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>
27 #include <linux/ima.h>
30 * This is needed for the following functions:
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_lru);
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;
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 inodes_stat.nr_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 return proc_dointvec(table, write, buffer, lenp, ppos);
138 static void wake_up_inode(struct inode *inode)
141 * Prevent speculative execution through spin_unlock(&inode_lock);
144 wake_up_bit(&inode->i_state, __I_NEW);
148 * inode_init_always - perform inode structure intialisation
149 * @sb: superblock inode belongs to
150 * @inode: inode to initialise
152 * These are initializations that need to be done on every inode
153 * allocation as the fields are not initialised by slab allocation.
155 int inode_init_always(struct super_block *sb, struct inode *inode)
157 static const struct address_space_operations empty_aops;
158 static const struct inode_operations empty_iops;
159 static const struct file_operations empty_fops;
160 struct address_space *const mapping = &inode->i_data;
163 inode->i_blkbits = sb->s_blocksize_bits;
165 atomic_set(&inode->i_count, 1);
166 inode->i_op = &empty_iops;
167 inode->i_fop = &empty_fops;
171 atomic_set(&inode->i_writecount, 0);
175 inode->i_generation = 0;
177 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
179 inode->i_pipe = NULL;
180 inode->i_bdev = NULL;
181 inode->i_cdev = NULL;
183 inode->dirtied_when = 0;
185 if (security_inode_alloc(inode))
187 spin_lock_init(&inode->i_lock);
188 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
190 mutex_init(&inode->i_mutex);
191 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
193 init_rwsem(&inode->i_alloc_sem);
194 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
196 mapping->a_ops = &empty_aops;
197 mapping->host = inode;
199 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
200 mapping->assoc_mapping = NULL;
201 mapping->backing_dev_info = &default_backing_dev_info;
202 mapping->writeback_index = 0;
205 * If the block_device provides a backing_dev_info for client
206 * inodes then use that. Otherwise the inode share the bdev's
210 struct backing_dev_info *bdi;
212 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
213 mapping->backing_dev_info = bdi;
215 inode->i_private = NULL;
216 inode->i_mapping = mapping;
217 #ifdef CONFIG_FS_POSIX_ACL
218 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
221 #ifdef CONFIG_FSNOTIFY
222 inode->i_fsnotify_mask = 0;
225 percpu_counter_inc(&nr_inodes);
231 EXPORT_SYMBOL(inode_init_always);
233 static struct inode *alloc_inode(struct super_block *sb)
237 if (sb->s_op->alloc_inode)
238 inode = sb->s_op->alloc_inode(sb);
240 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
245 if (unlikely(inode_init_always(sb, inode))) {
246 if (inode->i_sb->s_op->destroy_inode)
247 inode->i_sb->s_op->destroy_inode(inode);
249 kmem_cache_free(inode_cachep, inode);
256 void __destroy_inode(struct inode *inode)
258 BUG_ON(inode_has_buffers(inode));
259 security_inode_free(inode);
260 fsnotify_inode_delete(inode);
261 #ifdef CONFIG_FS_POSIX_ACL
262 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
263 posix_acl_release(inode->i_acl);
264 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
265 posix_acl_release(inode->i_default_acl);
267 percpu_counter_dec(&nr_inodes);
269 EXPORT_SYMBOL(__destroy_inode);
271 static void destroy_inode(struct inode *inode)
273 BUG_ON(!list_empty(&inode->i_lru));
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_wb_list);
293 INIT_LIST_HEAD(&inode->i_lru);
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_lru)) {
335 list_add(&inode->i_lru, &inode_lru);
336 inodes_stat.nr_unused++;
340 static void inode_lru_list_del(struct inode *inode)
342 if (!list_empty(&inode->i_lru)) {
343 list_del_init(&inode->i_lru);
344 inodes_stat.nr_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_lru);
465 list_del_init(&inode->i_lru);
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 * evict_inodes - evict all evictable inodes for a superblock
481 * @sb: superblock to operate on
483 * Make sure that no inodes with zero refcount are retained. This is
484 * called by superblock shutdown after having MS_ACTIVE flag removed,
485 * so any inode reaching zero refcount during or after that call will
486 * be immediately evicted.
488 void evict_inodes(struct super_block *sb)
490 struct inode *inode, *next;
493 down_write(&iprune_sem);
495 spin_lock(&inode_lock);
496 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
497 if (atomic_read(&inode->i_count))
500 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
505 inode->i_state |= I_FREEING;
508 * Move the inode off the IO lists and LRU once I_FREEING is
509 * set so that it won't get moved back on there if it is dirty.
511 list_move(&inode->i_lru, &dispose);
512 list_del_init(&inode->i_wb_list);
513 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
514 inodes_stat.nr_unused--;
516 spin_unlock(&inode_lock);
518 dispose_list(&dispose);
519 up_write(&iprune_sem);
523 * invalidate_inodes - attempt to free all inodes on a superblock
524 * @sb: superblock to operate on
526 * Attempts to free all inodes for a given superblock. If there were any
527 * busy inodes return a non-zero value, else zero.
529 int invalidate_inodes(struct super_block *sb)
532 struct inode *inode, *next;
535 down_write(&iprune_sem);
537 spin_lock(&inode_lock);
538 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
539 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE))
541 if (atomic_read(&inode->i_count)) {
546 inode->i_state |= I_FREEING;
549 * Move the inode off the IO lists and LRU once I_FREEING is
550 * set so that it won't get moved back on there if it is dirty.
552 list_move(&inode->i_lru, &dispose);
553 list_del_init(&inode->i_wb_list);
554 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
555 inodes_stat.nr_unused--;
557 spin_unlock(&inode_lock);
559 dispose_list(&dispose);
560 up_write(&iprune_sem);
565 static int can_unuse(struct inode *inode)
567 if (inode->i_state & ~I_REFERENCED)
569 if (inode_has_buffers(inode))
571 if (atomic_read(&inode->i_count))
573 if (inode->i_data.nrpages)
579 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
580 * temporary list and then are freed outside inode_lock by dispose_list().
582 * Any inodes which are pinned purely because of attached pagecache have their
583 * pagecache removed. If the inode has metadata buffers attached to
584 * mapping->private_list then try to remove them.
586 * If the inode has the I_REFERENCED flag set, then it means that it has been
587 * used recently - the flag is set in iput_final(). When we encounter such an
588 * inode, clear the flag and move it to the back of the LRU so it gets another
589 * pass through the LRU before it gets reclaimed. This is necessary because of
590 * the fact we are doing lazy LRU updates to minimise lock contention so the
591 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
592 * with this flag set because they are the inodes that are out of order.
594 static void prune_icache(int nr_to_scan)
598 unsigned long reap = 0;
600 down_read(&iprune_sem);
601 spin_lock(&inode_lock);
602 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
605 if (list_empty(&inode_lru))
608 inode = list_entry(inode_lru.prev, struct inode, i_lru);
611 * Referenced or dirty inodes are still in use. Give them
612 * another pass through the LRU as we canot reclaim them now.
614 if (atomic_read(&inode->i_count) ||
615 (inode->i_state & ~I_REFERENCED)) {
616 list_del_init(&inode->i_lru);
617 inodes_stat.nr_unused--;
621 /* recently referenced inodes get one more pass */
622 if (inode->i_state & I_REFERENCED) {
623 list_move(&inode->i_lru, &inode_lru);
624 inode->i_state &= ~I_REFERENCED;
627 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
629 spin_unlock(&inode_lock);
630 if (remove_inode_buffers(inode))
631 reap += invalidate_mapping_pages(&inode->i_data,
634 spin_lock(&inode_lock);
636 if (inode != list_entry(inode_lru.next,
637 struct inode, i_lru))
638 continue; /* wrong inode or list_empty */
639 if (!can_unuse(inode))
642 WARN_ON(inode->i_state & I_NEW);
643 inode->i_state |= I_FREEING;
646 * Move the inode off the IO lists and LRU once I_FREEING is
647 * set so that it won't get moved back on there if it is dirty.
649 list_move(&inode->i_lru, &freeable);
650 list_del_init(&inode->i_wb_list);
651 inodes_stat.nr_unused--;
653 if (current_is_kswapd())
654 __count_vm_events(KSWAPD_INODESTEAL, reap);
656 __count_vm_events(PGINODESTEAL, reap);
657 spin_unlock(&inode_lock);
659 dispose_list(&freeable);
660 up_read(&iprune_sem);
664 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
665 * "unused" means that no dentries are referring to the inodes: the files are
666 * not open and the dcache references to those inodes have already been
669 * This function is passed the number of inodes to scan, and it returns the
670 * total number of remaining possibly-reclaimable inodes.
672 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
676 * Nasty deadlock avoidance. We may hold various FS locks,
677 * and we don't want to recurse into the FS that called us
678 * in clear_inode() and friends..
680 if (!(gfp_mask & __GFP_FS))
684 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
687 static struct shrinker icache_shrinker = {
688 .shrink = shrink_icache_memory,
689 .seeks = DEFAULT_SEEKS,
692 static void __wait_on_freeing_inode(struct inode *inode);
694 * Called with the inode lock held.
696 static struct inode *find_inode(struct super_block *sb,
697 struct hlist_head *head,
698 int (*test)(struct inode *, void *),
701 struct hlist_node *node;
702 struct inode *inode = NULL;
705 hlist_for_each_entry(inode, node, head, i_hash) {
706 if (inode->i_sb != sb)
708 if (!test(inode, data))
710 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
711 __wait_on_freeing_inode(inode);
721 * find_inode_fast is the fast path version of find_inode, see the comment at
722 * iget_locked for details.
724 static struct inode *find_inode_fast(struct super_block *sb,
725 struct hlist_head *head, unsigned long ino)
727 struct hlist_node *node;
728 struct inode *inode = NULL;
731 hlist_for_each_entry(inode, node, head, i_hash) {
732 if (inode->i_ino != ino)
734 if (inode->i_sb != sb)
736 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
737 __wait_on_freeing_inode(inode);
747 * Each cpu owns a range of LAST_INO_BATCH numbers.
748 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
749 * to renew the exhausted range.
751 * This does not significantly increase overflow rate because every CPU can
752 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
753 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
754 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
755 * overflow rate by 2x, which does not seem too significant.
757 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
758 * error if st_ino won't fit in target struct field. Use 32bit counter
759 * here to attempt to avoid that.
761 #define LAST_INO_BATCH 1024
762 static DEFINE_PER_CPU(unsigned int, last_ino);
764 unsigned int get_next_ino(void)
766 unsigned int *p = &get_cpu_var(last_ino);
767 unsigned int res = *p;
770 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
771 static atomic_t shared_last_ino;
772 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
774 res = next - LAST_INO_BATCH;
779 put_cpu_var(last_ino);
782 EXPORT_SYMBOL(get_next_ino);
785 * new_inode - obtain an inode
788 * Allocates a new inode for given superblock. The default gfp_mask
789 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
790 * If HIGHMEM pages are unsuitable or it is known that pages allocated
791 * for the page cache are not reclaimable or migratable,
792 * mapping_set_gfp_mask() must be called with suitable flags on the
793 * newly created inode's mapping
796 struct inode *new_inode(struct super_block *sb)
800 spin_lock_prefetch(&inode_lock);
802 inode = alloc_inode(sb);
804 spin_lock(&inode_lock);
805 __inode_sb_list_add(inode);
807 spin_unlock(&inode_lock);
811 EXPORT_SYMBOL(new_inode);
813 void unlock_new_inode(struct inode *inode)
815 #ifdef CONFIG_DEBUG_LOCK_ALLOC
816 if (S_ISDIR(inode->i_mode)) {
817 struct file_system_type *type = inode->i_sb->s_type;
819 /* Set new key only if filesystem hasn't already changed it */
820 if (!lockdep_match_class(&inode->i_mutex,
821 &type->i_mutex_key)) {
823 * ensure nobody is actually holding i_mutex
825 mutex_destroy(&inode->i_mutex);
826 mutex_init(&inode->i_mutex);
827 lockdep_set_class(&inode->i_mutex,
828 &type->i_mutex_dir_key);
833 * This is special! We do not need the spinlock when clearing I_NEW,
834 * because we're guaranteed that nobody else tries to do anything about
835 * the state of the inode when it is locked, as we just created it (so
836 * there can be no old holders that haven't tested I_NEW).
837 * However we must emit the memory barrier so that other CPUs reliably
838 * see the clearing of I_NEW after the other inode initialisation has
842 WARN_ON(!(inode->i_state & I_NEW));
843 inode->i_state &= ~I_NEW;
844 wake_up_inode(inode);
846 EXPORT_SYMBOL(unlock_new_inode);
849 * This is called without the inode lock held.. Be careful.
851 * We no longer cache the sb_flags in i_flags - see fs.h
852 * -- rmk@arm.uk.linux.org
854 static struct inode *get_new_inode(struct super_block *sb,
855 struct hlist_head *head,
856 int (*test)(struct inode *, void *),
857 int (*set)(struct inode *, void *),
862 inode = alloc_inode(sb);
866 spin_lock(&inode_lock);
867 /* We released the lock, so.. */
868 old = find_inode(sb, head, test, data);
870 if (set(inode, data))
873 hlist_add_head(&inode->i_hash, head);
874 __inode_sb_list_add(inode);
875 inode->i_state = I_NEW;
876 spin_unlock(&inode_lock);
878 /* Return the locked inode with I_NEW set, the
879 * caller is responsible for filling in the contents
885 * Uhhuh, somebody else created the same inode under
886 * us. Use the old inode instead of the one we just
889 spin_unlock(&inode_lock);
890 destroy_inode(inode);
892 wait_on_inode(inode);
897 spin_unlock(&inode_lock);
898 destroy_inode(inode);
903 * get_new_inode_fast is the fast path version of get_new_inode, see the
904 * comment at iget_locked for details.
906 static struct inode *get_new_inode_fast(struct super_block *sb,
907 struct hlist_head *head, unsigned long ino)
911 inode = alloc_inode(sb);
915 spin_lock(&inode_lock);
916 /* We released the lock, so.. */
917 old = find_inode_fast(sb, head, ino);
920 hlist_add_head(&inode->i_hash, head);
921 __inode_sb_list_add(inode);
922 inode->i_state = I_NEW;
923 spin_unlock(&inode_lock);
925 /* Return the locked inode with I_NEW set, the
926 * caller is responsible for filling in the contents
932 * Uhhuh, somebody else created the same inode under
933 * us. Use the old inode instead of the one we just
936 spin_unlock(&inode_lock);
937 destroy_inode(inode);
939 wait_on_inode(inode);
945 * search the inode cache for a matching inode number.
946 * If we find one, then the inode number we are trying to
947 * allocate is not unique and so we should not use it.
949 * Returns 1 if the inode number is unique, 0 if it is not.
951 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
953 struct hlist_head *b = inode_hashtable + hash(sb, ino);
954 struct hlist_node *node;
957 hlist_for_each_entry(inode, node, b, i_hash) {
958 if (inode->i_ino == ino && inode->i_sb == sb)
966 * iunique - get a unique inode number
968 * @max_reserved: highest reserved inode number
970 * Obtain an inode number that is unique on the system for a given
971 * superblock. This is used by file systems that have no natural
972 * permanent inode numbering system. An inode number is returned that
973 * is higher than the reserved limit but unique.
976 * With a large number of inodes live on the file system this function
977 * currently becomes quite slow.
979 ino_t iunique(struct super_block *sb, ino_t max_reserved)
982 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
983 * error if st_ino won't fit in target struct field. Use 32bit counter
984 * here to attempt to avoid that.
986 static DEFINE_SPINLOCK(iunique_lock);
987 static unsigned int counter;
990 spin_lock(&inode_lock);
991 spin_lock(&iunique_lock);
993 if (counter <= max_reserved)
994 counter = max_reserved + 1;
996 } while (!test_inode_iunique(sb, res));
997 spin_unlock(&iunique_lock);
998 spin_unlock(&inode_lock);
1002 EXPORT_SYMBOL(iunique);
1004 struct inode *igrab(struct inode *inode)
1006 spin_lock(&inode_lock);
1007 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
1011 * Handle the case where s_op->clear_inode is not been
1012 * called yet, and somebody is calling igrab
1013 * while the inode is getting freed.
1016 spin_unlock(&inode_lock);
1019 EXPORT_SYMBOL(igrab);
1022 * ifind - internal function, you want ilookup5() or iget5().
1023 * @sb: super block of file system to search
1024 * @head: the head of the list to search
1025 * @test: callback used for comparisons between inodes
1026 * @data: opaque data pointer to pass to @test
1027 * @wait: if true wait for the inode to be unlocked, if false do not
1029 * ifind() searches for the inode specified by @data in the inode
1030 * cache. This is a generalized version of ifind_fast() for file systems where
1031 * the inode number is not sufficient for unique identification of an inode.
1033 * If the inode is in the cache, the inode is returned with an incremented
1036 * Otherwise NULL is returned.
1038 * Note, @test is called with the inode_lock held, so can't sleep.
1040 static struct inode *ifind(struct super_block *sb,
1041 struct hlist_head *head, int (*test)(struct inode *, void *),
1042 void *data, const int wait)
1044 struct inode *inode;
1046 spin_lock(&inode_lock);
1047 inode = find_inode(sb, head, test, data);
1049 spin_unlock(&inode_lock);
1051 wait_on_inode(inode);
1054 spin_unlock(&inode_lock);
1059 * ifind_fast - internal function, you want ilookup() or iget().
1060 * @sb: super block of file system to search
1061 * @head: head of the list to search
1062 * @ino: inode number to search for
1064 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1065 * file systems where the inode number is sufficient for unique identification
1068 * If the inode is in the cache, the inode is returned with an incremented
1071 * Otherwise NULL is returned.
1073 static struct inode *ifind_fast(struct super_block *sb,
1074 struct hlist_head *head, unsigned long ino)
1076 struct inode *inode;
1078 spin_lock(&inode_lock);
1079 inode = find_inode_fast(sb, head, ino);
1081 spin_unlock(&inode_lock);
1082 wait_on_inode(inode);
1085 spin_unlock(&inode_lock);
1090 * ilookup5_nowait - search for an inode in the inode cache
1091 * @sb: super block of file system to search
1092 * @hashval: hash value (usually inode number) to search for
1093 * @test: callback used for comparisons between inodes
1094 * @data: opaque data pointer to pass to @test
1096 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1097 * @data in the inode cache. This is a generalized version of ilookup() for
1098 * file systems where the inode number is not sufficient for unique
1099 * identification of an inode.
1101 * If the inode is in the cache, the inode is returned with an incremented
1102 * reference count. Note, the inode lock is not waited upon so you have to be
1103 * very careful what you do with the returned inode. You probably should be
1104 * using ilookup5() instead.
1106 * Otherwise NULL is returned.
1108 * Note, @test is called with the inode_lock held, so can't sleep.
1110 struct inode *ilookup5_nowait(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, 0);
1117 EXPORT_SYMBOL(ilookup5_nowait);
1120 * ilookup5 - search for an inode in the inode cache
1121 * @sb: super block of file system to search
1122 * @hashval: hash value (usually inode number) to search for
1123 * @test: callback used for comparisons between inodes
1124 * @data: opaque data pointer to pass to @test
1126 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1127 * @data in the inode cache. This is a generalized version of ilookup() for
1128 * file systems where the inode number is not sufficient for unique
1129 * identification of an inode.
1131 * If the inode is in the cache, the inode lock is waited upon and the inode is
1132 * returned with an incremented reference count.
1134 * Otherwise NULL is returned.
1136 * Note, @test is called with the inode_lock held, so can't sleep.
1138 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1139 int (*test)(struct inode *, void *), void *data)
1141 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1143 return ifind(sb, head, test, data, 1);
1145 EXPORT_SYMBOL(ilookup5);
1148 * ilookup - search for an inode in the inode cache
1149 * @sb: super block of file system to search
1150 * @ino: inode number to search for
1152 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1153 * This is for file systems where the inode number is sufficient for unique
1154 * identification of an inode.
1156 * If the inode is in the cache, the inode is returned with an incremented
1159 * Otherwise NULL is returned.
1161 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1163 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1165 return ifind_fast(sb, head, ino);
1167 EXPORT_SYMBOL(ilookup);
1170 * iget5_locked - obtain an inode from a mounted file system
1171 * @sb: super block of file system
1172 * @hashval: hash value (usually inode number) to get
1173 * @test: callback used for comparisons between inodes
1174 * @set: callback used to initialize a new struct inode
1175 * @data: opaque data pointer to pass to @test and @set
1177 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1178 * and @data in the inode cache and if present it is returned with an increased
1179 * reference count. This is a generalized version of iget_locked() for file
1180 * systems where the inode number is not sufficient for unique identification
1183 * If the inode is not in cache, get_new_inode() is called to allocate a new
1184 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1185 * file system gets to fill it in before unlocking it via unlock_new_inode().
1187 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1189 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1190 int (*test)(struct inode *, void *),
1191 int (*set)(struct inode *, void *), void *data)
1193 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1194 struct inode *inode;
1196 inode = ifind(sb, head, test, data, 1);
1200 * get_new_inode() will do the right thing, re-trying the search
1201 * in case it had to block at any point.
1203 return get_new_inode(sb, head, test, set, data);
1205 EXPORT_SYMBOL(iget5_locked);
1208 * iget_locked - obtain an inode from a mounted file system
1209 * @sb: super block of file system
1210 * @ino: inode number to get
1212 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1213 * the inode cache and if present it is returned with an increased reference
1214 * count. This is for file systems where the inode number is sufficient for
1215 * unique identification of an inode.
1217 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1218 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1219 * The file system gets to fill it in before unlocking it via
1220 * unlock_new_inode().
1222 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1224 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1225 struct inode *inode;
1227 inode = ifind_fast(sb, head, ino);
1231 * get_new_inode_fast() will do the right thing, re-trying the search
1232 * in case it had to block at any point.
1234 return get_new_inode_fast(sb, head, ino);
1236 EXPORT_SYMBOL(iget_locked);
1238 int insert_inode_locked(struct inode *inode)
1240 struct super_block *sb = inode->i_sb;
1241 ino_t ino = inode->i_ino;
1242 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1244 inode->i_state |= I_NEW;
1246 struct hlist_node *node;
1247 struct inode *old = NULL;
1248 spin_lock(&inode_lock);
1249 hlist_for_each_entry(old, node, head, i_hash) {
1250 if (old->i_ino != ino)
1252 if (old->i_sb != sb)
1254 if (old->i_state & (I_FREEING|I_WILL_FREE))
1258 if (likely(!node)) {
1259 hlist_add_head(&inode->i_hash, head);
1260 spin_unlock(&inode_lock);
1264 spin_unlock(&inode_lock);
1266 if (unlikely(!inode_unhashed(old))) {
1273 EXPORT_SYMBOL(insert_inode_locked);
1275 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1276 int (*test)(struct inode *, void *), void *data)
1278 struct super_block *sb = inode->i_sb;
1279 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1281 inode->i_state |= I_NEW;
1284 struct hlist_node *node;
1285 struct inode *old = NULL;
1287 spin_lock(&inode_lock);
1288 hlist_for_each_entry(old, node, head, i_hash) {
1289 if (old->i_sb != sb)
1291 if (!test(old, data))
1293 if (old->i_state & (I_FREEING|I_WILL_FREE))
1297 if (likely(!node)) {
1298 hlist_add_head(&inode->i_hash, head);
1299 spin_unlock(&inode_lock);
1303 spin_unlock(&inode_lock);
1305 if (unlikely(!inode_unhashed(old))) {
1312 EXPORT_SYMBOL(insert_inode_locked4);
1315 int generic_delete_inode(struct inode *inode)
1319 EXPORT_SYMBOL(generic_delete_inode);
1322 * Normal UNIX filesystem behaviour: delete the
1323 * inode when the usage count drops to zero, and
1326 int generic_drop_inode(struct inode *inode)
1328 return !inode->i_nlink || inode_unhashed(inode);
1330 EXPORT_SYMBOL_GPL(generic_drop_inode);
1333 * Called when we're dropping the last reference
1336 * Call the FS "drop_inode()" function, defaulting to
1337 * the legacy UNIX filesystem behaviour. If it tells
1338 * us to evict inode, do so. Otherwise, retain inode
1339 * in cache if fs is alive, sync and evict if fs is
1342 static void iput_final(struct inode *inode)
1344 struct super_block *sb = inode->i_sb;
1345 const struct super_operations *op = inode->i_sb->s_op;
1348 if (op && op->drop_inode)
1349 drop = op->drop_inode(inode);
1351 drop = generic_drop_inode(inode);
1354 if (sb->s_flags & MS_ACTIVE) {
1355 inode->i_state |= I_REFERENCED;
1356 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1357 inode_lru_list_add(inode);
1359 spin_unlock(&inode_lock);
1362 WARN_ON(inode->i_state & I_NEW);
1363 inode->i_state |= I_WILL_FREE;
1364 spin_unlock(&inode_lock);
1365 write_inode_now(inode, 1);
1366 spin_lock(&inode_lock);
1367 WARN_ON(inode->i_state & I_NEW);
1368 inode->i_state &= ~I_WILL_FREE;
1369 __remove_inode_hash(inode);
1372 WARN_ON(inode->i_state & I_NEW);
1373 inode->i_state |= I_FREEING;
1376 * Move the inode off the IO lists and LRU once I_FREEING is
1377 * set so that it won't get moved back on there if it is dirty.
1379 inode_lru_list_del(inode);
1380 list_del_init(&inode->i_wb_list);
1382 __inode_sb_list_del(inode);
1383 spin_unlock(&inode_lock);
1385 remove_inode_hash(inode);
1386 wake_up_inode(inode);
1387 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1388 destroy_inode(inode);
1392 * iput - put an inode
1393 * @inode: inode to put
1395 * Puts an inode, dropping its usage count. If the inode use count hits
1396 * zero, the inode is then freed and may also be destroyed.
1398 * Consequently, iput() can sleep.
1400 void iput(struct inode *inode)
1403 BUG_ON(inode->i_state & I_CLEAR);
1405 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1409 EXPORT_SYMBOL(iput);
1412 * bmap - find a block number in a file
1413 * @inode: inode of file
1414 * @block: block to find
1416 * Returns the block number on the device holding the inode that
1417 * is the disk block number for the block of the file requested.
1418 * That is, asked for block 4 of inode 1 the function will return the
1419 * disk block relative to the disk start that holds that block of the
1422 sector_t bmap(struct inode *inode, sector_t block)
1425 if (inode->i_mapping->a_ops->bmap)
1426 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1429 EXPORT_SYMBOL(bmap);
1432 * With relative atime, only update atime if the previous atime is
1433 * earlier than either the ctime or mtime or if at least a day has
1434 * passed since the last atime update.
1436 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1437 struct timespec now)
1440 if (!(mnt->mnt_flags & MNT_RELATIME))
1443 * Is mtime younger than atime? If yes, update atime:
1445 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1448 * Is ctime younger than atime? If yes, update atime:
1450 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1454 * Is the previous atime value older than a day? If yes,
1457 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1460 * Good, we can skip the atime update:
1466 * touch_atime - update the access time
1467 * @mnt: mount the inode is accessed on
1468 * @dentry: dentry accessed
1470 * Update the accessed time on an inode and mark it for writeback.
1471 * This function automatically handles read only file systems and media,
1472 * as well as the "noatime" flag and inode specific "noatime" markers.
1474 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1476 struct inode *inode = dentry->d_inode;
1477 struct timespec now;
1479 if (inode->i_flags & S_NOATIME)
1481 if (IS_NOATIME(inode))
1483 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1486 if (mnt->mnt_flags & MNT_NOATIME)
1488 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1491 now = current_fs_time(inode->i_sb);
1493 if (!relatime_need_update(mnt, inode, now))
1496 if (timespec_equal(&inode->i_atime, &now))
1499 if (mnt_want_write(mnt))
1502 inode->i_atime = now;
1503 mark_inode_dirty_sync(inode);
1504 mnt_drop_write(mnt);
1506 EXPORT_SYMBOL(touch_atime);
1509 * file_update_time - update mtime and ctime time
1510 * @file: file accessed
1512 * Update the mtime and ctime members of an inode and mark the inode
1513 * for writeback. Note that this function is meant exclusively for
1514 * usage in the file write path of filesystems, and filesystems may
1515 * choose to explicitly ignore update via this function with the
1516 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1517 * timestamps are handled by the server.
1520 void file_update_time(struct file *file)
1522 struct inode *inode = file->f_path.dentry->d_inode;
1523 struct timespec now;
1524 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1526 /* First try to exhaust all avenues to not sync */
1527 if (IS_NOCMTIME(inode))
1530 now = current_fs_time(inode->i_sb);
1531 if (!timespec_equal(&inode->i_mtime, &now))
1534 if (!timespec_equal(&inode->i_ctime, &now))
1537 if (IS_I_VERSION(inode))
1538 sync_it |= S_VERSION;
1543 /* Finally allowed to write? Takes lock. */
1544 if (mnt_want_write_file(file))
1547 /* Only change inode inside the lock region */
1548 if (sync_it & S_VERSION)
1549 inode_inc_iversion(inode);
1550 if (sync_it & S_CTIME)
1551 inode->i_ctime = now;
1552 if (sync_it & S_MTIME)
1553 inode->i_mtime = now;
1554 mark_inode_dirty_sync(inode);
1555 mnt_drop_write(file->f_path.mnt);
1557 EXPORT_SYMBOL(file_update_time);
1559 int inode_needs_sync(struct inode *inode)
1563 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1567 EXPORT_SYMBOL(inode_needs_sync);
1569 int inode_wait(void *word)
1574 EXPORT_SYMBOL(inode_wait);
1577 * If we try to find an inode in the inode hash while it is being
1578 * deleted, we have to wait until the filesystem completes its
1579 * deletion before reporting that it isn't found. This function waits
1580 * until the deletion _might_ have completed. Callers are responsible
1581 * to recheck inode state.
1583 * It doesn't matter if I_NEW is not set initially, a call to
1584 * wake_up_inode() after removing from the hash list will DTRT.
1586 * This is called with inode_lock held.
1588 static void __wait_on_freeing_inode(struct inode *inode)
1590 wait_queue_head_t *wq;
1591 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1592 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1593 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1594 spin_unlock(&inode_lock);
1596 finish_wait(wq, &wait.wait);
1597 spin_lock(&inode_lock);
1600 static __initdata unsigned long ihash_entries;
1601 static int __init set_ihash_entries(char *str)
1605 ihash_entries = simple_strtoul(str, &str, 0);
1608 __setup("ihash_entries=", set_ihash_entries);
1611 * Initialize the waitqueues and inode hash table.
1613 void __init inode_init_early(void)
1617 /* If hashes are distributed across NUMA nodes, defer
1618 * hash allocation until vmalloc space is available.
1624 alloc_large_system_hash("Inode-cache",
1625 sizeof(struct hlist_head),
1633 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1634 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1637 void __init inode_init(void)
1641 /* inode slab cache */
1642 inode_cachep = kmem_cache_create("inode_cache",
1643 sizeof(struct inode),
1645 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1648 register_shrinker(&icache_shrinker);
1649 percpu_counter_init(&nr_inodes, 0);
1651 /* Hash may have been set up in inode_init_early */
1656 alloc_large_system_hash("Inode-cache",
1657 sizeof(struct hlist_head),
1665 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1666 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1669 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1671 inode->i_mode = mode;
1672 if (S_ISCHR(mode)) {
1673 inode->i_fop = &def_chr_fops;
1674 inode->i_rdev = rdev;
1675 } else if (S_ISBLK(mode)) {
1676 inode->i_fop = &def_blk_fops;
1677 inode->i_rdev = rdev;
1678 } else if (S_ISFIFO(mode))
1679 inode->i_fop = &def_fifo_fops;
1680 else if (S_ISSOCK(mode))
1681 inode->i_fop = &bad_sock_fops;
1683 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1684 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1687 EXPORT_SYMBOL(init_special_inode);
1690 * Init uid,gid,mode for new inode according to posix standards
1692 * @dir: Directory inode
1693 * @mode: mode of the new inode
1695 void inode_init_owner(struct inode *inode, const struct inode *dir,
1698 inode->i_uid = current_fsuid();
1699 if (dir && dir->i_mode & S_ISGID) {
1700 inode->i_gid = dir->i_gid;
1704 inode->i_gid = current_fsgid();
1705 inode->i_mode = mode;
1707 EXPORT_SYMBOL(inode_init_owner);