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);
324 * get additional reference to inode; caller must already hold one.
326 void ihold(struct inode *inode)
328 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
330 EXPORT_SYMBOL(ihold);
332 static void inode_lru_list_add(struct inode *inode)
334 if (list_empty(&inode->i_list)) {
335 list_add(&inode->i_list, &inode_unused);
336 percpu_counter_inc(&nr_inodes_unused);
340 static void inode_lru_list_del(struct inode *inode)
342 if (!list_empty(&inode->i_list)) {
343 list_del_init(&inode->i_list);
344 percpu_counter_dec(&nr_inodes_unused);
348 static inline void __inode_sb_list_add(struct inode *inode)
350 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
354 * inode_sb_list_add - add inode to the superblock list of inodes
355 * @inode: inode to add
357 void inode_sb_list_add(struct inode *inode)
359 spin_lock(&inode_lock);
360 __inode_sb_list_add(inode);
361 spin_unlock(&inode_lock);
363 EXPORT_SYMBOL_GPL(inode_sb_list_add);
365 static inline void __inode_sb_list_del(struct inode *inode)
367 list_del_init(&inode->i_sb_list);
370 static unsigned long hash(struct super_block *sb, unsigned long hashval)
374 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
376 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
377 return tmp & I_HASHMASK;
381 * __insert_inode_hash - hash an inode
382 * @inode: unhashed inode
383 * @hashval: unsigned long value used to locate this object in the
386 * Add an inode to the inode hash for this superblock.
388 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
390 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
392 spin_lock(&inode_lock);
393 hlist_add_head(&inode->i_hash, b);
394 spin_unlock(&inode_lock);
396 EXPORT_SYMBOL(__insert_inode_hash);
399 * __remove_inode_hash - remove an inode from the hash
400 * @inode: inode to unhash
402 * Remove an inode from the superblock.
404 static void __remove_inode_hash(struct inode *inode)
406 hlist_del_init(&inode->i_hash);
410 * remove_inode_hash - remove an inode from the hash
411 * @inode: inode to unhash
413 * Remove an inode from the superblock.
415 void remove_inode_hash(struct inode *inode)
417 spin_lock(&inode_lock);
418 hlist_del_init(&inode->i_hash);
419 spin_unlock(&inode_lock);
421 EXPORT_SYMBOL(remove_inode_hash);
423 void end_writeback(struct inode *inode)
426 BUG_ON(inode->i_data.nrpages);
427 BUG_ON(!list_empty(&inode->i_data.private_list));
428 BUG_ON(!(inode->i_state & I_FREEING));
429 BUG_ON(inode->i_state & I_CLEAR);
430 inode_sync_wait(inode);
431 inode->i_state = I_FREEING | I_CLEAR;
433 EXPORT_SYMBOL(end_writeback);
435 static void evict(struct inode *inode)
437 const struct super_operations *op = inode->i_sb->s_op;
439 if (op->evict_inode) {
440 op->evict_inode(inode);
442 if (inode->i_data.nrpages)
443 truncate_inode_pages(&inode->i_data, 0);
444 end_writeback(inode);
446 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
448 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
453 * dispose_list - dispose of the contents of a local list
454 * @head: the head of the list to free
456 * Dispose-list gets a local list with local inodes in it, so it doesn't
457 * need to worry about list corruption and SMP locks.
459 static void dispose_list(struct list_head *head)
461 while (!list_empty(head)) {
464 inode = list_first_entry(head, struct inode, i_list);
465 list_del_init(&inode->i_list);
469 spin_lock(&inode_lock);
470 __remove_inode_hash(inode);
471 __inode_sb_list_del(inode);
472 spin_unlock(&inode_lock);
474 wake_up_inode(inode);
475 destroy_inode(inode);
480 * Invalidate all inodes for a device.
482 static int invalidate_list(struct list_head *head, struct list_head *dispose)
484 struct list_head *next;
489 struct list_head *tmp = next;
493 * We can reschedule here without worrying about the list's
494 * consistency because the per-sb list of inodes must not
495 * change during umount anymore, and because iprune_sem keeps
496 * shrink_icache_memory() away.
498 cond_resched_lock(&inode_lock);
503 inode = list_entry(tmp, struct inode, i_sb_list);
504 if (inode->i_state & I_NEW)
506 invalidate_inode_buffers(inode);
507 if (!atomic_read(&inode->i_count)) {
508 list_move(&inode->i_list, dispose);
509 WARN_ON(inode->i_state & I_NEW);
510 inode->i_state |= I_FREEING;
511 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
512 percpu_counter_dec(&nr_inodes_unused);
521 * invalidate_inodes - discard the inodes on a device
524 * Discard all of the inodes for a given superblock. If the discard
525 * fails because there are busy inodes then a non zero value is returned.
526 * If the discard is successful all the inodes have been discarded.
528 int invalidate_inodes(struct super_block *sb)
531 LIST_HEAD(throw_away);
533 down_write(&iprune_sem);
534 spin_lock(&inode_lock);
535 fsnotify_unmount_inodes(&sb->s_inodes);
536 busy = invalidate_list(&sb->s_inodes, &throw_away);
537 spin_unlock(&inode_lock);
539 dispose_list(&throw_away);
540 up_write(&iprune_sem);
545 static int can_unuse(struct inode *inode)
547 if (inode->i_state & ~I_REFERENCED)
549 if (inode_has_buffers(inode))
551 if (atomic_read(&inode->i_count))
553 if (inode->i_data.nrpages)
559 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
560 * temporary list and then are freed outside inode_lock by dispose_list().
562 * Any inodes which are pinned purely because of attached pagecache have their
563 * pagecache removed. If the inode has metadata buffers attached to
564 * mapping->private_list then try to remove them.
566 * If the inode has the I_REFERENCED flag set, then it means that it has been
567 * used recently - the flag is set in iput_final(). When we encounter such an
568 * inode, clear the flag and move it to the back of the LRU so it gets another
569 * pass through the LRU before it gets reclaimed. This is necessary because of
570 * the fact we are doing lazy LRU updates to minimise lock contention so the
571 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
572 * with this flag set because they are the inodes that are out of order.
574 static void prune_icache(int nr_to_scan)
578 unsigned long reap = 0;
580 down_read(&iprune_sem);
581 spin_lock(&inode_lock);
582 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
585 if (list_empty(&inode_unused))
588 inode = list_entry(inode_unused.prev, struct inode, i_list);
591 * Referenced or dirty inodes are still in use. Give them
592 * another pass through the LRU as we canot reclaim them now.
594 if (atomic_read(&inode->i_count) ||
595 (inode->i_state & ~I_REFERENCED)) {
596 list_del_init(&inode->i_list);
597 percpu_counter_dec(&nr_inodes_unused);
601 /* recently referenced inodes get one more pass */
602 if (inode->i_state & I_REFERENCED) {
603 list_move(&inode->i_list, &inode_unused);
604 inode->i_state &= ~I_REFERENCED;
607 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
609 spin_unlock(&inode_lock);
610 if (remove_inode_buffers(inode))
611 reap += invalidate_mapping_pages(&inode->i_data,
614 spin_lock(&inode_lock);
616 if (inode != list_entry(inode_unused.next,
617 struct inode, i_list))
618 continue; /* wrong inode or list_empty */
619 if (!can_unuse(inode))
622 list_move(&inode->i_list, &freeable);
623 WARN_ON(inode->i_state & I_NEW);
624 inode->i_state |= I_FREEING;
625 percpu_counter_dec(&nr_inodes_unused);
627 if (current_is_kswapd())
628 __count_vm_events(KSWAPD_INODESTEAL, reap);
630 __count_vm_events(PGINODESTEAL, reap);
631 spin_unlock(&inode_lock);
633 dispose_list(&freeable);
634 up_read(&iprune_sem);
638 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
639 * "unused" means that no dentries are referring to the inodes: the files are
640 * not open and the dcache references to those inodes have already been
643 * This function is passed the number of inodes to scan, and it returns the
644 * total number of remaining possibly-reclaimable inodes.
646 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
650 * Nasty deadlock avoidance. We may hold various FS locks,
651 * and we don't want to recurse into the FS that called us
652 * in clear_inode() and friends..
654 if (!(gfp_mask & __GFP_FS))
658 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
661 static struct shrinker icache_shrinker = {
662 .shrink = shrink_icache_memory,
663 .seeks = DEFAULT_SEEKS,
666 static void __wait_on_freeing_inode(struct inode *inode);
668 * Called with the inode lock held.
670 static struct inode *find_inode(struct super_block *sb,
671 struct hlist_head *head,
672 int (*test)(struct inode *, void *),
675 struct hlist_node *node;
676 struct inode *inode = NULL;
679 hlist_for_each_entry(inode, node, head, i_hash) {
680 if (inode->i_sb != sb)
682 if (!test(inode, data))
684 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
685 __wait_on_freeing_inode(inode);
695 * find_inode_fast is the fast path version of find_inode, see the comment at
696 * iget_locked for details.
698 static struct inode *find_inode_fast(struct super_block *sb,
699 struct hlist_head *head, unsigned long ino)
701 struct hlist_node *node;
702 struct inode *inode = NULL;
705 hlist_for_each_entry(inode, node, head, i_hash) {
706 if (inode->i_ino != ino)
708 if (inode->i_sb != sb)
710 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
711 __wait_on_freeing_inode(inode);
721 * Each cpu owns a range of LAST_INO_BATCH numbers.
722 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
723 * to renew the exhausted range.
725 * This does not significantly increase overflow rate because every CPU can
726 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
727 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
728 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
729 * overflow rate by 2x, which does not seem too significant.
731 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
732 * error if st_ino won't fit in target struct field. Use 32bit counter
733 * here to attempt to avoid that.
735 #define LAST_INO_BATCH 1024
736 static DEFINE_PER_CPU(unsigned int, last_ino);
738 static unsigned int get_next_ino(void)
740 unsigned int *p = &get_cpu_var(last_ino);
741 unsigned int res = *p;
744 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
745 static atomic_t shared_last_ino;
746 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
748 res = next - LAST_INO_BATCH;
753 put_cpu_var(last_ino);
758 * new_inode - obtain an inode
761 * Allocates a new inode for given superblock. The default gfp_mask
762 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
763 * If HIGHMEM pages are unsuitable or it is known that pages allocated
764 * for the page cache are not reclaimable or migratable,
765 * mapping_set_gfp_mask() must be called with suitable flags on the
766 * newly created inode's mapping
769 struct inode *new_inode(struct super_block *sb)
773 spin_lock_prefetch(&inode_lock);
775 inode = alloc_inode(sb);
777 spin_lock(&inode_lock);
778 __inode_sb_list_add(inode);
779 inode->i_ino = get_next_ino();
781 spin_unlock(&inode_lock);
785 EXPORT_SYMBOL(new_inode);
787 void unlock_new_inode(struct inode *inode)
789 #ifdef CONFIG_DEBUG_LOCK_ALLOC
790 if (S_ISDIR(inode->i_mode)) {
791 struct file_system_type *type = inode->i_sb->s_type;
793 /* Set new key only if filesystem hasn't already changed it */
794 if (!lockdep_match_class(&inode->i_mutex,
795 &type->i_mutex_key)) {
797 * ensure nobody is actually holding i_mutex
799 mutex_destroy(&inode->i_mutex);
800 mutex_init(&inode->i_mutex);
801 lockdep_set_class(&inode->i_mutex,
802 &type->i_mutex_dir_key);
807 * This is special! We do not need the spinlock when clearing I_NEW,
808 * because we're guaranteed that nobody else tries to do anything about
809 * the state of the inode when it is locked, as we just created it (so
810 * there can be no old holders that haven't tested I_NEW).
811 * However we must emit the memory barrier so that other CPUs reliably
812 * see the clearing of I_NEW after the other inode initialisation has
816 WARN_ON(!(inode->i_state & I_NEW));
817 inode->i_state &= ~I_NEW;
818 wake_up_inode(inode);
820 EXPORT_SYMBOL(unlock_new_inode);
823 * This is called without the inode lock held.. Be careful.
825 * We no longer cache the sb_flags in i_flags - see fs.h
826 * -- rmk@arm.uk.linux.org
828 static struct inode *get_new_inode(struct super_block *sb,
829 struct hlist_head *head,
830 int (*test)(struct inode *, void *),
831 int (*set)(struct inode *, void *),
836 inode = alloc_inode(sb);
840 spin_lock(&inode_lock);
841 /* We released the lock, so.. */
842 old = find_inode(sb, head, test, data);
844 if (set(inode, data))
847 hlist_add_head(&inode->i_hash, head);
848 __inode_sb_list_add(inode);
849 inode->i_state = I_NEW;
850 spin_unlock(&inode_lock);
852 /* Return the locked inode with I_NEW set, the
853 * caller is responsible for filling in the contents
859 * Uhhuh, somebody else created the same inode under
860 * us. Use the old inode instead of the one we just
863 spin_unlock(&inode_lock);
864 destroy_inode(inode);
866 wait_on_inode(inode);
871 spin_unlock(&inode_lock);
872 destroy_inode(inode);
877 * get_new_inode_fast is the fast path version of get_new_inode, see the
878 * comment at iget_locked for details.
880 static struct inode *get_new_inode_fast(struct super_block *sb,
881 struct hlist_head *head, unsigned long ino)
885 inode = alloc_inode(sb);
889 spin_lock(&inode_lock);
890 /* We released the lock, so.. */
891 old = find_inode_fast(sb, head, ino);
894 hlist_add_head(&inode->i_hash, head);
895 __inode_sb_list_add(inode);
896 inode->i_state = I_NEW;
897 spin_unlock(&inode_lock);
899 /* Return the locked inode with I_NEW set, the
900 * caller is responsible for filling in the contents
906 * Uhhuh, somebody else created the same inode under
907 * us. Use the old inode instead of the one we just
910 spin_unlock(&inode_lock);
911 destroy_inode(inode);
913 wait_on_inode(inode);
919 * search the inode cache for a matching inode number.
920 * If we find one, then the inode number we are trying to
921 * allocate is not unique and so we should not use it.
923 * Returns 1 if the inode number is unique, 0 if it is not.
925 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
927 struct hlist_head *b = inode_hashtable + hash(sb, ino);
928 struct hlist_node *node;
931 hlist_for_each_entry(inode, node, b, i_hash) {
932 if (inode->i_ino == ino && inode->i_sb == sb)
940 * iunique - get a unique inode number
942 * @max_reserved: highest reserved inode number
944 * Obtain an inode number that is unique on the system for a given
945 * superblock. This is used by file systems that have no natural
946 * permanent inode numbering system. An inode number is returned that
947 * is higher than the reserved limit but unique.
950 * With a large number of inodes live on the file system this function
951 * currently becomes quite slow.
953 ino_t iunique(struct super_block *sb, ino_t max_reserved)
956 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
957 * error if st_ino won't fit in target struct field. Use 32bit counter
958 * here to attempt to avoid that.
960 static DEFINE_SPINLOCK(iunique_lock);
961 static unsigned int counter;
964 spin_lock(&inode_lock);
965 spin_lock(&iunique_lock);
967 if (counter <= max_reserved)
968 counter = max_reserved + 1;
970 } while (!test_inode_iunique(sb, res));
971 spin_unlock(&iunique_lock);
972 spin_unlock(&inode_lock);
976 EXPORT_SYMBOL(iunique);
978 struct inode *igrab(struct inode *inode)
980 spin_lock(&inode_lock);
981 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
985 * Handle the case where s_op->clear_inode is not been
986 * called yet, and somebody is calling igrab
987 * while the inode is getting freed.
990 spin_unlock(&inode_lock);
993 EXPORT_SYMBOL(igrab);
996 * ifind - internal function, you want ilookup5() or iget5().
997 * @sb: super block of file system to search
998 * @head: the head of the list to search
999 * @test: callback used for comparisons between inodes
1000 * @data: opaque data pointer to pass to @test
1001 * @wait: if true wait for the inode to be unlocked, if false do not
1003 * ifind() searches for the inode specified by @data in the inode
1004 * cache. This is a generalized version of ifind_fast() for file systems where
1005 * the inode number is not sufficient for unique identification of an inode.
1007 * If the inode is in the cache, the inode is returned with an incremented
1010 * Otherwise NULL is returned.
1012 * Note, @test is called with the inode_lock held, so can't sleep.
1014 static struct inode *ifind(struct super_block *sb,
1015 struct hlist_head *head, int (*test)(struct inode *, void *),
1016 void *data, const int wait)
1018 struct inode *inode;
1020 spin_lock(&inode_lock);
1021 inode = find_inode(sb, head, test, data);
1023 spin_unlock(&inode_lock);
1025 wait_on_inode(inode);
1028 spin_unlock(&inode_lock);
1033 * ifind_fast - internal function, you want ilookup() or iget().
1034 * @sb: super block of file system to search
1035 * @head: head of the list to search
1036 * @ino: inode number to search for
1038 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1039 * file systems where the inode number is sufficient for unique identification
1042 * If the inode is in the cache, the inode is returned with an incremented
1045 * Otherwise NULL is returned.
1047 static struct inode *ifind_fast(struct super_block *sb,
1048 struct hlist_head *head, unsigned long ino)
1050 struct inode *inode;
1052 spin_lock(&inode_lock);
1053 inode = find_inode_fast(sb, head, ino);
1055 spin_unlock(&inode_lock);
1056 wait_on_inode(inode);
1059 spin_unlock(&inode_lock);
1064 * ilookup5_nowait - search for an inode in the inode cache
1065 * @sb: super block of file system to search
1066 * @hashval: hash value (usually inode number) to search for
1067 * @test: callback used for comparisons between inodes
1068 * @data: opaque data pointer to pass to @test
1070 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1071 * @data in the inode cache. This is a generalized version of ilookup() for
1072 * file systems where the inode number is not sufficient for unique
1073 * identification of an inode.
1075 * If the inode is in the cache, the inode is returned with an incremented
1076 * reference count. Note, the inode lock is not waited upon so you have to be
1077 * very careful what you do with the returned inode. You probably should be
1078 * using ilookup5() instead.
1080 * Otherwise NULL is returned.
1082 * Note, @test is called with the inode_lock held, so can't sleep.
1084 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1085 int (*test)(struct inode *, void *), void *data)
1087 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1089 return ifind(sb, head, test, data, 0);
1091 EXPORT_SYMBOL(ilookup5_nowait);
1094 * ilookup5 - search for an inode in the inode cache
1095 * @sb: super block of file system to search
1096 * @hashval: hash value (usually inode number) to search for
1097 * @test: callback used for comparisons between inodes
1098 * @data: opaque data pointer to pass to @test
1100 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1101 * @data in the inode cache. This is a generalized version of ilookup() for
1102 * file systems where the inode number is not sufficient for unique
1103 * identification of an inode.
1105 * If the inode is in the cache, the inode lock is waited upon and the inode is
1106 * returned with an incremented reference count.
1108 * Otherwise NULL is returned.
1110 * Note, @test is called with the inode_lock held, so can't sleep.
1112 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1113 int (*test)(struct inode *, void *), void *data)
1115 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1117 return ifind(sb, head, test, data, 1);
1119 EXPORT_SYMBOL(ilookup5);
1122 * ilookup - search for an inode in the inode cache
1123 * @sb: super block of file system to search
1124 * @ino: inode number to search for
1126 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1127 * This is for file systems where the inode number is sufficient for unique
1128 * identification of an inode.
1130 * If the inode is in the cache, the inode is returned with an incremented
1133 * Otherwise NULL is returned.
1135 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1137 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1139 return ifind_fast(sb, head, ino);
1141 EXPORT_SYMBOL(ilookup);
1144 * iget5_locked - obtain an inode from a mounted file system
1145 * @sb: super block of file system
1146 * @hashval: hash value (usually inode number) to get
1147 * @test: callback used for comparisons between inodes
1148 * @set: callback used to initialize a new struct inode
1149 * @data: opaque data pointer to pass to @test and @set
1151 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1152 * and @data in the inode cache and if present it is returned with an increased
1153 * reference count. This is a generalized version of iget_locked() for file
1154 * systems where the inode number is not sufficient for unique identification
1157 * If the inode is not in cache, get_new_inode() is called to allocate a new
1158 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1159 * file system gets to fill it in before unlocking it via unlock_new_inode().
1161 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1163 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1164 int (*test)(struct inode *, void *),
1165 int (*set)(struct inode *, void *), void *data)
1167 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1168 struct inode *inode;
1170 inode = ifind(sb, head, test, data, 1);
1174 * get_new_inode() will do the right thing, re-trying the search
1175 * in case it had to block at any point.
1177 return get_new_inode(sb, head, test, set, data);
1179 EXPORT_SYMBOL(iget5_locked);
1182 * iget_locked - obtain an inode from a mounted file system
1183 * @sb: super block of file system
1184 * @ino: inode number to get
1186 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1187 * the inode cache and if present it is returned with an increased reference
1188 * count. This is for file systems where the inode number is sufficient for
1189 * unique identification of an inode.
1191 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1192 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1193 * The file system gets to fill it in before unlocking it via
1194 * unlock_new_inode().
1196 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1198 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1199 struct inode *inode;
1201 inode = ifind_fast(sb, head, ino);
1205 * get_new_inode_fast() will do the right thing, re-trying the search
1206 * in case it had to block at any point.
1208 return get_new_inode_fast(sb, head, ino);
1210 EXPORT_SYMBOL(iget_locked);
1212 int insert_inode_locked(struct inode *inode)
1214 struct super_block *sb = inode->i_sb;
1215 ino_t ino = inode->i_ino;
1216 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1218 inode->i_state |= I_NEW;
1220 struct hlist_node *node;
1221 struct inode *old = NULL;
1222 spin_lock(&inode_lock);
1223 hlist_for_each_entry(old, node, head, i_hash) {
1224 if (old->i_ino != ino)
1226 if (old->i_sb != sb)
1228 if (old->i_state & (I_FREEING|I_WILL_FREE))
1232 if (likely(!node)) {
1233 hlist_add_head(&inode->i_hash, head);
1234 spin_unlock(&inode_lock);
1238 spin_unlock(&inode_lock);
1240 if (unlikely(!inode_unhashed(old))) {
1247 EXPORT_SYMBOL(insert_inode_locked);
1249 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1250 int (*test)(struct inode *, void *), void *data)
1252 struct super_block *sb = inode->i_sb;
1253 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1255 inode->i_state |= I_NEW;
1258 struct hlist_node *node;
1259 struct inode *old = NULL;
1261 spin_lock(&inode_lock);
1262 hlist_for_each_entry(old, node, head, i_hash) {
1263 if (old->i_sb != sb)
1265 if (!test(old, data))
1267 if (old->i_state & (I_FREEING|I_WILL_FREE))
1271 if (likely(!node)) {
1272 hlist_add_head(&inode->i_hash, head);
1273 spin_unlock(&inode_lock);
1277 spin_unlock(&inode_lock);
1279 if (unlikely(!inode_unhashed(old))) {
1286 EXPORT_SYMBOL(insert_inode_locked4);
1289 int generic_delete_inode(struct inode *inode)
1293 EXPORT_SYMBOL(generic_delete_inode);
1296 * Normal UNIX filesystem behaviour: delete the
1297 * inode when the usage count drops to zero, and
1300 int generic_drop_inode(struct inode *inode)
1302 return !inode->i_nlink || inode_unhashed(inode);
1304 EXPORT_SYMBOL_GPL(generic_drop_inode);
1307 * Called when we're dropping the last reference
1310 * Call the FS "drop_inode()" function, defaulting to
1311 * the legacy UNIX filesystem behaviour. If it tells
1312 * us to evict inode, do so. Otherwise, retain inode
1313 * in cache if fs is alive, sync and evict if fs is
1316 static void iput_final(struct inode *inode)
1318 struct super_block *sb = inode->i_sb;
1319 const struct super_operations *op = inode->i_sb->s_op;
1322 if (op && op->drop_inode)
1323 drop = op->drop_inode(inode);
1325 drop = generic_drop_inode(inode);
1328 if (sb->s_flags & MS_ACTIVE) {
1329 inode->i_state |= I_REFERENCED;
1330 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1331 inode_lru_list_add(inode);
1333 spin_unlock(&inode_lock);
1336 WARN_ON(inode->i_state & I_NEW);
1337 inode->i_state |= I_WILL_FREE;
1338 spin_unlock(&inode_lock);
1339 write_inode_now(inode, 1);
1340 spin_lock(&inode_lock);
1341 WARN_ON(inode->i_state & I_NEW);
1342 inode->i_state &= ~I_WILL_FREE;
1343 __remove_inode_hash(inode);
1345 WARN_ON(inode->i_state & I_NEW);
1346 inode->i_state |= I_FREEING;
1349 * After we delete the inode from the LRU here, we avoid moving dirty
1350 * inodes back onto the LRU now because I_FREEING is set and hence
1351 * writeback_single_inode() won't move the inode around.
1353 inode_lru_list_del(inode);
1355 __inode_sb_list_del(inode);
1356 spin_unlock(&inode_lock);
1358 remove_inode_hash(inode);
1359 wake_up_inode(inode);
1360 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1361 destroy_inode(inode);
1365 * iput - put an inode
1366 * @inode: inode to put
1368 * Puts an inode, dropping its usage count. If the inode use count hits
1369 * zero, the inode is then freed and may also be destroyed.
1371 * Consequently, iput() can sleep.
1373 void iput(struct inode *inode)
1376 BUG_ON(inode->i_state & I_CLEAR);
1378 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1382 EXPORT_SYMBOL(iput);
1385 * bmap - find a block number in a file
1386 * @inode: inode of file
1387 * @block: block to find
1389 * Returns the block number on the device holding the inode that
1390 * is the disk block number for the block of the file requested.
1391 * That is, asked for block 4 of inode 1 the function will return the
1392 * disk block relative to the disk start that holds that block of the
1395 sector_t bmap(struct inode *inode, sector_t block)
1398 if (inode->i_mapping->a_ops->bmap)
1399 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1402 EXPORT_SYMBOL(bmap);
1405 * With relative atime, only update atime if the previous atime is
1406 * earlier than either the ctime or mtime or if at least a day has
1407 * passed since the last atime update.
1409 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1410 struct timespec now)
1413 if (!(mnt->mnt_flags & MNT_RELATIME))
1416 * Is mtime younger than atime? If yes, update atime:
1418 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1421 * Is ctime younger than atime? If yes, update atime:
1423 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1427 * Is the previous atime value older than a day? If yes,
1430 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1433 * Good, we can skip the atime update:
1439 * touch_atime - update the access time
1440 * @mnt: mount the inode is accessed on
1441 * @dentry: dentry accessed
1443 * Update the accessed time on an inode and mark it for writeback.
1444 * This function automatically handles read only file systems and media,
1445 * as well as the "noatime" flag and inode specific "noatime" markers.
1447 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1449 struct inode *inode = dentry->d_inode;
1450 struct timespec now;
1452 if (inode->i_flags & S_NOATIME)
1454 if (IS_NOATIME(inode))
1456 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1459 if (mnt->mnt_flags & MNT_NOATIME)
1461 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1464 now = current_fs_time(inode->i_sb);
1466 if (!relatime_need_update(mnt, inode, now))
1469 if (timespec_equal(&inode->i_atime, &now))
1472 if (mnt_want_write(mnt))
1475 inode->i_atime = now;
1476 mark_inode_dirty_sync(inode);
1477 mnt_drop_write(mnt);
1479 EXPORT_SYMBOL(touch_atime);
1482 * file_update_time - update mtime and ctime time
1483 * @file: file accessed
1485 * Update the mtime and ctime members of an inode and mark the inode
1486 * for writeback. Note that this function is meant exclusively for
1487 * usage in the file write path of filesystems, and filesystems may
1488 * choose to explicitly ignore update via this function with the
1489 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1490 * timestamps are handled by the server.
1493 void file_update_time(struct file *file)
1495 struct inode *inode = file->f_path.dentry->d_inode;
1496 struct timespec now;
1497 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1499 /* First try to exhaust all avenues to not sync */
1500 if (IS_NOCMTIME(inode))
1503 now = current_fs_time(inode->i_sb);
1504 if (!timespec_equal(&inode->i_mtime, &now))
1507 if (!timespec_equal(&inode->i_ctime, &now))
1510 if (IS_I_VERSION(inode))
1511 sync_it |= S_VERSION;
1516 /* Finally allowed to write? Takes lock. */
1517 if (mnt_want_write_file(file))
1520 /* Only change inode inside the lock region */
1521 if (sync_it & S_VERSION)
1522 inode_inc_iversion(inode);
1523 if (sync_it & S_CTIME)
1524 inode->i_ctime = now;
1525 if (sync_it & S_MTIME)
1526 inode->i_mtime = now;
1527 mark_inode_dirty_sync(inode);
1528 mnt_drop_write(file->f_path.mnt);
1530 EXPORT_SYMBOL(file_update_time);
1532 int inode_needs_sync(struct inode *inode)
1536 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1540 EXPORT_SYMBOL(inode_needs_sync);
1542 int inode_wait(void *word)
1547 EXPORT_SYMBOL(inode_wait);
1550 * If we try to find an inode in the inode hash while it is being
1551 * deleted, we have to wait until the filesystem completes its
1552 * deletion before reporting that it isn't found. This function waits
1553 * until the deletion _might_ have completed. Callers are responsible
1554 * to recheck inode state.
1556 * It doesn't matter if I_NEW is not set initially, a call to
1557 * wake_up_inode() after removing from the hash list will DTRT.
1559 * This is called with inode_lock held.
1561 static void __wait_on_freeing_inode(struct inode *inode)
1563 wait_queue_head_t *wq;
1564 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1565 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1566 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1567 spin_unlock(&inode_lock);
1569 finish_wait(wq, &wait.wait);
1570 spin_lock(&inode_lock);
1573 static __initdata unsigned long ihash_entries;
1574 static int __init set_ihash_entries(char *str)
1578 ihash_entries = simple_strtoul(str, &str, 0);
1581 __setup("ihash_entries=", set_ihash_entries);
1584 * Initialize the waitqueues and inode hash table.
1586 void __init inode_init_early(void)
1590 /* If hashes are distributed across NUMA nodes, defer
1591 * hash allocation until vmalloc space is available.
1597 alloc_large_system_hash("Inode-cache",
1598 sizeof(struct hlist_head),
1606 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1607 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1610 void __init inode_init(void)
1614 /* inode slab cache */
1615 inode_cachep = kmem_cache_create("inode_cache",
1616 sizeof(struct inode),
1618 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1621 register_shrinker(&icache_shrinker);
1622 percpu_counter_init(&nr_inodes, 0);
1623 percpu_counter_init(&nr_inodes_unused, 0);
1625 /* Hash may have been set up in inode_init_early */
1630 alloc_large_system_hash("Inode-cache",
1631 sizeof(struct hlist_head),
1639 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1640 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1643 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1645 inode->i_mode = mode;
1646 if (S_ISCHR(mode)) {
1647 inode->i_fop = &def_chr_fops;
1648 inode->i_rdev = rdev;
1649 } else if (S_ISBLK(mode)) {
1650 inode->i_fop = &def_blk_fops;
1651 inode->i_rdev = rdev;
1652 } else if (S_ISFIFO(mode))
1653 inode->i_fop = &def_fifo_fops;
1654 else if (S_ISSOCK(mode))
1655 inode->i_fop = &bad_sock_fops;
1657 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1658 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1661 EXPORT_SYMBOL(init_special_inode);
1664 * Init uid,gid,mode for new inode according to posix standards
1666 * @dir: Directory inode
1667 * @mode: mode of the new inode
1669 void inode_init_owner(struct inode *inode, const struct inode *dir,
1672 inode->i_uid = current_fsuid();
1673 if (dir && dir->i_mode & S_ISGID) {
1674 inode->i_gid = dir->i_gid;
1678 inode->i_gid = current_fsgid();
1679 inode->i_mode = mode;
1681 EXPORT_SYMBOL(inode_init_owner);