4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/quotaops.h>
12 #include <linux/slab.h>
13 #include <linux/writeback.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/wait.h>
17 #include <linux/rwsem.h>
18 #include <linux/hash.h>
19 #include <linux/swap.h>
20 #include <linux/security.h>
21 #include <linux/pagemap.h>
22 #include <linux/cdev.h>
23 #include <linux/bootmem.h>
24 #include <linux/inotify.h>
25 #include <linux/fsnotify.h>
26 #include <linux/mount.h>
27 #include <linux/async.h>
28 #include <linux/posix_acl.h>
31 * This is needed for the following functions:
33 * - invalidate_inode_buffers
36 * FIXME: remove all knowledge of the buffer layer from this file
38 #include <linux/buffer_head.h>
41 * New inode.c implementation.
43 * This implementation has the basic premise of trying
44 * to be extremely low-overhead and SMP-safe, yet be
45 * simple enough to be "obviously correct".
50 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
52 /* #define INODE_PARANOIA 1 */
53 /* #define INODE_DEBUG 1 */
56 * Inode lookup is no longer as critical as it used to be:
57 * most of the lookups are going to be through the dcache.
59 #define I_HASHBITS i_hash_shift
60 #define I_HASHMASK i_hash_mask
62 static unsigned int i_hash_mask __read_mostly;
63 static unsigned int i_hash_shift __read_mostly;
66 * Each inode can be on two separate lists. One is
67 * the hash list of the inode, used for lookups. The
68 * other linked list is the "type" list:
69 * "in_use" - valid inode, i_count > 0, i_nlink > 0
70 * "dirty" - as "in_use" but also dirty
71 * "unused" - valid inode, i_count = 0
73 * A "dirty" list is maintained for each super block,
74 * allowing for low-overhead inode sync() operations.
77 LIST_HEAD(inode_in_use);
78 LIST_HEAD(inode_unused);
79 static struct hlist_head *inode_hashtable __read_mostly;
82 * A simple spinlock to protect the list manipulations.
84 * NOTE! You also have to own the lock if you change
85 * the i_state of an inode while it is in use..
87 DEFINE_SPINLOCK(inode_lock);
90 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
91 * icache shrinking path, and the umount path. Without this exclusion,
92 * by the time prune_icache calls iput for the inode whose pages it has
93 * been invalidating, or by the time it calls clear_inode & destroy_inode
94 * from its final dispose_list, the struct super_block they refer to
95 * (for inode->i_sb->s_op) may already have been freed and reused.
97 * We make this an rwsem because the fastpath is icache shrinking. In
98 * some cases a filesystem may be doing a significant amount of work in
99 * its inode reclaim code, so this should improve parallelism.
101 static DECLARE_RWSEM(iprune_sem);
104 * Statistics gathering..
106 struct inodes_stat_t inodes_stat;
108 static struct kmem_cache *inode_cachep __read_mostly;
110 static void wake_up_inode(struct inode *inode)
113 * Prevent speculative execution through spin_unlock(&inode_lock);
116 wake_up_bit(&inode->i_state, __I_NEW);
120 * inode_init_always - perform inode structure intialisation
121 * @sb: superblock inode belongs to
122 * @inode: inode to initialise
124 * These are initializations that need to be done on every inode
125 * allocation as the fields are not initialised by slab allocation.
127 int inode_init_always(struct super_block *sb, struct inode *inode)
129 static const struct address_space_operations empty_aops;
130 static const struct inode_operations empty_iops;
131 static const struct file_operations empty_fops;
132 struct address_space *const mapping = &inode->i_data;
135 inode->i_blkbits = sb->s_blocksize_bits;
137 atomic_set(&inode->i_count, 1);
138 inode->i_op = &empty_iops;
139 inode->i_fop = &empty_fops;
143 atomic_set(&inode->i_writecount, 0);
147 inode->i_generation = 0;
149 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
151 inode->i_pipe = NULL;
152 inode->i_bdev = NULL;
153 inode->i_cdev = NULL;
155 inode->dirtied_when = 0;
157 if (security_inode_alloc(inode))
159 spin_lock_init(&inode->i_lock);
160 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
162 mutex_init(&inode->i_mutex);
163 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
165 init_rwsem(&inode->i_alloc_sem);
166 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
168 mapping->a_ops = &empty_aops;
169 mapping->host = inode;
171 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
172 mapping->assoc_mapping = NULL;
173 mapping->backing_dev_info = &default_backing_dev_info;
174 mapping->writeback_index = 0;
177 * If the block_device provides a backing_dev_info for client
178 * inodes then use that. Otherwise the inode share the bdev's
182 struct backing_dev_info *bdi;
184 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
185 mapping->backing_dev_info = bdi;
187 inode->i_private = NULL;
188 inode->i_mapping = mapping;
189 #ifdef CONFIG_FS_POSIX_ACL
190 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
193 #ifdef CONFIG_FSNOTIFY
194 inode->i_fsnotify_mask = 0;
201 EXPORT_SYMBOL(inode_init_always);
203 static struct inode *alloc_inode(struct super_block *sb)
207 if (sb->s_op->alloc_inode)
208 inode = sb->s_op->alloc_inode(sb);
210 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
215 if (unlikely(inode_init_always(sb, inode))) {
216 if (inode->i_sb->s_op->destroy_inode)
217 inode->i_sb->s_op->destroy_inode(inode);
219 kmem_cache_free(inode_cachep, inode);
226 void __destroy_inode(struct inode *inode)
228 BUG_ON(inode_has_buffers(inode));
229 security_inode_free(inode);
230 fsnotify_inode_delete(inode);
231 #ifdef CONFIG_FS_POSIX_ACL
232 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
233 posix_acl_release(inode->i_acl);
234 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
235 posix_acl_release(inode->i_default_acl);
238 EXPORT_SYMBOL(__destroy_inode);
240 void destroy_inode(struct inode *inode)
242 __destroy_inode(inode);
243 if (inode->i_sb->s_op->destroy_inode)
244 inode->i_sb->s_op->destroy_inode(inode);
246 kmem_cache_free(inode_cachep, (inode));
250 * These are initializations that only need to be done
251 * once, because the fields are idempotent across use
252 * of the inode, so let the slab aware of that.
254 void inode_init_once(struct inode *inode)
256 memset(inode, 0, sizeof(*inode));
257 INIT_HLIST_NODE(&inode->i_hash);
258 INIT_LIST_HEAD(&inode->i_dentry);
259 INIT_LIST_HEAD(&inode->i_devices);
260 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
261 spin_lock_init(&inode->i_data.tree_lock);
262 spin_lock_init(&inode->i_data.i_mmap_lock);
263 INIT_LIST_HEAD(&inode->i_data.private_list);
264 spin_lock_init(&inode->i_data.private_lock);
265 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
266 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
267 i_size_ordered_init(inode);
268 #ifdef CONFIG_INOTIFY
269 INIT_LIST_HEAD(&inode->inotify_watches);
270 mutex_init(&inode->inotify_mutex);
272 #ifdef CONFIG_FSNOTIFY
273 INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries);
276 EXPORT_SYMBOL(inode_init_once);
278 static void init_once(void *foo)
280 struct inode *inode = (struct inode *) foo;
282 inode_init_once(inode);
286 * inode_lock must be held
288 void __iget(struct inode *inode)
290 if (atomic_read(&inode->i_count)) {
291 atomic_inc(&inode->i_count);
294 atomic_inc(&inode->i_count);
295 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
296 list_move(&inode->i_list, &inode_in_use);
297 inodes_stat.nr_unused--;
301 * clear_inode - clear an inode
302 * @inode: inode to clear
304 * This is called by the filesystem to tell us
305 * that the inode is no longer useful. We just
306 * terminate it with extreme prejudice.
308 void clear_inode(struct inode *inode)
311 invalidate_inode_buffers(inode);
313 BUG_ON(inode->i_data.nrpages);
314 BUG_ON(!(inode->i_state & I_FREEING));
315 BUG_ON(inode->i_state & I_CLEAR);
316 inode_sync_wait(inode);
317 if (inode->i_sb->s_op->clear_inode)
318 inode->i_sb->s_op->clear_inode(inode);
319 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
321 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
323 inode->i_state = I_CLEAR;
325 EXPORT_SYMBOL(clear_inode);
328 * dispose_list - dispose of the contents of a local list
329 * @head: the head of the list to free
331 * Dispose-list gets a local list with local inodes in it, so it doesn't
332 * need to worry about list corruption and SMP locks.
334 static void dispose_list(struct list_head *head)
338 while (!list_empty(head)) {
341 inode = list_first_entry(head, struct inode, i_list);
342 list_del(&inode->i_list);
344 if (inode->i_data.nrpages)
345 truncate_inode_pages(&inode->i_data, 0);
348 spin_lock(&inode_lock);
349 hlist_del_init(&inode->i_hash);
350 list_del_init(&inode->i_sb_list);
351 spin_unlock(&inode_lock);
353 wake_up_inode(inode);
354 destroy_inode(inode);
357 spin_lock(&inode_lock);
358 inodes_stat.nr_inodes -= nr_disposed;
359 spin_unlock(&inode_lock);
363 * Invalidate all inodes for a device.
365 static int invalidate_list(struct list_head *head, struct list_head *dispose)
367 struct list_head *next;
368 int busy = 0, count = 0;
372 struct list_head *tmp = next;
376 * We can reschedule here without worrying about the list's
377 * consistency because the per-sb list of inodes must not
378 * change during umount anymore, and because iprune_sem keeps
379 * shrink_icache_memory() away.
381 cond_resched_lock(&inode_lock);
386 inode = list_entry(tmp, struct inode, i_sb_list);
387 if (inode->i_state & I_NEW)
389 invalidate_inode_buffers(inode);
390 if (!atomic_read(&inode->i_count)) {
391 list_move(&inode->i_list, dispose);
392 WARN_ON(inode->i_state & I_NEW);
393 inode->i_state |= I_FREEING;
399 /* only unused inodes may be cached with i_count zero */
400 inodes_stat.nr_unused -= count;
405 * invalidate_inodes - discard the inodes on a device
408 * Discard all of the inodes for a given superblock. If the discard
409 * fails because there are busy inodes then a non zero value is returned.
410 * If the discard is successful all the inodes have been discarded.
412 int invalidate_inodes(struct super_block *sb)
415 LIST_HEAD(throw_away);
417 down_write(&iprune_sem);
418 spin_lock(&inode_lock);
419 inotify_unmount_inodes(&sb->s_inodes);
420 fsnotify_unmount_inodes(&sb->s_inodes);
421 busy = invalidate_list(&sb->s_inodes, &throw_away);
422 spin_unlock(&inode_lock);
424 dispose_list(&throw_away);
425 up_write(&iprune_sem);
429 EXPORT_SYMBOL(invalidate_inodes);
431 static int can_unuse(struct inode *inode)
435 if (inode_has_buffers(inode))
437 if (atomic_read(&inode->i_count))
439 if (inode->i_data.nrpages)
445 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
446 * a temporary list and then are freed outside inode_lock by dispose_list().
448 * Any inodes which are pinned purely because of attached pagecache have their
449 * pagecache removed. We expect the final iput() on that inode to add it to
450 * the front of the inode_unused list. So look for it there and if the
451 * inode is still freeable, proceed. The right inode is found 99.9% of the
452 * time in testing on a 4-way.
454 * If the inode has metadata buffers attached to mapping->private_list then
455 * try to remove them.
457 static void prune_icache(int nr_to_scan)
462 unsigned long reap = 0;
464 down_read(&iprune_sem);
465 spin_lock(&inode_lock);
466 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
469 if (list_empty(&inode_unused))
472 inode = list_entry(inode_unused.prev, struct inode, i_list);
474 if (inode->i_state || atomic_read(&inode->i_count)) {
475 list_move(&inode->i_list, &inode_unused);
478 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
480 spin_unlock(&inode_lock);
481 if (remove_inode_buffers(inode))
482 reap += invalidate_mapping_pages(&inode->i_data,
485 spin_lock(&inode_lock);
487 if (inode != list_entry(inode_unused.next,
488 struct inode, i_list))
489 continue; /* wrong inode or list_empty */
490 if (!can_unuse(inode))
493 list_move(&inode->i_list, &freeable);
494 WARN_ON(inode->i_state & I_NEW);
495 inode->i_state |= I_FREEING;
498 inodes_stat.nr_unused -= nr_pruned;
499 if (current_is_kswapd())
500 __count_vm_events(KSWAPD_INODESTEAL, reap);
502 __count_vm_events(PGINODESTEAL, reap);
503 spin_unlock(&inode_lock);
505 dispose_list(&freeable);
506 up_read(&iprune_sem);
510 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
511 * "unused" means that no dentries are referring to the inodes: the files are
512 * not open and the dcache references to those inodes have already been
515 * This function is passed the number of inodes to scan, and it returns the
516 * total number of remaining possibly-reclaimable inodes.
518 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
522 * Nasty deadlock avoidance. We may hold various FS locks,
523 * and we don't want to recurse into the FS that called us
524 * in clear_inode() and friends..
526 if (!(gfp_mask & __GFP_FS))
530 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
533 static struct shrinker icache_shrinker = {
534 .shrink = shrink_icache_memory,
535 .seeks = DEFAULT_SEEKS,
538 static void __wait_on_freeing_inode(struct inode *inode);
540 * Called with the inode lock held.
541 * NOTE: we are not increasing the inode-refcount, you must call __iget()
542 * by hand after calling find_inode now! This simplifies iunique and won't
543 * add any additional branch in the common code.
545 static struct inode *find_inode(struct super_block *sb,
546 struct hlist_head *head,
547 int (*test)(struct inode *, void *),
550 struct hlist_node *node;
551 struct inode *inode = NULL;
554 hlist_for_each_entry(inode, node, head, i_hash) {
555 if (inode->i_sb != sb)
557 if (!test(inode, data))
559 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
560 __wait_on_freeing_inode(inode);
565 return node ? inode : NULL;
569 * find_inode_fast is the fast path version of find_inode, see the comment at
570 * iget_locked for details.
572 static struct inode *find_inode_fast(struct super_block *sb,
573 struct hlist_head *head, unsigned long ino)
575 struct hlist_node *node;
576 struct inode *inode = NULL;
579 hlist_for_each_entry(inode, node, head, i_hash) {
580 if (inode->i_ino != ino)
582 if (inode->i_sb != sb)
584 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
585 __wait_on_freeing_inode(inode);
590 return node ? inode : NULL;
593 static unsigned long hash(struct super_block *sb, unsigned long hashval)
597 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
599 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
600 return tmp & I_HASHMASK;
604 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
607 inodes_stat.nr_inodes++;
608 list_add(&inode->i_list, &inode_in_use);
609 list_add(&inode->i_sb_list, &sb->s_inodes);
611 hlist_add_head(&inode->i_hash, head);
615 * inode_add_to_lists - add a new inode to relevant lists
616 * @sb: superblock inode belongs to
617 * @inode: inode to mark in use
619 * When an inode is allocated it needs to be accounted for, added to the in use
620 * list, the owning superblock and the inode hash. This needs to be done under
621 * the inode_lock, so export a function to do this rather than the inode lock
622 * itself. We calculate the hash list to add to here so it is all internal
623 * which requires the caller to have already set up the inode number in the
626 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
628 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
630 spin_lock(&inode_lock);
631 __inode_add_to_lists(sb, head, inode);
632 spin_unlock(&inode_lock);
634 EXPORT_SYMBOL_GPL(inode_add_to_lists);
637 * new_inode - obtain an inode
640 * Allocates a new inode for given superblock. The default gfp_mask
641 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
642 * If HIGHMEM pages are unsuitable or it is known that pages allocated
643 * for the page cache are not reclaimable or migratable,
644 * mapping_set_gfp_mask() must be called with suitable flags on the
645 * newly created inode's mapping
648 struct inode *new_inode(struct super_block *sb)
651 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
652 * error if st_ino won't fit in target struct field. Use 32bit counter
653 * here to attempt to avoid that.
655 static unsigned int last_ino;
658 spin_lock_prefetch(&inode_lock);
660 inode = alloc_inode(sb);
662 spin_lock(&inode_lock);
663 __inode_add_to_lists(sb, NULL, inode);
664 inode->i_ino = ++last_ino;
666 spin_unlock(&inode_lock);
670 EXPORT_SYMBOL(new_inode);
672 void unlock_new_inode(struct inode *inode)
674 #ifdef CONFIG_DEBUG_LOCK_ALLOC
675 if (inode->i_mode & S_IFDIR) {
676 struct file_system_type *type = inode->i_sb->s_type;
678 /* Set new key only if filesystem hasn't already changed it */
679 if (!lockdep_match_class(&inode->i_mutex,
680 &type->i_mutex_key)) {
682 * ensure nobody is actually holding i_mutex
684 mutex_destroy(&inode->i_mutex);
685 mutex_init(&inode->i_mutex);
686 lockdep_set_class(&inode->i_mutex,
687 &type->i_mutex_dir_key);
692 * This is special! We do not need the spinlock when clearing I_NEW,
693 * because we're guaranteed that nobody else tries to do anything about
694 * the state of the inode when it is locked, as we just created it (so
695 * there can be no old holders that haven't tested I_NEW).
696 * However we must emit the memory barrier so that other CPUs reliably
697 * see the clearing of I_NEW after the other inode initialisation has
701 WARN_ON(!(inode->i_state & I_NEW));
702 inode->i_state &= ~I_NEW;
703 wake_up_inode(inode);
705 EXPORT_SYMBOL(unlock_new_inode);
708 * This is called without the inode lock held.. Be careful.
710 * We no longer cache the sb_flags in i_flags - see fs.h
711 * -- rmk@arm.uk.linux.org
713 static struct inode *get_new_inode(struct super_block *sb,
714 struct hlist_head *head,
715 int (*test)(struct inode *, void *),
716 int (*set)(struct inode *, void *),
721 inode = alloc_inode(sb);
725 spin_lock(&inode_lock);
726 /* We released the lock, so.. */
727 old = find_inode(sb, head, test, data);
729 if (set(inode, data))
732 __inode_add_to_lists(sb, head, inode);
733 inode->i_state = I_NEW;
734 spin_unlock(&inode_lock);
736 /* Return the locked inode with I_NEW set, the
737 * caller is responsible for filling in the contents
743 * Uhhuh, somebody else created the same inode under
744 * us. Use the old inode instead of the one we just
748 spin_unlock(&inode_lock);
749 destroy_inode(inode);
751 wait_on_inode(inode);
756 spin_unlock(&inode_lock);
757 destroy_inode(inode);
762 * get_new_inode_fast is the fast path version of get_new_inode, see the
763 * comment at iget_locked for details.
765 static struct inode *get_new_inode_fast(struct super_block *sb,
766 struct hlist_head *head, unsigned long ino)
770 inode = alloc_inode(sb);
774 spin_lock(&inode_lock);
775 /* We released the lock, so.. */
776 old = find_inode_fast(sb, head, ino);
779 __inode_add_to_lists(sb, head, inode);
780 inode->i_state = I_NEW;
781 spin_unlock(&inode_lock);
783 /* Return the locked inode with I_NEW set, the
784 * caller is responsible for filling in the contents
790 * Uhhuh, somebody else created the same inode under
791 * us. Use the old inode instead of the one we just
795 spin_unlock(&inode_lock);
796 destroy_inode(inode);
798 wait_on_inode(inode);
804 * iunique - get a unique inode number
806 * @max_reserved: highest reserved inode number
808 * Obtain an inode number that is unique on the system for a given
809 * superblock. This is used by file systems that have no natural
810 * permanent inode numbering system. An inode number is returned that
811 * is higher than the reserved limit but unique.
814 * With a large number of inodes live on the file system this function
815 * currently becomes quite slow.
817 ino_t iunique(struct super_block *sb, ino_t max_reserved)
820 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
821 * error if st_ino won't fit in target struct field. Use 32bit counter
822 * here to attempt to avoid that.
824 static unsigned int counter;
826 struct hlist_head *head;
829 spin_lock(&inode_lock);
831 if (counter <= max_reserved)
832 counter = max_reserved + 1;
834 head = inode_hashtable + hash(sb, res);
835 inode = find_inode_fast(sb, head, res);
836 } while (inode != NULL);
837 spin_unlock(&inode_lock);
841 EXPORT_SYMBOL(iunique);
843 struct inode *igrab(struct inode *inode)
845 spin_lock(&inode_lock);
846 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
850 * Handle the case where s_op->clear_inode is not been
851 * called yet, and somebody is calling igrab
852 * while the inode is getting freed.
855 spin_unlock(&inode_lock);
858 EXPORT_SYMBOL(igrab);
861 * ifind - internal function, you want ilookup5() or iget5().
862 * @sb: super block of file system to search
863 * @head: the head of the list to search
864 * @test: callback used for comparisons between inodes
865 * @data: opaque data pointer to pass to @test
866 * @wait: if true wait for the inode to be unlocked, if false do not
868 * ifind() searches for the inode specified by @data in the inode
869 * cache. This is a generalized version of ifind_fast() for file systems where
870 * the inode number is not sufficient for unique identification of an inode.
872 * If the inode is in the cache, the inode is returned with an incremented
875 * Otherwise NULL is returned.
877 * Note, @test is called with the inode_lock held, so can't sleep.
879 static struct inode *ifind(struct super_block *sb,
880 struct hlist_head *head, int (*test)(struct inode *, void *),
881 void *data, const int wait)
885 spin_lock(&inode_lock);
886 inode = find_inode(sb, head, test, data);
889 spin_unlock(&inode_lock);
891 wait_on_inode(inode);
894 spin_unlock(&inode_lock);
899 * ifind_fast - internal function, you want ilookup() or iget().
900 * @sb: super block of file system to search
901 * @head: head of the list to search
902 * @ino: inode number to search for
904 * ifind_fast() searches for the inode @ino in the inode cache. This is for
905 * file systems where the inode number is sufficient for unique identification
908 * If the inode is in the cache, the inode is returned with an incremented
911 * Otherwise NULL is returned.
913 static struct inode *ifind_fast(struct super_block *sb,
914 struct hlist_head *head, unsigned long ino)
918 spin_lock(&inode_lock);
919 inode = find_inode_fast(sb, head, ino);
922 spin_unlock(&inode_lock);
923 wait_on_inode(inode);
926 spin_unlock(&inode_lock);
931 * ilookup5_nowait - search for an inode in the inode cache
932 * @sb: super block of file system to search
933 * @hashval: hash value (usually inode number) to search for
934 * @test: callback used for comparisons between inodes
935 * @data: opaque data pointer to pass to @test
937 * ilookup5() uses ifind() to search for the inode specified by @hashval and
938 * @data in the inode cache. This is a generalized version of ilookup() for
939 * file systems where the inode number is not sufficient for unique
940 * identification of an inode.
942 * If the inode is in the cache, the inode is returned with an incremented
943 * reference count. Note, the inode lock is not waited upon so you have to be
944 * very careful what you do with the returned inode. You probably should be
945 * using ilookup5() instead.
947 * Otherwise NULL is returned.
949 * Note, @test is called with the inode_lock held, so can't sleep.
951 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
952 int (*test)(struct inode *, void *), void *data)
954 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
956 return ifind(sb, head, test, data, 0);
958 EXPORT_SYMBOL(ilookup5_nowait);
961 * ilookup5 - search for an inode in the inode cache
962 * @sb: super block of file system to search
963 * @hashval: hash value (usually inode number) to search for
964 * @test: callback used for comparisons between inodes
965 * @data: opaque data pointer to pass to @test
967 * ilookup5() uses ifind() to search for the inode specified by @hashval and
968 * @data in the inode cache. This is a generalized version of ilookup() for
969 * file systems where the inode number is not sufficient for unique
970 * identification of an inode.
972 * If the inode is in the cache, the inode lock is waited upon and the inode is
973 * returned with an incremented reference count.
975 * Otherwise NULL is returned.
977 * Note, @test is called with the inode_lock held, so can't sleep.
979 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
980 int (*test)(struct inode *, void *), void *data)
982 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
984 return ifind(sb, head, test, data, 1);
986 EXPORT_SYMBOL(ilookup5);
989 * ilookup - search for an inode in the inode cache
990 * @sb: super block of file system to search
991 * @ino: inode number to search for
993 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
994 * This is for file systems where the inode number is sufficient for unique
995 * identification of an inode.
997 * If the inode is in the cache, the inode is returned with an incremented
1000 * Otherwise NULL is returned.
1002 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1004 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1006 return ifind_fast(sb, head, ino);
1008 EXPORT_SYMBOL(ilookup);
1011 * iget5_locked - obtain an inode from a mounted file system
1012 * @sb: super block of file system
1013 * @hashval: hash value (usually inode number) to get
1014 * @test: callback used for comparisons between inodes
1015 * @set: callback used to initialize a new struct inode
1016 * @data: opaque data pointer to pass to @test and @set
1018 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1019 * and @data in the inode cache and if present it is returned with an increased
1020 * reference count. This is a generalized version of iget_locked() for file
1021 * systems where the inode number is not sufficient for unique identification
1024 * If the inode is not in cache, get_new_inode() is called to allocate a new
1025 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1026 * file system gets to fill it in before unlocking it via unlock_new_inode().
1028 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1030 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1031 int (*test)(struct inode *, void *),
1032 int (*set)(struct inode *, void *), void *data)
1034 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1035 struct inode *inode;
1037 inode = ifind(sb, head, test, data, 1);
1041 * get_new_inode() will do the right thing, re-trying the search
1042 * in case it had to block at any point.
1044 return get_new_inode(sb, head, test, set, data);
1046 EXPORT_SYMBOL(iget5_locked);
1049 * iget_locked - obtain an inode from a mounted file system
1050 * @sb: super block of file system
1051 * @ino: inode number to get
1053 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1054 * the inode cache and if present it is returned with an increased reference
1055 * count. This is for file systems where the inode number is sufficient for
1056 * unique identification of an inode.
1058 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1059 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1060 * The file system gets to fill it in before unlocking it via
1061 * unlock_new_inode().
1063 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1065 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1066 struct inode *inode;
1068 inode = ifind_fast(sb, head, ino);
1072 * get_new_inode_fast() will do the right thing, re-trying the search
1073 * in case it had to block at any point.
1075 return get_new_inode_fast(sb, head, ino);
1077 EXPORT_SYMBOL(iget_locked);
1079 int insert_inode_locked(struct inode *inode)
1081 struct super_block *sb = inode->i_sb;
1082 ino_t ino = inode->i_ino;
1083 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1085 inode->i_state |= I_NEW;
1087 struct hlist_node *node;
1088 struct inode *old = NULL;
1089 spin_lock(&inode_lock);
1090 hlist_for_each_entry(old, node, head, i_hash) {
1091 if (old->i_ino != ino)
1093 if (old->i_sb != sb)
1095 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1099 if (likely(!node)) {
1100 hlist_add_head(&inode->i_hash, head);
1101 spin_unlock(&inode_lock);
1105 spin_unlock(&inode_lock);
1107 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1114 EXPORT_SYMBOL(insert_inode_locked);
1116 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1117 int (*test)(struct inode *, void *), void *data)
1119 struct super_block *sb = inode->i_sb;
1120 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1122 inode->i_state |= I_NEW;
1125 struct hlist_node *node;
1126 struct inode *old = NULL;
1128 spin_lock(&inode_lock);
1129 hlist_for_each_entry(old, node, head, i_hash) {
1130 if (old->i_sb != sb)
1132 if (!test(old, data))
1134 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1138 if (likely(!node)) {
1139 hlist_add_head(&inode->i_hash, head);
1140 spin_unlock(&inode_lock);
1144 spin_unlock(&inode_lock);
1146 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1153 EXPORT_SYMBOL(insert_inode_locked4);
1156 * __insert_inode_hash - hash an inode
1157 * @inode: unhashed inode
1158 * @hashval: unsigned long value used to locate this object in the
1161 * Add an inode to the inode hash for this superblock.
1163 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1165 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1166 spin_lock(&inode_lock);
1167 hlist_add_head(&inode->i_hash, head);
1168 spin_unlock(&inode_lock);
1170 EXPORT_SYMBOL(__insert_inode_hash);
1173 * remove_inode_hash - remove an inode from the hash
1174 * @inode: inode to unhash
1176 * Remove an inode from the superblock.
1178 void remove_inode_hash(struct inode *inode)
1180 spin_lock(&inode_lock);
1181 hlist_del_init(&inode->i_hash);
1182 spin_unlock(&inode_lock);
1184 EXPORT_SYMBOL(remove_inode_hash);
1187 * Tell the filesystem that this inode is no longer of any interest and should
1188 * be completely destroyed.
1190 * We leave the inode in the inode hash table until *after* the filesystem's
1191 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1192 * instigate) will always find up-to-date information either in the hash or on
1195 * I_FREEING is set so that no-one will take a new reference to the inode while
1196 * it is being deleted.
1198 void generic_delete_inode(struct inode *inode)
1200 const struct super_operations *op = inode->i_sb->s_op;
1202 list_del_init(&inode->i_list);
1203 list_del_init(&inode->i_sb_list);
1204 WARN_ON(inode->i_state & I_NEW);
1205 inode->i_state |= I_FREEING;
1206 inodes_stat.nr_inodes--;
1207 spin_unlock(&inode_lock);
1209 security_inode_delete(inode);
1211 if (op->delete_inode) {
1212 void (*delete)(struct inode *) = op->delete_inode;
1213 if (!is_bad_inode(inode))
1215 /* Filesystems implementing their own
1216 * s_op->delete_inode are required to call
1217 * truncate_inode_pages and clear_inode()
1221 truncate_inode_pages(&inode->i_data, 0);
1224 spin_lock(&inode_lock);
1225 hlist_del_init(&inode->i_hash);
1226 spin_unlock(&inode_lock);
1227 wake_up_inode(inode);
1228 BUG_ON(inode->i_state != I_CLEAR);
1229 destroy_inode(inode);
1231 EXPORT_SYMBOL(generic_delete_inode);
1234 * generic_detach_inode - remove inode from inode lists
1235 * @inode: inode to remove
1237 * Remove inode from inode lists, write it if it's dirty. This is just an
1238 * internal VFS helper exported for hugetlbfs. Do not use!
1240 * Returns 1 if inode should be completely destroyed.
1242 int generic_detach_inode(struct inode *inode)
1244 struct super_block *sb = inode->i_sb;
1246 if (!hlist_unhashed(&inode->i_hash)) {
1247 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1248 list_move(&inode->i_list, &inode_unused);
1249 inodes_stat.nr_unused++;
1250 if (sb->s_flags & MS_ACTIVE) {
1251 spin_unlock(&inode_lock);
1254 WARN_ON(inode->i_state & I_NEW);
1255 inode->i_state |= I_WILL_FREE;
1256 spin_unlock(&inode_lock);
1257 write_inode_now(inode, 1);
1258 spin_lock(&inode_lock);
1259 WARN_ON(inode->i_state & I_NEW);
1260 inode->i_state &= ~I_WILL_FREE;
1261 inodes_stat.nr_unused--;
1262 hlist_del_init(&inode->i_hash);
1264 list_del_init(&inode->i_list);
1265 list_del_init(&inode->i_sb_list);
1266 WARN_ON(inode->i_state & I_NEW);
1267 inode->i_state |= I_FREEING;
1268 inodes_stat.nr_inodes--;
1269 spin_unlock(&inode_lock);
1272 EXPORT_SYMBOL_GPL(generic_detach_inode);
1274 static void generic_forget_inode(struct inode *inode)
1276 if (!generic_detach_inode(inode))
1278 if (inode->i_data.nrpages)
1279 truncate_inode_pages(&inode->i_data, 0);
1281 wake_up_inode(inode);
1282 destroy_inode(inode);
1286 * Normal UNIX filesystem behaviour: delete the
1287 * inode when the usage count drops to zero, and
1290 void generic_drop_inode(struct inode *inode)
1292 if (!inode->i_nlink)
1293 generic_delete_inode(inode);
1295 generic_forget_inode(inode);
1297 EXPORT_SYMBOL_GPL(generic_drop_inode);
1300 * Called when we're dropping the last reference
1303 * Call the FS "drop()" function, defaulting to
1304 * the legacy UNIX filesystem behaviour..
1306 * NOTE! NOTE! NOTE! We're called with the inode lock
1307 * held, and the drop function is supposed to release
1310 static inline void iput_final(struct inode *inode)
1312 const struct super_operations *op = inode->i_sb->s_op;
1313 void (*drop)(struct inode *) = generic_drop_inode;
1315 if (op && op->drop_inode)
1316 drop = op->drop_inode;
1321 * iput - put an inode
1322 * @inode: inode to put
1324 * Puts an inode, dropping its usage count. If the inode use count hits
1325 * zero, the inode is then freed and may also be destroyed.
1327 * Consequently, iput() can sleep.
1329 void iput(struct inode *inode)
1332 BUG_ON(inode->i_state == I_CLEAR);
1334 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1338 EXPORT_SYMBOL(iput);
1341 * bmap - find a block number in a file
1342 * @inode: inode of file
1343 * @block: block to find
1345 * Returns the block number on the device holding the inode that
1346 * is the disk block number for the block of the file requested.
1347 * That is, asked for block 4 of inode 1 the function will return the
1348 * disk block relative to the disk start that holds that block of the
1351 sector_t bmap(struct inode *inode, sector_t block)
1354 if (inode->i_mapping->a_ops->bmap)
1355 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1358 EXPORT_SYMBOL(bmap);
1361 * With relative atime, only update atime if the previous atime is
1362 * earlier than either the ctime or mtime or if at least a day has
1363 * passed since the last atime update.
1365 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1366 struct timespec now)
1369 if (!(mnt->mnt_flags & MNT_RELATIME))
1372 * Is mtime younger than atime? If yes, update atime:
1374 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1377 * Is ctime younger than atime? If yes, update atime:
1379 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1383 * Is the previous atime value older than a day? If yes,
1386 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1389 * Good, we can skip the atime update:
1395 * touch_atime - update the access time
1396 * @mnt: mount the inode is accessed on
1397 * @dentry: dentry accessed
1399 * Update the accessed time on an inode and mark it for writeback.
1400 * This function automatically handles read only file systems and media,
1401 * as well as the "noatime" flag and inode specific "noatime" markers.
1403 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1405 struct inode *inode = dentry->d_inode;
1406 struct timespec now;
1408 if (inode->i_flags & S_NOATIME)
1410 if (IS_NOATIME(inode))
1412 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1415 if (mnt->mnt_flags & MNT_NOATIME)
1417 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1420 now = current_fs_time(inode->i_sb);
1422 if (!relatime_need_update(mnt, inode, now))
1425 if (timespec_equal(&inode->i_atime, &now))
1428 if (mnt_want_write(mnt))
1431 inode->i_atime = now;
1432 mark_inode_dirty_sync(inode);
1433 mnt_drop_write(mnt);
1435 EXPORT_SYMBOL(touch_atime);
1438 * file_update_time - update mtime and ctime time
1439 * @file: file accessed
1441 * Update the mtime and ctime members of an inode and mark the inode
1442 * for writeback. Note that this function is meant exclusively for
1443 * usage in the file write path of filesystems, and filesystems may
1444 * choose to explicitly ignore update via this function with the
1445 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1446 * timestamps are handled by the server.
1449 void file_update_time(struct file *file)
1451 struct inode *inode = file->f_path.dentry->d_inode;
1452 struct timespec now;
1453 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1455 /* First try to exhaust all avenues to not sync */
1456 if (IS_NOCMTIME(inode))
1459 now = current_fs_time(inode->i_sb);
1460 if (!timespec_equal(&inode->i_mtime, &now))
1463 if (!timespec_equal(&inode->i_ctime, &now))
1466 if (IS_I_VERSION(inode))
1467 sync_it |= S_VERSION;
1472 /* Finally allowed to write? Takes lock. */
1473 if (mnt_want_write_file(file))
1476 /* Only change inode inside the lock region */
1477 if (sync_it & S_VERSION)
1478 inode_inc_iversion(inode);
1479 if (sync_it & S_CTIME)
1480 inode->i_ctime = now;
1481 if (sync_it & S_MTIME)
1482 inode->i_mtime = now;
1483 mark_inode_dirty_sync(inode);
1484 mnt_drop_write(file->f_path.mnt);
1486 EXPORT_SYMBOL(file_update_time);
1488 int inode_needs_sync(struct inode *inode)
1492 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1496 EXPORT_SYMBOL(inode_needs_sync);
1498 int inode_wait(void *word)
1503 EXPORT_SYMBOL(inode_wait);
1506 * If we try to find an inode in the inode hash while it is being
1507 * deleted, we have to wait until the filesystem completes its
1508 * deletion before reporting that it isn't found. This function waits
1509 * until the deletion _might_ have completed. Callers are responsible
1510 * to recheck inode state.
1512 * It doesn't matter if I_NEW is not set initially, a call to
1513 * wake_up_inode() after removing from the hash list will DTRT.
1515 * This is called with inode_lock held.
1517 static void __wait_on_freeing_inode(struct inode *inode)
1519 wait_queue_head_t *wq;
1520 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1521 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1522 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1523 spin_unlock(&inode_lock);
1525 finish_wait(wq, &wait.wait);
1526 spin_lock(&inode_lock);
1529 static __initdata unsigned long ihash_entries;
1530 static int __init set_ihash_entries(char *str)
1534 ihash_entries = simple_strtoul(str, &str, 0);
1537 __setup("ihash_entries=", set_ihash_entries);
1540 * Initialize the waitqueues and inode hash table.
1542 void __init inode_init_early(void)
1546 /* If hashes are distributed across NUMA nodes, defer
1547 * hash allocation until vmalloc space is available.
1553 alloc_large_system_hash("Inode-cache",
1554 sizeof(struct hlist_head),
1562 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1563 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1566 void __init inode_init(void)
1570 /* inode slab cache */
1571 inode_cachep = kmem_cache_create("inode_cache",
1572 sizeof(struct inode),
1574 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1577 register_shrinker(&icache_shrinker);
1579 /* Hash may have been set up in inode_init_early */
1584 alloc_large_system_hash("Inode-cache",
1585 sizeof(struct hlist_head),
1593 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1594 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1597 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1599 inode->i_mode = mode;
1600 if (S_ISCHR(mode)) {
1601 inode->i_fop = &def_chr_fops;
1602 inode->i_rdev = rdev;
1603 } else if (S_ISBLK(mode)) {
1604 inode->i_fop = &def_blk_fops;
1605 inode->i_rdev = rdev;
1606 } else if (S_ISFIFO(mode))
1607 inode->i_fop = &def_fifo_fops;
1608 else if (S_ISSOCK(mode))
1609 inode->i_fop = &bad_sock_fops;
1611 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1612 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1615 EXPORT_SYMBOL(init_special_inode);