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/fsnotify.h>
26 #include <linux/mount.h>
27 #include <linux/async.h>
30 * This is needed for the following functions:
32 * - invalidate_inode_buffers
35 * FIXME: remove all knowledge of the buffer layer from this file
37 #include <linux/buffer_head.h>
40 * New inode.c implementation.
42 * This implementation has the basic premise of trying
43 * to be extremely low-overhead and SMP-safe, yet be
44 * simple enough to be "obviously correct".
49 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
51 /* #define INODE_PARANOIA 1 */
52 /* #define INODE_DEBUG 1 */
55 * Inode lookup is no longer as critical as it used to be:
56 * most of the lookups are going to be through the dcache.
58 #define I_HASHBITS i_hash_shift
59 #define I_HASHMASK i_hash_mask
61 static unsigned int i_hash_mask __read_mostly;
62 static unsigned int i_hash_shift __read_mostly;
65 * Each inode can be on two separate lists. One is
66 * the hash list of the inode, used for lookups. The
67 * other linked list is the "type" list:
68 * "in_use" - valid inode, i_count > 0, i_nlink > 0
69 * "dirty" - as "in_use" but also dirty
70 * "unused" - valid inode, i_count = 0
72 * A "dirty" list is maintained for each super block,
73 * allowing for low-overhead inode sync() operations.
76 LIST_HEAD(inode_in_use);
77 LIST_HEAD(inode_unused);
78 static struct hlist_head *inode_hashtable __read_mostly;
81 * A simple spinlock to protect the list manipulations.
83 * NOTE! You also have to own the lock if you change
84 * the i_state of an inode while it is in use..
86 DEFINE_SPINLOCK(inode_lock);
89 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
90 * icache shrinking path, and the umount path. Without this exclusion,
91 * by the time prune_icache calls iput for the inode whose pages it has
92 * been invalidating, or by the time it calls clear_inode & destroy_inode
93 * from its final dispose_list, the struct super_block they refer to
94 * (for inode->i_sb->s_op) may already have been freed and reused.
96 static DEFINE_MUTEX(iprune_mutex);
99 * Statistics gathering..
101 struct inodes_stat_t inodes_stat;
103 static struct kmem_cache *inode_cachep __read_mostly;
105 static void wake_up_inode(struct inode *inode)
108 * Prevent speculative execution through spin_unlock(&inode_lock);
111 wake_up_bit(&inode->i_state, __I_LOCK);
115 * inode_init_always - perform inode structure intialisation
116 * @sb: superblock inode belongs to
117 * @inode: inode to initialise
119 * These are initializations that need to be done on every inode
120 * allocation as the fields are not initialised by slab allocation.
122 struct inode *inode_init_always(struct super_block *sb, struct inode *inode)
124 static const struct address_space_operations empty_aops;
125 static struct inode_operations empty_iops;
126 static const struct file_operations empty_fops;
128 struct address_space *const mapping = &inode->i_data;
131 inode->i_blkbits = sb->s_blocksize_bits;
133 atomic_set(&inode->i_count, 1);
134 inode->i_op = &empty_iops;
135 inode->i_fop = &empty_fops;
139 atomic_set(&inode->i_writecount, 0);
143 inode->i_generation = 0;
145 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
147 inode->i_pipe = NULL;
148 inode->i_bdev = NULL;
149 inode->i_cdev = NULL;
151 inode->dirtied_when = 0;
153 if (security_inode_alloc(inode))
156 /* allocate and initialize an i_integrity */
157 if (ima_inode_alloc(inode))
158 goto out_free_security;
160 spin_lock_init(&inode->i_lock);
161 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
163 mutex_init(&inode->i_mutex);
164 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
166 init_rwsem(&inode->i_alloc_sem);
167 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
169 mapping->a_ops = &empty_aops;
170 mapping->host = inode;
172 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
173 mapping->assoc_mapping = NULL;
174 mapping->backing_dev_info = &default_backing_dev_info;
175 mapping->writeback_index = 0;
178 * If the block_device provides a backing_dev_info for client
179 * inodes then use that. Otherwise the inode share the bdev's
183 struct backing_dev_info *bdi;
185 bdi = sb->s_bdev->bd_inode_backing_dev_info;
187 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
188 mapping->backing_dev_info = bdi;
190 inode->i_private = NULL;
191 inode->i_mapping = mapping;
193 #ifdef CONFIG_FSNOTIFY
194 inode->i_fsnotify_mask = 0;
200 security_inode_free(inode);
202 if (inode->i_sb->s_op->destroy_inode)
203 inode->i_sb->s_op->destroy_inode(inode);
205 kmem_cache_free(inode_cachep, (inode));
208 EXPORT_SYMBOL(inode_init_always);
210 static struct inode *alloc_inode(struct super_block *sb)
214 if (sb->s_op->alloc_inode)
215 inode = sb->s_op->alloc_inode(sb);
217 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
220 return inode_init_always(sb, inode);
224 void destroy_inode(struct inode *inode)
226 BUG_ON(inode_has_buffers(inode));
227 ima_inode_free(inode);
228 security_inode_free(inode);
229 fsnotify_inode_delete(inode);
230 if (inode->i_sb->s_op->destroy_inode)
231 inode->i_sb->s_op->destroy_inode(inode);
233 kmem_cache_free(inode_cachep, (inode));
235 EXPORT_SYMBOL(destroy_inode);
239 * These are initializations that only need to be done
240 * once, because the fields are idempotent across use
241 * of the inode, so let the slab aware of that.
243 void inode_init_once(struct inode *inode)
245 memset(inode, 0, sizeof(*inode));
246 INIT_HLIST_NODE(&inode->i_hash);
247 INIT_LIST_HEAD(&inode->i_dentry);
248 INIT_LIST_HEAD(&inode->i_devices);
249 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
250 spin_lock_init(&inode->i_data.tree_lock);
251 spin_lock_init(&inode->i_data.i_mmap_lock);
252 INIT_LIST_HEAD(&inode->i_data.private_list);
253 spin_lock_init(&inode->i_data.private_lock);
254 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
255 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
256 i_size_ordered_init(inode);
257 #ifdef CONFIG_INOTIFY
258 INIT_LIST_HEAD(&inode->inotify_watches);
259 mutex_init(&inode->inotify_mutex);
261 #ifdef CONFIG_FSNOTIFY
262 INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries);
265 EXPORT_SYMBOL(inode_init_once);
267 static void init_once(void *foo)
269 struct inode *inode = (struct inode *) foo;
271 inode_init_once(inode);
275 * inode_lock must be held
277 void __iget(struct inode *inode)
279 if (atomic_read(&inode->i_count)) {
280 atomic_inc(&inode->i_count);
283 atomic_inc(&inode->i_count);
284 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
285 list_move(&inode->i_list, &inode_in_use);
286 inodes_stat.nr_unused--;
290 * clear_inode - clear an inode
291 * @inode: inode to clear
293 * This is called by the filesystem to tell us
294 * that the inode is no longer useful. We just
295 * terminate it with extreme prejudice.
297 void clear_inode(struct inode *inode)
300 invalidate_inode_buffers(inode);
302 BUG_ON(inode->i_data.nrpages);
303 BUG_ON(!(inode->i_state & I_FREEING));
304 BUG_ON(inode->i_state & I_CLEAR);
305 inode_sync_wait(inode);
307 if (inode->i_sb->s_op->clear_inode)
308 inode->i_sb->s_op->clear_inode(inode);
309 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
311 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
313 inode->i_state = I_CLEAR;
315 EXPORT_SYMBOL(clear_inode);
318 * dispose_list - dispose of the contents of a local list
319 * @head: the head of the list to free
321 * Dispose-list gets a local list with local inodes in it, so it doesn't
322 * need to worry about list corruption and SMP locks.
324 static void dispose_list(struct list_head *head)
328 while (!list_empty(head)) {
331 inode = list_first_entry(head, struct inode, i_list);
332 list_del(&inode->i_list);
334 if (inode->i_data.nrpages)
335 truncate_inode_pages(&inode->i_data, 0);
338 spin_lock(&inode_lock);
339 hlist_del_init(&inode->i_hash);
340 list_del_init(&inode->i_sb_list);
341 spin_unlock(&inode_lock);
343 wake_up_inode(inode);
344 destroy_inode(inode);
347 spin_lock(&inode_lock);
348 inodes_stat.nr_inodes -= nr_disposed;
349 spin_unlock(&inode_lock);
353 * Invalidate all inodes for a device.
355 static int invalidate_list(struct list_head *head, struct list_head *dispose)
357 struct list_head *next;
358 int busy = 0, count = 0;
362 struct list_head *tmp = next;
366 * We can reschedule here without worrying about the list's
367 * consistency because the per-sb list of inodes must not
368 * change during umount anymore, and because iprune_mutex keeps
369 * shrink_icache_memory() away.
371 cond_resched_lock(&inode_lock);
376 inode = list_entry(tmp, struct inode, i_sb_list);
377 if (inode->i_state & I_NEW)
379 invalidate_inode_buffers(inode);
380 if (!atomic_read(&inode->i_count)) {
381 list_move(&inode->i_list, dispose);
382 WARN_ON(inode->i_state & I_NEW);
383 inode->i_state |= I_FREEING;
389 /* only unused inodes may be cached with i_count zero */
390 inodes_stat.nr_unused -= count;
395 * invalidate_inodes - discard the inodes on a device
398 * Discard all of the inodes for a given superblock. If the discard
399 * fails because there are busy inodes then a non zero value is returned.
400 * If the discard is successful all the inodes have been discarded.
402 int invalidate_inodes(struct super_block *sb)
405 LIST_HEAD(throw_away);
407 mutex_lock(&iprune_mutex);
408 spin_lock(&inode_lock);
409 inotify_unmount_inodes(&sb->s_inodes);
410 busy = invalidate_list(&sb->s_inodes, &throw_away);
411 spin_unlock(&inode_lock);
413 dispose_list(&throw_away);
414 mutex_unlock(&iprune_mutex);
418 EXPORT_SYMBOL(invalidate_inodes);
420 static int can_unuse(struct inode *inode)
424 if (inode_has_buffers(inode))
426 if (atomic_read(&inode->i_count))
428 if (inode->i_data.nrpages)
434 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
435 * a temporary list and then are freed outside inode_lock by dispose_list().
437 * Any inodes which are pinned purely because of attached pagecache have their
438 * pagecache removed. We expect the final iput() on that inode to add it to
439 * the front of the inode_unused list. So look for it there and if the
440 * inode is still freeable, proceed. The right inode is found 99.9% of the
441 * time in testing on a 4-way.
443 * If the inode has metadata buffers attached to mapping->private_list then
444 * try to remove them.
446 static void prune_icache(int nr_to_scan)
451 unsigned long reap = 0;
453 mutex_lock(&iprune_mutex);
454 spin_lock(&inode_lock);
455 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
458 if (list_empty(&inode_unused))
461 inode = list_entry(inode_unused.prev, struct inode, i_list);
463 if (inode->i_state || atomic_read(&inode->i_count)) {
464 list_move(&inode->i_list, &inode_unused);
467 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
469 spin_unlock(&inode_lock);
470 if (remove_inode_buffers(inode))
471 reap += invalidate_mapping_pages(&inode->i_data,
474 spin_lock(&inode_lock);
476 if (inode != list_entry(inode_unused.next,
477 struct inode, i_list))
478 continue; /* wrong inode or list_empty */
479 if (!can_unuse(inode))
482 list_move(&inode->i_list, &freeable);
483 WARN_ON(inode->i_state & I_NEW);
484 inode->i_state |= I_FREEING;
487 inodes_stat.nr_unused -= nr_pruned;
488 if (current_is_kswapd())
489 __count_vm_events(KSWAPD_INODESTEAL, reap);
491 __count_vm_events(PGINODESTEAL, reap);
492 spin_unlock(&inode_lock);
494 dispose_list(&freeable);
495 mutex_unlock(&iprune_mutex);
499 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
500 * "unused" means that no dentries are referring to the inodes: the files are
501 * not open and the dcache references to those inodes have already been
504 * This function is passed the number of inodes to scan, and it returns the
505 * total number of remaining possibly-reclaimable inodes.
507 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
511 * Nasty deadlock avoidance. We may hold various FS locks,
512 * and we don't want to recurse into the FS that called us
513 * in clear_inode() and friends..
515 if (!(gfp_mask & __GFP_FS))
519 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
522 static struct shrinker icache_shrinker = {
523 .shrink = shrink_icache_memory,
524 .seeks = DEFAULT_SEEKS,
527 static void __wait_on_freeing_inode(struct inode *inode);
529 * Called with the inode lock held.
530 * NOTE: we are not increasing the inode-refcount, you must call __iget()
531 * by hand after calling find_inode now! This simplifies iunique and won't
532 * add any additional branch in the common code.
534 static struct inode *find_inode(struct super_block *sb,
535 struct hlist_head *head,
536 int (*test)(struct inode *, void *),
539 struct hlist_node *node;
540 struct inode *inode = NULL;
543 hlist_for_each_entry(inode, node, head, i_hash) {
544 if (inode->i_sb != sb)
546 if (!test(inode, data))
548 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
549 __wait_on_freeing_inode(inode);
554 return node ? inode : NULL;
558 * find_inode_fast is the fast path version of find_inode, see the comment at
559 * iget_locked for details.
561 static struct inode *find_inode_fast(struct super_block *sb,
562 struct hlist_head *head, unsigned long ino)
564 struct hlist_node *node;
565 struct inode *inode = NULL;
568 hlist_for_each_entry(inode, node, head, i_hash) {
569 if (inode->i_ino != ino)
571 if (inode->i_sb != sb)
573 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
574 __wait_on_freeing_inode(inode);
579 return node ? inode : NULL;
582 static unsigned long hash(struct super_block *sb, unsigned long hashval)
586 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
588 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
589 return tmp & I_HASHMASK;
593 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
596 inodes_stat.nr_inodes++;
597 list_add(&inode->i_list, &inode_in_use);
598 list_add(&inode->i_sb_list, &sb->s_inodes);
600 hlist_add_head(&inode->i_hash, head);
604 * inode_add_to_lists - add a new inode to relevant lists
605 * @sb: superblock inode belongs to
606 * @inode: inode to mark in use
608 * When an inode is allocated it needs to be accounted for, added to the in use
609 * list, the owning superblock and the inode hash. This needs to be done under
610 * the inode_lock, so export a function to do this rather than the inode lock
611 * itself. We calculate the hash list to add to here so it is all internal
612 * which requires the caller to have already set up the inode number in the
615 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
617 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
619 spin_lock(&inode_lock);
620 __inode_add_to_lists(sb, head, inode);
621 spin_unlock(&inode_lock);
623 EXPORT_SYMBOL_GPL(inode_add_to_lists);
626 * new_inode - obtain an inode
629 * Allocates a new inode for given superblock. The default gfp_mask
630 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
631 * If HIGHMEM pages are unsuitable or it is known that pages allocated
632 * for the page cache are not reclaimable or migratable,
633 * mapping_set_gfp_mask() must be called with suitable flags on the
634 * newly created inode's mapping
637 struct inode *new_inode(struct super_block *sb)
640 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
641 * error if st_ino won't fit in target struct field. Use 32bit counter
642 * here to attempt to avoid that.
644 static unsigned int last_ino;
647 spin_lock_prefetch(&inode_lock);
649 inode = alloc_inode(sb);
651 spin_lock(&inode_lock);
652 __inode_add_to_lists(sb, NULL, inode);
653 inode->i_ino = ++last_ino;
655 spin_unlock(&inode_lock);
659 EXPORT_SYMBOL(new_inode);
661 void unlock_new_inode(struct inode *inode)
663 #ifdef CONFIG_DEBUG_LOCK_ALLOC
664 if (inode->i_mode & S_IFDIR) {
665 struct file_system_type *type = inode->i_sb->s_type;
668 * ensure nobody is actually holding i_mutex
670 mutex_destroy(&inode->i_mutex);
671 mutex_init(&inode->i_mutex);
672 lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key);
676 * This is special! We do not need the spinlock
677 * when clearing I_LOCK, because we're guaranteed
678 * that nobody else tries to do anything about the
679 * state of the inode when it is locked, as we
680 * just created it (so there can be no old holders
681 * that haven't tested I_LOCK).
683 WARN_ON((inode->i_state & (I_LOCK|I_NEW)) != (I_LOCK|I_NEW));
684 inode->i_state &= ~(I_LOCK|I_NEW);
685 wake_up_inode(inode);
687 EXPORT_SYMBOL(unlock_new_inode);
690 * This is called without the inode lock held.. Be careful.
692 * We no longer cache the sb_flags in i_flags - see fs.h
693 * -- rmk@arm.uk.linux.org
695 static struct inode *get_new_inode(struct super_block *sb,
696 struct hlist_head *head,
697 int (*test)(struct inode *, void *),
698 int (*set)(struct inode *, void *),
703 inode = alloc_inode(sb);
707 spin_lock(&inode_lock);
708 /* We released the lock, so.. */
709 old = find_inode(sb, head, test, data);
711 if (set(inode, data))
714 __inode_add_to_lists(sb, head, inode);
715 inode->i_state = I_LOCK|I_NEW;
716 spin_unlock(&inode_lock);
718 /* Return the locked inode with I_NEW set, the
719 * caller is responsible for filling in the contents
725 * Uhhuh, somebody else created the same inode under
726 * us. Use the old inode instead of the one we just
730 spin_unlock(&inode_lock);
731 destroy_inode(inode);
733 wait_on_inode(inode);
738 spin_unlock(&inode_lock);
739 destroy_inode(inode);
744 * get_new_inode_fast is the fast path version of get_new_inode, see the
745 * comment at iget_locked for details.
747 static struct inode *get_new_inode_fast(struct super_block *sb,
748 struct hlist_head *head, unsigned long ino)
752 inode = alloc_inode(sb);
756 spin_lock(&inode_lock);
757 /* We released the lock, so.. */
758 old = find_inode_fast(sb, head, ino);
761 __inode_add_to_lists(sb, head, inode);
762 inode->i_state = I_LOCK|I_NEW;
763 spin_unlock(&inode_lock);
765 /* Return the locked inode with I_NEW set, the
766 * caller is responsible for filling in the contents
772 * Uhhuh, somebody else created the same inode under
773 * us. Use the old inode instead of the one we just
777 spin_unlock(&inode_lock);
778 destroy_inode(inode);
780 wait_on_inode(inode);
786 * iunique - get a unique inode number
788 * @max_reserved: highest reserved inode number
790 * Obtain an inode number that is unique on the system for a given
791 * superblock. This is used by file systems that have no natural
792 * permanent inode numbering system. An inode number is returned that
793 * is higher than the reserved limit but unique.
796 * With a large number of inodes live on the file system this function
797 * currently becomes quite slow.
799 ino_t iunique(struct super_block *sb, ino_t max_reserved)
802 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
803 * error if st_ino won't fit in target struct field. Use 32bit counter
804 * here to attempt to avoid that.
806 static unsigned int counter;
808 struct hlist_head *head;
811 spin_lock(&inode_lock);
813 if (counter <= max_reserved)
814 counter = max_reserved + 1;
816 head = inode_hashtable + hash(sb, res);
817 inode = find_inode_fast(sb, head, res);
818 } while (inode != NULL);
819 spin_unlock(&inode_lock);
823 EXPORT_SYMBOL(iunique);
825 struct inode *igrab(struct inode *inode)
827 spin_lock(&inode_lock);
828 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
832 * Handle the case where s_op->clear_inode is not been
833 * called yet, and somebody is calling igrab
834 * while the inode is getting freed.
837 spin_unlock(&inode_lock);
840 EXPORT_SYMBOL(igrab);
843 * ifind - internal function, you want ilookup5() or iget5().
844 * @sb: super block of file system to search
845 * @head: the head of the list to search
846 * @test: callback used for comparisons between inodes
847 * @data: opaque data pointer to pass to @test
848 * @wait: if true wait for the inode to be unlocked, if false do not
850 * ifind() searches for the inode specified by @data in the inode
851 * cache. This is a generalized version of ifind_fast() for file systems where
852 * the inode number is not sufficient for unique identification of an inode.
854 * If the inode is in the cache, the inode is returned with an incremented
857 * Otherwise NULL is returned.
859 * Note, @test is called with the inode_lock held, so can't sleep.
861 static struct inode *ifind(struct super_block *sb,
862 struct hlist_head *head, int (*test)(struct inode *, void *),
863 void *data, const int wait)
867 spin_lock(&inode_lock);
868 inode = find_inode(sb, head, test, data);
871 spin_unlock(&inode_lock);
873 wait_on_inode(inode);
876 spin_unlock(&inode_lock);
881 * ifind_fast - internal function, you want ilookup() or iget().
882 * @sb: super block of file system to search
883 * @head: head of the list to search
884 * @ino: inode number to search for
886 * ifind_fast() searches for the inode @ino in the inode cache. This is for
887 * file systems where the inode number is sufficient for unique identification
890 * If the inode is in the cache, the inode is returned with an incremented
893 * Otherwise NULL is returned.
895 static struct inode *ifind_fast(struct super_block *sb,
896 struct hlist_head *head, unsigned long ino)
900 spin_lock(&inode_lock);
901 inode = find_inode_fast(sb, head, ino);
904 spin_unlock(&inode_lock);
905 wait_on_inode(inode);
908 spin_unlock(&inode_lock);
913 * ilookup5_nowait - search for an inode in the inode cache
914 * @sb: super block of file system to search
915 * @hashval: hash value (usually inode number) to search for
916 * @test: callback used for comparisons between inodes
917 * @data: opaque data pointer to pass to @test
919 * ilookup5() uses ifind() to search for the inode specified by @hashval and
920 * @data in the inode cache. This is a generalized version of ilookup() for
921 * file systems where the inode number is not sufficient for unique
922 * identification of an inode.
924 * If the inode is in the cache, the inode is returned with an incremented
925 * reference count. Note, the inode lock is not waited upon so you have to be
926 * very careful what you do with the returned inode. You probably should be
927 * using ilookup5() instead.
929 * Otherwise NULL is returned.
931 * Note, @test is called with the inode_lock held, so can't sleep.
933 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
934 int (*test)(struct inode *, void *), void *data)
936 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
938 return ifind(sb, head, test, data, 0);
940 EXPORT_SYMBOL(ilookup5_nowait);
943 * ilookup5 - search for an inode in the inode cache
944 * @sb: super block of file system to search
945 * @hashval: hash value (usually inode number) to search for
946 * @test: callback used for comparisons between inodes
947 * @data: opaque data pointer to pass to @test
949 * ilookup5() uses ifind() to search for the inode specified by @hashval and
950 * @data in the inode cache. This is a generalized version of ilookup() for
951 * file systems where the inode number is not sufficient for unique
952 * identification of an inode.
954 * If the inode is in the cache, the inode lock is waited upon and the inode is
955 * returned with an incremented reference count.
957 * Otherwise NULL is returned.
959 * Note, @test is called with the inode_lock held, so can't sleep.
961 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
962 int (*test)(struct inode *, void *), void *data)
964 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
966 return ifind(sb, head, test, data, 1);
968 EXPORT_SYMBOL(ilookup5);
971 * ilookup - search for an inode in the inode cache
972 * @sb: super block of file system to search
973 * @ino: inode number to search for
975 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
976 * This is for file systems where the inode number is sufficient for unique
977 * identification of an inode.
979 * If the inode is in the cache, the inode is returned with an incremented
982 * Otherwise NULL is returned.
984 struct inode *ilookup(struct super_block *sb, unsigned long ino)
986 struct hlist_head *head = inode_hashtable + hash(sb, ino);
988 return ifind_fast(sb, head, ino);
990 EXPORT_SYMBOL(ilookup);
993 * iget5_locked - obtain an inode from a mounted file system
994 * @sb: super block of file system
995 * @hashval: hash value (usually inode number) to get
996 * @test: callback used for comparisons between inodes
997 * @set: callback used to initialize a new struct inode
998 * @data: opaque data pointer to pass to @test and @set
1000 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1001 * and @data in the inode cache and if present it is returned with an increased
1002 * reference count. This is a generalized version of iget_locked() for file
1003 * systems where the inode number is not sufficient for unique identification
1006 * If the inode is not in cache, get_new_inode() is called to allocate a new
1007 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1008 * file system gets to fill it in before unlocking it via unlock_new_inode().
1010 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1012 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1013 int (*test)(struct inode *, void *),
1014 int (*set)(struct inode *, void *), void *data)
1016 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1017 struct inode *inode;
1019 inode = ifind(sb, head, test, data, 1);
1023 * get_new_inode() will do the right thing, re-trying the search
1024 * in case it had to block at any point.
1026 return get_new_inode(sb, head, test, set, data);
1028 EXPORT_SYMBOL(iget5_locked);
1031 * iget_locked - obtain an inode from a mounted file system
1032 * @sb: super block of file system
1033 * @ino: inode number to get
1035 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1036 * the inode cache and if present it is returned with an increased reference
1037 * count. This is for file systems where the inode number is sufficient for
1038 * unique identification of an inode.
1040 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1041 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1042 * The file system gets to fill it in before unlocking it via
1043 * unlock_new_inode().
1045 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1047 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1048 struct inode *inode;
1050 inode = ifind_fast(sb, head, ino);
1054 * get_new_inode_fast() will do the right thing, re-trying the search
1055 * in case it had to block at any point.
1057 return get_new_inode_fast(sb, head, ino);
1059 EXPORT_SYMBOL(iget_locked);
1061 int insert_inode_locked(struct inode *inode)
1063 struct super_block *sb = inode->i_sb;
1064 ino_t ino = inode->i_ino;
1065 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1067 inode->i_state |= I_LOCK|I_NEW;
1069 struct hlist_node *node;
1070 struct inode *old = NULL;
1071 spin_lock(&inode_lock);
1072 hlist_for_each_entry(old, node, head, i_hash) {
1073 if (old->i_ino != ino)
1075 if (old->i_sb != sb)
1077 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1081 if (likely(!node)) {
1082 hlist_add_head(&inode->i_hash, head);
1083 spin_unlock(&inode_lock);
1087 spin_unlock(&inode_lock);
1089 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1096 EXPORT_SYMBOL(insert_inode_locked);
1098 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1099 int (*test)(struct inode *, void *), void *data)
1101 struct super_block *sb = inode->i_sb;
1102 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1104 inode->i_state |= I_LOCK|I_NEW;
1107 struct hlist_node *node;
1108 struct inode *old = NULL;
1110 spin_lock(&inode_lock);
1111 hlist_for_each_entry(old, node, head, i_hash) {
1112 if (old->i_sb != sb)
1114 if (!test(old, data))
1116 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1120 if (likely(!node)) {
1121 hlist_add_head(&inode->i_hash, head);
1122 spin_unlock(&inode_lock);
1126 spin_unlock(&inode_lock);
1128 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1135 EXPORT_SYMBOL(insert_inode_locked4);
1138 * __insert_inode_hash - hash an inode
1139 * @inode: unhashed inode
1140 * @hashval: unsigned long value used to locate this object in the
1143 * Add an inode to the inode hash for this superblock.
1145 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1147 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1148 spin_lock(&inode_lock);
1149 hlist_add_head(&inode->i_hash, head);
1150 spin_unlock(&inode_lock);
1152 EXPORT_SYMBOL(__insert_inode_hash);
1155 * remove_inode_hash - remove an inode from the hash
1156 * @inode: inode to unhash
1158 * Remove an inode from the superblock.
1160 void remove_inode_hash(struct inode *inode)
1162 spin_lock(&inode_lock);
1163 hlist_del_init(&inode->i_hash);
1164 spin_unlock(&inode_lock);
1166 EXPORT_SYMBOL(remove_inode_hash);
1169 * Tell the filesystem that this inode is no longer of any interest and should
1170 * be completely destroyed.
1172 * We leave the inode in the inode hash table until *after* the filesystem's
1173 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1174 * instigate) will always find up-to-date information either in the hash or on
1177 * I_FREEING is set so that no-one will take a new reference to the inode while
1178 * it is being deleted.
1180 void generic_delete_inode(struct inode *inode)
1182 const struct super_operations *op = inode->i_sb->s_op;
1184 list_del_init(&inode->i_list);
1185 list_del_init(&inode->i_sb_list);
1186 WARN_ON(inode->i_state & I_NEW);
1187 inode->i_state |= I_FREEING;
1188 inodes_stat.nr_inodes--;
1189 spin_unlock(&inode_lock);
1191 security_inode_delete(inode);
1193 if (op->delete_inode) {
1194 void (*delete)(struct inode *) = op->delete_inode;
1195 if (!is_bad_inode(inode))
1197 /* Filesystems implementing their own
1198 * s_op->delete_inode are required to call
1199 * truncate_inode_pages and clear_inode()
1203 truncate_inode_pages(&inode->i_data, 0);
1206 spin_lock(&inode_lock);
1207 hlist_del_init(&inode->i_hash);
1208 spin_unlock(&inode_lock);
1209 wake_up_inode(inode);
1210 BUG_ON(inode->i_state != I_CLEAR);
1211 destroy_inode(inode);
1213 EXPORT_SYMBOL(generic_delete_inode);
1215 static void generic_forget_inode(struct inode *inode)
1217 struct super_block *sb = inode->i_sb;
1219 if (!hlist_unhashed(&inode->i_hash)) {
1220 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1221 list_move(&inode->i_list, &inode_unused);
1222 inodes_stat.nr_unused++;
1223 if (sb->s_flags & MS_ACTIVE) {
1224 spin_unlock(&inode_lock);
1227 WARN_ON(inode->i_state & I_NEW);
1228 inode->i_state |= I_WILL_FREE;
1229 spin_unlock(&inode_lock);
1230 write_inode_now(inode, 1);
1231 spin_lock(&inode_lock);
1232 WARN_ON(inode->i_state & I_NEW);
1233 inode->i_state &= ~I_WILL_FREE;
1234 inodes_stat.nr_unused--;
1235 hlist_del_init(&inode->i_hash);
1237 list_del_init(&inode->i_list);
1238 list_del_init(&inode->i_sb_list);
1239 WARN_ON(inode->i_state & I_NEW);
1240 inode->i_state |= I_FREEING;
1241 inodes_stat.nr_inodes--;
1242 spin_unlock(&inode_lock);
1243 if (inode->i_data.nrpages)
1244 truncate_inode_pages(&inode->i_data, 0);
1246 wake_up_inode(inode);
1247 destroy_inode(inode);
1251 * Normal UNIX filesystem behaviour: delete the
1252 * inode when the usage count drops to zero, and
1255 void generic_drop_inode(struct inode *inode)
1257 if (!inode->i_nlink)
1258 generic_delete_inode(inode);
1260 generic_forget_inode(inode);
1262 EXPORT_SYMBOL_GPL(generic_drop_inode);
1265 * Called when we're dropping the last reference
1268 * Call the FS "drop()" function, defaulting to
1269 * the legacy UNIX filesystem behaviour..
1271 * NOTE! NOTE! NOTE! We're called with the inode lock
1272 * held, and the drop function is supposed to release
1275 static inline void iput_final(struct inode *inode)
1277 const struct super_operations *op = inode->i_sb->s_op;
1278 void (*drop)(struct inode *) = generic_drop_inode;
1280 if (op && op->drop_inode)
1281 drop = op->drop_inode;
1286 * iput - put an inode
1287 * @inode: inode to put
1289 * Puts an inode, dropping its usage count. If the inode use count hits
1290 * zero, the inode is then freed and may also be destroyed.
1292 * Consequently, iput() can sleep.
1294 void iput(struct inode *inode)
1297 BUG_ON(inode->i_state == I_CLEAR);
1299 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1303 EXPORT_SYMBOL(iput);
1306 * bmap - find a block number in a file
1307 * @inode: inode of file
1308 * @block: block to find
1310 * Returns the block number on the device holding the inode that
1311 * is the disk block number for the block of the file requested.
1312 * That is, asked for block 4 of inode 1 the function will return the
1313 * disk block relative to the disk start that holds that block of the
1316 sector_t bmap(struct inode *inode, sector_t block)
1319 if (inode->i_mapping->a_ops->bmap)
1320 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1323 EXPORT_SYMBOL(bmap);
1326 * With relative atime, only update atime if the previous atime is
1327 * earlier than either the ctime or mtime or if at least a day has
1328 * passed since the last atime update.
1330 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1331 struct timespec now)
1334 if (!(mnt->mnt_flags & MNT_RELATIME))
1337 * Is mtime younger than atime? If yes, update atime:
1339 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1342 * Is ctime younger than atime? If yes, update atime:
1344 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1348 * Is the previous atime value older than a day? If yes,
1351 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1354 * Good, we can skip the atime update:
1360 * touch_atime - update the access time
1361 * @mnt: mount the inode is accessed on
1362 * @dentry: dentry accessed
1364 * Update the accessed time on an inode and mark it for writeback.
1365 * This function automatically handles read only file systems and media,
1366 * as well as the "noatime" flag and inode specific "noatime" markers.
1368 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1370 struct inode *inode = dentry->d_inode;
1371 struct timespec now;
1373 if (mnt_want_write(mnt))
1375 if (inode->i_flags & S_NOATIME)
1377 if (IS_NOATIME(inode))
1379 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1382 if (mnt->mnt_flags & MNT_NOATIME)
1384 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1387 now = current_fs_time(inode->i_sb);
1389 if (!relatime_need_update(mnt, inode, now))
1392 if (timespec_equal(&inode->i_atime, &now))
1395 inode->i_atime = now;
1396 mark_inode_dirty_sync(inode);
1398 mnt_drop_write(mnt);
1400 EXPORT_SYMBOL(touch_atime);
1403 * file_update_time - update mtime and ctime time
1404 * @file: file accessed
1406 * Update the mtime and ctime members of an inode and mark the inode
1407 * for writeback. Note that this function is meant exclusively for
1408 * usage in the file write path of filesystems, and filesystems may
1409 * choose to explicitly ignore update via this function with the
1410 * S_NOCTIME inode flag, e.g. for network filesystem where these
1411 * timestamps are handled by the server.
1414 void file_update_time(struct file *file)
1416 struct inode *inode = file->f_path.dentry->d_inode;
1417 struct timespec now;
1421 if (IS_NOCMTIME(inode))
1424 err = mnt_want_write(file->f_path.mnt);
1428 now = current_fs_time(inode->i_sb);
1429 if (!timespec_equal(&inode->i_mtime, &now)) {
1430 inode->i_mtime = now;
1434 if (!timespec_equal(&inode->i_ctime, &now)) {
1435 inode->i_ctime = now;
1439 if (IS_I_VERSION(inode)) {
1440 inode_inc_iversion(inode);
1445 mark_inode_dirty_sync(inode);
1446 mnt_drop_write(file->f_path.mnt);
1448 EXPORT_SYMBOL(file_update_time);
1450 int inode_needs_sync(struct inode *inode)
1454 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1458 EXPORT_SYMBOL(inode_needs_sync);
1460 int inode_wait(void *word)
1465 EXPORT_SYMBOL(inode_wait);
1468 * If we try to find an inode in the inode hash while it is being
1469 * deleted, we have to wait until the filesystem completes its
1470 * deletion before reporting that it isn't found. This function waits
1471 * until the deletion _might_ have completed. Callers are responsible
1472 * to recheck inode state.
1474 * It doesn't matter if I_LOCK is not set initially, a call to
1475 * wake_up_inode() after removing from the hash list will DTRT.
1477 * This is called with inode_lock held.
1479 static void __wait_on_freeing_inode(struct inode *inode)
1481 wait_queue_head_t *wq;
1482 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1483 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1484 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1485 spin_unlock(&inode_lock);
1487 finish_wait(wq, &wait.wait);
1488 spin_lock(&inode_lock);
1491 static __initdata unsigned long ihash_entries;
1492 static int __init set_ihash_entries(char *str)
1496 ihash_entries = simple_strtoul(str, &str, 0);
1499 __setup("ihash_entries=", set_ihash_entries);
1502 * Initialize the waitqueues and inode hash table.
1504 void __init inode_init_early(void)
1508 /* If hashes are distributed across NUMA nodes, defer
1509 * hash allocation until vmalloc space is available.
1515 alloc_large_system_hash("Inode-cache",
1516 sizeof(struct hlist_head),
1524 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1525 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1528 void __init inode_init(void)
1532 /* inode slab cache */
1533 inode_cachep = kmem_cache_create("inode_cache",
1534 sizeof(struct inode),
1536 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1539 register_shrinker(&icache_shrinker);
1541 /* Hash may have been set up in inode_init_early */
1546 alloc_large_system_hash("Inode-cache",
1547 sizeof(struct hlist_head),
1555 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1556 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1559 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1561 inode->i_mode = mode;
1562 if (S_ISCHR(mode)) {
1563 inode->i_fop = &def_chr_fops;
1564 inode->i_rdev = rdev;
1565 } else if (S_ISBLK(mode)) {
1566 inode->i_fop = &def_blk_fops;
1567 inode->i_rdev = rdev;
1568 } else if (S_ISFIFO(mode))
1569 inode->i_fop = &def_fifo_fops;
1570 else if (S_ISSOCK(mode))
1571 inode->i_fop = &bad_sock_fops;
1573 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1576 EXPORT_SYMBOL(init_special_inode);