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:
33 * FIXME: remove all knowledge of the buffer layer from this file
35 #include <linux/buffer_head.h>
38 * New inode.c implementation.
40 * This implementation has the basic premise of trying
41 * to be extremely low-overhead and SMP-safe, yet be
42 * simple enough to be "obviously correct".
47 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
49 /* #define INODE_PARANOIA 1 */
50 /* #define INODE_DEBUG 1 */
53 * Inode lookup is no longer as critical as it used to be:
54 * most of the lookups are going to be through the dcache.
56 #define I_HASHBITS i_hash_shift
57 #define I_HASHMASK i_hash_mask
59 static unsigned int i_hash_mask __read_mostly;
60 static unsigned int i_hash_shift __read_mostly;
63 * Each inode can be on two separate lists. One is
64 * the hash list of the inode, used for lookups. The
65 * other linked list is the "type" list:
66 * "in_use" - valid inode, i_count > 0, i_nlink > 0
67 * "dirty" - as "in_use" but also dirty
68 * "unused" - valid inode, i_count = 0
70 * A "dirty" list is maintained for each super block,
71 * allowing for low-overhead inode sync() operations.
74 static LIST_HEAD(inode_unused);
75 static struct hlist_head *inode_hashtable __read_mostly;
78 * A simple spinlock to protect the list manipulations.
80 * NOTE! You also have to own the lock if you change
81 * the i_state of an inode while it is in use..
83 DEFINE_SPINLOCK(inode_lock);
86 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
87 * icache shrinking path, and the umount path. Without this exclusion,
88 * by the time prune_icache calls iput for the inode whose pages it has
89 * been invalidating, or by the time it calls clear_inode & destroy_inode
90 * from its final dispose_list, the struct super_block they refer to
91 * (for inode->i_sb->s_op) may already have been freed and reused.
93 * We make this an rwsem because the fastpath is icache shrinking. In
94 * some cases a filesystem may be doing a significant amount of work in
95 * its inode reclaim code, so this should improve parallelism.
97 static DECLARE_RWSEM(iprune_sem);
100 * Statistics gathering..
102 struct inodes_stat_t inodes_stat;
104 static struct percpu_counter nr_inodes __cacheline_aligned_in_smp;
105 static struct percpu_counter nr_inodes_unused __cacheline_aligned_in_smp;
107 static struct kmem_cache *inode_cachep __read_mostly;
109 static inline int get_nr_inodes(void)
111 return percpu_counter_sum_positive(&nr_inodes);
114 static inline int get_nr_inodes_unused(void)
116 return percpu_counter_sum_positive(&nr_inodes_unused);
119 int get_nr_dirty_inodes(void)
121 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
122 return nr_dirty > 0 ? nr_dirty : 0;
127 * Handle nr_inode sysctl
130 int proc_nr_inodes(ctl_table *table, int write,
131 void __user *buffer, size_t *lenp, loff_t *ppos)
133 inodes_stat.nr_inodes = get_nr_inodes();
134 inodes_stat.nr_unused = get_nr_inodes_unused();
135 return proc_dointvec(table, write, buffer, lenp, ppos);
139 static void wake_up_inode(struct inode *inode)
142 * Prevent speculative execution through spin_unlock(&inode_lock);
145 wake_up_bit(&inode->i_state, __I_NEW);
149 * inode_init_always - perform inode structure intialisation
150 * @sb: superblock inode belongs to
151 * @inode: inode to initialise
153 * These are initializations that need to be done on every inode
154 * allocation as the fields are not initialised by slab allocation.
156 int inode_init_always(struct super_block *sb, struct inode *inode)
158 static const struct address_space_operations empty_aops;
159 static const struct inode_operations empty_iops;
160 static const struct file_operations empty_fops;
161 struct address_space *const mapping = &inode->i_data;
164 inode->i_blkbits = sb->s_blocksize_bits;
166 atomic_set(&inode->i_count, 1);
167 inode->i_op = &empty_iops;
168 inode->i_fop = &empty_fops;
172 atomic_set(&inode->i_writecount, 0);
176 inode->i_generation = 0;
178 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
180 inode->i_pipe = NULL;
181 inode->i_bdev = NULL;
182 inode->i_cdev = NULL;
184 inode->dirtied_when = 0;
186 if (security_inode_alloc(inode))
188 spin_lock_init(&inode->i_lock);
189 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
191 mutex_init(&inode->i_mutex);
192 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
194 init_rwsem(&inode->i_alloc_sem);
195 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
197 mapping->a_ops = &empty_aops;
198 mapping->host = inode;
200 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
201 mapping->assoc_mapping = NULL;
202 mapping->backing_dev_info = &default_backing_dev_info;
203 mapping->writeback_index = 0;
206 * If the block_device provides a backing_dev_info for client
207 * inodes then use that. Otherwise the inode share the bdev's
211 struct backing_dev_info *bdi;
213 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
214 mapping->backing_dev_info = bdi;
216 inode->i_private = NULL;
217 inode->i_mapping = mapping;
218 #ifdef CONFIG_FS_POSIX_ACL
219 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
222 #ifdef CONFIG_FSNOTIFY
223 inode->i_fsnotify_mask = 0;
226 percpu_counter_inc(&nr_inodes);
232 EXPORT_SYMBOL(inode_init_always);
234 static struct inode *alloc_inode(struct super_block *sb)
238 if (sb->s_op->alloc_inode)
239 inode = sb->s_op->alloc_inode(sb);
241 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
246 if (unlikely(inode_init_always(sb, inode))) {
247 if (inode->i_sb->s_op->destroy_inode)
248 inode->i_sb->s_op->destroy_inode(inode);
250 kmem_cache_free(inode_cachep, inode);
257 void __destroy_inode(struct inode *inode)
259 BUG_ON(inode_has_buffers(inode));
260 security_inode_free(inode);
261 fsnotify_inode_delete(inode);
262 #ifdef CONFIG_FS_POSIX_ACL
263 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
264 posix_acl_release(inode->i_acl);
265 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
266 posix_acl_release(inode->i_default_acl);
268 percpu_counter_dec(&nr_inodes);
270 EXPORT_SYMBOL(__destroy_inode);
272 static void destroy_inode(struct inode *inode)
274 __destroy_inode(inode);
275 if (inode->i_sb->s_op->destroy_inode)
276 inode->i_sb->s_op->destroy_inode(inode);
278 kmem_cache_free(inode_cachep, (inode));
282 * These are initializations that only need to be done
283 * once, because the fields are idempotent across use
284 * of the inode, so let the slab aware of that.
286 void inode_init_once(struct inode *inode)
288 memset(inode, 0, sizeof(*inode));
289 INIT_HLIST_NODE(&inode->i_hash);
290 INIT_LIST_HEAD(&inode->i_dentry);
291 INIT_LIST_HEAD(&inode->i_devices);
292 INIT_LIST_HEAD(&inode->i_list);
293 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
294 spin_lock_init(&inode->i_data.tree_lock);
295 spin_lock_init(&inode->i_data.i_mmap_lock);
296 INIT_LIST_HEAD(&inode->i_data.private_list);
297 spin_lock_init(&inode->i_data.private_lock);
298 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
299 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
300 i_size_ordered_init(inode);
301 #ifdef CONFIG_FSNOTIFY
302 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
305 EXPORT_SYMBOL(inode_init_once);
307 static void init_once(void *foo)
309 struct inode *inode = (struct inode *) foo;
311 inode_init_once(inode);
315 * inode_lock must be held
317 void __iget(struct inode *inode)
319 atomic_inc(&inode->i_count);
323 * get additional reference to inode; caller must already hold one.
325 void ihold(struct inode *inode)
327 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
329 EXPORT_SYMBOL(ihold);
331 static void inode_lru_list_add(struct inode *inode)
333 if (list_empty(&inode->i_list)) {
334 list_add(&inode->i_list, &inode_unused);
335 percpu_counter_inc(&nr_inodes_unused);
339 static void inode_lru_list_del(struct inode *inode)
341 if (!list_empty(&inode->i_list)) {
342 list_del_init(&inode->i_list);
343 percpu_counter_dec(&nr_inodes_unused);
347 static inline void __inode_sb_list_add(struct inode *inode)
349 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
353 * inode_sb_list_add - add inode to the superblock list of inodes
354 * @inode: inode to add
356 void inode_sb_list_add(struct inode *inode)
358 spin_lock(&inode_lock);
359 __inode_sb_list_add(inode);
360 spin_unlock(&inode_lock);
362 EXPORT_SYMBOL_GPL(inode_sb_list_add);
364 static inline void __inode_sb_list_del(struct inode *inode)
366 list_del_init(&inode->i_sb_list);
369 static unsigned long hash(struct super_block *sb, unsigned long hashval)
373 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
375 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
376 return tmp & I_HASHMASK;
380 * __insert_inode_hash - hash an inode
381 * @inode: unhashed inode
382 * @hashval: unsigned long value used to locate this object in the
385 * Add an inode to the inode hash for this superblock.
387 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
389 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
391 spin_lock(&inode_lock);
392 hlist_add_head(&inode->i_hash, b);
393 spin_unlock(&inode_lock);
395 EXPORT_SYMBOL(__insert_inode_hash);
398 * __remove_inode_hash - remove an inode from the hash
399 * @inode: inode to unhash
401 * Remove an inode from the superblock.
403 static void __remove_inode_hash(struct inode *inode)
405 hlist_del_init(&inode->i_hash);
409 * remove_inode_hash - remove an inode from the hash
410 * @inode: inode to unhash
412 * Remove an inode from the superblock.
414 void remove_inode_hash(struct inode *inode)
416 spin_lock(&inode_lock);
417 hlist_del_init(&inode->i_hash);
418 spin_unlock(&inode_lock);
420 EXPORT_SYMBOL(remove_inode_hash);
422 void end_writeback(struct inode *inode)
425 BUG_ON(inode->i_data.nrpages);
426 BUG_ON(!list_empty(&inode->i_data.private_list));
427 BUG_ON(!(inode->i_state & I_FREEING));
428 BUG_ON(inode->i_state & I_CLEAR);
429 inode_sync_wait(inode);
430 inode->i_state = I_FREEING | I_CLEAR;
432 EXPORT_SYMBOL(end_writeback);
434 static void evict(struct inode *inode)
436 const struct super_operations *op = inode->i_sb->s_op;
438 if (op->evict_inode) {
439 op->evict_inode(inode);
441 if (inode->i_data.nrpages)
442 truncate_inode_pages(&inode->i_data, 0);
443 end_writeback(inode);
445 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
447 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
452 * dispose_list - dispose of the contents of a local list
453 * @head: the head of the list to free
455 * Dispose-list gets a local list with local inodes in it, so it doesn't
456 * need to worry about list corruption and SMP locks.
458 static void dispose_list(struct list_head *head)
460 while (!list_empty(head)) {
463 inode = list_first_entry(head, struct inode, i_list);
464 list_del_init(&inode->i_list);
468 spin_lock(&inode_lock);
469 __remove_inode_hash(inode);
470 __inode_sb_list_del(inode);
471 spin_unlock(&inode_lock);
473 wake_up_inode(inode);
474 destroy_inode(inode);
479 * Invalidate all inodes for a device.
481 static int invalidate_list(struct list_head *head, struct list_head *dispose)
483 struct list_head *next;
488 struct list_head *tmp = next;
492 * We can reschedule here without worrying about the list's
493 * consistency because the per-sb list of inodes must not
494 * change during umount anymore, and because iprune_sem keeps
495 * shrink_icache_memory() away.
497 cond_resched_lock(&inode_lock);
502 inode = list_entry(tmp, struct inode, i_sb_list);
503 if (inode->i_state & I_NEW)
505 if (atomic_read(&inode->i_count)) {
510 list_move(&inode->i_list, dispose);
511 inode->i_state |= I_FREEING;
512 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
513 percpu_counter_dec(&nr_inodes_unused);
519 * invalidate_inodes - discard the inodes on a device
522 * Discard all of the inodes for a given superblock. If the discard
523 * fails because there are busy inodes then a non zero value is returned.
524 * If the discard is successful all the inodes have been discarded.
526 int invalidate_inodes(struct super_block *sb)
529 LIST_HEAD(throw_away);
531 down_write(&iprune_sem);
532 spin_lock(&inode_lock);
533 fsnotify_unmount_inodes(&sb->s_inodes);
534 busy = invalidate_list(&sb->s_inodes, &throw_away);
535 spin_unlock(&inode_lock);
537 dispose_list(&throw_away);
538 up_write(&iprune_sem);
543 static int can_unuse(struct inode *inode)
545 if (inode->i_state & ~I_REFERENCED)
547 if (inode_has_buffers(inode))
549 if (atomic_read(&inode->i_count))
551 if (inode->i_data.nrpages)
557 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
558 * temporary list and then are freed outside inode_lock by dispose_list().
560 * Any inodes which are pinned purely because of attached pagecache have their
561 * pagecache removed. If the inode has metadata buffers attached to
562 * mapping->private_list then try to remove them.
564 * If the inode has the I_REFERENCED flag set, then it means that it has been
565 * used recently - the flag is set in iput_final(). When we encounter such an
566 * inode, clear the flag and move it to the back of the LRU so it gets another
567 * pass through the LRU before it gets reclaimed. This is necessary because of
568 * the fact we are doing lazy LRU updates to minimise lock contention so the
569 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
570 * with this flag set because they are the inodes that are out of order.
572 static void prune_icache(int nr_to_scan)
576 unsigned long reap = 0;
578 down_read(&iprune_sem);
579 spin_lock(&inode_lock);
580 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
583 if (list_empty(&inode_unused))
586 inode = list_entry(inode_unused.prev, struct inode, i_list);
589 * Referenced or dirty inodes are still in use. Give them
590 * another pass through the LRU as we canot reclaim them now.
592 if (atomic_read(&inode->i_count) ||
593 (inode->i_state & ~I_REFERENCED)) {
594 list_del_init(&inode->i_list);
595 percpu_counter_dec(&nr_inodes_unused);
599 /* recently referenced inodes get one more pass */
600 if (inode->i_state & I_REFERENCED) {
601 list_move(&inode->i_list, &inode_unused);
602 inode->i_state &= ~I_REFERENCED;
605 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
607 spin_unlock(&inode_lock);
608 if (remove_inode_buffers(inode))
609 reap += invalidate_mapping_pages(&inode->i_data,
612 spin_lock(&inode_lock);
614 if (inode != list_entry(inode_unused.next,
615 struct inode, i_list))
616 continue; /* wrong inode or list_empty */
617 if (!can_unuse(inode))
620 list_move(&inode->i_list, &freeable);
621 WARN_ON(inode->i_state & I_NEW);
622 inode->i_state |= I_FREEING;
623 percpu_counter_dec(&nr_inodes_unused);
625 if (current_is_kswapd())
626 __count_vm_events(KSWAPD_INODESTEAL, reap);
628 __count_vm_events(PGINODESTEAL, reap);
629 spin_unlock(&inode_lock);
631 dispose_list(&freeable);
632 up_read(&iprune_sem);
636 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
637 * "unused" means that no dentries are referring to the inodes: the files are
638 * not open and the dcache references to those inodes have already been
641 * This function is passed the number of inodes to scan, and it returns the
642 * total number of remaining possibly-reclaimable inodes.
644 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
648 * Nasty deadlock avoidance. We may hold various FS locks,
649 * and we don't want to recurse into the FS that called us
650 * in clear_inode() and friends..
652 if (!(gfp_mask & __GFP_FS))
656 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
659 static struct shrinker icache_shrinker = {
660 .shrink = shrink_icache_memory,
661 .seeks = DEFAULT_SEEKS,
664 static void __wait_on_freeing_inode(struct inode *inode);
666 * Called with the inode lock held.
668 static struct inode *find_inode(struct super_block *sb,
669 struct hlist_head *head,
670 int (*test)(struct inode *, void *),
673 struct hlist_node *node;
674 struct inode *inode = NULL;
677 hlist_for_each_entry(inode, node, head, i_hash) {
678 if (inode->i_sb != sb)
680 if (!test(inode, data))
682 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
683 __wait_on_freeing_inode(inode);
693 * find_inode_fast is the fast path version of find_inode, see the comment at
694 * iget_locked for details.
696 static struct inode *find_inode_fast(struct super_block *sb,
697 struct hlist_head *head, unsigned long ino)
699 struct hlist_node *node;
700 struct inode *inode = NULL;
703 hlist_for_each_entry(inode, node, head, i_hash) {
704 if (inode->i_ino != ino)
706 if (inode->i_sb != sb)
708 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
709 __wait_on_freeing_inode(inode);
719 * Each cpu owns a range of LAST_INO_BATCH numbers.
720 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
721 * to renew the exhausted range.
723 * This does not significantly increase overflow rate because every CPU can
724 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
725 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
726 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
727 * overflow rate by 2x, which does not seem too significant.
729 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
730 * error if st_ino won't fit in target struct field. Use 32bit counter
731 * here to attempt to avoid that.
733 #define LAST_INO_BATCH 1024
734 static DEFINE_PER_CPU(unsigned int, last_ino);
736 unsigned int get_next_ino(void)
738 unsigned int *p = &get_cpu_var(last_ino);
739 unsigned int res = *p;
742 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
743 static atomic_t shared_last_ino;
744 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
746 res = next - LAST_INO_BATCH;
751 put_cpu_var(last_ino);
754 EXPORT_SYMBOL(get_next_ino);
757 * new_inode - obtain an inode
760 * Allocates a new inode for given superblock. The default gfp_mask
761 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
762 * If HIGHMEM pages are unsuitable or it is known that pages allocated
763 * for the page cache are not reclaimable or migratable,
764 * mapping_set_gfp_mask() must be called with suitable flags on the
765 * newly created inode's mapping
768 struct inode *new_inode(struct super_block *sb)
772 spin_lock_prefetch(&inode_lock);
774 inode = alloc_inode(sb);
776 spin_lock(&inode_lock);
777 __inode_sb_list_add(inode);
779 spin_unlock(&inode_lock);
783 EXPORT_SYMBOL(new_inode);
785 void unlock_new_inode(struct inode *inode)
787 #ifdef CONFIG_DEBUG_LOCK_ALLOC
788 if (S_ISDIR(inode->i_mode)) {
789 struct file_system_type *type = inode->i_sb->s_type;
791 /* Set new key only if filesystem hasn't already changed it */
792 if (!lockdep_match_class(&inode->i_mutex,
793 &type->i_mutex_key)) {
795 * ensure nobody is actually holding i_mutex
797 mutex_destroy(&inode->i_mutex);
798 mutex_init(&inode->i_mutex);
799 lockdep_set_class(&inode->i_mutex,
800 &type->i_mutex_dir_key);
805 * This is special! We do not need the spinlock when clearing I_NEW,
806 * because we're guaranteed that nobody else tries to do anything about
807 * the state of the inode when it is locked, as we just created it (so
808 * there can be no old holders that haven't tested I_NEW).
809 * However we must emit the memory barrier so that other CPUs reliably
810 * see the clearing of I_NEW after the other inode initialisation has
814 WARN_ON(!(inode->i_state & I_NEW));
815 inode->i_state &= ~I_NEW;
816 wake_up_inode(inode);
818 EXPORT_SYMBOL(unlock_new_inode);
821 * This is called without the inode lock held.. Be careful.
823 * We no longer cache the sb_flags in i_flags - see fs.h
824 * -- rmk@arm.uk.linux.org
826 static struct inode *get_new_inode(struct super_block *sb,
827 struct hlist_head *head,
828 int (*test)(struct inode *, void *),
829 int (*set)(struct inode *, void *),
834 inode = alloc_inode(sb);
838 spin_lock(&inode_lock);
839 /* We released the lock, so.. */
840 old = find_inode(sb, head, test, data);
842 if (set(inode, data))
845 hlist_add_head(&inode->i_hash, head);
846 __inode_sb_list_add(inode);
847 inode->i_state = I_NEW;
848 spin_unlock(&inode_lock);
850 /* Return the locked inode with I_NEW set, the
851 * caller is responsible for filling in the contents
857 * Uhhuh, somebody else created the same inode under
858 * us. Use the old inode instead of the one we just
861 spin_unlock(&inode_lock);
862 destroy_inode(inode);
864 wait_on_inode(inode);
869 spin_unlock(&inode_lock);
870 destroy_inode(inode);
875 * get_new_inode_fast is the fast path version of get_new_inode, see the
876 * comment at iget_locked for details.
878 static struct inode *get_new_inode_fast(struct super_block *sb,
879 struct hlist_head *head, unsigned long ino)
883 inode = alloc_inode(sb);
887 spin_lock(&inode_lock);
888 /* We released the lock, so.. */
889 old = find_inode_fast(sb, head, ino);
892 hlist_add_head(&inode->i_hash, head);
893 __inode_sb_list_add(inode);
894 inode->i_state = I_NEW;
895 spin_unlock(&inode_lock);
897 /* Return the locked inode with I_NEW set, the
898 * caller is responsible for filling in the contents
904 * Uhhuh, somebody else created the same inode under
905 * us. Use the old inode instead of the one we just
908 spin_unlock(&inode_lock);
909 destroy_inode(inode);
911 wait_on_inode(inode);
917 * search the inode cache for a matching inode number.
918 * If we find one, then the inode number we are trying to
919 * allocate is not unique and so we should not use it.
921 * Returns 1 if the inode number is unique, 0 if it is not.
923 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
925 struct hlist_head *b = inode_hashtable + hash(sb, ino);
926 struct hlist_node *node;
929 hlist_for_each_entry(inode, node, b, i_hash) {
930 if (inode->i_ino == ino && inode->i_sb == sb)
938 * iunique - get a unique inode number
940 * @max_reserved: highest reserved inode number
942 * Obtain an inode number that is unique on the system for a given
943 * superblock. This is used by file systems that have no natural
944 * permanent inode numbering system. An inode number is returned that
945 * is higher than the reserved limit but unique.
948 * With a large number of inodes live on the file system this function
949 * currently becomes quite slow.
951 ino_t iunique(struct super_block *sb, ino_t max_reserved)
954 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
955 * error if st_ino won't fit in target struct field. Use 32bit counter
956 * here to attempt to avoid that.
958 static DEFINE_SPINLOCK(iunique_lock);
959 static unsigned int counter;
962 spin_lock(&inode_lock);
963 spin_lock(&iunique_lock);
965 if (counter <= max_reserved)
966 counter = max_reserved + 1;
968 } while (!test_inode_iunique(sb, res));
969 spin_unlock(&iunique_lock);
970 spin_unlock(&inode_lock);
974 EXPORT_SYMBOL(iunique);
976 struct inode *igrab(struct inode *inode)
978 spin_lock(&inode_lock);
979 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
983 * Handle the case where s_op->clear_inode is not been
984 * called yet, and somebody is calling igrab
985 * while the inode is getting freed.
988 spin_unlock(&inode_lock);
991 EXPORT_SYMBOL(igrab);
994 * ifind - internal function, you want ilookup5() or iget5().
995 * @sb: super block of file system to search
996 * @head: the head of the list to search
997 * @test: callback used for comparisons between inodes
998 * @data: opaque data pointer to pass to @test
999 * @wait: if true wait for the inode to be unlocked, if false do not
1001 * ifind() searches for the inode specified by @data in the inode
1002 * cache. This is a generalized version of ifind_fast() for file systems where
1003 * the inode number is not sufficient for unique identification of an inode.
1005 * If the inode is in the cache, the inode is returned with an incremented
1008 * Otherwise NULL is returned.
1010 * Note, @test is called with the inode_lock held, so can't sleep.
1012 static struct inode *ifind(struct super_block *sb,
1013 struct hlist_head *head, int (*test)(struct inode *, void *),
1014 void *data, const int wait)
1016 struct inode *inode;
1018 spin_lock(&inode_lock);
1019 inode = find_inode(sb, head, test, data);
1021 spin_unlock(&inode_lock);
1023 wait_on_inode(inode);
1026 spin_unlock(&inode_lock);
1031 * ifind_fast - internal function, you want ilookup() or iget().
1032 * @sb: super block of file system to search
1033 * @head: head of the list to search
1034 * @ino: inode number to search for
1036 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1037 * file systems where the inode number is sufficient for unique identification
1040 * If the inode is in the cache, the inode is returned with an incremented
1043 * Otherwise NULL is returned.
1045 static struct inode *ifind_fast(struct super_block *sb,
1046 struct hlist_head *head, unsigned long ino)
1048 struct inode *inode;
1050 spin_lock(&inode_lock);
1051 inode = find_inode_fast(sb, head, ino);
1053 spin_unlock(&inode_lock);
1054 wait_on_inode(inode);
1057 spin_unlock(&inode_lock);
1062 * ilookup5_nowait - search for an inode in the inode cache
1063 * @sb: super block of file system to search
1064 * @hashval: hash value (usually inode number) to search for
1065 * @test: callback used for comparisons between inodes
1066 * @data: opaque data pointer to pass to @test
1068 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1069 * @data in the inode cache. This is a generalized version of ilookup() for
1070 * file systems where the inode number is not sufficient for unique
1071 * identification of an inode.
1073 * If the inode is in the cache, the inode is returned with an incremented
1074 * reference count. Note, the inode lock is not waited upon so you have to be
1075 * very careful what you do with the returned inode. You probably should be
1076 * using ilookup5() instead.
1078 * Otherwise NULL is returned.
1080 * Note, @test is called with the inode_lock held, so can't sleep.
1082 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1083 int (*test)(struct inode *, void *), void *data)
1085 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1087 return ifind(sb, head, test, data, 0);
1089 EXPORT_SYMBOL(ilookup5_nowait);
1092 * ilookup5 - search for an inode in the inode cache
1093 * @sb: super block of file system to search
1094 * @hashval: hash value (usually inode number) to search for
1095 * @test: callback used for comparisons between inodes
1096 * @data: opaque data pointer to pass to @test
1098 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1099 * @data in the inode cache. This is a generalized version of ilookup() for
1100 * file systems where the inode number is not sufficient for unique
1101 * identification of an inode.
1103 * If the inode is in the cache, the inode lock is waited upon and the inode is
1104 * returned with an incremented reference count.
1106 * Otherwise NULL is returned.
1108 * Note, @test is called with the inode_lock held, so can't sleep.
1110 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1111 int (*test)(struct inode *, void *), void *data)
1113 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1115 return ifind(sb, head, test, data, 1);
1117 EXPORT_SYMBOL(ilookup5);
1120 * ilookup - search for an inode in the inode cache
1121 * @sb: super block of file system to search
1122 * @ino: inode number to search for
1124 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1125 * This is for file systems where the inode number is sufficient for unique
1126 * identification of an inode.
1128 * If the inode is in the cache, the inode is returned with an incremented
1131 * Otherwise NULL is returned.
1133 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1135 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1137 return ifind_fast(sb, head, ino);
1139 EXPORT_SYMBOL(ilookup);
1142 * iget5_locked - obtain an inode from a mounted file system
1143 * @sb: super block of file system
1144 * @hashval: hash value (usually inode number) to get
1145 * @test: callback used for comparisons between inodes
1146 * @set: callback used to initialize a new struct inode
1147 * @data: opaque data pointer to pass to @test and @set
1149 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1150 * and @data in the inode cache and if present it is returned with an increased
1151 * reference count. This is a generalized version of iget_locked() for file
1152 * systems where the inode number is not sufficient for unique identification
1155 * If the inode is not in cache, get_new_inode() is called to allocate a new
1156 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1157 * file system gets to fill it in before unlocking it via unlock_new_inode().
1159 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1161 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1162 int (*test)(struct inode *, void *),
1163 int (*set)(struct inode *, void *), void *data)
1165 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1166 struct inode *inode;
1168 inode = ifind(sb, head, test, data, 1);
1172 * get_new_inode() will do the right thing, re-trying the search
1173 * in case it had to block at any point.
1175 return get_new_inode(sb, head, test, set, data);
1177 EXPORT_SYMBOL(iget5_locked);
1180 * iget_locked - obtain an inode from a mounted file system
1181 * @sb: super block of file system
1182 * @ino: inode number to get
1184 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1185 * the inode cache and if present it is returned with an increased reference
1186 * count. This is for file systems where the inode number is sufficient for
1187 * unique identification of an inode.
1189 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1190 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1191 * The file system gets to fill it in before unlocking it via
1192 * unlock_new_inode().
1194 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1196 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1197 struct inode *inode;
1199 inode = ifind_fast(sb, head, ino);
1203 * get_new_inode_fast() will do the right thing, re-trying the search
1204 * in case it had to block at any point.
1206 return get_new_inode_fast(sb, head, ino);
1208 EXPORT_SYMBOL(iget_locked);
1210 int insert_inode_locked(struct inode *inode)
1212 struct super_block *sb = inode->i_sb;
1213 ino_t ino = inode->i_ino;
1214 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1216 inode->i_state |= I_NEW;
1218 struct hlist_node *node;
1219 struct inode *old = NULL;
1220 spin_lock(&inode_lock);
1221 hlist_for_each_entry(old, node, head, i_hash) {
1222 if (old->i_ino != ino)
1224 if (old->i_sb != sb)
1226 if (old->i_state & (I_FREEING|I_WILL_FREE))
1230 if (likely(!node)) {
1231 hlist_add_head(&inode->i_hash, head);
1232 spin_unlock(&inode_lock);
1236 spin_unlock(&inode_lock);
1238 if (unlikely(!inode_unhashed(old))) {
1245 EXPORT_SYMBOL(insert_inode_locked);
1247 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1248 int (*test)(struct inode *, void *), void *data)
1250 struct super_block *sb = inode->i_sb;
1251 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1253 inode->i_state |= I_NEW;
1256 struct hlist_node *node;
1257 struct inode *old = NULL;
1259 spin_lock(&inode_lock);
1260 hlist_for_each_entry(old, node, head, i_hash) {
1261 if (old->i_sb != sb)
1263 if (!test(old, data))
1265 if (old->i_state & (I_FREEING|I_WILL_FREE))
1269 if (likely(!node)) {
1270 hlist_add_head(&inode->i_hash, head);
1271 spin_unlock(&inode_lock);
1275 spin_unlock(&inode_lock);
1277 if (unlikely(!inode_unhashed(old))) {
1284 EXPORT_SYMBOL(insert_inode_locked4);
1287 int generic_delete_inode(struct inode *inode)
1291 EXPORT_SYMBOL(generic_delete_inode);
1294 * Normal UNIX filesystem behaviour: delete the
1295 * inode when the usage count drops to zero, and
1298 int generic_drop_inode(struct inode *inode)
1300 return !inode->i_nlink || inode_unhashed(inode);
1302 EXPORT_SYMBOL_GPL(generic_drop_inode);
1305 * Called when we're dropping the last reference
1308 * Call the FS "drop_inode()" function, defaulting to
1309 * the legacy UNIX filesystem behaviour. If it tells
1310 * us to evict inode, do so. Otherwise, retain inode
1311 * in cache if fs is alive, sync and evict if fs is
1314 static void iput_final(struct inode *inode)
1316 struct super_block *sb = inode->i_sb;
1317 const struct super_operations *op = inode->i_sb->s_op;
1320 if (op && op->drop_inode)
1321 drop = op->drop_inode(inode);
1323 drop = generic_drop_inode(inode);
1326 if (sb->s_flags & MS_ACTIVE) {
1327 inode->i_state |= I_REFERENCED;
1328 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1329 inode_lru_list_add(inode);
1331 spin_unlock(&inode_lock);
1334 WARN_ON(inode->i_state & I_NEW);
1335 inode->i_state |= I_WILL_FREE;
1336 spin_unlock(&inode_lock);
1337 write_inode_now(inode, 1);
1338 spin_lock(&inode_lock);
1339 WARN_ON(inode->i_state & I_NEW);
1340 inode->i_state &= ~I_WILL_FREE;
1341 __remove_inode_hash(inode);
1343 WARN_ON(inode->i_state & I_NEW);
1344 inode->i_state |= I_FREEING;
1347 * After we delete the inode from the LRU here, we avoid moving dirty
1348 * inodes back onto the LRU now because I_FREEING is set and hence
1349 * writeback_single_inode() won't move the inode around.
1351 inode_lru_list_del(inode);
1353 __inode_sb_list_del(inode);
1354 spin_unlock(&inode_lock);
1356 remove_inode_hash(inode);
1357 wake_up_inode(inode);
1358 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1359 destroy_inode(inode);
1363 * iput - put an inode
1364 * @inode: inode to put
1366 * Puts an inode, dropping its usage count. If the inode use count hits
1367 * zero, the inode is then freed and may also be destroyed.
1369 * Consequently, iput() can sleep.
1371 void iput(struct inode *inode)
1374 BUG_ON(inode->i_state & I_CLEAR);
1376 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1380 EXPORT_SYMBOL(iput);
1383 * bmap - find a block number in a file
1384 * @inode: inode of file
1385 * @block: block to find
1387 * Returns the block number on the device holding the inode that
1388 * is the disk block number for the block of the file requested.
1389 * That is, asked for block 4 of inode 1 the function will return the
1390 * disk block relative to the disk start that holds that block of the
1393 sector_t bmap(struct inode *inode, sector_t block)
1396 if (inode->i_mapping->a_ops->bmap)
1397 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1400 EXPORT_SYMBOL(bmap);
1403 * With relative atime, only update atime if the previous atime is
1404 * earlier than either the ctime or mtime or if at least a day has
1405 * passed since the last atime update.
1407 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1408 struct timespec now)
1411 if (!(mnt->mnt_flags & MNT_RELATIME))
1414 * Is mtime younger than atime? If yes, update atime:
1416 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1419 * Is ctime younger than atime? If yes, update atime:
1421 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1425 * Is the previous atime value older than a day? If yes,
1428 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1431 * Good, we can skip the atime update:
1437 * touch_atime - update the access time
1438 * @mnt: mount the inode is accessed on
1439 * @dentry: dentry accessed
1441 * Update the accessed time on an inode and mark it for writeback.
1442 * This function automatically handles read only file systems and media,
1443 * as well as the "noatime" flag and inode specific "noatime" markers.
1445 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1447 struct inode *inode = dentry->d_inode;
1448 struct timespec now;
1450 if (inode->i_flags & S_NOATIME)
1452 if (IS_NOATIME(inode))
1454 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1457 if (mnt->mnt_flags & MNT_NOATIME)
1459 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1462 now = current_fs_time(inode->i_sb);
1464 if (!relatime_need_update(mnt, inode, now))
1467 if (timespec_equal(&inode->i_atime, &now))
1470 if (mnt_want_write(mnt))
1473 inode->i_atime = now;
1474 mark_inode_dirty_sync(inode);
1475 mnt_drop_write(mnt);
1477 EXPORT_SYMBOL(touch_atime);
1480 * file_update_time - update mtime and ctime time
1481 * @file: file accessed
1483 * Update the mtime and ctime members of an inode and mark the inode
1484 * for writeback. Note that this function is meant exclusively for
1485 * usage in the file write path of filesystems, and filesystems may
1486 * choose to explicitly ignore update via this function with the
1487 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1488 * timestamps are handled by the server.
1491 void file_update_time(struct file *file)
1493 struct inode *inode = file->f_path.dentry->d_inode;
1494 struct timespec now;
1495 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1497 /* First try to exhaust all avenues to not sync */
1498 if (IS_NOCMTIME(inode))
1501 now = current_fs_time(inode->i_sb);
1502 if (!timespec_equal(&inode->i_mtime, &now))
1505 if (!timespec_equal(&inode->i_ctime, &now))
1508 if (IS_I_VERSION(inode))
1509 sync_it |= S_VERSION;
1514 /* Finally allowed to write? Takes lock. */
1515 if (mnt_want_write_file(file))
1518 /* Only change inode inside the lock region */
1519 if (sync_it & S_VERSION)
1520 inode_inc_iversion(inode);
1521 if (sync_it & S_CTIME)
1522 inode->i_ctime = now;
1523 if (sync_it & S_MTIME)
1524 inode->i_mtime = now;
1525 mark_inode_dirty_sync(inode);
1526 mnt_drop_write(file->f_path.mnt);
1528 EXPORT_SYMBOL(file_update_time);
1530 int inode_needs_sync(struct inode *inode)
1534 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1538 EXPORT_SYMBOL(inode_needs_sync);
1540 int inode_wait(void *word)
1545 EXPORT_SYMBOL(inode_wait);
1548 * If we try to find an inode in the inode hash while it is being
1549 * deleted, we have to wait until the filesystem completes its
1550 * deletion before reporting that it isn't found. This function waits
1551 * until the deletion _might_ have completed. Callers are responsible
1552 * to recheck inode state.
1554 * It doesn't matter if I_NEW is not set initially, a call to
1555 * wake_up_inode() after removing from the hash list will DTRT.
1557 * This is called with inode_lock held.
1559 static void __wait_on_freeing_inode(struct inode *inode)
1561 wait_queue_head_t *wq;
1562 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1563 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1564 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1565 spin_unlock(&inode_lock);
1567 finish_wait(wq, &wait.wait);
1568 spin_lock(&inode_lock);
1571 static __initdata unsigned long ihash_entries;
1572 static int __init set_ihash_entries(char *str)
1576 ihash_entries = simple_strtoul(str, &str, 0);
1579 __setup("ihash_entries=", set_ihash_entries);
1582 * Initialize the waitqueues and inode hash table.
1584 void __init inode_init_early(void)
1588 /* If hashes are distributed across NUMA nodes, defer
1589 * hash allocation until vmalloc space is available.
1595 alloc_large_system_hash("Inode-cache",
1596 sizeof(struct hlist_head),
1604 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1605 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1608 void __init inode_init(void)
1612 /* inode slab cache */
1613 inode_cachep = kmem_cache_create("inode_cache",
1614 sizeof(struct inode),
1616 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1619 register_shrinker(&icache_shrinker);
1620 percpu_counter_init(&nr_inodes, 0);
1621 percpu_counter_init(&nr_inodes_unused, 0);
1623 /* Hash may have been set up in inode_init_early */
1628 alloc_large_system_hash("Inode-cache",
1629 sizeof(struct hlist_head),
1637 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1638 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1641 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1643 inode->i_mode = mode;
1644 if (S_ISCHR(mode)) {
1645 inode->i_fop = &def_chr_fops;
1646 inode->i_rdev = rdev;
1647 } else if (S_ISBLK(mode)) {
1648 inode->i_fop = &def_blk_fops;
1649 inode->i_rdev = rdev;
1650 } else if (S_ISFIFO(mode))
1651 inode->i_fop = &def_fifo_fops;
1652 else if (S_ISSOCK(mode))
1653 inode->i_fop = &bad_sock_fops;
1655 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1656 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1659 EXPORT_SYMBOL(init_special_inode);
1662 * Init uid,gid,mode for new inode according to posix standards
1664 * @dir: Directory inode
1665 * @mode: mode of the new inode
1667 void inode_init_owner(struct inode *inode, const struct inode *dir,
1670 inode->i_uid = current_fsuid();
1671 if (dir && dir->i_mode & S_ISGID) {
1672 inode->i_gid = dir->i_gid;
1676 inode->i_gid = current_fsgid();
1677 inode->i_mode = mode;
1679 EXPORT_SYMBOL(inode_init_owner);