4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/slab.h>
12 #include <linux/writeback.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/wait.h>
16 #include <linux/rwsem.h>
17 #include <linux/hash.h>
18 #include <linux/swap.h>
19 #include <linux/security.h>
20 #include <linux/pagemap.h>
21 #include <linux/cdev.h>
22 #include <linux/bootmem.h>
23 #include <linux/fsnotify.h>
24 #include <linux/mount.h>
25 #include <linux/async.h>
26 #include <linux/posix_acl.h>
29 * This is needed for the following functions:
31 * - invalidate_inode_buffers
34 * FIXME: remove all knowledge of the buffer layer from this file
36 #include <linux/buffer_head.h>
39 * New inode.c implementation.
41 * This implementation has the basic premise of trying
42 * to be extremely low-overhead and SMP-safe, yet be
43 * simple enough to be "obviously correct".
48 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
50 /* #define INODE_PARANOIA 1 */
51 /* #define INODE_DEBUG 1 */
54 * Inode lookup is no longer as critical as it used to be:
55 * most of the lookups are going to be through the dcache.
57 #define I_HASHBITS i_hash_shift
58 #define I_HASHMASK i_hash_mask
60 static unsigned int i_hash_mask __read_mostly;
61 static unsigned int i_hash_shift __read_mostly;
64 * Each inode can be on two separate lists. One is
65 * the hash list of the inode, used for lookups. The
66 * other linked list is the "type" list:
67 * "in_use" - valid inode, i_count > 0, i_nlink > 0
68 * "dirty" - as "in_use" but also dirty
69 * "unused" - valid inode, i_count = 0
71 * A "dirty" list is maintained for each super block,
72 * allowing for low-overhead inode sync() operations.
75 static LIST_HEAD(inode_unused);
76 static struct hlist_head *inode_hashtable __read_mostly;
79 * A simple spinlock to protect the list manipulations.
81 * NOTE! You also have to own the lock if you change
82 * the i_state of an inode while it is in use..
84 DEFINE_SPINLOCK(inode_lock);
87 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
88 * icache shrinking path, and the umount path. Without this exclusion,
89 * by the time prune_icache calls iput for the inode whose pages it has
90 * been invalidating, or by the time it calls clear_inode & destroy_inode
91 * from its final dispose_list, the struct super_block they refer to
92 * (for inode->i_sb->s_op) may already have been freed and reused.
94 * We make this an rwsem because the fastpath is icache shrinking. In
95 * some cases a filesystem may be doing a significant amount of work in
96 * its inode reclaim code, so this should improve parallelism.
98 static DECLARE_RWSEM(iprune_sem);
101 * Statistics gathering..
103 struct inodes_stat_t inodes_stat;
105 static struct percpu_counter nr_inodes __cacheline_aligned_in_smp;
106 static struct percpu_counter nr_inodes_unused __cacheline_aligned_in_smp;
108 static struct kmem_cache *inode_cachep __read_mostly;
110 static inline int get_nr_inodes(void)
112 return percpu_counter_sum_positive(&nr_inodes);
115 static inline int get_nr_inodes_unused(void)
117 return percpu_counter_sum_positive(&nr_inodes_unused);
120 int get_nr_dirty_inodes(void)
122 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
123 return nr_dirty > 0 ? nr_dirty : 0;
128 * Handle nr_inode sysctl
131 int proc_nr_inodes(ctl_table *table, int write,
132 void __user *buffer, size_t *lenp, loff_t *ppos)
134 inodes_stat.nr_inodes = get_nr_inodes();
135 inodes_stat.nr_unused = get_nr_inodes_unused();
136 return proc_dointvec(table, write, buffer, lenp, ppos);
140 static void wake_up_inode(struct inode *inode)
143 * Prevent speculative execution through spin_unlock(&inode_lock);
146 wake_up_bit(&inode->i_state, __I_NEW);
150 * inode_init_always - perform inode structure intialisation
151 * @sb: superblock inode belongs to
152 * @inode: inode to initialise
154 * These are initializations that need to be done on every inode
155 * allocation as the fields are not initialised by slab allocation.
157 int inode_init_always(struct super_block *sb, struct inode *inode)
159 static const struct address_space_operations empty_aops;
160 static const struct inode_operations empty_iops;
161 static const struct file_operations empty_fops;
162 struct address_space *const mapping = &inode->i_data;
165 inode->i_blkbits = sb->s_blocksize_bits;
167 atomic_set(&inode->i_count, 1);
168 inode->i_op = &empty_iops;
169 inode->i_fop = &empty_fops;
173 atomic_set(&inode->i_writecount, 0);
177 inode->i_generation = 0;
179 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
181 inode->i_pipe = NULL;
182 inode->i_bdev = NULL;
183 inode->i_cdev = NULL;
185 inode->dirtied_when = 0;
187 if (security_inode_alloc(inode))
189 spin_lock_init(&inode->i_lock);
190 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
192 mutex_init(&inode->i_mutex);
193 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
195 init_rwsem(&inode->i_alloc_sem);
196 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
198 mapping->a_ops = &empty_aops;
199 mapping->host = inode;
201 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
202 mapping->assoc_mapping = NULL;
203 mapping->backing_dev_info = &default_backing_dev_info;
204 mapping->writeback_index = 0;
207 * If the block_device provides a backing_dev_info for client
208 * inodes then use that. Otherwise the inode share the bdev's
212 struct backing_dev_info *bdi;
214 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
215 mapping->backing_dev_info = bdi;
217 inode->i_private = NULL;
218 inode->i_mapping = mapping;
219 #ifdef CONFIG_FS_POSIX_ACL
220 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
223 #ifdef CONFIG_FSNOTIFY
224 inode->i_fsnotify_mask = 0;
227 percpu_counter_inc(&nr_inodes);
233 EXPORT_SYMBOL(inode_init_always);
235 static struct inode *alloc_inode(struct super_block *sb)
239 if (sb->s_op->alloc_inode)
240 inode = sb->s_op->alloc_inode(sb);
242 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
247 if (unlikely(inode_init_always(sb, inode))) {
248 if (inode->i_sb->s_op->destroy_inode)
249 inode->i_sb->s_op->destroy_inode(inode);
251 kmem_cache_free(inode_cachep, inode);
258 void __destroy_inode(struct inode *inode)
260 BUG_ON(inode_has_buffers(inode));
261 security_inode_free(inode);
262 fsnotify_inode_delete(inode);
263 #ifdef CONFIG_FS_POSIX_ACL
264 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
265 posix_acl_release(inode->i_acl);
266 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
267 posix_acl_release(inode->i_default_acl);
269 percpu_counter_dec(&nr_inodes);
271 EXPORT_SYMBOL(__destroy_inode);
273 static void destroy_inode(struct inode *inode)
275 __destroy_inode(inode);
276 if (inode->i_sb->s_op->destroy_inode)
277 inode->i_sb->s_op->destroy_inode(inode);
279 kmem_cache_free(inode_cachep, (inode));
283 * These are initializations that only need to be done
284 * once, because the fields are idempotent across use
285 * of the inode, so let the slab aware of that.
287 void inode_init_once(struct inode *inode)
289 memset(inode, 0, sizeof(*inode));
290 INIT_HLIST_NODE(&inode->i_hash);
291 INIT_LIST_HEAD(&inode->i_dentry);
292 INIT_LIST_HEAD(&inode->i_devices);
293 INIT_LIST_HEAD(&inode->i_list);
294 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
295 spin_lock_init(&inode->i_data.tree_lock);
296 spin_lock_init(&inode->i_data.i_mmap_lock);
297 INIT_LIST_HEAD(&inode->i_data.private_list);
298 spin_lock_init(&inode->i_data.private_lock);
299 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
300 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
301 i_size_ordered_init(inode);
302 #ifdef CONFIG_FSNOTIFY
303 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
306 EXPORT_SYMBOL(inode_init_once);
308 static void init_once(void *foo)
310 struct inode *inode = (struct inode *) foo;
312 inode_init_once(inode);
316 * inode_lock must be held
318 void __iget(struct inode *inode)
320 atomic_inc(&inode->i_count);
323 static void inode_lru_list_add(struct inode *inode)
325 if (list_empty(&inode->i_list)) {
326 list_add(&inode->i_list, &inode_unused);
327 percpu_counter_inc(&nr_inodes_unused);
331 static void inode_lru_list_del(struct inode *inode)
333 if (!list_empty(&inode->i_list)) {
334 list_del_init(&inode->i_list);
335 percpu_counter_dec(&nr_inodes_unused);
339 static unsigned long hash(struct super_block *sb, unsigned long hashval)
343 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
345 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
346 return tmp & I_HASHMASK;
350 * __insert_inode_hash - hash an inode
351 * @inode: unhashed inode
352 * @hashval: unsigned long value used to locate this object in the
355 * Add an inode to the inode hash for this superblock.
357 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
359 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
360 spin_lock(&inode_lock);
361 hlist_add_head(&inode->i_hash, head);
362 spin_unlock(&inode_lock);
364 EXPORT_SYMBOL(__insert_inode_hash);
367 * __remove_inode_hash - remove an inode from the hash
368 * @inode: inode to unhash
370 * Remove an inode from the superblock.
372 static void __remove_inode_hash(struct inode *inode)
374 hlist_del_init(&inode->i_hash);
378 * remove_inode_hash - remove an inode from the hash
379 * @inode: inode to unhash
381 * Remove an inode from the superblock.
383 void remove_inode_hash(struct inode *inode)
385 spin_lock(&inode_lock);
386 hlist_del_init(&inode->i_hash);
387 spin_unlock(&inode_lock);
389 EXPORT_SYMBOL(remove_inode_hash);
391 void end_writeback(struct inode *inode)
394 BUG_ON(inode->i_data.nrpages);
395 BUG_ON(!list_empty(&inode->i_data.private_list));
396 BUG_ON(!(inode->i_state & I_FREEING));
397 BUG_ON(inode->i_state & I_CLEAR);
398 inode_sync_wait(inode);
399 inode->i_state = I_FREEING | I_CLEAR;
401 EXPORT_SYMBOL(end_writeback);
403 static void evict(struct inode *inode)
405 const struct super_operations *op = inode->i_sb->s_op;
407 if (op->evict_inode) {
408 op->evict_inode(inode);
410 if (inode->i_data.nrpages)
411 truncate_inode_pages(&inode->i_data, 0);
412 end_writeback(inode);
414 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
416 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
421 * dispose_list - dispose of the contents of a local list
422 * @head: the head of the list to free
424 * Dispose-list gets a local list with local inodes in it, so it doesn't
425 * need to worry about list corruption and SMP locks.
427 static void dispose_list(struct list_head *head)
429 while (!list_empty(head)) {
432 inode = list_first_entry(head, struct inode, i_list);
433 list_del_init(&inode->i_list);
437 spin_lock(&inode_lock);
438 __remove_inode_hash(inode);
439 list_del_init(&inode->i_sb_list);
440 spin_unlock(&inode_lock);
442 wake_up_inode(inode);
443 destroy_inode(inode);
448 * Invalidate all inodes for a device.
450 static int invalidate_list(struct list_head *head, struct list_head *dispose)
452 struct list_head *next;
457 struct list_head *tmp = next;
461 * We can reschedule here without worrying about the list's
462 * consistency because the per-sb list of inodes must not
463 * change during umount anymore, and because iprune_sem keeps
464 * shrink_icache_memory() away.
466 cond_resched_lock(&inode_lock);
471 inode = list_entry(tmp, struct inode, i_sb_list);
472 if (inode->i_state & I_NEW)
474 invalidate_inode_buffers(inode);
475 if (!atomic_read(&inode->i_count)) {
476 list_move(&inode->i_list, dispose);
477 WARN_ON(inode->i_state & I_NEW);
478 inode->i_state |= I_FREEING;
479 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
480 percpu_counter_dec(&nr_inodes_unused);
489 * invalidate_inodes - discard the inodes on a device
492 * Discard all of the inodes for a given superblock. If the discard
493 * fails because there are busy inodes then a non zero value is returned.
494 * If the discard is successful all the inodes have been discarded.
496 int invalidate_inodes(struct super_block *sb)
499 LIST_HEAD(throw_away);
501 down_write(&iprune_sem);
502 spin_lock(&inode_lock);
503 fsnotify_unmount_inodes(&sb->s_inodes);
504 busy = invalidate_list(&sb->s_inodes, &throw_away);
505 spin_unlock(&inode_lock);
507 dispose_list(&throw_away);
508 up_write(&iprune_sem);
513 static int can_unuse(struct inode *inode)
515 if (inode->i_state & ~I_REFERENCED)
517 if (inode_has_buffers(inode))
519 if (atomic_read(&inode->i_count))
521 if (inode->i_data.nrpages)
527 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
528 * temporary list and then are freed outside inode_lock by dispose_list().
530 * Any inodes which are pinned purely because of attached pagecache have their
531 * pagecache removed. If the inode has metadata buffers attached to
532 * mapping->private_list then try to remove them.
534 * If the inode has the I_REFERENCED flag set, then it means that it has been
535 * used recently - the flag is set in iput_final(). When we encounter such an
536 * inode, clear the flag and move it to the back of the LRU so it gets another
537 * pass through the LRU before it gets reclaimed. This is necessary because of
538 * the fact we are doing lazy LRU updates to minimise lock contention so the
539 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
540 * with this flag set because they are the inodes that are out of order.
542 static void prune_icache(int nr_to_scan)
546 unsigned long reap = 0;
548 down_read(&iprune_sem);
549 spin_lock(&inode_lock);
550 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
553 if (list_empty(&inode_unused))
556 inode = list_entry(inode_unused.prev, struct inode, i_list);
559 * Referenced or dirty inodes are still in use. Give them
560 * another pass through the LRU as we canot reclaim them now.
562 if (atomic_read(&inode->i_count) ||
563 (inode->i_state & ~I_REFERENCED)) {
564 list_del_init(&inode->i_list);
565 percpu_counter_dec(&nr_inodes_unused);
569 /* recently referenced inodes get one more pass */
570 if (inode->i_state & I_REFERENCED) {
571 list_move(&inode->i_list, &inode_unused);
572 inode->i_state &= ~I_REFERENCED;
575 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
577 spin_unlock(&inode_lock);
578 if (remove_inode_buffers(inode))
579 reap += invalidate_mapping_pages(&inode->i_data,
582 spin_lock(&inode_lock);
584 if (inode != list_entry(inode_unused.next,
585 struct inode, i_list))
586 continue; /* wrong inode or list_empty */
587 if (!can_unuse(inode))
590 list_move(&inode->i_list, &freeable);
591 WARN_ON(inode->i_state & I_NEW);
592 inode->i_state |= I_FREEING;
593 percpu_counter_dec(&nr_inodes_unused);
595 if (current_is_kswapd())
596 __count_vm_events(KSWAPD_INODESTEAL, reap);
598 __count_vm_events(PGINODESTEAL, reap);
599 spin_unlock(&inode_lock);
601 dispose_list(&freeable);
602 up_read(&iprune_sem);
606 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
607 * "unused" means that no dentries are referring to the inodes: the files are
608 * not open and the dcache references to those inodes have already been
611 * This function is passed the number of inodes to scan, and it returns the
612 * total number of remaining possibly-reclaimable inodes.
614 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
618 * Nasty deadlock avoidance. We may hold various FS locks,
619 * and we don't want to recurse into the FS that called us
620 * in clear_inode() and friends..
622 if (!(gfp_mask & __GFP_FS))
626 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
629 static struct shrinker icache_shrinker = {
630 .shrink = shrink_icache_memory,
631 .seeks = DEFAULT_SEEKS,
634 static void __wait_on_freeing_inode(struct inode *inode);
636 * Called with the inode lock held.
637 * NOTE: we are not increasing the inode-refcount, you must call __iget()
638 * by hand after calling find_inode now! This simplifies iunique and won't
639 * add any additional branch in the common code.
641 static struct inode *find_inode(struct super_block *sb,
642 struct hlist_head *head,
643 int (*test)(struct inode *, void *),
646 struct hlist_node *node;
647 struct inode *inode = NULL;
650 hlist_for_each_entry(inode, node, head, i_hash) {
651 if (inode->i_sb != sb)
653 if (!test(inode, data))
655 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
656 __wait_on_freeing_inode(inode);
661 return node ? inode : NULL;
665 * find_inode_fast is the fast path version of find_inode, see the comment at
666 * iget_locked for details.
668 static struct inode *find_inode_fast(struct super_block *sb,
669 struct hlist_head *head, unsigned long ino)
671 struct hlist_node *node;
672 struct inode *inode = NULL;
675 hlist_for_each_entry(inode, node, head, i_hash) {
676 if (inode->i_ino != ino)
678 if (inode->i_sb != sb)
680 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
681 __wait_on_freeing_inode(inode);
686 return node ? inode : NULL;
690 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
693 list_add(&inode->i_sb_list, &sb->s_inodes);
695 hlist_add_head(&inode->i_hash, head);
699 * inode_add_to_lists - add a new inode to relevant lists
700 * @sb: superblock inode belongs to
701 * @inode: inode to mark in use
703 * When an inode is allocated it needs to be accounted for, added to the in use
704 * list, the owning superblock and the inode hash. This needs to be done under
705 * the inode_lock, so export a function to do this rather than the inode lock
706 * itself. We calculate the hash list to add to here so it is all internal
707 * which requires the caller to have already set up the inode number in the
710 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
712 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
714 spin_lock(&inode_lock);
715 __inode_add_to_lists(sb, head, inode);
716 spin_unlock(&inode_lock);
718 EXPORT_SYMBOL_GPL(inode_add_to_lists);
721 * new_inode - obtain an inode
724 * Allocates a new inode for given superblock. The default gfp_mask
725 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
726 * If HIGHMEM pages are unsuitable or it is known that pages allocated
727 * for the page cache are not reclaimable or migratable,
728 * mapping_set_gfp_mask() must be called with suitable flags on the
729 * newly created inode's mapping
732 struct inode *new_inode(struct super_block *sb)
735 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
736 * error if st_ino won't fit in target struct field. Use 32bit counter
737 * here to attempt to avoid that.
739 static unsigned int last_ino;
742 spin_lock_prefetch(&inode_lock);
744 inode = alloc_inode(sb);
746 spin_lock(&inode_lock);
747 __inode_add_to_lists(sb, NULL, inode);
748 inode->i_ino = ++last_ino;
750 spin_unlock(&inode_lock);
754 EXPORT_SYMBOL(new_inode);
756 void unlock_new_inode(struct inode *inode)
758 #ifdef CONFIG_DEBUG_LOCK_ALLOC
759 if (S_ISDIR(inode->i_mode)) {
760 struct file_system_type *type = inode->i_sb->s_type;
762 /* Set new key only if filesystem hasn't already changed it */
763 if (!lockdep_match_class(&inode->i_mutex,
764 &type->i_mutex_key)) {
766 * ensure nobody is actually holding i_mutex
768 mutex_destroy(&inode->i_mutex);
769 mutex_init(&inode->i_mutex);
770 lockdep_set_class(&inode->i_mutex,
771 &type->i_mutex_dir_key);
776 * This is special! We do not need the spinlock when clearing I_NEW,
777 * because we're guaranteed that nobody else tries to do anything about
778 * the state of the inode when it is locked, as we just created it (so
779 * there can be no old holders that haven't tested I_NEW).
780 * However we must emit the memory barrier so that other CPUs reliably
781 * see the clearing of I_NEW after the other inode initialisation has
785 WARN_ON(!(inode->i_state & I_NEW));
786 inode->i_state &= ~I_NEW;
787 wake_up_inode(inode);
789 EXPORT_SYMBOL(unlock_new_inode);
792 * This is called without the inode lock held.. Be careful.
794 * We no longer cache the sb_flags in i_flags - see fs.h
795 * -- rmk@arm.uk.linux.org
797 static struct inode *get_new_inode(struct super_block *sb,
798 struct hlist_head *head,
799 int (*test)(struct inode *, void *),
800 int (*set)(struct inode *, void *),
805 inode = alloc_inode(sb);
809 spin_lock(&inode_lock);
810 /* We released the lock, so.. */
811 old = find_inode(sb, head, test, data);
813 if (set(inode, data))
816 __inode_add_to_lists(sb, head, inode);
817 inode->i_state = I_NEW;
818 spin_unlock(&inode_lock);
820 /* Return the locked inode with I_NEW set, the
821 * caller is responsible for filling in the contents
827 * Uhhuh, somebody else created the same inode under
828 * us. Use the old inode instead of the one we just
832 spin_unlock(&inode_lock);
833 destroy_inode(inode);
835 wait_on_inode(inode);
840 spin_unlock(&inode_lock);
841 destroy_inode(inode);
846 * get_new_inode_fast is the fast path version of get_new_inode, see the
847 * comment at iget_locked for details.
849 static struct inode *get_new_inode_fast(struct super_block *sb,
850 struct hlist_head *head, unsigned long ino)
854 inode = alloc_inode(sb);
858 spin_lock(&inode_lock);
859 /* We released the lock, so.. */
860 old = find_inode_fast(sb, head, ino);
863 __inode_add_to_lists(sb, head, inode);
864 inode->i_state = I_NEW;
865 spin_unlock(&inode_lock);
867 /* Return the locked inode with I_NEW set, the
868 * caller is responsible for filling in the contents
874 * Uhhuh, somebody else created the same inode under
875 * us. Use the old inode instead of the one we just
879 spin_unlock(&inode_lock);
880 destroy_inode(inode);
882 wait_on_inode(inode);
888 * search the inode cache for a matching inode number.
889 * If we find one, then the inode number we are trying to
890 * allocate is not unique and so we should not use it.
892 * Returns 1 if the inode number is unique, 0 if it is not.
894 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
896 struct hlist_head *b = inode_hashtable + hash(sb, ino);
897 struct hlist_node *node;
900 hlist_for_each_entry(inode, node, b, i_hash) {
901 if (inode->i_ino == ino && inode->i_sb == sb)
909 * iunique - get a unique inode number
911 * @max_reserved: highest reserved inode number
913 * Obtain an inode number that is unique on the system for a given
914 * superblock. This is used by file systems that have no natural
915 * permanent inode numbering system. An inode number is returned that
916 * is higher than the reserved limit but unique.
919 * With a large number of inodes live on the file system this function
920 * currently becomes quite slow.
922 ino_t iunique(struct super_block *sb, ino_t max_reserved)
925 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
926 * error if st_ino won't fit in target struct field. Use 32bit counter
927 * here to attempt to avoid that.
929 static DEFINE_SPINLOCK(iunique_lock);
930 static unsigned int counter;
933 spin_lock(&inode_lock);
934 spin_lock(&iunique_lock);
936 if (counter <= max_reserved)
937 counter = max_reserved + 1;
939 } while (!test_inode_iunique(sb, res));
940 spin_unlock(&iunique_lock);
941 spin_unlock(&inode_lock);
945 EXPORT_SYMBOL(iunique);
947 struct inode *igrab(struct inode *inode)
949 spin_lock(&inode_lock);
950 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
954 * Handle the case where s_op->clear_inode is not been
955 * called yet, and somebody is calling igrab
956 * while the inode is getting freed.
959 spin_unlock(&inode_lock);
962 EXPORT_SYMBOL(igrab);
965 * ifind - internal function, you want ilookup5() or iget5().
966 * @sb: super block of file system to search
967 * @head: the head of the list to search
968 * @test: callback used for comparisons between inodes
969 * @data: opaque data pointer to pass to @test
970 * @wait: if true wait for the inode to be unlocked, if false do not
972 * ifind() searches for the inode specified by @data in the inode
973 * cache. This is a generalized version of ifind_fast() for file systems where
974 * the inode number is not sufficient for unique identification of an inode.
976 * If the inode is in the cache, the inode is returned with an incremented
979 * Otherwise NULL is returned.
981 * Note, @test is called with the inode_lock held, so can't sleep.
983 static struct inode *ifind(struct super_block *sb,
984 struct hlist_head *head, int (*test)(struct inode *, void *),
985 void *data, const int wait)
989 spin_lock(&inode_lock);
990 inode = find_inode(sb, head, test, data);
993 spin_unlock(&inode_lock);
995 wait_on_inode(inode);
998 spin_unlock(&inode_lock);
1003 * ifind_fast - internal function, you want ilookup() or iget().
1004 * @sb: super block of file system to search
1005 * @head: head of the list to search
1006 * @ino: inode number to search for
1008 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1009 * file systems where the inode number is sufficient for unique identification
1012 * If the inode is in the cache, the inode is returned with an incremented
1015 * Otherwise NULL is returned.
1017 static struct inode *ifind_fast(struct super_block *sb,
1018 struct hlist_head *head, unsigned long ino)
1020 struct inode *inode;
1022 spin_lock(&inode_lock);
1023 inode = find_inode_fast(sb, head, ino);
1026 spin_unlock(&inode_lock);
1027 wait_on_inode(inode);
1030 spin_unlock(&inode_lock);
1035 * ilookup5_nowait - search for an inode in the inode cache
1036 * @sb: super block of file system to search
1037 * @hashval: hash value (usually inode number) to search for
1038 * @test: callback used for comparisons between inodes
1039 * @data: opaque data pointer to pass to @test
1041 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1042 * @data in the inode cache. This is a generalized version of ilookup() for
1043 * file systems where the inode number is not sufficient for unique
1044 * identification of an inode.
1046 * If the inode is in the cache, the inode is returned with an incremented
1047 * reference count. Note, the inode lock is not waited upon so you have to be
1048 * very careful what you do with the returned inode. You probably should be
1049 * using ilookup5() instead.
1051 * Otherwise NULL is returned.
1053 * Note, @test is called with the inode_lock held, so can't sleep.
1055 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1056 int (*test)(struct inode *, void *), void *data)
1058 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1060 return ifind(sb, head, test, data, 0);
1062 EXPORT_SYMBOL(ilookup5_nowait);
1065 * ilookup5 - search for an inode in the inode cache
1066 * @sb: super block of file system to search
1067 * @hashval: hash value (usually inode number) to search for
1068 * @test: callback used for comparisons between inodes
1069 * @data: opaque data pointer to pass to @test
1071 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1072 * @data in the inode cache. This is a generalized version of ilookup() for
1073 * file systems where the inode number is not sufficient for unique
1074 * identification of an inode.
1076 * If the inode is in the cache, the inode lock is waited upon and the inode is
1077 * returned with an incremented reference count.
1079 * Otherwise NULL is returned.
1081 * Note, @test is called with the inode_lock held, so can't sleep.
1083 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1084 int (*test)(struct inode *, void *), void *data)
1086 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1088 return ifind(sb, head, test, data, 1);
1090 EXPORT_SYMBOL(ilookup5);
1093 * ilookup - search for an inode in the inode cache
1094 * @sb: super block of file system to search
1095 * @ino: inode number to search for
1097 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1098 * This is for file systems where the inode number is sufficient for unique
1099 * identification of an inode.
1101 * If the inode is in the cache, the inode is returned with an incremented
1104 * Otherwise NULL is returned.
1106 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1108 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1110 return ifind_fast(sb, head, ino);
1112 EXPORT_SYMBOL(ilookup);
1115 * iget5_locked - obtain an inode from a mounted file system
1116 * @sb: super block of file system
1117 * @hashval: hash value (usually inode number) to get
1118 * @test: callback used for comparisons between inodes
1119 * @set: callback used to initialize a new struct inode
1120 * @data: opaque data pointer to pass to @test and @set
1122 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1123 * and @data in the inode cache and if present it is returned with an increased
1124 * reference count. This is a generalized version of iget_locked() for file
1125 * systems where the inode number is not sufficient for unique identification
1128 * If the inode is not in cache, get_new_inode() is called to allocate a new
1129 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1130 * file system gets to fill it in before unlocking it via unlock_new_inode().
1132 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1134 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1135 int (*test)(struct inode *, void *),
1136 int (*set)(struct inode *, void *), void *data)
1138 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1139 struct inode *inode;
1141 inode = ifind(sb, head, test, data, 1);
1145 * get_new_inode() will do the right thing, re-trying the search
1146 * in case it had to block at any point.
1148 return get_new_inode(sb, head, test, set, data);
1150 EXPORT_SYMBOL(iget5_locked);
1153 * iget_locked - obtain an inode from a mounted file system
1154 * @sb: super block of file system
1155 * @ino: inode number to get
1157 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1158 * the inode cache and if present it is returned with an increased reference
1159 * count. This is for file systems where the inode number is sufficient for
1160 * unique identification of an inode.
1162 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1163 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1164 * The file system gets to fill it in before unlocking it via
1165 * unlock_new_inode().
1167 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1169 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1170 struct inode *inode;
1172 inode = ifind_fast(sb, head, ino);
1176 * get_new_inode_fast() will do the right thing, re-trying the search
1177 * in case it had to block at any point.
1179 return get_new_inode_fast(sb, head, ino);
1181 EXPORT_SYMBOL(iget_locked);
1183 int insert_inode_locked(struct inode *inode)
1185 struct super_block *sb = inode->i_sb;
1186 ino_t ino = inode->i_ino;
1187 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1189 inode->i_state |= I_NEW;
1191 struct hlist_node *node;
1192 struct inode *old = NULL;
1193 spin_lock(&inode_lock);
1194 hlist_for_each_entry(old, node, head, i_hash) {
1195 if (old->i_ino != ino)
1197 if (old->i_sb != sb)
1199 if (old->i_state & (I_FREEING|I_WILL_FREE))
1203 if (likely(!node)) {
1204 hlist_add_head(&inode->i_hash, head);
1205 spin_unlock(&inode_lock);
1209 spin_unlock(&inode_lock);
1211 if (unlikely(!inode_unhashed(old))) {
1218 EXPORT_SYMBOL(insert_inode_locked);
1220 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1221 int (*test)(struct inode *, void *), void *data)
1223 struct super_block *sb = inode->i_sb;
1224 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1226 inode->i_state |= I_NEW;
1229 struct hlist_node *node;
1230 struct inode *old = NULL;
1232 spin_lock(&inode_lock);
1233 hlist_for_each_entry(old, node, head, i_hash) {
1234 if (old->i_sb != sb)
1236 if (!test(old, data))
1238 if (old->i_state & (I_FREEING|I_WILL_FREE))
1242 if (likely(!node)) {
1243 hlist_add_head(&inode->i_hash, head);
1244 spin_unlock(&inode_lock);
1248 spin_unlock(&inode_lock);
1250 if (unlikely(!inode_unhashed(old))) {
1257 EXPORT_SYMBOL(insert_inode_locked4);
1260 int generic_delete_inode(struct inode *inode)
1264 EXPORT_SYMBOL(generic_delete_inode);
1267 * Normal UNIX filesystem behaviour: delete the
1268 * inode when the usage count drops to zero, and
1271 int generic_drop_inode(struct inode *inode)
1273 return !inode->i_nlink || inode_unhashed(inode);
1275 EXPORT_SYMBOL_GPL(generic_drop_inode);
1278 * Called when we're dropping the last reference
1281 * Call the FS "drop_inode()" function, defaulting to
1282 * the legacy UNIX filesystem behaviour. If it tells
1283 * us to evict inode, do so. Otherwise, retain inode
1284 * in cache if fs is alive, sync and evict if fs is
1287 static void iput_final(struct inode *inode)
1289 struct super_block *sb = inode->i_sb;
1290 const struct super_operations *op = inode->i_sb->s_op;
1293 if (op && op->drop_inode)
1294 drop = op->drop_inode(inode);
1296 drop = generic_drop_inode(inode);
1299 if (sb->s_flags & MS_ACTIVE) {
1300 inode->i_state |= I_REFERENCED;
1301 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1302 inode_lru_list_add(inode);
1304 spin_unlock(&inode_lock);
1307 WARN_ON(inode->i_state & I_NEW);
1308 inode->i_state |= I_WILL_FREE;
1309 spin_unlock(&inode_lock);
1310 write_inode_now(inode, 1);
1311 spin_lock(&inode_lock);
1312 WARN_ON(inode->i_state & I_NEW);
1313 inode->i_state &= ~I_WILL_FREE;
1314 __remove_inode_hash(inode);
1316 WARN_ON(inode->i_state & I_NEW);
1317 inode->i_state |= I_FREEING;
1320 * After we delete the inode from the LRU here, we avoid moving dirty
1321 * inodes back onto the LRU now because I_FREEING is set and hence
1322 * writeback_single_inode() won't move the inode around.
1324 inode_lru_list_del(inode);
1326 list_del_init(&inode->i_sb_list);
1327 spin_unlock(&inode_lock);
1329 remove_inode_hash(inode);
1330 wake_up_inode(inode);
1331 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1332 destroy_inode(inode);
1336 * iput - put an inode
1337 * @inode: inode to put
1339 * Puts an inode, dropping its usage count. If the inode use count hits
1340 * zero, the inode is then freed and may also be destroyed.
1342 * Consequently, iput() can sleep.
1344 void iput(struct inode *inode)
1347 BUG_ON(inode->i_state & I_CLEAR);
1349 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1353 EXPORT_SYMBOL(iput);
1356 * bmap - find a block number in a file
1357 * @inode: inode of file
1358 * @block: block to find
1360 * Returns the block number on the device holding the inode that
1361 * is the disk block number for the block of the file requested.
1362 * That is, asked for block 4 of inode 1 the function will return the
1363 * disk block relative to the disk start that holds that block of the
1366 sector_t bmap(struct inode *inode, sector_t block)
1369 if (inode->i_mapping->a_ops->bmap)
1370 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1373 EXPORT_SYMBOL(bmap);
1376 * With relative atime, only update atime if the previous atime is
1377 * earlier than either the ctime or mtime or if at least a day has
1378 * passed since the last atime update.
1380 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1381 struct timespec now)
1384 if (!(mnt->mnt_flags & MNT_RELATIME))
1387 * Is mtime younger than atime? If yes, update atime:
1389 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1392 * Is ctime younger than atime? If yes, update atime:
1394 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1398 * Is the previous atime value older than a day? If yes,
1401 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1404 * Good, we can skip the atime update:
1410 * touch_atime - update the access time
1411 * @mnt: mount the inode is accessed on
1412 * @dentry: dentry accessed
1414 * Update the accessed time on an inode and mark it for writeback.
1415 * This function automatically handles read only file systems and media,
1416 * as well as the "noatime" flag and inode specific "noatime" markers.
1418 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1420 struct inode *inode = dentry->d_inode;
1421 struct timespec now;
1423 if (inode->i_flags & S_NOATIME)
1425 if (IS_NOATIME(inode))
1427 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1430 if (mnt->mnt_flags & MNT_NOATIME)
1432 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1435 now = current_fs_time(inode->i_sb);
1437 if (!relatime_need_update(mnt, inode, now))
1440 if (timespec_equal(&inode->i_atime, &now))
1443 if (mnt_want_write(mnt))
1446 inode->i_atime = now;
1447 mark_inode_dirty_sync(inode);
1448 mnt_drop_write(mnt);
1450 EXPORT_SYMBOL(touch_atime);
1453 * file_update_time - update mtime and ctime time
1454 * @file: file accessed
1456 * Update the mtime and ctime members of an inode and mark the inode
1457 * for writeback. Note that this function is meant exclusively for
1458 * usage in the file write path of filesystems, and filesystems may
1459 * choose to explicitly ignore update via this function with the
1460 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1461 * timestamps are handled by the server.
1464 void file_update_time(struct file *file)
1466 struct inode *inode = file->f_path.dentry->d_inode;
1467 struct timespec now;
1468 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1470 /* First try to exhaust all avenues to not sync */
1471 if (IS_NOCMTIME(inode))
1474 now = current_fs_time(inode->i_sb);
1475 if (!timespec_equal(&inode->i_mtime, &now))
1478 if (!timespec_equal(&inode->i_ctime, &now))
1481 if (IS_I_VERSION(inode))
1482 sync_it |= S_VERSION;
1487 /* Finally allowed to write? Takes lock. */
1488 if (mnt_want_write_file(file))
1491 /* Only change inode inside the lock region */
1492 if (sync_it & S_VERSION)
1493 inode_inc_iversion(inode);
1494 if (sync_it & S_CTIME)
1495 inode->i_ctime = now;
1496 if (sync_it & S_MTIME)
1497 inode->i_mtime = now;
1498 mark_inode_dirty_sync(inode);
1499 mnt_drop_write(file->f_path.mnt);
1501 EXPORT_SYMBOL(file_update_time);
1503 int inode_needs_sync(struct inode *inode)
1507 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1511 EXPORT_SYMBOL(inode_needs_sync);
1513 int inode_wait(void *word)
1518 EXPORT_SYMBOL(inode_wait);
1521 * If we try to find an inode in the inode hash while it is being
1522 * deleted, we have to wait until the filesystem completes its
1523 * deletion before reporting that it isn't found. This function waits
1524 * until the deletion _might_ have completed. Callers are responsible
1525 * to recheck inode state.
1527 * It doesn't matter if I_NEW is not set initially, a call to
1528 * wake_up_inode() after removing from the hash list will DTRT.
1530 * This is called with inode_lock held.
1532 static void __wait_on_freeing_inode(struct inode *inode)
1534 wait_queue_head_t *wq;
1535 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1536 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1537 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1538 spin_unlock(&inode_lock);
1540 finish_wait(wq, &wait.wait);
1541 spin_lock(&inode_lock);
1544 static __initdata unsigned long ihash_entries;
1545 static int __init set_ihash_entries(char *str)
1549 ihash_entries = simple_strtoul(str, &str, 0);
1552 __setup("ihash_entries=", set_ihash_entries);
1555 * Initialize the waitqueues and inode hash table.
1557 void __init inode_init_early(void)
1561 /* If hashes are distributed across NUMA nodes, defer
1562 * hash allocation until vmalloc space is available.
1568 alloc_large_system_hash("Inode-cache",
1569 sizeof(struct hlist_head),
1577 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1578 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1581 void __init inode_init(void)
1585 /* inode slab cache */
1586 inode_cachep = kmem_cache_create("inode_cache",
1587 sizeof(struct inode),
1589 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1592 register_shrinker(&icache_shrinker);
1593 percpu_counter_init(&nr_inodes, 0);
1594 percpu_counter_init(&nr_inodes_unused, 0);
1596 /* Hash may have been set up in inode_init_early */
1601 alloc_large_system_hash("Inode-cache",
1602 sizeof(struct hlist_head),
1610 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1611 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1614 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1616 inode->i_mode = mode;
1617 if (S_ISCHR(mode)) {
1618 inode->i_fop = &def_chr_fops;
1619 inode->i_rdev = rdev;
1620 } else if (S_ISBLK(mode)) {
1621 inode->i_fop = &def_blk_fops;
1622 inode->i_rdev = rdev;
1623 } else if (S_ISFIFO(mode))
1624 inode->i_fop = &def_fifo_fops;
1625 else if (S_ISSOCK(mode))
1626 inode->i_fop = &bad_sock_fops;
1628 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1629 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1632 EXPORT_SYMBOL(init_special_inode);
1635 * Init uid,gid,mode for new inode according to posix standards
1637 * @dir: Directory inode
1638 * @mode: mode of the new inode
1640 void inode_init_owner(struct inode *inode, const struct inode *dir,
1643 inode->i_uid = current_fsuid();
1644 if (dir && dir->i_mode & S_ISGID) {
1645 inode->i_gid = dir->i_gid;
1649 inode->i_gid = current_fsgid();
1650 inode->i_mode = mode;
1652 EXPORT_SYMBOL(inode_init_owner);