[PATCH] wrong error path in dup_fd() leading to oopses in RCU
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / inode.c
blob32b7c3375021fdacaade973b59ccb350899b2dd6
1 /*
2 * linux/fs/inode.c
4 * (C) 1997 Linus Torvalds
5 */
7 #include <linux/config.h>
8 #include <linux/fs.h>
9 #include <linux/mm.h>
10 #include <linux/dcache.h>
11 #include <linux/init.h>
12 #include <linux/quotaops.h>
13 #include <linux/slab.h>
14 #include <linux/writeback.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/wait.h>
18 #include <linux/hash.h>
19 #include <linux/swap.h>
20 #include <linux/security.h>
21 #include <linux/pagemap.h>
22 #include <linux/cdev.h>
23 #include <linux/bootmem.h>
24 #include <linux/inotify.h>
25 #include <linux/mount.h>
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 kmem_cache_t * inode_cachep __read_mostly;
103 static struct inode *alloc_inode(struct super_block *sb)
105 static struct address_space_operations empty_aops;
106 static struct inode_operations empty_iops;
107 static const struct file_operations empty_fops;
108 struct inode *inode;
110 if (sb->s_op->alloc_inode)
111 inode = sb->s_op->alloc_inode(sb);
112 else
113 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
115 if (inode) {
116 struct address_space * const mapping = &inode->i_data;
118 inode->i_sb = sb;
119 inode->i_blkbits = sb->s_blocksize_bits;
120 inode->i_flags = 0;
121 atomic_set(&inode->i_count, 1);
122 inode->i_op = &empty_iops;
123 inode->i_fop = &empty_fops;
124 inode->i_nlink = 1;
125 atomic_set(&inode->i_writecount, 0);
126 inode->i_size = 0;
127 inode->i_blocks = 0;
128 inode->i_bytes = 0;
129 inode->i_generation = 0;
130 #ifdef CONFIG_QUOTA
131 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
132 #endif
133 inode->i_pipe = NULL;
134 inode->i_bdev = NULL;
135 inode->i_cdev = NULL;
136 inode->i_rdev = 0;
137 inode->i_security = NULL;
138 inode->dirtied_when = 0;
139 if (security_inode_alloc(inode)) {
140 if (inode->i_sb->s_op->destroy_inode)
141 inode->i_sb->s_op->destroy_inode(inode);
142 else
143 kmem_cache_free(inode_cachep, (inode));
144 return NULL;
147 mapping->a_ops = &empty_aops;
148 mapping->host = inode;
149 mapping->flags = 0;
150 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
151 mapping->assoc_mapping = NULL;
152 mapping->backing_dev_info = &default_backing_dev_info;
155 * If the block_device provides a backing_dev_info for client
156 * inodes then use that. Otherwise the inode share the bdev's
157 * backing_dev_info.
159 if (sb->s_bdev) {
160 struct backing_dev_info *bdi;
162 bdi = sb->s_bdev->bd_inode_backing_dev_info;
163 if (!bdi)
164 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
165 mapping->backing_dev_info = bdi;
167 memset(&inode->u, 0, sizeof(inode->u));
168 inode->i_mapping = mapping;
170 return inode;
173 void destroy_inode(struct inode *inode)
175 if (inode_has_buffers(inode))
176 BUG();
177 security_inode_free(inode);
178 if (inode->i_sb->s_op->destroy_inode)
179 inode->i_sb->s_op->destroy_inode(inode);
180 else
181 kmem_cache_free(inode_cachep, (inode));
186 * These are initializations that only need to be done
187 * once, because the fields are idempotent across use
188 * of the inode, so let the slab aware of that.
190 void inode_init_once(struct inode *inode)
192 memset(inode, 0, sizeof(*inode));
193 INIT_HLIST_NODE(&inode->i_hash);
194 INIT_LIST_HEAD(&inode->i_dentry);
195 INIT_LIST_HEAD(&inode->i_devices);
196 mutex_init(&inode->i_mutex);
197 init_rwsem(&inode->i_alloc_sem);
198 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
199 rwlock_init(&inode->i_data.tree_lock);
200 spin_lock_init(&inode->i_data.i_mmap_lock);
201 INIT_LIST_HEAD(&inode->i_data.private_list);
202 spin_lock_init(&inode->i_data.private_lock);
203 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
204 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
205 spin_lock_init(&inode->i_lock);
206 i_size_ordered_init(inode);
207 #ifdef CONFIG_INOTIFY
208 INIT_LIST_HEAD(&inode->inotify_watches);
209 mutex_init(&inode->inotify_mutex);
210 #endif
213 EXPORT_SYMBOL(inode_init_once);
215 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
217 struct inode * inode = (struct inode *) foo;
219 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
220 SLAB_CTOR_CONSTRUCTOR)
221 inode_init_once(inode);
225 * inode_lock must be held
227 void __iget(struct inode * inode)
229 if (atomic_read(&inode->i_count)) {
230 atomic_inc(&inode->i_count);
231 return;
233 atomic_inc(&inode->i_count);
234 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
235 list_move(&inode->i_list, &inode_in_use);
236 inodes_stat.nr_unused--;
240 * clear_inode - clear an inode
241 * @inode: inode to clear
243 * This is called by the filesystem to tell us
244 * that the inode is no longer useful. We just
245 * terminate it with extreme prejudice.
247 void clear_inode(struct inode *inode)
249 might_sleep();
250 invalidate_inode_buffers(inode);
252 if (inode->i_data.nrpages)
253 BUG();
254 if (!(inode->i_state & I_FREEING))
255 BUG();
256 if (inode->i_state & I_CLEAR)
257 BUG();
258 wait_on_inode(inode);
259 DQUOT_DROP(inode);
260 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
261 inode->i_sb->s_op->clear_inode(inode);
262 if (inode->i_bdev)
263 bd_forget(inode);
264 if (inode->i_cdev)
265 cd_forget(inode);
266 inode->i_state = I_CLEAR;
269 EXPORT_SYMBOL(clear_inode);
272 * dispose_list - dispose of the contents of a local list
273 * @head: the head of the list to free
275 * Dispose-list gets a local list with local inodes in it, so it doesn't
276 * need to worry about list corruption and SMP locks.
278 static void dispose_list(struct list_head *head)
280 int nr_disposed = 0;
282 while (!list_empty(head)) {
283 struct inode *inode;
285 inode = list_entry(head->next, struct inode, i_list);
286 list_del(&inode->i_list);
288 if (inode->i_data.nrpages)
289 truncate_inode_pages(&inode->i_data, 0);
290 clear_inode(inode);
292 spin_lock(&inode_lock);
293 hlist_del_init(&inode->i_hash);
294 list_del_init(&inode->i_sb_list);
295 spin_unlock(&inode_lock);
297 wake_up_inode(inode);
298 destroy_inode(inode);
299 nr_disposed++;
301 spin_lock(&inode_lock);
302 inodes_stat.nr_inodes -= nr_disposed;
303 spin_unlock(&inode_lock);
307 * Invalidate all inodes for a device.
309 static int invalidate_list(struct list_head *head, struct list_head *dispose)
311 struct list_head *next;
312 int busy = 0, count = 0;
314 next = head->next;
315 for (;;) {
316 struct list_head * tmp = next;
317 struct inode * inode;
320 * We can reschedule here without worrying about the list's
321 * consistency because the per-sb list of inodes must not
322 * change during umount anymore, and because iprune_mutex keeps
323 * shrink_icache_memory() away.
325 cond_resched_lock(&inode_lock);
327 next = next->next;
328 if (tmp == head)
329 break;
330 inode = list_entry(tmp, struct inode, i_sb_list);
331 invalidate_inode_buffers(inode);
332 if (!atomic_read(&inode->i_count)) {
333 list_move(&inode->i_list, dispose);
334 inode->i_state |= I_FREEING;
335 count++;
336 continue;
338 busy = 1;
340 /* only unused inodes may be cached with i_count zero */
341 inodes_stat.nr_unused -= count;
342 return busy;
346 * invalidate_inodes - discard the inodes on a device
347 * @sb: superblock
349 * Discard all of the inodes for a given superblock. If the discard
350 * fails because there are busy inodes then a non zero value is returned.
351 * If the discard is successful all the inodes have been discarded.
353 int invalidate_inodes(struct super_block * sb)
355 int busy;
356 LIST_HEAD(throw_away);
358 mutex_lock(&iprune_mutex);
359 spin_lock(&inode_lock);
360 inotify_unmount_inodes(&sb->s_inodes);
361 busy = invalidate_list(&sb->s_inodes, &throw_away);
362 spin_unlock(&inode_lock);
364 dispose_list(&throw_away);
365 mutex_unlock(&iprune_mutex);
367 return busy;
370 EXPORT_SYMBOL(invalidate_inodes);
372 int __invalidate_device(struct block_device *bdev)
374 struct super_block *sb = get_super(bdev);
375 int res = 0;
377 if (sb) {
379 * no need to lock the super, get_super holds the
380 * read mutex so the filesystem cannot go away
381 * under us (->put_super runs with the write lock
382 * hold).
384 shrink_dcache_sb(sb);
385 res = invalidate_inodes(sb);
386 drop_super(sb);
388 invalidate_bdev(bdev, 0);
389 return res;
391 EXPORT_SYMBOL(__invalidate_device);
393 static int can_unuse(struct inode *inode)
395 if (inode->i_state)
396 return 0;
397 if (inode_has_buffers(inode))
398 return 0;
399 if (atomic_read(&inode->i_count))
400 return 0;
401 if (inode->i_data.nrpages)
402 return 0;
403 return 1;
407 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
408 * a temporary list and then are freed outside inode_lock by dispose_list().
410 * Any inodes which are pinned purely because of attached pagecache have their
411 * pagecache removed. We expect the final iput() on that inode to add it to
412 * the front of the inode_unused list. So look for it there and if the
413 * inode is still freeable, proceed. The right inode is found 99.9% of the
414 * time in testing on a 4-way.
416 * If the inode has metadata buffers attached to mapping->private_list then
417 * try to remove them.
419 static void prune_icache(int nr_to_scan)
421 LIST_HEAD(freeable);
422 int nr_pruned = 0;
423 int nr_scanned;
424 unsigned long reap = 0;
426 mutex_lock(&iprune_mutex);
427 spin_lock(&inode_lock);
428 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
429 struct inode *inode;
431 if (list_empty(&inode_unused))
432 break;
434 inode = list_entry(inode_unused.prev, struct inode, i_list);
436 if (inode->i_state || atomic_read(&inode->i_count)) {
437 list_move(&inode->i_list, &inode_unused);
438 continue;
440 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
441 __iget(inode);
442 spin_unlock(&inode_lock);
443 if (remove_inode_buffers(inode))
444 reap += invalidate_inode_pages(&inode->i_data);
445 iput(inode);
446 spin_lock(&inode_lock);
448 if (inode != list_entry(inode_unused.next,
449 struct inode, i_list))
450 continue; /* wrong inode or list_empty */
451 if (!can_unuse(inode))
452 continue;
454 list_move(&inode->i_list, &freeable);
455 inode->i_state |= I_FREEING;
456 nr_pruned++;
458 inodes_stat.nr_unused -= nr_pruned;
459 spin_unlock(&inode_lock);
461 dispose_list(&freeable);
462 mutex_unlock(&iprune_mutex);
464 if (current_is_kswapd())
465 mod_page_state(kswapd_inodesteal, reap);
466 else
467 mod_page_state(pginodesteal, reap);
471 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
472 * "unused" means that no dentries are referring to the inodes: the files are
473 * not open and the dcache references to those inodes have already been
474 * reclaimed.
476 * This function is passed the number of inodes to scan, and it returns the
477 * total number of remaining possibly-reclaimable inodes.
479 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
481 if (nr) {
483 * Nasty deadlock avoidance. We may hold various FS locks,
484 * and we don't want to recurse into the FS that called us
485 * in clear_inode() and friends..
487 if (!(gfp_mask & __GFP_FS))
488 return -1;
489 prune_icache(nr);
491 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
494 static void __wait_on_freeing_inode(struct inode *inode);
496 * Called with the inode lock held.
497 * NOTE: we are not increasing the inode-refcount, you must call __iget()
498 * by hand after calling find_inode now! This simplifies iunique and won't
499 * add any additional branch in the common code.
501 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
503 struct hlist_node *node;
504 struct inode * inode = NULL;
506 repeat:
507 hlist_for_each (node, head) {
508 inode = hlist_entry(node, struct inode, i_hash);
509 if (inode->i_sb != sb)
510 continue;
511 if (!test(inode, data))
512 continue;
513 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
514 __wait_on_freeing_inode(inode);
515 goto repeat;
517 break;
519 return node ? inode : NULL;
523 * find_inode_fast is the fast path version of find_inode, see the comment at
524 * iget_locked for details.
526 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
528 struct hlist_node *node;
529 struct inode * inode = NULL;
531 repeat:
532 hlist_for_each (node, head) {
533 inode = hlist_entry(node, struct inode, i_hash);
534 if (inode->i_ino != ino)
535 continue;
536 if (inode->i_sb != sb)
537 continue;
538 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
539 __wait_on_freeing_inode(inode);
540 goto repeat;
542 break;
544 return node ? inode : NULL;
548 * new_inode - obtain an inode
549 * @sb: superblock
551 * Allocates a new inode for given superblock.
553 struct inode *new_inode(struct super_block *sb)
555 static unsigned long last_ino;
556 struct inode * inode;
558 spin_lock_prefetch(&inode_lock);
560 inode = alloc_inode(sb);
561 if (inode) {
562 spin_lock(&inode_lock);
563 inodes_stat.nr_inodes++;
564 list_add(&inode->i_list, &inode_in_use);
565 list_add(&inode->i_sb_list, &sb->s_inodes);
566 inode->i_ino = ++last_ino;
567 inode->i_state = 0;
568 spin_unlock(&inode_lock);
570 return inode;
573 EXPORT_SYMBOL(new_inode);
575 void unlock_new_inode(struct inode *inode)
578 * This is special! We do not need the spinlock
579 * when clearing I_LOCK, because we're guaranteed
580 * that nobody else tries to do anything about the
581 * state of the inode when it is locked, as we
582 * just created it (so there can be no old holders
583 * that haven't tested I_LOCK).
585 inode->i_state &= ~(I_LOCK|I_NEW);
586 wake_up_inode(inode);
589 EXPORT_SYMBOL(unlock_new_inode);
592 * This is called without the inode lock held.. Be careful.
594 * We no longer cache the sb_flags in i_flags - see fs.h
595 * -- rmk@arm.uk.linux.org
597 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)
599 struct inode * inode;
601 inode = alloc_inode(sb);
602 if (inode) {
603 struct inode * old;
605 spin_lock(&inode_lock);
606 /* We released the lock, so.. */
607 old = find_inode(sb, head, test, data);
608 if (!old) {
609 if (set(inode, data))
610 goto set_failed;
612 inodes_stat.nr_inodes++;
613 list_add(&inode->i_list, &inode_in_use);
614 list_add(&inode->i_sb_list, &sb->s_inodes);
615 hlist_add_head(&inode->i_hash, head);
616 inode->i_state = I_LOCK|I_NEW;
617 spin_unlock(&inode_lock);
619 /* Return the locked inode with I_NEW set, the
620 * caller is responsible for filling in the contents
622 return inode;
626 * Uhhuh, somebody else created the same inode under
627 * us. Use the old inode instead of the one we just
628 * allocated.
630 __iget(old);
631 spin_unlock(&inode_lock);
632 destroy_inode(inode);
633 inode = old;
634 wait_on_inode(inode);
636 return inode;
638 set_failed:
639 spin_unlock(&inode_lock);
640 destroy_inode(inode);
641 return NULL;
645 * get_new_inode_fast is the fast path version of get_new_inode, see the
646 * comment at iget_locked for details.
648 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
650 struct inode * inode;
652 inode = alloc_inode(sb);
653 if (inode) {
654 struct inode * old;
656 spin_lock(&inode_lock);
657 /* We released the lock, so.. */
658 old = find_inode_fast(sb, head, ino);
659 if (!old) {
660 inode->i_ino = ino;
661 inodes_stat.nr_inodes++;
662 list_add(&inode->i_list, &inode_in_use);
663 list_add(&inode->i_sb_list, &sb->s_inodes);
664 hlist_add_head(&inode->i_hash, head);
665 inode->i_state = I_LOCK|I_NEW;
666 spin_unlock(&inode_lock);
668 /* Return the locked inode with I_NEW set, the
669 * caller is responsible for filling in the contents
671 return inode;
675 * Uhhuh, somebody else created the same inode under
676 * us. Use the old inode instead of the one we just
677 * allocated.
679 __iget(old);
680 spin_unlock(&inode_lock);
681 destroy_inode(inode);
682 inode = old;
683 wait_on_inode(inode);
685 return inode;
688 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
690 unsigned long tmp;
692 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
693 L1_CACHE_BYTES;
694 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
695 return tmp & I_HASHMASK;
699 * iunique - get a unique inode number
700 * @sb: superblock
701 * @max_reserved: highest reserved inode number
703 * Obtain an inode number that is unique on the system for a given
704 * superblock. This is used by file systems that have no natural
705 * permanent inode numbering system. An inode number is returned that
706 * is higher than the reserved limit but unique.
708 * BUGS:
709 * With a large number of inodes live on the file system this function
710 * currently becomes quite slow.
712 ino_t iunique(struct super_block *sb, ino_t max_reserved)
714 static ino_t counter;
715 struct inode *inode;
716 struct hlist_head * head;
717 ino_t res;
718 spin_lock(&inode_lock);
719 retry:
720 if (counter > max_reserved) {
721 head = inode_hashtable + hash(sb,counter);
722 res = counter++;
723 inode = find_inode_fast(sb, head, res);
724 if (!inode) {
725 spin_unlock(&inode_lock);
726 return res;
728 } else {
729 counter = max_reserved + 1;
731 goto retry;
735 EXPORT_SYMBOL(iunique);
737 struct inode *igrab(struct inode *inode)
739 spin_lock(&inode_lock);
740 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
741 __iget(inode);
742 else
744 * Handle the case where s_op->clear_inode is not been
745 * called yet, and somebody is calling igrab
746 * while the inode is getting freed.
748 inode = NULL;
749 spin_unlock(&inode_lock);
750 return inode;
753 EXPORT_SYMBOL(igrab);
756 * ifind - internal function, you want ilookup5() or iget5().
757 * @sb: super block of file system to search
758 * @head: the head of the list to search
759 * @test: callback used for comparisons between inodes
760 * @data: opaque data pointer to pass to @test
761 * @wait: if true wait for the inode to be unlocked, if false do not
763 * ifind() searches for the inode specified by @data in the inode
764 * cache. This is a generalized version of ifind_fast() for file systems where
765 * the inode number is not sufficient for unique identification of an inode.
767 * If the inode is in the cache, the inode is returned with an incremented
768 * reference count.
770 * Otherwise NULL is returned.
772 * Note, @test is called with the inode_lock held, so can't sleep.
774 static struct inode *ifind(struct super_block *sb,
775 struct hlist_head *head, int (*test)(struct inode *, void *),
776 void *data, const int wait)
778 struct inode *inode;
780 spin_lock(&inode_lock);
781 inode = find_inode(sb, head, test, data);
782 if (inode) {
783 __iget(inode);
784 spin_unlock(&inode_lock);
785 if (likely(wait))
786 wait_on_inode(inode);
787 return inode;
789 spin_unlock(&inode_lock);
790 return NULL;
794 * ifind_fast - internal function, you want ilookup() or iget().
795 * @sb: super block of file system to search
796 * @head: head of the list to search
797 * @ino: inode number to search for
799 * ifind_fast() searches for the inode @ino in the inode cache. This is for
800 * file systems where the inode number is sufficient for unique identification
801 * of an inode.
803 * If the inode is in the cache, the inode is returned with an incremented
804 * reference count.
806 * Otherwise NULL is returned.
808 static struct inode *ifind_fast(struct super_block *sb,
809 struct hlist_head *head, unsigned long ino)
811 struct inode *inode;
813 spin_lock(&inode_lock);
814 inode = find_inode_fast(sb, head, ino);
815 if (inode) {
816 __iget(inode);
817 spin_unlock(&inode_lock);
818 wait_on_inode(inode);
819 return inode;
821 spin_unlock(&inode_lock);
822 return NULL;
826 * ilookup5_nowait - search for an inode in the inode cache
827 * @sb: super block of file system to search
828 * @hashval: hash value (usually inode number) to search for
829 * @test: callback used for comparisons between inodes
830 * @data: opaque data pointer to pass to @test
832 * ilookup5() uses ifind() to search for the inode specified by @hashval and
833 * @data in the inode cache. This is a generalized version of ilookup() for
834 * file systems where the inode number is not sufficient for unique
835 * identification of an inode.
837 * If the inode is in the cache, the inode is returned with an incremented
838 * reference count. Note, the inode lock is not waited upon so you have to be
839 * very careful what you do with the returned inode. You probably should be
840 * using ilookup5() instead.
842 * Otherwise NULL is returned.
844 * Note, @test is called with the inode_lock held, so can't sleep.
846 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
847 int (*test)(struct inode *, void *), void *data)
849 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
851 return ifind(sb, head, test, data, 0);
854 EXPORT_SYMBOL(ilookup5_nowait);
857 * ilookup5 - search for an inode in the inode cache
858 * @sb: super block of file system to search
859 * @hashval: hash value (usually inode number) to search for
860 * @test: callback used for comparisons between inodes
861 * @data: opaque data pointer to pass to @test
863 * ilookup5() uses ifind() to search for the inode specified by @hashval and
864 * @data in the inode cache. This is a generalized version of ilookup() for
865 * file systems where the inode number is not sufficient for unique
866 * identification of an inode.
868 * If the inode is in the cache, the inode lock is waited upon and the inode is
869 * returned with an incremented reference count.
871 * Otherwise NULL is returned.
873 * Note, @test is called with the inode_lock held, so can't sleep.
875 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
876 int (*test)(struct inode *, void *), void *data)
878 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
880 return ifind(sb, head, test, data, 1);
883 EXPORT_SYMBOL(ilookup5);
886 * ilookup - search for an inode in the inode cache
887 * @sb: super block of file system to search
888 * @ino: inode number to search for
890 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
891 * This is for file systems where the inode number is sufficient for unique
892 * identification of an inode.
894 * If the inode is in the cache, the inode is returned with an incremented
895 * reference count.
897 * Otherwise NULL is returned.
899 struct inode *ilookup(struct super_block *sb, unsigned long ino)
901 struct hlist_head *head = inode_hashtable + hash(sb, ino);
903 return ifind_fast(sb, head, ino);
906 EXPORT_SYMBOL(ilookup);
909 * iget5_locked - obtain an inode from a mounted file system
910 * @sb: super block of file system
911 * @hashval: hash value (usually inode number) to get
912 * @test: callback used for comparisons between inodes
913 * @set: callback used to initialize a new struct inode
914 * @data: opaque data pointer to pass to @test and @set
916 * This is iget() without the read_inode() portion of get_new_inode().
918 * iget5_locked() uses ifind() to search for the inode specified by @hashval
919 * and @data in the inode cache and if present it is returned with an increased
920 * reference count. This is a generalized version of iget_locked() for file
921 * systems where the inode number is not sufficient for unique identification
922 * of an inode.
924 * If the inode is not in cache, get_new_inode() is called to allocate a new
925 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
926 * file system gets to fill it in before unlocking it via unlock_new_inode().
928 * Note both @test and @set are called with the inode_lock held, so can't sleep.
930 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
931 int (*test)(struct inode *, void *),
932 int (*set)(struct inode *, void *), void *data)
934 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
935 struct inode *inode;
937 inode = ifind(sb, head, test, data, 1);
938 if (inode)
939 return inode;
941 * get_new_inode() will do the right thing, re-trying the search
942 * in case it had to block at any point.
944 return get_new_inode(sb, head, test, set, data);
947 EXPORT_SYMBOL(iget5_locked);
950 * iget_locked - obtain an inode from a mounted file system
951 * @sb: super block of file system
952 * @ino: inode number to get
954 * This is iget() without the read_inode() portion of get_new_inode_fast().
956 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
957 * the inode cache and if present it is returned with an increased reference
958 * count. This is for file systems where the inode number is sufficient for
959 * unique identification of an inode.
961 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
962 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
963 * The file system gets to fill it in before unlocking it via
964 * unlock_new_inode().
966 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
968 struct hlist_head *head = inode_hashtable + hash(sb, ino);
969 struct inode *inode;
971 inode = ifind_fast(sb, head, ino);
972 if (inode)
973 return inode;
975 * get_new_inode_fast() will do the right thing, re-trying the search
976 * in case it had to block at any point.
978 return get_new_inode_fast(sb, head, ino);
981 EXPORT_SYMBOL(iget_locked);
984 * __insert_inode_hash - hash an inode
985 * @inode: unhashed inode
986 * @hashval: unsigned long value used to locate this object in the
987 * inode_hashtable.
989 * Add an inode to the inode hash for this superblock.
991 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
993 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
994 spin_lock(&inode_lock);
995 hlist_add_head(&inode->i_hash, head);
996 spin_unlock(&inode_lock);
999 EXPORT_SYMBOL(__insert_inode_hash);
1002 * remove_inode_hash - remove an inode from the hash
1003 * @inode: inode to unhash
1005 * Remove an inode from the superblock.
1007 void remove_inode_hash(struct inode *inode)
1009 spin_lock(&inode_lock);
1010 hlist_del_init(&inode->i_hash);
1011 spin_unlock(&inode_lock);
1014 EXPORT_SYMBOL(remove_inode_hash);
1017 * Tell the filesystem that this inode is no longer of any interest and should
1018 * be completely destroyed.
1020 * We leave the inode in the inode hash table until *after* the filesystem's
1021 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1022 * instigate) will always find up-to-date information either in the hash or on
1023 * disk.
1025 * I_FREEING is set so that no-one will take a new reference to the inode while
1026 * it is being deleted.
1028 void generic_delete_inode(struct inode *inode)
1030 struct super_operations *op = inode->i_sb->s_op;
1032 list_del_init(&inode->i_list);
1033 list_del_init(&inode->i_sb_list);
1034 inode->i_state|=I_FREEING;
1035 inodes_stat.nr_inodes--;
1036 spin_unlock(&inode_lock);
1038 security_inode_delete(inode);
1040 if (op->delete_inode) {
1041 void (*delete)(struct inode *) = op->delete_inode;
1042 if (!is_bad_inode(inode))
1043 DQUOT_INIT(inode);
1044 /* Filesystems implementing their own
1045 * s_op->delete_inode are required to call
1046 * truncate_inode_pages and clear_inode()
1047 * internally */
1048 delete(inode);
1049 } else {
1050 truncate_inode_pages(&inode->i_data, 0);
1051 clear_inode(inode);
1053 spin_lock(&inode_lock);
1054 hlist_del_init(&inode->i_hash);
1055 spin_unlock(&inode_lock);
1056 wake_up_inode(inode);
1057 if (inode->i_state != I_CLEAR)
1058 BUG();
1059 destroy_inode(inode);
1062 EXPORT_SYMBOL(generic_delete_inode);
1064 static void generic_forget_inode(struct inode *inode)
1066 struct super_block *sb = inode->i_sb;
1068 if (!hlist_unhashed(&inode->i_hash)) {
1069 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1070 list_move(&inode->i_list, &inode_unused);
1071 inodes_stat.nr_unused++;
1072 if (!sb || (sb->s_flags & MS_ACTIVE)) {
1073 spin_unlock(&inode_lock);
1074 return;
1076 inode->i_state |= I_WILL_FREE;
1077 spin_unlock(&inode_lock);
1078 write_inode_now(inode, 1);
1079 spin_lock(&inode_lock);
1080 inode->i_state &= ~I_WILL_FREE;
1081 inodes_stat.nr_unused--;
1082 hlist_del_init(&inode->i_hash);
1084 list_del_init(&inode->i_list);
1085 list_del_init(&inode->i_sb_list);
1086 inode->i_state |= I_FREEING;
1087 inodes_stat.nr_inodes--;
1088 spin_unlock(&inode_lock);
1089 if (inode->i_data.nrpages)
1090 truncate_inode_pages(&inode->i_data, 0);
1091 clear_inode(inode);
1092 wake_up_inode(inode);
1093 destroy_inode(inode);
1097 * Normal UNIX filesystem behaviour: delete the
1098 * inode when the usage count drops to zero, and
1099 * i_nlink is zero.
1101 void generic_drop_inode(struct inode *inode)
1103 if (!inode->i_nlink)
1104 generic_delete_inode(inode);
1105 else
1106 generic_forget_inode(inode);
1109 EXPORT_SYMBOL_GPL(generic_drop_inode);
1112 * Called when we're dropping the last reference
1113 * to an inode.
1115 * Call the FS "drop()" function, defaulting to
1116 * the legacy UNIX filesystem behaviour..
1118 * NOTE! NOTE! NOTE! We're called with the inode lock
1119 * held, and the drop function is supposed to release
1120 * the lock!
1122 static inline void iput_final(struct inode *inode)
1124 struct super_operations *op = inode->i_sb->s_op;
1125 void (*drop)(struct inode *) = generic_drop_inode;
1127 if (op && op->drop_inode)
1128 drop = op->drop_inode;
1129 drop(inode);
1133 * iput - put an inode
1134 * @inode: inode to put
1136 * Puts an inode, dropping its usage count. If the inode use count hits
1137 * zero, the inode is then freed and may also be destroyed.
1139 * Consequently, iput() can sleep.
1141 void iput(struct inode *inode)
1143 if (inode) {
1144 struct super_operations *op = inode->i_sb->s_op;
1146 BUG_ON(inode->i_state == I_CLEAR);
1148 if (op && op->put_inode)
1149 op->put_inode(inode);
1151 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1152 iput_final(inode);
1156 EXPORT_SYMBOL(iput);
1159 * bmap - find a block number in a file
1160 * @inode: inode of file
1161 * @block: block to find
1163 * Returns the block number on the device holding the inode that
1164 * is the disk block number for the block of the file requested.
1165 * That is, asked for block 4 of inode 1 the function will return the
1166 * disk block relative to the disk start that holds that block of the
1167 * file.
1169 sector_t bmap(struct inode * inode, sector_t block)
1171 sector_t res = 0;
1172 if (inode->i_mapping->a_ops->bmap)
1173 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1174 return res;
1177 EXPORT_SYMBOL(bmap);
1180 * touch_atime - update the access time
1181 * @mnt: mount the inode is accessed on
1182 * @dentry: dentry accessed
1184 * Update the accessed time on an inode and mark it for writeback.
1185 * This function automatically handles read only file systems and media,
1186 * as well as the "noatime" flag and inode specific "noatime" markers.
1188 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1190 struct inode *inode = dentry->d_inode;
1191 struct timespec now;
1193 if (IS_RDONLY(inode))
1194 return;
1196 if ((inode->i_flags & S_NOATIME) ||
1197 (inode->i_sb->s_flags & MS_NOATIME) ||
1198 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
1199 return;
1202 * We may have a NULL vfsmount when coming from NFSD
1204 if (mnt &&
1205 ((mnt->mnt_flags & MNT_NOATIME) ||
1206 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))))
1207 return;
1209 now = current_fs_time(inode->i_sb);
1210 if (!timespec_equal(&inode->i_atime, &now)) {
1211 inode->i_atime = now;
1212 mark_inode_dirty_sync(inode);
1216 EXPORT_SYMBOL(touch_atime);
1219 * file_update_time - update mtime and ctime time
1220 * @file: file accessed
1222 * Update the mtime and ctime members of an inode and mark the inode
1223 * for writeback. Note that this function is meant exclusively for
1224 * usage in the file write path of filesystems, and filesystems may
1225 * choose to explicitly ignore update via this function with the
1226 * S_NOCTIME inode flag, e.g. for network filesystem where these
1227 * timestamps are handled by the server.
1230 void file_update_time(struct file *file)
1232 struct inode *inode = file->f_dentry->d_inode;
1233 struct timespec now;
1234 int sync_it = 0;
1236 if (IS_NOCMTIME(inode))
1237 return;
1238 if (IS_RDONLY(inode))
1239 return;
1241 now = current_fs_time(inode->i_sb);
1242 if (!timespec_equal(&inode->i_mtime, &now))
1243 sync_it = 1;
1244 inode->i_mtime = now;
1246 if (!timespec_equal(&inode->i_ctime, &now))
1247 sync_it = 1;
1248 inode->i_ctime = now;
1250 if (sync_it)
1251 mark_inode_dirty_sync(inode);
1254 EXPORT_SYMBOL(file_update_time);
1256 int inode_needs_sync(struct inode *inode)
1258 if (IS_SYNC(inode))
1259 return 1;
1260 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1261 return 1;
1262 return 0;
1265 EXPORT_SYMBOL(inode_needs_sync);
1268 * Quota functions that want to walk the inode lists..
1270 #ifdef CONFIG_QUOTA
1272 /* Function back in dquot.c */
1273 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1275 void remove_dquot_ref(struct super_block *sb, int type,
1276 struct list_head *tofree_head)
1278 struct inode *inode;
1280 if (!sb->dq_op)
1281 return; /* nothing to do */
1282 spin_lock(&inode_lock); /* This lock is for inodes code */
1285 * We don't have to lock against quota code - test IS_QUOTAINIT is
1286 * just for speedup...
1288 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1289 if (!IS_NOQUOTA(inode))
1290 remove_inode_dquot_ref(inode, type, tofree_head);
1292 spin_unlock(&inode_lock);
1295 #endif
1297 int inode_wait(void *word)
1299 schedule();
1300 return 0;
1304 * If we try to find an inode in the inode hash while it is being
1305 * deleted, we have to wait until the filesystem completes its
1306 * deletion before reporting that it isn't found. This function waits
1307 * until the deletion _might_ have completed. Callers are responsible
1308 * to recheck inode state.
1310 * It doesn't matter if I_LOCK is not set initially, a call to
1311 * wake_up_inode() after removing from the hash list will DTRT.
1313 * This is called with inode_lock held.
1315 static void __wait_on_freeing_inode(struct inode *inode)
1317 wait_queue_head_t *wq;
1318 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1319 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1320 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1321 spin_unlock(&inode_lock);
1322 schedule();
1323 finish_wait(wq, &wait.wait);
1324 spin_lock(&inode_lock);
1327 void wake_up_inode(struct inode *inode)
1330 * Prevent speculative execution through spin_unlock(&inode_lock);
1332 smp_mb();
1333 wake_up_bit(&inode->i_state, __I_LOCK);
1336 static __initdata unsigned long ihash_entries;
1337 static int __init set_ihash_entries(char *str)
1339 if (!str)
1340 return 0;
1341 ihash_entries = simple_strtoul(str, &str, 0);
1342 return 1;
1344 __setup("ihash_entries=", set_ihash_entries);
1347 * Initialize the waitqueues and inode hash table.
1349 void __init inode_init_early(void)
1351 int loop;
1353 /* If hashes are distributed across NUMA nodes, defer
1354 * hash allocation until vmalloc space is available.
1356 if (hashdist)
1357 return;
1359 inode_hashtable =
1360 alloc_large_system_hash("Inode-cache",
1361 sizeof(struct hlist_head),
1362 ihash_entries,
1364 HASH_EARLY,
1365 &i_hash_shift,
1366 &i_hash_mask,
1369 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1370 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1373 void __init inode_init(unsigned long mempages)
1375 int loop;
1377 /* inode slab cache */
1378 inode_cachep = kmem_cache_create("inode_cache",
1379 sizeof(struct inode),
1381 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1382 SLAB_MEM_SPREAD),
1383 init_once,
1384 NULL);
1385 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1387 /* Hash may have been set up in inode_init_early */
1388 if (!hashdist)
1389 return;
1391 inode_hashtable =
1392 alloc_large_system_hash("Inode-cache",
1393 sizeof(struct hlist_head),
1394 ihash_entries,
1397 &i_hash_shift,
1398 &i_hash_mask,
1401 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1402 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1405 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1407 inode->i_mode = mode;
1408 if (S_ISCHR(mode)) {
1409 inode->i_fop = &def_chr_fops;
1410 inode->i_rdev = rdev;
1411 } else if (S_ISBLK(mode)) {
1412 inode->i_fop = &def_blk_fops;
1413 inode->i_rdev = rdev;
1414 } else if (S_ISFIFO(mode))
1415 inode->i_fop = &def_fifo_fops;
1416 else if (S_ISSOCK(mode))
1417 inode->i_fop = &bad_sock_fops;
1418 else
1419 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1420 mode);
1422 EXPORT_SYMBOL(init_special_inode);