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 percpu_counter nr_inodes __cacheline_aligned_in_smp;
107 static struct percpu_counter nr_inodes_unused __cacheline_aligned_in_smp;
109 static struct kmem_cache *inode_cachep __read_mostly;
111 static inline int get_nr_inodes(void)
113 return percpu_counter_sum_positive(&nr_inodes);
116 static inline int get_nr_inodes_unused(void)
118 return percpu_counter_sum_positive(&nr_inodes_unused);
121 int get_nr_dirty_inodes(void)
123 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
124 return nr_dirty > 0 ? nr_dirty : 0;
129 * Handle nr_inode sysctl
132 int proc_nr_inodes(ctl_table *table, int write,
133 void __user *buffer, size_t *lenp, loff_t *ppos)
135 inodes_stat.nr_inodes = get_nr_inodes();
136 inodes_stat.nr_unused = get_nr_inodes_unused();
137 return proc_dointvec(table, write, buffer, lenp, ppos);
141 static void wake_up_inode(struct inode *inode)
144 * Prevent speculative execution through spin_unlock(&inode_lock);
147 wake_up_bit(&inode->i_state, __I_NEW);
151 * inode_init_always - perform inode structure intialisation
152 * @sb: superblock inode belongs to
153 * @inode: inode to initialise
155 * These are initializations that need to be done on every inode
156 * allocation as the fields are not initialised by slab allocation.
158 int inode_init_always(struct super_block *sb, struct inode *inode)
160 static const struct address_space_operations empty_aops;
161 static const struct inode_operations empty_iops;
162 static const struct file_operations empty_fops;
163 struct address_space *const mapping = &inode->i_data;
166 inode->i_blkbits = sb->s_blocksize_bits;
168 atomic_set(&inode->i_count, 1);
169 inode->i_op = &empty_iops;
170 inode->i_fop = &empty_fops;
174 atomic_set(&inode->i_writecount, 0);
178 inode->i_generation = 0;
180 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
182 inode->i_pipe = NULL;
183 inode->i_bdev = NULL;
184 inode->i_cdev = NULL;
186 inode->dirtied_when = 0;
188 if (security_inode_alloc(inode))
190 spin_lock_init(&inode->i_lock);
191 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
193 mutex_init(&inode->i_mutex);
194 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
196 init_rwsem(&inode->i_alloc_sem);
197 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
199 mapping->a_ops = &empty_aops;
200 mapping->host = inode;
202 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
203 mapping->assoc_mapping = NULL;
204 mapping->backing_dev_info = &default_backing_dev_info;
205 mapping->writeback_index = 0;
208 * If the block_device provides a backing_dev_info for client
209 * inodes then use that. Otherwise the inode share the bdev's
213 struct backing_dev_info *bdi;
215 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
216 mapping->backing_dev_info = bdi;
218 inode->i_private = NULL;
219 inode->i_mapping = mapping;
220 #ifdef CONFIG_FS_POSIX_ACL
221 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
224 #ifdef CONFIG_FSNOTIFY
225 inode->i_fsnotify_mask = 0;
228 percpu_counter_inc(&nr_inodes);
234 EXPORT_SYMBOL(inode_init_always);
236 static struct inode *alloc_inode(struct super_block *sb)
240 if (sb->s_op->alloc_inode)
241 inode = sb->s_op->alloc_inode(sb);
243 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
248 if (unlikely(inode_init_always(sb, inode))) {
249 if (inode->i_sb->s_op->destroy_inode)
250 inode->i_sb->s_op->destroy_inode(inode);
252 kmem_cache_free(inode_cachep, inode);
259 void __destroy_inode(struct inode *inode)
261 BUG_ON(inode_has_buffers(inode));
262 security_inode_free(inode);
263 fsnotify_inode_delete(inode);
264 #ifdef CONFIG_FS_POSIX_ACL
265 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
266 posix_acl_release(inode->i_acl);
267 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
268 posix_acl_release(inode->i_default_acl);
270 percpu_counter_dec(&nr_inodes);
272 EXPORT_SYMBOL(__destroy_inode);
274 static void destroy_inode(struct inode *inode)
276 __destroy_inode(inode);
277 if (inode->i_sb->s_op->destroy_inode)
278 inode->i_sb->s_op->destroy_inode(inode);
280 kmem_cache_free(inode_cachep, (inode));
284 * These are initializations that only need to be done
285 * once, because the fields are idempotent across use
286 * of the inode, so let the slab aware of that.
288 void inode_init_once(struct inode *inode)
290 memset(inode, 0, sizeof(*inode));
291 INIT_HLIST_NODE(&inode->i_hash);
292 INIT_LIST_HEAD(&inode->i_dentry);
293 INIT_LIST_HEAD(&inode->i_devices);
294 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
295 spin_lock_init(&inode->i_data.tree_lock);
296 spin_lock_init(&inode->i_data.i_mmap_lock);
297 INIT_LIST_HEAD(&inode->i_data.private_list);
298 spin_lock_init(&inode->i_data.private_lock);
299 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
300 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
301 i_size_ordered_init(inode);
302 #ifdef CONFIG_FSNOTIFY
303 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
306 EXPORT_SYMBOL(inode_init_once);
308 static void init_once(void *foo)
310 struct inode *inode = (struct inode *) foo;
312 inode_init_once(inode);
316 * inode_lock must be held
318 void __iget(struct inode *inode)
320 if (atomic_inc_return(&inode->i_count) != 1)
323 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
324 list_move(&inode->i_list, &inode_in_use);
325 percpu_counter_dec(&nr_inodes_unused);
328 void end_writeback(struct inode *inode)
331 BUG_ON(inode->i_data.nrpages);
332 BUG_ON(!list_empty(&inode->i_data.private_list));
333 BUG_ON(!(inode->i_state & I_FREEING));
334 BUG_ON(inode->i_state & I_CLEAR);
335 inode_sync_wait(inode);
336 inode->i_state = I_FREEING | I_CLEAR;
338 EXPORT_SYMBOL(end_writeback);
340 static void evict(struct inode *inode)
342 const struct super_operations *op = inode->i_sb->s_op;
344 if (op->evict_inode) {
345 op->evict_inode(inode);
347 if (inode->i_data.nrpages)
348 truncate_inode_pages(&inode->i_data, 0);
349 end_writeback(inode);
351 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
353 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
358 * dispose_list - dispose of the contents of a local list
359 * @head: the head of the list to free
361 * Dispose-list gets a local list with local inodes in it, so it doesn't
362 * need to worry about list corruption and SMP locks.
364 static void dispose_list(struct list_head *head)
366 while (!list_empty(head)) {
369 inode = list_first_entry(head, struct inode, i_list);
370 list_del(&inode->i_list);
374 spin_lock(&inode_lock);
375 hlist_del_init(&inode->i_hash);
376 list_del_init(&inode->i_sb_list);
377 spin_unlock(&inode_lock);
379 wake_up_inode(inode);
380 destroy_inode(inode);
385 * Invalidate all inodes for a device.
387 static int invalidate_list(struct list_head *head, struct list_head *dispose)
389 struct list_head *next;
394 struct list_head *tmp = next;
398 * We can reschedule here without worrying about the list's
399 * consistency because the per-sb list of inodes must not
400 * change during umount anymore, and because iprune_sem keeps
401 * shrink_icache_memory() away.
403 cond_resched_lock(&inode_lock);
408 inode = list_entry(tmp, struct inode, i_sb_list);
409 if (inode->i_state & I_NEW)
411 invalidate_inode_buffers(inode);
412 if (!atomic_read(&inode->i_count)) {
413 list_move(&inode->i_list, dispose);
414 WARN_ON(inode->i_state & I_NEW);
415 inode->i_state |= I_FREEING;
416 percpu_counter_dec(&nr_inodes_unused);
425 * invalidate_inodes - discard the inodes on a device
428 * Discard all of the inodes for a given superblock. If the discard
429 * fails because there are busy inodes then a non zero value is returned.
430 * If the discard is successful all the inodes have been discarded.
432 int invalidate_inodes(struct super_block *sb)
435 LIST_HEAD(throw_away);
437 down_write(&iprune_sem);
438 spin_lock(&inode_lock);
439 fsnotify_unmount_inodes(&sb->s_inodes);
440 busy = invalidate_list(&sb->s_inodes, &throw_away);
441 spin_unlock(&inode_lock);
443 dispose_list(&throw_away);
444 up_write(&iprune_sem);
449 static int can_unuse(struct inode *inode)
453 if (inode_has_buffers(inode))
455 if (atomic_read(&inode->i_count))
457 if (inode->i_data.nrpages)
463 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
464 * a temporary list and then are freed outside inode_lock by dispose_list().
466 * Any inodes which are pinned purely because of attached pagecache have their
467 * pagecache removed. We expect the final iput() on that inode to add it to
468 * the front of the inode_unused list. So look for it there and if the
469 * inode is still freeable, proceed. The right inode is found 99.9% of the
470 * time in testing on a 4-way.
472 * If the inode has metadata buffers attached to mapping->private_list then
473 * try to remove them.
475 static void prune_icache(int nr_to_scan)
479 unsigned long reap = 0;
481 down_read(&iprune_sem);
482 spin_lock(&inode_lock);
483 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
486 if (list_empty(&inode_unused))
489 inode = list_entry(inode_unused.prev, struct inode, i_list);
491 if (inode->i_state || atomic_read(&inode->i_count)) {
492 list_move(&inode->i_list, &inode_unused);
495 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
497 spin_unlock(&inode_lock);
498 if (remove_inode_buffers(inode))
499 reap += invalidate_mapping_pages(&inode->i_data,
502 spin_lock(&inode_lock);
504 if (inode != list_entry(inode_unused.next,
505 struct inode, i_list))
506 continue; /* wrong inode or list_empty */
507 if (!can_unuse(inode))
510 list_move(&inode->i_list, &freeable);
511 WARN_ON(inode->i_state & I_NEW);
512 inode->i_state |= I_FREEING;
513 percpu_counter_dec(&nr_inodes_unused);
515 if (current_is_kswapd())
516 __count_vm_events(KSWAPD_INODESTEAL, reap);
518 __count_vm_events(PGINODESTEAL, reap);
519 spin_unlock(&inode_lock);
521 dispose_list(&freeable);
522 up_read(&iprune_sem);
526 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
527 * "unused" means that no dentries are referring to the inodes: the files are
528 * not open and the dcache references to those inodes have already been
531 * This function is passed the number of inodes to scan, and it returns the
532 * total number of remaining possibly-reclaimable inodes.
534 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
538 * Nasty deadlock avoidance. We may hold various FS locks,
539 * and we don't want to recurse into the FS that called us
540 * in clear_inode() and friends..
542 if (!(gfp_mask & __GFP_FS))
546 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
549 static struct shrinker icache_shrinker = {
550 .shrink = shrink_icache_memory,
551 .seeks = DEFAULT_SEEKS,
554 static void __wait_on_freeing_inode(struct inode *inode);
556 * Called with the inode lock held.
557 * NOTE: we are not increasing the inode-refcount, you must call __iget()
558 * by hand after calling find_inode now! This simplifies iunique and won't
559 * add any additional branch in the common code.
561 static struct inode *find_inode(struct super_block *sb,
562 struct hlist_head *head,
563 int (*test)(struct inode *, void *),
566 struct hlist_node *node;
567 struct inode *inode = NULL;
570 hlist_for_each_entry(inode, node, head, i_hash) {
571 if (inode->i_sb != sb)
573 if (!test(inode, data))
575 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
576 __wait_on_freeing_inode(inode);
581 return node ? inode : NULL;
585 * find_inode_fast is the fast path version of find_inode, see the comment at
586 * iget_locked for details.
588 static struct inode *find_inode_fast(struct super_block *sb,
589 struct hlist_head *head, unsigned long ino)
591 struct hlist_node *node;
592 struct inode *inode = NULL;
595 hlist_for_each_entry(inode, node, head, i_hash) {
596 if (inode->i_ino != ino)
598 if (inode->i_sb != sb)
600 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
601 __wait_on_freeing_inode(inode);
606 return node ? inode : NULL;
609 static unsigned long hash(struct super_block *sb, unsigned long hashval)
613 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
615 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
616 return tmp & I_HASHMASK;
620 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
623 list_add(&inode->i_list, &inode_in_use);
624 list_add(&inode->i_sb_list, &sb->s_inodes);
626 hlist_add_head(&inode->i_hash, head);
630 * inode_add_to_lists - add a new inode to relevant lists
631 * @sb: superblock inode belongs to
632 * @inode: inode to mark in use
634 * When an inode is allocated it needs to be accounted for, added to the in use
635 * list, the owning superblock and the inode hash. This needs to be done under
636 * the inode_lock, so export a function to do this rather than the inode lock
637 * itself. We calculate the hash list to add to here so it is all internal
638 * which requires the caller to have already set up the inode number in the
641 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
643 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
645 spin_lock(&inode_lock);
646 __inode_add_to_lists(sb, head, inode);
647 spin_unlock(&inode_lock);
649 EXPORT_SYMBOL_GPL(inode_add_to_lists);
652 * new_inode - obtain an inode
655 * Allocates a new inode for given superblock. The default gfp_mask
656 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
657 * If HIGHMEM pages are unsuitable or it is known that pages allocated
658 * for the page cache are not reclaimable or migratable,
659 * mapping_set_gfp_mask() must be called with suitable flags on the
660 * newly created inode's mapping
663 struct inode *new_inode(struct super_block *sb)
666 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
667 * error if st_ino won't fit in target struct field. Use 32bit counter
668 * here to attempt to avoid that.
670 static unsigned int last_ino;
673 spin_lock_prefetch(&inode_lock);
675 inode = alloc_inode(sb);
677 spin_lock(&inode_lock);
678 __inode_add_to_lists(sb, NULL, inode);
679 inode->i_ino = ++last_ino;
681 spin_unlock(&inode_lock);
685 EXPORT_SYMBOL(new_inode);
687 void unlock_new_inode(struct inode *inode)
689 #ifdef CONFIG_DEBUG_LOCK_ALLOC
690 if (S_ISDIR(inode->i_mode)) {
691 struct file_system_type *type = inode->i_sb->s_type;
693 /* Set new key only if filesystem hasn't already changed it */
694 if (!lockdep_match_class(&inode->i_mutex,
695 &type->i_mutex_key)) {
697 * ensure nobody is actually holding i_mutex
699 mutex_destroy(&inode->i_mutex);
700 mutex_init(&inode->i_mutex);
701 lockdep_set_class(&inode->i_mutex,
702 &type->i_mutex_dir_key);
707 * This is special! We do not need the spinlock when clearing I_NEW,
708 * because we're guaranteed that nobody else tries to do anything about
709 * the state of the inode when it is locked, as we just created it (so
710 * there can be no old holders that haven't tested I_NEW).
711 * However we must emit the memory barrier so that other CPUs reliably
712 * see the clearing of I_NEW after the other inode initialisation has
716 WARN_ON(!(inode->i_state & I_NEW));
717 inode->i_state &= ~I_NEW;
718 wake_up_inode(inode);
720 EXPORT_SYMBOL(unlock_new_inode);
723 * This is called without the inode lock held.. Be careful.
725 * We no longer cache the sb_flags in i_flags - see fs.h
726 * -- rmk@arm.uk.linux.org
728 static struct inode *get_new_inode(struct super_block *sb,
729 struct hlist_head *head,
730 int (*test)(struct inode *, void *),
731 int (*set)(struct inode *, void *),
736 inode = alloc_inode(sb);
740 spin_lock(&inode_lock);
741 /* We released the lock, so.. */
742 old = find_inode(sb, head, test, data);
744 if (set(inode, data))
747 __inode_add_to_lists(sb, head, inode);
748 inode->i_state = I_NEW;
749 spin_unlock(&inode_lock);
751 /* Return the locked inode with I_NEW set, the
752 * caller is responsible for filling in the contents
758 * Uhhuh, somebody else created the same inode under
759 * us. Use the old inode instead of the one we just
763 spin_unlock(&inode_lock);
764 destroy_inode(inode);
766 wait_on_inode(inode);
771 spin_unlock(&inode_lock);
772 destroy_inode(inode);
777 * get_new_inode_fast is the fast path version of get_new_inode, see the
778 * comment at iget_locked for details.
780 static struct inode *get_new_inode_fast(struct super_block *sb,
781 struct hlist_head *head, unsigned long ino)
785 inode = alloc_inode(sb);
789 spin_lock(&inode_lock);
790 /* We released the lock, so.. */
791 old = find_inode_fast(sb, head, ino);
794 __inode_add_to_lists(sb, head, inode);
795 inode->i_state = I_NEW;
796 spin_unlock(&inode_lock);
798 /* Return the locked inode with I_NEW set, the
799 * caller is responsible for filling in the contents
805 * Uhhuh, somebody else created the same inode under
806 * us. Use the old inode instead of the one we just
810 spin_unlock(&inode_lock);
811 destroy_inode(inode);
813 wait_on_inode(inode);
819 * iunique - get a unique inode number
821 * @max_reserved: highest reserved inode number
823 * Obtain an inode number that is unique on the system for a given
824 * superblock. This is used by file systems that have no natural
825 * permanent inode numbering system. An inode number is returned that
826 * is higher than the reserved limit but unique.
829 * With a large number of inodes live on the file system this function
830 * currently becomes quite slow.
832 ino_t iunique(struct super_block *sb, ino_t max_reserved)
835 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
836 * error if st_ino won't fit in target struct field. Use 32bit counter
837 * here to attempt to avoid that.
839 static unsigned int counter;
841 struct hlist_head *head;
844 spin_lock(&inode_lock);
846 if (counter <= max_reserved)
847 counter = max_reserved + 1;
849 head = inode_hashtable + hash(sb, res);
850 inode = find_inode_fast(sb, head, res);
851 } while (inode != NULL);
852 spin_unlock(&inode_lock);
856 EXPORT_SYMBOL(iunique);
858 struct inode *igrab(struct inode *inode)
860 spin_lock(&inode_lock);
861 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
865 * Handle the case where s_op->clear_inode is not been
866 * called yet, and somebody is calling igrab
867 * while the inode is getting freed.
870 spin_unlock(&inode_lock);
873 EXPORT_SYMBOL(igrab);
876 * ifind - internal function, you want ilookup5() or iget5().
877 * @sb: super block of file system to search
878 * @head: the head of the list to search
879 * @test: callback used for comparisons between inodes
880 * @data: opaque data pointer to pass to @test
881 * @wait: if true wait for the inode to be unlocked, if false do not
883 * ifind() searches for the inode specified by @data in the inode
884 * cache. This is a generalized version of ifind_fast() for file systems where
885 * the inode number is not sufficient for unique identification of an inode.
887 * If the inode is in the cache, the inode is returned with an incremented
890 * Otherwise NULL is returned.
892 * Note, @test is called with the inode_lock held, so can't sleep.
894 static struct inode *ifind(struct super_block *sb,
895 struct hlist_head *head, int (*test)(struct inode *, void *),
896 void *data, const int wait)
900 spin_lock(&inode_lock);
901 inode = find_inode(sb, head, test, data);
904 spin_unlock(&inode_lock);
906 wait_on_inode(inode);
909 spin_unlock(&inode_lock);
914 * ifind_fast - internal function, you want ilookup() or iget().
915 * @sb: super block of file system to search
916 * @head: head of the list to search
917 * @ino: inode number to search for
919 * ifind_fast() searches for the inode @ino in the inode cache. This is for
920 * file systems where the inode number is sufficient for unique identification
923 * If the inode is in the cache, the inode is returned with an incremented
926 * Otherwise NULL is returned.
928 static struct inode *ifind_fast(struct super_block *sb,
929 struct hlist_head *head, unsigned long ino)
933 spin_lock(&inode_lock);
934 inode = find_inode_fast(sb, head, ino);
937 spin_unlock(&inode_lock);
938 wait_on_inode(inode);
941 spin_unlock(&inode_lock);
946 * ilookup5_nowait - search for an inode in the inode cache
947 * @sb: super block of file system to search
948 * @hashval: hash value (usually inode number) to search for
949 * @test: callback used for comparisons between inodes
950 * @data: opaque data pointer to pass to @test
952 * ilookup5() uses ifind() to search for the inode specified by @hashval and
953 * @data in the inode cache. This is a generalized version of ilookup() for
954 * file systems where the inode number is not sufficient for unique
955 * identification of an inode.
957 * If the inode is in the cache, the inode is returned with an incremented
958 * reference count. Note, the inode lock is not waited upon so you have to be
959 * very careful what you do with the returned inode. You probably should be
960 * using ilookup5() instead.
962 * Otherwise NULL is returned.
964 * Note, @test is called with the inode_lock held, so can't sleep.
966 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
967 int (*test)(struct inode *, void *), void *data)
969 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
971 return ifind(sb, head, test, data, 0);
973 EXPORT_SYMBOL(ilookup5_nowait);
976 * ilookup5 - search for an inode in the inode cache
977 * @sb: super block of file system to search
978 * @hashval: hash value (usually inode number) to search for
979 * @test: callback used for comparisons between inodes
980 * @data: opaque data pointer to pass to @test
982 * ilookup5() uses ifind() to search for the inode specified by @hashval and
983 * @data in the inode cache. This is a generalized version of ilookup() for
984 * file systems where the inode number is not sufficient for unique
985 * identification of an inode.
987 * If the inode is in the cache, the inode lock is waited upon and the inode is
988 * returned with an incremented reference count.
990 * Otherwise NULL is returned.
992 * Note, @test is called with the inode_lock held, so can't sleep.
994 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
995 int (*test)(struct inode *, void *), void *data)
997 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
999 return ifind(sb, head, test, data, 1);
1001 EXPORT_SYMBOL(ilookup5);
1004 * ilookup - search for an inode in the inode cache
1005 * @sb: super block of file system to search
1006 * @ino: inode number to search for
1008 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1009 * This is for file systems where the inode number is sufficient for unique
1010 * identification of an inode.
1012 * If the inode is in the cache, the inode is returned with an incremented
1015 * Otherwise NULL is returned.
1017 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1019 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1021 return ifind_fast(sb, head, ino);
1023 EXPORT_SYMBOL(ilookup);
1026 * iget5_locked - obtain an inode from a mounted file system
1027 * @sb: super block of file system
1028 * @hashval: hash value (usually inode number) to get
1029 * @test: callback used for comparisons between inodes
1030 * @set: callback used to initialize a new struct inode
1031 * @data: opaque data pointer to pass to @test and @set
1033 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1034 * and @data in the inode cache and if present it is returned with an increased
1035 * reference count. This is a generalized version of iget_locked() for file
1036 * systems where the inode number is not sufficient for unique identification
1039 * If the inode is not in cache, get_new_inode() is called to allocate a new
1040 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1041 * file system gets to fill it in before unlocking it via unlock_new_inode().
1043 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1045 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1046 int (*test)(struct inode *, void *),
1047 int (*set)(struct inode *, void *), void *data)
1049 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1050 struct inode *inode;
1052 inode = ifind(sb, head, test, data, 1);
1056 * get_new_inode() will do the right thing, re-trying the search
1057 * in case it had to block at any point.
1059 return get_new_inode(sb, head, test, set, data);
1061 EXPORT_SYMBOL(iget5_locked);
1064 * iget_locked - obtain an inode from a mounted file system
1065 * @sb: super block of file system
1066 * @ino: inode number to get
1068 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1069 * the inode cache and if present it is returned with an increased reference
1070 * count. This is for file systems where the inode number is sufficient for
1071 * unique identification of an inode.
1073 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1074 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1075 * The file system gets to fill it in before unlocking it via
1076 * unlock_new_inode().
1078 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1080 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1081 struct inode *inode;
1083 inode = ifind_fast(sb, head, ino);
1087 * get_new_inode_fast() will do the right thing, re-trying the search
1088 * in case it had to block at any point.
1090 return get_new_inode_fast(sb, head, ino);
1092 EXPORT_SYMBOL(iget_locked);
1094 int insert_inode_locked(struct inode *inode)
1096 struct super_block *sb = inode->i_sb;
1097 ino_t ino = inode->i_ino;
1098 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1100 inode->i_state |= I_NEW;
1102 struct hlist_node *node;
1103 struct inode *old = NULL;
1104 spin_lock(&inode_lock);
1105 hlist_for_each_entry(old, node, head, i_hash) {
1106 if (old->i_ino != ino)
1108 if (old->i_sb != sb)
1110 if (old->i_state & (I_FREEING|I_WILL_FREE))
1114 if (likely(!node)) {
1115 hlist_add_head(&inode->i_hash, head);
1116 spin_unlock(&inode_lock);
1120 spin_unlock(&inode_lock);
1122 if (unlikely(!inode_unhashed(old))) {
1129 EXPORT_SYMBOL(insert_inode_locked);
1131 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1132 int (*test)(struct inode *, void *), void *data)
1134 struct super_block *sb = inode->i_sb;
1135 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1137 inode->i_state |= I_NEW;
1140 struct hlist_node *node;
1141 struct inode *old = NULL;
1143 spin_lock(&inode_lock);
1144 hlist_for_each_entry(old, node, head, i_hash) {
1145 if (old->i_sb != sb)
1147 if (!test(old, data))
1149 if (old->i_state & (I_FREEING|I_WILL_FREE))
1153 if (likely(!node)) {
1154 hlist_add_head(&inode->i_hash, head);
1155 spin_unlock(&inode_lock);
1159 spin_unlock(&inode_lock);
1161 if (unlikely(!inode_unhashed(old))) {
1168 EXPORT_SYMBOL(insert_inode_locked4);
1171 * __insert_inode_hash - hash an inode
1172 * @inode: unhashed inode
1173 * @hashval: unsigned long value used to locate this object in the
1176 * Add an inode to the inode hash for this superblock.
1178 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1180 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1181 spin_lock(&inode_lock);
1182 hlist_add_head(&inode->i_hash, head);
1183 spin_unlock(&inode_lock);
1185 EXPORT_SYMBOL(__insert_inode_hash);
1188 * remove_inode_hash - remove an inode from the hash
1189 * @inode: inode to unhash
1191 * Remove an inode from the superblock.
1193 void remove_inode_hash(struct inode *inode)
1195 spin_lock(&inode_lock);
1196 hlist_del_init(&inode->i_hash);
1197 spin_unlock(&inode_lock);
1199 EXPORT_SYMBOL(remove_inode_hash);
1201 int generic_delete_inode(struct inode *inode)
1205 EXPORT_SYMBOL(generic_delete_inode);
1208 * Normal UNIX filesystem behaviour: delete the
1209 * inode when the usage count drops to zero, and
1212 int generic_drop_inode(struct inode *inode)
1214 return !inode->i_nlink || inode_unhashed(inode);
1216 EXPORT_SYMBOL_GPL(generic_drop_inode);
1219 * Called when we're dropping the last reference
1222 * Call the FS "drop_inode()" function, defaulting to
1223 * the legacy UNIX filesystem behaviour. If it tells
1224 * us to evict inode, do so. Otherwise, retain inode
1225 * in cache if fs is alive, sync and evict if fs is
1228 static void iput_final(struct inode *inode)
1230 struct super_block *sb = inode->i_sb;
1231 const struct super_operations *op = inode->i_sb->s_op;
1234 if (op && op->drop_inode)
1235 drop = op->drop_inode(inode);
1237 drop = generic_drop_inode(inode);
1240 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1241 list_move(&inode->i_list, &inode_unused);
1242 percpu_counter_inc(&nr_inodes_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 percpu_counter_dec(&nr_inodes_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 spin_unlock(&inode_lock);
1263 spin_lock(&inode_lock);
1264 hlist_del_init(&inode->i_hash);
1265 spin_unlock(&inode_lock);
1266 wake_up_inode(inode);
1267 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1268 destroy_inode(inode);
1272 * iput - put an inode
1273 * @inode: inode to put
1275 * Puts an inode, dropping its usage count. If the inode use count hits
1276 * zero, the inode is then freed and may also be destroyed.
1278 * Consequently, iput() can sleep.
1280 void iput(struct inode *inode)
1283 BUG_ON(inode->i_state & I_CLEAR);
1285 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1289 EXPORT_SYMBOL(iput);
1292 * bmap - find a block number in a file
1293 * @inode: inode of file
1294 * @block: block to find
1296 * Returns the block number on the device holding the inode that
1297 * is the disk block number for the block of the file requested.
1298 * That is, asked for block 4 of inode 1 the function will return the
1299 * disk block relative to the disk start that holds that block of the
1302 sector_t bmap(struct inode *inode, sector_t block)
1305 if (inode->i_mapping->a_ops->bmap)
1306 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1309 EXPORT_SYMBOL(bmap);
1312 * With relative atime, only update atime if the previous atime is
1313 * earlier than either the ctime or mtime or if at least a day has
1314 * passed since the last atime update.
1316 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1317 struct timespec now)
1320 if (!(mnt->mnt_flags & MNT_RELATIME))
1323 * Is mtime younger than atime? If yes, update atime:
1325 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1328 * Is ctime younger than atime? If yes, update atime:
1330 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1334 * Is the previous atime value older than a day? If yes,
1337 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1340 * Good, we can skip the atime update:
1346 * touch_atime - update the access time
1347 * @mnt: mount the inode is accessed on
1348 * @dentry: dentry accessed
1350 * Update the accessed time on an inode and mark it for writeback.
1351 * This function automatically handles read only file systems and media,
1352 * as well as the "noatime" flag and inode specific "noatime" markers.
1354 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1356 struct inode *inode = dentry->d_inode;
1357 struct timespec now;
1359 if (inode->i_flags & S_NOATIME)
1361 if (IS_NOATIME(inode))
1363 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1366 if (mnt->mnt_flags & MNT_NOATIME)
1368 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1371 now = current_fs_time(inode->i_sb);
1373 if (!relatime_need_update(mnt, inode, now))
1376 if (timespec_equal(&inode->i_atime, &now))
1379 if (mnt_want_write(mnt))
1382 inode->i_atime = now;
1383 mark_inode_dirty_sync(inode);
1384 mnt_drop_write(mnt);
1386 EXPORT_SYMBOL(touch_atime);
1389 * file_update_time - update mtime and ctime time
1390 * @file: file accessed
1392 * Update the mtime and ctime members of an inode and mark the inode
1393 * for writeback. Note that this function is meant exclusively for
1394 * usage in the file write path of filesystems, and filesystems may
1395 * choose to explicitly ignore update via this function with the
1396 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1397 * timestamps are handled by the server.
1400 void file_update_time(struct file *file)
1402 struct inode *inode = file->f_path.dentry->d_inode;
1403 struct timespec now;
1404 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1406 /* First try to exhaust all avenues to not sync */
1407 if (IS_NOCMTIME(inode))
1410 now = current_fs_time(inode->i_sb);
1411 if (!timespec_equal(&inode->i_mtime, &now))
1414 if (!timespec_equal(&inode->i_ctime, &now))
1417 if (IS_I_VERSION(inode))
1418 sync_it |= S_VERSION;
1423 /* Finally allowed to write? Takes lock. */
1424 if (mnt_want_write_file(file))
1427 /* Only change inode inside the lock region */
1428 if (sync_it & S_VERSION)
1429 inode_inc_iversion(inode);
1430 if (sync_it & S_CTIME)
1431 inode->i_ctime = now;
1432 if (sync_it & S_MTIME)
1433 inode->i_mtime = now;
1434 mark_inode_dirty_sync(inode);
1435 mnt_drop_write(file->f_path.mnt);
1437 EXPORT_SYMBOL(file_update_time);
1439 int inode_needs_sync(struct inode *inode)
1443 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1447 EXPORT_SYMBOL(inode_needs_sync);
1449 int inode_wait(void *word)
1454 EXPORT_SYMBOL(inode_wait);
1457 * If we try to find an inode in the inode hash while it is being
1458 * deleted, we have to wait until the filesystem completes its
1459 * deletion before reporting that it isn't found. This function waits
1460 * until the deletion _might_ have completed. Callers are responsible
1461 * to recheck inode state.
1463 * It doesn't matter if I_NEW is not set initially, a call to
1464 * wake_up_inode() after removing from the hash list will DTRT.
1466 * This is called with inode_lock held.
1468 static void __wait_on_freeing_inode(struct inode *inode)
1470 wait_queue_head_t *wq;
1471 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1472 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1473 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1474 spin_unlock(&inode_lock);
1476 finish_wait(wq, &wait.wait);
1477 spin_lock(&inode_lock);
1480 static __initdata unsigned long ihash_entries;
1481 static int __init set_ihash_entries(char *str)
1485 ihash_entries = simple_strtoul(str, &str, 0);
1488 __setup("ihash_entries=", set_ihash_entries);
1491 * Initialize the waitqueues and inode hash table.
1493 void __init inode_init_early(void)
1497 /* If hashes are distributed across NUMA nodes, defer
1498 * hash allocation until vmalloc space is available.
1504 alloc_large_system_hash("Inode-cache",
1505 sizeof(struct hlist_head),
1513 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1514 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1517 void __init inode_init(void)
1521 /* inode slab cache */
1522 inode_cachep = kmem_cache_create("inode_cache",
1523 sizeof(struct inode),
1525 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1528 register_shrinker(&icache_shrinker);
1529 percpu_counter_init(&nr_inodes, 0);
1530 percpu_counter_init(&nr_inodes_unused, 0);
1532 /* Hash may have been set up in inode_init_early */
1537 alloc_large_system_hash("Inode-cache",
1538 sizeof(struct hlist_head),
1546 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1547 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1550 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1552 inode->i_mode = mode;
1553 if (S_ISCHR(mode)) {
1554 inode->i_fop = &def_chr_fops;
1555 inode->i_rdev = rdev;
1556 } else if (S_ISBLK(mode)) {
1557 inode->i_fop = &def_blk_fops;
1558 inode->i_rdev = rdev;
1559 } else if (S_ISFIFO(mode))
1560 inode->i_fop = &def_fifo_fops;
1561 else if (S_ISSOCK(mode))
1562 inode->i_fop = &bad_sock_fops;
1564 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1565 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1568 EXPORT_SYMBOL(init_special_inode);
1571 * Init uid,gid,mode for new inode according to posix standards
1573 * @dir: Directory inode
1574 * @mode: mode of the new inode
1576 void inode_init_owner(struct inode *inode, const struct inode *dir,
1579 inode->i_uid = current_fsuid();
1580 if (dir && dir->i_mode & S_ISGID) {
1581 inode->i_gid = dir->i_gid;
1585 inode->i_gid = current_fsgid();
1586 inode->i_mode = mode;
1588 EXPORT_SYMBOL(inode_init_owner);