tmp & I_HASHMASK;
608 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
611 inodes_stat.nr_inodes++;
612 list_add(&inode->i_list, &inode_in_use);
613 list_add(&inode->i_sb_list, &sb->s_inodes);
615 hlist_add_head(&inode->i_hash, head);
619 * inode_add_to_lists - add a new inode to relevant lists
620 * @sb: superblock inode belongs to
621 * @inode: inode to mark in use
623 * When an inode is allocated it needs to be accounted for, added to the in use
624 * list, the owning superblock and the inode hash. This needs to be done under
625 * the inode_lock, so export a function to do this rather than the inode lock
626 * itself. We calculate the hash list to add to here so it is all internal
627 * which requires the caller to have already set up the inode number in the
630 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
632 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
634 spin_lock(&inode_lock);
635 __inode_add_to_lists(sb, head, inode);
636 spin_unlock(&inode_lock);
638 EXPORT_SYMBOL_GPL(inode_add_to_lists);
641 * new_inode - obtain an inode
644 * Allocates a new inode for given superblock. The default gfp_mask
645 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
646 * If HIGHMEM pages are unsuitable or it is known that pages allocated
647 * for the page cache are not reclaimable or migratable,
648 * mapping_set_gfp_mask() must be called with suitable flags on the
649 * newly created inode's mapping
652 struct inode *new_inode(struct super_block *sb)
655 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
656 * error if st_ino won't fit in target struct field. Use 32bit counter
657 * here to attempt to avoid that.
659 static unsigned int last_ino;
662 spin_lock_prefetch(&inode_lock);
664 inode = alloc_inode(sb);
666 spin_lock(&inode_lock);
667 __inode_add_to_lists(sb, NULL, inode);
668 inode->i_ino = ++last_ino;
670 spin_unlock(&inode_lock);
674 EXPORT_SYMBOL(new_inode);
676 void unlock_new_inode(struct inode *inode)
678 #ifdef CONFIG_DEBUG_LOCK_ALLOC
679 if (inode->i_mode & S_IFDIR) {
680 struct file_system_type *type = inode->i_sb->s_type;
682 /* Set new key only if filesystem hasn't already changed it */
683 if (!lockdep_match_class(&inode->i_mutex,
684 &type->i_mutex_key)) {
686 * ensure nobody is actually holding i_mutex
688 mutex_destroy(&inode->i_mutex);
689 mutex_init(&inode->i_mutex);
690 lockdep_set_class(&inode->i_mutex,
691 &type->i_mutex_dir_key);
696 * This is special! We do not need the spinlock
697 * when clearing I_LOCK, because we're guaranteed
698 * that nobody else tries to do anything about the
699 * state of the inode when it is locked, as we
700 * just created it (so there can be no old holders
701 * that haven't tested I_LOCK).
703 WARN_ON((inode->i_state & (I_LOCK|I_NEW)) != (I_LOCK|I_NEW));
704 inode->i_state &= ~(I_LOCK|I_NEW);
705 wake_up_inode(inode);
707 EXPORT_SYMBOL(unlock_new_inode);
710 * This is called without the inode lock held.. Be careful.
712 * We no longer cache the sb_flags in i_flags - see fs.h
713 * -- rmk@arm.uk.linux.org
715 static struct inode *get_new_inode(struct super_block *sb,
716 struct hlist_head *head,
717 int (*test)(struct inode *, void *),
718 int (*set)(struct inode *, void *),
723 inode = alloc_inode(sb);
727 spin_lock(&inode_lock);
728 /* We released the lock, so.. */
729 old = find_inode(sb, head, test, data);
731 if (set(inode, data))
734 __inode_add_to_lists(sb, head, inode);
735 inode->i_state = I_LOCK|I_NEW;
736 spin_unlock(&inode_lock);
738 /* Return the locked inode with I_NEW set, the
739 * caller is responsible for filling in the contents
745 * Uhhuh, somebody else created the same inode under
746 * us. Use the old inode instead of the one we just
750 spin_unlock(&inode_lock);
751 destroy_inode(inode);
753 wait_on_inode(inode);
758 spin_unlock(&inode_lock);
759 destroy_inode(inode);
764 * get_new_inode_fast is the fast path version of get_new_inode, see the
765 * comment at iget_locked for details.
767 static struct inode *get_new_inode_fast(struct super_block *sb,
768 struct hlist_head *head, unsigned long ino)
772 inode = alloc_inode(sb);
776 spin_lock(&inode_lock);
777 /* We released the lock, so.. */
778 old = find_inode_fast(sb, head, ino);
781 __inode_add_to_lists(sb, head, inode);
782 inode->i_state = I_LOCK|I_NEW;
783 spin_unlock(&inode_lock);
785 /* Return the locked inode with I_NEW set, the
786 * caller is responsible for filling in the contents
792 * Uhhuh, somebody else created the same inode under
793 * us. Use the old inode instead of the one we just
797 spin_unlock(&inode_lock);
798 destroy_inode(inode);
800 wait_on_inode(inode);
806 * iunique - get a unique inode number
808 * @max_reserved: highest reserved inode number
810 * Obtain an inode number that is unique on the system for a given
811 * superblock. This is used by file systems that have no natural
812 * permanent inode numbering system. An inode number is returned that
813 * is higher than the reserved limit but unique.
816 * With a large number of inodes live on the file system this function
817 * currently becomes quite slow.
819 ino_t iunique(struct super_block *sb, ino_t max_reserved)
822 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
823 * error if st_ino won't fit in target struct field. Use 32bit counter
824 * here to attempt to avoid that.
826 static unsigned int counter;
828 struct hlist_head *head;
831 spin_lock(&inode_lock);
833 if (counter <= max_reserved)
834 counter = max_reserved + 1;
836 head = inode_hashtable + hash(sb, res);
837 inode = find_inode_fast(sb, head, res);
838 } while (inode != NULL);
839 spin_unlock(&inode_lock);
843 EXPORT_SYMBOL(iunique);
845 struct inode *igrab(struct inode *inode)
847 spin_lock(&inode_lock);
848 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
852 * Handle the case where s_op->clear_inode is not been
853 * called yet, and somebody is calling igrab
854 * while the inode is getting freed.
857 spin_unlock(&inode_lock);
860 EXPORT_SYMBOL(igrab);
863 * ifind - internal function, you want ilookup5() or iget5().
864 * @sb: super block of file system to search
865 * @head: the head of the list to search
866 * @test: callback used for comparisons between inodes
867 * @data: opaque data pointer to pass to @test
868 * @wait: if true wait for the inode to be unlocked, if false do not
870 * ifind() searches for the inode specified by @data in the inode
871 * cache. This is a generalized version of ifind_fast() for file systems where
872 * the inode number is not sufficient for unique identification of an inode.
874 * If the inode is in the cache, the inode is returned with an incremented
877 * Otherwise NULL is returned.
879 * Note, @test is called with the inode_lock held, so can't sleep.
881 static struct inode *ifind(struct super_block *sb,
882 struct hlist_head *head, int (*test)(struct inode *, void *),
883 void *data, const int wait)
887 spin_lock(&inode_lock);
888 inode = find_inode(sb, head, test, data);
891 spin_unlock(&inode_lock);
893 wait_on_inode(inode);
896 spin_unlock(&inode_lock);
901 * ifind_fast - internal function, you want ilookup() or iget().
902 * @sb: super block of file system to search
903 * @head: head of the list to search
904 * @ino: inode number to search for
906 * ifind_fast() searches for the inode @ino in the inode cache. This is for
907 * file systems where the inode number is sufficient for unique identification
910 * If the inode is in the cache, the inode is returned with an incremented
913 * Otherwise NULL is returned.
915 static struct inode *ifind_fast(struct super_block *sb,
916 struct hlist_head *head, unsigned long ino)
920 spin_lock(&inode_lock);
921 inode = find_inode_fast(sb, head, ino);
924 spin_unlock(&inode_lock);
925 wait_on_inode(inode);
928 spin_unlock(&inode_lock);
933 * ilookup5_nowait - search for an inode in the inode cache
934 * @sb: super block of file system to search
935 * @hashval: hash value (usually inode number) to search for
936 * @test: callback used for comparisons between inodes
937 * @data: opaque data pointer to pass to @test
939 * ilookup5() uses ifind() to search for the inode specified by @hashval and
940 * @data in the inode cache. This is a generalized version of ilookup() for
941 * file systems where the inode number is not sufficient for unique
942 * identification of an inode.
944 * If the inode is in the cache, the inode is returned with an incremented
945 * reference count. Note, the inode lock is not waited upon so you have to be
946 * very careful what you do with the returned inode. You probably should be
947 * using ilookup5() instead.
949 * Otherwise NULL is returned.
951 * Note, @test is called with the inode_lock held, so can't sleep.
953 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
954 int (*test)(struct inode *, void *), void *data)
956 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
958 return ifind(sb, head, test, data, 0);
960 EXPORT_SYMBOL(ilookup5_nowait);
963 * ilookup5 - search for an inode in the inode cache
964 * @sb: super block of file system to search
965 * @hashval: hash value (usually inode number) to search for
966 * @test: callback used for comparisons between inodes
967 * @data: opaque data pointer to pass to @test
969 * ilookup5() uses ifind() to search for the inode specified by @hashval and
970 * @data in the inode cache. This is a generalized version of ilookup() for
971 * file systems where the inode number is not sufficient for unique
972 * identification of an inode.
974 * If the inode is in the cache, the inode lock is waited upon and the inode is
975 * returned with an incremented reference count.
977 * Otherwise NULL is returned.
979 * Note, @test is called with the inode_lock held, so can't sleep.
981 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
982 int (*test)(struct inode *, void *), void *data)
984 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
986 return ifind(sb, head, test, data, 1);
988 EXPORT_SYMBOL(ilookup5);
991 * ilookup - search for an inode in the inode cache
992 * @sb: super block of file system to search
993 * @ino: inode number to search for
995 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
996 * This is for file systems where the inode number is sufficient for unique
997 * identification of an inode.
999 * If the inode is in the cache, the inode is returned with an incremented
1002 * Otherwise NULL is returned.
1004 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1006 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1008 return ifind_fast(sb, head, ino);
1010 EXPORT_SYMBOL(ilookup);
1013 * iget5_locked - obtain an inode from a mounted file system
1014 * @sb: super block of file system
1015 * @hashval: hash value (usually inode number) to get
1016 * @test: callback used for comparisons between inodes
1017 * @set: callback used to initialize a new struct inode
1018 * @data: opaque data pointer to pass to @test and @set
1020 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1021 * and @data in the inode cache and if present it is returned with an increased
1022 * reference count. This is a generalized version of iget_locked() for file
1023 * systems where the inode number is not sufficient for unique identification
1026 * If the inode is not in cache, get_new_inode() is called to allocate a new
1027 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1028 * file system gets to fill it in before unlocking it via unlock_new_inode().
1030 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1032 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1033 int (*test)(struct inode *, void *),
1034 int (*set)(struct inode *, void *), void *data)
1036 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1037 struct inode *inode;
1039 inode = ifind(sb, head, test, data, 1);
1043 * get_new_inode() will do the right thing, re-trying the search
1044 * in case it had to block at any point.
1046 return get_new_inode(sb, head, test, set, data);
1048 EXPORT_SYMBOL(iget5_locked);
1051 * iget_locked - obtain an inode from a mounted file system
1052 * @sb: super block of file system
1053 * @ino: inode number to get
1055 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1056 * the inode cache and if present it is returned with an increased reference
1057 * count. This is for file systems where the inode number is sufficient for
1058 * unique identification of an inode.
1060 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1061 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1062 * The file system gets to fill it in before unlocking it via
1063 * unlock_new_inode().
1065 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1067 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1068 struct inode *inode;
1070 inode = ifind_fast(sb, head, ino);
1074 * get_new_inode_fast() will do the right thing, re-trying the search
1075 * in case it had to block at any point.
1077 return get_new_inode_fast(sb, head, ino);
1079 EXPORT_SYMBOL(iget_locked);
1081 int insert_inode_locked(struct inode *inode)
1083 struct super_block *sb = inode->i_sb;
1084 ino_t ino = inode->i_ino;
1085 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1087 inode->i_state |= I_LOCK|I_NEW;
1089 struct hlist_node *node;
1090 struct inode *old = NULL;
1091 spin_lock(&inode_lock);
1092 hlist_for_each_entry(old, node, head, i_hash) {
1093 if (old->i_ino != ino)
1095 if (old->i_sb != sb)
1097 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1101 if (likely(!node)) {
1102 hlist_add_head(&inode->i_hash, head);
1103 spin_unlock(&inode_lock);
1107 spin_unlock(&inode_lock);
1109 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1116 EXPORT_SYMBOL(insert_inode_locked);
1118 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1119 int (*test)(struct inode *, void *), void *data)
1121 struct super_block *sb = inode->i_sb;
1122 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1124 inode->i_state |= I_LOCK|I_NEW;
1127 struct hlist_node *node;
1128 struct inode *old = NULL;
1130 spin_lock(&inode_lock);
1131 hlist_for_each_entry(old, node, head, i_hash) {
1132 if (old->i_sb != sb)
1134 if (!test(old, data))
1136 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1140 if (likely(!node)) {
1141 hlist_add_head(&inode->i_hash, head);
1142 spin_unlock(&inode_lock);
1146 spin_unlock(&inode_lock);
1148 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1155 EXPORT_SYMBOL(insert_inode_locked4);
1158 * __insert_inode_hash - hash an inode
1159 * @inode: unhashed inode
1160 * @hashval: unsigned long value used to locate this object in the
1163 * Add an inode to the inode hash for this superblock.
1165 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1167 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1168 spin_lock(&inode_lock);
1169 hlist_add_head(&inode->i_hash, head);
1170 spin_unlock(&inode_lock);
1172 EXPORT_SYMBOL(__insert_inode_hash);
1175 * remove_inode_hash - remove an inode from the hash
1176 * @inode: inode to unhash
1178 * Remove an inode from the superblock.
1180 void remove_inode_hash(struct inode *inode)
1182 spin_lock(&inode_lock);
1183 hlist_del_init(&inode->i_hash);
1184 spin_unlock(&inode_lock);
1186 EXPORT_SYMBOL(remove_inode_hash);
1189 * Tell the filesystem that this inode is no longer of any interest and should
1190 * be completely destroyed.
1192 * We leave the inode in the inode hash table until *after* the filesystem's
1193 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1194 * instigate) will always find up-to-date information either in the hash or on
1197 * I_FREEING is set so that no-one will take a new reference to the inode while
1198 * it is being deleted.
1200 void generic_delete_inode(struct inode *inode)
1202 const struct super_operations *op = inode->i_sb->s_op;
1204 list_del_init(&inode->i_list);
1205 list_del_init(&inode->i_sb_list);
1206 WARN_ON(inode->i_state & I_NEW);
1207 inode->i_state |= I_FREEING;
1208 inodes_stat.nr_inodes--;
1209 spin_unlock(&inode_lock);
1211 security_inode_delete(inode);
1213 if (op->delete_inode) {
1214 void (*delete)(struct inode *) = op->delete_inode;
1215 if (!is_bad_inode(inode))
1217 /* Filesystems implementing their own
1218 * s_op->delete_inode are required to call
1219 * truncate_inode_pages and clear_inode()
1223 truncate_inode_pages(&inode->i_data, 0);
1226 spin_lock(&inode_lock);
1227 hlist_del_init(&inode->i_hash);
1228 spin_unlock(&inode_lock);
1229 wake_up_inode(inode);
1230 BUG_ON(inode->i_state != I_CLEAR);
1231 destroy_inode(inode);
1233 EXPORT_SYMBOL(generic_delete_inode);
1235 static void generic_forget_inode(struct inode *inode)
1237 struct super_block *sb = inode->i_sb;
1239 if (!hlist_unhashed(&inode->i_hash)) {
1240 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1241 list_move(&inode->i_list, &inode_unused);
1242 inodes_stat.nr_unused++;
1243 if (sb->s_flags & MS_ACTIVE) {
1244 spin_unlock(&inode_lock);
1247 WARN_ON(inode->i_state & I_NEW);
1248 inode->i_state |= I_WILL_FREE;
1249 spin_unlock(&inode_lock);
1250 write_inode_now(inode, 1);
1251 spin_lock(&inode_lock);
1252 WARN_ON(inode->i_state & I_NEW);
1253 inode->i_state &= ~I_WILL_FREE;
1254 inodes_stat.nr_unused--;
1255 hlist_del_init(&inode->i_hash);
1257 list_del_init(&inode->i_list);
1258 list_del_init(&inode->i_sb_list);
1259 WARN_ON(inode->i_state & I_NEW);
1260 inode->i_state |= I_FREEING;
1261 inodes_stat.nr_inodes--;
1262 spin_unlock(&inode_lock);
1263 if (inode->i_data.nrpages)
1264 truncate_inode_pages(&inode->i_data, 0);
1266 wake_up_inode(inode);
1267 destroy_inode(inode);
1271 * Normal UNIX filesystem behaviour: delete the
1272 * inode when the usage count drops to zero, and
1275 void generic_drop_inode(struct inode *inode)
1277 if (!inode->i_nlink)
1278 generic_delete_inode(inode);
1280 generic_forget_inode(inode);
1282 EXPORT_SYMBOL_GPL(generic_drop_inode);
1285 * Called when we're dropping the last reference
1288 * Call the FS "drop()" function, defaulting to
1289 * the legacy UNIX filesystem behaviour..
1291 * NOTE! NOTE! NOTE! We're called with the inode lock
1292 * held, and the drop function is supposed to release
1295 static inline void iput_final(struct inode *inode)
1297 const struct super_operations *op = inode->i_sb->s_op;
1298 void (*drop)(struct inode *) = generic_drop_inode;
1300 if (op && op->drop_inode)
1301 drop = op->drop_inode;
1306 * iput - put an inode
1307 * @inode: inode to put
1309 * Puts an inode, dropping its usage count. If the inode use count hits
1310 * zero, the inode is then freed and may also be destroyed.
1312 * Consequently, iput() can sleep.
1314 void iput(struct inode *inode)
1317 BUG_ON(inode->i_state == I_CLEAR);
1319 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1323 EXPORT_SYMBOL(iput);
1326 * bmap - find a block number in a file
1327 * @inode: inode of file
1328 * @block: block to find
1330 * Returns the block number on the device holding the inode that
1331 * is the disk block number for the block of the file requested.
1332 * That is, asked for block 4 of inode 1 the function will return the
1333 * disk block relative to the disk start that holds that block of the
1336 sector_t bmap(struct inode *inode, sector_t block)
1339 if (inode->i_mapping->a_ops->bmap)
1340 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1343 EXPORT_SYMBOL(bmap);
1346 * With relative atime, only update atime if the previous atime is
1347 * earlier than either the ctime or mtime or if at least a day has
1348 * passed since the last atime update.
1350 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1351 struct timespec now)
1354 if (!(mnt->mnt_flags & MNT_RELATIME))
1357 * Is mtime younger than atime? If yes, update atime:
1359 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1362 * Is ctime younger than atime? If yes, update atime:
1364 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1368 * Is the previous atime value older than a day? If yes,
1371 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1374 * Good, we can skip the atime update:
1380 * touch_atime - update the access time
1381 * @mnt: mount the inode is accessed on
1382 * @dentry: dentry accessed
1384 * Update the accessed time on an inode and mark it for writeback.
1385 * This function automatically handles read only file systems and media,
1386 * as well as the "noatime" flag and inode specific "noatime" markers.
1388 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1390 struct inode *inode = dentry->d_inode;
1391 struct timespec now;
1393 if (mnt_want_write(mnt))
1395 if (inode->i_flags & S_NOATIME)
1397 if (IS_NOATIME(inode))
1399 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1402 if (mnt->mnt_flags & MNT_NOATIME)
1404 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1407 now = current_fs_time(inode->i_sb);
1409 if (!relatime_need_update(mnt, inode, now))
1412 if (timespec_equal(&inode->i_atime, &now))
1415 inode->i_atime = now;
1416 mark_inode_dirty_sync(inode);
1418 mnt_drop_write(mnt);
1420 EXPORT_SYMBOL(touch_atime);
1423 * file_update_time - update mtime and ctime time
1424 * @file: file accessed
1426 * Update the mtime and ctime members of an inode and mark the inode
1427 * for writeback. Note that this function is meant exclusively for
1428 * usage in the file write path of filesystems, and filesystems may
1429 * choose to explicitly ignore update via this function with the
1430 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1431 * timestamps are handled by the server.
1434 void file_update_time(struct file *file)
1436 struct inode *inode = file->f_path.dentry->d_inode;
1437 struct timespec now;
1441 if (IS_NOCMTIME(inode))
1444 err = mnt_want_write_file(file);
1448 now = current_fs_time(inode->i_sb);
1449 if (!timespec_equal(&inode->i_mtime, &now)) {
1450 inode->i_mtime = now;
1454 if (!timespec_equal(&inode->i_ctime, &now)) {
1455 inode->i_ctime = now;
1459 if (IS_I_VERSION(inode)) {
1460 inode_inc_iversion(inode);
1465 mark_inode_dirty_sync(inode);
1466 mnt_drop_write(file->f_path.mnt);
1468 EXPORT_SYMBOL(file_update_time);
1470 int inode_needs_sync(struct inode *inode)
1474 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1478 EXPORT_SYMBOL(inode_needs_sync);
1480 int inode_wait(void *word)
1485 EXPORT_SYMBOL(inode_wait);
1488 * If we try to find an inode in the inode hash while it is being
1489 * deleted, we have to wait until the filesystem completes its
1490 * deletion before reporting that it isn't found. This function waits
1491 * until the deletion _might_ have completed. Callers are responsible
1492 * to recheck inode state.
1494 * It doesn't matter if I_LOCK is not set initially, a call to
1495 * wake_up_inode() after removing from the hash list will DTRT.
1497 * This is called with inode_lock held.
1499 static void __wait_on_freeing_inode(struct inode *inode)
1501 wait_queue_head_t *wq;
1502 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1503 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1504 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1505 spin_unlock(&inode_lock);
1507 finish_wait(wq, &wait.wait);
1508 spin_lock(&inode_lock);
1511 static __initdata unsigned long ihash_entries;
1512 static int __init set_ihash_entries(char *str)
1516 ihash_entries = simple_strtoul(str, &str, 0);
1519 __setup("ihash_entries=", set_ihash_entries);
1522 * Initialize the waitqueues and inode hash table.
1524 void __init inode_init_early(void)
1528 /* If hashes are distributed across NUMA nodes, defer
1529 * hash allocation until vmalloc space is available.
1535 alloc_large_system_hash("Inode-cache",
1536 sizeof(struct hlist_head),
1544 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1545 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1548 void __init inode_init(void)
1552 /* inode slab cache */
1553 inode_cachep = kmem_cache_create("inode_cache",
1554 sizeof(struct inode),
1556 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1559 register_shrinker(&icache_shrinker);
1561 /* Hash may have been set up in inode_init_early */
1566 alloc_large_system_hash("Inode-cache",
1567 sizeof(struct hlist_head),
1575 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1576 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1579 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1581 inode->i_mode = mode;
1582 if (S_ISCHR(mode)) {
1583 inode->i_fop = &def_chr_fops;
1584 inode->i_rdev = rdev;
1585 } else if (S_ISBLK(mode)) {
1586 inode->i_fop = &def_blk_fops;
1587 inode->i_rdev = rdev;
1588 } else if (S_ISFIFO(mode))
1589 inode->i_fop = &def_fifo_fops;
1590 else if (S_ISSOCK(mode))
1591 inode->i_fop = &bad_sock_fops;
1593 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1596 EXPORT_SYMBOL(init_special_inode);