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 void prune_icache(int nr_to_scan);
106 #define INODE_UNUSED_THRESHOLD 15000
107 #define PRUNE_BATCH_COUNT 32
109 void try_to_clip_inodes(void)
111 unsigned long count = 0;
112 /* if there are a LOT of unused inodes in cache, better shrink a few first */
114 /* check lockless first to not take the lock always here; racing occasionally isn't a big deal */
115 if (inodes_stat.nr_unused > INODE_UNUSED_THRESHOLD) {
116 spin_lock(&inode_lock);
117 if (inodes_stat.nr_unused > INODE_UNUSED_THRESHOLD)
118 count = inodes_stat.nr_unused - INODE_UNUSED_THRESHOLD;
119 spin_unlock(&inode_lock);
126 static struct inode *alloc_inode(struct super_block *sb)
128 static struct address_space_operations empty_aops;
129 static struct inode_operations empty_iops;
130 static struct file_operations empty_fops;
133 if (sb->s_op->alloc_inode)
134 inode = sb->s_op->alloc_inode(sb);
136 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
139 struct address_space * const mapping = &inode->i_data;
142 if (sb->s_flags & MS_TAGXID)
143 inode->i_xid = current->xid;
145 inode->i_xid = 0; /* maybe xid -1 would be better? */
146 // inode->i_dqh = dqhget(sb->s_dqh);
147 inode->i_blkbits = sb->s_blocksize_bits;
149 atomic_set(&inode->i_count, 1);
151 inode->i_op = &empty_iops;
152 inode->i_fop = &empty_fops;
154 atomic_set(&inode->i_writecount, 0);
158 inode->i_generation = 0;
160 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
162 inode->i_pipe = NULL;
163 inode->i_bdev = NULL;
164 inode->i_cdev = NULL;
166 // inode->i_xid = 0; /* maybe not too wise ... */
167 inode->i_security = NULL;
168 inode->dirtied_when = 0;
169 if (security_inode_alloc(inode)) {
170 if (inode->i_sb->s_op->destroy_inode)
171 inode->i_sb->s_op->destroy_inode(inode);
173 kmem_cache_free(inode_cachep, (inode));
177 mapping->a_ops = &empty_aops;
178 mapping->host = inode;
180 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
181 mapping->assoc_mapping = NULL;
182 mapping->backing_dev_info = &default_backing_dev_info;
185 * If the block_device provides a backing_dev_info for client
186 * inodes then use that. Otherwise the inode share the bdev's
190 struct backing_dev_info *bdi;
192 bdi = sb->s_bdev->bd_inode_backing_dev_info;
194 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
195 mapping->backing_dev_info = bdi;
197 memset(&inode->u, 0, sizeof(inode->u));
198 inode->i_mapping = mapping;
203 void destroy_inode(struct inode *inode)
205 if (inode_has_buffers(inode))
207 security_inode_free(inode);
208 if (inode->i_sb->s_op->destroy_inode)
209 inode->i_sb->s_op->destroy_inode(inode);
211 kmem_cache_free(inode_cachep, (inode));
216 * These are initializations that only need to be done
217 * once, because the fields are idempotent across use
218 * of the inode, so let the slab aware of that.
220 void inode_init_once(struct inode *inode)
222 memset(inode, 0, sizeof(*inode));
223 INIT_HLIST_NODE(&inode->i_hash);
224 INIT_LIST_HEAD(&inode->i_dentry);
225 INIT_LIST_HEAD(&inode->i_devices);
226 sema_init(&inode->i_sem, 1);
227 init_rwsem(&inode->i_alloc_sem);
228 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
229 spin_lock_init(&inode->i_data.tree_lock);
230 spin_lock_init(&inode->i_data.i_mmap_lock);
231 atomic_set(&inode->i_data.truncate_count, 0);
232 INIT_LIST_HEAD(&inode->i_data.private_list);
233 spin_lock_init(&inode->i_data.private_lock);
234 INIT_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
235 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
236 spin_lock_init(&inode->i_lock);
237 i_size_ordered_init(inode);
240 EXPORT_SYMBOL(inode_init_once);
242 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
244 struct inode * inode = (struct inode *) foo;
246 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
247 SLAB_CTOR_CONSTRUCTOR)
248 inode_init_once(inode);
252 * inode_lock must be held
254 void __iget(struct inode * inode)
256 if (atomic_read(&inode->i_count)) {
257 atomic_inc(&inode->i_count);
260 atomic_inc(&inode->i_count);
261 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
262 list_move(&inode->i_list, &inode_in_use);
263 inodes_stat.nr_unused--;
267 * clear_inode - clear an inode
268 * @inode: inode to clear
270 * This is called by the filesystem to tell us
271 * that the inode is no longer useful. We just
272 * terminate it with extreme prejudice.
274 void clear_inode(struct inode *inode)
277 invalidate_inode_buffers(inode);
279 if (inode->i_data.nrpages)
281 if (!(inode->i_state & I_FREEING))
283 if (inode->i_state & I_CLEAR)
285 wait_on_inode(inode);
287 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
288 inode->i_sb->s_op->clear_inode(inode);
293 inode->i_state = I_CLEAR;
296 EXPORT_SYMBOL(clear_inode);
299 * dispose_list - dispose of the contents of a local list
300 * @head: the head of the list to free
302 * Dispose-list gets a local list with local inodes in it, so it doesn't
303 * need to worry about list corruption and SMP locks.
305 static void dispose_list(struct list_head *head)
309 while (!list_empty(head)) {
312 inode = list_entry(head->next, struct inode, i_list);
313 list_del(&inode->i_list);
315 if (inode->i_data.nrpages)
316 truncate_inode_pages(&inode->i_data, 0);
318 destroy_inode(inode);
321 spin_lock(&inode_lock);
322 inodes_stat.nr_inodes -= nr_disposed;
323 spin_unlock(&inode_lock);
327 * Invalidate all inodes for a device.
329 static int invalidate_list(struct list_head *head, struct super_block * sb, struct list_head * dispose)
331 struct list_head *next;
332 int busy = 0, count = 0;
336 struct list_head * tmp = next;
337 struct inode * inode;
342 inode = list_entry(tmp, struct inode, i_list);
343 if (inode->i_sb != sb)
345 invalidate_inode_buffers(inode);
346 if (!atomic_read(&inode->i_count)) {
347 hlist_del_init(&inode->i_hash);
348 list_move(&inode->i_list, dispose);
349 inode->i_state |= I_FREEING;
355 /* only unused inodes may be cached with i_count zero */
356 inodes_stat.nr_unused -= count;
361 * This is a two-stage process. First we collect all
362 * offending inodes onto the throw-away list, and in
363 * the second stage we actually dispose of them. This
364 * is because we don't want to sleep while messing
365 * with the global lists..
369 * invalidate_inodes - discard the inodes on a device
372 * Discard all of the inodes for a given superblock. If the discard
373 * fails because there are busy inodes then a non zero value is returned.
374 * If the discard is successful all the inodes have been discarded.
376 int invalidate_inodes(struct super_block * sb)
379 LIST_HEAD(throw_away);
382 spin_lock(&inode_lock);
383 busy = invalidate_list(&inode_in_use, sb, &throw_away);
384 busy |= invalidate_list(&inode_unused, sb, &throw_away);
385 busy |= invalidate_list(&sb->s_dirty, sb, &throw_away);
386 busy |= invalidate_list(&sb->s_io, sb, &throw_away);
387 spin_unlock(&inode_lock);
389 dispose_list(&throw_away);
395 EXPORT_SYMBOL(invalidate_inodes);
397 int __invalidate_device(struct block_device *bdev, int do_sync)
399 struct super_block *sb;
406 sb = get_super(bdev);
409 * no need to lock the super, get_super holds the
410 * read semaphore so the filesystem cannot go away
411 * under us (->put_super runs with the write lock
414 shrink_dcache_sb(sb);
415 res = invalidate_inodes(sb);
418 invalidate_bdev(bdev, 0);
422 EXPORT_SYMBOL(__invalidate_device);
424 static int can_unuse(struct inode *inode)
428 if (inode_has_buffers(inode))
430 if (atomic_read(&inode->i_count))
432 if (inode->i_data.nrpages)
438 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
439 * a temporary list and then are freed outside inode_lock by dispose_list().
441 * Any inodes which are pinned purely because of attached pagecache have their
442 * pagecache removed. We expect the final iput() on that inode to add it to
443 * the front of the inode_unused list. So look for it there and if the
444 * inode is still freeable, proceed. The right inode is found 99.9% of the
445 * time in testing on a 4-way.
447 * If the inode has metadata buffers attached to mapping->private_list then
448 * try to remove them.
450 static void prune_icache(int nr_to_scan)
455 unsigned long reap = 0;
458 spin_lock(&inode_lock);
459 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
462 if (list_empty(&inode_unused))
465 inode = list_entry(inode_unused.prev, struct inode, i_list);
467 if (inode->i_state || atomic_read(&inode->i_count)) {
468 list_move(&inode->i_list, &inode_unused);
471 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
473 spin_unlock(&inode_lock);
474 if (remove_inode_buffers(inode))
475 reap += invalidate_inode_pages(&inode->i_data);
477 spin_lock(&inode_lock);
479 if (inode != list_entry(inode_unused.next,
480 struct inode, i_list))
481 continue; /* wrong inode or list_empty */
482 if (!can_unuse(inode))
485 hlist_del_init(&inode->i_hash);
486 list_move(&inode->i_list, &freeable);
487 inode->i_state |= I_FREEING;
490 inodes_stat.nr_unused -= nr_pruned;
491 spin_unlock(&inode_lock);
493 dispose_list(&freeable);
496 if (current_is_kswapd())
497 mod_page_state(kswapd_inodesteal, reap);
499 mod_page_state(pginodesteal, reap);
503 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
504 * "unused" means that no dentries are referring to the inodes: the files are
505 * not open and the dcache references to those inodes have already been
508 * This function is passed the number of inodes to scan, and it returns the
509 * total number of remaining possibly-reclaimable inodes.
511 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
515 * Nasty deadlock avoidance. We may hold various FS locks,
516 * and we don't want to recurse into the FS that called us
517 * in clear_inode() and friends..
519 if (gfp_mask & __GFP_FS)
522 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
525 static void __wait_on_freeing_inode(struct inode *inode);
527 * Called with the inode lock held.
528 * NOTE: we are not increasing the inode-refcount, you must call __iget()
529 * by hand after calling find_inode now! This simplifies iunique and won't
530 * add any additional branch in the common code.
532 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
534 struct hlist_node *node;
535 struct inode * inode = NULL;
538 hlist_for_each (node, head) {
539 inode = hlist_entry(node, struct inode, i_hash);
540 if (inode->i_sb != sb)
542 if (!test(inode, data))
544 if (inode->i_state & (I_FREEING|I_CLEAR)) {
545 __wait_on_freeing_inode(inode);
550 return node ? inode : NULL;
554 * find_inode_fast is the fast path version of find_inode, see the comment at
555 * iget_locked for details.
557 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
559 struct hlist_node *node;
560 struct inode * inode = NULL;
563 hlist_for_each (node, head) {
564 inode = hlist_entry(node, struct inode, i_hash);
565 if (inode->i_ino != ino)
567 if (inode->i_sb != sb)
569 if (inode->i_state & (I_FREEING|I_CLEAR)) {
570 __wait_on_freeing_inode(inode);
575 return node ? inode : NULL;
579 * new_inode - obtain an inode
582 * Allocates a new inode for given superblock.
584 struct inode *new_inode(struct super_block *sb)
586 static unsigned long last_ino;
587 struct inode * inode;
589 spin_lock_prefetch(&inode_lock);
591 inode = alloc_inode(sb);
593 spin_lock(&inode_lock);
594 inodes_stat.nr_inodes++;
595 list_add(&inode->i_list, &inode_in_use);
596 inode->i_ino = ++last_ino;
598 inode->i_xid = vx_current_xid();
599 spin_unlock(&inode_lock);
604 EXPORT_SYMBOL(new_inode);
606 void unlock_new_inode(struct inode *inode)
609 * This is special! We do not need the spinlock
610 * when clearing I_LOCK, because we're guaranteed
611 * that nobody else tries to do anything about the
612 * state of the inode when it is locked, as we
613 * just created it (so there can be no old holders
614 * that haven't tested I_LOCK).
616 inode->i_state &= ~(I_LOCK|I_NEW);
617 wake_up_inode(inode);
620 EXPORT_SYMBOL(unlock_new_inode);
623 * This is called without the inode lock held.. Be careful.
625 * We no longer cache the sb_flags in i_flags - see fs.h
626 * -- rmk@arm.uk.linux.org
628 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)
630 struct inode * inode;
632 inode = alloc_inode(sb);
636 spin_lock(&inode_lock);
637 /* We released the lock, so.. */
638 old = find_inode(sb, head, test, data);
640 if (set(inode, data))
643 inodes_stat.nr_inodes++;
644 list_add(&inode->i_list, &inode_in_use);
645 hlist_add_head(&inode->i_hash, head);
646 inode->i_state = I_LOCK|I_NEW;
647 spin_unlock(&inode_lock);
649 /* Return the locked inode with I_NEW set, the
650 * caller is responsible for filling in the contents
656 * Uhhuh, somebody else created the same inode under
657 * us. Use the old inode instead of the one we just
661 spin_unlock(&inode_lock);
662 destroy_inode(inode);
664 wait_on_inode(inode);
669 spin_unlock(&inode_lock);
670 destroy_inode(inode);
675 * get_new_inode_fast is the fast path version of get_new_inode, see the
676 * comment at iget_locked for details.
678 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
680 struct inode * inode;
682 inode = alloc_inode(sb);
686 spin_lock(&inode_lock);
687 /* We released the lock, so.. */
688 old = find_inode_fast(sb, head, ino);
691 inodes_stat.nr_inodes++;
692 list_add(&inode->i_list, &inode_in_use);
693 hlist_add_head(&inode->i_hash, head);
694 inode->i_state = I_LOCK|I_NEW;
695 spin_unlock(&inode_lock);
697 /* Return the locked inode with I_NEW set, the
698 * caller is responsible for filling in the contents
704 * Uhhuh, somebody else created the same inode under
705 * us. Use the old inode instead of the one we just
709 spin_unlock(&inode_lock);
710 destroy_inode(inode);
712 wait_on_inode(inode);
717 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
721 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
723 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
724 return tmp & I_HASHMASK;
728 * iunique - get a unique inode number
730 * @max_reserved: highest reserved inode number
732 * Obtain an inode number that is unique on the system for a given
733 * superblock. This is used by file systems that have no natural
734 * permanent inode numbering system. An inode number is returned that
735 * is higher than the reserved limit but unique.
738 * With a large number of inodes live on the file system this function
739 * currently becomes quite slow.
741 ino_t iunique(struct super_block *sb, ino_t max_reserved)
743 static ino_t counter;
745 struct hlist_head * head;
747 spin_lock(&inode_lock);
749 if (counter > max_reserved) {
750 head = inode_hashtable + hash(sb,counter);
752 inode = find_inode_fast(sb, head, res);
754 spin_unlock(&inode_lock);
758 counter = max_reserved + 1;
764 EXPORT_SYMBOL(iunique);
766 struct inode *igrab(struct inode *inode)
768 spin_lock(&inode_lock);
769 if (!(inode->i_state & I_FREEING))
773 * Handle the case where s_op->clear_inode is not been
774 * called yet, and somebody is calling igrab
775 * while the inode is getting freed.
778 spin_unlock(&inode_lock);
782 EXPORT_SYMBOL(igrab);
785 * ifind - internal function, you want ilookup5() or iget5().
786 * @sb: super block of file system to search
787 * @head: the head of the list to search
788 * @test: callback used for comparisons between inodes
789 * @data: opaque data pointer to pass to @test
791 * ifind() searches for the inode specified by @data in the inode
792 * cache. This is a generalized version of ifind_fast() for file systems where
793 * the inode number is not sufficient for unique identification of an inode.
795 * If the inode is in the cache, the inode is returned with an incremented
798 * Otherwise NULL is returned.
800 * Note, @test is called with the inode_lock held, so can't sleep.
802 static inline struct inode *ifind(struct super_block *sb,
803 struct hlist_head *head, int (*test)(struct inode *, void *),
808 spin_lock(&inode_lock);
809 inode = find_inode(sb, head, test, data);
812 spin_unlock(&inode_lock);
813 wait_on_inode(inode);
816 spin_unlock(&inode_lock);
821 * ifind_fast - internal function, you want ilookup() or iget().
822 * @sb: super block of file system to search
823 * @head: head of the list to search
824 * @ino: inode number to search for
826 * ifind_fast() searches for the inode @ino in the inode cache. This is for
827 * file systems where the inode number is sufficient for unique identification
830 * If the inode is in the cache, the inode is returned with an incremented
833 * Otherwise NULL is returned.
835 static inline struct inode *ifind_fast(struct super_block *sb,
836 struct hlist_head *head, unsigned long ino)
840 spin_lock(&inode_lock);
841 inode = find_inode_fast(sb, head, ino);
844 spin_unlock(&inode_lock);
845 wait_on_inode(inode);
848 spin_unlock(&inode_lock);
853 * ilookup5 - search for an inode in the inode cache
854 * @sb: super block of file system to search
855 * @hashval: hash value (usually inode number) to search for
856 * @test: callback used for comparisons between inodes
857 * @data: opaque data pointer to pass to @test
859 * ilookup5() uses ifind() to search for the inode specified by @hashval and
860 * @data in the inode cache. This is a generalized version of ilookup() for
861 * file systems where the inode number is not sufficient for unique
862 * identification of an inode.
864 * If the inode is in the cache, the inode is returned with an incremented
867 * Otherwise NULL is returned.
869 * Note, @test is called with the inode_lock held, so can't sleep.
871 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
872 int (*test)(struct inode *, void *), void *data)
874 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
876 return ifind(sb, head, test, data);
879 EXPORT_SYMBOL(ilookup5);
882 * ilookup - search for an inode in the inode cache
883 * @sb: super block of file system to search
884 * @ino: inode number to search for
886 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
887 * This is for file systems where the inode number is sufficient for unique
888 * identification of an inode.
890 * If the inode is in the cache, the inode is returned with an incremented
893 * Otherwise NULL is returned.
895 struct inode *ilookup(struct super_block *sb, unsigned long ino)
897 struct hlist_head *head = inode_hashtable + hash(sb, ino);
899 return ifind_fast(sb, head, ino);
902 EXPORT_SYMBOL(ilookup);
905 * iget5_locked - obtain an inode from a mounted file system
906 * @sb: super block of file system
907 * @hashval: hash value (usually inode number) to get
908 * @test: callback used for comparisons between inodes
909 * @set: callback used to initialize a new struct inode
910 * @data: opaque data pointer to pass to @test and @set
912 * This is iget() without the read_inode() portion of get_new_inode().
914 * iget5_locked() uses ifind() to search for the inode specified by @hashval
915 * and @data in the inode cache and if present it is returned with an increased
916 * reference count. This is a generalized version of iget_locked() for file
917 * systems where the inode number is not sufficient for unique identification
920 * If the inode is not in cache, get_new_inode() is called to allocate a new
921 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
922 * file system gets to fill it in before unlocking it via unlock_new_inode().
924 * Note both @test and @set are called with the inode_lock held, so can't sleep.
926 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
927 int (*test)(struct inode *, void *),
928 int (*set)(struct inode *, void *), void *data)
930 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
933 inode = ifind(sb, head, test, data);
937 * get_new_inode() will do the right thing, re-trying the search
938 * in case it had to block at any point.
940 return get_new_inode(sb, head, test, set, data);
943 EXPORT_SYMBOL(iget5_locked);
946 * iget_locked - obtain an inode from a mounted file system
947 * @sb: super block of file system
948 * @ino: inode number to get
950 * This is iget() without the read_inode() portion of get_new_inode_fast().
952 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
953 * the inode cache and if present it is returned with an increased reference
954 * count. This is for file systems where the inode number is sufficient for
955 * unique identification of an inode.
957 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
958 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
959 * The file system gets to fill it in before unlocking it via
960 * unlock_new_inode().
962 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
964 struct hlist_head *head = inode_hashtable + hash(sb, ino);
967 inode = ifind_fast(sb, head, ino);
971 * get_new_inode_fast() will do the right thing, re-trying the search
972 * in case it had to block at any point.
974 return get_new_inode_fast(sb, head, ino);
977 EXPORT_SYMBOL(iget_locked);
980 * __insert_inode_hash - hash an inode
981 * @inode: unhashed inode
982 * @hashval: unsigned long value used to locate this object in the
985 * Add an inode to the inode hash for this superblock.
987 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
989 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
990 spin_lock(&inode_lock);
991 hlist_add_head(&inode->i_hash, head);
992 spin_unlock(&inode_lock);
995 EXPORT_SYMBOL(__insert_inode_hash);
998 * remove_inode_hash - remove an inode from the hash
999 * @inode: inode to unhash
1001 * Remove an inode from the superblock.
1003 void remove_inode_hash(struct inode *inode)
1005 spin_lock(&inode_lock);
1006 hlist_del_init(&inode->i_hash);
1007 spin_unlock(&inode_lock);
1010 EXPORT_SYMBOL(remove_inode_hash);
1013 * Tell the filesystem that this inode is no longer of any interest and should
1014 * be completely destroyed.
1016 * We leave the inode in the inode hash table until *after* the filesystem's
1017 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1018 * instigate) will always find up-to-date information either in the hash or on
1021 * I_FREEING is set so that no-one will take a new reference to the inode while
1022 * it is being deleted.
1024 void generic_delete_inode(struct inode *inode)
1026 struct super_operations *op = inode->i_sb->s_op;
1028 list_del_init(&inode->i_list);
1029 inode->i_state|=I_FREEING;
1030 inodes_stat.nr_inodes--;
1031 spin_unlock(&inode_lock);
1033 if (inode->i_data.nrpages)
1034 truncate_inode_pages(&inode->i_data, 0);
1036 security_inode_delete(inode);
1038 if (op->delete_inode) {
1039 void (*delete)(struct inode *) = op->delete_inode;
1040 if (!is_bad_inode(inode))
1042 /* s_op->delete_inode internally recalls clear_inode() */
1046 spin_lock(&inode_lock);
1047 hlist_del_init(&inode->i_hash);
1048 spin_unlock(&inode_lock);
1049 wake_up_inode(inode);
1050 if (inode->i_state != I_CLEAR)
1052 destroy_inode(inode);
1055 EXPORT_SYMBOL(generic_delete_inode);
1057 static void generic_forget_inode(struct inode *inode)
1059 struct super_block *sb = inode->i_sb;
1061 if (!hlist_unhashed(&inode->i_hash)) {
1062 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1063 list_move(&inode->i_list, &inode_unused);
1064 inodes_stat.nr_unused++;
1065 spin_unlock(&inode_lock);
1066 if (!sb || (sb->s_flags & MS_ACTIVE))
1068 write_inode_now(inode, 1);
1069 spin_lock(&inode_lock);
1070 inodes_stat.nr_unused--;
1071 hlist_del_init(&inode->i_hash);
1073 list_del_init(&inode->i_list);
1074 inode->i_state|=I_FREEING;
1075 inodes_stat.nr_inodes--;
1076 spin_unlock(&inode_lock);
1077 if (inode->i_data.nrpages)
1078 truncate_inode_pages(&inode->i_data, 0);
1080 destroy_inode(inode);
1084 * Normal UNIX filesystem behaviour: delete the
1085 * inode when the usage count drops to zero, and
1088 static void generic_drop_inode(struct inode *inode)
1090 if (!inode->i_nlink)
1091 generic_delete_inode(inode);
1093 generic_forget_inode(inode);
1097 * Called when we're dropping the last reference
1100 * Call the FS "drop()" function, defaulting to
1101 * the legacy UNIX filesystem behaviour..
1103 * NOTE! NOTE! NOTE! We're called with the inode lock
1104 * held, and the drop function is supposed to release
1107 static inline void iput_final(struct inode *inode)
1109 struct super_operations *op = inode->i_sb->s_op;
1110 void (*drop)(struct inode *) = generic_drop_inode;
1112 if (op && op->drop_inode)
1113 drop = op->drop_inode;
1118 * iput - put an inode
1119 * @inode: inode to put
1121 * Puts an inode, dropping its usage count. If the inode use count hits
1122 * zero the inode is also then freed and may be destroyed.
1124 void iput(struct inode *inode)
1127 struct super_operations *op = inode->i_sb->s_op;
1129 if (inode->i_state == I_CLEAR)
1132 if (op && op->put_inode)
1133 op->put_inode(inode);
1135 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1140 EXPORT_SYMBOL(iput);
1143 * bmap - find a block number in a file
1144 * @inode: inode of file
1145 * @block: block to find
1147 * Returns the block number on the device holding the inode that
1148 * is the disk block number for the block of the file requested.
1149 * That is, asked for block 4 of inode 1 the function will return the
1150 * disk block relative to the disk start that holds that block of the
1153 sector_t bmap(struct inode * inode, sector_t block)
1156 if (inode->i_mapping->a_ops->bmap)
1157 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1161 EXPORT_SYMBOL(bmap);
1164 * Return true if the filesystem which backs this inode considers the two
1165 * passed timespecs to be sufficiently different to warrant flushing the
1166 * altered time out to disk.
1168 static int inode_times_differ(struct inode *inode,
1169 struct timespec *old, struct timespec *new)
1171 if (IS_ONE_SECOND(inode))
1172 return old->tv_sec != new->tv_sec;
1173 return !timespec_equal(old, new);
1177 * update_atime - update the access time
1178 * @inode: inode accessed
1180 * Update the accessed time on an inode and mark it for writeback.
1181 * This function automatically handles read only file systems and media,
1182 * as well as the "noatime" flag and inode specific "noatime" markers.
1184 void update_atime(struct inode *inode)
1186 struct timespec now;
1188 if (IS_NOATIME(inode))
1190 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1192 if (IS_RDONLY(inode))
1195 now = current_kernel_time();
1196 if (inode_times_differ(inode, &inode->i_atime, &now)) {
1197 inode->i_atime = now;
1198 mark_inode_dirty_sync(inode);
1200 if (!timespec_equal(&inode->i_atime, &now))
1201 inode->i_atime = now;
1205 EXPORT_SYMBOL(update_atime);
1208 * inode_update_time - update mtime and ctime time
1209 * @inode: inode accessed
1210 * @ctime_too: update ctime too
1212 * Update the mtime time on an inode and mark it for writeback.
1213 * When ctime_too is specified update the ctime too.
1216 void inode_update_time(struct inode *inode, int ctime_too)
1218 struct timespec now;
1221 if (IS_NOCMTIME(inode))
1223 if (IS_RDONLY(inode))
1226 now = current_kernel_time();
1228 if (inode_times_differ(inode, &inode->i_mtime, &now))
1230 inode->i_mtime = now;
1233 if (inode_times_differ(inode, &inode->i_ctime, &now))
1235 inode->i_ctime = now;
1238 mark_inode_dirty_sync(inode);
1241 EXPORT_SYMBOL(inode_update_time);
1243 int inode_needs_sync(struct inode *inode)
1247 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1252 EXPORT_SYMBOL(inode_needs_sync);
1255 * Quota functions that want to walk the inode lists..
1259 /* Function back in dquot.c */
1260 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1262 void remove_dquot_ref(struct super_block *sb, int type, struct list_head *tofree_head)
1264 struct inode *inode;
1265 struct list_head *act_head;
1268 return; /* nothing to do */
1269 spin_lock(&inode_lock); /* This lock is for inodes code */
1271 /* We hold dqptr_sem so we are safe against the quota code */
1272 list_for_each(act_head, &inode_in_use) {
1273 inode = list_entry(act_head, struct inode, i_list);
1274 if (inode->i_sb == sb && !IS_NOQUOTA(inode))
1275 remove_inode_dquot_ref(inode, type, tofree_head);
1277 list_for_each(act_head, &inode_unused) {
1278 inode = list_entry(act_head, struct inode, i_list);
1279 if (inode->i_sb == sb && !IS_NOQUOTA(inode))
1280 remove_inode_dquot_ref(inode, type, tofree_head);
1282 list_for_each(act_head, &sb->s_dirty) {
1283 inode = list_entry(act_head, struct inode, i_list);
1284 if (!IS_NOQUOTA(inode))
1285 remove_inode_dquot_ref(inode, type, tofree_head);
1287 list_for_each(act_head, &sb->s_io) {
1288 inode = list_entry(act_head, struct inode, i_list);
1289 if (!IS_NOQUOTA(inode))
1290 remove_inode_dquot_ref(inode, type, tofree_head);
1292 spin_unlock(&inode_lock);
1298 * Hashed waitqueues for wait_on_inode(). The table is pretty small - the
1299 * kernel doesn't lock many inodes at the same time.
1301 #define I_WAIT_TABLE_ORDER 3
1302 static struct i_wait_queue_head {
1303 wait_queue_head_t wqh;
1304 } ____cacheline_aligned_in_smp i_wait_queue_heads[1<<I_WAIT_TABLE_ORDER];
1307 * Return the address of the waitqueue_head to be used for this inode
1309 static wait_queue_head_t *i_waitq_head(struct inode *inode)
1311 return &i_wait_queue_heads[hash_ptr(inode, I_WAIT_TABLE_ORDER)].wqh;
1314 void __wait_on_inode(struct inode *inode)
1316 DECLARE_WAITQUEUE(wait, current);
1317 wait_queue_head_t *wq = i_waitq_head(inode);
1319 add_wait_queue(wq, &wait);
1321 set_current_state(TASK_UNINTERRUPTIBLE);
1322 if (inode->i_state & I_LOCK) {
1326 remove_wait_queue(wq, &wait);
1327 __set_current_state(TASK_RUNNING);
1331 * If we try to find an inode in the inode hash while it is being deleted, we
1332 * have to wait until the filesystem completes its deletion before reporting
1333 * that it isn't found. This is because iget will immediately call
1334 * ->read_inode, and we want to be sure that evidence of the deletion is found
1337 * This call might return early if an inode which shares the waitq is woken up.
1338 * This is most easily handled by the caller which will loop around again
1339 * looking for the inode.
1341 * This is called with inode_lock held.
1343 static void __wait_on_freeing_inode(struct inode *inode)
1345 DECLARE_WAITQUEUE(wait, current);
1346 wait_queue_head_t *wq = i_waitq_head(inode);
1348 add_wait_queue(wq, &wait);
1349 set_current_state(TASK_UNINTERRUPTIBLE);
1350 spin_unlock(&inode_lock);
1352 remove_wait_queue(wq, &wait);
1353 spin_lock(&inode_lock);
1356 void wake_up_inode(struct inode *inode)
1358 wait_queue_head_t *wq = i_waitq_head(inode);
1361 * Prevent speculative execution through spin_unlock(&inode_lock);
1364 if (waitqueue_active(wq))
1368 static __initdata unsigned long ihash_entries;
1369 static int __init set_ihash_entries(char *str)
1373 ihash_entries = simple_strtoul(str, &str, 0);
1376 __setup("ihash_entries=", set_ihash_entries);
1379 * Initialize the waitqueues and inode hash table.
1381 void __init inode_init_early(void)
1386 alloc_large_system_hash("Inode-cache",
1387 sizeof(struct hlist_head),
1394 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1395 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1398 void __init inode_init(unsigned long mempages)
1402 for (i = 0; i < ARRAY_SIZE(i_wait_queue_heads); i++)
1403 init_waitqueue_head(&i_wait_queue_heads[i].wqh);
1405 /* inode slab cache */
1406 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1407 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, init_once,
1409 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1412 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1414 inode->i_mode = mode;
1415 if (S_ISCHR(mode)) {
1416 inode->i_fop = &def_chr_fops;
1417 inode->i_rdev = rdev;
1418 } else if (S_ISBLK(mode)) {
1419 inode->i_fop = &def_blk_fops;
1420 inode->i_rdev = rdev;
1421 } else if (S_ISFIFO(mode))
1422 inode->i_fop = &def_fifo_fops;
1423 else if (S_ISSOCK(mode))
1424 inode->i_fop = &bad_sock_fops;
1426 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1429 EXPORT_SYMBOL(init_special_inode);