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 DEFINE_PER_CPU(unsigned int, nr_inodes);
107 static struct kmem_cache *inode_cachep __read_mostly;
109 static int get_nr_inodes(void)
113 for_each_possible_cpu(i)
114 sum += per_cpu(nr_inodes, i);
115 return sum < 0 ? 0 : sum;
118 static inline int get_nr_inodes_unused(void)
120 return inodes_stat.nr_unused;
123 int get_nr_dirty_inodes(void)
125 /* not actually dirty inodes, but a wild approximation */
126 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
127 return nr_dirty > 0 ? nr_dirty : 0;
131 * Handle nr_inode sysctl
134 int proc_nr_inodes(ctl_table *table, int write,
135 void __user *buffer, size_t *lenp, loff_t *ppos)
137 inodes_stat.nr_inodes = get_nr_inodes();
138 return proc_dointvec(table, write, buffer, lenp, ppos);
142 static void wake_up_inode(struct inode *inode)
145 * Prevent speculative execution through spin_unlock(&inode_lock);
148 wake_up_bit(&inode->i_state, __I_NEW);
152 * inode_init_always - perform inode structure intialisation
153 * @sb: superblock inode belongs to
154 * @inode: inode to initialise
156 * These are initializations that need to be done on every inode
157 * allocation as the fields are not initialised by slab allocation.
159 int inode_init_always(struct super_block *sb, struct inode *inode)
161 static const struct address_space_operations empty_aops;
162 static const struct inode_operations empty_iops;
163 static const struct file_operations empty_fops;
164 struct address_space *const mapping = &inode->i_data;
167 inode->i_blkbits = sb->s_blocksize_bits;
169 atomic_set(&inode->i_count, 1);
170 inode->i_op = &empty_iops;
171 inode->i_fop = &empty_fops;
175 atomic_set(&inode->i_writecount, 0);
179 inode->i_generation = 0;
181 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
183 inode->i_pipe = NULL;
184 inode->i_bdev = NULL;
185 inode->i_cdev = NULL;
187 inode->dirtied_when = 0;
189 if (security_inode_alloc(inode))
191 spin_lock_init(&inode->i_lock);
192 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
194 mutex_init(&inode->i_mutex);
195 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
197 init_rwsem(&inode->i_alloc_sem);
198 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
200 mapping->a_ops = &empty_aops;
201 mapping->host = inode;
203 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
204 mapping->assoc_mapping = NULL;
205 mapping->backing_dev_info = &default_backing_dev_info;
206 mapping->writeback_index = 0;
209 * If the block_device provides a backing_dev_info for client
210 * inodes then use that. Otherwise the inode share the bdev's
214 struct backing_dev_info *bdi;
216 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
217 mapping->backing_dev_info = bdi;
219 inode->i_private = NULL;
220 inode->i_mapping = mapping;
221 #ifdef CONFIG_FS_POSIX_ACL
222 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
225 #ifdef CONFIG_FSNOTIFY
226 inode->i_fsnotify_mask = 0;
229 this_cpu_inc(nr_inodes);
235 EXPORT_SYMBOL(inode_init_always);
237 static struct inode *alloc_inode(struct super_block *sb)
241 if (sb->s_op->alloc_inode)
242 inode = sb->s_op->alloc_inode(sb);
244 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
249 if (unlikely(inode_init_always(sb, inode))) {
250 if (inode->i_sb->s_op->destroy_inode)
251 inode->i_sb->s_op->destroy_inode(inode);
253 kmem_cache_free(inode_cachep, inode);
260 void free_inode_nonrcu(struct inode *inode)
262 kmem_cache_free(inode_cachep, inode);
264 EXPORT_SYMBOL(free_inode_nonrcu);
266 void __destroy_inode(struct inode *inode)
268 BUG_ON(inode_has_buffers(inode));
269 security_inode_free(inode);
270 fsnotify_inode_delete(inode);
271 #ifdef CONFIG_FS_POSIX_ACL
272 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
273 posix_acl_release(inode->i_acl);
274 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
275 posix_acl_release(inode->i_default_acl);
277 this_cpu_dec(nr_inodes);
279 EXPORT_SYMBOL(__destroy_inode);
281 static void i_callback(struct rcu_head *head)
283 struct inode *inode = container_of(head, struct inode, i_rcu);
284 INIT_LIST_HEAD(&inode->i_dentry);
285 kmem_cache_free(inode_cachep, inode);
288 static void destroy_inode(struct inode *inode)
290 BUG_ON(!list_empty(&inode->i_lru));
291 __destroy_inode(inode);
292 if (inode->i_sb->s_op->destroy_inode)
293 inode->i_sb->s_op->destroy_inode(inode);
295 call_rcu(&inode->i_rcu, i_callback);
299 * These are initializations that only need to be done
300 * once, because the fields are idempotent across use
301 * of the inode, so let the slab aware of that.
303 void inode_init_once(struct inode *inode)
305 memset(inode, 0, sizeof(*inode));
306 INIT_HLIST_NODE(&inode->i_hash);
307 INIT_LIST_HEAD(&inode->i_dentry);
308 INIT_LIST_HEAD(&inode->i_devices);
309 INIT_LIST_HEAD(&inode->i_wb_list);
310 INIT_LIST_HEAD(&inode->i_lru);
311 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
312 spin_lock_init(&inode->i_data.tree_lock);
313 spin_lock_init(&inode->i_data.i_mmap_lock);
314 INIT_LIST_HEAD(&inode->i_data.private_list);
315 spin_lock_init(&inode->i_data.private_lock);
316 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
317 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
318 i_size_ordered_init(inode);
319 #ifdef CONFIG_FSNOTIFY
320 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
323 EXPORT_SYMBOL(inode_init_once);
325 static void init_once(void *foo)
327 struct inode *inode = (struct inode *) foo;
329 inode_init_once(inode);
333 * inode_lock must be held
335 void __iget(struct inode *inode)
337 atomic_inc(&inode->i_count);
341 * get additional reference to inode; caller must already hold one.
343 void ihold(struct inode *inode)
345 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
347 EXPORT_SYMBOL(ihold);
349 static void inode_lru_list_add(struct inode *inode)
351 if (list_empty(&inode->i_lru)) {
352 list_add(&inode->i_lru, &inode_lru);
353 inodes_stat.nr_unused++;
357 static void inode_lru_list_del(struct inode *inode)
359 if (!list_empty(&inode->i_lru)) {
360 list_del_init(&inode->i_lru);
361 inodes_stat.nr_unused--;
365 static inline void __inode_sb_list_add(struct inode *inode)
367 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
371 * inode_sb_list_add - add inode to the superblock list of inodes
372 * @inode: inode to add
374 void inode_sb_list_add(struct inode *inode)
376 spin_lock(&inode_lock);
377 __inode_sb_list_add(inode);
378 spin_unlock(&inode_lock);
380 EXPORT_SYMBOL_GPL(inode_sb_list_add);
382 static inline void __inode_sb_list_del(struct inode *inode)
384 list_del_init(&inode->i_sb_list);
387 static unsigned long hash(struct super_block *sb, unsigned long hashval)
391 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
393 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
394 return tmp & I_HASHMASK;
398 * __insert_inode_hash - hash an inode
399 * @inode: unhashed inode
400 * @hashval: unsigned long value used to locate this object in the
403 * Add an inode to the inode hash for this superblock.
405 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
407 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
409 spin_lock(&inode_lock);
410 hlist_add_head(&inode->i_hash, b);
411 spin_unlock(&inode_lock);
413 EXPORT_SYMBOL(__insert_inode_hash);
416 * __remove_inode_hash - remove an inode from the hash
417 * @inode: inode to unhash
419 * Remove an inode from the superblock.
421 static void __remove_inode_hash(struct inode *inode)
423 hlist_del_init(&inode->i_hash);
427 * remove_inode_hash - remove an inode from the hash
428 * @inode: inode to unhash
430 * Remove an inode from the superblock.
432 void remove_inode_hash(struct inode *inode)
434 spin_lock(&inode_lock);
435 hlist_del_init(&inode->i_hash);
436 spin_unlock(&inode_lock);
438 EXPORT_SYMBOL(remove_inode_hash);
440 void end_writeback(struct inode *inode)
443 BUG_ON(inode->i_data.nrpages);
444 BUG_ON(!list_empty(&inode->i_data.private_list));
445 BUG_ON(!(inode->i_state & I_FREEING));
446 BUG_ON(inode->i_state & I_CLEAR);
447 inode_sync_wait(inode);
448 /* don't need i_lock here, no concurrent mods to i_state */
449 inode->i_state = I_FREEING | I_CLEAR;
451 EXPORT_SYMBOL(end_writeback);
453 static void evict(struct inode *inode)
455 const struct super_operations *op = inode->i_sb->s_op;
457 if (op->evict_inode) {
458 op->evict_inode(inode);
460 if (inode->i_data.nrpages)
461 truncate_inode_pages(&inode->i_data, 0);
462 end_writeback(inode);
464 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
466 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
471 * dispose_list - dispose of the contents of a local list
472 * @head: the head of the list to free
474 * Dispose-list gets a local list with local inodes in it, so it doesn't
475 * need to worry about list corruption and SMP locks.
477 static void dispose_list(struct list_head *head)
479 while (!list_empty(head)) {
482 inode = list_first_entry(head, struct inode, i_lru);
483 list_del_init(&inode->i_lru);
487 spin_lock(&inode_lock);
488 __remove_inode_hash(inode);
489 __inode_sb_list_del(inode);
490 spin_unlock(&inode_lock);
492 wake_up_inode(inode);
493 destroy_inode(inode);
498 * evict_inodes - evict all evictable inodes for a superblock
499 * @sb: superblock to operate on
501 * Make sure that no inodes with zero refcount are retained. This is
502 * called by superblock shutdown after having MS_ACTIVE flag removed,
503 * so any inode reaching zero refcount during or after that call will
504 * be immediately evicted.
506 void evict_inodes(struct super_block *sb)
508 struct inode *inode, *next;
511 down_write(&iprune_sem);
513 spin_lock(&inode_lock);
514 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
515 if (atomic_read(&inode->i_count))
518 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
523 inode->i_state |= I_FREEING;
526 * Move the inode off the IO lists and LRU once I_FREEING is
527 * set so that it won't get moved back on there if it is dirty.
529 list_move(&inode->i_lru, &dispose);
530 list_del_init(&inode->i_wb_list);
531 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
532 inodes_stat.nr_unused--;
534 spin_unlock(&inode_lock);
536 dispose_list(&dispose);
537 up_write(&iprune_sem);
541 * invalidate_inodes - attempt to free all inodes on a superblock
542 * @sb: superblock to operate on
543 * @kill_dirty: flag to guide handling of dirty inodes
545 * Attempts to free all inodes for a given superblock. If there were any
546 * busy inodes return a non-zero value, else zero.
547 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
550 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
553 struct inode *inode, *next;
556 down_write(&iprune_sem);
558 spin_lock(&inode_lock);
559 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
560 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE))
562 if (inode->i_state & I_DIRTY && !kill_dirty) {
566 if (atomic_read(&inode->i_count)) {
571 inode->i_state |= I_FREEING;
574 * Move the inode off the IO lists and LRU once I_FREEING is
575 * set so that it won't get moved back on there if it is dirty.
577 list_move(&inode->i_lru, &dispose);
578 list_del_init(&inode->i_wb_list);
579 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
580 inodes_stat.nr_unused--;
582 spin_unlock(&inode_lock);
584 dispose_list(&dispose);
585 up_write(&iprune_sem);
590 static int can_unuse(struct inode *inode)
592 if (inode->i_state & ~I_REFERENCED)
594 if (inode_has_buffers(inode))
596 if (atomic_read(&inode->i_count))
598 if (inode->i_data.nrpages)
604 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
605 * temporary list and then are freed outside inode_lock by dispose_list().
607 * Any inodes which are pinned purely because of attached pagecache have their
608 * pagecache removed. If the inode has metadata buffers attached to
609 * mapping->private_list then try to remove them.
611 * If the inode has the I_REFERENCED flag set, then it means that it has been
612 * used recently - the flag is set in iput_final(). When we encounter such an
613 * inode, clear the flag and move it to the back of the LRU so it gets another
614 * pass through the LRU before it gets reclaimed. This is necessary because of
615 * the fact we are doing lazy LRU updates to minimise lock contention so the
616 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
617 * with this flag set because they are the inodes that are out of order.
619 static void prune_icache(int nr_to_scan)
623 unsigned long reap = 0;
625 down_read(&iprune_sem);
626 spin_lock(&inode_lock);
627 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
630 if (list_empty(&inode_lru))
633 inode = list_entry(inode_lru.prev, struct inode, i_lru);
636 * Referenced or dirty inodes are still in use. Give them
637 * another pass through the LRU as we canot reclaim them now.
639 if (atomic_read(&inode->i_count) ||
640 (inode->i_state & ~I_REFERENCED)) {
641 list_del_init(&inode->i_lru);
642 inodes_stat.nr_unused--;
646 /* recently referenced inodes get one more pass */
647 if (inode->i_state & I_REFERENCED) {
648 list_move(&inode->i_lru, &inode_lru);
649 inode->i_state &= ~I_REFERENCED;
652 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
654 spin_unlock(&inode_lock);
655 if (remove_inode_buffers(inode))
656 reap += invalidate_mapping_pages(&inode->i_data,
659 spin_lock(&inode_lock);
661 if (inode != list_entry(inode_lru.next,
662 struct inode, i_lru))
663 continue; /* wrong inode or list_empty */
664 if (!can_unuse(inode))
667 WARN_ON(inode->i_state & I_NEW);
668 inode->i_state |= I_FREEING;
671 * Move the inode off the IO lists and LRU once I_FREEING is
672 * set so that it won't get moved back on there if it is dirty.
674 list_move(&inode->i_lru, &freeable);
675 list_del_init(&inode->i_wb_list);
676 inodes_stat.nr_unused--;
678 if (current_is_kswapd())
679 __count_vm_events(KSWAPD_INODESTEAL, reap);
681 __count_vm_events(PGINODESTEAL, reap);
682 spin_unlock(&inode_lock);
684 dispose_list(&freeable);
685 up_read(&iprune_sem);
689 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
690 * "unused" means that no dentries are referring to the inodes: the files are
691 * not open and the dcache references to those inodes have already been
694 * This function is passed the number of inodes to scan, and it returns the
695 * total number of remaining possibly-reclaimable inodes.
697 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
701 * Nasty deadlock avoidance. We may hold various FS locks,
702 * and we don't want to recurse into the FS that called us
703 * in clear_inode() and friends..
705 if (!(gfp_mask & __GFP_FS))
709 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
712 static struct shrinker icache_shrinker = {
713 .shrink = shrink_icache_memory,
714 .seeks = DEFAULT_SEEKS,
717 static void __wait_on_freeing_inode(struct inode *inode);
719 * Called with the inode lock held.
721 static struct inode *find_inode(struct super_block *sb,
722 struct hlist_head *head,
723 int (*test)(struct inode *, void *),
726 struct hlist_node *node;
727 struct inode *inode = NULL;
730 hlist_for_each_entry(inode, node, head, i_hash) {
731 if (inode->i_sb != sb)
733 if (!test(inode, data))
735 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
736 __wait_on_freeing_inode(inode);
746 * find_inode_fast is the fast path version of find_inode, see the comment at
747 * iget_locked for details.
749 static struct inode *find_inode_fast(struct super_block *sb,
750 struct hlist_head *head, unsigned long ino)
752 struct hlist_node *node;
753 struct inode *inode = NULL;
756 hlist_for_each_entry(inode, node, head, i_hash) {
757 if (inode->i_ino != ino)
759 if (inode->i_sb != sb)
761 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
762 __wait_on_freeing_inode(inode);
772 * Each cpu owns a range of LAST_INO_BATCH numbers.
773 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
774 * to renew the exhausted range.
776 * This does not significantly increase overflow rate because every CPU can
777 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
778 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
779 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
780 * overflow rate by 2x, which does not seem too significant.
782 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
783 * error if st_ino won't fit in target struct field. Use 32bit counter
784 * here to attempt to avoid that.
786 #define LAST_INO_BATCH 1024
787 static DEFINE_PER_CPU(unsigned int, last_ino);
789 unsigned int get_next_ino(void)
791 unsigned int *p = &get_cpu_var(last_ino);
792 unsigned int res = *p;
795 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
796 static atomic_t shared_last_ino;
797 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
799 res = next - LAST_INO_BATCH;
804 put_cpu_var(last_ino);
807 EXPORT_SYMBOL(get_next_ino);
810 * new_inode - obtain an inode
813 * Allocates a new inode for given superblock. The default gfp_mask
814 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
815 * If HIGHMEM pages are unsuitable or it is known that pages allocated
816 * for the page cache are not reclaimable or migratable,
817 * mapping_set_gfp_mask() must be called with suitable flags on the
818 * newly created inode's mapping
821 struct inode *new_inode(struct super_block *sb)
825 spin_lock_prefetch(&inode_lock);
827 inode = alloc_inode(sb);
829 spin_lock(&inode_lock);
830 __inode_sb_list_add(inode);
832 spin_unlock(&inode_lock);
836 EXPORT_SYMBOL(new_inode);
838 void unlock_new_inode(struct inode *inode)
840 #ifdef CONFIG_DEBUG_LOCK_ALLOC
841 if (S_ISDIR(inode->i_mode)) {
842 struct file_system_type *type = inode->i_sb->s_type;
844 /* Set new key only if filesystem hasn't already changed it */
845 if (!lockdep_match_class(&inode->i_mutex,
846 &type->i_mutex_key)) {
848 * ensure nobody is actually holding i_mutex
850 mutex_destroy(&inode->i_mutex);
851 mutex_init(&inode->i_mutex);
852 lockdep_set_class(&inode->i_mutex,
853 &type->i_mutex_dir_key);
858 * This is special! We do not need the spinlock when clearing I_NEW,
859 * because we're guaranteed that nobody else tries to do anything about
860 * the state of the inode when it is locked, as we just created it (so
861 * there can be no old holders that haven't tested I_NEW).
862 * However we must emit the memory barrier so that other CPUs reliably
863 * see the clearing of I_NEW after the other inode initialisation has
867 WARN_ON(!(inode->i_state & I_NEW));
868 inode->i_state &= ~I_NEW;
869 wake_up_inode(inode);
871 EXPORT_SYMBOL(unlock_new_inode);
874 * This is called without the inode lock held.. Be careful.
876 * We no longer cache the sb_flags in i_flags - see fs.h
877 * -- rmk@arm.uk.linux.org
879 static struct inode *get_new_inode(struct super_block *sb,
880 struct hlist_head *head,
881 int (*test)(struct inode *, void *),
882 int (*set)(struct inode *, void *),
887 inode = alloc_inode(sb);
891 spin_lock(&inode_lock);
892 /* We released the lock, so.. */
893 old = find_inode(sb, head, test, data);
895 if (set(inode, data))
898 hlist_add_head(&inode->i_hash, head);
899 __inode_sb_list_add(inode);
900 inode->i_state = I_NEW;
901 spin_unlock(&inode_lock);
903 /* Return the locked inode with I_NEW set, the
904 * caller is responsible for filling in the contents
910 * Uhhuh, somebody else created the same inode under
911 * us. Use the old inode instead of the one we just
914 spin_unlock(&inode_lock);
915 destroy_inode(inode);
917 wait_on_inode(inode);
922 spin_unlock(&inode_lock);
923 destroy_inode(inode);
928 * get_new_inode_fast is the fast path version of get_new_inode, see the
929 * comment at iget_locked for details.
931 static struct inode *get_new_inode_fast(struct super_block *sb,
932 struct hlist_head *head, unsigned long ino)
936 inode = alloc_inode(sb);
940 spin_lock(&inode_lock);
941 /* We released the lock, so.. */
942 old = find_inode_fast(sb, head, ino);
945 hlist_add_head(&inode->i_hash, head);
946 __inode_sb_list_add(inode);
947 inode->i_state = I_NEW;
948 spin_unlock(&inode_lock);
950 /* Return the locked inode with I_NEW set, the
951 * caller is responsible for filling in the contents
957 * Uhhuh, somebody else created the same inode under
958 * us. Use the old inode instead of the one we just
961 spin_unlock(&inode_lock);
962 destroy_inode(inode);
964 wait_on_inode(inode);
970 * search the inode cache for a matching inode number.
971 * If we find one, then the inode number we are trying to
972 * allocate is not unique and so we should not use it.
974 * Returns 1 if the inode number is unique, 0 if it is not.
976 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
978 struct hlist_head *b = inode_hashtable + hash(sb, ino);
979 struct hlist_node *node;
982 hlist_for_each_entry(inode, node, b, i_hash) {
983 if (inode->i_ino == ino && inode->i_sb == sb)
991 * iunique - get a unique inode number
993 * @max_reserved: highest reserved inode number
995 * Obtain an inode number that is unique on the system for a given
996 * superblock. This is used by file systems that have no natural
997 * permanent inode numbering system. An inode number is returned that
998 * is higher than the reserved limit but unique.
1001 * With a large number of inodes live on the file system this function
1002 * currently becomes quite slow.
1004 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1007 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1008 * error if st_ino won't fit in target struct field. Use 32bit counter
1009 * here to attempt to avoid that.
1011 static DEFINE_SPINLOCK(iunique_lock);
1012 static unsigned int counter;
1015 spin_lock(&inode_lock);
1016 spin_lock(&iunique_lock);
1018 if (counter <= max_reserved)
1019 counter = max_reserved + 1;
1021 } while (!test_inode_iunique(sb, res));
1022 spin_unlock(&iunique_lock);
1023 spin_unlock(&inode_lock);
1027 EXPORT_SYMBOL(iunique);
1029 struct inode *igrab(struct inode *inode)
1031 spin_lock(&inode_lock);
1032 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
1036 * Handle the case where s_op->clear_inode is not been
1037 * called yet, and somebody is calling igrab
1038 * while the inode is getting freed.
1041 spin_unlock(&inode_lock);
1044 EXPORT_SYMBOL(igrab);
1047 * ifind - internal function, you want ilookup5() or iget5().
1048 * @sb: super block of file system to search
1049 * @head: the head of the list to search
1050 * @test: callback used for comparisons between inodes
1051 * @data: opaque data pointer to pass to @test
1052 * @wait: if true wait for the inode to be unlocked, if false do not
1054 * ifind() searches for the inode specified by @data in the inode
1055 * cache. This is a generalized version of ifind_fast() for file systems where
1056 * the inode number is not sufficient for unique identification of an inode.
1058 * If the inode is in the cache, the inode is returned with an incremented
1061 * Otherwise NULL is returned.
1063 * Note, @test is called with the inode_lock held, so can't sleep.
1065 static struct inode *ifind(struct super_block *sb,
1066 struct hlist_head *head, int (*test)(struct inode *, void *),
1067 void *data, const int wait)
1069 struct inode *inode;
1071 spin_lock(&inode_lock);
1072 inode = find_inode(sb, head, test, data);
1074 spin_unlock(&inode_lock);
1076 wait_on_inode(inode);
1079 spin_unlock(&inode_lock);
1084 * ifind_fast - internal function, you want ilookup() or iget().
1085 * @sb: super block of file system to search
1086 * @head: head of the list to search
1087 * @ino: inode number to search for
1089 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1090 * file systems where the inode number is sufficient for unique identification
1093 * If the inode is in the cache, the inode is returned with an incremented
1096 * Otherwise NULL is returned.
1098 static struct inode *ifind_fast(struct super_block *sb,
1099 struct hlist_head *head, unsigned long ino)
1101 struct inode *inode;
1103 spin_lock(&inode_lock);
1104 inode = find_inode_fast(sb, head, ino);
1106 spin_unlock(&inode_lock);
1107 wait_on_inode(inode);
1110 spin_unlock(&inode_lock);
1115 * ilookup5_nowait - search for an inode in the inode cache
1116 * @sb: super block of file system to search
1117 * @hashval: hash value (usually inode number) to search for
1118 * @test: callback used for comparisons between inodes
1119 * @data: opaque data pointer to pass to @test
1121 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1122 * @data in the inode cache. This is a generalized version of ilookup() for
1123 * file systems where the inode number is not sufficient for unique
1124 * identification of an inode.
1126 * If the inode is in the cache, the inode is returned with an incremented
1127 * reference count. Note, the inode lock is not waited upon so you have to be
1128 * very careful what you do with the returned inode. You probably should be
1129 * using ilookup5() instead.
1131 * Otherwise NULL is returned.
1133 * Note, @test is called with the inode_lock held, so can't sleep.
1135 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1136 int (*test)(struct inode *, void *), void *data)
1138 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1140 return ifind(sb, head, test, data, 0);
1142 EXPORT_SYMBOL(ilookup5_nowait);
1145 * ilookup5 - search for an inode in the inode cache
1146 * @sb: super block of file system to search
1147 * @hashval: hash value (usually inode number) to search for
1148 * @test: callback used for comparisons between inodes
1149 * @data: opaque data pointer to pass to @test
1151 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1152 * @data in the inode cache. This is a generalized version of ilookup() for
1153 * file systems where the inode number is not sufficient for unique
1154 * identification of an inode.
1156 * If the inode is in the cache, the inode lock is waited upon and the inode is
1157 * returned with an incremented reference count.
1159 * Otherwise NULL is returned.
1161 * Note, @test is called with the inode_lock held, so can't sleep.
1163 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1164 int (*test)(struct inode *, void *), void *data)
1166 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1168 return ifind(sb, head, test, data, 1);
1170 EXPORT_SYMBOL(ilookup5);
1173 * ilookup - search for an inode in the inode cache
1174 * @sb: super block of file system to search
1175 * @ino: inode number to search for
1177 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1178 * This is for file systems where the inode number is sufficient for unique
1179 * identification of an inode.
1181 * If the inode is in the cache, the inode is returned with an incremented
1184 * Otherwise NULL is returned.
1186 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1188 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1190 return ifind_fast(sb, head, ino);
1192 EXPORT_SYMBOL(ilookup);
1195 * iget5_locked - obtain an inode from a mounted file system
1196 * @sb: super block of file system
1197 * @hashval: hash value (usually inode number) to get
1198 * @test: callback used for comparisons between inodes
1199 * @set: callback used to initialize a new struct inode
1200 * @data: opaque data pointer to pass to @test and @set
1202 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1203 * and @data in the inode cache and if present it is returned with an increased
1204 * reference count. This is a generalized version of iget_locked() for file
1205 * systems where the inode number is not sufficient for unique identification
1208 * If the inode is not in cache, get_new_inode() is called to allocate a new
1209 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1210 * file system gets to fill it in before unlocking it via unlock_new_inode().
1212 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1214 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1215 int (*test)(struct inode *, void *),
1216 int (*set)(struct inode *, void *), void *data)
1218 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1219 struct inode *inode;
1221 inode = ifind(sb, head, test, data, 1);
1225 * get_new_inode() will do the right thing, re-trying the search
1226 * in case it had to block at any point.
1228 return get_new_inode(sb, head, test, set, data);
1230 EXPORT_SYMBOL(iget5_locked);
1233 * iget_locked - obtain an inode from a mounted file system
1234 * @sb: super block of file system
1235 * @ino: inode number to get
1237 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1238 * the inode cache and if present it is returned with an increased reference
1239 * count. This is for file systems where the inode number is sufficient for
1240 * unique identification of an inode.
1242 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1243 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1244 * The file system gets to fill it in before unlocking it via
1245 * unlock_new_inode().
1247 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1249 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1250 struct inode *inode;
1252 inode = ifind_fast(sb, head, ino);
1256 * get_new_inode_fast() will do the right thing, re-trying the search
1257 * in case it had to block at any point.
1259 return get_new_inode_fast(sb, head, ino);
1261 EXPORT_SYMBOL(iget_locked);
1263 int insert_inode_locked(struct inode *inode)
1265 struct super_block *sb = inode->i_sb;
1266 ino_t ino = inode->i_ino;
1267 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1269 inode->i_state |= I_NEW;
1271 struct hlist_node *node;
1272 struct inode *old = NULL;
1273 spin_lock(&inode_lock);
1274 hlist_for_each_entry(old, node, head, i_hash) {
1275 if (old->i_ino != ino)
1277 if (old->i_sb != sb)
1279 if (old->i_state & (I_FREEING|I_WILL_FREE))
1283 if (likely(!node)) {
1284 hlist_add_head(&inode->i_hash, head);
1285 spin_unlock(&inode_lock);
1289 spin_unlock(&inode_lock);
1291 if (unlikely(!inode_unhashed(old))) {
1298 EXPORT_SYMBOL(insert_inode_locked);
1300 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1301 int (*test)(struct inode *, void *), void *data)
1303 struct super_block *sb = inode->i_sb;
1304 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1306 inode->i_state |= I_NEW;
1309 struct hlist_node *node;
1310 struct inode *old = NULL;
1312 spin_lock(&inode_lock);
1313 hlist_for_each_entry(old, node, head, i_hash) {
1314 if (old->i_sb != sb)
1316 if (!test(old, data))
1318 if (old->i_state & (I_FREEING|I_WILL_FREE))
1322 if (likely(!node)) {
1323 hlist_add_head(&inode->i_hash, head);
1324 spin_unlock(&inode_lock);
1328 spin_unlock(&inode_lock);
1330 if (unlikely(!inode_unhashed(old))) {
1337 EXPORT_SYMBOL(insert_inode_locked4);
1340 int generic_delete_inode(struct inode *inode)
1344 EXPORT_SYMBOL(generic_delete_inode);
1347 * Normal UNIX filesystem behaviour: delete the
1348 * inode when the usage count drops to zero, and
1351 int generic_drop_inode(struct inode *inode)
1353 return !inode->i_nlink || inode_unhashed(inode);
1355 EXPORT_SYMBOL_GPL(generic_drop_inode);
1358 * Called when we're dropping the last reference
1361 * Call the FS "drop_inode()" function, defaulting to
1362 * the legacy UNIX filesystem behaviour. If it tells
1363 * us to evict inode, do so. Otherwise, retain inode
1364 * in cache if fs is alive, sync and evict if fs is
1367 static void iput_final(struct inode *inode)
1369 struct super_block *sb = inode->i_sb;
1370 const struct super_operations *op = inode->i_sb->s_op;
1373 if (op && op->drop_inode)
1374 drop = op->drop_inode(inode);
1376 drop = generic_drop_inode(inode);
1379 if (sb->s_flags & MS_ACTIVE) {
1380 inode->i_state |= I_REFERENCED;
1381 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1382 inode_lru_list_add(inode);
1384 spin_unlock(&inode_lock);
1387 WARN_ON(inode->i_state & I_NEW);
1388 inode->i_state |= I_WILL_FREE;
1389 spin_unlock(&inode_lock);
1390 write_inode_now(inode, 1);
1391 spin_lock(&inode_lock);
1392 WARN_ON(inode->i_state & I_NEW);
1393 inode->i_state &= ~I_WILL_FREE;
1394 __remove_inode_hash(inode);
1397 WARN_ON(inode->i_state & I_NEW);
1398 inode->i_state |= I_FREEING;
1401 * Move the inode off the IO lists and LRU once I_FREEING is
1402 * set so that it won't get moved back on there if it is dirty.
1404 inode_lru_list_del(inode);
1405 list_del_init(&inode->i_wb_list);
1407 __inode_sb_list_del(inode);
1408 spin_unlock(&inode_lock);
1410 remove_inode_hash(inode);
1411 wake_up_inode(inode);
1412 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1413 destroy_inode(inode);
1417 * iput - put an inode
1418 * @inode: inode to put
1420 * Puts an inode, dropping its usage count. If the inode use count hits
1421 * zero, the inode is then freed and may also be destroyed.
1423 * Consequently, iput() can sleep.
1425 void iput(struct inode *inode)
1428 BUG_ON(inode->i_state & I_CLEAR);
1430 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1434 EXPORT_SYMBOL(iput);
1437 * bmap - find a block number in a file
1438 * @inode: inode of file
1439 * @block: block to find
1441 * Returns the block number on the device holding the inode that
1442 * is the disk block number for the block of the file requested.
1443 * That is, asked for block 4 of inode 1 the function will return the
1444 * disk block relative to the disk start that holds that block of the
1447 sector_t bmap(struct inode *inode, sector_t block)
1450 if (inode->i_mapping->a_ops->bmap)
1451 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1454 EXPORT_SYMBOL(bmap);
1457 * With relative atime, only update atime if the previous atime is
1458 * earlier than either the ctime or mtime or if at least a day has
1459 * passed since the last atime update.
1461 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1462 struct timespec now)
1465 if (!(mnt->mnt_flags & MNT_RELATIME))
1468 * Is mtime younger than atime? If yes, update atime:
1470 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1473 * Is ctime younger than atime? If yes, update atime:
1475 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1479 * Is the previous atime value older than a day? If yes,
1482 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1485 * Good, we can skip the atime update:
1491 * touch_atime - update the access time
1492 * @mnt: mount the inode is accessed on
1493 * @dentry: dentry accessed
1495 * Update the accessed time on an inode and mark it for writeback.
1496 * This function automatically handles read only file systems and media,
1497 * as well as the "noatime" flag and inode specific "noatime" markers.
1499 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1501 struct inode *inode = dentry->d_inode;
1502 struct timespec now;
1504 if (inode->i_flags & S_NOATIME)
1506 if (IS_NOATIME(inode))
1508 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1511 if (mnt->mnt_flags & MNT_NOATIME)
1513 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1516 now = current_fs_time(inode->i_sb);
1518 if (!relatime_need_update(mnt, inode, now))
1521 if (timespec_equal(&inode->i_atime, &now))
1524 if (mnt_want_write(mnt))
1527 inode->i_atime = now;
1528 mark_inode_dirty_sync(inode);
1529 mnt_drop_write(mnt);
1531 EXPORT_SYMBOL(touch_atime);
1534 * file_update_time - update mtime and ctime time
1535 * @file: file accessed
1537 * Update the mtime and ctime members of an inode and mark the inode
1538 * for writeback. Note that this function is meant exclusively for
1539 * usage in the file write path of filesystems, and filesystems may
1540 * choose to explicitly ignore update via this function with the
1541 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1542 * timestamps are handled by the server.
1545 void file_update_time(struct file *file)
1547 struct inode *inode = file->f_path.dentry->d_inode;
1548 struct timespec now;
1549 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1551 /* First try to exhaust all avenues to not sync */
1552 if (IS_NOCMTIME(inode))
1555 now = current_fs_time(inode->i_sb);
1556 if (!timespec_equal(&inode->i_mtime, &now))
1559 if (!timespec_equal(&inode->i_ctime, &now))
1562 if (IS_I_VERSION(inode))
1563 sync_it |= S_VERSION;
1568 /* Finally allowed to write? Takes lock. */
1569 if (mnt_want_write_file(file))
1572 /* Only change inode inside the lock region */
1573 if (sync_it & S_VERSION)
1574 inode_inc_iversion(inode);
1575 if (sync_it & S_CTIME)
1576 inode->i_ctime = now;
1577 if (sync_it & S_MTIME)
1578 inode->i_mtime = now;
1579 mark_inode_dirty_sync(inode);
1580 mnt_drop_write(file->f_path.mnt);
1582 EXPORT_SYMBOL(file_update_time);
1584 int inode_needs_sync(struct inode *inode)
1588 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1592 EXPORT_SYMBOL(inode_needs_sync);
1594 int inode_wait(void *word)
1599 EXPORT_SYMBOL(inode_wait);
1602 * If we try to find an inode in the inode hash while it is being
1603 * deleted, we have to wait until the filesystem completes its
1604 * deletion before reporting that it isn't found. This function waits
1605 * until the deletion _might_ have completed. Callers are responsible
1606 * to recheck inode state.
1608 * It doesn't matter if I_NEW is not set initially, a call to
1609 * wake_up_inode() after removing from the hash list will DTRT.
1611 * This is called with inode_lock held.
1613 static void __wait_on_freeing_inode(struct inode *inode)
1615 wait_queue_head_t *wq;
1616 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1617 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1618 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1619 spin_unlock(&inode_lock);
1621 finish_wait(wq, &wait.wait);
1622 spin_lock(&inode_lock);
1625 static __initdata unsigned long ihash_entries;
1626 static int __init set_ihash_entries(char *str)
1630 ihash_entries = simple_strtoul(str, &str, 0);
1633 __setup("ihash_entries=", set_ihash_entries);
1636 * Initialize the waitqueues and inode hash table.
1638 void __init inode_init_early(void)
1642 /* If hashes are distributed across NUMA nodes, defer
1643 * hash allocation until vmalloc space is available.
1649 alloc_large_system_hash("Inode-cache",
1650 sizeof(struct hlist_head),
1658 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1659 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1662 void __init inode_init(void)
1666 /* inode slab cache */
1667 inode_cachep = kmem_cache_create("inode_cache",
1668 sizeof(struct inode),
1670 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1673 register_shrinker(&icache_shrinker);
1675 /* Hash may have been set up in inode_init_early */
1680 alloc_large_system_hash("Inode-cache",
1681 sizeof(struct hlist_head),
1689 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1690 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1693 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1695 inode->i_mode = mode;
1696 if (S_ISCHR(mode)) {
1697 inode->i_fop = &def_chr_fops;
1698 inode->i_rdev = rdev;
1699 } else if (S_ISBLK(mode)) {
1700 inode->i_fop = &def_blk_fops;
1701 inode->i_rdev = rdev;
1702 } else if (S_ISFIFO(mode))
1703 inode->i_fop = &def_fifo_fops;
1704 else if (S_ISSOCK(mode))
1705 inode->i_fop = &bad_sock_fops;
1707 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1708 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1711 EXPORT_SYMBOL(init_special_inode);
1714 * Init uid,gid,mode for new inode according to posix standards
1716 * @dir: Directory inode
1717 * @mode: mode of the new inode
1719 void inode_init_owner(struct inode *inode, const struct inode *dir,
1722 inode->i_uid = current_fsuid();
1723 if (dir && dir->i_mode & S_ISGID) {
1724 inode->i_gid = dir->i_gid;
1728 inode->i_gid = current_fsgid();
1729 inode->i_mode = mode;
1731 EXPORT_SYMBOL(inode_init_owner);