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>
25 * This is needed for the following functions:
27 * - invalidate_inode_buffers
31 * FIXME: remove all knowledge of the buffer layer from this file
33 #include <linux/buffer_head.h>
36 * New inode.c implementation.
38 * This implementation has the basic premise of trying
39 * to be extremely low-overhead and SMP-safe, yet be
40 * simple enough to be "obviously correct".
45 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
47 /* #define INODE_PARANOIA 1 */
48 /* #define INODE_DEBUG 1 */
51 * Inode lookup is no longer as critical as it used to be:
52 * most of the lookups are going to be through the dcache.
54 #define I_HASHBITS i_hash_shift
55 #define I_HASHMASK i_hash_mask
57 static unsigned int i_hash_mask;
58 static unsigned int i_hash_shift;
61 * Each inode can be on two separate lists. One is
62 * the hash list of the inode, used for lookups. The
63 * other linked list is the "type" list:
64 * "in_use" - valid inode, i_count > 0, i_nlink > 0
65 * "dirty" - as "in_use" but also dirty
66 * "unused" - valid inode, i_count = 0
68 * A "dirty" list is maintained for each super block,
69 * allowing for low-overhead inode sync() operations.
72 LIST_HEAD(inode_in_use);
73 LIST_HEAD(inode_unused);
74 static struct hlist_head *inode_hashtable;
77 * A simple spinlock to protect the list manipulations.
79 * NOTE! You also have to own the lock if you change
80 * the i_state of an inode while it is in use..
82 spinlock_t inode_lock = SPIN_LOCK_UNLOCKED;
85 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
86 * icache shrinking path, and the umount path. Without this exclusion,
87 * by the time prune_icache calls iput for the inode whose pages it has
88 * been invalidating, or by the time it calls clear_inode & destroy_inode
89 * from its final dispose_list, the struct super_block they refer to
90 * (for inode->i_sb->s_op) may already have been freed and reused.
92 static DECLARE_MUTEX(iprune_sem);
95 * Statistics gathering..
97 struct inodes_stat_t inodes_stat;
99 static kmem_cache_t * inode_cachep;
101 static void prune_icache(int nr_to_scan);
104 #define INODE_UNUSED_THRESHOLD 15000
105 #define PRUNE_BATCH_COUNT 32
107 void try_to_clip_inodes(void)
109 unsigned long count = 0;
110 /* if there are a LOT of unused inodes in cache, better shrink a few first */
112 /* check lockless first to not take the lock always here; racing occasionally isn't a big deal */
113 if (inodes_stat.nr_unused > INODE_UNUSED_THRESHOLD) {
114 spin_lock(&inode_lock);
115 if (inodes_stat.nr_unused > INODE_UNUSED_THRESHOLD)
116 count = inodes_stat.nr_unused - INODE_UNUSED_THRESHOLD;
117 spin_unlock(&inode_lock);
124 static struct inode *alloc_inode(struct super_block *sb)
126 static struct address_space_operations empty_aops;
127 static struct inode_operations empty_iops;
128 static struct file_operations empty_fops;
131 if (sb->s_op->alloc_inode)
132 inode = sb->s_op->alloc_inode(sb);
134 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
137 struct address_space * const mapping = &inode->i_data;
140 if (sb->s_flags & MS_TAGXID)
141 inode->i_xid = current->xid;
143 inode->i_xid = 0; /* maybe xid -1 would be better? */
144 // inode->i_dqh = dqhget(sb->s_dqh);
145 inode->i_blkbits = sb->s_blocksize_bits;
147 atomic_set(&inode->i_count, 1);
149 inode->i_op = &empty_iops;
150 inode->i_fop = &empty_fops;
152 atomic_set(&inode->i_writecount, 0);
156 inode->i_generation = 0;
158 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
160 inode->i_pipe = NULL;
161 inode->i_bdev = NULL;
162 inode->i_cdev = NULL;
164 // inode->i_xid = 0; /* maybe not too wise ... */
165 inode->i_security = NULL;
166 inode->dirtied_when = 0;
167 if (security_inode_alloc(inode)) {
168 if (inode->i_sb->s_op->destroy_inode)
169 inode->i_sb->s_op->destroy_inode(inode);
171 kmem_cache_free(inode_cachep, (inode));
175 mapping->a_ops = &empty_aops;
176 mapping->host = inode;
178 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
179 mapping->assoc_mapping = NULL;
180 mapping->backing_dev_info = &default_backing_dev_info;
183 * If the block_device provides a backing_dev_info for client
184 * inodes then use that. Otherwise the inode share the bdev's
188 struct backing_dev_info *bdi;
190 bdi = sb->s_bdev->bd_inode_backing_dev_info;
192 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
193 mapping->backing_dev_info = bdi;
195 memset(&inode->u, 0, sizeof(inode->u));
196 inode->i_mapping = mapping;
201 void destroy_inode(struct inode *inode)
203 if (inode_has_buffers(inode))
205 security_inode_free(inode);
206 if (inode->i_sb->s_op->destroy_inode)
207 inode->i_sb->s_op->destroy_inode(inode);
209 kmem_cache_free(inode_cachep, (inode));
214 * These are initializations that only need to be done
215 * once, because the fields are idempotent across use
216 * of the inode, so let the slab aware of that.
218 void inode_init_once(struct inode *inode)
220 memset(inode, 0, sizeof(*inode));
221 INIT_HLIST_NODE(&inode->i_hash);
222 INIT_LIST_HEAD(&inode->i_dentry);
223 INIT_LIST_HEAD(&inode->i_devices);
224 sema_init(&inode->i_sem, 1);
225 init_rwsem(&inode->i_alloc_sem);
226 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
227 spin_lock_init(&inode->i_data.tree_lock);
228 spin_lock_init(&inode->i_data.i_mmap_lock);
229 atomic_set(&inode->i_data.truncate_count, 0);
230 INIT_LIST_HEAD(&inode->i_data.private_list);
231 spin_lock_init(&inode->i_data.private_lock);
232 INIT_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
233 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
234 spin_lock_init(&inode->i_lock);
235 i_size_ordered_init(inode);
238 EXPORT_SYMBOL(inode_init_once);
240 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
242 struct inode * inode = (struct inode *) foo;
244 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
245 SLAB_CTOR_CONSTRUCTOR)
246 inode_init_once(inode);
250 * inode_lock must be held
252 void __iget(struct inode * inode)
254 if (atomic_read(&inode->i_count)) {
255 atomic_inc(&inode->i_count);
258 atomic_inc(&inode->i_count);
259 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
260 list_move(&inode->i_list, &inode_in_use);
261 inodes_stat.nr_unused--;
265 * clear_inode - clear an inode
266 * @inode: inode to clear
268 * This is called by the filesystem to tell us
269 * that the inode is no longer useful. We just
270 * terminate it with extreme prejudice.
272 void clear_inode(struct inode *inode)
274 invalidate_inode_buffers(inode);
276 if (inode->i_data.nrpages)
278 if (!(inode->i_state & I_FREEING))
280 if (inode->i_state & I_CLEAR)
282 wait_on_inode(inode);
284 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
285 inode->i_sb->s_op->clear_inode(inode);
290 inode->i_state = I_CLEAR;
293 EXPORT_SYMBOL(clear_inode);
296 * dispose_list - dispose of the contents of a local list
297 * @head: the head of the list to free
299 * Dispose-list gets a local list with local inodes in it, so it doesn't
300 * need to worry about list corruption and SMP locks.
302 static void dispose_list(struct list_head *head)
306 while (!list_empty(head)) {
309 inode = list_entry(head->next, struct inode, i_list);
310 list_del(&inode->i_list);
312 if (inode->i_data.nrpages)
313 truncate_inode_pages(&inode->i_data, 0);
315 destroy_inode(inode);
318 spin_lock(&inode_lock);
319 inodes_stat.nr_inodes -= nr_disposed;
320 spin_unlock(&inode_lock);
324 * Invalidate all inodes for a device.
326 static int invalidate_list(struct list_head *head, struct super_block * sb, struct list_head * dispose)
328 struct list_head *next;
329 int busy = 0, count = 0;
333 struct list_head * tmp = next;
334 struct inode * inode;
339 inode = list_entry(tmp, struct inode, i_list);
340 if (inode->i_sb != sb)
342 invalidate_inode_buffers(inode);
343 if (!atomic_read(&inode->i_count)) {
344 hlist_del_init(&inode->i_hash);
345 list_move(&inode->i_list, dispose);
346 inode->i_state |= I_FREEING;
352 /* only unused inodes may be cached with i_count zero */
353 inodes_stat.nr_unused -= count;
358 * This is a two-stage process. First we collect all
359 * offending inodes onto the throw-away list, and in
360 * the second stage we actually dispose of them. This
361 * is because we don't want to sleep while messing
362 * with the global lists..
366 * invalidate_inodes - discard the inodes on a device
369 * Discard all of the inodes for a given superblock. If the discard
370 * fails because there are busy inodes then a non zero value is returned.
371 * If the discard is successful all the inodes have been discarded.
373 int invalidate_inodes(struct super_block * sb)
376 LIST_HEAD(throw_away);
379 spin_lock(&inode_lock);
380 busy = invalidate_list(&inode_in_use, sb, &throw_away);
381 busy |= invalidate_list(&inode_unused, sb, &throw_away);
382 busy |= invalidate_list(&sb->s_dirty, sb, &throw_away);
383 busy |= invalidate_list(&sb->s_io, sb, &throw_away);
384 spin_unlock(&inode_lock);
386 dispose_list(&throw_away);
392 EXPORT_SYMBOL(invalidate_inodes);
394 int __invalidate_device(struct block_device *bdev, int do_sync)
396 struct super_block *sb;
403 sb = get_super(bdev);
406 * no need to lock the super, get_super holds the
407 * read semaphore so the filesystem cannot go away
408 * under us (->put_super runs with the write lock
411 shrink_dcache_sb(sb);
412 res = invalidate_inodes(sb);
415 invalidate_bdev(bdev, 0);
419 EXPORT_SYMBOL(__invalidate_device);
421 static int can_unuse(struct inode *inode)
425 if (inode_has_buffers(inode))
427 if (atomic_read(&inode->i_count))
429 if (inode->i_data.nrpages)
435 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
436 * a temporary list and then are freed outside inode_lock by dispose_list().
438 * Any inodes which are pinned purely because of attached pagecache have their
439 * pagecache removed. We expect the final iput() on that inode to add it to
440 * the front of the inode_unused list. So look for it there and if the
441 * inode is still freeable, proceed. The right inode is found 99.9% of the
442 * time in testing on a 4-way.
444 * If the inode has metadata buffers attached to mapping->private_list then
445 * try to remove them.
447 static void prune_icache(int nr_to_scan)
452 unsigned long reap = 0;
455 spin_lock(&inode_lock);
456 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
459 if (list_empty(&inode_unused))
462 inode = list_entry(inode_unused.prev, struct inode, i_list);
464 if (inode->i_state || atomic_read(&inode->i_count)) {
465 list_move(&inode->i_list, &inode_unused);
468 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
470 spin_unlock(&inode_lock);
471 if (remove_inode_buffers(inode))
472 reap += invalidate_inode_pages(&inode->i_data);
474 spin_lock(&inode_lock);
476 if (inode != list_entry(inode_unused.next,
477 struct inode, i_list))
478 continue; /* wrong inode or list_empty */
479 if (!can_unuse(inode))
482 hlist_del_init(&inode->i_hash);
483 list_move(&inode->i_list, &freeable);
484 inode->i_state |= I_FREEING;
487 inodes_stat.nr_unused -= nr_pruned;
488 spin_unlock(&inode_lock);
490 dispose_list(&freeable);
493 if (current_is_kswapd())
494 mod_page_state(kswapd_inodesteal, reap);
496 mod_page_state(pginodesteal, reap);
500 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
501 * "unused" means that no dentries are referring to the inodes: the files are
502 * not open and the dcache references to those inodes have already been
505 * This function is passed the number of inodes to scan, and it returns the
506 * total number of remaining possibly-reclaimable inodes.
508 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
512 * Nasty deadlock avoidance. We may hold various FS locks,
513 * and we don't want to recurse into the FS that called us
514 * in clear_inode() and friends..
516 if (gfp_mask & __GFP_FS)
519 return inodes_stat.nr_unused;
522 static void __wait_on_freeing_inode(struct inode *inode);
524 * Called with the inode lock held.
525 * NOTE: we are not increasing the inode-refcount, you must call __iget()
526 * by hand after calling find_inode now! This simplifies iunique and won't
527 * add any additional branch in the common code.
529 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
531 struct hlist_node *node;
532 struct inode * inode = NULL;
535 hlist_for_each (node, head) {
536 inode = hlist_entry(node, struct inode, i_hash);
537 if (inode->i_sb != sb)
539 if (!test(inode, data))
541 if (inode->i_state & (I_FREEING|I_CLEAR)) {
542 __wait_on_freeing_inode(inode);
547 return node ? inode : NULL;
551 * find_inode_fast is the fast path version of find_inode, see the comment at
552 * iget_locked for details.
554 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
556 struct hlist_node *node;
557 struct inode * inode = NULL;
560 hlist_for_each (node, head) {
561 inode = hlist_entry(node, struct inode, i_hash);
562 if (inode->i_ino != ino)
564 if (inode->i_sb != sb)
566 if (inode->i_state & (I_FREEING|I_CLEAR)) {
567 __wait_on_freeing_inode(inode);
572 return node ? inode : NULL;
576 * new_inode - obtain an inode
579 * Allocates a new inode for given superblock.
581 struct inode *new_inode(struct super_block *sb)
583 static unsigned long last_ino;
584 struct inode * inode;
586 spin_lock_prefetch(&inode_lock);
588 inode = alloc_inode(sb);
590 spin_lock(&inode_lock);
591 inodes_stat.nr_inodes++;
592 list_add(&inode->i_list, &inode_in_use);
593 inode->i_ino = ++last_ino;
595 spin_unlock(&inode_lock);
600 EXPORT_SYMBOL(new_inode);
602 void unlock_new_inode(struct inode *inode)
605 * This is special! We do not need the spinlock
606 * when clearing I_LOCK, because we're guaranteed
607 * that nobody else tries to do anything about the
608 * state of the inode when it is locked, as we
609 * just created it (so there can be no old holders
610 * that haven't tested I_LOCK).
612 inode->i_state &= ~(I_LOCK|I_NEW);
613 wake_up_inode(inode);
616 EXPORT_SYMBOL(unlock_new_inode);
619 * This is called without the inode lock held.. Be careful.
621 * We no longer cache the sb_flags in i_flags - see fs.h
622 * -- rmk@arm.uk.linux.org
624 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)
626 struct inode * inode;
628 inode = alloc_inode(sb);
632 spin_lock(&inode_lock);
633 /* We released the lock, so.. */
634 old = find_inode(sb, head, test, data);
636 if (set(inode, data))
639 inodes_stat.nr_inodes++;
640 list_add(&inode->i_list, &inode_in_use);
641 hlist_add_head(&inode->i_hash, head);
642 inode->i_state = I_LOCK|I_NEW;
643 spin_unlock(&inode_lock);
645 /* Return the locked inode with I_NEW set, the
646 * caller is responsible for filling in the contents
652 * Uhhuh, somebody else created the same inode under
653 * us. Use the old inode instead of the one we just
657 spin_unlock(&inode_lock);
658 destroy_inode(inode);
660 wait_on_inode(inode);
665 spin_unlock(&inode_lock);
666 destroy_inode(inode);
671 * get_new_inode_fast is the fast path version of get_new_inode, see the
672 * comment at iget_locked for details.
674 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
676 struct inode * inode;
678 inode = alloc_inode(sb);
682 spin_lock(&inode_lock);
683 /* We released the lock, so.. */
684 old = find_inode_fast(sb, head, ino);
687 inodes_stat.nr_inodes++;
688 list_add(&inode->i_list, &inode_in_use);
689 hlist_add_head(&inode->i_hash, head);
690 inode->i_state = I_LOCK|I_NEW;
691 spin_unlock(&inode_lock);
693 /* Return the locked inode with I_NEW set, the
694 * caller is responsible for filling in the contents
700 * Uhhuh, somebody else created the same inode under
701 * us. Use the old inode instead of the one we just
705 spin_unlock(&inode_lock);
706 destroy_inode(inode);
708 wait_on_inode(inode);
713 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
717 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
719 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
720 return tmp & I_HASHMASK;
724 * iunique - get a unique inode number
726 * @max_reserved: highest reserved inode number
728 * Obtain an inode number that is unique on the system for a given
729 * superblock. This is used by file systems that have no natural
730 * permanent inode numbering system. An inode number is returned that
731 * is higher than the reserved limit but unique.
734 * With a large number of inodes live on the file system this function
735 * currently becomes quite slow.
737 ino_t iunique(struct super_block *sb, ino_t max_reserved)
739 static ino_t counter;
741 struct hlist_head * head;
743 spin_lock(&inode_lock);
745 if (counter > max_reserved) {
746 head = inode_hashtable + hash(sb,counter);
748 inode = find_inode_fast(sb, head, res);
750 spin_unlock(&inode_lock);
754 counter = max_reserved + 1;
760 EXPORT_SYMBOL(iunique);
762 struct inode *igrab(struct inode *inode)
764 spin_lock(&inode_lock);
765 if (!(inode->i_state & I_FREEING))
769 * Handle the case where s_op->clear_inode is not been
770 * called yet, and somebody is calling igrab
771 * while the inode is getting freed.
774 spin_unlock(&inode_lock);
778 EXPORT_SYMBOL(igrab);
781 * ifind - internal function, you want ilookup5() or iget5().
782 * @sb: super block of file system to search
783 * @head: the head of the list to search
784 * @test: callback used for comparisons between inodes
785 * @data: opaque data pointer to pass to @test
787 * ifind() searches for the inode specified by @data in the inode
788 * cache. This is a generalized version of ifind_fast() for file systems where
789 * the inode number is not sufficient for unique identification of an inode.
791 * If the inode is in the cache, the inode is returned with an incremented
794 * Otherwise NULL is returned.
796 * Note, @test is called with the inode_lock held, so can't sleep.
798 static inline struct inode *ifind(struct super_block *sb,
799 struct hlist_head *head, int (*test)(struct inode *, void *),
804 spin_lock(&inode_lock);
805 inode = find_inode(sb, head, test, data);
808 spin_unlock(&inode_lock);
809 wait_on_inode(inode);
812 spin_unlock(&inode_lock);
817 * ifind_fast - internal function, you want ilookup() or iget().
818 * @sb: super block of file system to search
819 * @head: head of the list to search
820 * @ino: inode number to search for
822 * ifind_fast() searches for the inode @ino in the inode cache. This is for
823 * file systems where the inode number is sufficient for unique identification
826 * If the inode is in the cache, the inode is returned with an incremented
829 * Otherwise NULL is returned.
831 static inline struct inode *ifind_fast(struct super_block *sb,
832 struct hlist_head *head, unsigned long ino)
836 spin_lock(&inode_lock);
837 inode = find_inode_fast(sb, head, ino);
840 spin_unlock(&inode_lock);
841 wait_on_inode(inode);
844 spin_unlock(&inode_lock);
849 * ilookup5 - search for an inode in the inode cache
850 * @sb: super block of file system to search
851 * @hashval: hash value (usually inode number) to search for
852 * @test: callback used for comparisons between inodes
853 * @data: opaque data pointer to pass to @test
855 * ilookup5() uses ifind() to search for the inode specified by @hashval and
856 * @data in the inode cache. This is a generalized version of ilookup() for
857 * file systems where the inode number is not sufficient for unique
858 * identification of an inode.
860 * If the inode is in the cache, the inode is returned with an incremented
863 * Otherwise NULL is returned.
865 * Note, @test is called with the inode_lock held, so can't sleep.
867 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
868 int (*test)(struct inode *, void *), void *data)
870 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
872 return ifind(sb, head, test, data);
875 EXPORT_SYMBOL(ilookup5);
878 * ilookup - search for an inode in the inode cache
879 * @sb: super block of file system to search
880 * @ino: inode number to search for
882 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
883 * This is for file systems where the inode number is sufficient for unique
884 * identification of an inode.
886 * If the inode is in the cache, the inode is returned with an incremented
889 * Otherwise NULL is returned.
891 struct inode *ilookup(struct super_block *sb, unsigned long ino)
893 struct hlist_head *head = inode_hashtable + hash(sb, ino);
895 return ifind_fast(sb, head, ino);
898 EXPORT_SYMBOL(ilookup);
901 * iget5_locked - obtain an inode from a mounted file system
902 * @sb: super block of file system
903 * @hashval: hash value (usually inode number) to get
904 * @test: callback used for comparisons between inodes
905 * @set: callback used to initialize a new struct inode
906 * @data: opaque data pointer to pass to @test and @set
908 * This is iget() without the read_inode() portion of get_new_inode().
910 * iget5_locked() uses ifind() to search for the inode specified by @hashval
911 * and @data in the inode cache and if present it is returned with an increased
912 * reference count. This is a generalized version of iget_locked() for file
913 * systems where the inode number is not sufficient for unique identification
916 * If the inode is not in cache, get_new_inode() is called to allocate a new
917 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
918 * file system gets to fill it in before unlocking it via unlock_new_inode().
920 * Note both @test and @set are called with the inode_lock held, so can't sleep.
922 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
923 int (*test)(struct inode *, void *),
924 int (*set)(struct inode *, void *), void *data)
926 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
929 inode = ifind(sb, head, test, data);
933 * get_new_inode() will do the right thing, re-trying the search
934 * in case it had to block at any point.
936 return get_new_inode(sb, head, test, set, data);
939 EXPORT_SYMBOL(iget5_locked);
942 * iget_locked - obtain an inode from a mounted file system
943 * @sb: super block of file system
944 * @ino: inode number to get
946 * This is iget() without the read_inode() portion of get_new_inode_fast().
948 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
949 * the inode cache and if present it is returned with an increased reference
950 * count. This is for file systems where the inode number is sufficient for
951 * unique identification of an inode.
953 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
954 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
955 * The file system gets to fill it in before unlocking it via
956 * unlock_new_inode().
958 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
960 struct hlist_head *head = inode_hashtable + hash(sb, ino);
963 inode = ifind_fast(sb, head, ino);
967 * get_new_inode_fast() will do the right thing, re-trying the search
968 * in case it had to block at any point.
970 return get_new_inode_fast(sb, head, ino);
973 EXPORT_SYMBOL(iget_locked);
976 * __insert_inode_hash - hash an inode
977 * @inode: unhashed inode
978 * @hashval: unsigned long value used to locate this object in the
981 * Add an inode to the inode hash for this superblock.
983 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
985 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
986 spin_lock(&inode_lock);
987 hlist_add_head(&inode->i_hash, head);
988 spin_unlock(&inode_lock);
991 EXPORT_SYMBOL(__insert_inode_hash);
994 * remove_inode_hash - remove an inode from the hash
995 * @inode: inode to unhash
997 * Remove an inode from the superblock.
999 void remove_inode_hash(struct inode *inode)
1001 spin_lock(&inode_lock);
1002 hlist_del_init(&inode->i_hash);
1003 spin_unlock(&inode_lock);
1006 EXPORT_SYMBOL(remove_inode_hash);
1009 * Tell the filesystem that this inode is no longer of any interest and should
1010 * be completely destroyed.
1012 * We leave the inode in the inode hash table until *after* the filesystem's
1013 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1014 * instigate) will always find up-to-date information either in the hash or on
1017 * I_FREEING is set so that no-one will take a new reference to the inode while
1018 * it is being deleted.
1020 void generic_delete_inode(struct inode *inode)
1022 struct super_operations *op = inode->i_sb->s_op;
1024 list_del_init(&inode->i_list);
1025 inode->i_state|=I_FREEING;
1026 inodes_stat.nr_inodes--;
1027 spin_unlock(&inode_lock);
1029 if (inode->i_data.nrpages)
1030 truncate_inode_pages(&inode->i_data, 0);
1032 security_inode_delete(inode);
1034 if (op->delete_inode) {
1035 void (*delete)(struct inode *) = op->delete_inode;
1036 if (!is_bad_inode(inode))
1038 /* s_op->delete_inode internally recalls clear_inode() */
1042 spin_lock(&inode_lock);
1043 hlist_del_init(&inode->i_hash);
1044 spin_unlock(&inode_lock);
1045 wake_up_inode(inode);
1046 if (inode->i_state != I_CLEAR)
1048 destroy_inode(inode);
1051 EXPORT_SYMBOL(generic_delete_inode);
1053 static void generic_forget_inode(struct inode *inode)
1055 struct super_block *sb = inode->i_sb;
1057 if (!hlist_unhashed(&inode->i_hash)) {
1058 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1059 list_move(&inode->i_list, &inode_unused);
1060 inodes_stat.nr_unused++;
1061 spin_unlock(&inode_lock);
1062 if (!sb || (sb->s_flags & MS_ACTIVE))
1064 write_inode_now(inode, 1);
1065 spin_lock(&inode_lock);
1066 inodes_stat.nr_unused--;
1067 hlist_del_init(&inode->i_hash);
1069 list_del_init(&inode->i_list);
1070 inode->i_state|=I_FREEING;
1071 inodes_stat.nr_inodes--;
1072 spin_unlock(&inode_lock);
1073 if (inode->i_data.nrpages)
1074 truncate_inode_pages(&inode->i_data, 0);
1076 destroy_inode(inode);
1080 * Normal UNIX filesystem behaviour: delete the
1081 * inode when the usage count drops to zero, and
1084 static void generic_drop_inode(struct inode *inode)
1086 if (!inode->i_nlink)
1087 generic_delete_inode(inode);
1089 generic_forget_inode(inode);
1093 * Called when we're dropping the last reference
1096 * Call the FS "drop()" function, defaulting to
1097 * the legacy UNIX filesystem behaviour..
1099 * NOTE! NOTE! NOTE! We're called with the inode lock
1100 * held, and the drop function is supposed to release
1103 static inline void iput_final(struct inode *inode)
1105 struct super_operations *op = inode->i_sb->s_op;
1106 void (*drop)(struct inode *) = generic_drop_inode;
1108 if (op && op->drop_inode)
1109 drop = op->drop_inode;
1114 * iput - put an inode
1115 * @inode: inode to put
1117 * Puts an inode, dropping its usage count. If the inode use count hits
1118 * zero the inode is also then freed and may be destroyed.
1120 void iput(struct inode *inode)
1123 struct super_operations *op = inode->i_sb->s_op;
1125 if (inode->i_state == I_CLEAR)
1128 if (op && op->put_inode)
1129 op->put_inode(inode);
1131 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1136 EXPORT_SYMBOL(iput);
1139 * bmap - find a block number in a file
1140 * @inode: inode of file
1141 * @block: block to find
1143 * Returns the block number on the device holding the inode that
1144 * is the disk block number for the block of the file requested.
1145 * That is, asked for block 4 of inode 1 the function will return the
1146 * disk block relative to the disk start that holds that block of the
1149 sector_t bmap(struct inode * inode, sector_t block)
1152 if (inode->i_mapping->a_ops->bmap)
1153 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1157 EXPORT_SYMBOL(bmap);
1160 * Return true if the filesystem which backs this inode considers the two
1161 * passed timespecs to be sufficiently different to warrant flushing the
1162 * altered time out to disk.
1164 static int inode_times_differ(struct inode *inode,
1165 struct timespec *old, struct timespec *new)
1167 if (IS_ONE_SECOND(inode))
1168 return old->tv_sec != new->tv_sec;
1169 return !timespec_equal(old, new);
1173 * update_atime - update the access time
1174 * @inode: inode accessed
1176 * Update the accessed time on an inode and mark it for writeback.
1177 * This function automatically handles read only file systems and media,
1178 * as well as the "noatime" flag and inode specific "noatime" markers.
1180 void update_atime(struct inode *inode)
1182 struct timespec now;
1184 if (IS_NOATIME(inode))
1186 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1188 if (IS_RDONLY(inode))
1191 now = current_kernel_time();
1192 if (inode_times_differ(inode, &inode->i_atime, &now)) {
1193 inode->i_atime = now;
1194 mark_inode_dirty_sync(inode);
1196 if (!timespec_equal(&inode->i_atime, &now))
1197 inode->i_atime = now;
1201 EXPORT_SYMBOL(update_atime);
1204 * inode_update_time - update mtime and ctime time
1205 * @inode: inode accessed
1206 * @ctime_too: update ctime too
1208 * Update the mtime time on an inode and mark it for writeback.
1209 * When ctime_too is specified update the ctime too.
1212 void inode_update_time(struct inode *inode, int ctime_too)
1214 struct timespec now;
1217 if (IS_NOCMTIME(inode))
1219 if (IS_RDONLY(inode))
1222 now = current_kernel_time();
1224 if (inode_times_differ(inode, &inode->i_mtime, &now))
1226 inode->i_mtime = now;
1229 if (inode_times_differ(inode, &inode->i_ctime, &now))
1231 inode->i_ctime = now;
1234 mark_inode_dirty_sync(inode);
1237 EXPORT_SYMBOL(inode_update_time);
1239 int inode_needs_sync(struct inode *inode)
1243 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1248 EXPORT_SYMBOL(inode_needs_sync);
1251 * Quota functions that want to walk the inode lists..
1255 /* Function back in dquot.c */
1256 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1258 void remove_dquot_ref(struct super_block *sb, int type, struct list_head *tofree_head)
1260 struct inode *inode;
1261 struct list_head *act_head;
1264 return; /* nothing to do */
1265 spin_lock(&inode_lock); /* This lock is for inodes code */
1266 /* We don't have to lock against quota code - test IS_QUOTAINIT is just for speedup... */
1268 list_for_each(act_head, &inode_in_use) {
1269 inode = list_entry(act_head, struct inode, i_list);
1270 if (inode->i_sb == sb && IS_QUOTAINIT(inode))
1271 remove_inode_dquot_ref(inode, type, tofree_head);
1273 list_for_each(act_head, &inode_unused) {
1274 inode = list_entry(act_head, struct inode, i_list);
1275 if (inode->i_sb == sb && IS_QUOTAINIT(inode))
1276 remove_inode_dquot_ref(inode, type, tofree_head);
1278 list_for_each(act_head, &sb->s_dirty) {
1279 inode = list_entry(act_head, struct inode, i_list);
1280 if (IS_QUOTAINIT(inode))
1281 remove_inode_dquot_ref(inode, type, tofree_head);
1283 list_for_each(act_head, &sb->s_io) {
1284 inode = list_entry(act_head, struct inode, i_list);
1285 if (IS_QUOTAINIT(inode))
1286 remove_inode_dquot_ref(inode, type, tofree_head);
1288 spin_unlock(&inode_lock);
1294 * Hashed waitqueues for wait_on_inode(). The table is pretty small - the
1295 * kernel doesn't lock many inodes at the same time.
1297 #define I_WAIT_TABLE_ORDER 3
1298 static struct i_wait_queue_head {
1299 wait_queue_head_t wqh;
1300 } ____cacheline_aligned_in_smp i_wait_queue_heads[1<<I_WAIT_TABLE_ORDER];
1303 * Return the address of the waitqueue_head to be used for this inode
1305 static wait_queue_head_t *i_waitq_head(struct inode *inode)
1307 return &i_wait_queue_heads[hash_ptr(inode, I_WAIT_TABLE_ORDER)].wqh;
1310 void __wait_on_inode(struct inode *inode)
1312 DECLARE_WAITQUEUE(wait, current);
1313 wait_queue_head_t *wq = i_waitq_head(inode);
1315 add_wait_queue(wq, &wait);
1317 set_current_state(TASK_UNINTERRUPTIBLE);
1318 if (inode->i_state & I_LOCK) {
1322 remove_wait_queue(wq, &wait);
1323 __set_current_state(TASK_RUNNING);
1327 * If we try to find an inode in the inode hash while it is being deleted, we
1328 * have to wait until the filesystem completes its deletion before reporting
1329 * that it isn't found. This is because iget will immediately call
1330 * ->read_inode, and we want to be sure that evidence of the deletion is found
1333 * This call might return early if an inode which shares the waitq is woken up.
1334 * This is most easily handled by the caller which will loop around again
1335 * looking for the inode.
1337 * This is called with inode_lock held.
1339 static void __wait_on_freeing_inode(struct inode *inode)
1341 DECLARE_WAITQUEUE(wait, current);
1342 wait_queue_head_t *wq = i_waitq_head(inode);
1344 add_wait_queue(wq, &wait);
1345 set_current_state(TASK_UNINTERRUPTIBLE);
1346 spin_unlock(&inode_lock);
1348 remove_wait_queue(wq, &wait);
1349 spin_lock(&inode_lock);
1352 void wake_up_inode(struct inode *inode)
1354 wait_queue_head_t *wq = i_waitq_head(inode);
1357 * Prevent speculative execution through spin_unlock(&inode_lock);
1360 if (waitqueue_active(wq))
1364 static __initdata unsigned long ihash_entries;
1365 static int __init set_ihash_entries(char *str)
1369 ihash_entries = simple_strtoul(str, &str, 0);
1372 __setup("ihash_entries=", set_ihash_entries);
1375 * Initialize the waitqueues and inode hash table.
1377 void __init inode_init(unsigned long mempages)
1379 struct hlist_head *head;
1380 unsigned long order;
1381 unsigned int nr_hash;
1384 for (i = 0; i < ARRAY_SIZE(i_wait_queue_heads); i++)
1385 init_waitqueue_head(&i_wait_queue_heads[i].wqh);
1388 ihash_entries = PAGE_SHIFT < 14 ?
1389 mempages >> (14 - PAGE_SHIFT) :
1390 mempages << (PAGE_SHIFT - 14);
1392 ihash_entries *= sizeof(struct hlist_head);
1393 for (order = 0; ((1UL << order) << PAGE_SHIFT) < ihash_entries; order++)
1402 nr_hash = (1UL << order) * PAGE_SIZE /
1403 sizeof(struct hlist_head);
1404 i_hash_mask = (nr_hash - 1);
1408 while ((tmp >>= 1UL) != 0UL)
1411 inode_hashtable = (struct hlist_head *)
1412 __get_free_pages(GFP_ATOMIC, order);
1413 } while (inode_hashtable == NULL && --order >= 0);
1415 printk("Inode-cache hash table entries: %d (order: %ld, %ld bytes)\n",
1416 nr_hash, order, (PAGE_SIZE << order));
1418 if (!inode_hashtable)
1419 panic("Failed to allocate inode hash table\n");
1421 head = inode_hashtable;
1424 INIT_HLIST_HEAD(head);
1429 /* inode slab cache */
1430 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1431 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, init_once,
1433 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1436 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1438 inode->i_mode = mode;
1439 if (S_ISCHR(mode)) {
1440 inode->i_fop = &def_chr_fops;
1441 inode->i_rdev = rdev;
1442 } else if (S_ISBLK(mode)) {
1443 inode->i_fop = &def_blk_fops;
1444 inode->i_rdev = rdev;
1445 } else if (S_ISFIFO(mode))
1446 inode->i_fop = &def_fifo_fops;
1447 else if (S_ISSOCK(mode))
1448 inode->i_fop = &bad_sock_fops;
1450 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1453 EXPORT_SYMBOL(init_special_inode);