Rewrite MSI-HOWTO
[linux-2.6/mini2440.git] / fs / inode.c
blob826fb0b9d1c38f4dbe4ad16c6ec938a7e31cdc88
1 /*
2 * linux/fs/inode.c
4 * (C) 1997 Linus Torvalds
5 */
7 #include <linux/fs.h>
8 #include <linux/mm.h>
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/pagemap.h>
21 #include <linux/cdev.h>
22 #include <linux/bootmem.h>
23 #include <linux/inotify.h>
24 #include <linux/mount.h>
25 #include <linux/async.h>
28 * This is needed for the following functions:
29 * - inode_has_buffers
30 * - invalidate_inode_buffers
31 * - invalidate_bdev
33 * FIXME: remove all knowledge of the buffer layer from this file
35 #include <linux/buffer_head.h>
38 * New inode.c implementation.
40 * This implementation has the basic premise of trying
41 * to be extremely low-overhead and SMP-safe, yet be
42 * simple enough to be "obviously correct".
44 * Famous last words.
47 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
49 /* #define INODE_PARANOIA 1 */
50 /* #define INODE_DEBUG 1 */
53 * Inode lookup is no longer as critical as it used to be:
54 * most of the lookups are going to be through the dcache.
56 #define I_HASHBITS i_hash_shift
57 #define I_HASHMASK i_hash_mask
59 static unsigned int i_hash_mask __read_mostly;
60 static unsigned int i_hash_shift __read_mostly;
63 * Each inode can be on two separate lists. One is
64 * the hash list of the inode, used for lookups. The
65 * other linked list is the "type" list:
66 * "in_use" - valid inode, i_count > 0, i_nlink > 0
67 * "dirty" - as "in_use" but also dirty
68 * "unused" - valid inode, i_count = 0
70 * A "dirty" list is maintained for each super block,
71 * allowing for low-overhead inode sync() operations.
74 LIST_HEAD(inode_in_use);
75 LIST_HEAD(inode_unused);
76 static struct hlist_head *inode_hashtable __read_mostly;
79 * A simple spinlock to protect the list manipulations.
81 * NOTE! You also have to own the lock if you change
82 * the i_state of an inode while it is in use..
84 DEFINE_SPINLOCK(inode_lock);
87 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
88 * icache shrinking path, and the umount path. Without this exclusion,
89 * by the time prune_icache calls iput for the inode whose pages it has
90 * been invalidating, or by the time it calls clear_inode & destroy_inode
91 * from its final dispose_list, the struct super_block they refer to
92 * (for inode->i_sb->s_op) may already have been freed and reused.
94 static DEFINE_MUTEX(iprune_mutex);
97 * Statistics gathering..
99 struct inodes_stat_t inodes_stat;
101 static struct kmem_cache * inode_cachep __read_mostly;
103 static void wake_up_inode(struct inode *inode)
106 * Prevent speculative execution through spin_unlock(&inode_lock);
108 smp_mb();
109 wake_up_bit(&inode->i_state, __I_LOCK);
113 * inode_init_always - perform inode structure intialisation
114 * @sb: superblock inode belongs to
115 * @inode: inode to initialise
117 * These are initializations that need to be done on every inode
118 * allocation as the fields are not initialised by slab allocation.
120 struct inode *inode_init_always(struct super_block *sb, struct inode *inode)
122 static const struct address_space_operations empty_aops;
123 static struct inode_operations empty_iops;
124 static const struct file_operations empty_fops;
126 struct address_space * const mapping = &inode->i_data;
128 inode->i_sb = sb;
129 inode->i_blkbits = sb->s_blocksize_bits;
130 inode->i_flags = 0;
131 atomic_set(&inode->i_count, 1);
132 inode->i_op = &empty_iops;
133 inode->i_fop = &empty_fops;
134 inode->i_nlink = 1;
135 inode->i_uid = 0;
136 inode->i_gid = 0;
137 atomic_set(&inode->i_writecount, 0);
138 inode->i_size = 0;
139 inode->i_blocks = 0;
140 inode->i_bytes = 0;
141 inode->i_generation = 0;
142 #ifdef CONFIG_QUOTA
143 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
144 #endif
145 inode->i_pipe = NULL;
146 inode->i_bdev = NULL;
147 inode->i_cdev = NULL;
148 inode->i_rdev = 0;
149 inode->dirtied_when = 0;
150 if (security_inode_alloc(inode)) {
151 if (inode->i_sb->s_op->destroy_inode)
152 inode->i_sb->s_op->destroy_inode(inode);
153 else
154 kmem_cache_free(inode_cachep, (inode));
155 return NULL;
158 spin_lock_init(&inode->i_lock);
159 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
161 mutex_init(&inode->i_mutex);
162 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
164 init_rwsem(&inode->i_alloc_sem);
165 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
167 mapping->a_ops = &empty_aops;
168 mapping->host = inode;
169 mapping->flags = 0;
170 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
171 mapping->assoc_mapping = NULL;
172 mapping->backing_dev_info = &default_backing_dev_info;
173 mapping->writeback_index = 0;
176 * If the block_device provides a backing_dev_info for client
177 * inodes then use that. Otherwise the inode share the bdev's
178 * backing_dev_info.
180 if (sb->s_bdev) {
181 struct backing_dev_info *bdi;
183 bdi = sb->s_bdev->bd_inode_backing_dev_info;
184 if (!bdi)
185 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
186 mapping->backing_dev_info = bdi;
188 inode->i_private = NULL;
189 inode->i_mapping = mapping;
191 return inode;
193 EXPORT_SYMBOL(inode_init_always);
195 static struct inode *alloc_inode(struct super_block *sb)
197 struct inode *inode;
199 if (sb->s_op->alloc_inode)
200 inode = sb->s_op->alloc_inode(sb);
201 else
202 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
204 if (inode)
205 return inode_init_always(sb, inode);
206 return NULL;
209 void destroy_inode(struct inode *inode)
211 BUG_ON(inode_has_buffers(inode));
212 security_inode_free(inode);
213 if (inode->i_sb->s_op->destroy_inode)
214 inode->i_sb->s_op->destroy_inode(inode);
215 else
216 kmem_cache_free(inode_cachep, (inode));
218 EXPORT_SYMBOL(destroy_inode);
222 * These are initializations that only need to be done
223 * once, because the fields are idempotent across use
224 * of the inode, so let the slab aware of that.
226 void inode_init_once(struct inode *inode)
228 memset(inode, 0, sizeof(*inode));
229 INIT_HLIST_NODE(&inode->i_hash);
230 INIT_LIST_HEAD(&inode->i_dentry);
231 INIT_LIST_HEAD(&inode->i_devices);
232 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
233 spin_lock_init(&inode->i_data.tree_lock);
234 spin_lock_init(&inode->i_data.i_mmap_lock);
235 INIT_LIST_HEAD(&inode->i_data.private_list);
236 spin_lock_init(&inode->i_data.private_lock);
237 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
238 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
239 i_size_ordered_init(inode);
240 #ifdef CONFIG_INOTIFY
241 INIT_LIST_HEAD(&inode->inotify_watches);
242 mutex_init(&inode->inotify_mutex);
243 #endif
246 EXPORT_SYMBOL(inode_init_once);
248 static void init_once(void *foo)
250 struct inode * inode = (struct inode *) foo;
252 inode_init_once(inode);
256 * inode_lock must be held
258 void __iget(struct inode * inode)
260 if (atomic_read(&inode->i_count)) {
261 atomic_inc(&inode->i_count);
262 return;
264 atomic_inc(&inode->i_count);
265 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
266 list_move(&inode->i_list, &inode_in_use);
267 inodes_stat.nr_unused--;
271 * clear_inode - clear an inode
272 * @inode: inode to clear
274 * This is called by the filesystem to tell us
275 * that the inode is no longer useful. We just
276 * terminate it with extreme prejudice.
278 void clear_inode(struct inode *inode)
280 might_sleep();
281 invalidate_inode_buffers(inode);
283 BUG_ON(inode->i_data.nrpages);
284 BUG_ON(!(inode->i_state & I_FREEING));
285 BUG_ON(inode->i_state & I_CLEAR);
286 inode_sync_wait(inode);
287 DQUOT_DROP(inode);
288 if (inode->i_sb->s_op->clear_inode)
289 inode->i_sb->s_op->clear_inode(inode);
290 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
291 bd_forget(inode);
292 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
293 cd_forget(inode);
294 inode->i_state = I_CLEAR;
297 EXPORT_SYMBOL(clear_inode);
300 * dispose_list - dispose of the contents of a local list
301 * @head: the head of the list to free
303 * Dispose-list gets a local list with local inodes in it, so it doesn't
304 * need to worry about list corruption and SMP locks.
306 static void dispose_list(struct list_head *head)
308 int nr_disposed = 0;
310 while (!list_empty(head)) {
311 struct inode *inode;
313 inode = list_first_entry(head, struct inode, i_list);
314 list_del(&inode->i_list);
316 if (inode->i_data.nrpages)
317 truncate_inode_pages(&inode->i_data, 0);
318 clear_inode(inode);
320 spin_lock(&inode_lock);
321 hlist_del_init(&inode->i_hash);
322 list_del_init(&inode->i_sb_list);
323 spin_unlock(&inode_lock);
325 wake_up_inode(inode);
326 destroy_inode(inode);
327 nr_disposed++;
329 spin_lock(&inode_lock);
330 inodes_stat.nr_inodes -= nr_disposed;
331 spin_unlock(&inode_lock);
335 * Invalidate all inodes for a device.
337 static int invalidate_list(struct list_head *head, struct list_head *dispose)
339 struct list_head *next;
340 int busy = 0, count = 0;
342 next = head->next;
343 for (;;) {
344 struct list_head * tmp = next;
345 struct inode * inode;
348 * We can reschedule here without worrying about the list's
349 * consistency because the per-sb list of inodes must not
350 * change during umount anymore, and because iprune_mutex keeps
351 * shrink_icache_memory() away.
353 cond_resched_lock(&inode_lock);
355 next = next->next;
356 if (tmp == head)
357 break;
358 inode = list_entry(tmp, struct inode, i_sb_list);
359 invalidate_inode_buffers(inode);
360 if (!atomic_read(&inode->i_count)) {
361 list_move(&inode->i_list, dispose);
362 WARN_ON(inode->i_state & I_NEW);
363 inode->i_state |= I_FREEING;
364 count++;
365 continue;
367 busy = 1;
369 /* only unused inodes may be cached with i_count zero */
370 inodes_stat.nr_unused -= count;
371 return busy;
375 * invalidate_inodes - discard the inodes on a device
376 * @sb: superblock
378 * Discard all of the inodes for a given superblock. If the discard
379 * fails because there are busy inodes then a non zero value is returned.
380 * If the discard is successful all the inodes have been discarded.
382 int invalidate_inodes(struct super_block * sb)
384 int busy;
385 LIST_HEAD(throw_away);
387 mutex_lock(&iprune_mutex);
388 spin_lock(&inode_lock);
389 inotify_unmount_inodes(&sb->s_inodes);
390 busy = invalidate_list(&sb->s_inodes, &throw_away);
391 spin_unlock(&inode_lock);
393 dispose_list(&throw_away);
394 mutex_unlock(&iprune_mutex);
396 return busy;
399 EXPORT_SYMBOL(invalidate_inodes);
401 static int can_unuse(struct inode *inode)
403 if (inode->i_state)
404 return 0;
405 if (inode_has_buffers(inode))
406 return 0;
407 if (atomic_read(&inode->i_count))
408 return 0;
409 if (inode->i_data.nrpages)
410 return 0;
411 return 1;
415 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
416 * a temporary list and then are freed outside inode_lock by dispose_list().
418 * Any inodes which are pinned purely because of attached pagecache have their
419 * pagecache removed. We expect the final iput() on that inode to add it to
420 * the front of the inode_unused list. So look for it there and if the
421 * inode is still freeable, proceed. The right inode is found 99.9% of the
422 * time in testing on a 4-way.
424 * If the inode has metadata buffers attached to mapping->private_list then
425 * try to remove them.
427 static void prune_icache(int nr_to_scan)
429 LIST_HEAD(freeable);
430 int nr_pruned = 0;
431 int nr_scanned;
432 unsigned long reap = 0;
434 mutex_lock(&iprune_mutex);
435 spin_lock(&inode_lock);
436 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
437 struct inode *inode;
439 if (list_empty(&inode_unused))
440 break;
442 inode = list_entry(inode_unused.prev, struct inode, i_list);
444 if (inode->i_state || atomic_read(&inode->i_count)) {
445 list_move(&inode->i_list, &inode_unused);
446 continue;
448 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
449 __iget(inode);
450 spin_unlock(&inode_lock);
451 if (remove_inode_buffers(inode))
452 reap += invalidate_mapping_pages(&inode->i_data,
453 0, -1);
454 iput(inode);
455 spin_lock(&inode_lock);
457 if (inode != list_entry(inode_unused.next,
458 struct inode, i_list))
459 continue; /* wrong inode or list_empty */
460 if (!can_unuse(inode))
461 continue;
463 list_move(&inode->i_list, &freeable);
464 WARN_ON(inode->i_state & I_NEW);
465 inode->i_state |= I_FREEING;
466 nr_pruned++;
468 inodes_stat.nr_unused -= nr_pruned;
469 if (current_is_kswapd())
470 __count_vm_events(KSWAPD_INODESTEAL, reap);
471 else
472 __count_vm_events(PGINODESTEAL, reap);
473 spin_unlock(&inode_lock);
475 dispose_list(&freeable);
476 mutex_unlock(&iprune_mutex);
480 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
481 * "unused" means that no dentries are referring to the inodes: the files are
482 * not open and the dcache references to those inodes have already been
483 * reclaimed.
485 * This function is passed the number of inodes to scan, and it returns the
486 * total number of remaining possibly-reclaimable inodes.
488 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
490 if (nr) {
492 * Nasty deadlock avoidance. We may hold various FS locks,
493 * and we don't want to recurse into the FS that called us
494 * in clear_inode() and friends..
496 if (!(gfp_mask & __GFP_FS))
497 return -1;
498 prune_icache(nr);
500 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
503 static struct shrinker icache_shrinker = {
504 .shrink = shrink_icache_memory,
505 .seeks = DEFAULT_SEEKS,
508 static void __wait_on_freeing_inode(struct inode *inode);
510 * Called with the inode lock held.
511 * NOTE: we are not increasing the inode-refcount, you must call __iget()
512 * by hand after calling find_inode now! This simplifies iunique and won't
513 * add any additional branch in the common code.
515 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
517 struct hlist_node *node;
518 struct inode * inode = NULL;
520 repeat:
521 hlist_for_each_entry(inode, node, head, i_hash) {
522 if (inode->i_sb != sb)
523 continue;
524 if (!test(inode, data))
525 continue;
526 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
527 __wait_on_freeing_inode(inode);
528 goto repeat;
530 break;
532 return node ? inode : NULL;
536 * find_inode_fast is the fast path version of find_inode, see the comment at
537 * iget_locked for details.
539 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
541 struct hlist_node *node;
542 struct inode * inode = NULL;
544 repeat:
545 hlist_for_each_entry(inode, node, head, i_hash) {
546 if (inode->i_ino != ino)
547 continue;
548 if (inode->i_sb != sb)
549 continue;
550 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
551 __wait_on_freeing_inode(inode);
552 goto repeat;
554 break;
556 return node ? inode : NULL;
559 static unsigned long hash(struct super_block *sb, unsigned long hashval)
561 unsigned long tmp;
563 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
564 L1_CACHE_BYTES;
565 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
566 return tmp & I_HASHMASK;
569 static inline void
570 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
571 struct inode *inode)
573 inodes_stat.nr_inodes++;
574 list_add(&inode->i_list, &inode_in_use);
575 list_add(&inode->i_sb_list, &sb->s_inodes);
576 if (head)
577 hlist_add_head(&inode->i_hash, head);
581 * inode_add_to_lists - add a new inode to relevant lists
582 * @sb: superblock inode belongs to
583 * @inode: inode to mark in use
585 * When an inode is allocated it needs to be accounted for, added to the in use
586 * list, the owning superblock and the inode hash. This needs to be done under
587 * the inode_lock, so export a function to do this rather than the inode lock
588 * itself. We calculate the hash list to add to here so it is all internal
589 * which requires the caller to have already set up the inode number in the
590 * inode to add.
592 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
594 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
596 spin_lock(&inode_lock);
597 __inode_add_to_lists(sb, head, inode);
598 spin_unlock(&inode_lock);
600 EXPORT_SYMBOL_GPL(inode_add_to_lists);
603 * new_inode - obtain an inode
604 * @sb: superblock
606 * Allocates a new inode for given superblock. The default gfp_mask
607 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
608 * If HIGHMEM pages are unsuitable or it is known that pages allocated
609 * for the page cache are not reclaimable or migratable,
610 * mapping_set_gfp_mask() must be called with suitable flags on the
611 * newly created inode's mapping
614 struct inode *new_inode(struct super_block *sb)
617 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
618 * error if st_ino won't fit in target struct field. Use 32bit counter
619 * here to attempt to avoid that.
621 static unsigned int last_ino;
622 struct inode * inode;
624 spin_lock_prefetch(&inode_lock);
626 inode = alloc_inode(sb);
627 if (inode) {
628 spin_lock(&inode_lock);
629 __inode_add_to_lists(sb, NULL, inode);
630 inode->i_ino = ++last_ino;
631 inode->i_state = 0;
632 spin_unlock(&inode_lock);
634 return inode;
637 EXPORT_SYMBOL(new_inode);
639 void unlock_new_inode(struct inode *inode)
641 #ifdef CONFIG_DEBUG_LOCK_ALLOC
642 if (inode->i_mode & S_IFDIR) {
643 struct file_system_type *type = inode->i_sb->s_type;
646 * ensure nobody is actually holding i_mutex
648 mutex_destroy(&inode->i_mutex);
649 mutex_init(&inode->i_mutex);
650 lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key);
652 #endif
654 * This is special! We do not need the spinlock
655 * when clearing I_LOCK, because we're guaranteed
656 * that nobody else tries to do anything about the
657 * state of the inode when it is locked, as we
658 * just created it (so there can be no old holders
659 * that haven't tested I_LOCK).
661 WARN_ON((inode->i_state & (I_LOCK|I_NEW)) != (I_LOCK|I_NEW));
662 inode->i_state &= ~(I_LOCK|I_NEW);
663 wake_up_inode(inode);
666 EXPORT_SYMBOL(unlock_new_inode);
669 * This is called without the inode lock held.. Be careful.
671 * We no longer cache the sb_flags in i_flags - see fs.h
672 * -- rmk@arm.uk.linux.org
674 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)
676 struct inode * inode;
678 inode = alloc_inode(sb);
679 if (inode) {
680 struct inode * old;
682 spin_lock(&inode_lock);
683 /* We released the lock, so.. */
684 old = find_inode(sb, head, test, data);
685 if (!old) {
686 if (set(inode, data))
687 goto set_failed;
689 __inode_add_to_lists(sb, head, inode);
690 inode->i_state = I_LOCK|I_NEW;
691 spin_unlock(&inode_lock);
693 /* Return the locked inode with I_NEW set, the
694 * caller is responsible for filling in the contents
696 return inode;
700 * Uhhuh, somebody else created the same inode under
701 * us. Use the old inode instead of the one we just
702 * allocated.
704 __iget(old);
705 spin_unlock(&inode_lock);
706 destroy_inode(inode);
707 inode = old;
708 wait_on_inode(inode);
710 return inode;
712 set_failed:
713 spin_unlock(&inode_lock);
714 destroy_inode(inode);
715 return NULL;
719 * get_new_inode_fast is the fast path version of get_new_inode, see the
720 * comment at iget_locked for details.
722 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
724 struct inode * inode;
726 inode = alloc_inode(sb);
727 if (inode) {
728 struct inode * old;
730 spin_lock(&inode_lock);
731 /* We released the lock, so.. */
732 old = find_inode_fast(sb, head, ino);
733 if (!old) {
734 inode->i_ino = ino;
735 __inode_add_to_lists(sb, head, inode);
736 inode->i_state = I_LOCK|I_NEW;
737 spin_unlock(&inode_lock);
739 /* Return the locked inode with I_NEW set, the
740 * caller is responsible for filling in the contents
742 return inode;
746 * Uhhuh, somebody else created the same inode under
747 * us. Use the old inode instead of the one we just
748 * allocated.
750 __iget(old);
751 spin_unlock(&inode_lock);
752 destroy_inode(inode);
753 inode = old;
754 wait_on_inode(inode);
756 return inode;
760 * iunique - get a unique inode number
761 * @sb: superblock
762 * @max_reserved: highest reserved inode number
764 * Obtain an inode number that is unique on the system for a given
765 * superblock. This is used by file systems that have no natural
766 * permanent inode numbering system. An inode number is returned that
767 * is higher than the reserved limit but unique.
769 * BUGS:
770 * With a large number of inodes live on the file system this function
771 * currently becomes quite slow.
773 ino_t iunique(struct super_block *sb, ino_t max_reserved)
776 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
777 * error if st_ino won't fit in target struct field. Use 32bit counter
778 * here to attempt to avoid that.
780 static unsigned int counter;
781 struct inode *inode;
782 struct hlist_head *head;
783 ino_t res;
785 spin_lock(&inode_lock);
786 do {
787 if (counter <= max_reserved)
788 counter = max_reserved + 1;
789 res = counter++;
790 head = inode_hashtable + hash(sb, res);
791 inode = find_inode_fast(sb, head, res);
792 } while (inode != NULL);
793 spin_unlock(&inode_lock);
795 return res;
797 EXPORT_SYMBOL(iunique);
799 struct inode *igrab(struct inode *inode)
801 spin_lock(&inode_lock);
802 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
803 __iget(inode);
804 else
806 * Handle the case where s_op->clear_inode is not been
807 * called yet, and somebody is calling igrab
808 * while the inode is getting freed.
810 inode = NULL;
811 spin_unlock(&inode_lock);
812 return inode;
815 EXPORT_SYMBOL(igrab);
818 * ifind - internal function, you want ilookup5() or iget5().
819 * @sb: super block of file system to search
820 * @head: the head of the list to search
821 * @test: callback used for comparisons between inodes
822 * @data: opaque data pointer to pass to @test
823 * @wait: if true wait for the inode to be unlocked, if false do not
825 * ifind() searches for the inode specified by @data in the inode
826 * cache. This is a generalized version of ifind_fast() for file systems where
827 * the inode number is not sufficient for unique identification of an inode.
829 * If the inode is in the cache, the inode is returned with an incremented
830 * reference count.
832 * Otherwise NULL is returned.
834 * Note, @test is called with the inode_lock held, so can't sleep.
836 static struct inode *ifind(struct super_block *sb,
837 struct hlist_head *head, int (*test)(struct inode *, void *),
838 void *data, const int wait)
840 struct inode *inode;
842 spin_lock(&inode_lock);
843 inode = find_inode(sb, head, test, data);
844 if (inode) {
845 __iget(inode);
846 spin_unlock(&inode_lock);
847 if (likely(wait))
848 wait_on_inode(inode);
849 return inode;
851 spin_unlock(&inode_lock);
852 return NULL;
856 * ifind_fast - internal function, you want ilookup() or iget().
857 * @sb: super block of file system to search
858 * @head: head of the list to search
859 * @ino: inode number to search for
861 * ifind_fast() searches for the inode @ino in the inode cache. This is for
862 * file systems where the inode number is sufficient for unique identification
863 * of an inode.
865 * If the inode is in the cache, the inode is returned with an incremented
866 * reference count.
868 * Otherwise NULL is returned.
870 static struct inode *ifind_fast(struct super_block *sb,
871 struct hlist_head *head, unsigned long ino)
873 struct inode *inode;
875 spin_lock(&inode_lock);
876 inode = find_inode_fast(sb, head, ino);
877 if (inode) {
878 __iget(inode);
879 spin_unlock(&inode_lock);
880 wait_on_inode(inode);
881 return inode;
883 spin_unlock(&inode_lock);
884 return NULL;
888 * ilookup5_nowait - search for an inode in the inode cache
889 * @sb: super block of file system to search
890 * @hashval: hash value (usually inode number) to search for
891 * @test: callback used for comparisons between inodes
892 * @data: opaque data pointer to pass to @test
894 * ilookup5() uses ifind() to search for the inode specified by @hashval and
895 * @data in the inode cache. This is a generalized version of ilookup() for
896 * file systems where the inode number is not sufficient for unique
897 * identification of an inode.
899 * If the inode is in the cache, the inode is returned with an incremented
900 * reference count. Note, the inode lock is not waited upon so you have to be
901 * very careful what you do with the returned inode. You probably should be
902 * using ilookup5() instead.
904 * Otherwise NULL is returned.
906 * Note, @test is called with the inode_lock held, so can't sleep.
908 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
909 int (*test)(struct inode *, void *), void *data)
911 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
913 return ifind(sb, head, test, data, 0);
916 EXPORT_SYMBOL(ilookup5_nowait);
919 * ilookup5 - search for an inode in the inode cache
920 * @sb: super block of file system to search
921 * @hashval: hash value (usually inode number) to search for
922 * @test: callback used for comparisons between inodes
923 * @data: opaque data pointer to pass to @test
925 * ilookup5() uses ifind() to search for the inode specified by @hashval and
926 * @data in the inode cache. This is a generalized version of ilookup() for
927 * file systems where the inode number is not sufficient for unique
928 * identification of an inode.
930 * If the inode is in the cache, the inode lock is waited upon and the inode is
931 * returned with an incremented reference count.
933 * Otherwise NULL is returned.
935 * Note, @test is called with the inode_lock held, so can't sleep.
937 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
938 int (*test)(struct inode *, void *), void *data)
940 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
942 return ifind(sb, head, test, data, 1);
945 EXPORT_SYMBOL(ilookup5);
948 * ilookup - search for an inode in the inode cache
949 * @sb: super block of file system to search
950 * @ino: inode number to search for
952 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
953 * This is for file systems where the inode number is sufficient for unique
954 * identification of an inode.
956 * If the inode is in the cache, the inode is returned with an incremented
957 * reference count.
959 * Otherwise NULL is returned.
961 struct inode *ilookup(struct super_block *sb, unsigned long ino)
963 struct hlist_head *head = inode_hashtable + hash(sb, ino);
965 return ifind_fast(sb, head, ino);
968 EXPORT_SYMBOL(ilookup);
971 * iget5_locked - obtain an inode from a mounted file system
972 * @sb: super block of file system
973 * @hashval: hash value (usually inode number) to get
974 * @test: callback used for comparisons between inodes
975 * @set: callback used to initialize a new struct inode
976 * @data: opaque data pointer to pass to @test and @set
978 * iget5_locked() uses ifind() to search for the inode specified by @hashval
979 * and @data in the inode cache and if present it is returned with an increased
980 * reference count. This is a generalized version of iget_locked() for file
981 * systems where the inode number is not sufficient for unique identification
982 * of an inode.
984 * If the inode is not in cache, get_new_inode() is called to allocate a new
985 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
986 * file system gets to fill it in before unlocking it via unlock_new_inode().
988 * Note both @test and @set are called with the inode_lock held, so can't sleep.
990 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
991 int (*test)(struct inode *, void *),
992 int (*set)(struct inode *, void *), void *data)
994 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
995 struct inode *inode;
997 inode = ifind(sb, head, test, data, 1);
998 if (inode)
999 return inode;
1001 * get_new_inode() will do the right thing, re-trying the search
1002 * in case it had to block at any point.
1004 return get_new_inode(sb, head, test, set, data);
1007 EXPORT_SYMBOL(iget5_locked);
1010 * iget_locked - obtain an inode from a mounted file system
1011 * @sb: super block of file system
1012 * @ino: inode number to get
1014 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1015 * the inode cache and if present it is returned with an increased reference
1016 * count. This is for file systems where the inode number is sufficient for
1017 * unique identification of an inode.
1019 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1020 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1021 * The file system gets to fill it in before unlocking it via
1022 * unlock_new_inode().
1024 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1026 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1027 struct inode *inode;
1029 inode = ifind_fast(sb, head, ino);
1030 if (inode)
1031 return inode;
1033 * get_new_inode_fast() will do the right thing, re-trying the search
1034 * in case it had to block at any point.
1036 return get_new_inode_fast(sb, head, ino);
1039 EXPORT_SYMBOL(iget_locked);
1041 int insert_inode_locked(struct inode *inode)
1043 struct super_block *sb = inode->i_sb;
1044 ino_t ino = inode->i_ino;
1045 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1046 struct inode *old;
1048 inode->i_state |= I_LOCK|I_NEW;
1049 while (1) {
1050 spin_lock(&inode_lock);
1051 old = find_inode_fast(sb, head, ino);
1052 if (likely(!old)) {
1053 hlist_add_head(&inode->i_hash, head);
1054 spin_unlock(&inode_lock);
1055 return 0;
1057 __iget(old);
1058 spin_unlock(&inode_lock);
1059 wait_on_inode(old);
1060 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1061 iput(old);
1062 return -EBUSY;
1064 iput(old);
1068 EXPORT_SYMBOL(insert_inode_locked);
1070 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1071 int (*test)(struct inode *, void *), void *data)
1073 struct super_block *sb = inode->i_sb;
1074 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1075 struct inode *old;
1077 inode->i_state |= I_LOCK|I_NEW;
1079 while (1) {
1080 spin_lock(&inode_lock);
1081 old = find_inode(sb, head, test, data);
1082 if (likely(!old)) {
1083 hlist_add_head(&inode->i_hash, head);
1084 spin_unlock(&inode_lock);
1085 return 0;
1087 __iget(old);
1088 spin_unlock(&inode_lock);
1089 wait_on_inode(old);
1090 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1091 iput(old);
1092 return -EBUSY;
1094 iput(old);
1098 EXPORT_SYMBOL(insert_inode_locked4);
1101 * __insert_inode_hash - hash an inode
1102 * @inode: unhashed inode
1103 * @hashval: unsigned long value used to locate this object in the
1104 * inode_hashtable.
1106 * Add an inode to the inode hash for this superblock.
1108 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1110 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1111 spin_lock(&inode_lock);
1112 hlist_add_head(&inode->i_hash, head);
1113 spin_unlock(&inode_lock);
1116 EXPORT_SYMBOL(__insert_inode_hash);
1119 * remove_inode_hash - remove an inode from the hash
1120 * @inode: inode to unhash
1122 * Remove an inode from the superblock.
1124 void remove_inode_hash(struct inode *inode)
1126 spin_lock(&inode_lock);
1127 hlist_del_init(&inode->i_hash);
1128 spin_unlock(&inode_lock);
1131 EXPORT_SYMBOL(remove_inode_hash);
1134 * Tell the filesystem that this inode is no longer of any interest and should
1135 * be completely destroyed.
1137 * We leave the inode in the inode hash table until *after* the filesystem's
1138 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1139 * instigate) will always find up-to-date information either in the hash or on
1140 * disk.
1142 * I_FREEING is set so that no-one will take a new reference to the inode while
1143 * it is being deleted.
1145 void generic_delete_inode(struct inode *inode)
1147 const struct super_operations *op = inode->i_sb->s_op;
1149 list_del_init(&inode->i_list);
1150 list_del_init(&inode->i_sb_list);
1151 WARN_ON(inode->i_state & I_NEW);
1152 inode->i_state |= I_FREEING;
1153 inodes_stat.nr_inodes--;
1154 spin_unlock(&inode_lock);
1156 security_inode_delete(inode);
1158 if (op->delete_inode) {
1159 void (*delete)(struct inode *) = op->delete_inode;
1160 if (!is_bad_inode(inode))
1161 DQUOT_INIT(inode);
1162 /* Filesystems implementing their own
1163 * s_op->delete_inode are required to call
1164 * truncate_inode_pages and clear_inode()
1165 * internally */
1166 delete(inode);
1167 } else {
1168 truncate_inode_pages(&inode->i_data, 0);
1169 clear_inode(inode);
1171 spin_lock(&inode_lock);
1172 hlist_del_init(&inode->i_hash);
1173 spin_unlock(&inode_lock);
1174 wake_up_inode(inode);
1175 BUG_ON(inode->i_state != I_CLEAR);
1176 destroy_inode(inode);
1179 EXPORT_SYMBOL(generic_delete_inode);
1181 static void generic_forget_inode(struct inode *inode)
1183 struct super_block *sb = inode->i_sb;
1185 if (!hlist_unhashed(&inode->i_hash)) {
1186 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1187 list_move(&inode->i_list, &inode_unused);
1188 inodes_stat.nr_unused++;
1189 if (sb->s_flags & MS_ACTIVE) {
1190 spin_unlock(&inode_lock);
1191 return;
1193 WARN_ON(inode->i_state & I_NEW);
1194 inode->i_state |= I_WILL_FREE;
1195 spin_unlock(&inode_lock);
1196 write_inode_now(inode, 1);
1197 spin_lock(&inode_lock);
1198 WARN_ON(inode->i_state & I_NEW);
1199 inode->i_state &= ~I_WILL_FREE;
1200 inodes_stat.nr_unused--;
1201 hlist_del_init(&inode->i_hash);
1203 list_del_init(&inode->i_list);
1204 list_del_init(&inode->i_sb_list);
1205 WARN_ON(inode->i_state & I_NEW);
1206 inode->i_state |= I_FREEING;
1207 inodes_stat.nr_inodes--;
1208 spin_unlock(&inode_lock);
1209 if (inode->i_data.nrpages)
1210 truncate_inode_pages(&inode->i_data, 0);
1211 clear_inode(inode);
1212 wake_up_inode(inode);
1213 destroy_inode(inode);
1217 * Normal UNIX filesystem behaviour: delete the
1218 * inode when the usage count drops to zero, and
1219 * i_nlink is zero.
1221 void generic_drop_inode(struct inode *inode)
1223 if (!inode->i_nlink)
1224 generic_delete_inode(inode);
1225 else
1226 generic_forget_inode(inode);
1229 EXPORT_SYMBOL_GPL(generic_drop_inode);
1232 * Called when we're dropping the last reference
1233 * to an inode.
1235 * Call the FS "drop()" function, defaulting to
1236 * the legacy UNIX filesystem behaviour..
1238 * NOTE! NOTE! NOTE! We're called with the inode lock
1239 * held, and the drop function is supposed to release
1240 * the lock!
1242 static inline void iput_final(struct inode *inode)
1244 const struct super_operations *op = inode->i_sb->s_op;
1245 void (*drop)(struct inode *) = generic_drop_inode;
1247 if (op && op->drop_inode)
1248 drop = op->drop_inode;
1249 drop(inode);
1253 * iput - put an inode
1254 * @inode: inode to put
1256 * Puts an inode, dropping its usage count. If the inode use count hits
1257 * zero, the inode is then freed and may also be destroyed.
1259 * Consequently, iput() can sleep.
1261 void iput(struct inode *inode)
1263 if (inode) {
1264 BUG_ON(inode->i_state == I_CLEAR);
1266 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1267 iput_final(inode);
1271 EXPORT_SYMBOL(iput);
1274 * bmap - find a block number in a file
1275 * @inode: inode of file
1276 * @block: block to find
1278 * Returns the block number on the device holding the inode that
1279 * is the disk block number for the block of the file requested.
1280 * That is, asked for block 4 of inode 1 the function will return the
1281 * disk block relative to the disk start that holds that block of the
1282 * file.
1284 sector_t bmap(struct inode * inode, sector_t block)
1286 sector_t res = 0;
1287 if (inode->i_mapping->a_ops->bmap)
1288 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1289 return res;
1291 EXPORT_SYMBOL(bmap);
1294 * touch_atime - update the access time
1295 * @mnt: mount the inode is accessed on
1296 * @dentry: dentry accessed
1298 * Update the accessed time on an inode and mark it for writeback.
1299 * This function automatically handles read only file systems and media,
1300 * as well as the "noatime" flag and inode specific "noatime" markers.
1302 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1304 struct inode *inode = dentry->d_inode;
1305 struct timespec now;
1307 if (mnt_want_write(mnt))
1308 return;
1309 if (inode->i_flags & S_NOATIME)
1310 goto out;
1311 if (IS_NOATIME(inode))
1312 goto out;
1313 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1314 goto out;
1316 if (mnt->mnt_flags & MNT_NOATIME)
1317 goto out;
1318 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1319 goto out;
1320 if (mnt->mnt_flags & MNT_RELATIME) {
1322 * With relative atime, only update atime if the previous
1323 * atime is earlier than either the ctime or mtime.
1325 if (timespec_compare(&inode->i_mtime, &inode->i_atime) < 0 &&
1326 timespec_compare(&inode->i_ctime, &inode->i_atime) < 0)
1327 goto out;
1330 now = current_fs_time(inode->i_sb);
1331 if (timespec_equal(&inode->i_atime, &now))
1332 goto out;
1334 inode->i_atime = now;
1335 mark_inode_dirty_sync(inode);
1336 out:
1337 mnt_drop_write(mnt);
1339 EXPORT_SYMBOL(touch_atime);
1342 * file_update_time - update mtime and ctime time
1343 * @file: file accessed
1345 * Update the mtime and ctime members of an inode and mark the inode
1346 * for writeback. Note that this function is meant exclusively for
1347 * usage in the file write path of filesystems, and filesystems may
1348 * choose to explicitly ignore update via this function with the
1349 * S_NOCTIME inode flag, e.g. for network filesystem where these
1350 * timestamps are handled by the server.
1353 void file_update_time(struct file *file)
1355 struct inode *inode = file->f_path.dentry->d_inode;
1356 struct timespec now;
1357 int sync_it = 0;
1358 int err;
1360 if (IS_NOCMTIME(inode))
1361 return;
1363 err = mnt_want_write(file->f_path.mnt);
1364 if (err)
1365 return;
1367 now = current_fs_time(inode->i_sb);
1368 if (!timespec_equal(&inode->i_mtime, &now)) {
1369 inode->i_mtime = now;
1370 sync_it = 1;
1373 if (!timespec_equal(&inode->i_ctime, &now)) {
1374 inode->i_ctime = now;
1375 sync_it = 1;
1378 if (IS_I_VERSION(inode)) {
1379 inode_inc_iversion(inode);
1380 sync_it = 1;
1383 if (sync_it)
1384 mark_inode_dirty_sync(inode);
1385 mnt_drop_write(file->f_path.mnt);
1388 EXPORT_SYMBOL(file_update_time);
1390 int inode_needs_sync(struct inode *inode)
1392 if (IS_SYNC(inode))
1393 return 1;
1394 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1395 return 1;
1396 return 0;
1399 EXPORT_SYMBOL(inode_needs_sync);
1401 int inode_wait(void *word)
1403 schedule();
1404 return 0;
1406 EXPORT_SYMBOL(inode_wait);
1409 * If we try to find an inode in the inode hash while it is being
1410 * deleted, we have to wait until the filesystem completes its
1411 * deletion before reporting that it isn't found. This function waits
1412 * until the deletion _might_ have completed. Callers are responsible
1413 * to recheck inode state.
1415 * It doesn't matter if I_LOCK is not set initially, a call to
1416 * wake_up_inode() after removing from the hash list will DTRT.
1418 * This is called with inode_lock held.
1420 static void __wait_on_freeing_inode(struct inode *inode)
1422 wait_queue_head_t *wq;
1423 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1424 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1425 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1426 spin_unlock(&inode_lock);
1427 schedule();
1428 finish_wait(wq, &wait.wait);
1429 spin_lock(&inode_lock);
1433 * We rarely want to lock two inodes that do not have a parent/child
1434 * relationship (such as directory, child inode) simultaneously. The
1435 * vast majority of file systems should be able to get along fine
1436 * without this. Do not use these functions except as a last resort.
1438 void inode_double_lock(struct inode *inode1, struct inode *inode2)
1440 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1441 if (inode1)
1442 mutex_lock(&inode1->i_mutex);
1443 else if (inode2)
1444 mutex_lock(&inode2->i_mutex);
1445 return;
1448 if (inode1 < inode2) {
1449 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1450 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1451 } else {
1452 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1453 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1456 EXPORT_SYMBOL(inode_double_lock);
1458 void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1460 if (inode1)
1461 mutex_unlock(&inode1->i_mutex);
1463 if (inode2 && inode2 != inode1)
1464 mutex_unlock(&inode2->i_mutex);
1466 EXPORT_SYMBOL(inode_double_unlock);
1468 static __initdata unsigned long ihash_entries;
1469 static int __init set_ihash_entries(char *str)
1471 if (!str)
1472 return 0;
1473 ihash_entries = simple_strtoul(str, &str, 0);
1474 return 1;
1476 __setup("ihash_entries=", set_ihash_entries);
1479 * Initialize the waitqueues and inode hash table.
1481 void __init inode_init_early(void)
1483 int loop;
1485 /* If hashes are distributed across NUMA nodes, defer
1486 * hash allocation until vmalloc space is available.
1488 if (hashdist)
1489 return;
1491 inode_hashtable =
1492 alloc_large_system_hash("Inode-cache",
1493 sizeof(struct hlist_head),
1494 ihash_entries,
1496 HASH_EARLY,
1497 &i_hash_shift,
1498 &i_hash_mask,
1501 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1502 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1505 void __init inode_init(void)
1507 int loop;
1509 /* inode slab cache */
1510 inode_cachep = kmem_cache_create("inode_cache",
1511 sizeof(struct inode),
1513 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1514 SLAB_MEM_SPREAD),
1515 init_once);
1516 register_shrinker(&icache_shrinker);
1518 /* Hash may have been set up in inode_init_early */
1519 if (!hashdist)
1520 return;
1522 inode_hashtable =
1523 alloc_large_system_hash("Inode-cache",
1524 sizeof(struct hlist_head),
1525 ihash_entries,
1528 &i_hash_shift,
1529 &i_hash_mask,
1532 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1533 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1536 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1538 inode->i_mode = mode;
1539 if (S_ISCHR(mode)) {
1540 inode->i_fop = &def_chr_fops;
1541 inode->i_rdev = rdev;
1542 } else if (S_ISBLK(mode)) {
1543 inode->i_fop = &def_blk_fops;
1544 inode->i_rdev = rdev;
1545 } else if (S_ISFIFO(mode))
1546 inode->i_fop = &def_fifo_fops;
1547 else if (S_ISSOCK(mode))
1548 inode->i_fop = &bad_sock_fops;
1549 else
1550 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1551 mode);
1553 EXPORT_SYMBOL(init_special_inode);