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>
26 * This is needed for the following functions:
28 * - invalidate_inode_buffers
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".
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;
59 static unsigned int i_hash_shift;
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;
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 spinlock_t inode_lock = SPIN_LOCK_UNLOCKED;
86 * iprune_sem 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 DECLARE_MUTEX(iprune_sem);
96 * Statistics gathering..
98 struct inodes_stat_t inodes_stat;
100 static kmem_cache_t * inode_cachep;
102 static struct inode *alloc_inode(struct super_block *sb)
104 static struct address_space_operations empty_aops;
105 static struct inode_operations empty_iops;
106 static struct file_operations empty_fops;
109 if (sb->s_op->alloc_inode)
110 inode = sb->s_op->alloc_inode(sb);
112 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
115 struct address_space * const mapping = &inode->i_data;
118 inode->i_blkbits = sb->s_blocksize_bits;
120 atomic_set(&inode->i_count, 1);
122 inode->i_op = &empty_iops;
123 inode->i_fop = &empty_fops;
125 atomic_set(&inode->i_writecount, 0);
129 inode->i_generation = 0;
131 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
133 inode->i_pipe = NULL;
134 inode->i_bdev = NULL;
135 inode->i_cdev = NULL;
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);
143 kmem_cache_free(inode_cachep, (inode));
147 mapping->a_ops = &empty_aops;
148 mapping->host = inode;
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
160 struct backing_dev_info *bdi;
162 bdi = sb->s_bdev->bd_inode_backing_dev_info;
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;
173 void destroy_inode(struct inode *inode)
175 if (inode_has_buffers(inode))
177 security_inode_free(inode);
178 if (inode->i_sb->s_op->destroy_inode)
179 inode->i_sb->s_op->destroy_inode(inode);
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 sema_init(&inode->i_sem, 1);
197 init_rwsem(&inode->i_alloc_sem);
198 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
199 spin_lock_init(&inode->i_data.tree_lock);
200 spin_lock_init(&inode->i_data.i_mmap_lock);
201 atomic_set(&inode->i_data.truncate_count, 0);
202 INIT_LIST_HEAD(&inode->i_data.private_list);
203 spin_lock_init(&inode->i_data.private_lock);
204 INIT_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
205 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
206 spin_lock_init(&inode->i_lock);
207 i_size_ordered_init(inode);
210 EXPORT_SYMBOL(inode_init_once);
212 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
214 struct inode * inode = (struct inode *) foo;
216 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
217 SLAB_CTOR_CONSTRUCTOR)
218 inode_init_once(inode);
222 * inode_lock must be held
224 void __iget(struct inode * inode)
226 if (atomic_read(&inode->i_count)) {
227 atomic_inc(&inode->i_count);
230 atomic_inc(&inode->i_count);
231 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
232 list_move(&inode->i_list, &inode_in_use);
233 inodes_stat.nr_unused--;
237 * clear_inode - clear an inode
238 * @inode: inode to clear
240 * This is called by the filesystem to tell us
241 * that the inode is no longer useful. We just
242 * terminate it with extreme prejudice.
244 void clear_inode(struct inode *inode)
246 invalidate_inode_buffers(inode);
248 if (inode->i_data.nrpages)
250 if (!(inode->i_state & I_FREEING))
252 if (inode->i_state & I_CLEAR)
254 wait_on_inode(inode);
256 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
257 inode->i_sb->s_op->clear_inode(inode);
262 inode->i_state = I_CLEAR;
265 EXPORT_SYMBOL(clear_inode);
268 * dispose_list - dispose of the contents of a local list
269 * @head: the head of the list to free
271 * Dispose-list gets a local list with local inodes in it, so it doesn't
272 * need to worry about list corruption and SMP locks.
274 static void dispose_list(struct list_head *head)
278 while (!list_empty(head)) {
281 inode = list_entry(head->next, struct inode, i_list);
282 list_del(&inode->i_list);
284 if (inode->i_data.nrpages)
285 truncate_inode_pages(&inode->i_data, 0);
287 destroy_inode(inode);
290 spin_lock(&inode_lock);
291 inodes_stat.nr_inodes -= nr_disposed;
292 spin_unlock(&inode_lock);
296 * Invalidate all inodes for a device.
298 static int invalidate_list(struct list_head *head, struct super_block * sb, struct list_head * dispose)
300 struct list_head *next;
301 int busy = 0, count = 0;
305 struct list_head * tmp = next;
306 struct inode * inode;
311 inode = list_entry(tmp, struct inode, i_list);
312 if (inode->i_sb != sb)
314 invalidate_inode_buffers(inode);
315 if (!atomic_read(&inode->i_count)) {
316 hlist_del_init(&inode->i_hash);
317 list_move(&inode->i_list, dispose);
318 inode->i_state |= I_FREEING;
324 /* only unused inodes may be cached with i_count zero */
325 inodes_stat.nr_unused -= count;
330 * This is a two-stage process. First we collect all
331 * offending inodes onto the throw-away list, and in
332 * the second stage we actually dispose of them. This
333 * is because we don't want to sleep while messing
334 * with the global lists..
338 * invalidate_inodes - discard the inodes on a device
341 * Discard all of the inodes for a given superblock. If the discard
342 * fails because there are busy inodes then a non zero value is returned.
343 * If the discard is successful all the inodes have been discarded.
345 int invalidate_inodes(struct super_block * sb)
348 LIST_HEAD(throw_away);
351 spin_lock(&inode_lock);
352 busy = invalidate_list(&inode_in_use, sb, &throw_away);
353 busy |= invalidate_list(&inode_unused, sb, &throw_away);
354 busy |= invalidate_list(&sb->s_dirty, sb, &throw_away);
355 busy |= invalidate_list(&sb->s_io, sb, &throw_away);
356 spin_unlock(&inode_lock);
358 dispose_list(&throw_away);
364 EXPORT_SYMBOL(invalidate_inodes);
366 int __invalidate_device(struct block_device *bdev, int do_sync)
368 struct super_block *sb;
375 sb = get_super(bdev);
378 * no need to lock the super, get_super holds the
379 * read semaphore so the filesystem cannot go away
380 * under us (->put_super runs with the write lock
383 shrink_dcache_sb(sb);
384 res = invalidate_inodes(sb);
387 invalidate_bdev(bdev, 0);
391 EXPORT_SYMBOL(__invalidate_device);
393 static int can_unuse(struct inode *inode)
397 if (inode_has_buffers(inode))
399 if (atomic_read(&inode->i_count))
401 if (inode->i_data.nrpages)
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)
424 unsigned long reap = 0;
427 spin_lock(&inode_lock);
428 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
431 if (list_empty(&inode_unused))
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);
440 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
442 spin_unlock(&inode_lock);
443 if (remove_inode_buffers(inode))
444 reap += invalidate_inode_pages(&inode->i_data);
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))
454 hlist_del_init(&inode->i_hash);
455 list_move(&inode->i_list, &freeable);
456 inode->i_state |= I_FREEING;
459 inodes_stat.nr_unused -= nr_pruned;
460 spin_unlock(&inode_lock);
462 dispose_list(&freeable);
465 if (current_is_kswapd())
466 mod_page_state(kswapd_inodesteal, reap);
468 mod_page_state(pginodesteal, reap);
472 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
473 * "unused" means that no dentries are referring to the inodes: the files are
474 * not open and the dcache references to those inodes have already been
477 * This function is passed the number of inodes to scan, and it returns the
478 * total number of remaining possibly-reclaimable inodes.
480 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
484 * Nasty deadlock avoidance. We may hold various FS locks,
485 * and we don't want to recurse into the FS that called us
486 * in clear_inode() and friends..
488 if (gfp_mask & __GFP_FS)
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;
507 hlist_for_each (node, head) {
508 inode = hlist_entry(node, struct inode, i_hash);
509 if (inode->i_sb != sb)
511 if (!test(inode, data))
513 if (inode->i_state & (I_FREEING|I_CLEAR)) {
514 __wait_on_freeing_inode(inode);
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;
532 hlist_for_each (node, head) {
533 inode = hlist_entry(node, struct inode, i_hash);
534 if (inode->i_ino != ino)
536 if (inode->i_sb != sb)
538 if (inode->i_state & (I_FREEING|I_CLEAR)) {
539 __wait_on_freeing_inode(inode);
544 return node ? inode : NULL;
548 * new_inode - obtain an inode
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);
562 spin_lock(&inode_lock);
563 inodes_stat.nr_inodes++;
564 list_add(&inode->i_list, &inode_in_use);
565 inode->i_ino = ++last_ino;
567 spin_unlock(&inode_lock);
572 EXPORT_SYMBOL(new_inode);
574 void unlock_new_inode(struct inode *inode)
577 * This is special! We do not need the spinlock
578 * when clearing I_LOCK, because we're guaranteed
579 * that nobody else tries to do anything about the
580 * state of the inode when it is locked, as we
581 * just created it (so there can be no old holders
582 * that haven't tested I_LOCK).
584 inode->i_state &= ~(I_LOCK|I_NEW);
585 wake_up_inode(inode);
588 EXPORT_SYMBOL(unlock_new_inode);
591 * This is called without the inode lock held.. Be careful.
593 * We no longer cache the sb_flags in i_flags - see fs.h
594 * -- rmk@arm.uk.linux.org
596 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)
598 struct inode * inode;
600 inode = alloc_inode(sb);
604 spin_lock(&inode_lock);
605 /* We released the lock, so.. */
606 old = find_inode(sb, head, test, data);
608 if (set(inode, data))
611 inodes_stat.nr_inodes++;
612 list_add(&inode->i_list, &inode_in_use);
613 hlist_add_head(&inode->i_hash, head);
614 inode->i_state = I_LOCK|I_NEW;
615 spin_unlock(&inode_lock);
617 /* Return the locked inode with I_NEW set, the
618 * caller is responsible for filling in the contents
624 * Uhhuh, somebody else created the same inode under
625 * us. Use the old inode instead of the one we just
629 spin_unlock(&inode_lock);
630 destroy_inode(inode);
632 wait_on_inode(inode);
637 spin_unlock(&inode_lock);
638 destroy_inode(inode);
643 * get_new_inode_fast is the fast path version of get_new_inode, see the
644 * comment at iget_locked for details.
646 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
648 struct inode * inode;
650 inode = alloc_inode(sb);
654 spin_lock(&inode_lock);
655 /* We released the lock, so.. */
656 old = find_inode_fast(sb, head, ino);
659 inodes_stat.nr_inodes++;
660 list_add(&inode->i_list, &inode_in_use);
661 hlist_add_head(&inode->i_hash, head);
662 inode->i_state = I_LOCK|I_NEW;
663 spin_unlock(&inode_lock);
665 /* Return the locked inode with I_NEW set, the
666 * caller is responsible for filling in the contents
672 * Uhhuh, somebody else created the same inode under
673 * us. Use the old inode instead of the one we just
677 spin_unlock(&inode_lock);
678 destroy_inode(inode);
680 wait_on_inode(inode);
685 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
689 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
691 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
692 return tmp & I_HASHMASK;
696 * iunique - get a unique inode number
698 * @max_reserved: highest reserved inode number
700 * Obtain an inode number that is unique on the system for a given
701 * superblock. This is used by file systems that have no natural
702 * permanent inode numbering system. An inode number is returned that
703 * is higher than the reserved limit but unique.
706 * With a large number of inodes live on the file system this function
707 * currently becomes quite slow.
709 ino_t iunique(struct super_block *sb, ino_t max_reserved)
711 static ino_t counter;
713 struct hlist_head * head;
715 spin_lock(&inode_lock);
717 if (counter > max_reserved) {
718 head = inode_hashtable + hash(sb,counter);
720 inode = find_inode_fast(sb, head, res);
722 spin_unlock(&inode_lock);
726 counter = max_reserved + 1;
732 EXPORT_SYMBOL(iunique);
734 struct inode *igrab(struct inode *inode)
736 spin_lock(&inode_lock);
737 if (!(inode->i_state & I_FREEING))
741 * Handle the case where s_op->clear_inode is not been
742 * called yet, and somebody is calling igrab
743 * while the inode is getting freed.
746 spin_unlock(&inode_lock);
750 EXPORT_SYMBOL(igrab);
753 * ifind - internal function, you want ilookup5() or iget5().
754 * @sb: super block of file system to search
755 * @head: the head of the list to search
756 * @test: callback used for comparisons between inodes
757 * @data: opaque data pointer to pass to @test
759 * ifind() searches for the inode specified by @data in the inode
760 * cache. This is a generalized version of ifind_fast() for file systems where
761 * the inode number is not sufficient for unique identification of an inode.
763 * If the inode is in the cache, the inode is returned with an incremented
766 * Otherwise NULL is returned.
768 * Note, @test is called with the inode_lock held, so can't sleep.
770 static inline struct inode *ifind(struct super_block *sb,
771 struct hlist_head *head, int (*test)(struct inode *, void *),
776 spin_lock(&inode_lock);
777 inode = find_inode(sb, head, test, data);
780 spin_unlock(&inode_lock);
781 wait_on_inode(inode);
784 spin_unlock(&inode_lock);
789 * ifind_fast - internal function, you want ilookup() or iget().
790 * @sb: super block of file system to search
791 * @head: head of the list to search
792 * @ino: inode number to search for
794 * ifind_fast() searches for the inode @ino in the inode cache. This is for
795 * file systems where the inode number is sufficient for unique identification
798 * If the inode is in the cache, the inode is returned with an incremented
801 * Otherwise NULL is returned.
803 static inline struct inode *ifind_fast(struct super_block *sb,
804 struct hlist_head *head, unsigned long ino)
808 spin_lock(&inode_lock);
809 inode = find_inode_fast(sb, head, ino);
812 spin_unlock(&inode_lock);
813 wait_on_inode(inode);
816 spin_unlock(&inode_lock);
821 * ilookup5 - search for an inode in the inode cache
822 * @sb: super block of file system to search
823 * @hashval: hash value (usually inode number) to search for
824 * @test: callback used for comparisons between inodes
825 * @data: opaque data pointer to pass to @test
827 * ilookup5() uses ifind() to search for the inode specified by @hashval and
828 * @data in the inode cache. This is a generalized version of ilookup() for
829 * file systems where the inode number is not sufficient for unique
830 * identification of an inode.
832 * If the inode is in the cache, the inode is returned with an incremented
835 * Otherwise NULL is returned.
837 * Note, @test is called with the inode_lock held, so can't sleep.
839 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
840 int (*test)(struct inode *, void *), void *data)
842 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
844 return ifind(sb, head, test, data);
847 EXPORT_SYMBOL(ilookup5);
850 * ilookup - search for an inode in the inode cache
851 * @sb: super block of file system to search
852 * @ino: inode number to search for
854 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
855 * This is for file systems where the inode number is sufficient for unique
856 * identification of an inode.
858 * If the inode is in the cache, the inode is returned with an incremented
861 * Otherwise NULL is returned.
863 struct inode *ilookup(struct super_block *sb, unsigned long ino)
865 struct hlist_head *head = inode_hashtable + hash(sb, ino);
867 return ifind_fast(sb, head, ino);
870 EXPORT_SYMBOL(ilookup);
873 * iget5_locked - obtain an inode from a mounted file system
874 * @sb: super block of file system
875 * @hashval: hash value (usually inode number) to get
876 * @test: callback used for comparisons between inodes
877 * @set: callback used to initialize a new struct inode
878 * @data: opaque data pointer to pass to @test and @set
880 * This is iget() without the read_inode() portion of get_new_inode().
882 * iget5_locked() uses ifind() to search for the inode specified by @hashval
883 * and @data in the inode cache and if present it is returned with an increased
884 * reference count. This is a generalized version of iget_locked() for file
885 * systems where the inode number is not sufficient for unique identification
888 * If the inode is not in cache, get_new_inode() is called to allocate a new
889 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
890 * file system gets to fill it in before unlocking it via unlock_new_inode().
892 * Note both @test and @set are called with the inode_lock held, so can't sleep.
894 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
895 int (*test)(struct inode *, void *),
896 int (*set)(struct inode *, void *), void *data)
898 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
901 inode = ifind(sb, head, test, data);
905 * get_new_inode() will do the right thing, re-trying the search
906 * in case it had to block at any point.
908 return get_new_inode(sb, head, test, set, data);
911 EXPORT_SYMBOL(iget5_locked);
914 * iget_locked - obtain an inode from a mounted file system
915 * @sb: super block of file system
916 * @ino: inode number to get
918 * This is iget() without the read_inode() portion of get_new_inode_fast().
920 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
921 * the inode cache and if present it is returned with an increased reference
922 * count. This is for file systems where the inode number is sufficient for
923 * unique identification of an inode.
925 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
926 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
927 * The file system gets to fill it in before unlocking it via
928 * unlock_new_inode().
930 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
932 struct hlist_head *head = inode_hashtable + hash(sb, ino);
935 inode = ifind_fast(sb, head, ino);
939 * get_new_inode_fast() will do the right thing, re-trying the search
940 * in case it had to block at any point.
942 return get_new_inode_fast(sb, head, ino);
945 EXPORT_SYMBOL(iget_locked);
948 * __insert_inode_hash - hash an inode
949 * @inode: unhashed inode
950 * @hashval: unsigned long value used to locate this object in the
953 * Add an inode to the inode hash for this superblock.
955 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
957 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
958 spin_lock(&inode_lock);
959 hlist_add_head(&inode->i_hash, head);
960 spin_unlock(&inode_lock);
963 EXPORT_SYMBOL(__insert_inode_hash);
966 * remove_inode_hash - remove an inode from the hash
967 * @inode: inode to unhash
969 * Remove an inode from the superblock.
971 void remove_inode_hash(struct inode *inode)
973 spin_lock(&inode_lock);
974 hlist_del_init(&inode->i_hash);
975 spin_unlock(&inode_lock);
978 EXPORT_SYMBOL(remove_inode_hash);
981 * Tell the filesystem that this inode is no longer of any interest and should
982 * be completely destroyed.
984 * We leave the inode in the inode hash table until *after* the filesystem's
985 * ->delete_inode completes. This ensures that an iget (such as nfsd might
986 * instigate) will always find up-to-date information either in the hash or on
989 * I_FREEING is set so that no-one will take a new reference to the inode while
990 * it is being deleted.
992 void generic_delete_inode(struct inode *inode)
994 struct super_operations *op = inode->i_sb->s_op;
996 list_del_init(&inode->i_list);
997 inode->i_state|=I_FREEING;
998 inodes_stat.nr_inodes--;
999 spin_unlock(&inode_lock);
1001 if (inode->i_data.nrpages)
1002 truncate_inode_pages(&inode->i_data, 0);
1004 security_inode_delete(inode);
1006 if (op->delete_inode) {
1007 void (*delete)(struct inode *) = op->delete_inode;
1008 if (!is_bad_inode(inode))
1010 /* s_op->delete_inode internally recalls clear_inode() */
1014 spin_lock(&inode_lock);
1015 hlist_del_init(&inode->i_hash);
1016 spin_unlock(&inode_lock);
1017 wake_up_inode(inode);
1018 if (inode->i_state != I_CLEAR)
1020 destroy_inode(inode);
1023 EXPORT_SYMBOL(generic_delete_inode);
1025 static void generic_forget_inode(struct inode *inode)
1027 struct super_block *sb = inode->i_sb;
1029 if (!hlist_unhashed(&inode->i_hash)) {
1030 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1031 list_move(&inode->i_list, &inode_unused);
1032 inodes_stat.nr_unused++;
1033 spin_unlock(&inode_lock);
1034 if (!sb || (sb->s_flags & MS_ACTIVE))
1036 write_inode_now(inode, 1);
1037 spin_lock(&inode_lock);
1038 inodes_stat.nr_unused--;
1039 hlist_del_init(&inode->i_hash);
1041 list_del_init(&inode->i_list);
1042 inode->i_state|=I_FREEING;
1043 inodes_stat.nr_inodes--;
1044 spin_unlock(&inode_lock);
1045 if (inode->i_data.nrpages)
1046 truncate_inode_pages(&inode->i_data, 0);
1048 destroy_inode(inode);
1052 * Normal UNIX filesystem behaviour: delete the
1053 * inode when the usage count drops to zero, and
1056 static void generic_drop_inode(struct inode *inode)
1058 if (!inode->i_nlink)
1059 generic_delete_inode(inode);
1061 generic_forget_inode(inode);
1065 * Called when we're dropping the last reference
1068 * Call the FS "drop()" function, defaulting to
1069 * the legacy UNIX filesystem behaviour..
1071 * NOTE! NOTE! NOTE! We're called with the inode lock
1072 * held, and the drop function is supposed to release
1075 static inline void iput_final(struct inode *inode)
1077 struct super_operations *op = inode->i_sb->s_op;
1078 void (*drop)(struct inode *) = generic_drop_inode;
1080 if (op && op->drop_inode)
1081 drop = op->drop_inode;
1086 * iput - put an inode
1087 * @inode: inode to put
1089 * Puts an inode, dropping its usage count. If the inode use count hits
1090 * zero the inode is also then freed and may be destroyed.
1092 void iput(struct inode *inode)
1095 struct super_operations *op = inode->i_sb->s_op;
1097 if (inode->i_state == I_CLEAR)
1100 if (op && op->put_inode)
1101 op->put_inode(inode);
1103 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1108 EXPORT_SYMBOL(iput);
1111 * bmap - find a block number in a file
1112 * @inode: inode of file
1113 * @block: block to find
1115 * Returns the block number on the device holding the inode that
1116 * is the disk block number for the block of the file requested.
1117 * That is, asked for block 4 of inode 1 the function will return the
1118 * disk block relative to the disk start that holds that block of the
1121 sector_t bmap(struct inode * inode, sector_t block)
1124 if (inode->i_mapping->a_ops->bmap)
1125 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1129 EXPORT_SYMBOL(bmap);
1132 * Return true if the filesystem which backs this inode considers the two
1133 * passed timespecs to be sufficiently different to warrant flushing the
1134 * altered time out to disk.
1136 static int inode_times_differ(struct inode *inode,
1137 struct timespec *old, struct timespec *new)
1139 if (IS_ONE_SECOND(inode))
1140 return old->tv_sec != new->tv_sec;
1141 return !timespec_equal(old, new);
1145 * update_atime - update the access time
1146 * @inode: inode accessed
1148 * Update the accessed time on an inode and mark it for writeback.
1149 * This function automatically handles read only file systems and media,
1150 * as well as the "noatime" flag and inode specific "noatime" markers.
1152 void update_atime(struct inode *inode)
1154 struct timespec now;
1156 if (IS_NOATIME(inode))
1158 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1160 if (IS_RDONLY(inode))
1163 now = current_kernel_time();
1164 if (inode_times_differ(inode, &inode->i_atime, &now)) {
1165 inode->i_atime = now;
1166 mark_inode_dirty_sync(inode);
1168 if (!timespec_equal(&inode->i_atime, &now))
1169 inode->i_atime = now;
1173 EXPORT_SYMBOL(update_atime);
1176 * inode_update_time - update mtime and ctime time
1177 * @inode: inode accessed
1178 * @ctime_too: update ctime too
1180 * Update the mtime time on an inode and mark it for writeback.
1181 * When ctime_too is specified update the ctime too.
1184 void inode_update_time(struct inode *inode, int ctime_too)
1186 struct timespec now;
1189 if (IS_NOCMTIME(inode))
1191 if (IS_RDONLY(inode))
1194 now = current_kernel_time();
1196 if (inode_times_differ(inode, &inode->i_mtime, &now))
1198 inode->i_mtime = now;
1201 if (inode_times_differ(inode, &inode->i_ctime, &now))
1203 inode->i_ctime = now;
1206 mark_inode_dirty_sync(inode);
1209 EXPORT_SYMBOL(inode_update_time);
1211 int inode_needs_sync(struct inode *inode)
1215 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1220 EXPORT_SYMBOL(inode_needs_sync);
1223 * Quota functions that want to walk the inode lists..
1227 /* Function back in dquot.c */
1228 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1230 void remove_dquot_ref(struct super_block *sb, int type, struct list_head *tofree_head)
1232 struct inode *inode;
1233 struct list_head *act_head;
1236 return; /* nothing to do */
1237 spin_lock(&inode_lock); /* This lock is for inodes code */
1239 /* We hold dqptr_sem so we are safe against the quota code */
1240 list_for_each(act_head, &inode_in_use) {
1241 inode = list_entry(act_head, struct inode, i_list);
1242 if (inode->i_sb == sb && !IS_NOQUOTA(inode))
1243 remove_inode_dquot_ref(inode, type, tofree_head);
1245 list_for_each(act_head, &inode_unused) {
1246 inode = list_entry(act_head, struct inode, i_list);
1247 if (inode->i_sb == sb && !IS_NOQUOTA(inode))
1248 remove_inode_dquot_ref(inode, type, tofree_head);
1250 list_for_each(act_head, &sb->s_dirty) {
1251 inode = list_entry(act_head, struct inode, i_list);
1252 if (!IS_NOQUOTA(inode))
1253 remove_inode_dquot_ref(inode, type, tofree_head);
1255 list_for_each(act_head, &sb->s_io) {
1256 inode = list_entry(act_head, struct inode, i_list);
1257 if (!IS_NOQUOTA(inode))
1258 remove_inode_dquot_ref(inode, type, tofree_head);
1260 spin_unlock(&inode_lock);
1266 * Hashed waitqueues for wait_on_inode(). The table is pretty small - the
1267 * kernel doesn't lock many inodes at the same time.
1269 #define I_WAIT_TABLE_ORDER 3
1270 static struct i_wait_queue_head {
1271 wait_queue_head_t wqh;
1272 } ____cacheline_aligned_in_smp i_wait_queue_heads[1<<I_WAIT_TABLE_ORDER];
1275 * Return the address of the waitqueue_head to be used for this inode
1277 static wait_queue_head_t *i_waitq_head(struct inode *inode)
1279 return &i_wait_queue_heads[hash_ptr(inode, I_WAIT_TABLE_ORDER)].wqh;
1282 void __wait_on_inode(struct inode *inode)
1284 DECLARE_WAITQUEUE(wait, current);
1285 wait_queue_head_t *wq = i_waitq_head(inode);
1287 add_wait_queue(wq, &wait);
1289 set_current_state(TASK_UNINTERRUPTIBLE);
1290 if (inode->i_state & I_LOCK) {
1294 remove_wait_queue(wq, &wait);
1295 __set_current_state(TASK_RUNNING);
1299 * If we try to find an inode in the inode hash while it is being deleted, we
1300 * have to wait until the filesystem completes its deletion before reporting
1301 * that it isn't found. This is because iget will immediately call
1302 * ->read_inode, and we want to be sure that evidence of the deletion is found
1305 * This call might return early if an inode which shares the waitq is woken up.
1306 * This is most easily handled by the caller which will loop around again
1307 * looking for the inode.
1309 * This is called with inode_lock held.
1311 static void __wait_on_freeing_inode(struct inode *inode)
1313 DECLARE_WAITQUEUE(wait, current);
1314 wait_queue_head_t *wq = i_waitq_head(inode);
1316 add_wait_queue(wq, &wait);
1317 set_current_state(TASK_UNINTERRUPTIBLE);
1318 spin_unlock(&inode_lock);
1320 remove_wait_queue(wq, &wait);
1321 spin_lock(&inode_lock);
1324 void wake_up_inode(struct inode *inode)
1326 wait_queue_head_t *wq = i_waitq_head(inode);
1329 * Prevent speculative execution through spin_unlock(&inode_lock);
1332 if (waitqueue_active(wq))
1336 static __initdata unsigned long ihash_entries;
1337 static int __init set_ihash_entries(char *str)
1341 ihash_entries = simple_strtoul(str, &str, 0);
1344 __setup("ihash_entries=", set_ihash_entries);
1347 * Initialize the waitqueues and inode hash table.
1349 void __init inode_init_early(void)
1354 alloc_large_system_hash("Inode-cache",
1355 sizeof(struct hlist_head),
1362 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1363 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1366 void __init inode_init(unsigned long mempages)
1370 for (i = 0; i < ARRAY_SIZE(i_wait_queue_heads); i++)
1371 init_waitqueue_head(&i_wait_queue_heads[i].wqh);
1373 /* inode slab cache */
1374 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1375 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, init_once,
1377 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1380 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1382 inode->i_mode = mode;
1383 if (S_ISCHR(mode)) {
1384 inode->i_fop = &def_chr_fops;
1385 inode->i_rdev = rdev;
1386 } else if (S_ISBLK(mode)) {
1387 inode->i_fop = &def_blk_fops;
1388 inode->i_rdev = rdev;
1389 } else if (S_ISFIFO(mode))
1390 inode->i_fop = &def_fifo_fops;
1391 else if (S_ISSOCK(mode))
1392 inode->i_fop = &bad_sock_fops;
1394 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1397 EXPORT_SYMBOL(init_special_inode);