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/vs_base.h>
27 * This is needed for the following functions:
29 * - 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 spinlock_t inode_lock = SPIN_LOCK_UNLOCKED;
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;
119 // inode->i_dqh = dqhget(sb->s_dqh);
121 /* important because of inode slab reuse */
123 inode->i_blkbits = sb->s_blocksize_bits;
125 atomic_set(&inode->i_count, 1);
127 inode->i_op = &empty_iops;
128 inode->i_fop = &empty_fops;
130 atomic_set(&inode->i_writecount, 0);
134 inode->i_generation = 0;
136 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
138 inode->i_pipe = NULL;
139 inode->i_bdev = NULL;
140 inode->i_cdev = NULL;
142 inode->i_security = NULL;
143 inode->dirtied_when = 0;
144 if (security_inode_alloc(inode)) {
145 if (inode->i_sb->s_op->destroy_inode)
146 inode->i_sb->s_op->destroy_inode(inode);
148 kmem_cache_free(inode_cachep, (inode));
152 mapping->a_ops = &empty_aops;
153 mapping->host = inode;
155 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
156 mapping->assoc_mapping = NULL;
157 mapping->backing_dev_info = &default_backing_dev_info;
160 * If the block_device provides a backing_dev_info for client
161 * inodes then use that. Otherwise the inode share the bdev's
165 struct backing_dev_info *bdi;
167 bdi = sb->s_bdev->bd_inode_backing_dev_info;
169 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
170 mapping->backing_dev_info = bdi;
172 memset(&inode->u, 0, sizeof(inode->u));
173 inode->i_mapping = mapping;
178 void destroy_inode(struct inode *inode)
180 if (inode_has_buffers(inode))
182 security_inode_free(inode);
183 if (inode->i_sb->s_op->destroy_inode)
184 inode->i_sb->s_op->destroy_inode(inode);
186 kmem_cache_free(inode_cachep, (inode));
191 * These are initializations that only need to be done
192 * once, because the fields are idempotent across use
193 * of the inode, so let the slab aware of that.
195 void inode_init_once(struct inode *inode)
197 memset(inode, 0, sizeof(*inode));
198 INIT_HLIST_NODE(&inode->i_hash);
199 INIT_LIST_HEAD(&inode->i_dentry);
200 INIT_LIST_HEAD(&inode->i_devices);
201 sema_init(&inode->i_sem, 1);
202 init_rwsem(&inode->i_alloc_sem);
203 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
204 spin_lock_init(&inode->i_data.tree_lock);
205 spin_lock_init(&inode->i_data.i_mmap_lock);
206 atomic_set(&inode->i_data.truncate_count, 0);
207 INIT_LIST_HEAD(&inode->i_data.private_list);
208 spin_lock_init(&inode->i_data.private_lock);
209 INIT_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
210 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
211 spin_lock_init(&inode->i_lock);
212 i_size_ordered_init(inode);
215 EXPORT_SYMBOL(inode_init_once);
217 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
219 struct inode * inode = (struct inode *) foo;
221 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
222 SLAB_CTOR_CONSTRUCTOR)
223 inode_init_once(inode);
227 * inode_lock must be held
229 void __iget(struct inode * inode)
231 if (atomic_read(&inode->i_count)) {
232 atomic_inc(&inode->i_count);
235 atomic_inc(&inode->i_count);
236 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
237 list_move(&inode->i_list, &inode_in_use);
238 inodes_stat.nr_unused--;
242 * clear_inode - clear an inode
243 * @inode: inode to clear
245 * This is called by the filesystem to tell us
246 * that the inode is no longer useful. We just
247 * terminate it with extreme prejudice.
249 void clear_inode(struct inode *inode)
251 invalidate_inode_buffers(inode);
253 if (inode->i_data.nrpages)
255 if (!(inode->i_state & I_FREEING))
257 if (inode->i_state & I_CLEAR)
259 wait_on_inode(inode);
261 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
262 inode->i_sb->s_op->clear_inode(inode);
267 inode->i_state = I_CLEAR;
270 EXPORT_SYMBOL(clear_inode);
273 * dispose_list - dispose of the contents of a local list
274 * @head: the head of the list to free
276 * Dispose-list gets a local list with local inodes in it, so it doesn't
277 * need to worry about list corruption and SMP locks.
279 static void dispose_list(struct list_head *head)
283 while (!list_empty(head)) {
286 inode = list_entry(head->next, struct inode, i_list);
287 list_del(&inode->i_list);
289 if (inode->i_data.nrpages)
290 truncate_inode_pages(&inode->i_data, 0);
292 destroy_inode(inode);
295 spin_lock(&inode_lock);
296 inodes_stat.nr_inodes -= nr_disposed;
297 spin_unlock(&inode_lock);
301 * Invalidate all inodes for a device.
303 static int invalidate_list(struct list_head *head, struct super_block * sb, struct list_head * dispose)
305 struct list_head *next;
306 int busy = 0, count = 0;
310 struct list_head * tmp = next;
311 struct inode * inode;
316 inode = list_entry(tmp, struct inode, i_list);
317 if (inode->i_sb != sb)
319 invalidate_inode_buffers(inode);
320 if (!atomic_read(&inode->i_count)) {
321 hlist_del_init(&inode->i_hash);
322 list_move(&inode->i_list, dispose);
323 inode->i_state |= I_FREEING;
329 /* only unused inodes may be cached with i_count zero */
330 inodes_stat.nr_unused -= count;
335 * This is a two-stage process. First we collect all
336 * offending inodes onto the throw-away list, and in
337 * the second stage we actually dispose of them. This
338 * is because we don't want to sleep while messing
339 * with the global lists..
343 * invalidate_inodes - discard the inodes on a device
346 * Discard all of the inodes for a given superblock. If the discard
347 * fails because there are busy inodes then a non zero value is returned.
348 * If the discard is successful all the inodes have been discarded.
350 int invalidate_inodes(struct super_block * sb)
353 LIST_HEAD(throw_away);
356 spin_lock(&inode_lock);
357 busy = invalidate_list(&inode_in_use, sb, &throw_away);
358 busy |= invalidate_list(&inode_unused, sb, &throw_away);
359 busy |= invalidate_list(&sb->s_dirty, sb, &throw_away);
360 busy |= invalidate_list(&sb->s_io, sb, &throw_away);
361 spin_unlock(&inode_lock);
363 dispose_list(&throw_away);
369 EXPORT_SYMBOL(invalidate_inodes);
371 int __invalidate_device(struct block_device *bdev, int do_sync)
373 struct super_block *sb;
380 sb = get_super(bdev);
383 * no need to lock the super, get_super holds the
384 * read semaphore so the filesystem cannot go away
385 * under us (->put_super runs with the write lock
388 shrink_dcache_sb(sb);
389 res = invalidate_inodes(sb);
392 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 hlist_del_init(&inode->i_hash);
460 list_move(&inode->i_list, &freeable);
461 inode->i_state |= I_FREEING;
464 inodes_stat.nr_unused -= nr_pruned;
465 spin_unlock(&inode_lock);
467 dispose_list(&freeable);
470 if (current_is_kswapd())
471 mod_page_state(kswapd_inodesteal, reap);
473 mod_page_state(pginodesteal, reap);
477 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
478 * "unused" means that no dentries are referring to the inodes: the files are
479 * not open and the dcache references to those inodes have already been
482 * This function is passed the number of inodes to scan, and it returns the
483 * total number of remaining possibly-reclaimable inodes.
485 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
489 * Nasty deadlock avoidance. We may hold various FS locks,
490 * and we don't want to recurse into the FS that called us
491 * in clear_inode() and friends..
493 if (gfp_mask & __GFP_FS)
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)) {
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)) {
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 inode->i_ino = ++last_ino;
572 spin_unlock(&inode_lock);
577 EXPORT_SYMBOL(new_inode);
579 void unlock_new_inode(struct inode *inode)
582 * This is special! We do not need the spinlock
583 * when clearing I_LOCK, because we're guaranteed
584 * that nobody else tries to do anything about the
585 * state of the inode when it is locked, as we
586 * just created it (so there can be no old holders
587 * that haven't tested I_LOCK).
589 inode->i_state &= ~(I_LOCK|I_NEW);
590 wake_up_inode(inode);
593 EXPORT_SYMBOL(unlock_new_inode);
596 * This is called without the inode lock held.. Be careful.
598 * We no longer cache the sb_flags in i_flags - see fs.h
599 * -- rmk@arm.uk.linux.org
601 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)
603 struct inode * inode;
605 inode = alloc_inode(sb);
609 spin_lock(&inode_lock);
610 /* We released the lock, so.. */
611 old = find_inode(sb, head, test, data);
613 if (set(inode, data))
616 inodes_stat.nr_inodes++;
617 list_add(&inode->i_list, &inode_in_use);
618 hlist_add_head(&inode->i_hash, head);
619 inode->i_state = I_LOCK|I_NEW;
620 spin_unlock(&inode_lock);
622 /* Return the locked inode with I_NEW set, the
623 * caller is responsible for filling in the contents
629 * Uhhuh, somebody else created the same inode under
630 * us. Use the old inode instead of the one we just
634 spin_unlock(&inode_lock);
635 destroy_inode(inode);
637 wait_on_inode(inode);
642 spin_unlock(&inode_lock);
643 destroy_inode(inode);
648 * get_new_inode_fast is the fast path version of get_new_inode, see the
649 * comment at iget_locked for details.
651 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
653 struct inode * inode;
655 inode = alloc_inode(sb);
659 spin_lock(&inode_lock);
660 /* We released the lock, so.. */
661 old = find_inode_fast(sb, head, ino);
664 inodes_stat.nr_inodes++;
665 list_add(&inode->i_list, &inode_in_use);
666 hlist_add_head(&inode->i_hash, head);
667 inode->i_state = I_LOCK|I_NEW;
668 spin_unlock(&inode_lock);
670 /* Return the locked inode with I_NEW set, the
671 * caller is responsible for filling in the contents
677 * Uhhuh, somebody else created the same inode under
678 * us. Use the old inode instead of the one we just
682 spin_unlock(&inode_lock);
683 destroy_inode(inode);
685 wait_on_inode(inode);
690 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
694 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
696 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
697 return tmp & I_HASHMASK;
701 * iunique - get a unique inode number
703 * @max_reserved: highest reserved inode number
705 * Obtain an inode number that is unique on the system for a given
706 * superblock. This is used by file systems that have no natural
707 * permanent inode numbering system. An inode number is returned that
708 * is higher than the reserved limit but unique.
711 * With a large number of inodes live on the file system this function
712 * currently becomes quite slow.
714 ino_t iunique(struct super_block *sb, ino_t max_reserved)
716 static ino_t counter;
718 struct hlist_head * head;
720 spin_lock(&inode_lock);
722 if (counter > max_reserved) {
723 head = inode_hashtable + hash(sb,counter);
725 inode = find_inode_fast(sb, head, res);
727 spin_unlock(&inode_lock);
731 counter = max_reserved + 1;
737 EXPORT_SYMBOL(iunique);
739 struct inode *igrab(struct inode *inode)
741 spin_lock(&inode_lock);
742 if (!(inode->i_state & I_FREEING))
746 * Handle the case where s_op->clear_inode is not been
747 * called yet, and somebody is calling igrab
748 * while the inode is getting freed.
751 spin_unlock(&inode_lock);
755 EXPORT_SYMBOL(igrab);
758 * ifind - internal function, you want ilookup5() or iget5().
759 * @sb: super block of file system to search
760 * @head: the head of the list to search
761 * @test: callback used for comparisons between inodes
762 * @data: opaque data pointer to pass to @test
764 * ifind() searches for the inode specified by @data in the inode
765 * cache. This is a generalized version of ifind_fast() for file systems where
766 * the inode number is not sufficient for unique identification of an inode.
768 * If the inode is in the cache, the inode is returned with an incremented
771 * Otherwise NULL is returned.
773 * Note, @test is called with the inode_lock held, so can't sleep.
775 static inline struct inode *ifind(struct super_block *sb,
776 struct hlist_head *head, int (*test)(struct inode *, void *),
781 spin_lock(&inode_lock);
782 inode = find_inode(sb, head, test, data);
785 spin_unlock(&inode_lock);
786 wait_on_inode(inode);
789 spin_unlock(&inode_lock);
794 * ifind_fast - internal function, you want ilookup() or iget().
795 * @sb: super block of file system to search
796 * @head: head of the list to search
797 * @ino: inode number to search for
799 * ifind_fast() searches for the inode @ino in the inode cache. This is for
800 * file systems where the inode number is sufficient for unique identification
803 * If the inode is in the cache, the inode is returned with an incremented
806 * Otherwise NULL is returned.
808 static inline struct inode *ifind_fast(struct super_block *sb,
809 struct hlist_head *head, unsigned long ino)
813 spin_lock(&inode_lock);
814 inode = find_inode_fast(sb, head, ino);
817 spin_unlock(&inode_lock);
818 wait_on_inode(inode);
821 spin_unlock(&inode_lock);
826 * ilookup5 - search for an inode in the inode cache
827 * @sb: super block of file system to search
828 * @hashval: hash value (usually inode number) to search for
829 * @test: callback used for comparisons between inodes
830 * @data: opaque data pointer to pass to @test
832 * ilookup5() uses ifind() to search for the inode specified by @hashval and
833 * @data in the inode cache. This is a generalized version of ilookup() for
834 * file systems where the inode number is not sufficient for unique
835 * identification of an inode.
837 * If the inode is in the cache, the inode is returned with an incremented
840 * Otherwise NULL is returned.
842 * Note, @test is called with the inode_lock held, so can't sleep.
844 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
845 int (*test)(struct inode *, void *), void *data)
847 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
849 return ifind(sb, head, test, data);
852 EXPORT_SYMBOL(ilookup5);
855 * ilookup - search for an inode in the inode cache
856 * @sb: super block of file system to search
857 * @ino: inode number to search for
859 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
860 * This is for file systems where the inode number is sufficient for unique
861 * identification of an inode.
863 * If the inode is in the cache, the inode is returned with an incremented
866 * Otherwise NULL is returned.
868 struct inode *ilookup(struct super_block *sb, unsigned long ino)
870 struct hlist_head *head = inode_hashtable + hash(sb, ino);
872 return ifind_fast(sb, head, ino);
875 EXPORT_SYMBOL(ilookup);
878 * iget5_locked - obtain an inode from a mounted file system
879 * @sb: super block of file system
880 * @hashval: hash value (usually inode number) to get
881 * @test: callback used for comparisons between inodes
882 * @set: callback used to initialize a new struct inode
883 * @data: opaque data pointer to pass to @test and @set
885 * This is iget() without the read_inode() portion of get_new_inode().
887 * iget5_locked() uses ifind() to search for the inode specified by @hashval
888 * and @data in the inode cache and if present it is returned with an increased
889 * reference count. This is a generalized version of iget_locked() for file
890 * systems where the inode number is not sufficient for unique identification
893 * If the inode is not in cache, get_new_inode() is called to allocate a new
894 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
895 * file system gets to fill it in before unlocking it via unlock_new_inode().
897 * Note both @test and @set are called with the inode_lock held, so can't sleep.
899 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
900 int (*test)(struct inode *, void *),
901 int (*set)(struct inode *, void *), void *data)
903 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
906 inode = ifind(sb, head, test, data);
910 * get_new_inode() will do the right thing, re-trying the search
911 * in case it had to block at any point.
913 return get_new_inode(sb, head, test, set, data);
916 EXPORT_SYMBOL(iget5_locked);
919 * iget_locked - obtain an inode from a mounted file system
920 * @sb: super block of file system
921 * @ino: inode number to get
923 * This is iget() without the read_inode() portion of get_new_inode_fast().
925 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
926 * the inode cache and if present it is returned with an increased reference
927 * count. This is for file systems where the inode number is sufficient for
928 * unique identification of an inode.
930 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
931 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
932 * The file system gets to fill it in before unlocking it via
933 * unlock_new_inode().
935 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
937 struct hlist_head *head = inode_hashtable + hash(sb, ino);
940 inode = ifind_fast(sb, head, ino);
944 * get_new_inode_fast() will do the right thing, re-trying the search
945 * in case it had to block at any point.
947 return get_new_inode_fast(sb, head, ino);
950 EXPORT_SYMBOL(iget_locked);
953 * __insert_inode_hash - hash an inode
954 * @inode: unhashed inode
955 * @hashval: unsigned long value used to locate this object in the
958 * Add an inode to the inode hash for this superblock.
960 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
962 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
963 spin_lock(&inode_lock);
964 hlist_add_head(&inode->i_hash, head);
965 spin_unlock(&inode_lock);
968 EXPORT_SYMBOL(__insert_inode_hash);
971 * remove_inode_hash - remove an inode from the hash
972 * @inode: inode to unhash
974 * Remove an inode from the superblock.
976 void remove_inode_hash(struct inode *inode)
978 spin_lock(&inode_lock);
979 hlist_del_init(&inode->i_hash);
980 spin_unlock(&inode_lock);
983 EXPORT_SYMBOL(remove_inode_hash);
986 * Tell the filesystem that this inode is no longer of any interest and should
987 * be completely destroyed.
989 * We leave the inode in the inode hash table until *after* the filesystem's
990 * ->delete_inode completes. This ensures that an iget (such as nfsd might
991 * instigate) will always find up-to-date information either in the hash or on
994 * I_FREEING is set so that no-one will take a new reference to the inode while
995 * it is being deleted.
997 void generic_delete_inode(struct inode *inode)
999 struct super_operations *op = inode->i_sb->s_op;
1001 list_del_init(&inode->i_list);
1002 inode->i_state|=I_FREEING;
1003 inodes_stat.nr_inodes--;
1004 spin_unlock(&inode_lock);
1006 if (inode->i_data.nrpages)
1007 truncate_inode_pages(&inode->i_data, 0);
1009 security_inode_delete(inode);
1011 if (op->delete_inode) {
1012 void (*delete)(struct inode *) = op->delete_inode;
1013 if (!is_bad_inode(inode))
1015 /* s_op->delete_inode internally recalls clear_inode() */
1019 spin_lock(&inode_lock);
1020 hlist_del_init(&inode->i_hash);
1021 spin_unlock(&inode_lock);
1022 wake_up_inode(inode);
1023 if (inode->i_state != I_CLEAR)
1025 destroy_inode(inode);
1028 EXPORT_SYMBOL(generic_delete_inode);
1030 static void generic_forget_inode(struct inode *inode)
1032 struct super_block *sb = inode->i_sb;
1034 if (!hlist_unhashed(&inode->i_hash)) {
1035 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1036 list_move(&inode->i_list, &inode_unused);
1037 inodes_stat.nr_unused++;
1038 spin_unlock(&inode_lock);
1039 if (!sb || (sb->s_flags & MS_ACTIVE))
1041 write_inode_now(inode, 1);
1042 spin_lock(&inode_lock);
1043 inodes_stat.nr_unused--;
1044 hlist_del_init(&inode->i_hash);
1046 list_del_init(&inode->i_list);
1047 inode->i_state|=I_FREEING;
1048 inodes_stat.nr_inodes--;
1049 spin_unlock(&inode_lock);
1050 if (inode->i_data.nrpages)
1051 truncate_inode_pages(&inode->i_data, 0);
1053 destroy_inode(inode);
1057 * Normal UNIX filesystem behaviour: delete the
1058 * inode when the usage count drops to zero, and
1061 static void generic_drop_inode(struct inode *inode)
1063 if (!inode->i_nlink)
1064 generic_delete_inode(inode);
1066 generic_forget_inode(inode);
1070 * Called when we're dropping the last reference
1073 * Call the FS "drop()" function, defaulting to
1074 * the legacy UNIX filesystem behaviour..
1076 * NOTE! NOTE! NOTE! We're called with the inode lock
1077 * held, and the drop function is supposed to release
1080 static inline void iput_final(struct inode *inode)
1082 struct super_operations *op = inode->i_sb->s_op;
1083 void (*drop)(struct inode *) = generic_drop_inode;
1085 if (op && op->drop_inode)
1086 drop = op->drop_inode;
1091 * iput - put an inode
1092 * @inode: inode to put
1094 * Puts an inode, dropping its usage count. If the inode use count hits
1095 * zero the inode is also then freed and may be destroyed.
1097 void iput(struct inode *inode)
1100 struct super_operations *op = inode->i_sb->s_op;
1102 if (inode->i_state == I_CLEAR)
1105 if (op && op->put_inode)
1106 op->put_inode(inode);
1108 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1113 EXPORT_SYMBOL(iput);
1116 * bmap - find a block number in a file
1117 * @inode: inode of file
1118 * @block: block to find
1120 * Returns the block number on the device holding the inode that
1121 * is the disk block number for the block of the file requested.
1122 * That is, asked for block 4 of inode 1 the function will return the
1123 * disk block relative to the disk start that holds that block of the
1126 sector_t bmap(struct inode * inode, sector_t block)
1129 if (inode->i_mapping->a_ops->bmap)
1130 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1134 EXPORT_SYMBOL(bmap);
1137 * Return true if the filesystem which backs this inode considers the two
1138 * passed timespecs to be sufficiently different to warrant flushing the
1139 * altered time out to disk.
1141 static int inode_times_differ(struct inode *inode,
1142 struct timespec *old, struct timespec *new)
1144 if (IS_ONE_SECOND(inode))
1145 return old->tv_sec != new->tv_sec;
1146 return !timespec_equal(old, new);
1150 * update_atime - update the access time
1151 * @inode: inode accessed
1153 * Update the accessed time on an inode and mark it for writeback.
1154 * This function automatically handles read only file systems and media,
1155 * as well as the "noatime" flag and inode specific "noatime" markers.
1157 void update_atime(struct inode *inode)
1159 struct timespec now;
1161 if (IS_NOATIME(inode))
1163 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1165 if (IS_RDONLY(inode))
1168 now = current_kernel_time();
1169 if (inode_times_differ(inode, &inode->i_atime, &now)) {
1170 inode->i_atime = now;
1171 mark_inode_dirty_sync(inode);
1173 if (!timespec_equal(&inode->i_atime, &now))
1174 inode->i_atime = now;
1178 EXPORT_SYMBOL(update_atime);
1181 * inode_update_time - update mtime and ctime time
1182 * @inode: inode accessed
1183 * @ctime_too: update ctime too
1185 * Update the mtime time on an inode and mark it for writeback.
1186 * When ctime_too is specified update the ctime too.
1189 void inode_update_time(struct inode *inode, int ctime_too)
1191 struct timespec now;
1194 if (IS_NOCMTIME(inode))
1196 if (IS_RDONLY(inode))
1199 now = current_kernel_time();
1201 if (inode_times_differ(inode, &inode->i_mtime, &now))
1203 inode->i_mtime = now;
1206 if (inode_times_differ(inode, &inode->i_ctime, &now))
1208 inode->i_ctime = now;
1211 mark_inode_dirty_sync(inode);
1214 EXPORT_SYMBOL(inode_update_time);
1216 int inode_needs_sync(struct inode *inode)
1220 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1225 EXPORT_SYMBOL(inode_needs_sync);
1228 * Quota functions that want to walk the inode lists..
1232 /* Function back in dquot.c */
1233 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1235 void remove_dquot_ref(struct super_block *sb, int type, struct list_head *tofree_head)
1237 struct inode *inode;
1238 struct list_head *act_head;
1241 return; /* nothing to do */
1242 spin_lock(&inode_lock); /* This lock is for inodes code */
1244 /* We hold dqptr_sem so we are safe against the quota code */
1245 list_for_each(act_head, &inode_in_use) {
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, &inode_unused) {
1251 inode = list_entry(act_head, struct inode, i_list);
1252 if (inode->i_sb == sb && !IS_NOQUOTA(inode))
1253 remove_inode_dquot_ref(inode, type, tofree_head);
1255 list_for_each(act_head, &sb->s_dirty) {
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 list_for_each(act_head, &sb->s_io) {
1261 inode = list_entry(act_head, struct inode, i_list);
1262 if (!IS_NOQUOTA(inode))
1263 remove_inode_dquot_ref(inode, type, tofree_head);
1265 spin_unlock(&inode_lock);
1271 * Hashed waitqueues for wait_on_inode(). The table is pretty small - the
1272 * kernel doesn't lock many inodes at the same time.
1274 #define I_WAIT_TABLE_ORDER 3
1275 static struct i_wait_queue_head {
1276 wait_queue_head_t wqh;
1277 } ____cacheline_aligned_in_smp i_wait_queue_heads[1<<I_WAIT_TABLE_ORDER];
1280 * Return the address of the waitqueue_head to be used for this inode
1282 static wait_queue_head_t *i_waitq_head(struct inode *inode)
1284 return &i_wait_queue_heads[hash_ptr(inode, I_WAIT_TABLE_ORDER)].wqh;
1287 void __wait_on_inode(struct inode *inode)
1289 DECLARE_WAITQUEUE(wait, current);
1290 wait_queue_head_t *wq = i_waitq_head(inode);
1292 add_wait_queue(wq, &wait);
1294 set_current_state(TASK_UNINTERRUPTIBLE);
1295 if (inode->i_state & I_LOCK) {
1299 remove_wait_queue(wq, &wait);
1300 __set_current_state(TASK_RUNNING);
1304 * If we try to find an inode in the inode hash while it is being deleted, we
1305 * have to wait until the filesystem completes its deletion before reporting
1306 * that it isn't found. This is because iget will immediately call
1307 * ->read_inode, and we want to be sure that evidence of the deletion is found
1310 * This call might return early if an inode which shares the waitq is woken up.
1311 * This is most easily handled by the caller which will loop around again
1312 * looking for the inode.
1314 * This is called with inode_lock held.
1316 static void __wait_on_freeing_inode(struct inode *inode)
1318 DECLARE_WAITQUEUE(wait, current);
1319 wait_queue_head_t *wq = i_waitq_head(inode);
1321 add_wait_queue(wq, &wait);
1322 set_current_state(TASK_UNINTERRUPTIBLE);
1323 spin_unlock(&inode_lock);
1325 remove_wait_queue(wq, &wait);
1326 spin_lock(&inode_lock);
1329 void wake_up_inode(struct inode *inode)
1331 wait_queue_head_t *wq = i_waitq_head(inode);
1334 * Prevent speculative execution through spin_unlock(&inode_lock);
1337 if (waitqueue_active(wq))
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)
1359 alloc_large_system_hash("Inode-cache",
1360 sizeof(struct hlist_head),
1367 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1368 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1371 void __init inode_init(unsigned long mempages)
1375 for (i = 0; i < ARRAY_SIZE(i_wait_queue_heads); i++)
1376 init_waitqueue_head(&i_wait_queue_heads[i].wqh);
1378 /* inode slab cache */
1379 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1380 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, init_once,
1382 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1385 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1387 inode->i_mode = mode;
1388 if (S_ISCHR(mode)) {
1389 inode->i_fop = &def_chr_fops;
1390 inode->i_rdev = rdev;
1391 } else if (S_ISBLK(mode)) {
1392 inode->i_fop = &def_blk_fops;
1393 inode->i_rdev = rdev;
1394 } else if (S_ISFIFO(mode))
1395 inode->i_fop = &def_fifo_fops;
1396 else if (S_ISSOCK(mode))
1397 inode->i_fop = &bad_sock_fops;
1399 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1402 EXPORT_SYMBOL(init_special_inode);