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/hash.h>
18 #include <linux/swap.h>
19 #include <linux/security.h>
20 #include <linux/ima.h>
21 #include <linux/pagemap.h>
22 #include <linux/cdev.h>
23 #include <linux/bootmem.h>
24 #include <linux/inotify.h>
25 #include <linux/mount.h>
26 #include <linux/async.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 LIST_HEAD(inode_in_use);
76 LIST_HEAD(inode_unused);
77 static struct hlist_head *inode_hashtable __read_mostly;
80 * A simple spinlock to protect the list manipulations.
82 * NOTE! You also have to own the lock if you change
83 * the i_state of an inode while it is in use..
85 DEFINE_SPINLOCK(inode_lock);
88 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
89 * icache shrinking path, and the umount path. Without this exclusion,
90 * by the time prune_icache calls iput for the inode whose pages it has
91 * been invalidating, or by the time it calls clear_inode & destroy_inode
92 * from its final dispose_list, the struct super_block they refer to
93 * (for inode->i_sb->s_op) may already have been freed and reused.
95 static DEFINE_MUTEX(iprune_mutex);
98 * Statistics gathering..
100 struct inodes_stat_t inodes_stat;
102 static struct kmem_cache *inode_cachep __read_mostly;
104 static void wake_up_inode(struct inode *inode)
107 * Prevent speculative execution through spin_unlock(&inode_lock);
110 wake_up_bit(&inode->i_state, __I_LOCK);
114 * inode_init_always - perform inode structure intialisation
115 * @sb: superblock inode belongs to
116 * @inode: inode to initialise
118 * These are initializations that need to be done on every inode
119 * allocation as the fields are not initialised by slab allocation.
121 struct inode *inode_init_always(struct super_block *sb, struct inode *inode)
123 static const struct address_space_operations empty_aops;
124 static struct inode_operations empty_iops;
125 static const struct file_operations empty_fops;
127 struct address_space *const mapping = &inode->i_data;
130 inode->i_blkbits = sb->s_blocksize_bits;
132 atomic_set(&inode->i_count, 1);
133 inode->i_op = &empty_iops;
134 inode->i_fop = &empty_fops;
138 atomic_set(&inode->i_writecount, 0);
142 inode->i_generation = 0;
144 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
146 inode->i_pipe = NULL;
147 inode->i_bdev = NULL;
148 inode->i_cdev = NULL;
150 inode->dirtied_when = 0;
152 if (security_inode_alloc(inode))
155 /* allocate and initialize an i_integrity */
156 if (ima_inode_alloc(inode))
157 goto out_free_security;
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_backing_dev_info;
186 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
187 mapping->backing_dev_info = bdi;
189 inode->i_private = NULL;
190 inode->i_mapping = mapping;
195 security_inode_free(inode);
197 if (inode->i_sb->s_op->destroy_inode)
198 inode->i_sb->s_op->destroy_inode(inode);
200 kmem_cache_free(inode_cachep, (inode));
203 EXPORT_SYMBOL(inode_init_always);
205 static struct inode *alloc_inode(struct super_block *sb)
209 if (sb->s_op->alloc_inode)
210 inode = sb->s_op->alloc_inode(sb);
212 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
215 return inode_init_always(sb, inode);
219 void destroy_inode(struct inode *inode)
221 BUG_ON(inode_has_buffers(inode));
222 ima_inode_free(inode);
223 security_inode_free(inode);
224 if (inode->i_sb->s_op->destroy_inode)
225 inode->i_sb->s_op->destroy_inode(inode);
227 kmem_cache_free(inode_cachep, (inode));
229 EXPORT_SYMBOL(destroy_inode);
233 * These are initializations that only need to be done
234 * once, because the fields are idempotent across use
235 * of the inode, so let the slab aware of that.
237 void inode_init_once(struct inode *inode)
239 memset(inode, 0, sizeof(*inode));
240 INIT_HLIST_NODE(&inode->i_hash);
241 INIT_LIST_HEAD(&inode->i_dentry);
242 INIT_LIST_HEAD(&inode->i_devices);
243 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
244 spin_lock_init(&inode->i_data.tree_lock);
245 spin_lock_init(&inode->i_data.i_mmap_lock);
246 INIT_LIST_HEAD(&inode->i_data.private_list);
247 spin_lock_init(&inode->i_data.private_lock);
248 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
249 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
250 i_size_ordered_init(inode);
251 #ifdef CONFIG_INOTIFY
252 INIT_LIST_HEAD(&inode->inotify_watches);
253 mutex_init(&inode->inotify_mutex);
256 EXPORT_SYMBOL(inode_init_once);
258 static void init_once(void *foo)
260 struct inode *inode = (struct inode *) foo;
262 inode_init_once(inode);
266 * inode_lock must be held
268 void __iget(struct inode *inode)
270 if (atomic_read(&inode->i_count)) {
271 atomic_inc(&inode->i_count);
274 atomic_inc(&inode->i_count);
275 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
276 list_move(&inode->i_list, &inode_in_use);
277 inodes_stat.nr_unused--;
281 * clear_inode - clear an inode
282 * @inode: inode to clear
284 * This is called by the filesystem to tell us
285 * that the inode is no longer useful. We just
286 * terminate it with extreme prejudice.
288 void clear_inode(struct inode *inode)
291 invalidate_inode_buffers(inode);
293 BUG_ON(inode->i_data.nrpages);
294 BUG_ON(!(inode->i_state & I_FREEING));
295 BUG_ON(inode->i_state & I_CLEAR);
296 inode_sync_wait(inode);
298 if (inode->i_sb->s_op->clear_inode)
299 inode->i_sb->s_op->clear_inode(inode);
300 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
302 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
304 inode->i_state = I_CLEAR;
306 EXPORT_SYMBOL(clear_inode);
309 * dispose_list - dispose of the contents of a local list
310 * @head: the head of the list to free
312 * Dispose-list gets a local list with local inodes in it, so it doesn't
313 * need to worry about list corruption and SMP locks.
315 static void dispose_list(struct list_head *head)
319 while (!list_empty(head)) {
322 inode = list_first_entry(head, struct inode, i_list);
323 list_del(&inode->i_list);
325 if (inode->i_data.nrpages)
326 truncate_inode_pages(&inode->i_data, 0);
329 spin_lock(&inode_lock);
330 hlist_del_init(&inode->i_hash);
331 list_del_init(&inode->i_sb_list);
332 spin_unlock(&inode_lock);
334 wake_up_inode(inode);
335 destroy_inode(inode);
338 spin_lock(&inode_lock);
339 inodes_stat.nr_inodes -= nr_disposed;
340 spin_unlock(&inode_lock);
344 * Invalidate all inodes for a device.
346 static int invalidate_list(struct list_head *head, struct list_head *dispose)
348 struct list_head *next;
349 int busy = 0, count = 0;
353 struct list_head *tmp = next;
357 * We can reschedule here without worrying about the list's
358 * consistency because the per-sb list of inodes must not
359 * change during umount anymore, and because iprune_mutex keeps
360 * shrink_icache_memory() away.
362 cond_resched_lock(&inode_lock);
367 inode = list_entry(tmp, struct inode, i_sb_list);
368 if (inode->i_state & I_NEW)
370 invalidate_inode_buffers(inode);
371 if (!atomic_read(&inode->i_count)) {
372 list_move(&inode->i_list, dispose);
373 WARN_ON(inode->i_state & I_NEW);
374 inode->i_state |= I_FREEING;
380 /* only unused inodes may be cached with i_count zero */
381 inodes_stat.nr_unused -= count;
386 * invalidate_inodes - discard the inodes on a device
389 * Discard all of the inodes for a given superblock. If the discard
390 * fails because there are busy inodes then a non zero value is returned.
391 * If the discard is successful all the inodes have been discarded.
393 int invalidate_inodes(struct super_block *sb)
396 LIST_HEAD(throw_away);
398 mutex_lock(&iprune_mutex);
399 spin_lock(&inode_lock);
400 inotify_unmount_inodes(&sb->s_inodes);
401 busy = invalidate_list(&sb->s_inodes, &throw_away);
402 spin_unlock(&inode_lock);
404 dispose_list(&throw_away);
405 mutex_unlock(&iprune_mutex);
409 EXPORT_SYMBOL(invalidate_inodes);
411 static int can_unuse(struct inode *inode)
415 if (inode_has_buffers(inode))
417 if (atomic_read(&inode->i_count))
419 if (inode->i_data.nrpages)
425 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
426 * a temporary list and then are freed outside inode_lock by dispose_list().
428 * Any inodes which are pinned purely because of attached pagecache have their
429 * pagecache removed. We expect the final iput() on that inode to add it to
430 * the front of the inode_unused list. So look for it there and if the
431 * inode is still freeable, proceed. The right inode is found 99.9% of the
432 * time in testing on a 4-way.
434 * If the inode has metadata buffers attached to mapping->private_list then
435 * try to remove them.
437 static void prune_icache(int nr_to_scan)
442 unsigned long reap = 0;
444 mutex_lock(&iprune_mutex);
445 spin_lock(&inode_lock);
446 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
449 if (list_empty(&inode_unused))
452 inode = list_entry(inode_unused.prev, struct inode, i_list);
454 if (inode->i_state || atomic_read(&inode->i_count)) {
455 list_move(&inode->i_list, &inode_unused);
458 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
460 spin_unlock(&inode_lock);
461 if (remove_inode_buffers(inode))
462 reap += invalidate_mapping_pages(&inode->i_data,
465 spin_lock(&inode_lock);
467 if (inode != list_entry(inode_unused.next,
468 struct inode, i_list))
469 continue; /* wrong inode or list_empty */
470 if (!can_unuse(inode))
473 list_move(&inode->i_list, &freeable);
474 WARN_ON(inode->i_state & I_NEW);
475 inode->i_state |= I_FREEING;
478 inodes_stat.nr_unused -= nr_pruned;
479 if (current_is_kswapd())
480 __count_vm_events(KSWAPD_INODESTEAL, reap);
482 __count_vm_events(PGINODESTEAL, reap);
483 spin_unlock(&inode_lock);
485 dispose_list(&freeable);
486 mutex_unlock(&iprune_mutex);
490 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
491 * "unused" means that no dentries are referring to the inodes: the files are
492 * not open and the dcache references to those inodes have already been
495 * This function is passed the number of inodes to scan, and it returns the
496 * total number of remaining possibly-reclaimable inodes.
498 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
502 * Nasty deadlock avoidance. We may hold various FS locks,
503 * and we don't want to recurse into the FS that called us
504 * in clear_inode() and friends..
506 if (!(gfp_mask & __GFP_FS))
510 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
513 static struct shrinker icache_shrinker = {
514 .shrink = shrink_icache_memory,
515 .seeks = DEFAULT_SEEKS,
518 static void __wait_on_freeing_inode(struct inode *inode);
520 * Called with the inode lock held.
521 * NOTE: we are not increasing the inode-refcount, you must call __iget()
522 * by hand after calling find_inode now! This simplifies iunique and won't
523 * add any additional branch in the common code.
525 static struct inode *find_inode(struct super_block *sb,
526 struct hlist_head *head,
527 int (*test)(struct inode *, void *),
530 struct hlist_node *node;
531 struct inode *inode = NULL;
534 hlist_for_each_entry(inode, node, head, i_hash) {
535 if (inode->i_sb != sb)
537 if (!test(inode, data))
539 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
540 __wait_on_freeing_inode(inode);
545 return node ? inode : NULL;
549 * find_inode_fast is the fast path version of find_inode, see the comment at
550 * iget_locked for details.
552 static struct inode *find_inode_fast(struct super_block *sb,
553 struct hlist_head *head, unsigned long ino)
555 struct hlist_node *node;
556 struct inode *inode = NULL;
559 hlist_for_each_entry(inode, node, head, i_hash) {
560 if (inode->i_ino != ino)
562 if (inode->i_sb != sb)
564 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
565 __wait_on_freeing_inode(inode);
570 return node ? inode : NULL;
573 static unsigned long hash(struct super_block *sb, unsigned long hashval)
577 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
579 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
580 return tmp & I_HASHMASK;
584 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
587 inodes_stat.nr_inodes++;
588 list_add(&inode->i_list, &inode_in_use);
589 list_add(&inode->i_sb_list, &sb->s_inodes);
591 hlist_add_head(&inode->i_hash, head);
595 * inode_add_to_lists - add a new inode to relevant lists
596 * @sb: superblock inode belongs to
597 * @inode: inode to mark in use
599 * When an inode is allocated it needs to be accounted for, added to the in use
600 * list, the owning superblock and the inode hash. This needs to be done under
601 * the inode_lock, so export a function to do this rather than the inode lock
602 * itself. We calculate the hash list to add to here so it is all internal
603 * which requires the caller to have already set up the inode number in the
606 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
608 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
610 spin_lock(&inode_lock);
611 __inode_add_to_lists(sb, head, inode);
612 spin_unlock(&inode_lock);
614 EXPORT_SYMBOL_GPL(inode_add_to_lists);
617 * new_inode - obtain an inode
620 * Allocates a new inode for given superblock. The default gfp_mask
621 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
622 * If HIGHMEM pages are unsuitable or it is known that pages allocated
623 * for the page cache are not reclaimable or migratable,
624 * mapping_set_gfp_mask() must be called with suitable flags on the
625 * newly created inode's mapping
628 struct inode *new_inode(struct super_block *sb)
631 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
632 * error if st_ino won't fit in target struct field. Use 32bit counter
633 * here to attempt to avoid that.
635 static unsigned int last_ino;
638 spin_lock_prefetch(&inode_lock);
640 inode = alloc_inode(sb);
642 spin_lock(&inode_lock);
643 __inode_add_to_lists(sb, NULL, inode);
644 inode->i_ino = ++last_ino;
646 spin_unlock(&inode_lock);
650 EXPORT_SYMBOL(new_inode);
652 void unlock_new_inode(struct inode *inode)
654 #ifdef CONFIG_DEBUG_LOCK_ALLOC
655 if (inode->i_mode & S_IFDIR) {
656 struct file_system_type *type = inode->i_sb->s_type;
659 * ensure nobody is actually holding i_mutex
661 mutex_destroy(&inode->i_mutex);
662 mutex_init(&inode->i_mutex);
663 lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key);
667 * This is special! We do not need the spinlock
668 * when clearing I_LOCK, because we're guaranteed
669 * that nobody else tries to do anything about the
670 * state of the inode when it is locked, as we
671 * just created it (so there can be no old holders
672 * that haven't tested I_LOCK).
674 WARN_ON((inode->i_state & (I_LOCK|I_NEW)) != (I_LOCK|I_NEW));
675 inode->i_state &= ~(I_LOCK|I_NEW);
676 wake_up_inode(inode);
678 EXPORT_SYMBOL(unlock_new_inode);
681 * This is called without the inode lock held.. Be careful.
683 * We no longer cache the sb_flags in i_flags - see fs.h
684 * -- rmk@arm.uk.linux.org
686 static struct inode *get_new_inode(struct super_block *sb,
687 struct hlist_head *head,
688 int (*test)(struct inode *, void *),
689 int (*set)(struct inode *, void *),
694 inode = alloc_inode(sb);
698 spin_lock(&inode_lock);
699 /* We released the lock, so.. */
700 old = find_inode(sb, head, test, data);
702 if (set(inode, data))
705 __inode_add_to_lists(sb, head, inode);
706 inode->i_state = I_LOCK|I_NEW;
707 spin_unlock(&inode_lock);
709 /* Return the locked inode with I_NEW set, the
710 * caller is responsible for filling in the contents
716 * Uhhuh, somebody else created the same inode under
717 * us. Use the old inode instead of the one we just
721 spin_unlock(&inode_lock);
722 destroy_inode(inode);
724 wait_on_inode(inode);
729 spin_unlock(&inode_lock);
730 destroy_inode(inode);
735 * get_new_inode_fast is the fast path version of get_new_inode, see the
736 * comment at iget_locked for details.
738 static struct inode *get_new_inode_fast(struct super_block *sb,
739 struct hlist_head *head, unsigned long ino)
743 inode = alloc_inode(sb);
747 spin_lock(&inode_lock);
748 /* We released the lock, so.. */
749 old = find_inode_fast(sb, head, ino);
752 __inode_add_to_lists(sb, head, inode);
753 inode->i_state = I_LOCK|I_NEW;
754 spin_unlock(&inode_lock);
756 /* Return the locked inode with I_NEW set, the
757 * caller is responsible for filling in the contents
763 * Uhhuh, somebody else created the same inode under
764 * us. Use the old inode instead of the one we just
768 spin_unlock(&inode_lock);
769 destroy_inode(inode);
771 wait_on_inode(inode);
777 * iunique - get a unique inode number
779 * @max_reserved: highest reserved inode number
781 * Obtain an inode number that is unique on the system for a given
782 * superblock. This is used by file systems that have no natural
783 * permanent inode numbering system. An inode number is returned that
784 * is higher than the reserved limit but unique.
787 * With a large number of inodes live on the file system this function
788 * currently becomes quite slow.
790 ino_t iunique(struct super_block *sb, ino_t max_reserved)
793 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
794 * error if st_ino won't fit in target struct field. Use 32bit counter
795 * here to attempt to avoid that.
797 static unsigned int counter;
799 struct hlist_head *head;
802 spin_lock(&inode_lock);
804 if (counter <= max_reserved)
805 counter = max_reserved + 1;
807 head = inode_hashtable + hash(sb, res);
808 inode = find_inode_fast(sb, head, res);
809 } while (inode != NULL);
810 spin_unlock(&inode_lock);
814 EXPORT_SYMBOL(iunique);
816 struct inode *igrab(struct inode *inode)
818 spin_lock(&inode_lock);
819 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
823 * Handle the case where s_op->clear_inode is not been
824 * called yet, and somebody is calling igrab
825 * while the inode is getting freed.
828 spin_unlock(&inode_lock);
831 EXPORT_SYMBOL(igrab);
834 * ifind - internal function, you want ilookup5() or iget5().
835 * @sb: super block of file system to search
836 * @head: the head of the list to search
837 * @test: callback used for comparisons between inodes
838 * @data: opaque data pointer to pass to @test
839 * @wait: if true wait for the inode to be unlocked, if false do not
841 * ifind() searches for the inode specified by @data in the inode
842 * cache. This is a generalized version of ifind_fast() for file systems where
843 * the inode number is not sufficient for unique identification of an inode.
845 * If the inode is in the cache, the inode is returned with an incremented
848 * Otherwise NULL is returned.
850 * Note, @test is called with the inode_lock held, so can't sleep.
852 static struct inode *ifind(struct super_block *sb,
853 struct hlist_head *head, int (*test)(struct inode *, void *),
854 void *data, const int wait)
858 spin_lock(&inode_lock);
859 inode = find_inode(sb, head, test, data);
862 spin_unlock(&inode_lock);
864 wait_on_inode(inode);
867 spin_unlock(&inode_lock);
872 * ifind_fast - internal function, you want ilookup() or iget().
873 * @sb: super block of file system to search
874 * @head: head of the list to search
875 * @ino: inode number to search for
877 * ifind_fast() searches for the inode @ino in the inode cache. This is for
878 * file systems where the inode number is sufficient for unique identification
881 * If the inode is in the cache, the inode is returned with an incremented
884 * Otherwise NULL is returned.
886 static struct inode *ifind_fast(struct super_block *sb,
887 struct hlist_head *head, unsigned long ino)
891 spin_lock(&inode_lock);
892 inode = find_inode_fast(sb, head, ino);
895 spin_unlock(&inode_lock);
896 wait_on_inode(inode);
899 spin_unlock(&inode_lock);
904 * ilookup5_nowait - search for an inode in the inode cache
905 * @sb: super block of file system to search
906 * @hashval: hash value (usually inode number) to search for
907 * @test: callback used for comparisons between inodes
908 * @data: opaque data pointer to pass to @test
910 * ilookup5() uses ifind() to search for the inode specified by @hashval and
911 * @data in the inode cache. This is a generalized version of ilookup() for
912 * file systems where the inode number is not sufficient for unique
913 * identification of an inode.
915 * If the inode is in the cache, the inode is returned with an incremented
916 * reference count. Note, the inode lock is not waited upon so you have to be
917 * very careful what you do with the returned inode. You probably should be
918 * using ilookup5() instead.
920 * Otherwise NULL is returned.
922 * Note, @test is called with the inode_lock held, so can't sleep.
924 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
925 int (*test)(struct inode *, void *), void *data)
927 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
929 return ifind(sb, head, test, data, 0);
931 EXPORT_SYMBOL(ilookup5_nowait);
934 * ilookup5 - search for an inode in the inode cache
935 * @sb: super block of file system to search
936 * @hashval: hash value (usually inode number) to search for
937 * @test: callback used for comparisons between inodes
938 * @data: opaque data pointer to pass to @test
940 * ilookup5() uses ifind() to search for the inode specified by @hashval and
941 * @data in the inode cache. This is a generalized version of ilookup() for
942 * file systems where the inode number is not sufficient for unique
943 * identification of an inode.
945 * If the inode is in the cache, the inode lock is waited upon and the inode is
946 * returned with an incremented reference count.
948 * Otherwise NULL is returned.
950 * Note, @test is called with the inode_lock held, so can't sleep.
952 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
953 int (*test)(struct inode *, void *), void *data)
955 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
957 return ifind(sb, head, test, data, 1);
959 EXPORT_SYMBOL(ilookup5);
962 * ilookup - search for an inode in the inode cache
963 * @sb: super block of file system to search
964 * @ino: inode number to search for
966 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
967 * This is for file systems where the inode number is sufficient for unique
968 * identification of an inode.
970 * If the inode is in the cache, the inode is returned with an incremented
973 * Otherwise NULL is returned.
975 struct inode *ilookup(struct super_block *sb, unsigned long ino)
977 struct hlist_head *head = inode_hashtable + hash(sb, ino);
979 return ifind_fast(sb, head, ino);
981 EXPORT_SYMBOL(ilookup);
984 * iget5_locked - obtain an inode from a mounted file system
985 * @sb: super block of file system
986 * @hashval: hash value (usually inode number) to get
987 * @test: callback used for comparisons between inodes
988 * @set: callback used to initialize a new struct inode
989 * @data: opaque data pointer to pass to @test and @set
991 * iget5_locked() uses ifind() to search for the inode specified by @hashval
992 * and @data in the inode cache and if present it is returned with an increased
993 * reference count. This is a generalized version of iget_locked() for file
994 * systems where the inode number is not sufficient for unique identification
997 * If the inode is not in cache, get_new_inode() is called to allocate a new
998 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
999 * file system gets to fill it in before unlocking it via unlock_new_inode().
1001 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1003 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1004 int (*test)(struct inode *, void *),
1005 int (*set)(struct inode *, void *), void *data)
1007 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1008 struct inode *inode;
1010 inode = ifind(sb, head, test, data, 1);
1014 * get_new_inode() will do the right thing, re-trying the search
1015 * in case it had to block at any point.
1017 return get_new_inode(sb, head, test, set, data);
1019 EXPORT_SYMBOL(iget5_locked);
1022 * iget_locked - obtain an inode from a mounted file system
1023 * @sb: super block of file system
1024 * @ino: inode number to get
1026 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1027 * the inode cache and if present it is returned with an increased reference
1028 * count. This is for file systems where the inode number is sufficient for
1029 * unique identification of an inode.
1031 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1032 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1033 * The file system gets to fill it in before unlocking it via
1034 * unlock_new_inode().
1036 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1038 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1039 struct inode *inode;
1041 inode = ifind_fast(sb, head, ino);
1045 * get_new_inode_fast() will do the right thing, re-trying the search
1046 * in case it had to block at any point.
1048 return get_new_inode_fast(sb, head, ino);
1050 EXPORT_SYMBOL(iget_locked);
1052 int insert_inode_locked(struct inode *inode)
1054 struct super_block *sb = inode->i_sb;
1055 ino_t ino = inode->i_ino;
1056 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1058 inode->i_state |= I_LOCK|I_NEW;
1060 struct hlist_node *node;
1061 struct inode *old = NULL;
1062 spin_lock(&inode_lock);
1063 hlist_for_each_entry(old, node, head, i_hash) {
1064 if (old->i_ino != ino)
1066 if (old->i_sb != sb)
1068 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1072 if (likely(!node)) {
1073 hlist_add_head(&inode->i_hash, head);
1074 spin_unlock(&inode_lock);
1078 spin_unlock(&inode_lock);
1080 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1087 EXPORT_SYMBOL(insert_inode_locked);
1089 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1090 int (*test)(struct inode *, void *), void *data)
1092 struct super_block *sb = inode->i_sb;
1093 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1095 inode->i_state |= I_LOCK|I_NEW;
1098 struct hlist_node *node;
1099 struct inode *old = NULL;
1101 spin_lock(&inode_lock);
1102 hlist_for_each_entry(old, node, head, i_hash) {
1103 if (old->i_sb != sb)
1105 if (!test(old, data))
1107 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1111 if (likely(!node)) {
1112 hlist_add_head(&inode->i_hash, head);
1113 spin_unlock(&inode_lock);
1117 spin_unlock(&inode_lock);
1119 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1126 EXPORT_SYMBOL(insert_inode_locked4);
1129 * __insert_inode_hash - hash an inode
1130 * @inode: unhashed inode
1131 * @hashval: unsigned long value used to locate this object in the
1134 * Add an inode to the inode hash for this superblock.
1136 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1138 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1139 spin_lock(&inode_lock);
1140 hlist_add_head(&inode->i_hash, head);
1141 spin_unlock(&inode_lock);
1143 EXPORT_SYMBOL(__insert_inode_hash);
1146 * remove_inode_hash - remove an inode from the hash
1147 * @inode: inode to unhash
1149 * Remove an inode from the superblock.
1151 void remove_inode_hash(struct inode *inode)
1153 spin_lock(&inode_lock);
1154 hlist_del_init(&inode->i_hash);
1155 spin_unlock(&inode_lock);
1157 EXPORT_SYMBOL(remove_inode_hash);
1160 * Tell the filesystem that this inode is no longer of any interest and should
1161 * be completely destroyed.
1163 * We leave the inode in the inode hash table until *after* the filesystem's
1164 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1165 * instigate) will always find up-to-date information either in the hash or on
1168 * I_FREEING is set so that no-one will take a new reference to the inode while
1169 * it is being deleted.
1171 void generic_delete_inode(struct inode *inode)
1173 const struct super_operations *op = inode->i_sb->s_op;
1175 list_del_init(&inode->i_list);
1176 list_del_init(&inode->i_sb_list);
1177 WARN_ON(inode->i_state & I_NEW);
1178 inode->i_state |= I_FREEING;
1179 inodes_stat.nr_inodes--;
1180 spin_unlock(&inode_lock);
1182 security_inode_delete(inode);
1184 if (op->delete_inode) {
1185 void (*delete)(struct inode *) = op->delete_inode;
1186 if (!is_bad_inode(inode))
1188 /* Filesystems implementing their own
1189 * s_op->delete_inode are required to call
1190 * truncate_inode_pages and clear_inode()
1194 truncate_inode_pages(&inode->i_data, 0);
1197 spin_lock(&inode_lock);
1198 hlist_del_init(&inode->i_hash);
1199 spin_unlock(&inode_lock);
1200 wake_up_inode(inode);
1201 BUG_ON(inode->i_state != I_CLEAR);
1202 destroy_inode(inode);
1204 EXPORT_SYMBOL(generic_delete_inode);
1206 static void generic_forget_inode(struct inode *inode)
1208 struct super_block *sb = inode->i_sb;
1210 if (!hlist_unhashed(&inode->i_hash)) {
1211 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1212 list_move(&inode->i_list, &inode_unused);
1213 inodes_stat.nr_unused++;
1214 if (sb->s_flags & MS_ACTIVE) {
1215 spin_unlock(&inode_lock);
1218 WARN_ON(inode->i_state & I_NEW);
1219 inode->i_state |= I_WILL_FREE;
1220 spin_unlock(&inode_lock);
1221 write_inode_now(inode, 1);
1222 spin_lock(&inode_lock);
1223 WARN_ON(inode->i_state & I_NEW);
1224 inode->i_state &= ~I_WILL_FREE;
1225 inodes_stat.nr_unused--;
1226 hlist_del_init(&inode->i_hash);
1228 list_del_init(&inode->i_list);
1229 list_del_init(&inode->i_sb_list);
1230 WARN_ON(inode->i_state & I_NEW);
1231 inode->i_state |= I_FREEING;
1232 inodes_stat.nr_inodes--;
1233 spin_unlock(&inode_lock);
1234 if (inode->i_data.nrpages)
1235 truncate_inode_pages(&inode->i_data, 0);
1237 wake_up_inode(inode);
1238 destroy_inode(inode);
1242 * Normal UNIX filesystem behaviour: delete the
1243 * inode when the usage count drops to zero, and
1246 void generic_drop_inode(struct inode *inode)
1248 if (!inode->i_nlink)
1249 generic_delete_inode(inode);
1251 generic_forget_inode(inode);
1253 EXPORT_SYMBOL_GPL(generic_drop_inode);
1256 * Called when we're dropping the last reference
1259 * Call the FS "drop()" function, defaulting to
1260 * the legacy UNIX filesystem behaviour..
1262 * NOTE! NOTE! NOTE! We're called with the inode lock
1263 * held, and the drop function is supposed to release
1266 static inline void iput_final(struct inode *inode)
1268 const struct super_operations *op = inode->i_sb->s_op;
1269 void (*drop)(struct inode *) = generic_drop_inode;
1271 if (op && op->drop_inode)
1272 drop = op->drop_inode;
1277 * iput - put an inode
1278 * @inode: inode to put
1280 * Puts an inode, dropping its usage count. If the inode use count hits
1281 * zero, the inode is then freed and may also be destroyed.
1283 * Consequently, iput() can sleep.
1285 void iput(struct inode *inode)
1288 BUG_ON(inode->i_state == I_CLEAR);
1290 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1294 EXPORT_SYMBOL(iput);
1297 * bmap - find a block number in a file
1298 * @inode: inode of file
1299 * @block: block to find
1301 * Returns the block number on the device holding the inode that
1302 * is the disk block number for the block of the file requested.
1303 * That is, asked for block 4 of inode 1 the function will return the
1304 * disk block relative to the disk start that holds that block of the
1307 sector_t bmap(struct inode *inode, sector_t block)
1310 if (inode->i_mapping->a_ops->bmap)
1311 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1314 EXPORT_SYMBOL(bmap);
1317 * With relative atime, only update atime if the previous atime is
1318 * earlier than either the ctime or mtime or if at least a day has
1319 * passed since the last atime update.
1321 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1322 struct timespec now)
1325 if (!(mnt->mnt_flags & MNT_RELATIME))
1328 * Is mtime younger than atime? If yes, update atime:
1330 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1333 * Is ctime younger than atime? If yes, update atime:
1335 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1339 * Is the previous atime value older than a day? If yes,
1342 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1345 * Good, we can skip the atime update:
1351 * touch_atime - update the access time
1352 * @mnt: mount the inode is accessed on
1353 * @dentry: dentry accessed
1355 * Update the accessed time on an inode and mark it for writeback.
1356 * This function automatically handles read only file systems and media,
1357 * as well as the "noatime" flag and inode specific "noatime" markers.
1359 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1361 struct inode *inode = dentry->d_inode;
1362 struct timespec now;
1364 if (mnt_want_write(mnt))
1366 if (inode->i_flags & S_NOATIME)
1368 if (IS_NOATIME(inode))
1370 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1373 if (mnt->mnt_flags & MNT_NOATIME)
1375 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1378 now = current_fs_time(inode->i_sb);
1380 if (!relatime_need_update(mnt, inode, now))
1383 if (timespec_equal(&inode->i_atime, &now))
1386 inode->i_atime = now;
1387 mark_inode_dirty_sync(inode);
1389 mnt_drop_write(mnt);
1391 EXPORT_SYMBOL(touch_atime);
1394 * file_update_time - update mtime and ctime time
1395 * @file: file accessed
1397 * Update the mtime and ctime members of an inode and mark the inode
1398 * for writeback. Note that this function is meant exclusively for
1399 * usage in the file write path of filesystems, and filesystems may
1400 * choose to explicitly ignore update via this function with the
1401 * S_NOCTIME inode flag, e.g. for network filesystem where these
1402 * timestamps are handled by the server.
1405 void file_update_time(struct file *file)
1407 struct inode *inode = file->f_path.dentry->d_inode;
1408 struct timespec now;
1412 if (IS_NOCMTIME(inode))
1415 err = mnt_want_write(file->f_path.mnt);
1419 now = current_fs_time(inode->i_sb);
1420 if (!timespec_equal(&inode->i_mtime, &now)) {
1421 inode->i_mtime = now;
1425 if (!timespec_equal(&inode->i_ctime, &now)) {
1426 inode->i_ctime = now;
1430 if (IS_I_VERSION(inode)) {
1431 inode_inc_iversion(inode);
1436 mark_inode_dirty_sync(inode);
1437 mnt_drop_write(file->f_path.mnt);
1439 EXPORT_SYMBOL(file_update_time);
1441 int inode_needs_sync(struct inode *inode)
1445 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1449 EXPORT_SYMBOL(inode_needs_sync);
1451 int inode_wait(void *word)
1456 EXPORT_SYMBOL(inode_wait);
1459 * If we try to find an inode in the inode hash while it is being
1460 * deleted, we have to wait until the filesystem completes its
1461 * deletion before reporting that it isn't found. This function waits
1462 * until the deletion _might_ have completed. Callers are responsible
1463 * to recheck inode state.
1465 * It doesn't matter if I_LOCK is not set initially, a call to
1466 * wake_up_inode() after removing from the hash list will DTRT.
1468 * This is called with inode_lock held.
1470 static void __wait_on_freeing_inode(struct inode *inode)
1472 wait_queue_head_t *wq;
1473 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1474 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1475 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1476 spin_unlock(&inode_lock);
1478 finish_wait(wq, &wait.wait);
1479 spin_lock(&inode_lock);
1482 static __initdata unsigned long ihash_entries;
1483 static int __init set_ihash_entries(char *str)
1487 ihash_entries = simple_strtoul(str, &str, 0);
1490 __setup("ihash_entries=", set_ihash_entries);
1493 * Initialize the waitqueues and inode hash table.
1495 void __init inode_init_early(void)
1499 /* If hashes are distributed across NUMA nodes, defer
1500 * hash allocation until vmalloc space is available.
1506 alloc_large_system_hash("Inode-cache",
1507 sizeof(struct hlist_head),
1515 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1516 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1519 void __init inode_init(void)
1523 /* inode slab cache */
1524 inode_cachep = kmem_cache_create("inode_cache",
1525 sizeof(struct inode),
1527 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1530 register_shrinker(&icache_shrinker);
1532 /* Hash may have been set up in inode_init_early */
1537 alloc_large_system_hash("Inode-cache",
1538 sizeof(struct hlist_head),
1546 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1547 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1550 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1552 inode->i_mode = mode;
1553 if (S_ISCHR(mode)) {
1554 inode->i_fop = &def_chr_fops;
1555 inode->i_rdev = rdev;
1556 } else if (S_ISBLK(mode)) {
1557 inode->i_fop = &def_blk_fops;
1558 inode->i_rdev = rdev;
1559 } else if (S_ISFIFO(mode))
1560 inode->i_fop = &def_fifo_fops;
1561 else if (S_ISSOCK(mode))
1562 inode->i_fop = &bad_sock_fops;
1564 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1567 EXPORT_SYMBOL(init_special_inode);