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/fsnotify.h>
24 #include <linux/mount.h>
25 #include <linux/async.h>
26 #include <linux/posix_acl.h>
29 * This is needed for the following functions:
31 * - invalidate_inode_buffers
34 * FIXME: remove all knowledge of the buffer layer from this file
36 #include <linux/buffer_head.h>
39 * New inode.c implementation.
41 * This implementation has the basic premise of trying
42 * to be extremely low-overhead and SMP-safe, yet be
43 * simple enough to be "obviously correct".
48 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
50 /* #define INODE_PARANOIA 1 */
51 /* #define INODE_DEBUG 1 */
54 * Inode lookup is no longer as critical as it used to be:
55 * most of the lookups are going to be through the dcache.
57 #define I_HASHBITS i_hash_shift
58 #define I_HASHMASK i_hash_mask
60 static unsigned int i_hash_mask __read_mostly;
61 static unsigned int i_hash_shift __read_mostly;
64 * Each inode can be on two separate lists. One is
65 * the hash list of the inode, used for lookups. The
66 * other linked list is the "type" list:
67 * "in_use" - valid inode, i_count > 0, i_nlink > 0
68 * "dirty" - as "in_use" but also dirty
69 * "unused" - valid inode, i_count = 0
71 * A "dirty" list is maintained for each super block,
72 * allowing for low-overhead inode sync() operations.
75 LIST_HEAD(inode_in_use);
76 LIST_HEAD(inode_unused);
77 static struct hlist_head *inode_hashtable __read_mostly;
80 * A simple spinlock to protect the list manipulations.
82 * NOTE! You also have to own the lock if you change
83 * the i_state of an inode while it is in use..
85 DEFINE_SPINLOCK(inode_lock);
88 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
89 * icache shrinking path, and the umount path. Without this exclusion,
90 * by the time prune_icache calls iput for the inode whose pages it has
91 * been invalidating, or by the time it calls clear_inode & destroy_inode
92 * from its final dispose_list, the struct super_block they refer to
93 * (for inode->i_sb->s_op) may already have been freed and reused.
95 * We make this an rwsem because the fastpath is icache shrinking. In
96 * some cases a filesystem may be doing a significant amount of work in
97 * its inode reclaim code, so this should improve parallelism.
99 static DECLARE_RWSEM(iprune_sem);
102 * Statistics gathering..
104 struct inodes_stat_t inodes_stat;
106 static struct kmem_cache *inode_cachep __read_mostly;
108 static void wake_up_inode(struct inode *inode)
111 * Prevent speculative execution through spin_unlock(&inode_lock);
114 wake_up_bit(&inode->i_state, __I_NEW);
118 * inode_init_always - perform inode structure intialisation
119 * @sb: superblock inode belongs to
120 * @inode: inode to initialise
122 * These are initializations that need to be done on every inode
123 * allocation as the fields are not initialised by slab allocation.
125 int inode_init_always(struct super_block *sb, struct inode *inode)
127 static const struct address_space_operations empty_aops;
128 static const struct inode_operations empty_iops;
129 static const struct file_operations empty_fops;
130 struct address_space *const mapping = &inode->i_data;
133 inode->i_blkbits = sb->s_blocksize_bits;
135 atomic_set(&inode->i_count, 1);
136 inode->i_op = &empty_iops;
137 inode->i_fop = &empty_fops;
141 atomic_set(&inode->i_writecount, 0);
145 inode->i_generation = 0;
147 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
149 inode->i_pipe = NULL;
150 inode->i_bdev = NULL;
151 inode->i_cdev = NULL;
153 inode->dirtied_when = 0;
155 if (security_inode_alloc(inode))
157 spin_lock_init(&inode->i_lock);
158 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
160 mutex_init(&inode->i_mutex);
161 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
163 init_rwsem(&inode->i_alloc_sem);
164 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
166 mapping->a_ops = &empty_aops;
167 mapping->host = inode;
169 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
170 mapping->assoc_mapping = NULL;
171 mapping->backing_dev_info = &default_backing_dev_info;
172 mapping->writeback_index = 0;
175 * If the block_device provides a backing_dev_info for client
176 * inodes then use that. Otherwise the inode share the bdev's
180 struct backing_dev_info *bdi;
182 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
183 mapping->backing_dev_info = bdi;
185 inode->i_private = NULL;
186 inode->i_mapping = mapping;
187 #ifdef CONFIG_FS_POSIX_ACL
188 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
191 #ifdef CONFIG_FSNOTIFY
192 inode->i_fsnotify_mask = 0;
199 EXPORT_SYMBOL(inode_init_always);
201 static struct inode *alloc_inode(struct super_block *sb)
205 if (sb->s_op->alloc_inode)
206 inode = sb->s_op->alloc_inode(sb);
208 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
213 if (unlikely(inode_init_always(sb, inode))) {
214 if (inode->i_sb->s_op->destroy_inode)
215 inode->i_sb->s_op->destroy_inode(inode);
217 kmem_cache_free(inode_cachep, inode);
224 void __destroy_inode(struct inode *inode)
226 BUG_ON(inode_has_buffers(inode));
227 security_inode_free(inode);
228 fsnotify_inode_delete(inode);
229 #ifdef CONFIG_FS_POSIX_ACL
230 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
231 posix_acl_release(inode->i_acl);
232 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
233 posix_acl_release(inode->i_default_acl);
236 EXPORT_SYMBOL(__destroy_inode);
238 void destroy_inode(struct inode *inode)
240 __destroy_inode(inode);
241 if (inode->i_sb->s_op->destroy_inode)
242 inode->i_sb->s_op->destroy_inode(inode);
244 kmem_cache_free(inode_cachep, (inode));
248 * These are initializations that only need to be done
249 * once, because the fields are idempotent across use
250 * of the inode, so let the slab aware of that.
252 void inode_init_once(struct inode *inode)
254 memset(inode, 0, sizeof(*inode));
255 INIT_HLIST_NODE(&inode->i_hash);
256 INIT_LIST_HEAD(&inode->i_dentry);
257 INIT_LIST_HEAD(&inode->i_devices);
258 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
259 spin_lock_init(&inode->i_data.tree_lock);
260 spin_lock_init(&inode->i_data.i_mmap_lock);
261 INIT_LIST_HEAD(&inode->i_data.private_list);
262 spin_lock_init(&inode->i_data.private_lock);
263 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
264 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
265 i_size_ordered_init(inode);
266 #ifdef CONFIG_FSNOTIFY
267 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
270 EXPORT_SYMBOL(inode_init_once);
272 static void init_once(void *foo)
274 struct inode *inode = (struct inode *) foo;
276 inode_init_once(inode);
280 * inode_lock must be held
282 void __iget(struct inode *inode)
284 if (atomic_inc_return(&inode->i_count) != 1)
287 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
288 list_move(&inode->i_list, &inode_in_use);
289 inodes_stat.nr_unused--;
293 * clear_inode - clear an inode
294 * @inode: inode to clear
296 * This is called by the filesystem to tell us
297 * that the inode is no longer useful. We just
298 * terminate it with extreme prejudice.
300 void clear_inode(struct inode *inode)
303 invalidate_inode_buffers(inode);
305 BUG_ON(inode->i_data.nrpages);
306 BUG_ON(!(inode->i_state & I_FREEING));
307 BUG_ON(inode->i_state & I_CLEAR);
308 inode_sync_wait(inode);
309 if (inode->i_sb->s_op->clear_inode)
310 inode->i_sb->s_op->clear_inode(inode);
311 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
313 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
315 inode->i_state = I_CLEAR;
317 EXPORT_SYMBOL(clear_inode);
320 * dispose_list - dispose of the contents of a local list
321 * @head: the head of the list to free
323 * Dispose-list gets a local list with local inodes in it, so it doesn't
324 * need to worry about list corruption and SMP locks.
326 static void dispose_list(struct list_head *head)
330 while (!list_empty(head)) {
333 inode = list_first_entry(head, struct inode, i_list);
334 list_del(&inode->i_list);
336 if (inode->i_data.nrpages)
337 truncate_inode_pages(&inode->i_data, 0);
340 spin_lock(&inode_lock);
341 hlist_del_init(&inode->i_hash);
342 list_del_init(&inode->i_sb_list);
343 spin_unlock(&inode_lock);
345 wake_up_inode(inode);
346 destroy_inode(inode);
349 spin_lock(&inode_lock);
350 inodes_stat.nr_inodes -= nr_disposed;
351 spin_unlock(&inode_lock);
355 * Invalidate all inodes for a device.
357 static int invalidate_list(struct list_head *head, struct list_head *dispose)
359 struct list_head *next;
360 int busy = 0, count = 0;
364 struct list_head *tmp = next;
368 * We can reschedule here without worrying about the list's
369 * consistency because the per-sb list of inodes must not
370 * change during umount anymore, and because iprune_sem keeps
371 * shrink_icache_memory() away.
373 cond_resched_lock(&inode_lock);
378 inode = list_entry(tmp, struct inode, i_sb_list);
379 if (inode->i_state & I_NEW)
381 invalidate_inode_buffers(inode);
382 if (!atomic_read(&inode->i_count)) {
383 list_move(&inode->i_list, dispose);
384 WARN_ON(inode->i_state & I_NEW);
385 inode->i_state |= I_FREEING;
391 /* only unused inodes may be cached with i_count zero */
392 inodes_stat.nr_unused -= count;
397 * invalidate_inodes - discard the inodes on a device
400 * Discard all of the inodes for a given superblock. If the discard
401 * fails because there are busy inodes then a non zero value is returned.
402 * If the discard is successful all the inodes have been discarded.
404 int invalidate_inodes(struct super_block *sb)
407 LIST_HEAD(throw_away);
409 down_write(&iprune_sem);
410 spin_lock(&inode_lock);
411 fsnotify_unmount_inodes(&sb->s_inodes);
412 busy = invalidate_list(&sb->s_inodes, &throw_away);
413 spin_unlock(&inode_lock);
415 dispose_list(&throw_away);
416 up_write(&iprune_sem);
420 EXPORT_SYMBOL(invalidate_inodes);
422 static int can_unuse(struct inode *inode)
426 if (inode_has_buffers(inode))
428 if (atomic_read(&inode->i_count))
430 if (inode->i_data.nrpages)
436 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
437 * a temporary list and then are freed outside inode_lock by dispose_list().
439 * Any inodes which are pinned purely because of attached pagecache have their
440 * pagecache removed. We expect the final iput() on that inode to add it to
441 * the front of the inode_unused list. So look for it there and if the
442 * inode is still freeable, proceed. The right inode is found 99.9% of the
443 * time in testing on a 4-way.
445 * If the inode has metadata buffers attached to mapping->private_list then
446 * try to remove them.
448 static void prune_icache(int nr_to_scan)
453 unsigned long reap = 0;
455 down_read(&iprune_sem);
456 spin_lock(&inode_lock);
457 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
460 if (list_empty(&inode_unused))
463 inode = list_entry(inode_unused.prev, struct inode, i_list);
465 if (inode->i_state || atomic_read(&inode->i_count)) {
466 list_move(&inode->i_list, &inode_unused);
469 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
471 spin_unlock(&inode_lock);
472 if (remove_inode_buffers(inode))
473 reap += invalidate_mapping_pages(&inode->i_data,
476 spin_lock(&inode_lock);
478 if (inode != list_entry(inode_unused.next,
479 struct inode, i_list))
480 continue; /* wrong inode or list_empty */
481 if (!can_unuse(inode))
484 list_move(&inode->i_list, &freeable);
485 WARN_ON(inode->i_state & I_NEW);
486 inode->i_state |= I_FREEING;
489 inodes_stat.nr_unused -= nr_pruned;
490 if (current_is_kswapd())
491 __count_vm_events(KSWAPD_INODESTEAL, reap);
493 __count_vm_events(PGINODESTEAL, reap);
494 spin_unlock(&inode_lock);
496 dispose_list(&freeable);
497 up_read(&iprune_sem);
501 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
502 * "unused" means that no dentries are referring to the inodes: the files are
503 * not open and the dcache references to those inodes have already been
506 * This function is passed the number of inodes to scan, and it returns the
507 * total number of remaining possibly-reclaimable inodes.
509 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
513 * Nasty deadlock avoidance. We may hold various FS locks,
514 * and we don't want to recurse into the FS that called us
515 * in clear_inode() and friends..
517 if (!(gfp_mask & __GFP_FS))
521 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
524 static struct shrinker icache_shrinker = {
525 .shrink = shrink_icache_memory,
526 .seeks = DEFAULT_SEEKS,
529 static void __wait_on_freeing_inode(struct inode *inode);
531 * Called with the inode lock held.
532 * NOTE: we are not increasing the inode-refcount, you must call __iget()
533 * by hand after calling find_inode now! This simplifies iunique and won't
534 * add any additional branch in the common code.
536 static struct inode *find_inode(struct super_block *sb,
537 struct hlist_head *head,
538 int (*test)(struct inode *, void *),
541 struct hlist_node *node;
542 struct inode *inode = NULL;
545 hlist_for_each_entry(inode, node, head, i_hash) {
546 if (inode->i_sb != sb)
548 if (!test(inode, data))
550 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
551 __wait_on_freeing_inode(inode);
556 return node ? inode : NULL;
560 * find_inode_fast is the fast path version of find_inode, see the comment at
561 * iget_locked for details.
563 static struct inode *find_inode_fast(struct super_block *sb,
564 struct hlist_head *head, unsigned long ino)
566 struct hlist_node *node;
567 struct inode *inode = NULL;
570 hlist_for_each_entry(inode, node, head, i_hash) {
571 if (inode->i_ino != ino)
573 if (inode->i_sb != sb)
575 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
576 __wait_on_freeing_inode(inode);
581 return node ? inode : NULL;
584 static unsigned long hash(struct super_block *sb, unsigned long hashval)
588 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
590 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
591 return tmp & I_HASHMASK;
595 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
598 inodes_stat.nr_inodes++;
599 list_add(&inode->i_list, &inode_in_use);
600 list_add(&inode->i_sb_list, &sb->s_inodes);
602 hlist_add_head(&inode->i_hash, head);
606 * inode_add_to_lists - add a new inode to relevant lists
607 * @sb: superblock inode belongs to
608 * @inode: inode to mark in use
610 * When an inode is allocated it needs to be accounted for, added to the in use
611 * list, the owning superblock and the inode hash. This needs to be done under
612 * the inode_lock, so export a function to do this rather than the inode lock
613 * itself. We calculate the hash list to add to here so it is all internal
614 * which requires the caller to have already set up the inode number in the
617 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
619 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
621 spin_lock(&inode_lock);
622 __inode_add_to_lists(sb, head, inode);
623 spin_unlock(&inode_lock);
625 EXPORT_SYMBOL_GPL(inode_add_to_lists);
628 * new_inode - obtain an inode
631 * Allocates a new inode for given superblock. The default gfp_mask
632 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
633 * If HIGHMEM pages are unsuitable or it is known that pages allocated
634 * for the page cache are not reclaimable or migratable,
635 * mapping_set_gfp_mask() must be called with suitable flags on the
636 * newly created inode's mapping
639 struct inode *new_inode(struct super_block *sb)
642 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
643 * error if st_ino won't fit in target struct field. Use 32bit counter
644 * here to attempt to avoid that.
646 static unsigned int last_ino;
649 spin_lock_prefetch(&inode_lock);
651 inode = alloc_inode(sb);
653 spin_lock(&inode_lock);
654 __inode_add_to_lists(sb, NULL, inode);
655 inode->i_ino = ++last_ino;
657 spin_unlock(&inode_lock);
661 EXPORT_SYMBOL(new_inode);
663 void unlock_new_inode(struct inode *inode)
665 #ifdef CONFIG_DEBUG_LOCK_ALLOC
666 if (inode->i_mode & S_IFDIR) {
667 struct file_system_type *type = inode->i_sb->s_type;
669 /* Set new key only if filesystem hasn't already changed it */
670 if (!lockdep_match_class(&inode->i_mutex,
671 &type->i_mutex_key)) {
673 * ensure nobody is actually holding i_mutex
675 mutex_destroy(&inode->i_mutex);
676 mutex_init(&inode->i_mutex);
677 lockdep_set_class(&inode->i_mutex,
678 &type->i_mutex_dir_key);
683 * This is special! We do not need the spinlock when clearing I_NEW,
684 * because we're guaranteed that nobody else tries to do anything about
685 * the state of the inode when it is locked, as we just created it (so
686 * there can be no old holders that haven't tested I_NEW).
687 * However we must emit the memory barrier so that other CPUs reliably
688 * see the clearing of I_NEW after the other inode initialisation has
692 WARN_ON(!(inode->i_state & I_NEW));
693 inode->i_state &= ~I_NEW;
694 wake_up_inode(inode);
696 EXPORT_SYMBOL(unlock_new_inode);
699 * This is called without the inode lock held.. Be careful.
701 * We no longer cache the sb_flags in i_flags - see fs.h
702 * -- rmk@arm.uk.linux.org
704 static struct inode *get_new_inode(struct super_block *sb,
705 struct hlist_head *head,
706 int (*test)(struct inode *, void *),
707 int (*set)(struct inode *, void *),
712 inode = alloc_inode(sb);
716 spin_lock(&inode_lock);
717 /* We released the lock, so.. */
718 old = find_inode(sb, head, test, data);
720 if (set(inode, data))
723 __inode_add_to_lists(sb, head, inode);
724 inode->i_state = I_NEW;
725 spin_unlock(&inode_lock);
727 /* Return the locked inode with I_NEW set, the
728 * caller is responsible for filling in the contents
734 * Uhhuh, somebody else created the same inode under
735 * us. Use the old inode instead of the one we just
739 spin_unlock(&inode_lock);
740 destroy_inode(inode);
742 wait_on_inode(inode);
747 spin_unlock(&inode_lock);
748 destroy_inode(inode);
753 * get_new_inode_fast is the fast path version of get_new_inode, see the
754 * comment at iget_locked for details.
756 static struct inode *get_new_inode_fast(struct super_block *sb,
757 struct hlist_head *head, unsigned long ino)
761 inode = alloc_inode(sb);
765 spin_lock(&inode_lock);
766 /* We released the lock, so.. */
767 old = find_inode_fast(sb, head, ino);
770 __inode_add_to_lists(sb, head, inode);
771 inode->i_state = I_NEW;
772 spin_unlock(&inode_lock);
774 /* Return the locked inode with I_NEW set, the
775 * caller is responsible for filling in the contents
781 * Uhhuh, somebody else created the same inode under
782 * us. Use the old inode instead of the one we just
786 spin_unlock(&inode_lock);
787 destroy_inode(inode);
789 wait_on_inode(inode);
795 * iunique - get a unique inode number
797 * @max_reserved: highest reserved inode number
799 * Obtain an inode number that is unique on the system for a given
800 * superblock. This is used by file systems that have no natural
801 * permanent inode numbering system. An inode number is returned that
802 * is higher than the reserved limit but unique.
805 * With a large number of inodes live on the file system this function
806 * currently becomes quite slow.
808 ino_t iunique(struct super_block *sb, ino_t max_reserved)
811 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
812 * error if st_ino won't fit in target struct field. Use 32bit counter
813 * here to attempt to avoid that.
815 static unsigned int counter;
817 struct hlist_head *head;
820 spin_lock(&inode_lock);
822 if (counter <= max_reserved)
823 counter = max_reserved + 1;
825 head = inode_hashtable + hash(sb, res);
826 inode = find_inode_fast(sb, head, res);
827 } while (inode != NULL);
828 spin_unlock(&inode_lock);
832 EXPORT_SYMBOL(iunique);
834 struct inode *igrab(struct inode *inode)
836 spin_lock(&inode_lock);
837 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
841 * Handle the case where s_op->clear_inode is not been
842 * called yet, and somebody is calling igrab
843 * while the inode is getting freed.
846 spin_unlock(&inode_lock);
849 EXPORT_SYMBOL(igrab);
852 * ifind - internal function, you want ilookup5() or iget5().
853 * @sb: super block of file system to search
854 * @head: the head of the list to search
855 * @test: callback used for comparisons between inodes
856 * @data: opaque data pointer to pass to @test
857 * @wait: if true wait for the inode to be unlocked, if false do not
859 * ifind() searches for the inode specified by @data in the inode
860 * cache. This is a generalized version of ifind_fast() for file systems where
861 * the inode number is not sufficient for unique identification of an inode.
863 * If the inode is in the cache, the inode is returned with an incremented
866 * Otherwise NULL is returned.
868 * Note, @test is called with the inode_lock held, so can't sleep.
870 static struct inode *ifind(struct super_block *sb,
871 struct hlist_head *head, int (*test)(struct inode *, void *),
872 void *data, const int wait)
876 spin_lock(&inode_lock);
877 inode = find_inode(sb, head, test, data);
880 spin_unlock(&inode_lock);
882 wait_on_inode(inode);
885 spin_unlock(&inode_lock);
890 * ifind_fast - internal function, you want ilookup() or iget().
891 * @sb: super block of file system to search
892 * @head: head of the list to search
893 * @ino: inode number to search for
895 * ifind_fast() searches for the inode @ino in the inode cache. This is for
896 * file systems where the inode number is sufficient for unique identification
899 * If the inode is in the cache, the inode is returned with an incremented
902 * Otherwise NULL is returned.
904 static struct inode *ifind_fast(struct super_block *sb,
905 struct hlist_head *head, unsigned long ino)
909 spin_lock(&inode_lock);
910 inode = find_inode_fast(sb, head, ino);
913 spin_unlock(&inode_lock);
914 wait_on_inode(inode);
917 spin_unlock(&inode_lock);
922 * ilookup5_nowait - search for an inode in the inode cache
923 * @sb: super block of file system to search
924 * @hashval: hash value (usually inode number) to search for
925 * @test: callback used for comparisons between inodes
926 * @data: opaque data pointer to pass to @test
928 * ilookup5() uses ifind() to search for the inode specified by @hashval and
929 * @data in the inode cache. This is a generalized version of ilookup() for
930 * file systems where the inode number is not sufficient for unique
931 * identification of an inode.
933 * If the inode is in the cache, the inode is returned with an incremented
934 * reference count. Note, the inode lock is not waited upon so you have to be
935 * very careful what you do with the returned inode. You probably should be
936 * using ilookup5() instead.
938 * Otherwise NULL is returned.
940 * Note, @test is called with the inode_lock held, so can't sleep.
942 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
943 int (*test)(struct inode *, void *), void *data)
945 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
947 return ifind(sb, head, test, data, 0);
949 EXPORT_SYMBOL(ilookup5_nowait);
952 * ilookup5 - search for an inode in the inode cache
953 * @sb: super block of file system to search
954 * @hashval: hash value (usually inode number) to search for
955 * @test: callback used for comparisons between inodes
956 * @data: opaque data pointer to pass to @test
958 * ilookup5() uses ifind() to search for the inode specified by @hashval and
959 * @data in the inode cache. This is a generalized version of ilookup() for
960 * file systems where the inode number is not sufficient for unique
961 * identification of an inode.
963 * If the inode is in the cache, the inode lock is waited upon and the inode is
964 * returned with an incremented reference count.
966 * Otherwise NULL is returned.
968 * Note, @test is called with the inode_lock held, so can't sleep.
970 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
971 int (*test)(struct inode *, void *), void *data)
973 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
975 return ifind(sb, head, test, data, 1);
977 EXPORT_SYMBOL(ilookup5);
980 * ilookup - search for an inode in the inode cache
981 * @sb: super block of file system to search
982 * @ino: inode number to search for
984 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
985 * This is for file systems where the inode number is sufficient for unique
986 * identification of an inode.
988 * If the inode is in the cache, the inode is returned with an incremented
991 * Otherwise NULL is returned.
993 struct inode *ilookup(struct super_block *sb, unsigned long ino)
995 struct hlist_head *head = inode_hashtable + hash(sb, ino);
997 return ifind_fast(sb, head, ino);
999 EXPORT_SYMBOL(ilookup);
1002 * iget5_locked - obtain an inode from a mounted file system
1003 * @sb: super block of file system
1004 * @hashval: hash value (usually inode number) to get
1005 * @test: callback used for comparisons between inodes
1006 * @set: callback used to initialize a new struct inode
1007 * @data: opaque data pointer to pass to @test and @set
1009 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1010 * and @data in the inode cache and if present it is returned with an increased
1011 * reference count. This is a generalized version of iget_locked() for file
1012 * systems where the inode number is not sufficient for unique identification
1015 * If the inode is not in cache, get_new_inode() is called to allocate a new
1016 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1017 * file system gets to fill it in before unlocking it via unlock_new_inode().
1019 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1021 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1022 int (*test)(struct inode *, void *),
1023 int (*set)(struct inode *, void *), void *data)
1025 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1026 struct inode *inode;
1028 inode = ifind(sb, head, test, data, 1);
1032 * get_new_inode() will do the right thing, re-trying the search
1033 * in case it had to block at any point.
1035 return get_new_inode(sb, head, test, set, data);
1037 EXPORT_SYMBOL(iget5_locked);
1040 * iget_locked - obtain an inode from a mounted file system
1041 * @sb: super block of file system
1042 * @ino: inode number to get
1044 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1045 * the inode cache and if present it is returned with an increased reference
1046 * count. This is for file systems where the inode number is sufficient for
1047 * unique identification of an inode.
1049 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1050 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1051 * The file system gets to fill it in before unlocking it via
1052 * unlock_new_inode().
1054 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1056 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1057 struct inode *inode;
1059 inode = ifind_fast(sb, head, ino);
1063 * get_new_inode_fast() will do the right thing, re-trying the search
1064 * in case it had to block at any point.
1066 return get_new_inode_fast(sb, head, ino);
1068 EXPORT_SYMBOL(iget_locked);
1070 int insert_inode_locked(struct inode *inode)
1072 struct super_block *sb = inode->i_sb;
1073 ino_t ino = inode->i_ino;
1074 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1076 inode->i_state |= I_NEW;
1078 struct hlist_node *node;
1079 struct inode *old = NULL;
1080 spin_lock(&inode_lock);
1081 hlist_for_each_entry(old, node, head, i_hash) {
1082 if (old->i_ino != ino)
1084 if (old->i_sb != sb)
1086 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1090 if (likely(!node)) {
1091 hlist_add_head(&inode->i_hash, head);
1092 spin_unlock(&inode_lock);
1096 spin_unlock(&inode_lock);
1098 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1105 EXPORT_SYMBOL(insert_inode_locked);
1107 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1108 int (*test)(struct inode *, void *), void *data)
1110 struct super_block *sb = inode->i_sb;
1111 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1113 inode->i_state |= I_NEW;
1116 struct hlist_node *node;
1117 struct inode *old = NULL;
1119 spin_lock(&inode_lock);
1120 hlist_for_each_entry(old, node, head, i_hash) {
1121 if (old->i_sb != sb)
1123 if (!test(old, data))
1125 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1129 if (likely(!node)) {
1130 hlist_add_head(&inode->i_hash, head);
1131 spin_unlock(&inode_lock);
1135 spin_unlock(&inode_lock);
1137 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1144 EXPORT_SYMBOL(insert_inode_locked4);
1147 * __insert_inode_hash - hash an inode
1148 * @inode: unhashed inode
1149 * @hashval: unsigned long value used to locate this object in the
1152 * Add an inode to the inode hash for this superblock.
1154 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1156 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1157 spin_lock(&inode_lock);
1158 hlist_add_head(&inode->i_hash, head);
1159 spin_unlock(&inode_lock);
1161 EXPORT_SYMBOL(__insert_inode_hash);
1164 * remove_inode_hash - remove an inode from the hash
1165 * @inode: inode to unhash
1167 * Remove an inode from the superblock.
1169 void remove_inode_hash(struct inode *inode)
1171 spin_lock(&inode_lock);
1172 hlist_del_init(&inode->i_hash);
1173 spin_unlock(&inode_lock);
1175 EXPORT_SYMBOL(remove_inode_hash);
1178 * Tell the filesystem that this inode is no longer of any interest and should
1179 * be completely destroyed.
1181 * We leave the inode in the inode hash table until *after* the filesystem's
1182 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1183 * instigate) will always find up-to-date information either in the hash or on
1186 * I_FREEING is set so that no-one will take a new reference to the inode while
1187 * it is being deleted.
1189 void generic_delete_inode(struct inode *inode)
1191 const struct super_operations *op = inode->i_sb->s_op;
1193 list_del_init(&inode->i_list);
1194 list_del_init(&inode->i_sb_list);
1195 WARN_ON(inode->i_state & I_NEW);
1196 inode->i_state |= I_FREEING;
1197 inodes_stat.nr_inodes--;
1198 spin_unlock(&inode_lock);
1200 if (op->delete_inode) {
1201 void (*delete)(struct inode *) = op->delete_inode;
1202 /* Filesystems implementing their own
1203 * s_op->delete_inode are required to call
1204 * truncate_inode_pages and clear_inode()
1208 truncate_inode_pages(&inode->i_data, 0);
1211 spin_lock(&inode_lock);
1212 hlist_del_init(&inode->i_hash);
1213 spin_unlock(&inode_lock);
1214 wake_up_inode(inode);
1215 BUG_ON(inode->i_state != I_CLEAR);
1216 destroy_inode(inode);
1218 EXPORT_SYMBOL(generic_delete_inode);
1221 * generic_detach_inode - remove inode from inode lists
1222 * @inode: inode to remove
1224 * Remove inode from inode lists, write it if it's dirty. This is just an
1225 * internal VFS helper exported for hugetlbfs. Do not use!
1227 * Returns 1 if inode should be completely destroyed.
1229 int generic_detach_inode(struct inode *inode)
1231 struct super_block *sb = inode->i_sb;
1233 if (!hlist_unhashed(&inode->i_hash)) {
1234 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1235 list_move(&inode->i_list, &inode_unused);
1236 inodes_stat.nr_unused++;
1237 if (sb->s_flags & MS_ACTIVE) {
1238 spin_unlock(&inode_lock);
1241 WARN_ON(inode->i_state & I_NEW);
1242 inode->i_state |= I_WILL_FREE;
1243 spin_unlock(&inode_lock);
1244 write_inode_now(inode, 1);
1245 spin_lock(&inode_lock);
1246 WARN_ON(inode->i_state & I_NEW);
1247 inode->i_state &= ~I_WILL_FREE;
1248 inodes_stat.nr_unused--;
1249 hlist_del_init(&inode->i_hash);
1251 list_del_init(&inode->i_list);
1252 list_del_init(&inode->i_sb_list);
1253 WARN_ON(inode->i_state & I_NEW);
1254 inode->i_state |= I_FREEING;
1255 inodes_stat.nr_inodes--;
1256 spin_unlock(&inode_lock);
1259 EXPORT_SYMBOL_GPL(generic_detach_inode);
1261 static void generic_forget_inode(struct inode *inode)
1263 if (!generic_detach_inode(inode))
1265 if (inode->i_data.nrpages)
1266 truncate_inode_pages(&inode->i_data, 0);
1268 wake_up_inode(inode);
1269 destroy_inode(inode);
1273 * Normal UNIX filesystem behaviour: delete the
1274 * inode when the usage count drops to zero, and
1277 void generic_drop_inode(struct inode *inode)
1279 if (!inode->i_nlink)
1280 generic_delete_inode(inode);
1282 generic_forget_inode(inode);
1284 EXPORT_SYMBOL_GPL(generic_drop_inode);
1287 * Called when we're dropping the last reference
1290 * Call the FS "drop()" function, defaulting to
1291 * the legacy UNIX filesystem behaviour..
1293 * NOTE! NOTE! NOTE! We're called with the inode lock
1294 * held, and the drop function is supposed to release
1297 static inline void iput_final(struct inode *inode)
1299 const struct super_operations *op = inode->i_sb->s_op;
1300 void (*drop)(struct inode *) = generic_drop_inode;
1302 if (op && op->drop_inode)
1303 drop = op->drop_inode;
1308 * iput - put an inode
1309 * @inode: inode to put
1311 * Puts an inode, dropping its usage count. If the inode use count hits
1312 * zero, the inode is then freed and may also be destroyed.
1314 * Consequently, iput() can sleep.
1316 void iput(struct inode *inode)
1319 BUG_ON(inode->i_state == I_CLEAR);
1321 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1325 EXPORT_SYMBOL(iput);
1328 * bmap - find a block number in a file
1329 * @inode: inode of file
1330 * @block: block to find
1332 * Returns the block number on the device holding the inode that
1333 * is the disk block number for the block of the file requested.
1334 * That is, asked for block 4 of inode 1 the function will return the
1335 * disk block relative to the disk start that holds that block of the
1338 sector_t bmap(struct inode *inode, sector_t block)
1341 if (inode->i_mapping->a_ops->bmap)
1342 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1345 EXPORT_SYMBOL(bmap);
1348 * With relative atime, only update atime if the previous atime is
1349 * earlier than either the ctime or mtime or if at least a day has
1350 * passed since the last atime update.
1352 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1353 struct timespec now)
1356 if (!(mnt->mnt_flags & MNT_RELATIME))
1359 * Is mtime younger than atime? If yes, update atime:
1361 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1364 * Is ctime younger than atime? If yes, update atime:
1366 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1370 * Is the previous atime value older than a day? If yes,
1373 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1376 * Good, we can skip the atime update:
1382 * touch_atime - update the access time
1383 * @mnt: mount the inode is accessed on
1384 * @dentry: dentry accessed
1386 * Update the accessed time on an inode and mark it for writeback.
1387 * This function automatically handles read only file systems and media,
1388 * as well as the "noatime" flag and inode specific "noatime" markers.
1390 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1392 struct inode *inode = dentry->d_inode;
1393 struct timespec now;
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 if (mnt_want_write(mnt))
1418 inode->i_atime = now;
1419 mark_inode_dirty_sync(inode);
1420 mnt_drop_write(mnt);
1422 EXPORT_SYMBOL(touch_atime);
1425 * file_update_time - update mtime and ctime time
1426 * @file: file accessed
1428 * Update the mtime and ctime members of an inode and mark the inode
1429 * for writeback. Note that this function is meant exclusively for
1430 * usage in the file write path of filesystems, and filesystems may
1431 * choose to explicitly ignore update via this function with the
1432 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1433 * timestamps are handled by the server.
1436 void file_update_time(struct file *file)
1438 struct inode *inode = file->f_path.dentry->d_inode;
1439 struct timespec now;
1440 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1442 /* First try to exhaust all avenues to not sync */
1443 if (IS_NOCMTIME(inode))
1446 now = current_fs_time(inode->i_sb);
1447 if (!timespec_equal(&inode->i_mtime, &now))
1450 if (!timespec_equal(&inode->i_ctime, &now))
1453 if (IS_I_VERSION(inode))
1454 sync_it |= S_VERSION;
1459 /* Finally allowed to write? Takes lock. */
1460 if (mnt_want_write_file(file))
1463 /* Only change inode inside the lock region */
1464 if (sync_it & S_VERSION)
1465 inode_inc_iversion(inode);
1466 if (sync_it & S_CTIME)
1467 inode->i_ctime = now;
1468 if (sync_it & S_MTIME)
1469 inode->i_mtime = now;
1470 mark_inode_dirty_sync(inode);
1471 mnt_drop_write(file->f_path.mnt);
1473 EXPORT_SYMBOL(file_update_time);
1475 int inode_needs_sync(struct inode *inode)
1479 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1483 EXPORT_SYMBOL(inode_needs_sync);
1485 int inode_wait(void *word)
1490 EXPORT_SYMBOL(inode_wait);
1493 * If we try to find an inode in the inode hash while it is being
1494 * deleted, we have to wait until the filesystem completes its
1495 * deletion before reporting that it isn't found. This function waits
1496 * until the deletion _might_ have completed. Callers are responsible
1497 * to recheck inode state.
1499 * It doesn't matter if I_NEW is not set initially, a call to
1500 * wake_up_inode() after removing from the hash list will DTRT.
1502 * This is called with inode_lock held.
1504 static void __wait_on_freeing_inode(struct inode *inode)
1506 wait_queue_head_t *wq;
1507 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1508 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1509 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1510 spin_unlock(&inode_lock);
1512 finish_wait(wq, &wait.wait);
1513 spin_lock(&inode_lock);
1516 static __initdata unsigned long ihash_entries;
1517 static int __init set_ihash_entries(char *str)
1521 ihash_entries = simple_strtoul(str, &str, 0);
1524 __setup("ihash_entries=", set_ihash_entries);
1527 * Initialize the waitqueues and inode hash table.
1529 void __init inode_init_early(void)
1533 /* If hashes are distributed across NUMA nodes, defer
1534 * hash allocation until vmalloc space is available.
1540 alloc_large_system_hash("Inode-cache",
1541 sizeof(struct hlist_head),
1549 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1550 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1553 void __init inode_init(void)
1557 /* inode slab cache */
1558 inode_cachep = kmem_cache_create("inode_cache",
1559 sizeof(struct inode),
1561 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1564 register_shrinker(&icache_shrinker);
1566 /* Hash may have been set up in inode_init_early */
1571 alloc_large_system_hash("Inode-cache",
1572 sizeof(struct hlist_head),
1580 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1581 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1584 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1586 inode->i_mode = mode;
1587 if (S_ISCHR(mode)) {
1588 inode->i_fop = &def_chr_fops;
1589 inode->i_rdev = rdev;
1590 } else if (S_ISBLK(mode)) {
1591 inode->i_fop = &def_blk_fops;
1592 inode->i_rdev = rdev;
1593 } else if (S_ISFIFO(mode))
1594 inode->i_fop = &def_fifo_fops;
1595 else if (S_ISSOCK(mode))
1596 inode->i_fop = &bad_sock_fops;
1598 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1599 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1602 EXPORT_SYMBOL(init_special_inode);
1605 * Init uid,gid,mode for new inode according to posix standards
1607 * @dir: Directory inode
1608 * @mode: mode of the new inode
1610 void inode_init_owner(struct inode *inode, const struct inode *dir,
1613 inode->i_uid = current_fsuid();
1614 if (dir && dir->i_mode & S_ISGID) {
1615 inode->i_gid = dir->i_gid;
1619 inode->i_gid = current_fsgid();
1620 inode->i_mode = mode;
1622 EXPORT_SYMBOL(inode_init_owner);