4 * (C) 1997 Linus Torvalds
7 #include <linux/config.h>
10 #include <linux/dcache.h>
11 #include <linux/init.h>
12 #include <linux/quotaops.h>
13 #include <linux/slab.h>
14 #include <linux/writeback.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/wait.h>
18 #include <linux/hash.h>
19 #include <linux/swap.h>
20 #include <linux/security.h>
21 #include <linux/pagemap.h>
22 #include <linux/cdev.h>
23 #include <linux/bootmem.h>
26 * This is needed for the following functions:
28 * - invalidate_inode_buffers
32 * FIXME: remove all knowledge of the buffer layer from this file
34 #include <linux/buffer_head.h>
37 * New inode.c implementation.
39 * This implementation has the basic premise of trying
40 * to be extremely low-overhead and SMP-safe, yet be
41 * simple enough to be "obviously correct".
46 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
48 /* #define INODE_PARANOIA 1 */
49 /* #define INODE_DEBUG 1 */
52 * Inode lookup is no longer as critical as it used to be:
53 * most of the lookups are going to be through the dcache.
55 #define I_HASHBITS i_hash_shift
56 #define I_HASHMASK i_hash_mask
58 static unsigned int i_hash_mask;
59 static unsigned int i_hash_shift;
62 * Each inode can be on two separate lists. One is
63 * the hash list of the inode, used for lookups. The
64 * other linked list is the "type" list:
65 * "in_use" - valid inode, i_count > 0, i_nlink > 0
66 * "dirty" - as "in_use" but also dirty
67 * "unused" - valid inode, i_count = 0
69 * A "dirty" list is maintained for each super block,
70 * allowing for low-overhead inode sync() operations.
73 LIST_HEAD(inode_in_use);
74 LIST_HEAD(inode_unused);
75 static struct hlist_head *inode_hashtable;
78 * A simple spinlock to protect the list manipulations.
80 * NOTE! You also have to own the lock if you change
81 * the i_state of an inode while it is in use..
83 spinlock_t inode_lock = SPIN_LOCK_UNLOCKED;
86 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
87 * icache shrinking path, and the umount path. Without this exclusion,
88 * by the time prune_icache calls iput for the inode whose pages it has
89 * been invalidating, or by the time it calls clear_inode & destroy_inode
90 * from its final dispose_list, the struct super_block they refer to
91 * (for inode->i_sb->s_op) may already have been freed and reused.
93 DECLARE_MUTEX(iprune_sem);
96 * Statistics gathering..
98 struct inodes_stat_t inodes_stat;
100 static kmem_cache_t * inode_cachep;
102 static void prune_icache(int nr_to_scan);
105 #define INODE_UNUSED_THRESHOLD 15000
106 #define PRUNE_BATCH_COUNT 32
108 void try_to_clip_inodes(void)
110 unsigned long count = 0;
111 /* if there are a LOT of unused inodes in cache, better shrink a few first */
113 /* check lockless first to not take the lock always here; racing occasionally isn't a big deal */
114 if (inodes_stat.nr_unused > INODE_UNUSED_THRESHOLD) {
115 spin_lock(&inode_lock);
116 if (inodes_stat.nr_unused > INODE_UNUSED_THRESHOLD)
117 count = inodes_stat.nr_unused - INODE_UNUSED_THRESHOLD;
118 spin_unlock(&inode_lock);
125 static struct inode *alloc_inode(struct super_block *sb)
127 static struct address_space_operations empty_aops;
128 static struct inode_operations empty_iops;
129 static struct file_operations empty_fops;
132 if (sb->s_op->alloc_inode)
133 inode = sb->s_op->alloc_inode(sb);
135 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
138 struct address_space * const mapping = &inode->i_data;
141 if (sb->s_flags & MS_TAGXID)
142 inode->i_xid = current->xid;
144 inode->i_xid = 0; /* maybe xid -1 would be better? */
145 // inode->i_dqh = dqhget(sb->s_dqh);
146 inode->i_blkbits = sb->s_blocksize_bits;
148 atomic_set(&inode->i_count, 1);
150 inode->i_op = &empty_iops;
151 inode->i_fop = &empty_fops;
153 atomic_set(&inode->i_writecount, 0);
157 inode->i_generation = 0;
159 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
161 inode->i_pipe = NULL;
162 inode->i_bdev = NULL;
163 inode->i_cdev = NULL;
165 // inode->i_xid = 0; /* maybe not too wise ... */
166 inode->i_security = NULL;
167 inode->dirtied_when = 0;
168 if (security_inode_alloc(inode)) {
169 if (inode->i_sb->s_op->destroy_inode)
170 inode->i_sb->s_op->destroy_inode(inode);
172 kmem_cache_free(inode_cachep, (inode));
176 mapping->a_ops = &empty_aops;
177 mapping->host = inode;
179 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
180 mapping->assoc_mapping = NULL;
181 mapping->backing_dev_info = &default_backing_dev_info;
184 * If the block_device provides a backing_dev_info for client
185 * inodes then use that. Otherwise the inode share the bdev's
189 struct backing_dev_info *bdi;
191 bdi = sb->s_bdev->bd_inode_backing_dev_info;
193 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
194 mapping->backing_dev_info = bdi;
196 memset(&inode->u, 0, sizeof(inode->u));
197 inode->i_mapping = mapping;
202 void destroy_inode(struct inode *inode)
204 if (inode_has_buffers(inode))
206 security_inode_free(inode);
207 if (inode->i_sb->s_op->destroy_inode)
208 inode->i_sb->s_op->destroy_inode(inode);
210 kmem_cache_free(inode_cachep, (inode));
215 * These are initializations that only need to be done
216 * once, because the fields are idempotent across use
217 * of the inode, so let the slab aware of that.
219 void inode_init_once(struct inode *inode)
221 memset(inode, 0, sizeof(*inode));
222 INIT_HLIST_NODE(&inode->i_hash);
223 INIT_LIST_HEAD(&inode->i_dentry);
224 INIT_LIST_HEAD(&inode->i_devices);
225 sema_init(&inode->i_sem, 1);
226 init_rwsem(&inode->i_alloc_sem);
227 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
228 spin_lock_init(&inode->i_data.tree_lock);
229 spin_lock_init(&inode->i_data.i_mmap_lock);
230 atomic_set(&inode->i_data.truncate_count, 0);
231 INIT_LIST_HEAD(&inode->i_data.private_list);
232 spin_lock_init(&inode->i_data.private_lock);
233 INIT_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
234 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
235 spin_lock_init(&inode->i_lock);
236 i_size_ordered_init(inode);
239 EXPORT_SYMBOL(inode_init_once);
241 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
243 struct inode * inode = (struct inode *) foo;
245 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
246 SLAB_CTOR_CONSTRUCTOR)
247 inode_init_once(inode);
251 * inode_lock must be held
253 void __iget(struct inode * inode)
255 if (atomic_read(&inode->i_count)) {
256 atomic_inc(&inode->i_count);
259 atomic_inc(&inode->i_count);
260 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
261 list_move(&inode->i_list, &inode_in_use);
262 inodes_stat.nr_unused--;
266 * clear_inode - clear an inode
267 * @inode: inode to clear
269 * This is called by the filesystem to tell us
270 * that the inode is no longer useful. We just
271 * terminate it with extreme prejudice.
273 void clear_inode(struct inode *inode)
276 invalidate_inode_buffers(inode);
278 if (inode->i_data.nrpages)
280 if (!(inode->i_state & I_FREEING))
282 if (inode->i_state & I_CLEAR)
284 wait_on_inode(inode);
286 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
287 inode->i_sb->s_op->clear_inode(inode);
292 inode->i_state = I_CLEAR;
295 EXPORT_SYMBOL(clear_inode);
298 * dispose_list - dispose of the contents of a local list
299 * @head: the head of the list to free
301 * Dispose-list gets a local list with local inodes in it, so it doesn't
302 * need to worry about list corruption and SMP locks.
304 static void dispose_list(struct list_head *head)
308 while (!list_empty(head)) {
311 inode = list_entry(head->next, struct inode, i_list);
312 list_del(&inode->i_list);
314 if (inode->i_data.nrpages)
315 truncate_inode_pages(&inode->i_data, 0);
317 destroy_inode(inode);
320 spin_lock(&inode_lock);
321 inodes_stat.nr_inodes -= nr_disposed;
322 spin_unlock(&inode_lock);
326 * Invalidate all inodes for a device.
328 static int invalidate_list(struct list_head *head, struct super_block * sb, struct list_head * dispose)
330 struct list_head *next;
331 int busy = 0, count = 0;
335 struct list_head * tmp = next;
336 struct inode * inode;
341 inode = list_entry(tmp, struct inode, i_list);
342 if (inode->i_sb != sb)
344 invalidate_inode_buffers(inode);
345 if (!atomic_read(&inode->i_count)) {
346 hlist_del_init(&inode->i_hash);
347 list_move(&inode->i_list, dispose);
348 inode->i_state |= I_FREEING;
354 /* only unused inodes may be cached with i_count zero */
355 inodes_stat.nr_unused -= count;
360 * This is a two-stage process. First we collect all
361 * offending inodes onto the throw-away list, and in
362 * the second stage we actually dispose of them. This
363 * is because we don't want to sleep while messing
364 * with the global lists..
368 * invalidate_inodes - discard the inodes on a device
371 * Discard all of the inodes for a given superblock. If the discard
372 * fails because there are busy inodes then a non zero value is returned.
373 * If the discard is successful all the inodes have been discarded.
375 int invalidate_inodes(struct super_block * sb)
378 LIST_HEAD(throw_away);
381 spin_lock(&inode_lock);
382 busy = invalidate_list(&inode_in_use, sb, &throw_away);
383 busy |= invalidate_list(&inode_unused, sb, &throw_away);
384 busy |= invalidate_list(&sb->s_dirty, sb, &throw_away);
385 busy |= invalidate_list(&sb->s_io, sb, &throw_away);
386 spin_unlock(&inode_lock);
388 dispose_list(&throw_away);
394 EXPORT_SYMBOL(invalidate_inodes);
396 int __invalidate_device(struct block_device *bdev, int do_sync)
398 struct super_block *sb;
405 sb = get_super(bdev);
408 * no need to lock the super, get_super holds the
409 * read semaphore so the filesystem cannot go away
410 * under us (->put_super runs with the write lock
413 shrink_dcache_sb(sb);
414 res = invalidate_inodes(sb);
417 invalidate_bdev(bdev, 0);
421 EXPORT_SYMBOL(__invalidate_device);
423 static int can_unuse(struct inode *inode)
427 if (inode_has_buffers(inode))
429 if (atomic_read(&inode->i_count))
431 if (inode->i_data.nrpages)
437 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
438 * a temporary list and then are freed outside inode_lock by dispose_list().
440 * Any inodes which are pinned purely because of attached pagecache have their
441 * pagecache removed. We expect the final iput() on that inode to add it to
442 * the front of the inode_unused list. So look for it there and if the
443 * inode is still freeable, proceed. The right inode is found 99.9% of the
444 * time in testing on a 4-way.
446 * If the inode has metadata buffers attached to mapping->private_list then
447 * try to remove them.
449 static void prune_icache(int nr_to_scan)
454 unsigned long reap = 0;
457 spin_lock(&inode_lock);
458 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
461 if (list_empty(&inode_unused))
464 inode = list_entry(inode_unused.prev, struct inode, i_list);
466 if (inode->i_state || atomic_read(&inode->i_count)) {
467 list_move(&inode->i_list, &inode_unused);
470 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
472 spin_unlock(&inode_lock);
473 if (remove_inode_buffers(inode))
474 reap += invalidate_inode_pages(&inode->i_data);
476 spin_lock(&inode_lock);
478 if (inode != list_entry(inode_unused.next,
479 struct inode, i_list))
480 continue; /* wrong inode or list_empty */
481 if (!can_unuse(inode))
484 hlist_del_init(&inode->i_hash);
485 list_move(&inode->i_list, &freeable);
486 inode->i_state |= I_FREEING;
489 inodes_stat.nr_unused -= nr_pruned;
490 spin_unlock(&inode_lock);
492 dispose_list(&freeable);
495 if (current_is_kswapd())
496 mod_page_state(kswapd_inodesteal, reap);
498 mod_page_state(pginodesteal, reap);
502 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
503 * "unused" means that no dentries are referring to the inodes: the files are
504 * not open and the dcache references to those inodes have already been
507 * This function is passed the number of inodes to scan, and it returns the
508 * total number of remaining possibly-reclaimable inodes.
510 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
514 * Nasty deadlock avoidance. We may hold various FS locks,
515 * and we don't want to recurse into the FS that called us
516 * in clear_inode() and friends..
518 if (gfp_mask & __GFP_FS)
521 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
524 static void __wait_on_freeing_inode(struct inode *inode);
526 * Called with the inode lock held.
527 * NOTE: we are not increasing the inode-refcount, you must call __iget()
528 * by hand after calling find_inode now! This simplifies iunique and won't
529 * add any additional branch in the common code.
531 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
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_sb != sb)
541 if (!test(inode, data))
543 if (inode->i_state & (I_FREEING|I_CLEAR)) {
544 __wait_on_freeing_inode(inode);
549 return node ? inode : NULL;
553 * find_inode_fast is the fast path version of find_inode, see the comment at
554 * iget_locked for details.
556 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
558 struct hlist_node *node;
559 struct inode * inode = NULL;
562 hlist_for_each (node, head) {
563 inode = hlist_entry(node, struct inode, i_hash);
564 if (inode->i_ino != ino)
566 if (inode->i_sb != sb)
568 if (inode->i_state & (I_FREEING|I_CLEAR)) {
569 __wait_on_freeing_inode(inode);
574 return node ? inode : NULL;
578 * new_inode - obtain an inode
581 * Allocates a new inode for given superblock.
583 struct inode *new_inode(struct super_block *sb)
585 static unsigned long last_ino;
586 struct inode * inode;
588 spin_lock_prefetch(&inode_lock);
590 inode = alloc_inode(sb);
592 spin_lock(&inode_lock);
593 inodes_stat.nr_inodes++;
594 list_add(&inode->i_list, &inode_in_use);
595 inode->i_ino = ++last_ino;
597 spin_unlock(&inode_lock);
602 EXPORT_SYMBOL(new_inode);
604 void unlock_new_inode(struct inode *inode)
607 * This is special! We do not need the spinlock
608 * when clearing I_LOCK, because we're guaranteed
609 * that nobody else tries to do anything about the
610 * state of the inode when it is locked, as we
611 * just created it (so there can be no old holders
612 * that haven't tested I_LOCK).
614 inode->i_state &= ~(I_LOCK|I_NEW);
615 wake_up_inode(inode);
618 EXPORT_SYMBOL(unlock_new_inode);
621 * This is called without the inode lock held.. Be careful.
623 * We no longer cache the sb_flags in i_flags - see fs.h
624 * -- rmk@arm.uk.linux.org
626 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)
628 struct inode * inode;
630 inode = alloc_inode(sb);
634 spin_lock(&inode_lock);
635 /* We released the lock, so.. */
636 old = find_inode(sb, head, test, data);
638 if (set(inode, data))
641 inodes_stat.nr_inodes++;
642 list_add(&inode->i_list, &inode_in_use);
643 hlist_add_head(&inode->i_hash, head);
644 inode->i_state = I_LOCK|I_NEW;
645 spin_unlock(&inode_lock);
647 /* Return the locked inode with I_NEW set, the
648 * caller is responsible for filling in the contents
654 * Uhhuh, somebody else created the same inode under
655 * us. Use the old inode instead of the one we just
659 spin_unlock(&inode_lock);
660 destroy_inode(inode);
662 wait_on_inode(inode);
667 spin_unlock(&inode_lock);
668 destroy_inode(inode);
673 * get_new_inode_fast is the fast path version of get_new_inode, see the
674 * comment at iget_locked for details.
676 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
678 struct inode * inode;
680 inode = alloc_inode(sb);
684 spin_lock(&inode_lock);
685 /* We released the lock, so.. */
686 old = find_inode_fast(sb, head, ino);
689 inodes_stat.nr_inodes++;
690 list_add(&inode->i_list, &inode_in_use);
691 hlist_add_head(&inode->i_hash, head);
692 inode->i_state = I_LOCK|I_NEW;
693 spin_unlock(&inode_lock);
695 /* Return the locked inode with I_NEW set, the
696 * caller is responsible for filling in the contents
702 * Uhhuh, somebody else created the same inode under
703 * us. Use the old inode instead of the one we just
707 spin_unlock(&inode_lock);
708 destroy_inode(inode);
710 wait_on_inode(inode);
715 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
719 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
721 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
722 return tmp & I_HASHMASK;
726 * iunique - get a unique inode number
728 * @max_reserved: highest reserved inode number
730 * Obtain an inode number that is unique on the system for a given
731 * superblock. This is used by file systems that have no natural
732 * permanent inode numbering system. An inode number is returned that
733 * is higher than the reserved limit but unique.
736 * With a large number of inodes live on the file system this function
737 * currently becomes quite slow.
739 ino_t iunique(struct super_block *sb, ino_t max_reserved)
741 static ino_t counter;
743 struct hlist_head * head;
745 spin_lock(&inode_lock);
747 if (counter > max_reserved) {
748 head = inode_hashtable + hash(sb,counter);
750 inode = find_inode_fast(sb, head, res);
752 spin_unlock(&inode_lock);
756 counter = max_reserved + 1;
762 EXPORT_SYMBOL(iunique);
764 struct inode *igrab(struct inode *inode)
766 spin_lock(&inode_lock);
767 if (!(inode->i_state & I_FREEING))
771 * Handle the case where s_op->clear_inode is not been
772 * called yet, and somebody is calling igrab
773 * while the inode is getting freed.
776 spin_unlock(&inode_lock);
780 EXPORT_SYMBOL(igrab);
783 * ifind - internal function, you want ilookup5() or iget5().
784 * @sb: super block of file system to search
785 * @head: the head of the list to search
786 * @test: callback used for comparisons between inodes
787 * @data: opaque data pointer to pass to @test
789 * ifind() searches for the inode specified by @data in the inode
790 * cache. This is a generalized version of ifind_fast() for file systems where
791 * the inode number is not sufficient for unique identification of an inode.
793 * If the inode is in the cache, the inode is returned with an incremented
796 * Otherwise NULL is returned.
798 * Note, @test is called with the inode_lock held, so can't sleep.
800 static inline struct inode *ifind(struct super_block *sb,
801 struct hlist_head *head, int (*test)(struct inode *, void *),
806 spin_lock(&inode_lock);
807 inode = find_inode(sb, head, test, data);
810 spin_unlock(&inode_lock);
811 wait_on_inode(inode);
814 spin_unlock(&inode_lock);
819 * ifind_fast - internal function, you want ilookup() or iget().
820 * @sb: super block of file system to search
821 * @head: head of the list to search
822 * @ino: inode number to search for
824 * ifind_fast() searches for the inode @ino in the inode cache. This is for
825 * file systems where the inode number is sufficient for unique identification
828 * If the inode is in the cache, the inode is returned with an incremented
831 * Otherwise NULL is returned.
833 static inline struct inode *ifind_fast(struct super_block *sb,
834 struct hlist_head *head, unsigned long ino)
838 spin_lock(&inode_lock);
839 inode = find_inode_fast(sb, head, ino);
842 spin_unlock(&inode_lock);
843 wait_on_inode(inode);
846 spin_unlock(&inode_lock);
851 * ilookup5 - search for an inode in the inode cache
852 * @sb: super block of file system to search
853 * @hashval: hash value (usually inode number) to search for
854 * @test: callback used for comparisons between inodes
855 * @data: opaque data pointer to pass to @test
857 * ilookup5() uses ifind() to search for the inode specified by @hashval and
858 * @data in the inode cache. This is a generalized version of ilookup() for
859 * file systems where the inode number is not sufficient for unique
860 * identification of an inode.
862 * If the inode is in the cache, the inode is returned with an incremented
865 * Otherwise NULL is returned.
867 * Note, @test is called with the inode_lock held, so can't sleep.
869 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
870 int (*test)(struct inode *, void *), void *data)
872 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
874 return ifind(sb, head, test, data);
877 EXPORT_SYMBOL(ilookup5);
880 * ilookup - search for an inode in the inode cache
881 * @sb: super block of file system to search
882 * @ino: inode number to search for
884 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
885 * This is for file systems where the inode number is sufficient for unique
886 * identification of an inode.
888 * If the inode is in the cache, the inode is returned with an incremented
891 * Otherwise NULL is returned.
893 struct inode *ilookup(struct super_block *sb, unsigned long ino)
895 struct hlist_head *head = inode_hashtable + hash(sb, ino);
897 return ifind_fast(sb, head, ino);
900 EXPORT_SYMBOL(ilookup);
903 * iget5_locked - obtain an inode from a mounted file system
904 * @sb: super block of file system
905 * @hashval: hash value (usually inode number) to get
906 * @test: callback used for comparisons between inodes
907 * @set: callback used to initialize a new struct inode
908 * @data: opaque data pointer to pass to @test and @set
910 * This is iget() without the read_inode() portion of get_new_inode().
912 * iget5_locked() uses ifind() to search for the inode specified by @hashval
913 * and @data in the inode cache and if present it is returned with an increased
914 * reference count. This is a generalized version of iget_locked() for file
915 * systems where the inode number is not sufficient for unique identification
918 * If the inode is not in cache, get_new_inode() is called to allocate a new
919 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
920 * file system gets to fill it in before unlocking it via unlock_new_inode().
922 * Note both @test and @set are called with the inode_lock held, so can't sleep.
924 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
925 int (*test)(struct inode *, void *),
926 int (*set)(struct inode *, void *), void *data)
928 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
931 inode = ifind(sb, head, test, data);
935 * get_new_inode() will do the right thing, re-trying the search
936 * in case it had to block at any point.
938 return get_new_inode(sb, head, test, set, data);
941 EXPORT_SYMBOL(iget5_locked);
944 * iget_locked - obtain an inode from a mounted file system
945 * @sb: super block of file system
946 * @ino: inode number to get
948 * This is iget() without the read_inode() portion of get_new_inode_fast().
950 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
951 * the inode cache and if present it is returned with an increased reference
952 * count. This is for file systems where the inode number is sufficient for
953 * unique identification of an inode.
955 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
956 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
957 * The file system gets to fill it in before unlocking it via
958 * unlock_new_inode().
960 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
962 struct hlist_head *head = inode_hashtable + hash(sb, ino);
965 inode = ifind_fast(sb, head, ino);
969 * get_new_inode_fast() will do the right thing, re-trying the search
970 * in case it had to block at any point.
972 return get_new_inode_fast(sb, head, ino);
975 EXPORT_SYMBOL(iget_locked);
978 * __insert_inode_hash - hash an inode
979 * @inode: unhashed inode
980 * @hashval: unsigned long value used to locate this object in the
983 * Add an inode to the inode hash for this superblock.
985 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
987 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
988 spin_lock(&inode_lock);
989 hlist_add_head(&inode->i_hash, head);
990 spin_unlock(&inode_lock);
993 EXPORT_SYMBOL(__insert_inode_hash);
996 * remove_inode_hash - remove an inode from the hash
997 * @inode: inode to unhash
999 * Remove an inode from the superblock.
1001 void remove_inode_hash(struct inode *inode)
1003 spin_lock(&inode_lock);
1004 hlist_del_init(&inode->i_hash);
1005 spin_unlock(&inode_lock);
1008 EXPORT_SYMBOL(remove_inode_hash);
1011 * Tell the filesystem that this inode is no longer of any interest and should
1012 * be completely destroyed.
1014 * We leave the inode in the inode hash table until *after* the filesystem's
1015 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1016 * instigate) will always find up-to-date information either in the hash or on
1019 * I_FREEING is set so that no-one will take a new reference to the inode while
1020 * it is being deleted.
1022 void generic_delete_inode(struct inode *inode)
1024 struct super_operations *op = inode->i_sb->s_op;
1026 list_del_init(&inode->i_list);
1027 inode->i_state|=I_FREEING;
1028 inodes_stat.nr_inodes--;
1029 spin_unlock(&inode_lock);
1031 if (inode->i_data.nrpages)
1032 truncate_inode_pages(&inode->i_data, 0);
1034 security_inode_delete(inode);
1036 if (op->delete_inode) {
1037 void (*delete)(struct inode *) = op->delete_inode;
1038 if (!is_bad_inode(inode))
1040 /* s_op->delete_inode internally recalls clear_inode() */
1044 spin_lock(&inode_lock);
1045 hlist_del_init(&inode->i_hash);
1046 spin_unlock(&inode_lock);
1047 wake_up_inode(inode);
1048 if (inode->i_state != I_CLEAR)
1050 destroy_inode(inode);
1053 EXPORT_SYMBOL(generic_delete_inode);
1055 static void generic_forget_inode(struct inode *inode)
1057 struct super_block *sb = inode->i_sb;
1059 if (!hlist_unhashed(&inode->i_hash)) {
1060 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1061 list_move(&inode->i_list, &inode_unused);
1062 inodes_stat.nr_unused++;
1063 spin_unlock(&inode_lock);
1064 if (!sb || (sb->s_flags & MS_ACTIVE))
1066 write_inode_now(inode, 1);
1067 spin_lock(&inode_lock);
1068 inodes_stat.nr_unused--;
1069 hlist_del_init(&inode->i_hash);
1071 list_del_init(&inode->i_list);
1072 inode->i_state|=I_FREEING;
1073 inodes_stat.nr_inodes--;
1074 spin_unlock(&inode_lock);
1075 if (inode->i_data.nrpages)
1076 truncate_inode_pages(&inode->i_data, 0);
1078 destroy_inode(inode);
1082 * Normal UNIX filesystem behaviour: delete the
1083 * inode when the usage count drops to zero, and
1086 static void generic_drop_inode(struct inode *inode)
1088 if (!inode->i_nlink)
1089 generic_delete_inode(inode);
1091 generic_forget_inode(inode);
1095 * Called when we're dropping the last reference
1098 * Call the FS "drop()" function, defaulting to
1099 * the legacy UNIX filesystem behaviour..
1101 * NOTE! NOTE! NOTE! We're called with the inode lock
1102 * held, and the drop function is supposed to release
1105 static inline void iput_final(struct inode *inode)
1107 struct super_operations *op = inode->i_sb->s_op;
1108 void (*drop)(struct inode *) = generic_drop_inode;
1110 if (op && op->drop_inode)
1111 drop = op->drop_inode;
1116 * iput - put an inode
1117 * @inode: inode to put
1119 * Puts an inode, dropping its usage count. If the inode use count hits
1120 * zero the inode is also then freed and may be destroyed.
1122 void iput(struct inode *inode)
1125 struct super_operations *op = inode->i_sb->s_op;
1127 if (inode->i_state == I_CLEAR)
1130 if (op && op->put_inode)
1131 op->put_inode(inode);
1133 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1138 EXPORT_SYMBOL(iput);
1141 * bmap - find a block number in a file
1142 * @inode: inode of file
1143 * @block: block to find
1145 * Returns the block number on the device holding the inode that
1146 * is the disk block number for the block of the file requested.
1147 * That is, asked for block 4 of inode 1 the function will return the
1148 * disk block relative to the disk start that holds that block of the
1151 sector_t bmap(struct inode * inode, sector_t block)
1154 if (inode->i_mapping->a_ops->bmap)
1155 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1159 EXPORT_SYMBOL(bmap);
1162 * Return true if the filesystem which backs this inode considers the two
1163 * passed timespecs to be sufficiently different to warrant flushing the
1164 * altered time out to disk.
1166 static int inode_times_differ(struct inode *inode,
1167 struct timespec *old, struct timespec *new)
1169 if (IS_ONE_SECOND(inode))
1170 return old->tv_sec != new->tv_sec;
1171 return !timespec_equal(old, new);
1175 * update_atime - update the access time
1176 * @inode: inode accessed
1178 * Update the accessed time on an inode and mark it for writeback.
1179 * This function automatically handles read only file systems and media,
1180 * as well as the "noatime" flag and inode specific "noatime" markers.
1182 void update_atime(struct inode *inode)
1184 struct timespec now;
1186 if (IS_NOATIME(inode))
1188 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1190 if (IS_RDONLY(inode))
1193 now = current_kernel_time();
1194 if (inode_times_differ(inode, &inode->i_atime, &now)) {
1195 inode->i_atime = now;
1196 mark_inode_dirty_sync(inode);
1198 if (!timespec_equal(&inode->i_atime, &now))
1199 inode->i_atime = now;
1203 EXPORT_SYMBOL(update_atime);
1206 * inode_update_time - update mtime and ctime time
1207 * @inode: inode accessed
1208 * @ctime_too: update ctime too
1210 * Update the mtime time on an inode and mark it for writeback.
1211 * When ctime_too is specified update the ctime too.
1214 void inode_update_time(struct inode *inode, int ctime_too)
1216 struct timespec now;
1219 if (IS_NOCMTIME(inode))
1221 if (IS_RDONLY(inode))
1224 now = current_kernel_time();
1226 if (inode_times_differ(inode, &inode->i_mtime, &now))
1228 inode->i_mtime = now;
1231 if (inode_times_differ(inode, &inode->i_ctime, &now))
1233 inode->i_ctime = now;
1236 mark_inode_dirty_sync(inode);
1239 EXPORT_SYMBOL(inode_update_time);
1241 int inode_needs_sync(struct inode *inode)
1245 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1250 EXPORT_SYMBOL(inode_needs_sync);
1253 * Quota functions that want to walk the inode lists..
1257 /* Function back in dquot.c */
1258 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1260 void remove_dquot_ref(struct super_block *sb, int type, struct list_head *tofree_head)
1262 struct inode *inode;
1263 struct list_head *act_head;
1266 return; /* nothing to do */
1267 spin_lock(&inode_lock); /* This lock is for inodes code */
1269 /* We hold dqptr_sem so we are safe against the quota code */
1270 list_for_each(act_head, &inode_in_use) {
1271 inode = list_entry(act_head, struct inode, i_list);
1272 if (inode->i_sb == sb && !IS_NOQUOTA(inode))
1273 remove_inode_dquot_ref(inode, type, tofree_head);
1275 list_for_each(act_head, &inode_unused) {
1276 inode = list_entry(act_head, struct inode, i_list);
1277 if (inode->i_sb == sb && !IS_NOQUOTA(inode))
1278 remove_inode_dquot_ref(inode, type, tofree_head);
1280 list_for_each(act_head, &sb->s_dirty) {
1281 inode = list_entry(act_head, struct inode, i_list);
1282 if (!IS_NOQUOTA(inode))
1283 remove_inode_dquot_ref(inode, type, tofree_head);
1285 list_for_each(act_head, &sb->s_io) {
1286 inode = list_entry(act_head, struct inode, i_list);
1287 if (!IS_NOQUOTA(inode))
1288 remove_inode_dquot_ref(inode, type, tofree_head);
1290 spin_unlock(&inode_lock);
1296 * Hashed waitqueues for wait_on_inode(). The table is pretty small - the
1297 * kernel doesn't lock many inodes at the same time.
1299 #define I_WAIT_TABLE_ORDER 3
1300 static struct i_wait_queue_head {
1301 wait_queue_head_t wqh;
1302 } ____cacheline_aligned_in_smp i_wait_queue_heads[1<<I_WAIT_TABLE_ORDER];
1305 * Return the address of the waitqueue_head to be used for this inode
1307 static wait_queue_head_t *i_waitq_head(struct inode *inode)
1309 return &i_wait_queue_heads[hash_ptr(inode, I_WAIT_TABLE_ORDER)].wqh;
1312 void __wait_on_inode(struct inode *inode)
1314 DECLARE_WAITQUEUE(wait, current);
1315 wait_queue_head_t *wq = i_waitq_head(inode);
1317 add_wait_queue(wq, &wait);
1319 set_current_state(TASK_UNINTERRUPTIBLE);
1320 if (inode->i_state & I_LOCK) {
1324 remove_wait_queue(wq, &wait);
1325 __set_current_state(TASK_RUNNING);
1329 * If we try to find an inode in the inode hash while it is being deleted, we
1330 * have to wait until the filesystem completes its deletion before reporting
1331 * that it isn't found. This is because iget will immediately call
1332 * ->read_inode, and we want to be sure that evidence of the deletion is found
1335 * This call might return early if an inode which shares the waitq is woken up.
1336 * This is most easily handled by the caller which will loop around again
1337 * looking for the inode.
1339 * This is called with inode_lock held.
1341 static void __wait_on_freeing_inode(struct inode *inode)
1343 DECLARE_WAITQUEUE(wait, current);
1344 wait_queue_head_t *wq = i_waitq_head(inode);
1346 add_wait_queue(wq, &wait);
1347 set_current_state(TASK_UNINTERRUPTIBLE);
1348 spin_unlock(&inode_lock);
1350 remove_wait_queue(wq, &wait);
1351 spin_lock(&inode_lock);
1354 void wake_up_inode(struct inode *inode)
1356 wait_queue_head_t *wq = i_waitq_head(inode);
1359 * Prevent speculative execution through spin_unlock(&inode_lock);
1362 if (waitqueue_active(wq))
1366 static __initdata unsigned long ihash_entries;
1367 static int __init set_ihash_entries(char *str)
1371 ihash_entries = simple_strtoul(str, &str, 0);
1374 __setup("ihash_entries=", set_ihash_entries);
1377 * Initialize the waitqueues and inode hash table.
1379 void __init inode_init_early(void)
1384 alloc_large_system_hash("Inode-cache",
1385 sizeof(struct hlist_head),
1392 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1393 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1396 void __init inode_init(unsigned long mempages)
1400 for (i = 0; i < ARRAY_SIZE(i_wait_queue_heads); i++)
1401 init_waitqueue_head(&i_wait_queue_heads[i].wqh);
1403 /* inode slab cache */
1404 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1405 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, init_once,
1407 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1410 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1412 inode->i_mode = mode;
1413 if (S_ISCHR(mode)) {
1414 inode->i_fop = &def_chr_fops;
1415 inode->i_rdev = rdev;
1416 } else if (S_ISBLK(mode)) {
1417 inode->i_fop = &def_blk_fops;
1418 inode->i_rdev = rdev;
1419 } else if (S_ISFIFO(mode))
1420 inode->i_fop = &def_fifo_fops;
1421 else if (S_ISSOCK(mode))
1422 inode->i_fop = &bad_sock_fops;
1424 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1427 EXPORT_SYMBOL(init_special_inode);