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
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 DEFINE_SPINLOCK(inode_lock);
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 DECLARE_MUTEX(iprune_sem);
95 * Statistics gathering..
97 struct inodes_stat_t inodes_stat;
99 static kmem_cache_t * inode_cachep;
101 static struct inode *alloc_inode(struct super_block *sb)
103 static struct address_space_operations empty_aops;
104 static struct inode_operations empty_iops;
105 static struct file_operations empty_fops;
108 if (sb->s_op->alloc_inode)
109 inode = sb->s_op->alloc_inode(sb);
111 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
114 struct address_space * const mapping = &inode->i_data;
118 /* essential because of inode slab reuse */
120 inode->i_blkbits = sb->s_blocksize_bits;
122 atomic_set(&inode->i_count, 1);
123 inode->i_op = &empty_iops;
124 inode->i_fop = &empty_fops;
126 atomic_set(&inode->i_writecount, 0);
130 inode->i_generation = 0;
132 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
134 inode->i_pipe = NULL;
135 inode->i_bdev = NULL;
136 inode->i_cdev = NULL;
138 inode->i_security = NULL;
139 inode->dirtied_when = 0;
140 if (security_inode_alloc(inode)) {
141 if (inode->i_sb->s_op->destroy_inode)
142 inode->i_sb->s_op->destroy_inode(inode);
144 kmem_cache_free(inode_cachep, (inode));
148 mapping->a_ops = &empty_aops;
149 mapping->host = inode;
151 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
152 mapping->assoc_mapping = NULL;
153 mapping->backing_dev_info = &default_backing_dev_info;
156 * If the block_device provides a backing_dev_info for client
157 * inodes then use that. Otherwise the inode share the bdev's
161 struct backing_dev_info *bdi;
163 bdi = sb->s_bdev->bd_inode_backing_dev_info;
165 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
166 mapping->backing_dev_info = bdi;
168 memset(&inode->u, 0, sizeof(inode->u));
169 inode->i_mapping = mapping;
174 void destroy_inode(struct inode *inode)
176 if (inode_has_buffers(inode))
178 security_inode_free(inode);
179 if (inode->i_sb->s_op->destroy_inode)
180 inode->i_sb->s_op->destroy_inode(inode);
182 kmem_cache_free(inode_cachep, (inode));
187 * These are initializations that only need to be done
188 * once, because the fields are idempotent across use
189 * of the inode, so let the slab aware of that.
191 void inode_init_once(struct inode *inode)
193 memset(inode, 0, sizeof(*inode));
194 INIT_HLIST_NODE(&inode->i_hash);
195 INIT_LIST_HEAD(&inode->i_dentry);
196 INIT_LIST_HEAD(&inode->i_devices);
197 sema_init(&inode->i_sem, 1);
198 init_rwsem(&inode->i_alloc_sem);
199 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
200 rwlock_init(&inode->i_data.tree_lock);
201 spin_lock_init(&inode->i_data.i_mmap_lock);
202 INIT_LIST_HEAD(&inode->i_data.private_list);
203 spin_lock_init(&inode->i_data.private_lock);
204 INIT_RAW_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--;
236 EXPORT_SYMBOL_GPL(__iget);
239 * clear_inode - clear an inode
240 * @inode: inode to clear
242 * This is called by the filesystem to tell us
243 * that the inode is no longer useful. We just
244 * terminate it with extreme prejudice.
246 void clear_inode(struct inode *inode)
249 invalidate_inode_buffers(inode);
251 if (inode->i_data.nrpages)
253 if (!(inode->i_state & I_FREEING))
255 if (inode->i_state & I_CLEAR)
257 wait_on_inode(inode);
259 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
260 inode->i_sb->s_op->clear_inode(inode);
265 inode->i_state = I_CLEAR;
268 EXPORT_SYMBOL(clear_inode);
271 * dispose_list - dispose of the contents of a local list
272 * @head: the head of the list to free
274 * Dispose-list gets a local list with local inodes in it, so it doesn't
275 * need to worry about list corruption and SMP locks.
277 static void dispose_list(struct list_head *head)
281 while (!list_empty(head)) {
284 inode = list_entry(head->next, struct inode, i_list);
285 list_del(&inode->i_list);
287 if (inode->i_data.nrpages)
288 truncate_inode_pages(&inode->i_data, 0);
290 destroy_inode(inode);
293 spin_lock(&inode_lock);
294 inodes_stat.nr_inodes -= nr_disposed;
295 spin_unlock(&inode_lock);
299 * Invalidate all inodes for a device.
301 static int invalidate_list(struct list_head *head, struct list_head *dispose)
303 struct list_head *next;
304 int busy = 0, count = 0;
308 struct list_head * tmp = next;
309 struct inode * inode;
312 * We can reschedule here without worrying about the list's
313 * consistency because the per-sb list of inodes must not
314 * change during umount anymore, and because iprune_sem keeps
315 * shrink_icache_memory() away.
317 cond_resched_lock(&inode_lock);
322 inode = list_entry(tmp, struct inode, i_sb_list);
323 invalidate_inode_buffers(inode);
324 if (!atomic_read(&inode->i_count)) {
325 hlist_del_init(&inode->i_hash);
326 list_del(&inode->i_sb_list);
327 list_move(&inode->i_list, dispose);
328 inode->i_state |= I_FREEING;
334 /* only unused inodes may be cached with i_count zero */
335 inodes_stat.nr_unused -= count;
340 * invalidate_inodes - discard the inodes on a device
343 * Discard all of the inodes for a given superblock. If the discard
344 * fails because there are busy inodes then a non zero value is returned.
345 * If the discard is successful all the inodes have been discarded.
347 int invalidate_inodes(struct super_block * sb)
350 LIST_HEAD(throw_away);
353 spin_lock(&inode_lock);
354 busy = invalidate_list(&sb->s_inodes, &throw_away);
355 spin_unlock(&inode_lock);
357 dispose_list(&throw_away);
363 EXPORT_SYMBOL(invalidate_inodes);
365 int __invalidate_device(struct block_device *bdev)
367 struct super_block *sb = get_super(bdev);
372 * no need to lock the super, get_super holds the
373 * read semaphore so the filesystem cannot go away
374 * under us (->put_super runs with the write lock
377 shrink_dcache_sb(sb);
378 res = invalidate_inodes(sb);
381 invalidate_bdev(bdev, 0);
384 EXPORT_SYMBOL(__invalidate_device);
386 static int can_unuse(struct inode *inode)
390 if (inode_has_buffers(inode))
392 if (atomic_read(&inode->i_count))
394 if (inode->i_data.nrpages)
400 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
401 * a temporary list and then are freed outside inode_lock by dispose_list().
403 * Any inodes which are pinned purely because of attached pagecache have their
404 * pagecache removed. We expect the final iput() on that inode to add it to
405 * the front of the inode_unused list. So look for it there and if the
406 * inode is still freeable, proceed. The right inode is found 99.9% of the
407 * time in testing on a 4-way.
409 * If the inode has metadata buffers attached to mapping->private_list then
410 * try to remove them.
412 static void prune_icache(int nr_to_scan)
417 unsigned long reap = 0;
420 spin_lock(&inode_lock);
421 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
424 if (list_empty(&inode_unused))
427 inode = list_entry(inode_unused.prev, struct inode, i_list);
429 if (inode->i_state || atomic_read(&inode->i_count)) {
430 list_move(&inode->i_list, &inode_unused);
433 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
435 spin_unlock(&inode_lock);
436 if (remove_inode_buffers(inode))
437 reap += invalidate_inode_pages(&inode->i_data);
439 spin_lock(&inode_lock);
441 if (inode != list_entry(inode_unused.next,
442 struct inode, i_list))
443 continue; /* wrong inode or list_empty */
444 if (!can_unuse(inode))
447 hlist_del_init(&inode->i_hash);
448 list_del_init(&inode->i_sb_list);
449 list_move(&inode->i_list, &freeable);
450 inode->i_state |= I_FREEING;
453 inodes_stat.nr_unused -= nr_pruned;
454 spin_unlock(&inode_lock);
456 dispose_list(&freeable);
459 if (current_is_kswapd())
460 mod_page_state(kswapd_inodesteal, reap);
462 mod_page_state(pginodesteal, reap);
466 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
467 * "unused" means that no dentries are referring to the inodes: the files are
468 * not open and the dcache references to those inodes have already been
471 * This function is passed the number of inodes to scan, and it returns the
472 * total number of remaining possibly-reclaimable inodes.
474 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
478 * Nasty deadlock avoidance. We may hold various FS locks,
479 * and we don't want to recurse into the FS that called us
480 * in clear_inode() and friends..
482 if (!(gfp_mask & __GFP_FS))
486 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
489 static void __wait_on_freeing_inode(struct inode *inode);
491 * Called with the inode lock held.
492 * NOTE: we are not increasing the inode-refcount, you must call __iget()
493 * by hand after calling find_inode now! This simplifies iunique and won't
494 * add any additional branch in the common code.
496 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
498 struct hlist_node *node;
499 struct inode * inode = NULL;
502 hlist_for_each (node, head) {
503 inode = hlist_entry(node, struct inode, i_hash);
504 if (inode->i_sb != sb)
506 if (!test(inode, data))
508 if (inode->i_state & (I_FREEING|I_CLEAR)) {
509 __wait_on_freeing_inode(inode);
514 return node ? inode : NULL;
518 * find_inode_fast is the fast path version of find_inode, see the comment at
519 * iget_locked for details.
521 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
523 struct hlist_node *node;
524 struct inode * inode = NULL;
527 hlist_for_each (node, head) {
528 inode = hlist_entry(node, struct inode, i_hash);
529 if (inode->i_ino != ino)
531 if (inode->i_sb != sb)
533 if (inode->i_state & (I_FREEING|I_CLEAR)) {
534 __wait_on_freeing_inode(inode);
539 return node ? inode : NULL;
543 * new_inode - obtain an inode
546 * Allocates a new inode for given superblock.
548 struct inode *new_inode(struct super_block *sb)
550 static unsigned long last_ino;
551 struct inode * inode;
553 spin_lock_prefetch(&inode_lock);
555 inode = alloc_inode(sb);
557 spin_lock(&inode_lock);
558 inodes_stat.nr_inodes++;
559 list_add(&inode->i_list, &inode_in_use);
560 list_add(&inode->i_sb_list, &sb->s_inodes);
561 inode->i_ino = ++last_ino;
563 spin_unlock(&inode_lock);
568 EXPORT_SYMBOL(new_inode);
570 void unlock_new_inode(struct inode *inode)
573 * This is special! We do not need the spinlock
574 * when clearing I_LOCK, because we're guaranteed
575 * that nobody else tries to do anything about the
576 * state of the inode when it is locked, as we
577 * just created it (so there can be no old holders
578 * that haven't tested I_LOCK).
580 inode->i_state &= ~(I_LOCK|I_NEW);
581 wake_up_inode(inode);
584 EXPORT_SYMBOL(unlock_new_inode);
587 * This is called without the inode lock held.. Be careful.
589 * We no longer cache the sb_flags in i_flags - see fs.h
590 * -- rmk@arm.uk.linux.org
592 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)
594 struct inode * inode;
596 inode = alloc_inode(sb);
600 spin_lock(&inode_lock);
601 /* We released the lock, so.. */
602 old = find_inode(sb, head, test, data);
604 if (set(inode, data))
607 inodes_stat.nr_inodes++;
608 list_add(&inode->i_list, &inode_in_use);
609 list_add(&inode->i_sb_list, &sb->s_inodes);
610 hlist_add_head(&inode->i_hash, head);
611 inode->i_state = I_LOCK|I_NEW;
612 spin_unlock(&inode_lock);
614 /* Return the locked inode with I_NEW set, the
615 * caller is responsible for filling in the contents
621 * Uhhuh, somebody else created the same inode under
622 * us. Use the old inode instead of the one we just
626 spin_unlock(&inode_lock);
627 destroy_inode(inode);
629 wait_on_inode(inode);
634 spin_unlock(&inode_lock);
635 destroy_inode(inode);
640 * get_new_inode_fast is the fast path version of get_new_inode, see the
641 * comment at iget_locked for details.
643 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
645 struct inode * inode;
647 inode = alloc_inode(sb);
651 spin_lock(&inode_lock);
652 /* We released the lock, so.. */
653 old = find_inode_fast(sb, head, ino);
656 inodes_stat.nr_inodes++;
657 list_add(&inode->i_list, &inode_in_use);
658 list_add(&inode->i_sb_list, &sb->s_inodes);
659 hlist_add_head(&inode->i_hash, head);
660 inode->i_state = I_LOCK|I_NEW;
661 spin_unlock(&inode_lock);
663 /* Return the locked inode with I_NEW set, the
664 * caller is responsible for filling in the contents
670 * Uhhuh, somebody else created the same inode under
671 * us. Use the old inode instead of the one we just
675 spin_unlock(&inode_lock);
676 destroy_inode(inode);
678 wait_on_inode(inode);
683 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
687 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
689 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
690 return tmp & I_HASHMASK;
694 * iunique - get a unique inode number
696 * @max_reserved: highest reserved inode number
698 * Obtain an inode number that is unique on the system for a given
699 * superblock. This is used by file systems that have no natural
700 * permanent inode numbering system. An inode number is returned that
701 * is higher than the reserved limit but unique.
704 * With a large number of inodes live on the file system this function
705 * currently becomes quite slow.
707 ino_t iunique(struct super_block *sb, ino_t max_reserved)
709 static ino_t counter;
711 struct hlist_head * head;
713 spin_lock(&inode_lock);
715 if (counter > max_reserved) {
716 head = inode_hashtable + hash(sb,counter);
718 inode = find_inode_fast(sb, head, res);
720 spin_unlock(&inode_lock);
724 counter = max_reserved + 1;
730 EXPORT_SYMBOL(iunique);
732 struct inode *igrab(struct inode *inode)
734 spin_lock(&inode_lock);
735 if (!(inode->i_state & I_FREEING))
739 * Handle the case where s_op->clear_inode is not been
740 * called yet, and somebody is calling igrab
741 * while the inode is getting freed.
744 spin_unlock(&inode_lock);
748 EXPORT_SYMBOL(igrab);
751 * ifind - internal function, you want ilookup5() or iget5().
752 * @sb: super block of file system to search
753 * @head: the head of the list to search
754 * @test: callback used for comparisons between inodes
755 * @data: opaque data pointer to pass to @test
757 * ifind() searches for the inode specified by @data in the inode
758 * cache. This is a generalized version of ifind_fast() for file systems where
759 * the inode number is not sufficient for unique identification of an inode.
761 * If the inode is in the cache, the inode is returned with an incremented
764 * Otherwise NULL is returned.
766 * Note, @test is called with the inode_lock held, so can't sleep.
768 static inline struct inode *ifind(struct super_block *sb,
769 struct hlist_head *head, int (*test)(struct inode *, void *),
774 spin_lock(&inode_lock);
775 inode = find_inode(sb, head, test, data);
778 spin_unlock(&inode_lock);
779 wait_on_inode(inode);
782 spin_unlock(&inode_lock);
787 * ifind_fast - internal function, you want ilookup() or iget().
788 * @sb: super block of file system to search
789 * @head: head of the list to search
790 * @ino: inode number to search for
792 * ifind_fast() searches for the inode @ino in the inode cache. This is for
793 * file systems where the inode number is sufficient for unique identification
796 * If the inode is in the cache, the inode is returned with an incremented
799 * Otherwise NULL is returned.
801 static inline struct inode *ifind_fast(struct super_block *sb,
802 struct hlist_head *head, unsigned long ino)
806 spin_lock(&inode_lock);
807 inode = find_inode_fast(sb, head, ino);
810 spin_unlock(&inode_lock);
811 wait_on_inode(inode);
814 spin_unlock(&inode_lock);
819 * ilookup5 - search for an inode in the inode cache
820 * @sb: super block of file system to search
821 * @hashval: hash value (usually inode number) to search for
822 * @test: callback used for comparisons between inodes
823 * @data: opaque data pointer to pass to @test
825 * ilookup5() uses ifind() to search for the inode specified by @hashval and
826 * @data in the inode cache. This is a generalized version of ilookup() for
827 * file systems where the inode number is not sufficient for unique
828 * identification of an inode.
830 * If the inode is in the cache, the inode is returned with an incremented
833 * Otherwise NULL is returned.
835 * Note, @test is called with the inode_lock held, so can't sleep.
837 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
838 int (*test)(struct inode *, void *), void *data)
840 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
842 return ifind(sb, head, test, data);
845 EXPORT_SYMBOL(ilookup5);
848 * ilookup - search for an inode in the inode cache
849 * @sb: super block of file system to search
850 * @ino: inode number to search for
852 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
853 * This is for file systems where the inode number is sufficient for unique
854 * identification of an inode.
856 * If the inode is in the cache, the inode is returned with an incremented
859 * Otherwise NULL is returned.
861 struct inode *ilookup(struct super_block *sb, unsigned long ino)
863 struct hlist_head *head = inode_hashtable + hash(sb, ino);
865 return ifind_fast(sb, head, ino);
868 EXPORT_SYMBOL(ilookup);
871 * iget5_locked - obtain an inode from a mounted file system
872 * @sb: super block of file system
873 * @hashval: hash value (usually inode number) to get
874 * @test: callback used for comparisons between inodes
875 * @set: callback used to initialize a new struct inode
876 * @data: opaque data pointer to pass to @test and @set
878 * This is iget() without the read_inode() portion of get_new_inode().
880 * iget5_locked() uses ifind() to search for the inode specified by @hashval
881 * and @data in the inode cache and if present it is returned with an increased
882 * reference count. This is a generalized version of iget_locked() for file
883 * systems where the inode number is not sufficient for unique identification
886 * If the inode is not in cache, get_new_inode() is called to allocate a new
887 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
888 * file system gets to fill it in before unlocking it via unlock_new_inode().
890 * Note both @test and @set are called with the inode_lock held, so can't sleep.
892 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
893 int (*test)(struct inode *, void *),
894 int (*set)(struct inode *, void *), void *data)
896 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
899 inode = ifind(sb, head, test, data);
903 * get_new_inode() will do the right thing, re-trying the search
904 * in case it had to block at any point.
906 return get_new_inode(sb, head, test, set, data);
909 EXPORT_SYMBOL(iget5_locked);
912 * iget_locked - obtain an inode from a mounted file system
913 * @sb: super block of file system
914 * @ino: inode number to get
916 * This is iget() without the read_inode() portion of get_new_inode_fast().
918 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
919 * the inode cache and if present it is returned with an increased reference
920 * count. This is for file systems where the inode number is sufficient for
921 * unique identification of an inode.
923 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
924 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
925 * The file system gets to fill it in before unlocking it via
926 * unlock_new_inode().
928 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
930 struct hlist_head *head = inode_hashtable + hash(sb, ino);
933 inode = ifind_fast(sb, head, ino);
937 * get_new_inode_fast() will do the right thing, re-trying the search
938 * in case it had to block at any point.
940 return get_new_inode_fast(sb, head, ino);
943 EXPORT_SYMBOL(iget_locked);
946 * __insert_inode_hash - hash an inode
947 * @inode: unhashed inode
948 * @hashval: unsigned long value used to locate this object in the
951 * Add an inode to the inode hash for this superblock.
953 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
955 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
956 spin_lock(&inode_lock);
957 hlist_add_head(&inode->i_hash, head);
958 spin_unlock(&inode_lock);
961 EXPORT_SYMBOL(__insert_inode_hash);
964 * remove_inode_hash - remove an inode from the hash
965 * @inode: inode to unhash
967 * Remove an inode from the superblock.
969 void remove_inode_hash(struct inode *inode)
971 spin_lock(&inode_lock);
972 hlist_del_init(&inode->i_hash);
973 spin_unlock(&inode_lock);
976 EXPORT_SYMBOL(remove_inode_hash);
979 * Tell the filesystem that this inode is no longer of any interest and should
980 * be completely destroyed.
982 * We leave the inode in the inode hash table until *after* the filesystem's
983 * ->delete_inode completes. This ensures that an iget (such as nfsd might
984 * instigate) will always find up-to-date information either in the hash or on
987 * I_FREEING is set so that no-one will take a new reference to the inode while
988 * it is being deleted.
990 void generic_delete_inode(struct inode *inode)
992 struct super_operations *op = inode->i_sb->s_op;
994 list_del_init(&inode->i_list);
995 list_del_init(&inode->i_sb_list);
996 inode->i_state|=I_FREEING;
997 inodes_stat.nr_inodes--;
998 spin_unlock(&inode_lock);
1000 if (inode->i_data.nrpages)
1001 truncate_inode_pages(&inode->i_data, 0);
1003 security_inode_delete(inode);
1005 if (op->delete_inode) {
1006 void (*delete)(struct inode *) = op->delete_inode;
1007 if (!is_bad_inode(inode))
1009 /* s_op->delete_inode internally recalls clear_inode() */
1013 spin_lock(&inode_lock);
1014 hlist_del_init(&inode->i_hash);
1015 spin_unlock(&inode_lock);
1016 wake_up_inode(inode);
1017 if (inode->i_state != I_CLEAR)
1019 destroy_inode(inode);
1022 EXPORT_SYMBOL(generic_delete_inode);
1024 static void generic_forget_inode(struct inode *inode)
1026 struct super_block *sb = inode->i_sb;
1028 if (!hlist_unhashed(&inode->i_hash)) {
1029 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1030 list_move(&inode->i_list, &inode_unused);
1031 inodes_stat.nr_unused++;
1032 spin_unlock(&inode_lock);
1033 if (!sb || (sb->s_flags & MS_ACTIVE))
1035 write_inode_now(inode, 1);
1036 spin_lock(&inode_lock);
1037 inodes_stat.nr_unused--;
1038 hlist_del_init(&inode->i_hash);
1040 list_del_init(&inode->i_list);
1041 list_del_init(&inode->i_sb_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 then freed and may also be destroyed.
1092 * Consequently, iput() can sleep.
1094 void iput(struct inode *inode)
1097 struct super_operations *op = inode->i_sb->s_op;
1099 BUG_ON(inode->i_state == I_CLEAR);
1101 if (op && op->put_inode)
1102 op->put_inode(inode);
1104 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1109 EXPORT_SYMBOL(iput);
1112 * bmap - find a block number in a file
1113 * @inode: inode of file
1114 * @block: block to find
1116 * Returns the block number on the device holding the inode that
1117 * is the disk block number for the block of the file requested.
1118 * That is, asked for block 4 of inode 1 the function will return the
1119 * disk block relative to the disk start that holds that block of the
1122 sector_t bmap(struct inode * inode, sector_t block)
1125 if (inode->i_mapping->a_ops->bmap)
1126 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1130 EXPORT_SYMBOL(bmap);
1133 * update_atime - update the access time
1134 * @inode: inode accessed
1136 * Update the accessed time on an inode and mark it for writeback.
1137 * This function automatically handles read only file systems and media,
1138 * as well as the "noatime" flag and inode specific "noatime" markers.
1140 void update_atime(struct inode *inode)
1142 struct timespec now;
1144 if (IS_NOATIME(inode))
1146 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1148 if (IS_RDONLY(inode))
1151 now = current_fs_time(inode->i_sb);
1152 if (!timespec_equal(&inode->i_atime, &now)) {
1153 inode->i_atime = now;
1154 mark_inode_dirty_sync(inode);
1156 if (!timespec_equal(&inode->i_atime, &now))
1157 inode->i_atime = now;
1161 EXPORT_SYMBOL(update_atime);
1164 * inode_update_time - update mtime and ctime time
1165 * @inode: inode accessed
1166 * @ctime_too: update ctime too
1168 * Update the mtime time on an inode and mark it for writeback.
1169 * When ctime_too is specified update the ctime too.
1172 void inode_update_time(struct inode *inode, struct vfsmount *mnt, int ctime_too)
1174 struct timespec now;
1177 if (IS_NOCMTIME(inode))
1179 if (IS_RDONLY(inode) || MNT_IS_RDONLY(mnt))
1182 now = current_fs_time(inode->i_sb);
1183 if (!timespec_equal(&inode->i_mtime, &now))
1185 inode->i_mtime = now;
1188 if (!timespec_equal(&inode->i_ctime, &now))
1190 inode->i_ctime = now;
1193 mark_inode_dirty_sync(inode);
1196 EXPORT_SYMBOL(inode_update_time);
1198 int inode_needs_sync(struct inode *inode)
1202 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1207 EXPORT_SYMBOL(inode_needs_sync);
1210 * Quota functions that want to walk the inode lists..
1214 /* Function back in dquot.c */
1215 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1217 void remove_dquot_ref(struct super_block *sb, int type,
1218 struct list_head *tofree_head)
1220 struct inode *inode;
1223 return; /* nothing to do */
1224 spin_lock(&inode_lock); /* This lock is for inodes code */
1227 * We don't have to lock against quota code - test IS_QUOTAINIT is
1228 * just for speedup...
1230 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1231 if (!IS_NOQUOTA(inode))
1232 remove_inode_dquot_ref(inode, type, tofree_head);
1234 spin_unlock(&inode_lock);
1239 int inode_wait(void *word)
1246 * If we try to find an inode in the inode hash while it is being deleted, we
1247 * have to wait until the filesystem completes its deletion before reporting
1248 * that it isn't found. This is because iget will immediately call
1249 * ->read_inode, and we want to be sure that evidence of the deletion is found
1251 * This is called with inode_lock held.
1253 static void __wait_on_freeing_inode(struct inode *inode)
1255 wait_queue_head_t *wq;
1256 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1259 * I_FREEING and I_CLEAR are cleared in process context under
1260 * inode_lock, so we have to give the tasks who would clear them
1261 * a chance to run and acquire inode_lock.
1263 if (!(inode->i_state & I_LOCK)) {
1264 spin_unlock(&inode_lock);
1266 spin_lock(&inode_lock);
1269 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1270 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1271 spin_unlock(&inode_lock);
1273 finish_wait(wq, &wait.wait);
1274 spin_lock(&inode_lock);
1277 void wake_up_inode(struct inode *inode)
1280 * Prevent speculative execution through spin_unlock(&inode_lock);
1283 wake_up_bit(&inode->i_state, __I_LOCK);
1286 static __initdata unsigned long ihash_entries;
1287 static int __init set_ihash_entries(char *str)
1291 ihash_entries = simple_strtoul(str, &str, 0);
1294 __setup("ihash_entries=", set_ihash_entries);
1297 * Initialize the waitqueues and inode hash table.
1299 void __init inode_init_early(void)
1303 /* If hashes are distributed across NUMA nodes, defer
1304 * hash allocation until vmalloc space is available.
1310 alloc_large_system_hash("Inode-cache",
1311 sizeof(struct hlist_head),
1319 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1320 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1323 void __init inode_init(unsigned long mempages)
1327 /* inode slab cache */
1328 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1329 0, SLAB_RECLAIM_ACCOUNT|SLAB_PANIC, init_once, NULL);
1330 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1332 /* Hash may have been set up in inode_init_early */
1337 alloc_large_system_hash("Inode-cache",
1338 sizeof(struct hlist_head),
1346 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1347 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1350 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1352 inode->i_mode = mode;
1353 if (S_ISCHR(mode)) {
1354 inode->i_fop = &def_chr_fops;
1355 inode->i_rdev = rdev;
1356 } else if (S_ISBLK(mode)) {
1357 inode->i_fop = &def_blk_fops;
1358 inode->i_rdev = rdev;
1359 } else if (S_ISFIFO(mode))
1360 inode->i_fop = &def_fifo_fops;
1361 else if (S_ISSOCK(mode))
1362 inode->i_fop = &bad_sock_fops;
1364 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1367 EXPORT_SYMBOL(init_special_inode);