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 static LIST_HEAD(inode_unused);
76 static struct hlist_head *inode_hashtable __read_mostly;
79 * A simple spinlock to protect the list manipulations.
81 * NOTE! You also have to own the lock if you change
82 * the i_state of an inode while it is in use..
84 DEFINE_SPINLOCK(inode_lock);
87 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
88 * icache shrinking path, and the umount path. Without this exclusion,
89 * by the time prune_icache calls iput for the inode whose pages it has
90 * been invalidating, or by the time it calls clear_inode & destroy_inode
91 * from its final dispose_list, the struct super_block they refer to
92 * (for inode->i_sb->s_op) may already have been freed and reused.
94 * We make this an rwsem because the fastpath is icache shrinking. In
95 * some cases a filesystem may be doing a significant amount of work in
96 * its inode reclaim code, so this should improve parallelism.
98 static DECLARE_RWSEM(iprune_sem);
101 * Statistics gathering..
103 struct inodes_stat_t inodes_stat;
105 static struct percpu_counter nr_inodes __cacheline_aligned_in_smp;
106 static struct percpu_counter nr_inodes_unused __cacheline_aligned_in_smp;
108 static struct kmem_cache *inode_cachep __read_mostly;
110 static inline int get_nr_inodes(void)
112 return percpu_counter_sum_positive(&nr_inodes);
115 static inline int get_nr_inodes_unused(void)
117 return percpu_counter_sum_positive(&nr_inodes_unused);
120 int get_nr_dirty_inodes(void)
122 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
123 return nr_dirty > 0 ? nr_dirty : 0;
128 * Handle nr_inode sysctl
131 int proc_nr_inodes(ctl_table *table, int write,
132 void __user *buffer, size_t *lenp, loff_t *ppos)
134 inodes_stat.nr_inodes = get_nr_inodes();
135 inodes_stat.nr_unused = get_nr_inodes_unused();
136 return proc_dointvec(table, write, buffer, lenp, ppos);
140 static void wake_up_inode(struct inode *inode)
143 * Prevent speculative execution through spin_unlock(&inode_lock);
146 wake_up_bit(&inode->i_state, __I_NEW);
150 * inode_init_always - perform inode structure intialisation
151 * @sb: superblock inode belongs to
152 * @inode: inode to initialise
154 * These are initializations that need to be done on every inode
155 * allocation as the fields are not initialised by slab allocation.
157 int inode_init_always(struct super_block *sb, struct inode *inode)
159 static const struct address_space_operations empty_aops;
160 static const struct inode_operations empty_iops;
161 static const struct file_operations empty_fops;
162 struct address_space *const mapping = &inode->i_data;
165 inode->i_blkbits = sb->s_blocksize_bits;
167 atomic_set(&inode->i_count, 1);
168 inode->i_op = &empty_iops;
169 inode->i_fop = &empty_fops;
173 atomic_set(&inode->i_writecount, 0);
177 inode->i_generation = 0;
179 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
181 inode->i_pipe = NULL;
182 inode->i_bdev = NULL;
183 inode->i_cdev = NULL;
185 inode->dirtied_when = 0;
187 if (security_inode_alloc(inode))
189 spin_lock_init(&inode->i_lock);
190 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
192 mutex_init(&inode->i_mutex);
193 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
195 init_rwsem(&inode->i_alloc_sem);
196 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
198 mapping->a_ops = &empty_aops;
199 mapping->host = inode;
201 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
202 mapping->assoc_mapping = NULL;
203 mapping->backing_dev_info = &default_backing_dev_info;
204 mapping->writeback_index = 0;
207 * If the block_device provides a backing_dev_info for client
208 * inodes then use that. Otherwise the inode share the bdev's
212 struct backing_dev_info *bdi;
214 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
215 mapping->backing_dev_info = bdi;
217 inode->i_private = NULL;
218 inode->i_mapping = mapping;
219 #ifdef CONFIG_FS_POSIX_ACL
220 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
223 #ifdef CONFIG_FSNOTIFY
224 inode->i_fsnotify_mask = 0;
227 percpu_counter_inc(&nr_inodes);
233 EXPORT_SYMBOL(inode_init_always);
235 static struct inode *alloc_inode(struct super_block *sb)
239 if (sb->s_op->alloc_inode)
240 inode = sb->s_op->alloc_inode(sb);
242 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
247 if (unlikely(inode_init_always(sb, inode))) {
248 if (inode->i_sb->s_op->destroy_inode)
249 inode->i_sb->s_op->destroy_inode(inode);
251 kmem_cache_free(inode_cachep, inode);
258 void __destroy_inode(struct inode *inode)
260 BUG_ON(inode_has_buffers(inode));
261 security_inode_free(inode);
262 fsnotify_inode_delete(inode);
263 #ifdef CONFIG_FS_POSIX_ACL
264 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
265 posix_acl_release(inode->i_acl);
266 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
267 posix_acl_release(inode->i_default_acl);
269 percpu_counter_dec(&nr_inodes);
271 EXPORT_SYMBOL(__destroy_inode);
273 static void destroy_inode(struct inode *inode)
275 __destroy_inode(inode);
276 if (inode->i_sb->s_op->destroy_inode)
277 inode->i_sb->s_op->destroy_inode(inode);
279 kmem_cache_free(inode_cachep, (inode));
283 * These are initializations that only need to be done
284 * once, because the fields are idempotent across use
285 * of the inode, so let the slab aware of that.
287 void inode_init_once(struct inode *inode)
289 memset(inode, 0, sizeof(*inode));
290 INIT_HLIST_NODE(&inode->i_hash);
291 INIT_LIST_HEAD(&inode->i_dentry);
292 INIT_LIST_HEAD(&inode->i_devices);
293 INIT_LIST_HEAD(&inode->i_list);
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 atomic_inc(&inode->i_count);
323 static void inode_lru_list_add(struct inode *inode)
325 if (list_empty(&inode->i_list)) {
326 list_add(&inode->i_list, &inode_unused);
327 percpu_counter_inc(&nr_inodes_unused);
331 static void inode_lru_list_del(struct inode *inode)
333 if (!list_empty(&inode->i_list)) {
334 list_del_init(&inode->i_list);
335 percpu_counter_dec(&nr_inodes_unused);
339 void end_writeback(struct inode *inode)
342 BUG_ON(inode->i_data.nrpages);
343 BUG_ON(!list_empty(&inode->i_data.private_list));
344 BUG_ON(!(inode->i_state & I_FREEING));
345 BUG_ON(inode->i_state & I_CLEAR);
346 inode_sync_wait(inode);
347 inode->i_state = I_FREEING | I_CLEAR;
349 EXPORT_SYMBOL(end_writeback);
351 static void evict(struct inode *inode)
353 const struct super_operations *op = inode->i_sb->s_op;
355 if (op->evict_inode) {
356 op->evict_inode(inode);
358 if (inode->i_data.nrpages)
359 truncate_inode_pages(&inode->i_data, 0);
360 end_writeback(inode);
362 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
364 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
369 * dispose_list - dispose of the contents of a local list
370 * @head: the head of the list to free
372 * Dispose-list gets a local list with local inodes in it, so it doesn't
373 * need to worry about list corruption and SMP locks.
375 static void dispose_list(struct list_head *head)
377 while (!list_empty(head)) {
380 inode = list_first_entry(head, struct inode, i_list);
381 list_del_init(&inode->i_list);
385 spin_lock(&inode_lock);
386 hlist_del_init(&inode->i_hash);
387 list_del_init(&inode->i_sb_list);
388 spin_unlock(&inode_lock);
390 wake_up_inode(inode);
391 destroy_inode(inode);
396 * Invalidate all inodes for a device.
398 static int invalidate_list(struct list_head *head, struct list_head *dispose)
400 struct list_head *next;
405 struct list_head *tmp = next;
409 * We can reschedule here without worrying about the list's
410 * consistency because the per-sb list of inodes must not
411 * change during umount anymore, and because iprune_sem keeps
412 * shrink_icache_memory() away.
414 cond_resched_lock(&inode_lock);
419 inode = list_entry(tmp, struct inode, i_sb_list);
420 if (inode->i_state & I_NEW)
422 invalidate_inode_buffers(inode);
423 if (!atomic_read(&inode->i_count)) {
424 list_move(&inode->i_list, dispose);
425 WARN_ON(inode->i_state & I_NEW);
426 inode->i_state |= I_FREEING;
427 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
428 percpu_counter_dec(&nr_inodes_unused);
437 * invalidate_inodes - discard the inodes on a device
440 * Discard all of the inodes for a given superblock. If the discard
441 * fails because there are busy inodes then a non zero value is returned.
442 * If the discard is successful all the inodes have been discarded.
444 int invalidate_inodes(struct super_block *sb)
447 LIST_HEAD(throw_away);
449 down_write(&iprune_sem);
450 spin_lock(&inode_lock);
451 fsnotify_unmount_inodes(&sb->s_inodes);
452 busy = invalidate_list(&sb->s_inodes, &throw_away);
453 spin_unlock(&inode_lock);
455 dispose_list(&throw_away);
456 up_write(&iprune_sem);
461 static int can_unuse(struct inode *inode)
463 if (inode->i_state & ~I_REFERENCED)
465 if (inode_has_buffers(inode))
467 if (atomic_read(&inode->i_count))
469 if (inode->i_data.nrpages)
475 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
476 * temporary list and then are freed outside inode_lock by dispose_list().
478 * Any inodes which are pinned purely because of attached pagecache have their
479 * pagecache removed. If the inode has metadata buffers attached to
480 * mapping->private_list then try to remove them.
482 * If the inode has the I_REFERENCED flag set, then it means that it has been
483 * used recently - the flag is set in iput_final(). When we encounter such an
484 * inode, clear the flag and move it to the back of the LRU so it gets another
485 * pass through the LRU before it gets reclaimed. This is necessary because of
486 * the fact we are doing lazy LRU updates to minimise lock contention so the
487 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
488 * with this flag set because they are the inodes that are out of order.
490 static void prune_icache(int nr_to_scan)
494 unsigned long reap = 0;
496 down_read(&iprune_sem);
497 spin_lock(&inode_lock);
498 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
501 if (list_empty(&inode_unused))
504 inode = list_entry(inode_unused.prev, struct inode, i_list);
507 * Referenced or dirty inodes are still in use. Give them
508 * another pass through the LRU as we canot reclaim them now.
510 if (atomic_read(&inode->i_count) ||
511 (inode->i_state & ~I_REFERENCED)) {
512 list_del_init(&inode->i_list);
513 percpu_counter_dec(&nr_inodes_unused);
517 /* recently referenced inodes get one more pass */
518 if (inode->i_state & I_REFERENCED) {
519 list_move(&inode->i_list, &inode_unused);
520 inode->i_state &= ~I_REFERENCED;
523 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
525 spin_unlock(&inode_lock);
526 if (remove_inode_buffers(inode))
527 reap += invalidate_mapping_pages(&inode->i_data,
530 spin_lock(&inode_lock);
532 if (inode != list_entry(inode_unused.next,
533 struct inode, i_list))
534 continue; /* wrong inode or list_empty */
535 if (!can_unuse(inode))
538 list_move(&inode->i_list, &freeable);
539 WARN_ON(inode->i_state & I_NEW);
540 inode->i_state |= I_FREEING;
541 percpu_counter_dec(&nr_inodes_unused);
543 if (current_is_kswapd())
544 __count_vm_events(KSWAPD_INODESTEAL, reap);
546 __count_vm_events(PGINODESTEAL, reap);
547 spin_unlock(&inode_lock);
549 dispose_list(&freeable);
550 up_read(&iprune_sem);
554 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
555 * "unused" means that no dentries are referring to the inodes: the files are
556 * not open and the dcache references to those inodes have already been
559 * This function is passed the number of inodes to scan, and it returns the
560 * total number of remaining possibly-reclaimable inodes.
562 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
566 * Nasty deadlock avoidance. We may hold various FS locks,
567 * and we don't want to recurse into the FS that called us
568 * in clear_inode() and friends..
570 if (!(gfp_mask & __GFP_FS))
574 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
577 static struct shrinker icache_shrinker = {
578 .shrink = shrink_icache_memory,
579 .seeks = DEFAULT_SEEKS,
582 static void __wait_on_freeing_inode(struct inode *inode);
584 * Called with the inode lock held.
585 * NOTE: we are not increasing the inode-refcount, you must call __iget()
586 * by hand after calling find_inode now! This simplifies iunique and won't
587 * add any additional branch in the common code.
589 static struct inode *find_inode(struct super_block *sb,
590 struct hlist_head *head,
591 int (*test)(struct inode *, void *),
594 struct hlist_node *node;
595 struct inode *inode = NULL;
598 hlist_for_each_entry(inode, node, head, i_hash) {
599 if (inode->i_sb != sb)
601 if (!test(inode, data))
603 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
604 __wait_on_freeing_inode(inode);
609 return node ? inode : NULL;
613 * find_inode_fast is the fast path version of find_inode, see the comment at
614 * iget_locked for details.
616 static struct inode *find_inode_fast(struct super_block *sb,
617 struct hlist_head *head, unsigned long ino)
619 struct hlist_node *node;
620 struct inode *inode = NULL;
623 hlist_for_each_entry(inode, node, head, i_hash) {
624 if (inode->i_ino != ino)
626 if (inode->i_sb != sb)
628 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
629 __wait_on_freeing_inode(inode);
634 return node ? inode : NULL;
637 static unsigned long hash(struct super_block *sb, unsigned long hashval)
641 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
643 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
644 return tmp & I_HASHMASK;
648 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
651 list_add(&inode->i_sb_list, &sb->s_inodes);
653 hlist_add_head(&inode->i_hash, head);
657 * inode_add_to_lists - add a new inode to relevant lists
658 * @sb: superblock inode belongs to
659 * @inode: inode to mark in use
661 * When an inode is allocated it needs to be accounted for, added to the in use
662 * list, the owning superblock and the inode hash. This needs to be done under
663 * the inode_lock, so export a function to do this rather than the inode lock
664 * itself. We calculate the hash list to add to here so it is all internal
665 * which requires the caller to have already set up the inode number in the
668 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
670 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
672 spin_lock(&inode_lock);
673 __inode_add_to_lists(sb, head, inode);
674 spin_unlock(&inode_lock);
676 EXPORT_SYMBOL_GPL(inode_add_to_lists);
679 * new_inode - obtain an inode
682 * Allocates a new inode for given superblock. The default gfp_mask
683 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
684 * If HIGHMEM pages are unsuitable or it is known that pages allocated
685 * for the page cache are not reclaimable or migratable,
686 * mapping_set_gfp_mask() must be called with suitable flags on the
687 * newly created inode's mapping
690 struct inode *new_inode(struct super_block *sb)
693 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
694 * error if st_ino won't fit in target struct field. Use 32bit counter
695 * here to attempt to avoid that.
697 static unsigned int last_ino;
700 spin_lock_prefetch(&inode_lock);
702 inode = alloc_inode(sb);
704 spin_lock(&inode_lock);
705 __inode_add_to_lists(sb, NULL, inode);
706 inode->i_ino = ++last_ino;
708 spin_unlock(&inode_lock);
712 EXPORT_SYMBOL(new_inode);
714 void unlock_new_inode(struct inode *inode)
716 #ifdef CONFIG_DEBUG_LOCK_ALLOC
717 if (S_ISDIR(inode->i_mode)) {
718 struct file_system_type *type = inode->i_sb->s_type;
720 /* Set new key only if filesystem hasn't already changed it */
721 if (!lockdep_match_class(&inode->i_mutex,
722 &type->i_mutex_key)) {
724 * ensure nobody is actually holding i_mutex
726 mutex_destroy(&inode->i_mutex);
727 mutex_init(&inode->i_mutex);
728 lockdep_set_class(&inode->i_mutex,
729 &type->i_mutex_dir_key);
734 * This is special! We do not need the spinlock when clearing I_NEW,
735 * because we're guaranteed that nobody else tries to do anything about
736 * the state of the inode when it is locked, as we just created it (so
737 * there can be no old holders that haven't tested I_NEW).
738 * However we must emit the memory barrier so that other CPUs reliably
739 * see the clearing of I_NEW after the other inode initialisation has
743 WARN_ON(!(inode->i_state & I_NEW));
744 inode->i_state &= ~I_NEW;
745 wake_up_inode(inode);
747 EXPORT_SYMBOL(unlock_new_inode);
750 * This is called without the inode lock held.. Be careful.
752 * We no longer cache the sb_flags in i_flags - see fs.h
753 * -- rmk@arm.uk.linux.org
755 static struct inode *get_new_inode(struct super_block *sb,
756 struct hlist_head *head,
757 int (*test)(struct inode *, void *),
758 int (*set)(struct inode *, void *),
763 inode = alloc_inode(sb);
767 spin_lock(&inode_lock);
768 /* We released the lock, so.. */
769 old = find_inode(sb, head, test, data);
771 if (set(inode, data))
774 __inode_add_to_lists(sb, head, inode);
775 inode->i_state = I_NEW;
776 spin_unlock(&inode_lock);
778 /* Return the locked inode with I_NEW set, the
779 * caller is responsible for filling in the contents
785 * Uhhuh, somebody else created the same inode under
786 * us. Use the old inode instead of the one we just
790 spin_unlock(&inode_lock);
791 destroy_inode(inode);
793 wait_on_inode(inode);
798 spin_unlock(&inode_lock);
799 destroy_inode(inode);
804 * get_new_inode_fast is the fast path version of get_new_inode, see the
805 * comment at iget_locked for details.
807 static struct inode *get_new_inode_fast(struct super_block *sb,
808 struct hlist_head *head, unsigned long ino)
812 inode = alloc_inode(sb);
816 spin_lock(&inode_lock);
817 /* We released the lock, so.. */
818 old = find_inode_fast(sb, head, ino);
821 __inode_add_to_lists(sb, head, inode);
822 inode->i_state = I_NEW;
823 spin_unlock(&inode_lock);
825 /* Return the locked inode with I_NEW set, the
826 * caller is responsible for filling in the contents
832 * Uhhuh, somebody else created the same inode under
833 * us. Use the old inode instead of the one we just
837 spin_unlock(&inode_lock);
838 destroy_inode(inode);
840 wait_on_inode(inode);
846 * iunique - get a unique inode number
848 * @max_reserved: highest reserved inode number
850 * Obtain an inode number that is unique on the system for a given
851 * superblock. This is used by file systems that have no natural
852 * permanent inode numbering system. An inode number is returned that
853 * is higher than the reserved limit but unique.
856 * With a large number of inodes live on the file system this function
857 * currently becomes quite slow.
859 ino_t iunique(struct super_block *sb, ino_t max_reserved)
862 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
863 * error if st_ino won't fit in target struct field. Use 32bit counter
864 * here to attempt to avoid that.
866 static unsigned int counter;
868 struct hlist_head *head;
871 spin_lock(&inode_lock);
873 if (counter <= max_reserved)
874 counter = max_reserved + 1;
876 head = inode_hashtable + hash(sb, res);
877 inode = find_inode_fast(sb, head, res);
878 } while (inode != NULL);
879 spin_unlock(&inode_lock);
883 EXPORT_SYMBOL(iunique);
885 struct inode *igrab(struct inode *inode)
887 spin_lock(&inode_lock);
888 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
892 * Handle the case where s_op->clear_inode is not been
893 * called yet, and somebody is calling igrab
894 * while the inode is getting freed.
897 spin_unlock(&inode_lock);
900 EXPORT_SYMBOL(igrab);
903 * ifind - internal function, you want ilookup5() or iget5().
904 * @sb: super block of file system to search
905 * @head: the head of the list to search
906 * @test: callback used for comparisons between inodes
907 * @data: opaque data pointer to pass to @test
908 * @wait: if true wait for the inode to be unlocked, if false do not
910 * ifind() searches for the inode specified by @data in the inode
911 * cache. This is a generalized version of ifind_fast() for file systems where
912 * the inode number is not sufficient for unique identification of an inode.
914 * If the inode is in the cache, the inode is returned with an incremented
917 * Otherwise NULL is returned.
919 * Note, @test is called with the inode_lock held, so can't sleep.
921 static struct inode *ifind(struct super_block *sb,
922 struct hlist_head *head, int (*test)(struct inode *, void *),
923 void *data, const int wait)
927 spin_lock(&inode_lock);
928 inode = find_inode(sb, head, test, data);
931 spin_unlock(&inode_lock);
933 wait_on_inode(inode);
936 spin_unlock(&inode_lock);
941 * ifind_fast - internal function, you want ilookup() or iget().
942 * @sb: super block of file system to search
943 * @head: head of the list to search
944 * @ino: inode number to search for
946 * ifind_fast() searches for the inode @ino in the inode cache. This is for
947 * file systems where the inode number is sufficient for unique identification
950 * If the inode is in the cache, the inode is returned with an incremented
953 * Otherwise NULL is returned.
955 static struct inode *ifind_fast(struct super_block *sb,
956 struct hlist_head *head, unsigned long ino)
960 spin_lock(&inode_lock);
961 inode = find_inode_fast(sb, head, ino);
964 spin_unlock(&inode_lock);
965 wait_on_inode(inode);
968 spin_unlock(&inode_lock);
973 * ilookup5_nowait - search for an inode in the inode cache
974 * @sb: super block of file system to search
975 * @hashval: hash value (usually inode number) to search for
976 * @test: callback used for comparisons between inodes
977 * @data: opaque data pointer to pass to @test
979 * ilookup5() uses ifind() to search for the inode specified by @hashval and
980 * @data in the inode cache. This is a generalized version of ilookup() for
981 * file systems where the inode number is not sufficient for unique
982 * identification of an inode.
984 * If the inode is in the cache, the inode is returned with an incremented
985 * reference count. Note, the inode lock is not waited upon so you have to be
986 * very careful what you do with the returned inode. You probably should be
987 * using ilookup5() instead.
989 * Otherwise NULL is returned.
991 * Note, @test is called with the inode_lock held, so can't sleep.
993 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
994 int (*test)(struct inode *, void *), void *data)
996 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
998 return ifind(sb, head, test, data, 0);
1000 EXPORT_SYMBOL(ilookup5_nowait);
1003 * ilookup5 - search for an inode in the inode cache
1004 * @sb: super block of file system to search
1005 * @hashval: hash value (usually inode number) to search for
1006 * @test: callback used for comparisons between inodes
1007 * @data: opaque data pointer to pass to @test
1009 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1010 * @data in the inode cache. This is a generalized version of ilookup() for
1011 * file systems where the inode number is not sufficient for unique
1012 * identification of an inode.
1014 * If the inode is in the cache, the inode lock is waited upon and the inode is
1015 * returned with an incremented reference count.
1017 * Otherwise NULL is returned.
1019 * Note, @test is called with the inode_lock held, so can't sleep.
1021 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1022 int (*test)(struct inode *, void *), void *data)
1024 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1026 return ifind(sb, head, test, data, 1);
1028 EXPORT_SYMBOL(ilookup5);
1031 * ilookup - search for an inode in the inode cache
1032 * @sb: super block of file system to search
1033 * @ino: inode number to search for
1035 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1036 * This is for file systems where the inode number is sufficient for unique
1037 * identification of an inode.
1039 * If the inode is in the cache, the inode is returned with an incremented
1042 * Otherwise NULL is returned.
1044 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1046 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1048 return ifind_fast(sb, head, ino);
1050 EXPORT_SYMBOL(ilookup);
1053 * iget5_locked - obtain an inode from a mounted file system
1054 * @sb: super block of file system
1055 * @hashval: hash value (usually inode number) to get
1056 * @test: callback used for comparisons between inodes
1057 * @set: callback used to initialize a new struct inode
1058 * @data: opaque data pointer to pass to @test and @set
1060 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1061 * and @data in the inode cache and if present it is returned with an increased
1062 * reference count. This is a generalized version of iget_locked() for file
1063 * systems where the inode number is not sufficient for unique identification
1066 * If the inode is not in cache, get_new_inode() is called to allocate a new
1067 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1068 * file system gets to fill it in before unlocking it via unlock_new_inode().
1070 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1072 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1073 int (*test)(struct inode *, void *),
1074 int (*set)(struct inode *, void *), void *data)
1076 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1077 struct inode *inode;
1079 inode = ifind(sb, head, test, data, 1);
1083 * get_new_inode() will do the right thing, re-trying the search
1084 * in case it had to block at any point.
1086 return get_new_inode(sb, head, test, set, data);
1088 EXPORT_SYMBOL(iget5_locked);
1091 * iget_locked - obtain an inode from a mounted file system
1092 * @sb: super block of file system
1093 * @ino: inode number to get
1095 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1096 * the inode cache and if present it is returned with an increased reference
1097 * count. This is for file systems where the inode number is sufficient for
1098 * unique identification of an inode.
1100 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1101 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1102 * The file system gets to fill it in before unlocking it via
1103 * unlock_new_inode().
1105 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1107 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1108 struct inode *inode;
1110 inode = ifind_fast(sb, head, ino);
1114 * get_new_inode_fast() will do the right thing, re-trying the search
1115 * in case it had to block at any point.
1117 return get_new_inode_fast(sb, head, ino);
1119 EXPORT_SYMBOL(iget_locked);
1121 int insert_inode_locked(struct inode *inode)
1123 struct super_block *sb = inode->i_sb;
1124 ino_t ino = inode->i_ino;
1125 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1127 inode->i_state |= I_NEW;
1129 struct hlist_node *node;
1130 struct inode *old = NULL;
1131 spin_lock(&inode_lock);
1132 hlist_for_each_entry(old, node, head, i_hash) {
1133 if (old->i_ino != ino)
1135 if (old->i_sb != sb)
1137 if (old->i_state & (I_FREEING|I_WILL_FREE))
1141 if (likely(!node)) {
1142 hlist_add_head(&inode->i_hash, head);
1143 spin_unlock(&inode_lock);
1147 spin_unlock(&inode_lock);
1149 if (unlikely(!inode_unhashed(old))) {
1156 EXPORT_SYMBOL(insert_inode_locked);
1158 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1159 int (*test)(struct inode *, void *), void *data)
1161 struct super_block *sb = inode->i_sb;
1162 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1164 inode->i_state |= I_NEW;
1167 struct hlist_node *node;
1168 struct inode *old = NULL;
1170 spin_lock(&inode_lock);
1171 hlist_for_each_entry(old, node, head, i_hash) {
1172 if (old->i_sb != sb)
1174 if (!test(old, data))
1176 if (old->i_state & (I_FREEING|I_WILL_FREE))
1180 if (likely(!node)) {
1181 hlist_add_head(&inode->i_hash, head);
1182 spin_unlock(&inode_lock);
1186 spin_unlock(&inode_lock);
1188 if (unlikely(!inode_unhashed(old))) {
1195 EXPORT_SYMBOL(insert_inode_locked4);
1198 * __insert_inode_hash - hash an inode
1199 * @inode: unhashed inode
1200 * @hashval: unsigned long value used to locate this object in the
1203 * Add an inode to the inode hash for this superblock.
1205 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1207 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1208 spin_lock(&inode_lock);
1209 hlist_add_head(&inode->i_hash, head);
1210 spin_unlock(&inode_lock);
1212 EXPORT_SYMBOL(__insert_inode_hash);
1215 * remove_inode_hash - remove an inode from the hash
1216 * @inode: inode to unhash
1218 * Remove an inode from the superblock.
1220 void remove_inode_hash(struct inode *inode)
1222 spin_lock(&inode_lock);
1223 hlist_del_init(&inode->i_hash);
1224 spin_unlock(&inode_lock);
1226 EXPORT_SYMBOL(remove_inode_hash);
1228 int generic_delete_inode(struct inode *inode)
1232 EXPORT_SYMBOL(generic_delete_inode);
1235 * Normal UNIX filesystem behaviour: delete the
1236 * inode when the usage count drops to zero, and
1239 int generic_drop_inode(struct inode *inode)
1241 return !inode->i_nlink || inode_unhashed(inode);
1243 EXPORT_SYMBOL_GPL(generic_drop_inode);
1246 * Called when we're dropping the last reference
1249 * Call the FS "drop_inode()" function, defaulting to
1250 * the legacy UNIX filesystem behaviour. If it tells
1251 * us to evict inode, do so. Otherwise, retain inode
1252 * in cache if fs is alive, sync and evict if fs is
1255 static void iput_final(struct inode *inode)
1257 struct super_block *sb = inode->i_sb;
1258 const struct super_operations *op = inode->i_sb->s_op;
1261 if (op && op->drop_inode)
1262 drop = op->drop_inode(inode);
1264 drop = generic_drop_inode(inode);
1267 if (sb->s_flags & MS_ACTIVE) {
1268 inode->i_state |= I_REFERENCED;
1269 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1270 inode_lru_list_add(inode);
1272 spin_unlock(&inode_lock);
1275 WARN_ON(inode->i_state & I_NEW);
1276 inode->i_state |= I_WILL_FREE;
1277 spin_unlock(&inode_lock);
1278 write_inode_now(inode, 1);
1279 spin_lock(&inode_lock);
1280 WARN_ON(inode->i_state & I_NEW);
1281 inode->i_state &= ~I_WILL_FREE;
1282 hlist_del_init(&inode->i_hash);
1284 WARN_ON(inode->i_state & I_NEW);
1285 inode->i_state |= I_FREEING;
1288 * After we delete the inode from the LRU here, we avoid moving dirty
1289 * inodes back onto the LRU now because I_FREEING is set and hence
1290 * writeback_single_inode() won't move the inode around.
1292 inode_lru_list_del(inode);
1294 list_del_init(&inode->i_sb_list);
1295 spin_unlock(&inode_lock);
1297 spin_lock(&inode_lock);
1298 hlist_del_init(&inode->i_hash);
1299 spin_unlock(&inode_lock);
1300 wake_up_inode(inode);
1301 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1302 destroy_inode(inode);
1306 * iput - put an inode
1307 * @inode: inode to put
1309 * Puts an inode, dropping its usage count. If the inode use count hits
1310 * zero, the inode is then freed and may also be destroyed.
1312 * Consequently, iput() can sleep.
1314 void iput(struct inode *inode)
1317 BUG_ON(inode->i_state & I_CLEAR);
1319 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1323 EXPORT_SYMBOL(iput);
1326 * bmap - find a block number in a file
1327 * @inode: inode of file
1328 * @block: block to find
1330 * Returns the block number on the device holding the inode that
1331 * is the disk block number for the block of the file requested.
1332 * That is, asked for block 4 of inode 1 the function will return the
1333 * disk block relative to the disk start that holds that block of the
1336 sector_t bmap(struct inode *inode, sector_t block)
1339 if (inode->i_mapping->a_ops->bmap)
1340 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1343 EXPORT_SYMBOL(bmap);
1346 * With relative atime, only update atime if the previous atime is
1347 * earlier than either the ctime or mtime or if at least a day has
1348 * passed since the last atime update.
1350 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1351 struct timespec now)
1354 if (!(mnt->mnt_flags & MNT_RELATIME))
1357 * Is mtime younger than atime? If yes, update atime:
1359 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1362 * Is ctime younger than atime? If yes, update atime:
1364 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1368 * Is the previous atime value older than a day? If yes,
1371 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1374 * Good, we can skip the atime update:
1380 * touch_atime - update the access time
1381 * @mnt: mount the inode is accessed on
1382 * @dentry: dentry accessed
1384 * Update the accessed time on an inode and mark it for writeback.
1385 * This function automatically handles read only file systems and media,
1386 * as well as the "noatime" flag and inode specific "noatime" markers.
1388 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1390 struct inode *inode = dentry->d_inode;
1391 struct timespec now;
1393 if (inode->i_flags & S_NOATIME)
1395 if (IS_NOATIME(inode))
1397 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1400 if (mnt->mnt_flags & MNT_NOATIME)
1402 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1405 now = current_fs_time(inode->i_sb);
1407 if (!relatime_need_update(mnt, inode, now))
1410 if (timespec_equal(&inode->i_atime, &now))
1413 if (mnt_want_write(mnt))
1416 inode->i_atime = now;
1417 mark_inode_dirty_sync(inode);
1418 mnt_drop_write(mnt);
1420 EXPORT_SYMBOL(touch_atime);
1423 * file_update_time - update mtime and ctime time
1424 * @file: file accessed
1426 * Update the mtime and ctime members of an inode and mark the inode
1427 * for writeback. Note that this function is meant exclusively for
1428 * usage in the file write path of filesystems, and filesystems may
1429 * choose to explicitly ignore update via this function with the
1430 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1431 * timestamps are handled by the server.
1434 void file_update_time(struct file *file)
1436 struct inode *inode = file->f_path.dentry->d_inode;
1437 struct timespec now;
1438 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1440 /* First try to exhaust all avenues to not sync */
1441 if (IS_NOCMTIME(inode))
1444 now = current_fs_time(inode->i_sb);
1445 if (!timespec_equal(&inode->i_mtime, &now))
1448 if (!timespec_equal(&inode->i_ctime, &now))
1451 if (IS_I_VERSION(inode))
1452 sync_it |= S_VERSION;
1457 /* Finally allowed to write? Takes lock. */
1458 if (mnt_want_write_file(file))
1461 /* Only change inode inside the lock region */
1462 if (sync_it & S_VERSION)
1463 inode_inc_iversion(inode);
1464 if (sync_it & S_CTIME)
1465 inode->i_ctime = now;
1466 if (sync_it & S_MTIME)
1467 inode->i_mtime = now;
1468 mark_inode_dirty_sync(inode);
1469 mnt_drop_write(file->f_path.mnt);
1471 EXPORT_SYMBOL(file_update_time);
1473 int inode_needs_sync(struct inode *inode)
1477 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1481 EXPORT_SYMBOL(inode_needs_sync);
1483 int inode_wait(void *word)
1488 EXPORT_SYMBOL(inode_wait);
1491 * If we try to find an inode in the inode hash while it is being
1492 * deleted, we have to wait until the filesystem completes its
1493 * deletion before reporting that it isn't found. This function waits
1494 * until the deletion _might_ have completed. Callers are responsible
1495 * to recheck inode state.
1497 * It doesn't matter if I_NEW is not set initially, a call to
1498 * wake_up_inode() after removing from the hash list will DTRT.
1500 * This is called with inode_lock held.
1502 static void __wait_on_freeing_inode(struct inode *inode)
1504 wait_queue_head_t *wq;
1505 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1506 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1507 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1508 spin_unlock(&inode_lock);
1510 finish_wait(wq, &wait.wait);
1511 spin_lock(&inode_lock);
1514 static __initdata unsigned long ihash_entries;
1515 static int __init set_ihash_entries(char *str)
1519 ihash_entries = simple_strtoul(str, &str, 0);
1522 __setup("ihash_entries=", set_ihash_entries);
1525 * Initialize the waitqueues and inode hash table.
1527 void __init inode_init_early(void)
1531 /* If hashes are distributed across NUMA nodes, defer
1532 * hash allocation until vmalloc space is available.
1538 alloc_large_system_hash("Inode-cache",
1539 sizeof(struct hlist_head),
1547 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1548 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1551 void __init inode_init(void)
1555 /* inode slab cache */
1556 inode_cachep = kmem_cache_create("inode_cache",
1557 sizeof(struct inode),
1559 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1562 register_shrinker(&icache_shrinker);
1563 percpu_counter_init(&nr_inodes, 0);
1564 percpu_counter_init(&nr_inodes_unused, 0);
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);