4 * (C) 1997 Linus Torvalds
7 #include <linux/config.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:
30 * - invalidate_inode_buffers
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".
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;
60 static unsigned int i_hash_shift;
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;
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_sem 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 DECLARE_MUTEX(iprune_sem);
97 * Statistics gathering..
99 struct inodes_stat_t inodes_stat;
101 static kmem_cache_t * inode_cachep;
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 struct file_operations empty_fops;
110 if (sb->s_op->alloc_inode)
111 inode = sb->s_op->alloc_inode(sb);
113 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
116 struct address_space * const mapping = &inode->i_data;
120 /* essential because of inode slab reuse */
122 inode->i_blkbits = sb->s_blocksize_bits;
124 atomic_set(&inode->i_count, 1);
125 inode->i_op = &empty_iops;
126 inode->i_fop = &empty_fops;
128 atomic_set(&inode->i_writecount, 0);
132 inode->i_generation = 0;
134 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
136 inode->i_pipe = NULL;
137 inode->i_bdev = NULL;
138 inode->i_cdev = NULL;
140 inode->i_security = NULL;
141 inode->dirtied_when = 0;
142 if (security_inode_alloc(inode)) {
143 if (inode->i_sb->s_op->destroy_inode)
144 inode->i_sb->s_op->destroy_inode(inode);
146 kmem_cache_free(inode_cachep, (inode));
150 mapping->a_ops = &empty_aops;
151 mapping->host = inode;
153 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
154 mapping->assoc_mapping = NULL;
155 mapping->backing_dev_info = &default_backing_dev_info;
158 * If the block_device provides a backing_dev_info for client
159 * inodes then use that. Otherwise the inode share the bdev's
163 struct backing_dev_info *bdi;
165 bdi = sb->s_bdev->bd_inode_backing_dev_info;
167 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
168 mapping->backing_dev_info = bdi;
170 memset(&inode->u, 0, sizeof(inode->u));
171 inode->i_mapping = mapping;
176 void destroy_inode(struct inode *inode)
178 if (inode_has_buffers(inode))
180 security_inode_free(inode);
181 if (inode->i_sb->s_op->destroy_inode)
182 inode->i_sb->s_op->destroy_inode(inode);
184 kmem_cache_free(inode_cachep, (inode));
189 * These are initializations that only need to be done
190 * once, because the fields are idempotent across use
191 * of the inode, so let the slab aware of that.
193 void inode_init_once(struct inode *inode)
195 memset(inode, 0, sizeof(*inode));
196 INIT_HLIST_NODE(&inode->i_hash);
197 INIT_LIST_HEAD(&inode->i_dentry);
198 INIT_LIST_HEAD(&inode->i_devices);
199 mutex_init(&inode->i_mutex);
200 init_rwsem(&inode->i_alloc_sem);
201 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
202 rwlock_init(&inode->i_data.tree_lock);
203 spin_lock_init(&inode->i_data.i_mmap_lock);
204 INIT_LIST_HEAD(&inode->i_data.private_list);
205 spin_lock_init(&inode->i_data.private_lock);
206 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
207 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
208 spin_lock_init(&inode->i_lock);
209 i_size_ordered_init(inode);
210 #ifdef CONFIG_INOTIFY
211 INIT_LIST_HEAD(&inode->inotify_watches);
212 sema_init(&inode->inotify_sem, 1);
216 EXPORT_SYMBOL(inode_init_once);
218 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
220 struct inode * inode = (struct inode *) foo;
222 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
223 SLAB_CTOR_CONSTRUCTOR)
224 inode_init_once(inode);
228 * inode_lock must be held
230 void __iget(struct inode * inode)
232 if (atomic_read(&inode->i_count)) {
233 atomic_inc(&inode->i_count);
236 atomic_inc(&inode->i_count);
237 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
238 list_move(&inode->i_list, &inode_in_use);
239 inodes_stat.nr_unused--;
242 EXPORT_SYMBOL_GPL(__iget);
245 * clear_inode - clear an inode
246 * @inode: inode to clear
248 * This is called by the filesystem to tell us
249 * that the inode is no longer useful. We just
250 * terminate it with extreme prejudice.
252 void clear_inode(struct inode *inode)
255 invalidate_inode_buffers(inode);
257 if (inode->i_data.nrpages)
259 if (!(inode->i_state & I_FREEING))
261 if (inode->i_state & I_CLEAR)
263 wait_on_inode(inode);
265 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
266 inode->i_sb->s_op->clear_inode(inode);
271 inode->i_state = I_CLEAR;
274 EXPORT_SYMBOL(clear_inode);
277 * dispose_list - dispose of the contents of a local list
278 * @head: the head of the list to free
280 * Dispose-list gets a local list with local inodes in it, so it doesn't
281 * need to worry about list corruption and SMP locks.
283 static void dispose_list(struct list_head *head)
287 while (!list_empty(head)) {
290 inode = list_entry(head->next, struct inode, i_list);
291 list_del(&inode->i_list);
293 if (inode->i_data.nrpages)
294 truncate_inode_pages(&inode->i_data, 0);
297 spin_lock(&inode_lock);
298 hlist_del_init(&inode->i_hash);
299 list_del_init(&inode->i_sb_list);
300 spin_unlock(&inode_lock);
302 wake_up_inode(inode);
303 destroy_inode(inode);
306 spin_lock(&inode_lock);
307 inodes_stat.nr_inodes -= nr_disposed;
308 spin_unlock(&inode_lock);
312 * Invalidate all inodes for a device.
314 static int invalidate_list(struct list_head *head, struct list_head *dispose)
316 struct list_head *next;
317 int busy = 0, count = 0;
321 struct list_head * tmp = next;
322 struct inode * inode;
325 * We can reschedule here without worrying about the list's
326 * consistency because the per-sb list of inodes must not
327 * change during umount anymore, and because iprune_sem keeps
328 * shrink_icache_memory() away.
330 cond_resched_lock(&inode_lock);
335 inode = list_entry(tmp, struct inode, i_sb_list);
336 invalidate_inode_buffers(inode);
337 if (!atomic_read(&inode->i_count)) {
338 list_move(&inode->i_list, dispose);
339 inode->i_state |= I_FREEING;
345 /* only unused inodes may be cached with i_count zero */
346 inodes_stat.nr_unused -= count;
351 * invalidate_inodes - discard the inodes on a device
354 * Discard all of the inodes for a given superblock. If the discard
355 * fails because there are busy inodes then a non zero value is returned.
356 * If the discard is successful all the inodes have been discarded.
358 int invalidate_inodes(struct super_block * sb)
361 LIST_HEAD(throw_away);
364 spin_lock(&inode_lock);
365 inotify_unmount_inodes(&sb->s_inodes);
366 busy = invalidate_list(&sb->s_inodes, &throw_away);
367 spin_unlock(&inode_lock);
369 dispose_list(&throw_away);
375 EXPORT_SYMBOL(invalidate_inodes);
377 int __invalidate_device(struct block_device *bdev)
379 struct super_block *sb = get_super(bdev);
384 * no need to lock the super, get_super holds the
385 * read semaphore so the filesystem cannot go away
386 * under us (->put_super runs with the write lock
389 shrink_dcache_sb(sb);
390 res = invalidate_inodes(sb);
393 invalidate_bdev(bdev, 0);
396 EXPORT_SYMBOL(__invalidate_device);
398 static int can_unuse(struct inode *inode)
402 if (inode_has_buffers(inode))
404 if (atomic_read(&inode->i_count))
406 if (inode->i_data.nrpages)
412 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
413 * a temporary list and then are freed outside inode_lock by dispose_list().
415 * Any inodes which are pinned purely because of attached pagecache have their
416 * pagecache removed. We expect the final iput() on that inode to add it to
417 * the front of the inode_unused list. So look for it there and if the
418 * inode is still freeable, proceed. The right inode is found 99.9% of the
419 * time in testing on a 4-way.
421 * If the inode has metadata buffers attached to mapping->private_list then
422 * try to remove them.
424 static void prune_icache(int nr_to_scan)
429 unsigned long reap = 0;
432 spin_lock(&inode_lock);
433 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
436 if (list_empty(&inode_unused))
439 inode = list_entry(inode_unused.prev, struct inode, i_list);
441 if (inode->i_state || atomic_read(&inode->i_count)) {
442 list_move(&inode->i_list, &inode_unused);
445 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
447 spin_unlock(&inode_lock);
448 if (remove_inode_buffers(inode))
449 reap += invalidate_inode_pages(&inode->i_data);
451 spin_lock(&inode_lock);
453 if (inode != list_entry(inode_unused.next,
454 struct inode, i_list))
455 continue; /* wrong inode or list_empty */
456 if (!can_unuse(inode))
459 list_move(&inode->i_list, &freeable);
460 inode->i_state |= I_FREEING;
463 inodes_stat.nr_unused -= nr_pruned;
464 spin_unlock(&inode_lock);
466 dispose_list(&freeable);
469 if (current_is_kswapd())
470 mod_page_state(kswapd_inodesteal, reap);
472 mod_page_state(pginodesteal, reap);
476 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
477 * "unused" means that no dentries are referring to the inodes: the files are
478 * not open and the dcache references to those inodes have already been
481 * This function is passed the number of inodes to scan, and it returns the
482 * total number of remaining possibly-reclaimable inodes.
484 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
488 * Nasty deadlock avoidance. We may hold various FS locks,
489 * and we don't want to recurse into the FS that called us
490 * in clear_inode() and friends..
492 if (!(gfp_mask & __GFP_FS))
496 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
499 static void __wait_on_freeing_inode(struct inode *inode);
501 * Called with the inode lock held.
502 * NOTE: we are not increasing the inode-refcount, you must call __iget()
503 * by hand after calling find_inode now! This simplifies iunique and won't
504 * add any additional branch in the common code.
506 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
508 struct hlist_node *node;
509 struct inode * inode = NULL;
512 hlist_for_each (node, head) {
513 inode = hlist_entry(node, struct inode, i_hash);
514 if (inode->i_sb != sb)
516 if (!test(inode, data))
518 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
519 __wait_on_freeing_inode(inode);
524 return node ? inode : NULL;
528 * find_inode_fast is the fast path version of find_inode, see the comment at
529 * iget_locked for details.
531 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
533 struct hlist_node *node;
534 struct inode * inode = NULL;
537 hlist_for_each (node, head) {
538 inode = hlist_entry(node, struct inode, i_hash);
539 if (inode->i_ino != ino)
541 if (inode->i_sb != sb)
543 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
544 __wait_on_freeing_inode(inode);
549 return node ? inode : NULL;
553 * new_inode - obtain an inode
556 * Allocates a new inode for given superblock.
558 struct inode *new_inode(struct super_block *sb)
560 static unsigned long last_ino;
561 struct inode * inode;
563 spin_lock_prefetch(&inode_lock);
565 inode = alloc_inode(sb);
567 spin_lock(&inode_lock);
568 inodes_stat.nr_inodes++;
569 list_add(&inode->i_list, &inode_in_use);
570 list_add(&inode->i_sb_list, &sb->s_inodes);
571 inode->i_ino = ++last_ino;
573 spin_unlock(&inode_lock);
578 EXPORT_SYMBOL(new_inode);
580 void unlock_new_inode(struct inode *inode)
583 * This is special! We do not need the spinlock
584 * when clearing I_LOCK, because we're guaranteed
585 * that nobody else tries to do anything about the
586 * state of the inode when it is locked, as we
587 * just created it (so there can be no old holders
588 * that haven't tested I_LOCK).
590 inode->i_state &= ~(I_LOCK|I_NEW);
591 wake_up_inode(inode);
594 EXPORT_SYMBOL(unlock_new_inode);
597 * This is called without the inode lock held.. Be careful.
599 * We no longer cache the sb_flags in i_flags - see fs.h
600 * -- rmk@arm.uk.linux.org
602 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)
604 struct inode * inode;
606 inode = alloc_inode(sb);
610 spin_lock(&inode_lock);
611 /* We released the lock, so.. */
612 old = find_inode(sb, head, test, data);
614 if (set(inode, data))
617 inodes_stat.nr_inodes++;
618 list_add(&inode->i_list, &inode_in_use);
619 list_add(&inode->i_sb_list, &sb->s_inodes);
620 hlist_add_head(&inode->i_hash, head);
621 inode->i_state = I_LOCK|I_NEW;
622 spin_unlock(&inode_lock);
624 /* Return the locked inode with I_NEW set, the
625 * caller is responsible for filling in the contents
631 * Uhhuh, somebody else created the same inode under
632 * us. Use the old inode instead of the one we just
636 spin_unlock(&inode_lock);
637 destroy_inode(inode);
639 wait_on_inode(inode);
644 spin_unlock(&inode_lock);
645 destroy_inode(inode);
650 * get_new_inode_fast is the fast path version of get_new_inode, see the
651 * comment at iget_locked for details.
653 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
655 struct inode * inode;
657 inode = alloc_inode(sb);
661 spin_lock(&inode_lock);
662 /* We released the lock, so.. */
663 old = find_inode_fast(sb, head, ino);
666 inodes_stat.nr_inodes++;
667 list_add(&inode->i_list, &inode_in_use);
668 list_add(&inode->i_sb_list, &sb->s_inodes);
669 hlist_add_head(&inode->i_hash, head);
670 inode->i_state = I_LOCK|I_NEW;
671 spin_unlock(&inode_lock);
673 /* Return the locked inode with I_NEW set, the
674 * caller is responsible for filling in the contents
680 * Uhhuh, somebody else created the same inode under
681 * us. Use the old inode instead of the one we just
685 spin_unlock(&inode_lock);
686 destroy_inode(inode);
688 wait_on_inode(inode);
693 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
697 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
699 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
700 return tmp & I_HASHMASK;
704 * iunique - get a unique inode number
706 * @max_reserved: highest reserved inode number
708 * Obtain an inode number that is unique on the system for a given
709 * superblock. This is used by file systems that have no natural
710 * permanent inode numbering system. An inode number is returned that
711 * is higher than the reserved limit but unique.
714 * With a large number of inodes live on the file system this function
715 * currently becomes quite slow.
717 ino_t iunique(struct super_block *sb, ino_t max_reserved)
719 static ino_t counter;
721 struct hlist_head * head;
723 spin_lock(&inode_lock);
725 if (counter > max_reserved) {
726 head = inode_hashtable + hash(sb,counter);
728 inode = find_inode_fast(sb, head, res);
730 spin_unlock(&inode_lock);
734 counter = max_reserved + 1;
740 EXPORT_SYMBOL(iunique);
742 struct inode *igrab(struct inode *inode)
744 spin_lock(&inode_lock);
745 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
749 * Handle the case where s_op->clear_inode is not been
750 * called yet, and somebody is calling igrab
751 * while the inode is getting freed.
754 spin_unlock(&inode_lock);
758 EXPORT_SYMBOL(igrab);
761 * ifind - internal function, you want ilookup5() or iget5().
762 * @sb: super block of file system to search
763 * @head: the head of the list to search
764 * @test: callback used for comparisons between inodes
765 * @data: opaque data pointer to pass to @test
766 * @wait: if true wait for the inode to be unlocked, if false do not
768 * ifind() searches for the inode specified by @data in the inode
769 * cache. This is a generalized version of ifind_fast() for file systems where
770 * the inode number is not sufficient for unique identification of an inode.
772 * If the inode is in the cache, the inode is returned with an incremented
775 * Otherwise NULL is returned.
777 * Note, @test is called with the inode_lock held, so can't sleep.
779 static struct inode *ifind(struct super_block *sb,
780 struct hlist_head *head, int (*test)(struct inode *, void *),
781 void *data, const int wait)
785 spin_lock(&inode_lock);
786 inode = find_inode(sb, head, test, data);
789 spin_unlock(&inode_lock);
791 wait_on_inode(inode);
794 spin_unlock(&inode_lock);
799 * ifind_fast - internal function, you want ilookup() or iget().
800 * @sb: super block of file system to search
801 * @head: head of the list to search
802 * @ino: inode number to search for
804 * ifind_fast() searches for the inode @ino in the inode cache. This is for
805 * file systems where the inode number is sufficient for unique identification
808 * If the inode is in the cache, the inode is returned with an incremented
811 * Otherwise NULL is returned.
813 static struct inode *ifind_fast(struct super_block *sb,
814 struct hlist_head *head, unsigned long ino)
818 spin_lock(&inode_lock);
819 inode = find_inode_fast(sb, head, ino);
822 spin_unlock(&inode_lock);
823 wait_on_inode(inode);
826 spin_unlock(&inode_lock);
831 * ilookup5_nowait - search for an inode in the inode cache
832 * @sb: super block of file system to search
833 * @hashval: hash value (usually inode number) to search for
834 * @test: callback used for comparisons between inodes
835 * @data: opaque data pointer to pass to @test
837 * ilookup5() uses ifind() to search for the inode specified by @hashval and
838 * @data in the inode cache. This is a generalized version of ilookup() for
839 * file systems where the inode number is not sufficient for unique
840 * identification of an inode.
842 * If the inode is in the cache, the inode is returned with an incremented
843 * reference count. Note, the inode lock is not waited upon so you have to be
844 * very careful what you do with the returned inode. You probably should be
845 * using ilookup5() instead.
847 * Otherwise NULL is returned.
849 * Note, @test is called with the inode_lock held, so can't sleep.
851 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
852 int (*test)(struct inode *, void *), void *data)
854 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
856 return ifind(sb, head, test, data, 0);
859 EXPORT_SYMBOL(ilookup5_nowait);
862 * ilookup5 - search for an inode in the inode cache
863 * @sb: super block of file system to search
864 * @hashval: hash value (usually inode number) to search for
865 * @test: callback used for comparisons between inodes
866 * @data: opaque data pointer to pass to @test
868 * ilookup5() uses ifind() to search for the inode specified by @hashval and
869 * @data in the inode cache. This is a generalized version of ilookup() for
870 * file systems where the inode number is not sufficient for unique
871 * identification of an inode.
873 * If the inode is in the cache, the inode lock is waited upon and the inode is
874 * returned with an incremented reference count.
876 * Otherwise NULL is returned.
878 * Note, @test is called with the inode_lock held, so can't sleep.
880 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
881 int (*test)(struct inode *, void *), void *data)
883 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
885 return ifind(sb, head, test, data, 1);
888 EXPORT_SYMBOL(ilookup5);
891 * ilookup - search for an inode in the inode cache
892 * @sb: super block of file system to search
893 * @ino: inode number to search for
895 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
896 * This is for file systems where the inode number is sufficient for unique
897 * identification of an inode.
899 * If the inode is in the cache, the inode is returned with an incremented
902 * Otherwise NULL is returned.
904 struct inode *ilookup(struct super_block *sb, unsigned long ino)
906 struct hlist_head *head = inode_hashtable + hash(sb, ino);
908 return ifind_fast(sb, head, ino);
911 EXPORT_SYMBOL(ilookup);
914 * iget5_locked - obtain an inode from a mounted file system
915 * @sb: super block of file system
916 * @hashval: hash value (usually inode number) to get
917 * @test: callback used for comparisons between inodes
918 * @set: callback used to initialize a new struct inode
919 * @data: opaque data pointer to pass to @test and @set
921 * This is iget() without the read_inode() portion of get_new_inode().
923 * iget5_locked() uses ifind() to search for the inode specified by @hashval
924 * and @data in the inode cache and if present it is returned with an increased
925 * reference count. This is a generalized version of iget_locked() for file
926 * systems where the inode number is not sufficient for unique identification
929 * If the inode is not in cache, get_new_inode() is called to allocate a new
930 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
931 * file system gets to fill it in before unlocking it via unlock_new_inode().
933 * Note both @test and @set are called with the inode_lock held, so can't sleep.
935 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
936 int (*test)(struct inode *, void *),
937 int (*set)(struct inode *, void *), void *data)
939 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
942 inode = ifind(sb, head, test, data, 1);
946 * get_new_inode() will do the right thing, re-trying the search
947 * in case it had to block at any point.
949 return get_new_inode(sb, head, test, set, data);
952 EXPORT_SYMBOL(iget5_locked);
955 * iget_locked - obtain an inode from a mounted file system
956 * @sb: super block of file system
957 * @ino: inode number to get
959 * This is iget() without the read_inode() portion of get_new_inode_fast().
961 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
962 * the inode cache and if present it is returned with an increased reference
963 * count. This is for file systems where the inode number is sufficient for
964 * unique identification of an inode.
966 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
967 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
968 * The file system gets to fill it in before unlocking it via
969 * unlock_new_inode().
971 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
973 struct hlist_head *head = inode_hashtable + hash(sb, ino);
976 inode = ifind_fast(sb, head, ino);
980 * get_new_inode_fast() will do the right thing, re-trying the search
981 * in case it had to block at any point.
983 return get_new_inode_fast(sb, head, ino);
986 EXPORT_SYMBOL(iget_locked);
989 * __insert_inode_hash - hash an inode
990 * @inode: unhashed inode
991 * @hashval: unsigned long value used to locate this object in the
994 * Add an inode to the inode hash for this superblock.
996 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
998 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
999 spin_lock(&inode_lock);
1000 hlist_add_head(&inode->i_hash, head);
1001 spin_unlock(&inode_lock);
1004 EXPORT_SYMBOL(__insert_inode_hash);
1007 * remove_inode_hash - remove an inode from the hash
1008 * @inode: inode to unhash
1010 * Remove an inode from the superblock.
1012 void remove_inode_hash(struct inode *inode)
1014 spin_lock(&inode_lock);
1015 hlist_del_init(&inode->i_hash);
1016 spin_unlock(&inode_lock);
1019 EXPORT_SYMBOL(remove_inode_hash);
1022 * Tell the filesystem that this inode is no longer of any interest and should
1023 * be completely destroyed.
1025 * We leave the inode in the inode hash table until *after* the filesystem's
1026 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1027 * instigate) will always find up-to-date information either in the hash or on
1030 * I_FREEING is set so that no-one will take a new reference to the inode while
1031 * it is being deleted.
1033 void generic_delete_inode(struct inode *inode)
1035 struct super_operations *op = inode->i_sb->s_op;
1037 list_del_init(&inode->i_list);
1038 list_del_init(&inode->i_sb_list);
1039 inode->i_state|=I_FREEING;
1040 inodes_stat.nr_inodes--;
1041 spin_unlock(&inode_lock);
1043 security_inode_delete(inode);
1045 if (op->delete_inode) {
1046 void (*delete)(struct inode *) = op->delete_inode;
1047 if (!is_bad_inode(inode))
1049 /* Filesystems implementing their own
1050 * s_op->delete_inode are required to call
1051 * truncate_inode_pages and clear_inode()
1055 truncate_inode_pages(&inode->i_data, 0);
1058 spin_lock(&inode_lock);
1059 hlist_del_init(&inode->i_hash);
1060 spin_unlock(&inode_lock);
1061 wake_up_inode(inode);
1062 if (inode->i_state != I_CLEAR)
1064 destroy_inode(inode);
1067 EXPORT_SYMBOL(generic_delete_inode);
1069 static void generic_forget_inode(struct inode *inode)
1071 struct super_block *sb = inode->i_sb;
1073 if (!hlist_unhashed(&inode->i_hash)) {
1074 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1075 list_move(&inode->i_list, &inode_unused);
1076 inodes_stat.nr_unused++;
1077 if (!sb || (sb->s_flags & MS_ACTIVE)) {
1078 spin_unlock(&inode_lock);
1081 inode->i_state |= I_WILL_FREE;
1082 spin_unlock(&inode_lock);
1083 write_inode_now(inode, 1);
1084 spin_lock(&inode_lock);
1085 inode->i_state &= ~I_WILL_FREE;
1086 inodes_stat.nr_unused--;
1087 hlist_del_init(&inode->i_hash);
1089 list_del_init(&inode->i_list);
1090 list_del_init(&inode->i_sb_list);
1091 inode->i_state |= I_FREEING;
1092 inodes_stat.nr_inodes--;
1093 spin_unlock(&inode_lock);
1094 if (inode->i_data.nrpages)
1095 truncate_inode_pages(&inode->i_data, 0);
1097 wake_up_inode(inode);
1098 destroy_inode(inode);
1102 * Normal UNIX filesystem behaviour: delete the
1103 * inode when the usage count drops to zero, and
1106 void generic_drop_inode(struct inode *inode)
1108 if (!inode->i_nlink)
1109 generic_delete_inode(inode);
1111 generic_forget_inode(inode);
1114 EXPORT_SYMBOL_GPL(generic_drop_inode);
1117 * Called when we're dropping the last reference
1120 * Call the FS "drop()" function, defaulting to
1121 * the legacy UNIX filesystem behaviour..
1123 * NOTE! NOTE! NOTE! We're called with the inode lock
1124 * held, and the drop function is supposed to release
1127 static inline void iput_final(struct inode *inode)
1129 struct super_operations *op = inode->i_sb->s_op;
1130 void (*drop)(struct inode *) = generic_drop_inode;
1132 if (op && op->drop_inode)
1133 drop = op->drop_inode;
1138 * iput - put an inode
1139 * @inode: inode to put
1141 * Puts an inode, dropping its usage count. If the inode use count hits
1142 * zero, the inode is then freed and may also be destroyed.
1144 * Consequently, iput() can sleep.
1146 void iput(struct inode *inode)
1149 struct super_operations *op = inode->i_sb->s_op;
1151 BUG_ON(inode->i_state == I_CLEAR);
1153 if (op && op->put_inode)
1154 op->put_inode(inode);
1156 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1161 EXPORT_SYMBOL(iput);
1164 * bmap - find a block number in a file
1165 * @inode: inode of file
1166 * @block: block to find
1168 * Returns the block number on the device holding the inode that
1169 * is the disk block number for the block of the file requested.
1170 * That is, asked for block 4 of inode 1 the function will return the
1171 * disk block relative to the disk start that holds that block of the
1174 sector_t bmap(struct inode * inode, sector_t block)
1177 if (inode->i_mapping->a_ops->bmap)
1178 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1182 EXPORT_SYMBOL(bmap);
1185 * touch_atime - update the access time
1186 * @mnt: mount the inode is accessed on
1187 * @dentry: dentry accessed
1189 * Update the accessed time on an inode and mark it for writeback.
1190 * This function automatically handles read only file systems and media,
1191 * as well as the "noatime" flag and inode specific "noatime" markers.
1193 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1195 struct inode *inode = dentry->d_inode;
1196 struct timespec now;
1198 if (IS_RDONLY(inode))
1201 if ((inode->i_flags & S_NOATIME) ||
1202 (inode->i_sb->s_flags & MS_NOATIME) ||
1203 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
1207 * We may have a NULL vfsmount when coming from NFSD
1210 ((mnt->mnt_flags & MNT_NOATIME) ||
1211 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))))
1214 now = current_fs_time(inode->i_sb);
1215 if (!timespec_equal(&inode->i_atime, &now)) {
1216 inode->i_atime = now;
1217 mark_inode_dirty_sync(inode);
1221 EXPORT_SYMBOL(touch_atime);
1224 * file_update_time - update mtime and ctime time
1225 * @file: file accessed
1227 * Update the mtime and ctime members of an inode and mark the inode
1228 * for writeback. Note that this function is meant exclusively for
1229 * usage in the file write path of filesystems, and filesystems may
1230 * choose to explicitly ignore update via this function with the
1231 * S_NOCTIME inode flag, e.g. for network filesystem where these
1232 * timestamps are handled by the server.
1235 void file_update_time(struct file *file)
1237 struct inode *inode = file->f_dentry->d_inode;
1238 struct timespec now;
1241 if (IS_NOCMTIME(inode))
1243 if (IS_RDONLY(inode))
1246 now = current_fs_time(inode->i_sb);
1247 if (!timespec_equal(&inode->i_mtime, &now))
1249 inode->i_mtime = now;
1251 if (!timespec_equal(&inode->i_ctime, &now))
1253 inode->i_ctime = now;
1256 mark_inode_dirty_sync(inode);
1259 EXPORT_SYMBOL(file_update_time);
1261 int inode_needs_sync(struct inode *inode)
1265 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1270 EXPORT_SYMBOL(inode_needs_sync);
1273 * Quota functions that want to walk the inode lists..
1277 /* Function back in dquot.c */
1278 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1280 void remove_dquot_ref(struct super_block *sb, int type,
1281 struct list_head *tofree_head)
1283 struct inode *inode;
1286 return; /* nothing to do */
1287 spin_lock(&inode_lock); /* This lock is for inodes code */
1290 * We don't have to lock against quota code - test IS_QUOTAINIT is
1291 * just for speedup...
1293 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1294 if (!IS_NOQUOTA(inode))
1295 remove_inode_dquot_ref(inode, type, tofree_head);
1297 spin_unlock(&inode_lock);
1302 int inode_wait(void *word)
1309 * If we try to find an inode in the inode hash while it is being
1310 * deleted, we have to wait until the filesystem completes its
1311 * deletion before reporting that it isn't found. This function waits
1312 * until the deletion _might_ have completed. Callers are responsible
1313 * to recheck inode state.
1315 * It doesn't matter if I_LOCK is not set initially, a call to
1316 * wake_up_inode() after removing from the hash list will DTRT.
1318 * This is called with inode_lock held.
1320 static void __wait_on_freeing_inode(struct inode *inode)
1322 wait_queue_head_t *wq;
1323 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1324 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1325 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1326 spin_unlock(&inode_lock);
1328 finish_wait(wq, &wait.wait);
1329 spin_lock(&inode_lock);
1332 void wake_up_inode(struct inode *inode)
1335 * Prevent speculative execution through spin_unlock(&inode_lock);
1338 wake_up_bit(&inode->i_state, __I_LOCK);
1341 static __initdata unsigned long ihash_entries;
1342 static int __init set_ihash_entries(char *str)
1346 ihash_entries = simple_strtoul(str, &str, 0);
1349 __setup("ihash_entries=", set_ihash_entries);
1352 * Initialize the waitqueues and inode hash table.
1354 void __init inode_init_early(void)
1358 /* If hashes are distributed across NUMA nodes, defer
1359 * hash allocation until vmalloc space is available.
1365 alloc_large_system_hash("Inode-cache",
1366 sizeof(struct hlist_head),
1374 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1375 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1378 void __init inode_init(unsigned long mempages)
1382 /* inode slab cache */
1383 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1384 0, SLAB_RECLAIM_ACCOUNT|SLAB_PANIC, init_once, NULL);
1385 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1387 /* Hash may have been set up in inode_init_early */
1392 alloc_large_system_hash("Inode-cache",
1393 sizeof(struct hlist_head),
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;
1419 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1422 EXPORT_SYMBOL(init_special_inode);