4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/slab.h>
12 #include <linux/writeback.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/wait.h>
16 #include <linux/rwsem.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/fsnotify.h>
25 #include <linux/mount.h>
26 #include <linux/async.h>
27 #include <linux/posix_acl.h>
30 * This is needed for the following functions:
32 * - invalidate_inode_buffers
35 * FIXME: remove all knowledge of the buffer layer from this file
37 #include <linux/buffer_head.h>
40 * New inode.c implementation.
42 * This implementation has the basic premise of trying
43 * to be extremely low-overhead and SMP-safe, yet be
44 * simple enough to be "obviously correct".
49 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
51 /* #define INODE_PARANOIA 1 */
52 /* #define INODE_DEBUG 1 */
55 * Inode lookup is no longer as critical as it used to be:
56 * most of the lookups are going to be through the dcache.
58 #define I_HASHBITS i_hash_shift
59 #define I_HASHMASK i_hash_mask
61 static unsigned int i_hash_mask __read_mostly;
62 static unsigned int i_hash_shift __read_mostly;
65 * Each inode can be on two separate lists. One is
66 * the hash list of the inode, used for lookups. The
67 * other linked list is the "type" list:
68 * "in_use" - valid inode, i_count > 0, i_nlink > 0
69 * "dirty" - as "in_use" but also dirty
70 * "unused" - valid inode, i_count = 0
72 * A "dirty" list is maintained for each super block,
73 * allowing for low-overhead inode sync() operations.
76 LIST_HEAD(inode_in_use);
77 LIST_HEAD(inode_unused);
78 static struct hlist_head *inode_hashtable __read_mostly;
81 * A simple spinlock to protect the list manipulations.
83 * NOTE! You also have to own the lock if you change
84 * the i_state of an inode while it is in use..
86 DEFINE_SPINLOCK(inode_lock);
89 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
90 * icache shrinking path, and the umount path. Without this exclusion,
91 * by the time prune_icache calls iput for the inode whose pages it has
92 * been invalidating, or by the time it calls clear_inode & destroy_inode
93 * from its final dispose_list, the struct super_block they refer to
94 * (for inode->i_sb->s_op) may already have been freed and reused.
96 * We make this an rwsem because the fastpath is icache shrinking. In
97 * some cases a filesystem may be doing a significant amount of work in
98 * its inode reclaim code, so this should improve parallelism.
100 static DECLARE_RWSEM(iprune_sem);
103 * Statistics gathering..
105 struct inodes_stat_t inodes_stat;
107 static struct kmem_cache *inode_cachep __read_mostly;
109 static void wake_up_inode(struct inode *inode)
112 * Prevent speculative execution through spin_unlock(&inode_lock);
115 wake_up_bit(&inode->i_state, __I_NEW);
119 * inode_init_always - perform inode structure intialisation
120 * @sb: superblock inode belongs to
121 * @inode: inode to initialise
123 * These are initializations that need to be done on every inode
124 * allocation as the fields are not initialised by slab allocation.
126 int inode_init_always(struct super_block *sb, struct inode *inode)
128 static const struct address_space_operations empty_aops;
129 static const struct inode_operations empty_iops;
130 static const struct file_operations empty_fops;
131 struct address_space *const mapping = &inode->i_data;
134 inode->i_blkbits = sb->s_blocksize_bits;
136 atomic_set(&inode->i_count, 1);
137 inode->i_op = &empty_iops;
138 inode->i_fop = &empty_fops;
142 atomic_set(&inode->i_writecount, 0);
146 inode->i_generation = 0;
148 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
150 inode->i_pipe = NULL;
151 inode->i_bdev = NULL;
152 inode->i_cdev = NULL;
154 inode->dirtied_when = 0;
156 if (security_inode_alloc(inode))
158 spin_lock_init(&inode->i_lock);
159 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
161 mutex_init(&inode->i_mutex);
162 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
164 init_rwsem(&inode->i_alloc_sem);
165 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
167 mapping->a_ops = &empty_aops;
168 mapping->host = inode;
170 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
171 mapping->assoc_mapping = NULL;
172 mapping->backing_dev_info = &default_backing_dev_info;
173 mapping->writeback_index = 0;
176 * If the block_device provides a backing_dev_info for client
177 * inodes then use that. Otherwise the inode share the bdev's
181 struct backing_dev_info *bdi;
183 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
184 mapping->backing_dev_info = bdi;
186 inode->i_private = NULL;
187 inode->i_mapping = mapping;
188 #ifdef CONFIG_FS_POSIX_ACL
189 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
192 #ifdef CONFIG_FSNOTIFY
193 inode->i_fsnotify_mask = 0;
200 EXPORT_SYMBOL(inode_init_always);
202 static struct inode *alloc_inode(struct super_block *sb)
206 if (sb->s_op->alloc_inode)
207 inode = sb->s_op->alloc_inode(sb);
209 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
214 if (unlikely(inode_init_always(sb, inode))) {
215 if (inode->i_sb->s_op->destroy_inode)
216 inode->i_sb->s_op->destroy_inode(inode);
218 kmem_cache_free(inode_cachep, inode);
225 void __destroy_inode(struct inode *inode)
227 BUG_ON(inode_has_buffers(inode));
228 security_inode_free(inode);
229 fsnotify_inode_delete(inode);
230 #ifdef CONFIG_FS_POSIX_ACL
231 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
232 posix_acl_release(inode->i_acl);
233 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
234 posix_acl_release(inode->i_default_acl);
237 EXPORT_SYMBOL(__destroy_inode);
239 void destroy_inode(struct inode *inode)
241 __destroy_inode(inode);
242 if (inode->i_sb->s_op->destroy_inode)
243 inode->i_sb->s_op->destroy_inode(inode);
245 kmem_cache_free(inode_cachep, (inode));
249 * These are initializations that only need to be done
250 * once, because the fields are idempotent across use
251 * of the inode, so let the slab aware of that.
253 void inode_init_once(struct inode *inode)
255 memset(inode, 0, sizeof(*inode));
256 INIT_HLIST_NODE(&inode->i_hash);
257 INIT_LIST_HEAD(&inode->i_dentry);
258 INIT_LIST_HEAD(&inode->i_devices);
259 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
260 spin_lock_init(&inode->i_data.tree_lock);
261 spin_lock_init(&inode->i_data.i_mmap_lock);
262 INIT_LIST_HEAD(&inode->i_data.private_list);
263 spin_lock_init(&inode->i_data.private_lock);
264 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
265 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
266 i_size_ordered_init(inode);
267 #ifdef CONFIG_INOTIFY
268 INIT_LIST_HEAD(&inode->inotify_watches);
269 mutex_init(&inode->inotify_mutex);
271 #ifdef CONFIG_FSNOTIFY
272 INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries);
275 EXPORT_SYMBOL(inode_init_once);
277 static void init_once(void *foo)
279 struct inode *inode = (struct inode *) foo;
281 inode_init_once(inode);
285 * inode_lock must be held
287 void __iget(struct inode *inode)
289 if (atomic_inc_return(&inode->i_count) != 1)
292 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
293 list_move(&inode->i_list, &inode_in_use);
294 inodes_stat.nr_unused--;
297 void end_writeback(struct inode *inode)
300 BUG_ON(inode->i_data.nrpages);
301 BUG_ON(!list_empty(&inode->i_data.private_list));
302 BUG_ON(!(inode->i_state & I_FREEING));
303 BUG_ON(inode->i_state & I_CLEAR);
304 inode_sync_wait(inode);
305 inode->i_state = I_FREEING | I_CLEAR;
307 EXPORT_SYMBOL(end_writeback);
309 static void evict(struct inode *inode)
311 const struct super_operations *op = inode->i_sb->s_op;
313 if (op->evict_inode) {
314 op->evict_inode(inode);
316 if (inode->i_data.nrpages)
317 truncate_inode_pages(&inode->i_data, 0);
318 invalidate_inode_buffers(inode);
319 end_writeback(inode);
321 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
323 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
328 * dispose_list - dispose of the contents of a local list
329 * @head: the head of the list to free
331 * Dispose-list gets a local list with local inodes in it, so it doesn't
332 * need to worry about list corruption and SMP locks.
334 static void dispose_list(struct list_head *head)
338 while (!list_empty(head)) {
341 inode = list_first_entry(head, struct inode, i_list);
342 list_del(&inode->i_list);
346 spin_lock(&inode_lock);
347 hlist_del_init(&inode->i_hash);
348 list_del_init(&inode->i_sb_list);
349 spin_unlock(&inode_lock);
351 wake_up_inode(inode);
352 destroy_inode(inode);
355 spin_lock(&inode_lock);
356 inodes_stat.nr_inodes -= nr_disposed;
357 spin_unlock(&inode_lock);
361 * Invalidate all inodes for a device.
363 static int invalidate_list(struct list_head *head, struct list_head *dispose)
365 struct list_head *next;
366 int busy = 0, count = 0;
370 struct list_head *tmp = next;
374 * We can reschedule here without worrying about the list's
375 * consistency because the per-sb list of inodes must not
376 * change during umount anymore, and because iprune_sem keeps
377 * shrink_icache_memory() away.
379 cond_resched_lock(&inode_lock);
384 inode = list_entry(tmp, struct inode, i_sb_list);
385 if (inode->i_state & I_NEW)
387 invalidate_inode_buffers(inode);
388 if (!atomic_read(&inode->i_count)) {
389 list_move(&inode->i_list, dispose);
390 WARN_ON(inode->i_state & I_NEW);
391 inode->i_state |= I_FREEING;
397 /* only unused inodes may be cached with i_count zero */
398 inodes_stat.nr_unused -= count;
403 * invalidate_inodes - discard the inodes on a device
406 * Discard all of the inodes for a given superblock. If the discard
407 * fails because there are busy inodes then a non zero value is returned.
408 * If the discard is successful all the inodes have been discarded.
410 int invalidate_inodes(struct super_block *sb)
413 LIST_HEAD(throw_away);
415 down_write(&iprune_sem);
416 spin_lock(&inode_lock);
417 inotify_unmount_inodes(&sb->s_inodes);
418 fsnotify_unmount_inodes(&sb->s_inodes);
419 busy = invalidate_list(&sb->s_inodes, &throw_away);
420 spin_unlock(&inode_lock);
422 dispose_list(&throw_away);
423 up_write(&iprune_sem);
427 EXPORT_SYMBOL(invalidate_inodes);
429 static int can_unuse(struct inode *inode)
433 if (inode_has_buffers(inode))
435 if (atomic_read(&inode->i_count))
437 if (inode->i_data.nrpages)
443 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
444 * a temporary list and then are freed outside inode_lock by dispose_list().
446 * Any inodes which are pinned purely because of attached pagecache have their
447 * pagecache removed. We expect the final iput() on that inode to add it to
448 * the front of the inode_unused list. So look for it there and if the
449 * inode is still freeable, proceed. The right inode is found 99.9% of the
450 * time in testing on a 4-way.
452 * If the inode has metadata buffers attached to mapping->private_list then
453 * try to remove them.
455 static void prune_icache(int nr_to_scan)
460 unsigned long reap = 0;
462 down_read(&iprune_sem);
463 spin_lock(&inode_lock);
464 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
467 if (list_empty(&inode_unused))
470 inode = list_entry(inode_unused.prev, struct inode, i_list);
472 if (inode->i_state || atomic_read(&inode->i_count)) {
473 list_move(&inode->i_list, &inode_unused);
476 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
478 spin_unlock(&inode_lock);
479 if (remove_inode_buffers(inode))
480 reap += invalidate_mapping_pages(&inode->i_data,
483 spin_lock(&inode_lock);
485 if (inode != list_entry(inode_unused.next,
486 struct inode, i_list))
487 continue; /* wrong inode or list_empty */
488 if (!can_unuse(inode))
491 list_move(&inode->i_list, &freeable);
492 WARN_ON(inode->i_state & I_NEW);
493 inode->i_state |= I_FREEING;
496 inodes_stat.nr_unused -= nr_pruned;
497 if (current_is_kswapd())
498 __count_vm_events(KSWAPD_INODESTEAL, reap);
500 __count_vm_events(PGINODESTEAL, reap);
501 spin_unlock(&inode_lock);
503 dispose_list(&freeable);
504 up_read(&iprune_sem);
508 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
509 * "unused" means that no dentries are referring to the inodes: the files are
510 * not open and the dcache references to those inodes have already been
513 * This function is passed the number of inodes to scan, and it returns the
514 * total number of remaining possibly-reclaimable inodes.
516 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
520 * Nasty deadlock avoidance. We may hold various FS locks,
521 * and we don't want to recurse into the FS that called us
522 * in clear_inode() and friends..
524 if (!(gfp_mask & __GFP_FS))
528 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
531 static struct shrinker icache_shrinker = {
532 .shrink = shrink_icache_memory,
533 .seeks = DEFAULT_SEEKS,
536 static void __wait_on_freeing_inode(struct inode *inode);
538 * Called with the inode lock held.
539 * NOTE: we are not increasing the inode-refcount, you must call __iget()
540 * by hand after calling find_inode now! This simplifies iunique and won't
541 * add any additional branch in the common code.
543 static struct inode *find_inode(struct super_block *sb,
544 struct hlist_head *head,
545 int (*test)(struct inode *, void *),
548 struct hlist_node *node;
549 struct inode *inode = NULL;
552 hlist_for_each_entry(inode, node, head, i_hash) {
553 if (inode->i_sb != sb)
555 if (!test(inode, data))
557 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
558 __wait_on_freeing_inode(inode);
563 return node ? inode : NULL;
567 * find_inode_fast is the fast path version of find_inode, see the comment at
568 * iget_locked for details.
570 static struct inode *find_inode_fast(struct super_block *sb,
571 struct hlist_head *head, unsigned long ino)
573 struct hlist_node *node;
574 struct inode *inode = NULL;
577 hlist_for_each_entry(inode, node, head, i_hash) {
578 if (inode->i_ino != ino)
580 if (inode->i_sb != sb)
582 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
583 __wait_on_freeing_inode(inode);
588 return node ? inode : NULL;
591 static unsigned long hash(struct super_block *sb, unsigned long hashval)
595 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
597 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
598 return tmp & I_HASHMASK;
602 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
605 inodes_stat.nr_inodes++;
606 list_add(&inode->i_list, &inode_in_use);
607 list_add(&inode->i_sb_list, &sb->s_inodes);
609 hlist_add_head(&inode->i_hash, head);
613 * inode_add_to_lists - add a new inode to relevant lists
614 * @sb: superblock inode belongs to
615 * @inode: inode to mark in use
617 * When an inode is allocated it needs to be accounted for, added to the in use
618 * list, the owning superblock and the inode hash. This needs to be done under
619 * the inode_lock, so export a function to do this rather than the inode lock
620 * itself. We calculate the hash list to add to here so it is all internal
621 * which requires the caller to have already set up the inode number in the
624 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
626 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
628 spin_lock(&inode_lock);
629 __inode_add_to_lists(sb, head, inode);
630 spin_unlock(&inode_lock);
632 EXPORT_SYMBOL_GPL(inode_add_to_lists);
635 * new_inode - obtain an inode
638 * Allocates a new inode for given superblock. The default gfp_mask
639 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
640 * If HIGHMEM pages are unsuitable or it is known that pages allocated
641 * for the page cache are not reclaimable or migratable,
642 * mapping_set_gfp_mask() must be called with suitable flags on the
643 * newly created inode's mapping
646 struct inode *new_inode(struct super_block *sb)
649 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
650 * error if st_ino won't fit in target struct field. Use 32bit counter
651 * here to attempt to avoid that.
653 static unsigned int last_ino;
656 spin_lock_prefetch(&inode_lock);
658 inode = alloc_inode(sb);
660 spin_lock(&inode_lock);
661 __inode_add_to_lists(sb, NULL, inode);
662 inode->i_ino = ++last_ino;
664 spin_unlock(&inode_lock);
668 EXPORT_SYMBOL(new_inode);
670 void unlock_new_inode(struct inode *inode)
672 #ifdef CONFIG_DEBUG_LOCK_ALLOC
673 if (inode->i_mode & S_IFDIR) {
674 struct file_system_type *type = inode->i_sb->s_type;
676 /* Set new key only if filesystem hasn't already changed it */
677 if (!lockdep_match_class(&inode->i_mutex,
678 &type->i_mutex_key)) {
680 * ensure nobody is actually holding i_mutex
682 mutex_destroy(&inode->i_mutex);
683 mutex_init(&inode->i_mutex);
684 lockdep_set_class(&inode->i_mutex,
685 &type->i_mutex_dir_key);
690 * This is special! We do not need the spinlock when clearing I_NEW,
691 * because we're guaranteed that nobody else tries to do anything about
692 * the state of the inode when it is locked, as we just created it (so
693 * there can be no old holders that haven't tested I_NEW).
694 * However we must emit the memory barrier so that other CPUs reliably
695 * see the clearing of I_NEW after the other inode initialisation has
699 WARN_ON(!(inode->i_state & I_NEW));
700 inode->i_state &= ~I_NEW;
701 wake_up_inode(inode);
703 EXPORT_SYMBOL(unlock_new_inode);
706 * This is called without the inode lock held.. Be careful.
708 * We no longer cache the sb_flags in i_flags - see fs.h
709 * -- rmk@arm.uk.linux.org
711 static struct inode *get_new_inode(struct super_block *sb,
712 struct hlist_head *head,
713 int (*test)(struct inode *, void *),
714 int (*set)(struct inode *, void *),
719 inode = alloc_inode(sb);
723 spin_lock(&inode_lock);
724 /* We released the lock, so.. */
725 old = find_inode(sb, head, test, data);
727 if (set(inode, data))
730 __inode_add_to_lists(sb, head, inode);
731 inode->i_state = I_NEW;
732 spin_unlock(&inode_lock);
734 /* Return the locked inode with I_NEW set, the
735 * caller is responsible for filling in the contents
741 * Uhhuh, somebody else created the same inode under
742 * us. Use the old inode instead of the one we just
746 spin_unlock(&inode_lock);
747 destroy_inode(inode);
749 wait_on_inode(inode);
754 spin_unlock(&inode_lock);
755 destroy_inode(inode);
760 * get_new_inode_fast is the fast path version of get_new_inode, see the
761 * comment at iget_locked for details.
763 static struct inode *get_new_inode_fast(struct super_block *sb,
764 struct hlist_head *head, unsigned long ino)
768 inode = alloc_inode(sb);
772 spin_lock(&inode_lock);
773 /* We released the lock, so.. */
774 old = find_inode_fast(sb, head, ino);
777 __inode_add_to_lists(sb, head, inode);
778 inode->i_state = I_NEW;
779 spin_unlock(&inode_lock);
781 /* Return the locked inode with I_NEW set, the
782 * caller is responsible for filling in the contents
788 * Uhhuh, somebody else created the same inode under
789 * us. Use the old inode instead of the one we just
793 spin_unlock(&inode_lock);
794 destroy_inode(inode);
796 wait_on_inode(inode);
802 * iunique - get a unique inode number
804 * @max_reserved: highest reserved inode number
806 * Obtain an inode number that is unique on the system for a given
807 * superblock. This is used by file systems that have no natural
808 * permanent inode numbering system. An inode number is returned that
809 * is higher than the reserved limit but unique.
812 * With a large number of inodes live on the file system this function
813 * currently becomes quite slow.
815 ino_t iunique(struct super_block *sb, ino_t max_reserved)
818 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
819 * error if st_ino won't fit in target struct field. Use 32bit counter
820 * here to attempt to avoid that.
822 static unsigned int counter;
824 struct hlist_head *head;
827 spin_lock(&inode_lock);
829 if (counter <= max_reserved)
830 counter = max_reserved + 1;
832 head = inode_hashtable + hash(sb, res);
833 inode = find_inode_fast(sb, head, res);
834 } while (inode != NULL);
835 spin_unlock(&inode_lock);
839 EXPORT_SYMBOL(iunique);
841 struct inode *igrab(struct inode *inode)
843 spin_lock(&inode_lock);
844 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
848 * Handle the case where s_op->clear_inode is not been
849 * called yet, and somebody is calling igrab
850 * while the inode is getting freed.
853 spin_unlock(&inode_lock);
856 EXPORT_SYMBOL(igrab);
859 * ifind - internal function, you want ilookup5() or iget5().
860 * @sb: super block of file system to search
861 * @head: the head of the list to search
862 * @test: callback used for comparisons between inodes
863 * @data: opaque data pointer to pass to @test
864 * @wait: if true wait for the inode to be unlocked, if false do not
866 * ifind() searches for the inode specified by @data in the inode
867 * cache. This is a generalized version of ifind_fast() for file systems where
868 * the inode number is not sufficient for unique identification of an inode.
870 * If the inode is in the cache, the inode is returned with an incremented
873 * Otherwise NULL is returned.
875 * Note, @test is called with the inode_lock held, so can't sleep.
877 static struct inode *ifind(struct super_block *sb,
878 struct hlist_head *head, int (*test)(struct inode *, void *),
879 void *data, const int wait)
883 spin_lock(&inode_lock);
884 inode = find_inode(sb, head, test, data);
887 spin_unlock(&inode_lock);
889 wait_on_inode(inode);
892 spin_unlock(&inode_lock);
897 * ifind_fast - internal function, you want ilookup() or iget().
898 * @sb: super block of file system to search
899 * @head: head of the list to search
900 * @ino: inode number to search for
902 * ifind_fast() searches for the inode @ino in the inode cache. This is for
903 * file systems where the inode number is sufficient for unique identification
906 * If the inode is in the cache, the inode is returned with an incremented
909 * Otherwise NULL is returned.
911 static struct inode *ifind_fast(struct super_block *sb,
912 struct hlist_head *head, unsigned long ino)
916 spin_lock(&inode_lock);
917 inode = find_inode_fast(sb, head, ino);
920 spin_unlock(&inode_lock);
921 wait_on_inode(inode);
924 spin_unlock(&inode_lock);
929 * ilookup5_nowait - search for an inode in the inode cache
930 * @sb: super block of file system to search
931 * @hashval: hash value (usually inode number) to search for
932 * @test: callback used for comparisons between inodes
933 * @data: opaque data pointer to pass to @test
935 * ilookup5() uses ifind() to search for the inode specified by @hashval and
936 * @data in the inode cache. This is a generalized version of ilookup() for
937 * file systems where the inode number is not sufficient for unique
938 * identification of an inode.
940 * If the inode is in the cache, the inode is returned with an incremented
941 * reference count. Note, the inode lock is not waited upon so you have to be
942 * very careful what you do with the returned inode. You probably should be
943 * using ilookup5() instead.
945 * Otherwise NULL is returned.
947 * Note, @test is called with the inode_lock held, so can't sleep.
949 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
950 int (*test)(struct inode *, void *), void *data)
952 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
954 return ifind(sb, head, test, data, 0);
956 EXPORT_SYMBOL(ilookup5_nowait);
959 * ilookup5 - search for an inode in the inode cache
960 * @sb: super block of file system to search
961 * @hashval: hash value (usually inode number) to search for
962 * @test: callback used for comparisons between inodes
963 * @data: opaque data pointer to pass to @test
965 * ilookup5() uses ifind() to search for the inode specified by @hashval and
966 * @data in the inode cache. This is a generalized version of ilookup() for
967 * file systems where the inode number is not sufficient for unique
968 * identification of an inode.
970 * If the inode is in the cache, the inode lock is waited upon and the inode is
971 * returned with an incremented reference count.
973 * Otherwise NULL is returned.
975 * Note, @test is called with the inode_lock held, so can't sleep.
977 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
978 int (*test)(struct inode *, void *), void *data)
980 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
982 return ifind(sb, head, test, data, 1);
984 EXPORT_SYMBOL(ilookup5);
987 * ilookup - search for an inode in the inode cache
988 * @sb: super block of file system to search
989 * @ino: inode number to search for
991 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
992 * This is for file systems where the inode number is sufficient for unique
993 * identification of an inode.
995 * If the inode is in the cache, the inode is returned with an incremented
998 * Otherwise NULL is returned.
1000 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1002 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1004 return ifind_fast(sb, head, ino);
1006 EXPORT_SYMBOL(ilookup);
1009 * iget5_locked - obtain an inode from a mounted file system
1010 * @sb: super block of file system
1011 * @hashval: hash value (usually inode number) to get
1012 * @test: callback used for comparisons between inodes
1013 * @set: callback used to initialize a new struct inode
1014 * @data: opaque data pointer to pass to @test and @set
1016 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1017 * and @data in the inode cache and if present it is returned with an increased
1018 * reference count. This is a generalized version of iget_locked() for file
1019 * systems where the inode number is not sufficient for unique identification
1022 * If the inode is not in cache, get_new_inode() is called to allocate a new
1023 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1024 * file system gets to fill it in before unlocking it via unlock_new_inode().
1026 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1028 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1029 int (*test)(struct inode *, void *),
1030 int (*set)(struct inode *, void *), void *data)
1032 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1033 struct inode *inode;
1035 inode = ifind(sb, head, test, data, 1);
1039 * get_new_inode() will do the right thing, re-trying the search
1040 * in case it had to block at any point.
1042 return get_new_inode(sb, head, test, set, data);
1044 EXPORT_SYMBOL(iget5_locked);
1047 * iget_locked - obtain an inode from a mounted file system
1048 * @sb: super block of file system
1049 * @ino: inode number to get
1051 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1052 * the inode cache and if present it is returned with an increased reference
1053 * count. This is for file systems where the inode number is sufficient for
1054 * unique identification of an inode.
1056 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1057 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1058 * The file system gets to fill it in before unlocking it via
1059 * unlock_new_inode().
1061 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1063 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1064 struct inode *inode;
1066 inode = ifind_fast(sb, head, ino);
1070 * get_new_inode_fast() will do the right thing, re-trying the search
1071 * in case it had to block at any point.
1073 return get_new_inode_fast(sb, head, ino);
1075 EXPORT_SYMBOL(iget_locked);
1077 int insert_inode_locked(struct inode *inode)
1079 struct super_block *sb = inode->i_sb;
1080 ino_t ino = inode->i_ino;
1081 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1083 inode->i_state |= I_NEW;
1085 struct hlist_node *node;
1086 struct inode *old = NULL;
1087 spin_lock(&inode_lock);
1088 hlist_for_each_entry(old, node, head, i_hash) {
1089 if (old->i_ino != ino)
1091 if (old->i_sb != sb)
1093 if (old->i_state & (I_FREEING|I_WILL_FREE))
1097 if (likely(!node)) {
1098 hlist_add_head(&inode->i_hash, head);
1099 spin_unlock(&inode_lock);
1103 spin_unlock(&inode_lock);
1105 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1112 EXPORT_SYMBOL(insert_inode_locked);
1114 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1115 int (*test)(struct inode *, void *), void *data)
1117 struct super_block *sb = inode->i_sb;
1118 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1120 inode->i_state |= I_NEW;
1123 struct hlist_node *node;
1124 struct inode *old = NULL;
1126 spin_lock(&inode_lock);
1127 hlist_for_each_entry(old, node, head, i_hash) {
1128 if (old->i_sb != sb)
1130 if (!test(old, data))
1132 if (old->i_state & (I_FREEING|I_WILL_FREE))
1136 if (likely(!node)) {
1137 hlist_add_head(&inode->i_hash, head);
1138 spin_unlock(&inode_lock);
1142 spin_unlock(&inode_lock);
1144 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1151 EXPORT_SYMBOL(insert_inode_locked4);
1154 * __insert_inode_hash - hash an inode
1155 * @inode: unhashed inode
1156 * @hashval: unsigned long value used to locate this object in the
1159 * Add an inode to the inode hash for this superblock.
1161 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1163 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1164 spin_lock(&inode_lock);
1165 hlist_add_head(&inode->i_hash, head);
1166 spin_unlock(&inode_lock);
1168 EXPORT_SYMBOL(__insert_inode_hash);
1171 * remove_inode_hash - remove an inode from the hash
1172 * @inode: inode to unhash
1174 * Remove an inode from the superblock.
1176 void remove_inode_hash(struct inode *inode)
1178 spin_lock(&inode_lock);
1179 hlist_del_init(&inode->i_hash);
1180 spin_unlock(&inode_lock);
1182 EXPORT_SYMBOL(remove_inode_hash);
1184 int generic_delete_inode(struct inode *inode)
1188 EXPORT_SYMBOL(generic_delete_inode);
1191 * Normal UNIX filesystem behaviour: delete the
1192 * inode when the usage count drops to zero, and
1195 int generic_drop_inode(struct inode *inode)
1197 return !inode->i_nlink || hlist_unhashed(&inode->i_hash);
1199 EXPORT_SYMBOL_GPL(generic_drop_inode);
1202 * Called when we're dropping the last reference
1205 * Call the FS "drop_inode()" function, defaulting to
1206 * the legacy UNIX filesystem behaviour. If it tells
1207 * us to evict inode, do so. Otherwise, retain inode
1208 * in cache if fs is alive, sync and evict if fs is
1211 static void iput_final(struct inode *inode)
1213 struct super_block *sb = inode->i_sb;
1214 const struct super_operations *op = inode->i_sb->s_op;
1217 if (op && op->drop_inode)
1218 drop = op->drop_inode(inode);
1220 drop = generic_drop_inode(inode);
1223 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1224 list_move(&inode->i_list, &inode_unused);
1225 inodes_stat.nr_unused++;
1226 if (sb->s_flags & MS_ACTIVE) {
1227 spin_unlock(&inode_lock);
1230 WARN_ON(inode->i_state & I_NEW);
1231 inode->i_state |= I_WILL_FREE;
1232 spin_unlock(&inode_lock);
1233 write_inode_now(inode, 1);
1234 spin_lock(&inode_lock);
1235 WARN_ON(inode->i_state & I_NEW);
1236 inode->i_state &= ~I_WILL_FREE;
1237 inodes_stat.nr_unused--;
1238 hlist_del_init(&inode->i_hash);
1240 list_del_init(&inode->i_list);
1241 list_del_init(&inode->i_sb_list);
1242 WARN_ON(inode->i_state & I_NEW);
1243 inode->i_state |= I_FREEING;
1244 inodes_stat.nr_inodes--;
1245 spin_unlock(&inode_lock);
1247 spin_lock(&inode_lock);
1248 hlist_del_init(&inode->i_hash);
1249 spin_unlock(&inode_lock);
1250 wake_up_inode(inode);
1251 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1252 destroy_inode(inode);
1256 * iput - put an inode
1257 * @inode: inode to put
1259 * Puts an inode, dropping its usage count. If the inode use count hits
1260 * zero, the inode is then freed and may also be destroyed.
1262 * Consequently, iput() can sleep.
1264 void iput(struct inode *inode)
1267 BUG_ON(inode->i_state & I_CLEAR);
1269 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1273 EXPORT_SYMBOL(iput);
1276 * bmap - find a block number in a file
1277 * @inode: inode of file
1278 * @block: block to find
1280 * Returns the block number on the device holding the inode that
1281 * is the disk block number for the block of the file requested.
1282 * That is, asked for block 4 of inode 1 the function will return the
1283 * disk block relative to the disk start that holds that block of the
1286 sector_t bmap(struct inode *inode, sector_t block)
1289 if (inode->i_mapping->a_ops->bmap)
1290 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1293 EXPORT_SYMBOL(bmap);
1296 * With relative atime, only update atime if the previous atime is
1297 * earlier than either the ctime or mtime or if at least a day has
1298 * passed since the last atime update.
1300 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1301 struct timespec now)
1304 if (!(mnt->mnt_flags & MNT_RELATIME))
1307 * Is mtime younger than atime? If yes, update atime:
1309 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1312 * Is ctime younger than atime? If yes, update atime:
1314 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1318 * Is the previous atime value older than a day? If yes,
1321 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1324 * Good, we can skip the atime update:
1330 * touch_atime - update the access time
1331 * @mnt: mount the inode is accessed on
1332 * @dentry: dentry accessed
1334 * Update the accessed time on an inode and mark it for writeback.
1335 * This function automatically handles read only file systems and media,
1336 * as well as the "noatime" flag and inode specific "noatime" markers.
1338 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1340 struct inode *inode = dentry->d_inode;
1341 struct timespec now;
1343 if (inode->i_flags & S_NOATIME)
1345 if (IS_NOATIME(inode))
1347 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1350 if (mnt->mnt_flags & MNT_NOATIME)
1352 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1355 now = current_fs_time(inode->i_sb);
1357 if (!relatime_need_update(mnt, inode, now))
1360 if (timespec_equal(&inode->i_atime, &now))
1363 if (mnt_want_write(mnt))
1366 inode->i_atime = now;
1367 mark_inode_dirty_sync(inode);
1368 mnt_drop_write(mnt);
1370 EXPORT_SYMBOL(touch_atime);
1373 * file_update_time - update mtime and ctime time
1374 * @file: file accessed
1376 * Update the mtime and ctime members of an inode and mark the inode
1377 * for writeback. Note that this function is meant exclusively for
1378 * usage in the file write path of filesystems, and filesystems may
1379 * choose to explicitly ignore update via this function with the
1380 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1381 * timestamps are handled by the server.
1384 void file_update_time(struct file *file)
1386 struct inode *inode = file->f_path.dentry->d_inode;
1387 struct timespec now;
1388 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1390 /* First try to exhaust all avenues to not sync */
1391 if (IS_NOCMTIME(inode))
1394 now = current_fs_time(inode->i_sb);
1395 if (!timespec_equal(&inode->i_mtime, &now))
1398 if (!timespec_equal(&inode->i_ctime, &now))
1401 if (IS_I_VERSION(inode))
1402 sync_it |= S_VERSION;
1407 /* Finally allowed to write? Takes lock. */
1408 if (mnt_want_write_file(file))
1411 /* Only change inode inside the lock region */
1412 if (sync_it & S_VERSION)
1413 inode_inc_iversion(inode);
1414 if (sync_it & S_CTIME)
1415 inode->i_ctime = now;
1416 if (sync_it & S_MTIME)
1417 inode->i_mtime = now;
1418 mark_inode_dirty_sync(inode);
1419 mnt_drop_write(file->f_path.mnt);
1421 EXPORT_SYMBOL(file_update_time);
1423 int inode_needs_sync(struct inode *inode)
1427 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1431 EXPORT_SYMBOL(inode_needs_sync);
1433 int inode_wait(void *word)
1438 EXPORT_SYMBOL(inode_wait);
1441 * If we try to find an inode in the inode hash while it is being
1442 * deleted, we have to wait until the filesystem completes its
1443 * deletion before reporting that it isn't found. This function waits
1444 * until the deletion _might_ have completed. Callers are responsible
1445 * to recheck inode state.
1447 * It doesn't matter if I_NEW is not set initially, a call to
1448 * wake_up_inode() after removing from the hash list will DTRT.
1450 * This is called with inode_lock held.
1452 static void __wait_on_freeing_inode(struct inode *inode)
1454 wait_queue_head_t *wq;
1455 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1456 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1457 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1458 spin_unlock(&inode_lock);
1460 finish_wait(wq, &wait.wait);
1461 spin_lock(&inode_lock);
1464 static __initdata unsigned long ihash_entries;
1465 static int __init set_ihash_entries(char *str)
1469 ihash_entries = simple_strtoul(str, &str, 0);
1472 __setup("ihash_entries=", set_ihash_entries);
1475 * Initialize the waitqueues and inode hash table.
1477 void __init inode_init_early(void)
1481 /* If hashes are distributed across NUMA nodes, defer
1482 * hash allocation until vmalloc space is available.
1488 alloc_large_system_hash("Inode-cache",
1489 sizeof(struct hlist_head),
1497 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1498 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1501 void __init inode_init(void)
1505 /* inode slab cache */
1506 inode_cachep = kmem_cache_create("inode_cache",
1507 sizeof(struct inode),
1509 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1512 register_shrinker(&icache_shrinker);
1514 /* Hash may have been set up in inode_init_early */
1519 alloc_large_system_hash("Inode-cache",
1520 sizeof(struct hlist_head),
1528 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1529 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1532 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1534 inode->i_mode = mode;
1535 if (S_ISCHR(mode)) {
1536 inode->i_fop = &def_chr_fops;
1537 inode->i_rdev = rdev;
1538 } else if (S_ISBLK(mode)) {
1539 inode->i_fop = &def_blk_fops;
1540 inode->i_rdev = rdev;
1541 } else if (S_ISFIFO(mode))
1542 inode->i_fop = &def_fifo_fops;
1543 else if (S_ISSOCK(mode))
1544 inode->i_fop = &bad_sock_fops;
1546 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1547 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1550 EXPORT_SYMBOL(init_special_inode);
1553 * Init uid,gid,mode for new inode according to posix standards
1555 * @dir: Directory inode
1556 * @mode: mode of the new inode
1558 void inode_init_owner(struct inode *inode, const struct inode *dir,
1561 inode->i_uid = current_fsuid();
1562 if (dir && dir->i_mode & S_ISGID) {
1563 inode->i_gid = dir->i_gid;
1567 inode->i_gid = current_fsgid();
1568 inode->i_mode = mode;
1570 EXPORT_SYMBOL(inode_init_owner);