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 security_inode_free(inode);
223 if (inode->i_sb->s_op->destroy_inode)
224 inode->i_sb->s_op->destroy_inode(inode);
226 kmem_cache_free(inode_cachep, (inode));
228 EXPORT_SYMBOL(destroy_inode);
232 * These are initializations that only need to be done
233 * once, because the fields are idempotent across use
234 * of the inode, so let the slab aware of that.
236 void inode_init_once(struct inode *inode)
238 memset(inode, 0, sizeof(*inode));
239 INIT_HLIST_NODE(&inode->i_hash);
240 INIT_LIST_HEAD(&inode->i_dentry);
241 INIT_LIST_HEAD(&inode->i_devices);
242 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
243 spin_lock_init(&inode->i_data.tree_lock);
244 spin_lock_init(&inode->i_data.i_mmap_lock);
245 INIT_LIST_HEAD(&inode->i_data.private_list);
246 spin_lock_init(&inode->i_data.private_lock);
247 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
248 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
249 i_size_ordered_init(inode);
250 #ifdef CONFIG_INOTIFY
251 INIT_LIST_HEAD(&inode->inotify_watches);
252 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;
307 EXPORT_SYMBOL(clear_inode);
310 * dispose_list - dispose of the contents of a local list
311 * @head: the head of the list to free
313 * Dispose-list gets a local list with local inodes in it, so it doesn't
314 * need to worry about list corruption and SMP locks.
316 static void dispose_list(struct list_head *head)
320 while (!list_empty(head)) {
323 inode = list_first_entry(head, struct inode, i_list);
324 list_del(&inode->i_list);
326 if (inode->i_data.nrpages)
327 truncate_inode_pages(&inode->i_data, 0);
330 spin_lock(&inode_lock);
331 hlist_del_init(&inode->i_hash);
332 list_del_init(&inode->i_sb_list);
333 spin_unlock(&inode_lock);
335 wake_up_inode(inode);
336 destroy_inode(inode);
339 spin_lock(&inode_lock);
340 inodes_stat.nr_inodes -= nr_disposed;
341 spin_unlock(&inode_lock);
345 * Invalidate all inodes for a device.
347 static int invalidate_list(struct list_head *head, struct list_head *dispose)
349 struct list_head *next;
350 int busy = 0, count = 0;
354 struct list_head * tmp = next;
355 struct inode * inode;
358 * We can reschedule here without worrying about the list's
359 * consistency because the per-sb list of inodes must not
360 * change during umount anymore, and because iprune_mutex keeps
361 * shrink_icache_memory() away.
363 cond_resched_lock(&inode_lock);
368 inode = list_entry(tmp, struct inode, i_sb_list);
369 invalidate_inode_buffers(inode);
370 if (!atomic_read(&inode->i_count)) {
371 list_move(&inode->i_list, dispose);
372 inode->i_state |= I_FREEING;
378 /* only unused inodes may be cached with i_count zero */
379 inodes_stat.nr_unused -= count;
384 * invalidate_inodes - discard the inodes on a device
387 * Discard all of the inodes for a given superblock. If the discard
388 * fails because there are busy inodes then a non zero value is returned.
389 * If the discard is successful all the inodes have been discarded.
391 int invalidate_inodes(struct super_block * sb)
394 LIST_HEAD(throw_away);
396 mutex_lock(&iprune_mutex);
397 spin_lock(&inode_lock);
398 inotify_unmount_inodes(&sb->s_inodes);
399 busy = invalidate_list(&sb->s_inodes, &throw_away);
400 spin_unlock(&inode_lock);
402 dispose_list(&throw_away);
403 mutex_unlock(&iprune_mutex);
408 EXPORT_SYMBOL(invalidate_inodes);
410 static int can_unuse(struct inode *inode)
414 if (inode_has_buffers(inode))
416 if (atomic_read(&inode->i_count))
418 if (inode->i_data.nrpages)
424 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
425 * a temporary list and then are freed outside inode_lock by dispose_list().
427 * Any inodes which are pinned purely because of attached pagecache have their
428 * pagecache removed. We expect the final iput() on that inode to add it to
429 * the front of the inode_unused list. So look for it there and if the
430 * inode is still freeable, proceed. The right inode is found 99.9% of the
431 * time in testing on a 4-way.
433 * If the inode has metadata buffers attached to mapping->private_list then
434 * try to remove them.
436 static void prune_icache(int nr_to_scan)
441 unsigned long reap = 0;
443 mutex_lock(&iprune_mutex);
444 spin_lock(&inode_lock);
445 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
448 if (list_empty(&inode_unused))
451 inode = list_entry(inode_unused.prev, struct inode, i_list);
453 if (inode->i_state || atomic_read(&inode->i_count)) {
454 list_move(&inode->i_list, &inode_unused);
457 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
459 spin_unlock(&inode_lock);
460 if (remove_inode_buffers(inode))
461 reap += invalidate_mapping_pages(&inode->i_data,
464 spin_lock(&inode_lock);
466 if (inode != list_entry(inode_unused.next,
467 struct inode, i_list))
468 continue; /* wrong inode or list_empty */
469 if (!can_unuse(inode))
472 list_move(&inode->i_list, &freeable);
473 inode->i_state |= I_FREEING;
476 inodes_stat.nr_unused -= nr_pruned;
477 if (current_is_kswapd())
478 __count_vm_events(KSWAPD_INODESTEAL, reap);
480 __count_vm_events(PGINODESTEAL, reap);
481 spin_unlock(&inode_lock);
483 dispose_list(&freeable);
484 mutex_unlock(&iprune_mutex);
488 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
489 * "unused" means that no dentries are referring to the inodes: the files are
490 * not open and the dcache references to those inodes have already been
493 * This function is passed the number of inodes to scan, and it returns the
494 * total number of remaining possibly-reclaimable inodes.
496 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
500 * Nasty deadlock avoidance. We may hold various FS locks,
501 * and we don't want to recurse into the FS that called us
502 * in clear_inode() and friends..
504 if (!(gfp_mask & __GFP_FS))
508 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
511 static struct shrinker icache_shrinker = {
512 .shrink = shrink_icache_memory,
513 .seeks = DEFAULT_SEEKS,
516 static void __wait_on_freeing_inode(struct inode *inode);
518 * Called with the inode lock held.
519 * NOTE: we are not increasing the inode-refcount, you must call __iget()
520 * by hand after calling find_inode now! This simplifies iunique and won't
521 * add any additional branch in the common code.
523 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
525 struct hlist_node *node;
526 struct inode * inode = NULL;
529 hlist_for_each_entry(inode, node, head, i_hash) {
530 if (inode->i_sb != sb)
532 if (!test(inode, data))
534 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
535 __wait_on_freeing_inode(inode);
540 return node ? inode : NULL;
544 * find_inode_fast is the fast path version of find_inode, see the comment at
545 * iget_locked for details.
547 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
549 struct hlist_node *node;
550 struct inode * inode = NULL;
553 hlist_for_each_entry(inode, node, head, i_hash) {
554 if (inode->i_ino != ino)
556 if (inode->i_sb != sb)
558 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
559 __wait_on_freeing_inode(inode);
564 return node ? inode : NULL;
567 static unsigned long hash(struct super_block *sb, unsigned long hashval)
571 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
573 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
574 return tmp & I_HASHMASK;
578 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
581 inodes_stat.nr_inodes++;
582 list_add(&inode->i_list, &inode_in_use);
583 list_add(&inode->i_sb_list, &sb->s_inodes);
585 hlist_add_head(&inode->i_hash, head);
589 * inode_add_to_lists - add a new inode to relevant lists
590 * @sb: superblock inode belongs to
591 * @inode: inode to mark in use
593 * When an inode is allocated it needs to be accounted for, added to the in use
594 * list, the owning superblock and the inode hash. This needs to be done under
595 * the inode_lock, so export a function to do this rather than the inode lock
596 * itself. We calculate the hash list to add to here so it is all internal
597 * which requires the caller to have already set up the inode number in the
600 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
602 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
604 spin_lock(&inode_lock);
605 __inode_add_to_lists(sb, head, inode);
606 spin_unlock(&inode_lock);
608 EXPORT_SYMBOL_GPL(inode_add_to_lists);
611 * new_inode - obtain an inode
614 * Allocates a new inode for given superblock. The default gfp_mask
615 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
616 * If HIGHMEM pages are unsuitable or it is known that pages allocated
617 * for the page cache are not reclaimable or migratable,
618 * mapping_set_gfp_mask() must be called with suitable flags on the
619 * newly created inode's mapping
622 struct inode *new_inode(struct super_block *sb)
625 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
626 * error if st_ino won't fit in target struct field. Use 32bit counter
627 * here to attempt to avoid that.
629 static unsigned int last_ino;
630 struct inode * inode;
632 spin_lock_prefetch(&inode_lock);
634 inode = alloc_inode(sb);
636 spin_lock(&inode_lock);
637 __inode_add_to_lists(sb, NULL, inode);
638 inode->i_ino = ++last_ino;
640 spin_unlock(&inode_lock);
645 EXPORT_SYMBOL(new_inode);
647 void unlock_new_inode(struct inode *inode)
649 #ifdef CONFIG_DEBUG_LOCK_ALLOC
650 if (inode->i_mode & S_IFDIR) {
651 struct file_system_type *type = inode->i_sb->s_type;
654 * ensure nobody is actually holding i_mutex
656 mutex_destroy(&inode->i_mutex);
657 mutex_init(&inode->i_mutex);
658 lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key);
662 * This is special! We do not need the spinlock
663 * when clearing I_LOCK, because we're guaranteed
664 * that nobody else tries to do anything about the
665 * state of the inode when it is locked, as we
666 * just created it (so there can be no old holders
667 * that haven't tested I_LOCK).
669 inode->i_state &= ~(I_LOCK|I_NEW);
670 wake_up_inode(inode);
673 EXPORT_SYMBOL(unlock_new_inode);
676 * This is called without the inode lock held.. Be careful.
678 * We no longer cache the sb_flags in i_flags - see fs.h
679 * -- rmk@arm.uk.linux.org
681 static struct inode * get_new_inode(struct super_block *sb, struct hlist_head *head, int (*test)(struct inode *, void *), int (*set)(struct inode *, void *), void *data)
683 struct inode * inode;
685 inode = alloc_inode(sb);
689 spin_lock(&inode_lock);
690 /* We released the lock, so.. */
691 old = find_inode(sb, head, test, data);
693 if (set(inode, data))
696 __inode_add_to_lists(sb, head, inode);
697 inode->i_state = I_LOCK|I_NEW;
698 spin_unlock(&inode_lock);
700 /* Return the locked inode with I_NEW set, the
701 * caller is responsible for filling in the contents
707 * Uhhuh, somebody else created the same inode under
708 * us. Use the old inode instead of the one we just
712 spin_unlock(&inode_lock);
713 destroy_inode(inode);
715 wait_on_inode(inode);
720 spin_unlock(&inode_lock);
721 destroy_inode(inode);
726 * get_new_inode_fast is the fast path version of get_new_inode, see the
727 * comment at iget_locked for details.
729 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
731 struct inode * inode;
733 inode = alloc_inode(sb);
737 spin_lock(&inode_lock);
738 /* We released the lock, so.. */
739 old = find_inode_fast(sb, head, ino);
742 __inode_add_to_lists(sb, head, inode);
743 inode->i_state = I_LOCK|I_NEW;
744 spin_unlock(&inode_lock);
746 /* Return the locked inode with I_NEW set, the
747 * caller is responsible for filling in the contents
753 * Uhhuh, somebody else created the same inode under
754 * us. Use the old inode instead of the one we just
758 spin_unlock(&inode_lock);
759 destroy_inode(inode);
761 wait_on_inode(inode);
767 * iunique - get a unique inode number
769 * @max_reserved: highest reserved inode number
771 * Obtain an inode number that is unique on the system for a given
772 * superblock. This is used by file systems that have no natural
773 * permanent inode numbering system. An inode number is returned that
774 * is higher than the reserved limit but unique.
777 * With a large number of inodes live on the file system this function
778 * currently becomes quite slow.
780 ino_t iunique(struct super_block *sb, ino_t max_reserved)
783 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
784 * error if st_ino won't fit in target struct field. Use 32bit counter
785 * here to attempt to avoid that.
787 static unsigned int counter;
789 struct hlist_head *head;
792 spin_lock(&inode_lock);
794 if (counter <= max_reserved)
795 counter = max_reserved + 1;
797 head = inode_hashtable + hash(sb, res);
798 inode = find_inode_fast(sb, head, res);
799 } while (inode != NULL);
800 spin_unlock(&inode_lock);
804 EXPORT_SYMBOL(iunique);
806 struct inode *igrab(struct inode *inode)
808 spin_lock(&inode_lock);
809 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
813 * Handle the case where s_op->clear_inode is not been
814 * called yet, and somebody is calling igrab
815 * while the inode is getting freed.
818 spin_unlock(&inode_lock);
822 EXPORT_SYMBOL(igrab);
825 * ifind - internal function, you want ilookup5() or iget5().
826 * @sb: super block of file system to search
827 * @head: the head of the list to search
828 * @test: callback used for comparisons between inodes
829 * @data: opaque data pointer to pass to @test
830 * @wait: if true wait for the inode to be unlocked, if false do not
832 * ifind() searches for the inode specified by @data in the inode
833 * cache. This is a generalized version of ifind_fast() for file systems where
834 * the inode number is not sufficient for unique identification of an inode.
836 * If the inode is in the cache, the inode is returned with an incremented
839 * Otherwise NULL is returned.
841 * Note, @test is called with the inode_lock held, so can't sleep.
843 static struct inode *ifind(struct super_block *sb,
844 struct hlist_head *head, int (*test)(struct inode *, void *),
845 void *data, const int wait)
849 spin_lock(&inode_lock);
850 inode = find_inode(sb, head, test, data);
853 spin_unlock(&inode_lock);
855 wait_on_inode(inode);
858 spin_unlock(&inode_lock);
863 * ifind_fast - internal function, you want ilookup() or iget().
864 * @sb: super block of file system to search
865 * @head: head of the list to search
866 * @ino: inode number to search for
868 * ifind_fast() searches for the inode @ino in the inode cache. This is for
869 * file systems where the inode number is sufficient for unique identification
872 * If the inode is in the cache, the inode is returned with an incremented
875 * Otherwise NULL is returned.
877 static struct inode *ifind_fast(struct super_block *sb,
878 struct hlist_head *head, unsigned long ino)
882 spin_lock(&inode_lock);
883 inode = find_inode_fast(sb, head, ino);
886 spin_unlock(&inode_lock);
887 wait_on_inode(inode);
890 spin_unlock(&inode_lock);
895 * ilookup5_nowait - search for an inode in the inode cache
896 * @sb: super block of file system to search
897 * @hashval: hash value (usually inode number) to search for
898 * @test: callback used for comparisons between inodes
899 * @data: opaque data pointer to pass to @test
901 * ilookup5() uses ifind() to search for the inode specified by @hashval and
902 * @data in the inode cache. This is a generalized version of ilookup() for
903 * file systems where the inode number is not sufficient for unique
904 * identification of an inode.
906 * If the inode is in the cache, the inode is returned with an incremented
907 * reference count. Note, the inode lock is not waited upon so you have to be
908 * very careful what you do with the returned inode. You probably should be
909 * using ilookup5() instead.
911 * Otherwise NULL is returned.
913 * Note, @test is called with the inode_lock held, so can't sleep.
915 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
916 int (*test)(struct inode *, void *), void *data)
918 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
920 return ifind(sb, head, test, data, 0);
923 EXPORT_SYMBOL(ilookup5_nowait);
926 * ilookup5 - search for an inode in the inode cache
927 * @sb: super block of file system to search
928 * @hashval: hash value (usually inode number) to search for
929 * @test: callback used for comparisons between inodes
930 * @data: opaque data pointer to pass to @test
932 * ilookup5() uses ifind() to search for the inode specified by @hashval and
933 * @data in the inode cache. This is a generalized version of ilookup() for
934 * file systems where the inode number is not sufficient for unique
935 * identification of an inode.
937 * If the inode is in the cache, the inode lock is waited upon and the inode is
938 * returned with an incremented reference count.
940 * Otherwise NULL is returned.
942 * Note, @test is called with the inode_lock held, so can't sleep.
944 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
945 int (*test)(struct inode *, void *), void *data)
947 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
949 return ifind(sb, head, test, data, 1);
952 EXPORT_SYMBOL(ilookup5);
955 * ilookup - search for an inode in the inode cache
956 * @sb: super block of file system to search
957 * @ino: inode number to search for
959 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
960 * This is for file systems where the inode number is sufficient for unique
961 * identification of an inode.
963 * If the inode is in the cache, the inode is returned with an incremented
966 * Otherwise NULL is returned.
968 struct inode *ilookup(struct super_block *sb, unsigned long ino)
970 struct hlist_head *head = inode_hashtable + hash(sb, ino);
972 return ifind_fast(sb, head, ino);
975 EXPORT_SYMBOL(ilookup);
978 * iget5_locked - obtain an inode from a mounted file system
979 * @sb: super block of file system
980 * @hashval: hash value (usually inode number) to get
981 * @test: callback used for comparisons between inodes
982 * @set: callback used to initialize a new struct inode
983 * @data: opaque data pointer to pass to @test and @set
985 * iget5_locked() uses ifind() to search for the inode specified by @hashval
986 * and @data in the inode cache and if present it is returned with an increased
987 * reference count. This is a generalized version of iget_locked() for file
988 * systems where the inode number is not sufficient for unique identification
991 * If the inode is not in cache, get_new_inode() is called to allocate a new
992 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
993 * file system gets to fill it in before unlocking it via unlock_new_inode().
995 * Note both @test and @set are called with the inode_lock held, so can't sleep.
997 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
998 int (*test)(struct inode *, void *),
999 int (*set)(struct inode *, void *), void *data)
1001 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1002 struct inode *inode;
1004 inode = ifind(sb, head, test, data, 1);
1008 * get_new_inode() will do the right thing, re-trying the search
1009 * in case it had to block at any point.
1011 return get_new_inode(sb, head, test, set, data);
1014 EXPORT_SYMBOL(iget5_locked);
1017 * iget_locked - obtain an inode from a mounted file system
1018 * @sb: super block of file system
1019 * @ino: inode number to get
1021 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1022 * the inode cache and if present it is returned with an increased reference
1023 * count. This is for file systems where the inode number is sufficient for
1024 * unique identification of an inode.
1026 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1027 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1028 * The file system gets to fill it in before unlocking it via
1029 * unlock_new_inode().
1031 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1033 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1034 struct inode *inode;
1036 inode = ifind_fast(sb, head, ino);
1040 * get_new_inode_fast() will do the right thing, re-trying the search
1041 * in case it had to block at any point.
1043 return get_new_inode_fast(sb, head, ino);
1046 EXPORT_SYMBOL(iget_locked);
1048 int insert_inode_locked(struct inode *inode)
1050 struct super_block *sb = inode->i_sb;
1051 ino_t ino = inode->i_ino;
1052 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1055 inode->i_state |= I_LOCK|I_NEW;
1057 spin_lock(&inode_lock);
1058 old = find_inode_fast(sb, head, ino);
1060 hlist_add_head(&inode->i_hash, head);
1061 spin_unlock(&inode_lock);
1065 spin_unlock(&inode_lock);
1067 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1075 EXPORT_SYMBOL(insert_inode_locked);
1077 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1078 int (*test)(struct inode *, void *), void *data)
1080 struct super_block *sb = inode->i_sb;
1081 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1084 inode->i_state |= I_LOCK|I_NEW;
1087 spin_lock(&inode_lock);
1088 old = find_inode(sb, head, test, data);
1090 hlist_add_head(&inode->i_hash, head);
1091 spin_unlock(&inode_lock);
1095 spin_unlock(&inode_lock);
1097 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1105 EXPORT_SYMBOL(insert_inode_locked4);
1108 * __insert_inode_hash - hash an inode
1109 * @inode: unhashed inode
1110 * @hashval: unsigned long value used to locate this object in the
1113 * Add an inode to the inode hash for this superblock.
1115 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1117 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1118 spin_lock(&inode_lock);
1119 hlist_add_head(&inode->i_hash, head);
1120 spin_unlock(&inode_lock);
1123 EXPORT_SYMBOL(__insert_inode_hash);
1126 * remove_inode_hash - remove an inode from the hash
1127 * @inode: inode to unhash
1129 * Remove an inode from the superblock.
1131 void remove_inode_hash(struct inode *inode)
1133 spin_lock(&inode_lock);
1134 hlist_del_init(&inode->i_hash);
1135 spin_unlock(&inode_lock);
1138 EXPORT_SYMBOL(remove_inode_hash);
1141 * Tell the filesystem that this inode is no longer of any interest and should
1142 * be completely destroyed.
1144 * We leave the inode in the inode hash table until *after* the filesystem's
1145 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1146 * instigate) will always find up-to-date information either in the hash or on
1149 * I_FREEING is set so that no-one will take a new reference to the inode while
1150 * it is being deleted.
1152 void generic_delete_inode(struct inode *inode)
1154 const struct super_operations *op = inode->i_sb->s_op;
1156 list_del_init(&inode->i_list);
1157 list_del_init(&inode->i_sb_list);
1158 inode->i_state |= I_FREEING;
1159 inodes_stat.nr_inodes--;
1160 spin_unlock(&inode_lock);
1162 security_inode_delete(inode);
1164 if (op->delete_inode) {
1165 void (*delete)(struct inode *) = op->delete_inode;
1166 if (!is_bad_inode(inode))
1168 /* Filesystems implementing their own
1169 * s_op->delete_inode are required to call
1170 * truncate_inode_pages and clear_inode()
1174 truncate_inode_pages(&inode->i_data, 0);
1177 spin_lock(&inode_lock);
1178 hlist_del_init(&inode->i_hash);
1179 spin_unlock(&inode_lock);
1180 wake_up_inode(inode);
1181 BUG_ON(inode->i_state != I_CLEAR);
1182 destroy_inode(inode);
1185 EXPORT_SYMBOL(generic_delete_inode);
1187 static void generic_forget_inode(struct inode *inode)
1189 struct super_block *sb = inode->i_sb;
1191 if (!hlist_unhashed(&inode->i_hash)) {
1192 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1193 list_move(&inode->i_list, &inode_unused);
1194 inodes_stat.nr_unused++;
1195 if (sb->s_flags & MS_ACTIVE) {
1196 spin_unlock(&inode_lock);
1199 inode->i_state |= I_WILL_FREE;
1200 spin_unlock(&inode_lock);
1201 write_inode_now(inode, 1);
1202 spin_lock(&inode_lock);
1203 inode->i_state &= ~I_WILL_FREE;
1204 inodes_stat.nr_unused--;
1205 hlist_del_init(&inode->i_hash);
1207 list_del_init(&inode->i_list);
1208 list_del_init(&inode->i_sb_list);
1209 inode->i_state |= I_FREEING;
1210 inodes_stat.nr_inodes--;
1211 spin_unlock(&inode_lock);
1212 if (inode->i_data.nrpages)
1213 truncate_inode_pages(&inode->i_data, 0);
1215 wake_up_inode(inode);
1216 destroy_inode(inode);
1220 * Normal UNIX filesystem behaviour: delete the
1221 * inode when the usage count drops to zero, and
1224 void generic_drop_inode(struct inode *inode)
1226 if (!inode->i_nlink)
1227 generic_delete_inode(inode);
1229 generic_forget_inode(inode);
1232 EXPORT_SYMBOL_GPL(generic_drop_inode);
1235 * Called when we're dropping the last reference
1238 * Call the FS "drop()" function, defaulting to
1239 * the legacy UNIX filesystem behaviour..
1241 * NOTE! NOTE! NOTE! We're called with the inode lock
1242 * held, and the drop function is supposed to release
1245 static inline void iput_final(struct inode *inode)
1247 const struct super_operations *op = inode->i_sb->s_op;
1248 void (*drop)(struct inode *) = generic_drop_inode;
1250 if (op && op->drop_inode)
1251 drop = op->drop_inode;
1256 * iput - put an inode
1257 * @inode: inode to put
1259 * Puts an inode, dropping its usage count. If the inode use count hits
1260 * zero, the inode is then freed and may also be destroyed.
1262 * Consequently, iput() can sleep.
1264 void iput(struct inode *inode)
1267 BUG_ON(inode->i_state == I_CLEAR);
1269 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1274 EXPORT_SYMBOL(iput);
1277 * bmap - find a block number in a file
1278 * @inode: inode of file
1279 * @block: block to find
1281 * Returns the block number on the device holding the inode that
1282 * is the disk block number for the block of the file requested.
1283 * That is, asked for block 4 of inode 1 the function will return the
1284 * disk block relative to the disk start that holds that block of the
1287 sector_t bmap(struct inode * inode, sector_t block)
1290 if (inode->i_mapping->a_ops->bmap)
1291 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1294 EXPORT_SYMBOL(bmap);
1297 * touch_atime - update the access time
1298 * @mnt: mount the inode is accessed on
1299 * @dentry: dentry accessed
1301 * Update the accessed time on an inode and mark it for writeback.
1302 * This function automatically handles read only file systems and media,
1303 * as well as the "noatime" flag and inode specific "noatime" markers.
1305 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1307 struct inode *inode = dentry->d_inode;
1308 struct timespec now;
1310 if (mnt_want_write(mnt))
1312 if (inode->i_flags & S_NOATIME)
1314 if (IS_NOATIME(inode))
1316 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1319 if (mnt->mnt_flags & MNT_NOATIME)
1321 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1323 if (mnt->mnt_flags & MNT_RELATIME) {
1325 * With relative atime, only update atime if the previous
1326 * atime is earlier than either the ctime or mtime.
1328 if (timespec_compare(&inode->i_mtime, &inode->i_atime) < 0 &&
1329 timespec_compare(&inode->i_ctime, &inode->i_atime) < 0)
1333 now = current_fs_time(inode->i_sb);
1334 if (timespec_equal(&inode->i_atime, &now))
1337 inode->i_atime = now;
1338 mark_inode_dirty_sync(inode);
1340 mnt_drop_write(mnt);
1342 EXPORT_SYMBOL(touch_atime);
1345 * file_update_time - update mtime and ctime time
1346 * @file: file accessed
1348 * Update the mtime and ctime members of an inode and mark the inode
1349 * for writeback. Note that this function is meant exclusively for
1350 * usage in the file write path of filesystems, and filesystems may
1351 * choose to explicitly ignore update via this function with the
1352 * S_NOCTIME inode flag, e.g. for network filesystem where these
1353 * timestamps are handled by the server.
1356 void file_update_time(struct file *file)
1358 struct inode *inode = file->f_path.dentry->d_inode;
1359 struct timespec now;
1363 if (IS_NOCMTIME(inode))
1366 err = mnt_want_write(file->f_path.mnt);
1370 now = current_fs_time(inode->i_sb);
1371 if (!timespec_equal(&inode->i_mtime, &now)) {
1372 inode->i_mtime = now;
1376 if (!timespec_equal(&inode->i_ctime, &now)) {
1377 inode->i_ctime = now;
1381 if (IS_I_VERSION(inode)) {
1382 inode_inc_iversion(inode);
1387 mark_inode_dirty_sync(inode);
1388 mnt_drop_write(file->f_path.mnt);
1391 EXPORT_SYMBOL(file_update_time);
1393 int inode_needs_sync(struct inode *inode)
1397 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1402 EXPORT_SYMBOL(inode_needs_sync);
1404 int inode_wait(void *word)
1409 EXPORT_SYMBOL(inode_wait);
1412 * If we try to find an inode in the inode hash while it is being
1413 * deleted, we have to wait until the filesystem completes its
1414 * deletion before reporting that it isn't found. This function waits
1415 * until the deletion _might_ have completed. Callers are responsible
1416 * to recheck inode state.
1418 * It doesn't matter if I_LOCK is not set initially, a call to
1419 * wake_up_inode() after removing from the hash list will DTRT.
1421 * This is called with inode_lock held.
1423 static void __wait_on_freeing_inode(struct inode *inode)
1425 wait_queue_head_t *wq;
1426 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1427 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1428 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1429 spin_unlock(&inode_lock);
1431 finish_wait(wq, &wait.wait);
1432 spin_lock(&inode_lock);
1436 * We rarely want to lock two inodes that do not have a parent/child
1437 * relationship (such as directory, child inode) simultaneously. The
1438 * vast majority of file systems should be able to get along fine
1439 * without this. Do not use these functions except as a last resort.
1441 void inode_double_lock(struct inode *inode1, struct inode *inode2)
1443 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1445 mutex_lock(&inode1->i_mutex);
1447 mutex_lock(&inode2->i_mutex);
1451 if (inode1 < inode2) {
1452 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1453 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1455 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1456 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1459 EXPORT_SYMBOL(inode_double_lock);
1461 void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1464 mutex_unlock(&inode1->i_mutex);
1466 if (inode2 && inode2 != inode1)
1467 mutex_unlock(&inode2->i_mutex);
1469 EXPORT_SYMBOL(inode_double_unlock);
1471 static __initdata unsigned long ihash_entries;
1472 static int __init set_ihash_entries(char *str)
1476 ihash_entries = simple_strtoul(str, &str, 0);
1479 __setup("ihash_entries=", set_ihash_entries);
1482 * Initialize the waitqueues and inode hash table.
1484 void __init inode_init_early(void)
1488 /* If hashes are distributed across NUMA nodes, defer
1489 * hash allocation until vmalloc space is available.
1495 alloc_large_system_hash("Inode-cache",
1496 sizeof(struct hlist_head),
1504 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1505 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1508 void __init inode_init(void)
1512 /* inode slab cache */
1513 inode_cachep = kmem_cache_create("inode_cache",
1514 sizeof(struct inode),
1516 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1519 register_shrinker(&icache_shrinker);
1521 /* Hash may have been set up in inode_init_early */
1526 alloc_large_system_hash("Inode-cache",
1527 sizeof(struct hlist_head),
1535 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1536 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1539 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1541 inode->i_mode = mode;
1542 if (S_ISCHR(mode)) {
1543 inode->i_fop = &def_chr_fops;
1544 inode->i_rdev = rdev;
1545 } else if (S_ISBLK(mode)) {
1546 inode->i_fop = &def_blk_fops;
1547 inode->i_rdev = rdev;
1548 } else if (S_ISFIFO(mode))
1549 inode->i_fop = &def_fifo_fops;
1550 else if (S_ISSOCK(mode))
1551 inode->i_fop = &bad_sock_fops;
1553 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1556 EXPORT_SYMBOL(init_special_inode);