module: add MODULE_STATE_LIVE notify
[linux-2.6/mini2440.git] / fs / inode.c
blob7a6e8c2ff7b16b472b2bc7fd40894cfc3d8d0904
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>
27 * This is needed for the following functions:
28 * - inode_has_buffers
29 * - invalidate_inode_buffers
30 * - invalidate_bdev
32 * FIXME: remove all knowledge of the buffer layer from this file
34 #include <linux/buffer_head.h>
37 * New inode.c implementation.
39 * This implementation has the basic premise of trying
40 * to be extremely low-overhead and SMP-safe, yet be
41 * simple enough to be "obviously correct".
43 * Famous last words.
46 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
48 /* #define INODE_PARANOIA 1 */
49 /* #define INODE_DEBUG 1 */
52 * Inode lookup is no longer as critical as it used to be:
53 * most of the lookups are going to be through the dcache.
55 #define I_HASHBITS i_hash_shift
56 #define I_HASHMASK i_hash_mask
58 static unsigned int i_hash_mask __read_mostly;
59 static unsigned int i_hash_shift __read_mostly;
62 * Each inode can be on two separate lists. One is
63 * the hash list of the inode, used for lookups. The
64 * other linked list is the "type" list:
65 * "in_use" - valid inode, i_count > 0, i_nlink > 0
66 * "dirty" - as "in_use" but also dirty
67 * "unused" - valid inode, i_count = 0
69 * A "dirty" list is maintained for each super block,
70 * allowing for low-overhead inode sync() operations.
73 LIST_HEAD(inode_in_use);
74 LIST_HEAD(inode_unused);
75 static struct hlist_head *inode_hashtable __read_mostly;
78 * A simple spinlock to protect the list manipulations.
80 * NOTE! You also have to own the lock if you change
81 * the i_state of an inode while it is in use..
83 DEFINE_SPINLOCK(inode_lock);
86 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
87 * icache shrinking path, and the umount path. Without this exclusion,
88 * by the time prune_icache calls iput for the inode whose pages it has
89 * been invalidating, or by the time it calls clear_inode & destroy_inode
90 * from its final dispose_list, the struct super_block they refer to
91 * (for inode->i_sb->s_op) may already have been freed and reused.
93 static DEFINE_MUTEX(iprune_mutex);
96 * Statistics gathering..
98 struct inodes_stat_t inodes_stat;
100 static struct kmem_cache * inode_cachep __read_mostly;
102 static void wake_up_inode(struct inode *inode)
105 * Prevent speculative execution through spin_unlock(&inode_lock);
107 smp_mb();
108 wake_up_bit(&inode->i_state, __I_LOCK);
112 * inode_init_always - perform inode structure intialisation
113 * @sb: superblock inode belongs to
114 * @inode: inode to initialise
116 * These are initializations that need to be done on every inode
117 * allocation as the fields are not initialised by slab allocation.
119 struct inode *inode_init_always(struct super_block *sb, struct inode *inode)
121 static const struct address_space_operations empty_aops;
122 static struct inode_operations empty_iops;
123 static const struct file_operations empty_fops;
125 struct address_space * const mapping = &inode->i_data;
127 inode->i_sb = sb;
128 inode->i_blkbits = sb->s_blocksize_bits;
129 inode->i_flags = 0;
130 atomic_set(&inode->i_count, 1);
131 inode->i_op = &empty_iops;
132 inode->i_fop = &empty_fops;
133 inode->i_nlink = 1;
134 inode->i_uid = 0;
135 inode->i_gid = 0;
136 atomic_set(&inode->i_writecount, 0);
137 inode->i_size = 0;
138 inode->i_blocks = 0;
139 inode->i_bytes = 0;
140 inode->i_generation = 0;
141 #ifdef CONFIG_QUOTA
142 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
143 #endif
144 inode->i_pipe = NULL;
145 inode->i_bdev = NULL;
146 inode->i_cdev = NULL;
147 inode->i_rdev = 0;
148 inode->dirtied_when = 0;
149 if (security_inode_alloc(inode)) {
150 if (inode->i_sb->s_op->destroy_inode)
151 inode->i_sb->s_op->destroy_inode(inode);
152 else
153 kmem_cache_free(inode_cachep, (inode));
154 return NULL;
157 spin_lock_init(&inode->i_lock);
158 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
160 mutex_init(&inode->i_mutex);
161 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
163 init_rwsem(&inode->i_alloc_sem);
164 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
166 mapping->a_ops = &empty_aops;
167 mapping->host = inode;
168 mapping->flags = 0;
169 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
170 mapping->assoc_mapping = NULL;
171 mapping->backing_dev_info = &default_backing_dev_info;
172 mapping->writeback_index = 0;
175 * If the block_device provides a backing_dev_info for client
176 * inodes then use that. Otherwise the inode share the bdev's
177 * backing_dev_info.
179 if (sb->s_bdev) {
180 struct backing_dev_info *bdi;
182 bdi = sb->s_bdev->bd_inode_backing_dev_info;
183 if (!bdi)
184 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
185 mapping->backing_dev_info = bdi;
187 inode->i_private = NULL;
188 inode->i_mapping = mapping;
190 return inode;
192 EXPORT_SYMBOL(inode_init_always);
194 static struct inode *alloc_inode(struct super_block *sb)
196 struct inode *inode;
198 if (sb->s_op->alloc_inode)
199 inode = sb->s_op->alloc_inode(sb);
200 else
201 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
203 if (inode)
204 return inode_init_always(sb, inode);
205 return NULL;
208 void destroy_inode(struct inode *inode)
210 BUG_ON(inode_has_buffers(inode));
211 security_inode_free(inode);
212 if (inode->i_sb->s_op->destroy_inode)
213 inode->i_sb->s_op->destroy_inode(inode);
214 else
215 kmem_cache_free(inode_cachep, (inode));
217 EXPORT_SYMBOL(destroy_inode);
221 * These are initializations that only need to be done
222 * once, because the fields are idempotent across use
223 * of the inode, so let the slab aware of that.
225 void inode_init_once(struct inode *inode)
227 memset(inode, 0, sizeof(*inode));
228 INIT_HLIST_NODE(&inode->i_hash);
229 INIT_LIST_HEAD(&inode->i_dentry);
230 INIT_LIST_HEAD(&inode->i_devices);
231 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
232 spin_lock_init(&inode->i_data.tree_lock);
233 spin_lock_init(&inode->i_data.i_mmap_lock);
234 INIT_LIST_HEAD(&inode->i_data.private_list);
235 spin_lock_init(&inode->i_data.private_lock);
236 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
237 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
238 i_size_ordered_init(inode);
239 #ifdef CONFIG_INOTIFY
240 INIT_LIST_HEAD(&inode->inotify_watches);
241 mutex_init(&inode->inotify_mutex);
242 #endif
245 EXPORT_SYMBOL(inode_init_once);
247 static void init_once(void *foo)
249 struct inode * inode = (struct inode *) foo;
251 inode_init_once(inode);
255 * inode_lock must be held
257 void __iget(struct inode * inode)
259 if (atomic_read(&inode->i_count)) {
260 atomic_inc(&inode->i_count);
261 return;
263 atomic_inc(&inode->i_count);
264 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
265 list_move(&inode->i_list, &inode_in_use);
266 inodes_stat.nr_unused--;
270 * clear_inode - clear an inode
271 * @inode: inode to clear
273 * This is called by the filesystem to tell us
274 * that the inode is no longer useful. We just
275 * terminate it with extreme prejudice.
277 void clear_inode(struct inode *inode)
279 might_sleep();
280 invalidate_inode_buffers(inode);
282 BUG_ON(inode->i_data.nrpages);
283 BUG_ON(!(inode->i_state & I_FREEING));
284 BUG_ON(inode->i_state & I_CLEAR);
285 inode_sync_wait(inode);
286 DQUOT_DROP(inode);
287 if (inode->i_sb->s_op->clear_inode)
288 inode->i_sb->s_op->clear_inode(inode);
289 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
290 bd_forget(inode);
291 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
292 cd_forget(inode);
293 inode->i_state = I_CLEAR;
296 EXPORT_SYMBOL(clear_inode);
299 * dispose_list - dispose of the contents of a local list
300 * @head: the head of the list to free
302 * Dispose-list gets a local list with local inodes in it, so it doesn't
303 * need to worry about list corruption and SMP locks.
305 static void dispose_list(struct list_head *head)
307 int nr_disposed = 0;
309 while (!list_empty(head)) {
310 struct inode *inode;
312 inode = list_first_entry(head, struct inode, i_list);
313 list_del(&inode->i_list);
315 if (inode->i_data.nrpages)
316 truncate_inode_pages(&inode->i_data, 0);
317 clear_inode(inode);
319 spin_lock(&inode_lock);
320 hlist_del_init(&inode->i_hash);
321 list_del_init(&inode->i_sb_list);
322 spin_unlock(&inode_lock);
324 wake_up_inode(inode);
325 destroy_inode(inode);
326 nr_disposed++;
328 spin_lock(&inode_lock);
329 inodes_stat.nr_inodes -= nr_disposed;
330 spin_unlock(&inode_lock);
334 * Invalidate all inodes for a device.
336 static int invalidate_list(struct list_head *head, struct list_head *dispose)
338 struct list_head *next;
339 int busy = 0, count = 0;
341 next = head->next;
342 for (;;) {
343 struct list_head * tmp = next;
344 struct inode * inode;
347 * We can reschedule here without worrying about the list's
348 * consistency because the per-sb list of inodes must not
349 * change during umount anymore, and because iprune_mutex keeps
350 * shrink_icache_memory() away.
352 cond_resched_lock(&inode_lock);
354 next = next->next;
355 if (tmp == head)
356 break;
357 inode = list_entry(tmp, struct inode, i_sb_list);
358 invalidate_inode_buffers(inode);
359 if (!atomic_read(&inode->i_count)) {
360 list_move(&inode->i_list, dispose);
361 inode->i_state |= I_FREEING;
362 count++;
363 continue;
365 busy = 1;
367 /* only unused inodes may be cached with i_count zero */
368 inodes_stat.nr_unused -= count;
369 return busy;
373 * invalidate_inodes - discard the inodes on a device
374 * @sb: superblock
376 * Discard all of the inodes for a given superblock. If the discard
377 * fails because there are busy inodes then a non zero value is returned.
378 * If the discard is successful all the inodes have been discarded.
380 int invalidate_inodes(struct super_block * sb)
382 int busy;
383 LIST_HEAD(throw_away);
385 mutex_lock(&iprune_mutex);
386 spin_lock(&inode_lock);
387 inotify_unmount_inodes(&sb->s_inodes);
388 busy = invalidate_list(&sb->s_inodes, &throw_away);
389 spin_unlock(&inode_lock);
391 dispose_list(&throw_away);
392 mutex_unlock(&iprune_mutex);
394 return busy;
397 EXPORT_SYMBOL(invalidate_inodes);
399 static int can_unuse(struct inode *inode)
401 if (inode->i_state)
402 return 0;
403 if (inode_has_buffers(inode))
404 return 0;
405 if (atomic_read(&inode->i_count))
406 return 0;
407 if (inode->i_data.nrpages)
408 return 0;
409 return 1;
413 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
414 * a temporary list and then are freed outside inode_lock by dispose_list().
416 * Any inodes which are pinned purely because of attached pagecache have their
417 * pagecache removed. We expect the final iput() on that inode to add it to
418 * the front of the inode_unused list. So look for it there and if the
419 * inode is still freeable, proceed. The right inode is found 99.9% of the
420 * time in testing on a 4-way.
422 * If the inode has metadata buffers attached to mapping->private_list then
423 * try to remove them.
425 static void prune_icache(int nr_to_scan)
427 LIST_HEAD(freeable);
428 int nr_pruned = 0;
429 int nr_scanned;
430 unsigned long reap = 0;
432 mutex_lock(&iprune_mutex);
433 spin_lock(&inode_lock);
434 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
435 struct inode *inode;
437 if (list_empty(&inode_unused))
438 break;
440 inode = list_entry(inode_unused.prev, struct inode, i_list);
442 if (inode->i_state || atomic_read(&inode->i_count)) {
443 list_move(&inode->i_list, &inode_unused);
444 continue;
446 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
447 __iget(inode);
448 spin_unlock(&inode_lock);
449 if (remove_inode_buffers(inode))
450 reap += invalidate_mapping_pages(&inode->i_data,
451 0, -1);
452 iput(inode);
453 spin_lock(&inode_lock);
455 if (inode != list_entry(inode_unused.next,
456 struct inode, i_list))
457 continue; /* wrong inode or list_empty */
458 if (!can_unuse(inode))
459 continue;
461 list_move(&inode->i_list, &freeable);
462 inode->i_state |= I_FREEING;
463 nr_pruned++;
465 inodes_stat.nr_unused -= nr_pruned;
466 if (current_is_kswapd())
467 __count_vm_events(KSWAPD_INODESTEAL, reap);
468 else
469 __count_vm_events(PGINODESTEAL, reap);
470 spin_unlock(&inode_lock);
472 dispose_list(&freeable);
473 mutex_unlock(&iprune_mutex);
477 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
478 * "unused" means that no dentries are referring to the inodes: the files are
479 * not open and the dcache references to those inodes have already been
480 * reclaimed.
482 * This function is passed the number of inodes to scan, and it returns the
483 * total number of remaining possibly-reclaimable inodes.
485 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
487 if (nr) {
489 * Nasty deadlock avoidance. We may hold various FS locks,
490 * and we don't want to recurse into the FS that called us
491 * in clear_inode() and friends..
493 if (!(gfp_mask & __GFP_FS))
494 return -1;
495 prune_icache(nr);
497 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
500 static struct shrinker icache_shrinker = {
501 .shrink = shrink_icache_memory,
502 .seeks = DEFAULT_SEEKS,
505 static void __wait_on_freeing_inode(struct inode *inode);
507 * Called with the inode lock held.
508 * NOTE: we are not increasing the inode-refcount, you must call __iget()
509 * by hand after calling find_inode now! This simplifies iunique and won't
510 * add any additional branch in the common code.
512 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
514 struct hlist_node *node;
515 struct inode * inode = NULL;
517 repeat:
518 hlist_for_each_entry(inode, node, head, i_hash) {
519 if (inode->i_sb != sb)
520 continue;
521 if (!test(inode, data))
522 continue;
523 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
524 __wait_on_freeing_inode(inode);
525 goto repeat;
527 break;
529 return node ? inode : NULL;
533 * find_inode_fast is the fast path version of find_inode, see the comment at
534 * iget_locked for details.
536 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
538 struct hlist_node *node;
539 struct inode * inode = NULL;
541 repeat:
542 hlist_for_each_entry(inode, node, head, i_hash) {
543 if (inode->i_ino != ino)
544 continue;
545 if (inode->i_sb != sb)
546 continue;
547 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
548 __wait_on_freeing_inode(inode);
549 goto repeat;
551 break;
553 return node ? inode : NULL;
556 static unsigned long hash(struct super_block *sb, unsigned long hashval)
558 unsigned long tmp;
560 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
561 L1_CACHE_BYTES;
562 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
563 return tmp & I_HASHMASK;
566 static inline void
567 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
568 struct inode *inode)
570 inodes_stat.nr_inodes++;
571 list_add(&inode->i_list, &inode_in_use);
572 list_add(&inode->i_sb_list, &sb->s_inodes);
573 if (head)
574 hlist_add_head(&inode->i_hash, head);
578 * inode_add_to_lists - add a new inode to relevant lists
579 * @sb: superblock inode belongs to
580 * @inode: inode to mark in use
582 * When an inode is allocated it needs to be accounted for, added to the in use
583 * list, the owning superblock and the inode hash. This needs to be done under
584 * the inode_lock, so export a function to do this rather than the inode lock
585 * itself. We calculate the hash list to add to here so it is all internal
586 * which requires the caller to have already set up the inode number in the
587 * inode to add.
589 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
591 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
593 spin_lock(&inode_lock);
594 __inode_add_to_lists(sb, head, inode);
595 spin_unlock(&inode_lock);
597 EXPORT_SYMBOL_GPL(inode_add_to_lists);
600 * new_inode - obtain an inode
601 * @sb: superblock
603 * Allocates a new inode for given superblock. The default gfp_mask
604 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
605 * If HIGHMEM pages are unsuitable or it is known that pages allocated
606 * for the page cache are not reclaimable or migratable,
607 * mapping_set_gfp_mask() must be called with suitable flags on the
608 * newly created inode's mapping
611 struct inode *new_inode(struct super_block *sb)
614 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
615 * error if st_ino won't fit in target struct field. Use 32bit counter
616 * here to attempt to avoid that.
618 static unsigned int last_ino;
619 struct inode * inode;
621 spin_lock_prefetch(&inode_lock);
623 inode = alloc_inode(sb);
624 if (inode) {
625 spin_lock(&inode_lock);
626 __inode_add_to_lists(sb, NULL, inode);
627 inode->i_ino = ++last_ino;
628 inode->i_state = 0;
629 spin_unlock(&inode_lock);
631 return inode;
634 EXPORT_SYMBOL(new_inode);
636 void unlock_new_inode(struct inode *inode)
638 #ifdef CONFIG_DEBUG_LOCK_ALLOC
639 if (inode->i_mode & S_IFDIR) {
640 struct file_system_type *type = inode->i_sb->s_type;
643 * ensure nobody is actually holding i_mutex
645 mutex_destroy(&inode->i_mutex);
646 mutex_init(&inode->i_mutex);
647 lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key);
649 #endif
651 * This is special! We do not need the spinlock
652 * when clearing I_LOCK, because we're guaranteed
653 * that nobody else tries to do anything about the
654 * state of the inode when it is locked, as we
655 * just created it (so there can be no old holders
656 * that haven't tested I_LOCK).
658 inode->i_state &= ~(I_LOCK|I_NEW);
659 wake_up_inode(inode);
662 EXPORT_SYMBOL(unlock_new_inode);
665 * This is called without the inode lock held.. Be careful.
667 * We no longer cache the sb_flags in i_flags - see fs.h
668 * -- rmk@arm.uk.linux.org
670 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)
672 struct inode * inode;
674 inode = alloc_inode(sb);
675 if (inode) {
676 struct inode * old;
678 spin_lock(&inode_lock);
679 /* We released the lock, so.. */
680 old = find_inode(sb, head, test, data);
681 if (!old) {
682 if (set(inode, data))
683 goto set_failed;
685 __inode_add_to_lists(sb, head, inode);
686 inode->i_state = I_LOCK|I_NEW;
687 spin_unlock(&inode_lock);
689 /* Return the locked inode with I_NEW set, the
690 * caller is responsible for filling in the contents
692 return inode;
696 * Uhhuh, somebody else created the same inode under
697 * us. Use the old inode instead of the one we just
698 * allocated.
700 __iget(old);
701 spin_unlock(&inode_lock);
702 destroy_inode(inode);
703 inode = old;
704 wait_on_inode(inode);
706 return inode;
708 set_failed:
709 spin_unlock(&inode_lock);
710 destroy_inode(inode);
711 return NULL;
715 * get_new_inode_fast is the fast path version of get_new_inode, see the
716 * comment at iget_locked for details.
718 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
720 struct inode * inode;
722 inode = alloc_inode(sb);
723 if (inode) {
724 struct inode * old;
726 spin_lock(&inode_lock);
727 /* We released the lock, so.. */
728 old = find_inode_fast(sb, head, ino);
729 if (!old) {
730 inode->i_ino = ino;
731 __inode_add_to_lists(sb, head, inode);
732 inode->i_state = I_LOCK|I_NEW;
733 spin_unlock(&inode_lock);
735 /* Return the locked inode with I_NEW set, the
736 * caller is responsible for filling in the contents
738 return inode;
742 * Uhhuh, somebody else created the same inode under
743 * us. Use the old inode instead of the one we just
744 * allocated.
746 __iget(old);
747 spin_unlock(&inode_lock);
748 destroy_inode(inode);
749 inode = old;
750 wait_on_inode(inode);
752 return inode;
756 * iunique - get a unique inode number
757 * @sb: superblock
758 * @max_reserved: highest reserved inode number
760 * Obtain an inode number that is unique on the system for a given
761 * superblock. This is used by file systems that have no natural
762 * permanent inode numbering system. An inode number is returned that
763 * is higher than the reserved limit but unique.
765 * BUGS:
766 * With a large number of inodes live on the file system this function
767 * currently becomes quite slow.
769 ino_t iunique(struct super_block *sb, ino_t max_reserved)
772 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
773 * error if st_ino won't fit in target struct field. Use 32bit counter
774 * here to attempt to avoid that.
776 static unsigned int counter;
777 struct inode *inode;
778 struct hlist_head *head;
779 ino_t res;
781 spin_lock(&inode_lock);
782 do {
783 if (counter <= max_reserved)
784 counter = max_reserved + 1;
785 res = counter++;
786 head = inode_hashtable + hash(sb, res);
787 inode = find_inode_fast(sb, head, res);
788 } while (inode != NULL);
789 spin_unlock(&inode_lock);
791 return res;
793 EXPORT_SYMBOL(iunique);
795 struct inode *igrab(struct inode *inode)
797 spin_lock(&inode_lock);
798 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
799 __iget(inode);
800 else
802 * Handle the case where s_op->clear_inode is not been
803 * called yet, and somebody is calling igrab
804 * while the inode is getting freed.
806 inode = NULL;
807 spin_unlock(&inode_lock);
808 return inode;
811 EXPORT_SYMBOL(igrab);
814 * ifind - internal function, you want ilookup5() or iget5().
815 * @sb: super block of file system to search
816 * @head: the head of the list to search
817 * @test: callback used for comparisons between inodes
818 * @data: opaque data pointer to pass to @test
819 * @wait: if true wait for the inode to be unlocked, if false do not
821 * ifind() searches for the inode specified by @data in the inode
822 * cache. This is a generalized version of ifind_fast() for file systems where
823 * the inode number is not sufficient for unique identification of an inode.
825 * If the inode is in the cache, the inode is returned with an incremented
826 * reference count.
828 * Otherwise NULL is returned.
830 * Note, @test is called with the inode_lock held, so can't sleep.
832 static struct inode *ifind(struct super_block *sb,
833 struct hlist_head *head, int (*test)(struct inode *, void *),
834 void *data, const int wait)
836 struct inode *inode;
838 spin_lock(&inode_lock);
839 inode = find_inode(sb, head, test, data);
840 if (inode) {
841 __iget(inode);
842 spin_unlock(&inode_lock);
843 if (likely(wait))
844 wait_on_inode(inode);
845 return inode;
847 spin_unlock(&inode_lock);
848 return NULL;
852 * ifind_fast - internal function, you want ilookup() or iget().
853 * @sb: super block of file system to search
854 * @head: head of the list to search
855 * @ino: inode number to search for
857 * ifind_fast() searches for the inode @ino in the inode cache. This is for
858 * file systems where the inode number is sufficient for unique identification
859 * of an inode.
861 * If the inode is in the cache, the inode is returned with an incremented
862 * reference count.
864 * Otherwise NULL is returned.
866 static struct inode *ifind_fast(struct super_block *sb,
867 struct hlist_head *head, unsigned long ino)
869 struct inode *inode;
871 spin_lock(&inode_lock);
872 inode = find_inode_fast(sb, head, ino);
873 if (inode) {
874 __iget(inode);
875 spin_unlock(&inode_lock);
876 wait_on_inode(inode);
877 return inode;
879 spin_unlock(&inode_lock);
880 return NULL;
884 * ilookup5_nowait - search for an inode in the inode cache
885 * @sb: super block of file system to search
886 * @hashval: hash value (usually inode number) to search for
887 * @test: callback used for comparisons between inodes
888 * @data: opaque data pointer to pass to @test
890 * ilookup5() uses ifind() to search for the inode specified by @hashval and
891 * @data in the inode cache. This is a generalized version of ilookup() for
892 * file systems where the inode number is not sufficient for unique
893 * identification of an inode.
895 * If the inode is in the cache, the inode is returned with an incremented
896 * reference count. Note, the inode lock is not waited upon so you have to be
897 * very careful what you do with the returned inode. You probably should be
898 * using ilookup5() instead.
900 * Otherwise NULL is returned.
902 * Note, @test is called with the inode_lock held, so can't sleep.
904 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
905 int (*test)(struct inode *, void *), void *data)
907 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
909 return ifind(sb, head, test, data, 0);
912 EXPORT_SYMBOL(ilookup5_nowait);
915 * ilookup5 - search for an inode in the inode cache
916 * @sb: super block of file system to search
917 * @hashval: hash value (usually inode number) to search for
918 * @test: callback used for comparisons between inodes
919 * @data: opaque data pointer to pass to @test
921 * ilookup5() uses ifind() to search for the inode specified by @hashval and
922 * @data in the inode cache. This is a generalized version of ilookup() for
923 * file systems where the inode number is not sufficient for unique
924 * identification of an inode.
926 * If the inode is in the cache, the inode lock is waited upon and the inode is
927 * returned with an incremented reference count.
929 * Otherwise NULL is returned.
931 * Note, @test is called with the inode_lock held, so can't sleep.
933 struct inode *ilookup5(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, 1);
941 EXPORT_SYMBOL(ilookup5);
944 * ilookup - search for an inode in the inode cache
945 * @sb: super block of file system to search
946 * @ino: inode number to search for
948 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
949 * This is for file systems where the inode number is sufficient for unique
950 * identification of an inode.
952 * If the inode is in the cache, the inode is returned with an incremented
953 * reference count.
955 * Otherwise NULL is returned.
957 struct inode *ilookup(struct super_block *sb, unsigned long ino)
959 struct hlist_head *head = inode_hashtable + hash(sb, ino);
961 return ifind_fast(sb, head, ino);
964 EXPORT_SYMBOL(ilookup);
967 * iget5_locked - obtain an inode from a mounted file system
968 * @sb: super block of file system
969 * @hashval: hash value (usually inode number) to get
970 * @test: callback used for comparisons between inodes
971 * @set: callback used to initialize a new struct inode
972 * @data: opaque data pointer to pass to @test and @set
974 * iget5_locked() uses ifind() to search for the inode specified by @hashval
975 * and @data in the inode cache and if present it is returned with an increased
976 * reference count. This is a generalized version of iget_locked() for file
977 * systems where the inode number is not sufficient for unique identification
978 * of an inode.
980 * If the inode is not in cache, get_new_inode() is called to allocate a new
981 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
982 * file system gets to fill it in before unlocking it via unlock_new_inode().
984 * Note both @test and @set are called with the inode_lock held, so can't sleep.
986 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
987 int (*test)(struct inode *, void *),
988 int (*set)(struct inode *, void *), void *data)
990 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
991 struct inode *inode;
993 inode = ifind(sb, head, test, data, 1);
994 if (inode)
995 return inode;
997 * get_new_inode() will do the right thing, re-trying the search
998 * in case it had to block at any point.
1000 return get_new_inode(sb, head, test, set, data);
1003 EXPORT_SYMBOL(iget5_locked);
1006 * iget_locked - obtain an inode from a mounted file system
1007 * @sb: super block of file system
1008 * @ino: inode number to get
1010 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1011 * the inode cache and if present it is returned with an increased reference
1012 * count. This is for file systems where the inode number is sufficient for
1013 * unique identification of an inode.
1015 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1016 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1017 * The file system gets to fill it in before unlocking it via
1018 * unlock_new_inode().
1020 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1022 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1023 struct inode *inode;
1025 inode = ifind_fast(sb, head, ino);
1026 if (inode)
1027 return inode;
1029 * get_new_inode_fast() will do the right thing, re-trying the search
1030 * in case it had to block at any point.
1032 return get_new_inode_fast(sb, head, ino);
1035 EXPORT_SYMBOL(iget_locked);
1037 int insert_inode_locked(struct inode *inode)
1039 struct super_block *sb = inode->i_sb;
1040 ino_t ino = inode->i_ino;
1041 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1042 struct inode *old;
1044 inode->i_state |= I_LOCK|I_NEW;
1045 while (1) {
1046 spin_lock(&inode_lock);
1047 old = find_inode_fast(sb, head, ino);
1048 if (likely(!old)) {
1049 hlist_add_head(&inode->i_hash, head);
1050 spin_unlock(&inode_lock);
1051 return 0;
1053 __iget(old);
1054 spin_unlock(&inode_lock);
1055 wait_on_inode(old);
1056 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1057 iput(old);
1058 return -EBUSY;
1060 iput(old);
1064 EXPORT_SYMBOL(insert_inode_locked);
1066 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1067 int (*test)(struct inode *, void *), void *data)
1069 struct super_block *sb = inode->i_sb;
1070 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1071 struct inode *old;
1073 inode->i_state |= I_LOCK|I_NEW;
1075 while (1) {
1076 spin_lock(&inode_lock);
1077 old = find_inode(sb, head, test, data);
1078 if (likely(!old)) {
1079 hlist_add_head(&inode->i_hash, head);
1080 spin_unlock(&inode_lock);
1081 return 0;
1083 __iget(old);
1084 spin_unlock(&inode_lock);
1085 wait_on_inode(old);
1086 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1087 iput(old);
1088 return -EBUSY;
1090 iput(old);
1094 EXPORT_SYMBOL(insert_inode_locked4);
1097 * __insert_inode_hash - hash an inode
1098 * @inode: unhashed inode
1099 * @hashval: unsigned long value used to locate this object in the
1100 * inode_hashtable.
1102 * Add an inode to the inode hash for this superblock.
1104 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1106 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1107 spin_lock(&inode_lock);
1108 hlist_add_head(&inode->i_hash, head);
1109 spin_unlock(&inode_lock);
1112 EXPORT_SYMBOL(__insert_inode_hash);
1115 * remove_inode_hash - remove an inode from the hash
1116 * @inode: inode to unhash
1118 * Remove an inode from the superblock.
1120 void remove_inode_hash(struct inode *inode)
1122 spin_lock(&inode_lock);
1123 hlist_del_init(&inode->i_hash);
1124 spin_unlock(&inode_lock);
1127 EXPORT_SYMBOL(remove_inode_hash);
1130 * Tell the filesystem that this inode is no longer of any interest and should
1131 * be completely destroyed.
1133 * We leave the inode in the inode hash table until *after* the filesystem's
1134 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1135 * instigate) will always find up-to-date information either in the hash or on
1136 * disk.
1138 * I_FREEING is set so that no-one will take a new reference to the inode while
1139 * it is being deleted.
1141 void generic_delete_inode(struct inode *inode)
1143 const struct super_operations *op = inode->i_sb->s_op;
1145 list_del_init(&inode->i_list);
1146 list_del_init(&inode->i_sb_list);
1147 inode->i_state |= I_FREEING;
1148 inodes_stat.nr_inodes--;
1149 spin_unlock(&inode_lock);
1151 security_inode_delete(inode);
1153 if (op->delete_inode) {
1154 void (*delete)(struct inode *) = op->delete_inode;
1155 if (!is_bad_inode(inode))
1156 DQUOT_INIT(inode);
1157 /* Filesystems implementing their own
1158 * s_op->delete_inode are required to call
1159 * truncate_inode_pages and clear_inode()
1160 * internally */
1161 delete(inode);
1162 } else {
1163 truncate_inode_pages(&inode->i_data, 0);
1164 clear_inode(inode);
1166 spin_lock(&inode_lock);
1167 hlist_del_init(&inode->i_hash);
1168 spin_unlock(&inode_lock);
1169 wake_up_inode(inode);
1170 BUG_ON(inode->i_state != I_CLEAR);
1171 destroy_inode(inode);
1174 EXPORT_SYMBOL(generic_delete_inode);
1176 static void generic_forget_inode(struct inode *inode)
1178 struct super_block *sb = inode->i_sb;
1180 if (!hlist_unhashed(&inode->i_hash)) {
1181 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1182 list_move(&inode->i_list, &inode_unused);
1183 inodes_stat.nr_unused++;
1184 if (sb->s_flags & MS_ACTIVE) {
1185 spin_unlock(&inode_lock);
1186 return;
1188 inode->i_state |= I_WILL_FREE;
1189 spin_unlock(&inode_lock);
1190 write_inode_now(inode, 1);
1191 spin_lock(&inode_lock);
1192 inode->i_state &= ~I_WILL_FREE;
1193 inodes_stat.nr_unused--;
1194 hlist_del_init(&inode->i_hash);
1196 list_del_init(&inode->i_list);
1197 list_del_init(&inode->i_sb_list);
1198 inode->i_state |= I_FREEING;
1199 inodes_stat.nr_inodes--;
1200 spin_unlock(&inode_lock);
1201 if (inode->i_data.nrpages)
1202 truncate_inode_pages(&inode->i_data, 0);
1203 clear_inode(inode);
1204 wake_up_inode(inode);
1205 destroy_inode(inode);
1209 * Normal UNIX filesystem behaviour: delete the
1210 * inode when the usage count drops to zero, and
1211 * i_nlink is zero.
1213 void generic_drop_inode(struct inode *inode)
1215 if (!inode->i_nlink)
1216 generic_delete_inode(inode);
1217 else
1218 generic_forget_inode(inode);
1221 EXPORT_SYMBOL_GPL(generic_drop_inode);
1224 * Called when we're dropping the last reference
1225 * to an inode.
1227 * Call the FS "drop()" function, defaulting to
1228 * the legacy UNIX filesystem behaviour..
1230 * NOTE! NOTE! NOTE! We're called with the inode lock
1231 * held, and the drop function is supposed to release
1232 * the lock!
1234 static inline void iput_final(struct inode *inode)
1236 const struct super_operations *op = inode->i_sb->s_op;
1237 void (*drop)(struct inode *) = generic_drop_inode;
1239 if (op && op->drop_inode)
1240 drop = op->drop_inode;
1241 drop(inode);
1245 * iput - put an inode
1246 * @inode: inode to put
1248 * Puts an inode, dropping its usage count. If the inode use count hits
1249 * zero, the inode is then freed and may also be destroyed.
1251 * Consequently, iput() can sleep.
1253 void iput(struct inode *inode)
1255 if (inode) {
1256 BUG_ON(inode->i_state == I_CLEAR);
1258 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1259 iput_final(inode);
1263 EXPORT_SYMBOL(iput);
1266 * bmap - find a block number in a file
1267 * @inode: inode of file
1268 * @block: block to find
1270 * Returns the block number on the device holding the inode that
1271 * is the disk block number for the block of the file requested.
1272 * That is, asked for block 4 of inode 1 the function will return the
1273 * disk block relative to the disk start that holds that block of the
1274 * file.
1276 sector_t bmap(struct inode * inode, sector_t block)
1278 sector_t res = 0;
1279 if (inode->i_mapping->a_ops->bmap)
1280 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1281 return res;
1283 EXPORT_SYMBOL(bmap);
1286 * touch_atime - update the access time
1287 * @mnt: mount the inode is accessed on
1288 * @dentry: dentry accessed
1290 * Update the accessed time on an inode and mark it for writeback.
1291 * This function automatically handles read only file systems and media,
1292 * as well as the "noatime" flag and inode specific "noatime" markers.
1294 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1296 struct inode *inode = dentry->d_inode;
1297 struct timespec now;
1299 if (mnt_want_write(mnt))
1300 return;
1301 if (inode->i_flags & S_NOATIME)
1302 goto out;
1303 if (IS_NOATIME(inode))
1304 goto out;
1305 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1306 goto out;
1308 if (mnt->mnt_flags & MNT_NOATIME)
1309 goto out;
1310 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1311 goto out;
1312 if (mnt->mnt_flags & MNT_RELATIME) {
1314 * With relative atime, only update atime if the previous
1315 * atime is earlier than either the ctime or mtime.
1317 if (timespec_compare(&inode->i_mtime, &inode->i_atime) < 0 &&
1318 timespec_compare(&inode->i_ctime, &inode->i_atime) < 0)
1319 goto out;
1322 now = current_fs_time(inode->i_sb);
1323 if (timespec_equal(&inode->i_atime, &now))
1324 goto out;
1326 inode->i_atime = now;
1327 mark_inode_dirty_sync(inode);
1328 out:
1329 mnt_drop_write(mnt);
1331 EXPORT_SYMBOL(touch_atime);
1334 * file_update_time - update mtime and ctime time
1335 * @file: file accessed
1337 * Update the mtime and ctime members of an inode and mark the inode
1338 * for writeback. Note that this function is meant exclusively for
1339 * usage in the file write path of filesystems, and filesystems may
1340 * choose to explicitly ignore update via this function with the
1341 * S_NOCTIME inode flag, e.g. for network filesystem where these
1342 * timestamps are handled by the server.
1345 void file_update_time(struct file *file)
1347 struct inode *inode = file->f_path.dentry->d_inode;
1348 struct timespec now;
1349 int sync_it = 0;
1350 int err;
1352 if (IS_NOCMTIME(inode))
1353 return;
1355 err = mnt_want_write(file->f_path.mnt);
1356 if (err)
1357 return;
1359 now = current_fs_time(inode->i_sb);
1360 if (!timespec_equal(&inode->i_mtime, &now)) {
1361 inode->i_mtime = now;
1362 sync_it = 1;
1365 if (!timespec_equal(&inode->i_ctime, &now)) {
1366 inode->i_ctime = now;
1367 sync_it = 1;
1370 if (IS_I_VERSION(inode)) {
1371 inode_inc_iversion(inode);
1372 sync_it = 1;
1375 if (sync_it)
1376 mark_inode_dirty_sync(inode);
1377 mnt_drop_write(file->f_path.mnt);
1380 EXPORT_SYMBOL(file_update_time);
1382 int inode_needs_sync(struct inode *inode)
1384 if (IS_SYNC(inode))
1385 return 1;
1386 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1387 return 1;
1388 return 0;
1391 EXPORT_SYMBOL(inode_needs_sync);
1393 int inode_wait(void *word)
1395 schedule();
1396 return 0;
1398 EXPORT_SYMBOL(inode_wait);
1401 * If we try to find an inode in the inode hash while it is being
1402 * deleted, we have to wait until the filesystem completes its
1403 * deletion before reporting that it isn't found. This function waits
1404 * until the deletion _might_ have completed. Callers are responsible
1405 * to recheck inode state.
1407 * It doesn't matter if I_LOCK is not set initially, a call to
1408 * wake_up_inode() after removing from the hash list will DTRT.
1410 * This is called with inode_lock held.
1412 static void __wait_on_freeing_inode(struct inode *inode)
1414 wait_queue_head_t *wq;
1415 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1416 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1417 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1418 spin_unlock(&inode_lock);
1419 schedule();
1420 finish_wait(wq, &wait.wait);
1421 spin_lock(&inode_lock);
1425 * We rarely want to lock two inodes that do not have a parent/child
1426 * relationship (such as directory, child inode) simultaneously. The
1427 * vast majority of file systems should be able to get along fine
1428 * without this. Do not use these functions except as a last resort.
1430 void inode_double_lock(struct inode *inode1, struct inode *inode2)
1432 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1433 if (inode1)
1434 mutex_lock(&inode1->i_mutex);
1435 else if (inode2)
1436 mutex_lock(&inode2->i_mutex);
1437 return;
1440 if (inode1 < inode2) {
1441 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1442 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1443 } else {
1444 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1445 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1448 EXPORT_SYMBOL(inode_double_lock);
1450 void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1452 if (inode1)
1453 mutex_unlock(&inode1->i_mutex);
1455 if (inode2 && inode2 != inode1)
1456 mutex_unlock(&inode2->i_mutex);
1458 EXPORT_SYMBOL(inode_double_unlock);
1460 static __initdata unsigned long ihash_entries;
1461 static int __init set_ihash_entries(char *str)
1463 if (!str)
1464 return 0;
1465 ihash_entries = simple_strtoul(str, &str, 0);
1466 return 1;
1468 __setup("ihash_entries=", set_ihash_entries);
1471 * Initialize the waitqueues and inode hash table.
1473 void __init inode_init_early(void)
1475 int loop;
1477 /* If hashes are distributed across NUMA nodes, defer
1478 * hash allocation until vmalloc space is available.
1480 if (hashdist)
1481 return;
1483 inode_hashtable =
1484 alloc_large_system_hash("Inode-cache",
1485 sizeof(struct hlist_head),
1486 ihash_entries,
1488 HASH_EARLY,
1489 &i_hash_shift,
1490 &i_hash_mask,
1493 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1494 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1497 void __init inode_init(void)
1499 int loop;
1501 /* inode slab cache */
1502 inode_cachep = kmem_cache_create("inode_cache",
1503 sizeof(struct inode),
1505 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1506 SLAB_MEM_SPREAD),
1507 init_once);
1508 register_shrinker(&icache_shrinker);
1510 /* Hash may have been set up in inode_init_early */
1511 if (!hashdist)
1512 return;
1514 inode_hashtable =
1515 alloc_large_system_hash("Inode-cache",
1516 sizeof(struct hlist_head),
1517 ihash_entries,
1520 &i_hash_shift,
1521 &i_hash_mask,
1524 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1525 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1528 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1530 inode->i_mode = mode;
1531 if (S_ISCHR(mode)) {
1532 inode->i_fop = &def_chr_fops;
1533 inode->i_rdev = rdev;
1534 } else if (S_ISBLK(mode)) {
1535 inode->i_fop = &def_blk_fops;
1536 inode->i_rdev = rdev;
1537 } else if (S_ISFIFO(mode))
1538 inode->i_fop = &def_fifo_fops;
1539 else if (S_ISSOCK(mode))
1540 inode->i_fop = &bad_sock_fops;
1541 else
1542 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1543 mode);
1545 EXPORT_SYMBOL(init_special_inode);