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 inline void __inode_sb_list_add(struct inode *inode)
341 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
345 * inode_sb_list_add - add inode to the superblock list of inodes
346 * @inode: inode to add
348 void inode_sb_list_add(struct inode *inode)
350 spin_lock(&inode_lock);
351 __inode_sb_list_add(inode);
352 spin_unlock(&inode_lock);
354 EXPORT_SYMBOL_GPL(inode_sb_list_add);
356 static inline void __inode_sb_list_del(struct inode *inode)
358 list_del_init(&inode->i_sb_list);
361 static unsigned long hash(struct super_block *sb, unsigned long hashval)
365 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
367 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
368 return tmp & I_HASHMASK;
372 * __insert_inode_hash - hash an inode
373 * @inode: unhashed inode
374 * @hashval: unsigned long value used to locate this object in the
377 * Add an inode to the inode hash for this superblock.
379 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
381 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
383 spin_lock(&inode_lock);
384 hlist_add_head(&inode->i_hash, b);
385 spin_unlock(&inode_lock);
387 EXPORT_SYMBOL(__insert_inode_hash);
390 * __remove_inode_hash - remove an inode from the hash
391 * @inode: inode to unhash
393 * Remove an inode from the superblock.
395 static void __remove_inode_hash(struct inode *inode)
397 hlist_del_init(&inode->i_hash);
401 * remove_inode_hash - remove an inode from the hash
402 * @inode: inode to unhash
404 * Remove an inode from the superblock.
406 void remove_inode_hash(struct inode *inode)
408 spin_lock(&inode_lock);
409 hlist_del_init(&inode->i_hash);
410 spin_unlock(&inode_lock);
412 EXPORT_SYMBOL(remove_inode_hash);
414 void end_writeback(struct inode *inode)
417 BUG_ON(inode->i_data.nrpages);
418 BUG_ON(!list_empty(&inode->i_data.private_list));
419 BUG_ON(!(inode->i_state & I_FREEING));
420 BUG_ON(inode->i_state & I_CLEAR);
421 inode_sync_wait(inode);
422 inode->i_state = I_FREEING | I_CLEAR;
424 EXPORT_SYMBOL(end_writeback);
426 static void evict(struct inode *inode)
428 const struct super_operations *op = inode->i_sb->s_op;
430 if (op->evict_inode) {
431 op->evict_inode(inode);
433 if (inode->i_data.nrpages)
434 truncate_inode_pages(&inode->i_data, 0);
435 end_writeback(inode);
437 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
439 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
444 * dispose_list - dispose of the contents of a local list
445 * @head: the head of the list to free
447 * Dispose-list gets a local list with local inodes in it, so it doesn't
448 * need to worry about list corruption and SMP locks.
450 static void dispose_list(struct list_head *head)
452 while (!list_empty(head)) {
455 inode = list_first_entry(head, struct inode, i_list);
456 list_del_init(&inode->i_list);
460 spin_lock(&inode_lock);
461 __remove_inode_hash(inode);
462 __inode_sb_list_del(inode);
463 spin_unlock(&inode_lock);
465 wake_up_inode(inode);
466 destroy_inode(inode);
471 * Invalidate all inodes for a device.
473 static int invalidate_list(struct list_head *head, struct list_head *dispose)
475 struct list_head *next;
480 struct list_head *tmp = next;
484 * We can reschedule here without worrying about the list's
485 * consistency because the per-sb list of inodes must not
486 * change during umount anymore, and because iprune_sem keeps
487 * shrink_icache_memory() away.
489 cond_resched_lock(&inode_lock);
494 inode = list_entry(tmp, struct inode, i_sb_list);
495 if (inode->i_state & I_NEW)
497 invalidate_inode_buffers(inode);
498 if (!atomic_read(&inode->i_count)) {
499 list_move(&inode->i_list, dispose);
500 WARN_ON(inode->i_state & I_NEW);
501 inode->i_state |= I_FREEING;
502 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
503 percpu_counter_dec(&nr_inodes_unused);
512 * invalidate_inodes - discard the inodes on a device
515 * Discard all of the inodes for a given superblock. If the discard
516 * fails because there are busy inodes then a non zero value is returned.
517 * If the discard is successful all the inodes have been discarded.
519 int invalidate_inodes(struct super_block *sb)
522 LIST_HEAD(throw_away);
524 down_write(&iprune_sem);
525 spin_lock(&inode_lock);
526 fsnotify_unmount_inodes(&sb->s_inodes);
527 busy = invalidate_list(&sb->s_inodes, &throw_away);
528 spin_unlock(&inode_lock);
530 dispose_list(&throw_away);
531 up_write(&iprune_sem);
536 static int can_unuse(struct inode *inode)
538 if (inode->i_state & ~I_REFERENCED)
540 if (inode_has_buffers(inode))
542 if (atomic_read(&inode->i_count))
544 if (inode->i_data.nrpages)
550 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
551 * temporary list and then are freed outside inode_lock by dispose_list().
553 * Any inodes which are pinned purely because of attached pagecache have their
554 * pagecache removed. If the inode has metadata buffers attached to
555 * mapping->private_list then try to remove them.
557 * If the inode has the I_REFERENCED flag set, then it means that it has been
558 * used recently - the flag is set in iput_final(). When we encounter such an
559 * inode, clear the flag and move it to the back of the LRU so it gets another
560 * pass through the LRU before it gets reclaimed. This is necessary because of
561 * the fact we are doing lazy LRU updates to minimise lock contention so the
562 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
563 * with this flag set because they are the inodes that are out of order.
565 static void prune_icache(int nr_to_scan)
569 unsigned long reap = 0;
571 down_read(&iprune_sem);
572 spin_lock(&inode_lock);
573 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
576 if (list_empty(&inode_unused))
579 inode = list_entry(inode_unused.prev, struct inode, i_list);
582 * Referenced or dirty inodes are still in use. Give them
583 * another pass through the LRU as we canot reclaim them now.
585 if (atomic_read(&inode->i_count) ||
586 (inode->i_state & ~I_REFERENCED)) {
587 list_del_init(&inode->i_list);
588 percpu_counter_dec(&nr_inodes_unused);
592 /* recently referenced inodes get one more pass */
593 if (inode->i_state & I_REFERENCED) {
594 list_move(&inode->i_list, &inode_unused);
595 inode->i_state &= ~I_REFERENCED;
598 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
600 spin_unlock(&inode_lock);
601 if (remove_inode_buffers(inode))
602 reap += invalidate_mapping_pages(&inode->i_data,
605 spin_lock(&inode_lock);
607 if (inode != list_entry(inode_unused.next,
608 struct inode, i_list))
609 continue; /* wrong inode or list_empty */
610 if (!can_unuse(inode))
613 list_move(&inode->i_list, &freeable);
614 WARN_ON(inode->i_state & I_NEW);
615 inode->i_state |= I_FREEING;
616 percpu_counter_dec(&nr_inodes_unused);
618 if (current_is_kswapd())
619 __count_vm_events(KSWAPD_INODESTEAL, reap);
621 __count_vm_events(PGINODESTEAL, reap);
622 spin_unlock(&inode_lock);
624 dispose_list(&freeable);
625 up_read(&iprune_sem);
629 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
630 * "unused" means that no dentries are referring to the inodes: the files are
631 * not open and the dcache references to those inodes have already been
634 * This function is passed the number of inodes to scan, and it returns the
635 * total number of remaining possibly-reclaimable inodes.
637 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
641 * Nasty deadlock avoidance. We may hold various FS locks,
642 * and we don't want to recurse into the FS that called us
643 * in clear_inode() and friends..
645 if (!(gfp_mask & __GFP_FS))
649 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
652 static struct shrinker icache_shrinker = {
653 .shrink = shrink_icache_memory,
654 .seeks = DEFAULT_SEEKS,
657 static void __wait_on_freeing_inode(struct inode *inode);
659 * Called with the inode lock held.
661 static struct inode *find_inode(struct super_block *sb,
662 struct hlist_head *head,
663 int (*test)(struct inode *, void *),
666 struct hlist_node *node;
667 struct inode *inode = NULL;
670 hlist_for_each_entry(inode, node, head, i_hash) {
671 if (inode->i_sb != sb)
673 if (!test(inode, data))
675 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
676 __wait_on_freeing_inode(inode);
686 * find_inode_fast is the fast path version of find_inode, see the comment at
687 * iget_locked for details.
689 static struct inode *find_inode_fast(struct super_block *sb,
690 struct hlist_head *head, unsigned long ino)
692 struct hlist_node *node;
693 struct inode *inode = NULL;
696 hlist_for_each_entry(inode, node, head, i_hash) {
697 if (inode->i_ino != ino)
699 if (inode->i_sb != sb)
701 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
702 __wait_on_freeing_inode(inode);
712 * new_inode - obtain an inode
715 * Allocates a new inode for given superblock. The default gfp_mask
716 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
717 * If HIGHMEM pages are unsuitable or it is known that pages allocated
718 * for the page cache are not reclaimable or migratable,
719 * mapping_set_gfp_mask() must be called with suitable flags on the
720 * newly created inode's mapping
723 struct inode *new_inode(struct super_block *sb)
726 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
727 * error if st_ino won't fit in target struct field. Use 32bit counter
728 * here to attempt to avoid that.
730 static unsigned int last_ino;
733 spin_lock_prefetch(&inode_lock);
735 inode = alloc_inode(sb);
737 spin_lock(&inode_lock);
738 __inode_sb_list_add(inode);
739 inode->i_ino = ++last_ino;
741 spin_unlock(&inode_lock);
745 EXPORT_SYMBOL(new_inode);
747 void unlock_new_inode(struct inode *inode)
749 #ifdef CONFIG_DEBUG_LOCK_ALLOC
750 if (S_ISDIR(inode->i_mode)) {
751 struct file_system_type *type = inode->i_sb->s_type;
753 /* Set new key only if filesystem hasn't already changed it */
754 if (!lockdep_match_class(&inode->i_mutex,
755 &type->i_mutex_key)) {
757 * ensure nobody is actually holding i_mutex
759 mutex_destroy(&inode->i_mutex);
760 mutex_init(&inode->i_mutex);
761 lockdep_set_class(&inode->i_mutex,
762 &type->i_mutex_dir_key);
767 * This is special! We do not need the spinlock when clearing I_NEW,
768 * because we're guaranteed that nobody else tries to do anything about
769 * the state of the inode when it is locked, as we just created it (so
770 * there can be no old holders that haven't tested I_NEW).
771 * However we must emit the memory barrier so that other CPUs reliably
772 * see the clearing of I_NEW after the other inode initialisation has
776 WARN_ON(!(inode->i_state & I_NEW));
777 inode->i_state &= ~I_NEW;
778 wake_up_inode(inode);
780 EXPORT_SYMBOL(unlock_new_inode);
783 * This is called without the inode lock held.. Be careful.
785 * We no longer cache the sb_flags in i_flags - see fs.h
786 * -- rmk@arm.uk.linux.org
788 static struct inode *get_new_inode(struct super_block *sb,
789 struct hlist_head *head,
790 int (*test)(struct inode *, void *),
791 int (*set)(struct inode *, void *),
796 inode = alloc_inode(sb);
800 spin_lock(&inode_lock);
801 /* We released the lock, so.. */
802 old = find_inode(sb, head, test, data);
804 if (set(inode, data))
807 hlist_add_head(&inode->i_hash, head);
808 __inode_sb_list_add(inode);
809 inode->i_state = I_NEW;
810 spin_unlock(&inode_lock);
812 /* Return the locked inode with I_NEW set, the
813 * caller is responsible for filling in the contents
819 * Uhhuh, somebody else created the same inode under
820 * us. Use the old inode instead of the one we just
823 spin_unlock(&inode_lock);
824 destroy_inode(inode);
826 wait_on_inode(inode);
831 spin_unlock(&inode_lock);
832 destroy_inode(inode);
837 * get_new_inode_fast is the fast path version of get_new_inode, see the
838 * comment at iget_locked for details.
840 static struct inode *get_new_inode_fast(struct super_block *sb,
841 struct hlist_head *head, unsigned long ino)
845 inode = alloc_inode(sb);
849 spin_lock(&inode_lock);
850 /* We released the lock, so.. */
851 old = find_inode_fast(sb, head, ino);
854 hlist_add_head(&inode->i_hash, head);
855 __inode_sb_list_add(inode);
856 inode->i_state = I_NEW;
857 spin_unlock(&inode_lock);
859 /* Return the locked inode with I_NEW set, the
860 * caller is responsible for filling in the contents
866 * Uhhuh, somebody else created the same inode under
867 * us. Use the old inode instead of the one we just
870 spin_unlock(&inode_lock);
871 destroy_inode(inode);
873 wait_on_inode(inode);
879 * search the inode cache for a matching inode number.
880 * If we find one, then the inode number we are trying to
881 * allocate is not unique and so we should not use it.
883 * Returns 1 if the inode number is unique, 0 if it is not.
885 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
887 struct hlist_head *b = inode_hashtable + hash(sb, ino);
888 struct hlist_node *node;
891 hlist_for_each_entry(inode, node, b, i_hash) {
892 if (inode->i_ino == ino && inode->i_sb == sb)
900 * iunique - get a unique inode number
902 * @max_reserved: highest reserved inode number
904 * Obtain an inode number that is unique on the system for a given
905 * superblock. This is used by file systems that have no natural
906 * permanent inode numbering system. An inode number is returned that
907 * is higher than the reserved limit but unique.
910 * With a large number of inodes live on the file system this function
911 * currently becomes quite slow.
913 ino_t iunique(struct super_block *sb, ino_t max_reserved)
916 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
917 * error if st_ino won't fit in target struct field. Use 32bit counter
918 * here to attempt to avoid that.
920 static DEFINE_SPINLOCK(iunique_lock);
921 static unsigned int counter;
924 spin_lock(&inode_lock);
925 spin_lock(&iunique_lock);
927 if (counter <= max_reserved)
928 counter = max_reserved + 1;
930 } while (!test_inode_iunique(sb, res));
931 spin_unlock(&iunique_lock);
932 spin_unlock(&inode_lock);
936 EXPORT_SYMBOL(iunique);
938 struct inode *igrab(struct inode *inode)
940 spin_lock(&inode_lock);
941 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
945 * Handle the case where s_op->clear_inode is not been
946 * called yet, and somebody is calling igrab
947 * while the inode is getting freed.
950 spin_unlock(&inode_lock);
953 EXPORT_SYMBOL(igrab);
956 * ifind - internal function, you want ilookup5() or iget5().
957 * @sb: super block of file system to search
958 * @head: the head of the list to search
959 * @test: callback used for comparisons between inodes
960 * @data: opaque data pointer to pass to @test
961 * @wait: if true wait for the inode to be unlocked, if false do not
963 * ifind() searches for the inode specified by @data in the inode
964 * cache. This is a generalized version of ifind_fast() for file systems where
965 * the inode number is not sufficient for unique identification of an inode.
967 * If the inode is in the cache, the inode is returned with an incremented
970 * Otherwise NULL is returned.
972 * Note, @test is called with the inode_lock held, so can't sleep.
974 static struct inode *ifind(struct super_block *sb,
975 struct hlist_head *head, int (*test)(struct inode *, void *),
976 void *data, const int wait)
980 spin_lock(&inode_lock);
981 inode = find_inode(sb, head, test, data);
983 spin_unlock(&inode_lock);
985 wait_on_inode(inode);
988 spin_unlock(&inode_lock);
993 * ifind_fast - internal function, you want ilookup() or iget().
994 * @sb: super block of file system to search
995 * @head: head of the list to search
996 * @ino: inode number to search for
998 * ifind_fast() searches for the inode @ino in the inode cache. This is for
999 * file systems where the inode number is sufficient for unique identification
1002 * If the inode is in the cache, the inode is returned with an incremented
1005 * Otherwise NULL is returned.
1007 static struct inode *ifind_fast(struct super_block *sb,
1008 struct hlist_head *head, unsigned long ino)
1010 struct inode *inode;
1012 spin_lock(&inode_lock);
1013 inode = find_inode_fast(sb, head, ino);
1015 spin_unlock(&inode_lock);
1016 wait_on_inode(inode);
1019 spin_unlock(&inode_lock);
1024 * ilookup5_nowait - search for an inode in the inode cache
1025 * @sb: super block of file system to search
1026 * @hashval: hash value (usually inode number) to search for
1027 * @test: callback used for comparisons between inodes
1028 * @data: opaque data pointer to pass to @test
1030 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1031 * @data in the inode cache. This is a generalized version of ilookup() for
1032 * file systems where the inode number is not sufficient for unique
1033 * identification of an inode.
1035 * If the inode is in the cache, the inode is returned with an incremented
1036 * reference count. Note, the inode lock is not waited upon so you have to be
1037 * very careful what you do with the returned inode. You probably should be
1038 * using ilookup5() instead.
1040 * Otherwise NULL is returned.
1042 * Note, @test is called with the inode_lock held, so can't sleep.
1044 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1045 int (*test)(struct inode *, void *), void *data)
1047 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1049 return ifind(sb, head, test, data, 0);
1051 EXPORT_SYMBOL(ilookup5_nowait);
1054 * ilookup5 - search for an inode in the inode cache
1055 * @sb: super block of file system to search
1056 * @hashval: hash value (usually inode number) to search for
1057 * @test: callback used for comparisons between inodes
1058 * @data: opaque data pointer to pass to @test
1060 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1061 * @data in the inode cache. This is a generalized version of ilookup() for
1062 * file systems where the inode number is not sufficient for unique
1063 * identification of an inode.
1065 * If the inode is in the cache, the inode lock is waited upon and the inode is
1066 * returned with an incremented reference count.
1068 * Otherwise NULL is returned.
1070 * Note, @test is called with the inode_lock held, so can't sleep.
1072 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1073 int (*test)(struct inode *, void *), void *data)
1075 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1077 return ifind(sb, head, test, data, 1);
1079 EXPORT_SYMBOL(ilookup5);
1082 * ilookup - search for an inode in the inode cache
1083 * @sb: super block of file system to search
1084 * @ino: inode number to search for
1086 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1087 * This is for file systems where the inode number is sufficient for unique
1088 * identification of an inode.
1090 * If the inode is in the cache, the inode is returned with an incremented
1093 * Otherwise NULL is returned.
1095 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1097 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1099 return ifind_fast(sb, head, ino);
1101 EXPORT_SYMBOL(ilookup);
1104 * iget5_locked - obtain an inode from a mounted file system
1105 * @sb: super block of file system
1106 * @hashval: hash value (usually inode number) to get
1107 * @test: callback used for comparisons between inodes
1108 * @set: callback used to initialize a new struct inode
1109 * @data: opaque data pointer to pass to @test and @set
1111 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1112 * and @data in the inode cache and if present it is returned with an increased
1113 * reference count. This is a generalized version of iget_locked() for file
1114 * systems where the inode number is not sufficient for unique identification
1117 * If the inode is not in cache, get_new_inode() is called to allocate a new
1118 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1119 * file system gets to fill it in before unlocking it via unlock_new_inode().
1121 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1123 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1124 int (*test)(struct inode *, void *),
1125 int (*set)(struct inode *, void *), void *data)
1127 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1128 struct inode *inode;
1130 inode = ifind(sb, head, test, data, 1);
1134 * get_new_inode() will do the right thing, re-trying the search
1135 * in case it had to block at any point.
1137 return get_new_inode(sb, head, test, set, data);
1139 EXPORT_SYMBOL(iget5_locked);
1142 * iget_locked - obtain an inode from a mounted file system
1143 * @sb: super block of file system
1144 * @ino: inode number to get
1146 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1147 * the inode cache and if present it is returned with an increased reference
1148 * count. This is for file systems where the inode number is sufficient for
1149 * unique identification of an inode.
1151 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1152 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1153 * The file system gets to fill it in before unlocking it via
1154 * unlock_new_inode().
1156 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1158 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1159 struct inode *inode;
1161 inode = ifind_fast(sb, head, ino);
1165 * get_new_inode_fast() will do the right thing, re-trying the search
1166 * in case it had to block at any point.
1168 return get_new_inode_fast(sb, head, ino);
1170 EXPORT_SYMBOL(iget_locked);
1172 int insert_inode_locked(struct inode *inode)
1174 struct super_block *sb = inode->i_sb;
1175 ino_t ino = inode->i_ino;
1176 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1178 inode->i_state |= I_NEW;
1180 struct hlist_node *node;
1181 struct inode *old = NULL;
1182 spin_lock(&inode_lock);
1183 hlist_for_each_entry(old, node, head, i_hash) {
1184 if (old->i_ino != ino)
1186 if (old->i_sb != sb)
1188 if (old->i_state & (I_FREEING|I_WILL_FREE))
1192 if (likely(!node)) {
1193 hlist_add_head(&inode->i_hash, head);
1194 spin_unlock(&inode_lock);
1198 spin_unlock(&inode_lock);
1200 if (unlikely(!inode_unhashed(old))) {
1207 EXPORT_SYMBOL(insert_inode_locked);
1209 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1210 int (*test)(struct inode *, void *), void *data)
1212 struct super_block *sb = inode->i_sb;
1213 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1215 inode->i_state |= I_NEW;
1218 struct hlist_node *node;
1219 struct inode *old = NULL;
1221 spin_lock(&inode_lock);
1222 hlist_for_each_entry(old, node, head, i_hash) {
1223 if (old->i_sb != sb)
1225 if (!test(old, data))
1227 if (old->i_state & (I_FREEING|I_WILL_FREE))
1231 if (likely(!node)) {
1232 hlist_add_head(&inode->i_hash, head);
1233 spin_unlock(&inode_lock);
1237 spin_unlock(&inode_lock);
1239 if (unlikely(!inode_unhashed(old))) {
1246 EXPORT_SYMBOL(insert_inode_locked4);
1249 int generic_delete_inode(struct inode *inode)
1253 EXPORT_SYMBOL(generic_delete_inode);
1256 * Normal UNIX filesystem behaviour: delete the
1257 * inode when the usage count drops to zero, and
1260 int generic_drop_inode(struct inode *inode)
1262 return !inode->i_nlink || inode_unhashed(inode);
1264 EXPORT_SYMBOL_GPL(generic_drop_inode);
1267 * Called when we're dropping the last reference
1270 * Call the FS "drop_inode()" function, defaulting to
1271 * the legacy UNIX filesystem behaviour. If it tells
1272 * us to evict inode, do so. Otherwise, retain inode
1273 * in cache if fs is alive, sync and evict if fs is
1276 static void iput_final(struct inode *inode)
1278 struct super_block *sb = inode->i_sb;
1279 const struct super_operations *op = inode->i_sb->s_op;
1282 if (op && op->drop_inode)
1283 drop = op->drop_inode(inode);
1285 drop = generic_drop_inode(inode);
1288 if (sb->s_flags & MS_ACTIVE) {
1289 inode->i_state |= I_REFERENCED;
1290 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1291 inode_lru_list_add(inode);
1293 spin_unlock(&inode_lock);
1296 WARN_ON(inode->i_state & I_NEW);
1297 inode->i_state |= I_WILL_FREE;
1298 spin_unlock(&inode_lock);
1299 write_inode_now(inode, 1);
1300 spin_lock(&inode_lock);
1301 WARN_ON(inode->i_state & I_NEW);
1302 inode->i_state &= ~I_WILL_FREE;
1303 __remove_inode_hash(inode);
1305 WARN_ON(inode->i_state & I_NEW);
1306 inode->i_state |= I_FREEING;
1309 * After we delete the inode from the LRU here, we avoid moving dirty
1310 * inodes back onto the LRU now because I_FREEING is set and hence
1311 * writeback_single_inode() won't move the inode around.
1313 inode_lru_list_del(inode);
1315 __inode_sb_list_del(inode);
1316 spin_unlock(&inode_lock);
1318 remove_inode_hash(inode);
1319 wake_up_inode(inode);
1320 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1321 destroy_inode(inode);
1325 * iput - put an inode
1326 * @inode: inode to put
1328 * Puts an inode, dropping its usage count. If the inode use count hits
1329 * zero, the inode is then freed and may also be destroyed.
1331 * Consequently, iput() can sleep.
1333 void iput(struct inode *inode)
1336 BUG_ON(inode->i_state & I_CLEAR);
1338 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1342 EXPORT_SYMBOL(iput);
1345 * bmap - find a block number in a file
1346 * @inode: inode of file
1347 * @block: block to find
1349 * Returns the block number on the device holding the inode that
1350 * is the disk block number for the block of the file requested.
1351 * That is, asked for block 4 of inode 1 the function will return the
1352 * disk block relative to the disk start that holds that block of the
1355 sector_t bmap(struct inode *inode, sector_t block)
1358 if (inode->i_mapping->a_ops->bmap)
1359 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1362 EXPORT_SYMBOL(bmap);
1365 * With relative atime, only update atime if the previous atime is
1366 * earlier than either the ctime or mtime or if at least a day has
1367 * passed since the last atime update.
1369 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1370 struct timespec now)
1373 if (!(mnt->mnt_flags & MNT_RELATIME))
1376 * Is mtime younger than atime? If yes, update atime:
1378 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1381 * Is ctime younger than atime? If yes, update atime:
1383 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1387 * Is the previous atime value older than a day? If yes,
1390 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1393 * Good, we can skip the atime update:
1399 * touch_atime - update the access time
1400 * @mnt: mount the inode is accessed on
1401 * @dentry: dentry accessed
1403 * Update the accessed time on an inode and mark it for writeback.
1404 * This function automatically handles read only file systems and media,
1405 * as well as the "noatime" flag and inode specific "noatime" markers.
1407 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1409 struct inode *inode = dentry->d_inode;
1410 struct timespec now;
1412 if (inode->i_flags & S_NOATIME)
1414 if (IS_NOATIME(inode))
1416 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1419 if (mnt->mnt_flags & MNT_NOATIME)
1421 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1424 now = current_fs_time(inode->i_sb);
1426 if (!relatime_need_update(mnt, inode, now))
1429 if (timespec_equal(&inode->i_atime, &now))
1432 if (mnt_want_write(mnt))
1435 inode->i_atime = now;
1436 mark_inode_dirty_sync(inode);
1437 mnt_drop_write(mnt);
1439 EXPORT_SYMBOL(touch_atime);
1442 * file_update_time - update mtime and ctime time
1443 * @file: file accessed
1445 * Update the mtime and ctime members of an inode and mark the inode
1446 * for writeback. Note that this function is meant exclusively for
1447 * usage in the file write path of filesystems, and filesystems may
1448 * choose to explicitly ignore update via this function with the
1449 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1450 * timestamps are handled by the server.
1453 void file_update_time(struct file *file)
1455 struct inode *inode = file->f_path.dentry->d_inode;
1456 struct timespec now;
1457 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1459 /* First try to exhaust all avenues to not sync */
1460 if (IS_NOCMTIME(inode))
1463 now = current_fs_time(inode->i_sb);
1464 if (!timespec_equal(&inode->i_mtime, &now))
1467 if (!timespec_equal(&inode->i_ctime, &now))
1470 if (IS_I_VERSION(inode))
1471 sync_it |= S_VERSION;
1476 /* Finally allowed to write? Takes lock. */
1477 if (mnt_want_write_file(file))
1480 /* Only change inode inside the lock region */
1481 if (sync_it & S_VERSION)
1482 inode_inc_iversion(inode);
1483 if (sync_it & S_CTIME)
1484 inode->i_ctime = now;
1485 if (sync_it & S_MTIME)
1486 inode->i_mtime = now;
1487 mark_inode_dirty_sync(inode);
1488 mnt_drop_write(file->f_path.mnt);
1490 EXPORT_SYMBOL(file_update_time);
1492 int inode_needs_sync(struct inode *inode)
1496 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1500 EXPORT_SYMBOL(inode_needs_sync);
1502 int inode_wait(void *word)
1507 EXPORT_SYMBOL(inode_wait);
1510 * If we try to find an inode in the inode hash while it is being
1511 * deleted, we have to wait until the filesystem completes its
1512 * deletion before reporting that it isn't found. This function waits
1513 * until the deletion _might_ have completed. Callers are responsible
1514 * to recheck inode state.
1516 * It doesn't matter if I_NEW is not set initially, a call to
1517 * wake_up_inode() after removing from the hash list will DTRT.
1519 * This is called with inode_lock held.
1521 static void __wait_on_freeing_inode(struct inode *inode)
1523 wait_queue_head_t *wq;
1524 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1525 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1526 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1527 spin_unlock(&inode_lock);
1529 finish_wait(wq, &wait.wait);
1530 spin_lock(&inode_lock);
1533 static __initdata unsigned long ihash_entries;
1534 static int __init set_ihash_entries(char *str)
1538 ihash_entries = simple_strtoul(str, &str, 0);
1541 __setup("ihash_entries=", set_ihash_entries);
1544 * Initialize the waitqueues and inode hash table.
1546 void __init inode_init_early(void)
1550 /* If hashes are distributed across NUMA nodes, defer
1551 * hash allocation until vmalloc space is available.
1557 alloc_large_system_hash("Inode-cache",
1558 sizeof(struct hlist_head),
1566 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1567 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1570 void __init inode_init(void)
1574 /* inode slab cache */
1575 inode_cachep = kmem_cache_create("inode_cache",
1576 sizeof(struct inode),
1578 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1581 register_shrinker(&icache_shrinker);
1582 percpu_counter_init(&nr_inodes, 0);
1583 percpu_counter_init(&nr_inodes_unused, 0);
1585 /* Hash may have been set up in inode_init_early */
1590 alloc_large_system_hash("Inode-cache",
1591 sizeof(struct hlist_head),
1599 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1600 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1603 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1605 inode->i_mode = mode;
1606 if (S_ISCHR(mode)) {
1607 inode->i_fop = &def_chr_fops;
1608 inode->i_rdev = rdev;
1609 } else if (S_ISBLK(mode)) {
1610 inode->i_fop = &def_blk_fops;
1611 inode->i_rdev = rdev;
1612 } else if (S_ISFIFO(mode))
1613 inode->i_fop = &def_fifo_fops;
1614 else if (S_ISSOCK(mode))
1615 inode->i_fop = &bad_sock_fops;
1617 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1618 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1621 EXPORT_SYMBOL(init_special_inode);
1624 * Init uid,gid,mode for new inode according to posix standards
1626 * @dir: Directory inode
1627 * @mode: mode of the new inode
1629 void inode_init_owner(struct inode *inode, const struct inode *dir,
1632 inode->i_uid = current_fsuid();
1633 if (dir && dir->i_mode & S_ISGID) {
1634 inode->i_gid = dir->i_gid;
1638 inode->i_gid = current_fsgid();
1639 inode->i_mode = mode;
1641 EXPORT_SYMBOL(inode_init_owner);