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 DEFINE_SPINLOCK(inode_lock);
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 struct inode *alloc_inode(struct super_block *sb)
104 static struct address_space_operations empty_aops;
105 static struct inode_operations empty_iops;
106 static struct file_operations empty_fops;
109 if (sb->s_op->alloc_inode)
110 inode = sb->s_op->alloc_inode(sb);
112 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
115 struct address_space * const mapping = &inode->i_data;
118 // inode->i_dqh = dqhget(sb->s_dqh);
120 /* essential because of inode slab reuse */
122 inode->i_blkbits = sb->s_blocksize_bits;
124 atomic_set(&inode->i_count, 1);
126 inode->i_op = &empty_iops;
127 inode->i_fop = &empty_fops;
129 atomic_set(&inode->i_writecount, 0);
133 inode->i_generation = 0;
135 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
137 inode->i_pipe = NULL;
138 inode->i_bdev = NULL;
139 inode->i_cdev = NULL;
141 inode->i_security = NULL;
142 inode->dirtied_when = 0;
143 if (security_inode_alloc(inode)) {
144 if (inode->i_sb->s_op->destroy_inode)
145 inode->i_sb->s_op->destroy_inode(inode);
147 kmem_cache_free(inode_cachep, (inode));
151 mapping->a_ops = &empty_aops;
152 mapping->host = inode;
154 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
155 mapping->assoc_mapping = NULL;
156 mapping->backing_dev_info = &default_backing_dev_info;
159 * If the block_device provides a backing_dev_info for client
160 * inodes then use that. Otherwise the inode share the bdev's
164 struct backing_dev_info *bdi;
166 bdi = sb->s_bdev->bd_inode_backing_dev_info;
168 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
169 mapping->backing_dev_info = bdi;
171 memset(&inode->u, 0, sizeof(inode->u));
172 inode->i_mapping = mapping;
177 void destroy_inode(struct inode *inode)
179 if (inode_has_buffers(inode))
181 security_inode_free(inode);
182 if (inode->i_sb->s_op->destroy_inode)
183 inode->i_sb->s_op->destroy_inode(inode);
185 kmem_cache_free(inode_cachep, (inode));
190 * These are initializations that only need to be done
191 * once, because the fields are idempotent across use
192 * of the inode, so let the slab aware of that.
194 void inode_init_once(struct inode *inode)
196 memset(inode, 0, sizeof(*inode));
197 INIT_HLIST_NODE(&inode->i_hash);
198 INIT_LIST_HEAD(&inode->i_dentry);
199 INIT_LIST_HEAD(&inode->i_devices);
200 sema_init(&inode->i_sem, 1);
201 init_rwsem(&inode->i_alloc_sem);
202 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
203 spin_lock_init(&inode->i_data.tree_lock);
204 spin_lock_init(&inode->i_data.i_mmap_lock);
205 INIT_LIST_HEAD(&inode->i_data.private_list);
206 spin_lock_init(&inode->i_data.private_lock);
207 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
208 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
209 spin_lock_init(&inode->i_lock);
210 i_size_ordered_init(inode);
213 EXPORT_SYMBOL(inode_init_once);
215 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
217 struct inode * inode = (struct inode *) foo;
219 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
220 SLAB_CTOR_CONSTRUCTOR)
221 inode_init_once(inode);
225 * inode_lock must be held
227 void __iget(struct inode * inode)
229 if (atomic_read(&inode->i_count)) {
230 atomic_inc(&inode->i_count);
233 atomic_inc(&inode->i_count);
234 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
235 list_move(&inode->i_list, &inode_in_use);
236 inodes_stat.nr_unused--;
240 * clear_inode - clear an inode
241 * @inode: inode to clear
243 * This is called by the filesystem to tell us
244 * that the inode is no longer useful. We just
245 * terminate it with extreme prejudice.
247 void clear_inode(struct inode *inode)
250 invalidate_inode_buffers(inode);
252 if (inode->i_data.nrpages)
254 if (!(inode->i_state & I_FREEING))
256 if (inode->i_state & I_CLEAR)
258 wait_on_inode(inode);
260 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
261 inode->i_sb->s_op->clear_inode(inode);
266 inode->i_state = I_CLEAR;
269 EXPORT_SYMBOL(clear_inode);
272 * dispose_list - dispose of the contents of a local list
273 * @head: the head of the list to free
275 * Dispose-list gets a local list with local inodes in it, so it doesn't
276 * need to worry about list corruption and SMP locks.
278 static void dispose_list(struct list_head *head)
282 while (!list_empty(head)) {
285 inode = list_entry(head->next, struct inode, i_list);
286 list_del(&inode->i_list);
288 if (inode->i_data.nrpages)
289 truncate_inode_pages(&inode->i_data, 0);
291 destroy_inode(inode);
294 spin_lock(&inode_lock);
295 inodes_stat.nr_inodes -= nr_disposed;
296 spin_unlock(&inode_lock);
300 * Invalidate all inodes for a device.
302 static int invalidate_list(struct list_head *head, struct list_head *dispose)
304 struct list_head *next;
305 int busy = 0, count = 0;
309 struct list_head * tmp = next;
310 struct inode * inode;
313 * We can reschedule here without worrying about the list's
314 * consistency because the per-sb list of inodes must not
315 * change during umount anymore, and because iprune_sem keeps
316 * shrink_icache_memory() away.
318 cond_resched_lock(&inode_lock);
323 inode = list_entry(tmp, struct inode, i_sb_list);
324 invalidate_inode_buffers(inode);
325 if (!atomic_read(&inode->i_count)) {
326 hlist_del_init(&inode->i_hash);
327 list_del(&inode->i_sb_list);
328 list_move(&inode->i_list, dispose);
329 inode->i_state |= I_FREEING;
335 /* only unused inodes may be cached with i_count zero */
336 inodes_stat.nr_unused -= count;
341 * This is a two-stage process. First we collect all
342 * offending inodes onto the throw-away list, and in
343 * the second stage we actually dispose of them. This
344 * is because we don't want to sleep while messing
345 * with the global lists..
349 * invalidate_inodes - discard the inodes on a device
352 * Discard all of the inodes for a given superblock. If the discard
353 * fails because there are busy inodes then a non zero value is returned.
354 * If the discard is successful all the inodes have been discarded.
356 int invalidate_inodes(struct super_block * sb)
359 LIST_HEAD(throw_away);
362 spin_lock(&inode_lock);
363 busy = invalidate_list(&sb->s_inodes, &throw_away);
364 spin_unlock(&inode_lock);
366 dispose_list(&throw_away);
372 EXPORT_SYMBOL(invalidate_inodes);
374 int __invalidate_device(struct block_device *bdev, int do_sync)
376 struct super_block *sb;
383 sb = get_super(bdev);
386 * no need to lock the super, get_super holds the
387 * read semaphore so the filesystem cannot go away
388 * under us (->put_super runs with the write lock
391 shrink_dcache_sb(sb);
392 res = invalidate_inodes(sb);
395 invalidate_bdev(bdev, 0);
399 EXPORT_SYMBOL(__invalidate_device);
401 static int can_unuse(struct inode *inode)
405 if (inode_has_buffers(inode))
407 if (atomic_read(&inode->i_count))
409 if (inode->i_data.nrpages)
415 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
416 * a temporary list and then are freed outside inode_lock by dispose_list().
418 * Any inodes which are pinned purely because of attached pagecache have their
419 * pagecache removed. We expect the final iput() on that inode to add it to
420 * the front of the inode_unused list. So look for it there and if the
421 * inode is still freeable, proceed. The right inode is found 99.9% of the
422 * time in testing on a 4-way.
424 * If the inode has metadata buffers attached to mapping->private_list then
425 * try to remove them.
427 static void prune_icache(int nr_to_scan)
432 unsigned long reap = 0;
435 spin_lock(&inode_lock);
436 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
439 if (list_empty(&inode_unused))
442 inode = list_entry(inode_unused.prev, struct inode, i_list);
444 if (inode->i_state || atomic_read(&inode->i_count)) {
445 list_move(&inode->i_list, &inode_unused);
448 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
450 spin_unlock(&inode_lock);
451 if (remove_inode_buffers(inode))
452 reap += invalidate_inode_pages(&inode->i_data);
454 spin_lock(&inode_lock);
456 if (inode != list_entry(inode_unused.next,
457 struct inode, i_list))
458 continue; /* wrong inode or list_empty */
459 if (!can_unuse(inode))
462 hlist_del_init(&inode->i_hash);
463 list_del_init(&inode->i_sb_list);
464 list_move(&inode->i_list, &freeable);
465 inode->i_state |= I_FREEING;
468 inodes_stat.nr_unused -= nr_pruned;
469 spin_unlock(&inode_lock);
471 dispose_list(&freeable);
474 if (current_is_kswapd())
475 mod_page_state(kswapd_inodesteal, reap);
477 mod_page_state(pginodesteal, reap);
481 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
482 * "unused" means that no dentries are referring to the inodes: the files are
483 * not open and the dcache references to those inodes have already been
486 * This function is passed the number of inodes to scan, and it returns the
487 * total number of remaining possibly-reclaimable inodes.
489 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
493 * Nasty deadlock avoidance. We may hold various FS locks,
494 * and we don't want to recurse into the FS that called us
495 * in clear_inode() and friends..
497 if (!(gfp_mask & __GFP_FS))
501 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
504 static void __wait_on_freeing_inode(struct inode *inode);
506 * Called with the inode lock held.
507 * NOTE: we are not increasing the inode-refcount, you must call __iget()
508 * by hand after calling find_inode now! This simplifies iunique and won't
509 * add any additional branch in the common code.
511 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
513 struct hlist_node *node;
514 struct inode * inode = NULL;
517 hlist_for_each (node, head) {
518 inode = hlist_entry(node, struct inode, i_hash);
519 if (inode->i_sb != sb)
521 if (!test(inode, data))
523 if (inode->i_state & (I_FREEING|I_CLEAR)) {
524 __wait_on_freeing_inode(inode);
529 return node ? inode : NULL;
533 * find_inode_fast is the fast path version of find_inode, see the comment at
534 * iget_locked for details.
536 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
538 struct hlist_node *node;
539 struct inode * inode = NULL;
542 hlist_for_each (node, head) {
543 inode = hlist_entry(node, struct inode, i_hash);
544 if (inode->i_ino != ino)
546 if (inode->i_sb != sb)
548 if (inode->i_state & (I_FREEING|I_CLEAR)) {
549 __wait_on_freeing_inode(inode);
554 return node ? inode : NULL;
558 * new_inode - obtain an inode
561 * Allocates a new inode for given superblock.
563 struct inode *new_inode(struct super_block *sb)
565 static unsigned long last_ino;
566 struct inode * inode;
568 spin_lock_prefetch(&inode_lock);
570 inode = alloc_inode(sb);
572 spin_lock(&inode_lock);
573 inodes_stat.nr_inodes++;
574 list_add(&inode->i_list, &inode_in_use);
575 list_add(&inode->i_sb_list, &sb->s_inodes);
576 inode->i_ino = ++last_ino;
578 spin_unlock(&inode_lock);
583 EXPORT_SYMBOL(new_inode);
585 void unlock_new_inode(struct inode *inode)
588 * This is special! We do not need the spinlock
589 * when clearing I_LOCK, because we're guaranteed
590 * that nobody else tries to do anything about the
591 * state of the inode when it is locked, as we
592 * just created it (so there can be no old holders
593 * that haven't tested I_LOCK).
595 inode->i_state &= ~(I_LOCK|I_NEW);
596 wake_up_inode(inode);
599 EXPORT_SYMBOL(unlock_new_inode);
602 * This is called without the inode lock held.. Be careful.
604 * We no longer cache the sb_flags in i_flags - see fs.h
605 * -- rmk@arm.uk.linux.org
607 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)
609 struct inode * inode;
611 inode = alloc_inode(sb);
615 spin_lock(&inode_lock);
616 /* We released the lock, so.. */
617 old = find_inode(sb, head, test, data);
619 if (set(inode, data))
622 inodes_stat.nr_inodes++;
623 list_add(&inode->i_list, &inode_in_use);
624 list_add(&inode->i_sb_list, &sb->s_inodes);
625 hlist_add_head(&inode->i_hash, head);
626 inode->i_state = I_LOCK|I_NEW;
627 spin_unlock(&inode_lock);
629 /* Return the locked inode with I_NEW set, the
630 * caller is responsible for filling in the contents
636 * Uhhuh, somebody else created the same inode under
637 * us. Use the old inode instead of the one we just
641 spin_unlock(&inode_lock);
642 destroy_inode(inode);
644 wait_on_inode(inode);
649 spin_unlock(&inode_lock);
650 destroy_inode(inode);
655 * get_new_inode_fast is the fast path version of get_new_inode, see the
656 * comment at iget_locked for details.
658 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
660 struct inode * inode;
662 inode = alloc_inode(sb);
666 spin_lock(&inode_lock);
667 /* We released the lock, so.. */
668 old = find_inode_fast(sb, head, ino);
671 inodes_stat.nr_inodes++;
672 list_add(&inode->i_list, &inode_in_use);
673 list_add(&inode->i_sb_list, &sb->s_inodes);
674 hlist_add_head(&inode->i_hash, head);
675 inode->i_state = I_LOCK|I_NEW;
676 spin_unlock(&inode_lock);
678 /* Return the locked inode with I_NEW set, the
679 * caller is responsible for filling in the contents
685 * Uhhuh, somebody else created the same inode under
686 * us. Use the old inode instead of the one we just
690 spin_unlock(&inode_lock);
691 destroy_inode(inode);
693 wait_on_inode(inode);
698 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
702 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
704 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
705 return tmp & I_HASHMASK;
709 * iunique - get a unique inode number
711 * @max_reserved: highest reserved inode number
713 * Obtain an inode number that is unique on the system for a given
714 * superblock. This is used by file systems that have no natural
715 * permanent inode numbering system. An inode number is returned that
716 * is higher than the reserved limit but unique.
719 * With a large number of inodes live on the file system this function
720 * currently becomes quite slow.
722 ino_t iunique(struct super_block *sb, ino_t max_reserved)
724 static ino_t counter;
726 struct hlist_head * head;
728 spin_lock(&inode_lock);
730 if (counter > max_reserved) {
731 head = inode_hashtable + hash(sb,counter);
733 inode = find_inode_fast(sb, head, res);
735 spin_unlock(&inode_lock);
739 counter = max_reserved + 1;
745 EXPORT_SYMBOL(iunique);
747 struct inode *igrab(struct inode *inode)
749 spin_lock(&inode_lock);
750 if (!(inode->i_state & I_FREEING))
754 * Handle the case where s_op->clear_inode is not been
755 * called yet, and somebody is calling igrab
756 * while the inode is getting freed.
759 spin_unlock(&inode_lock);
763 EXPORT_SYMBOL(igrab);
766 * ifind - internal function, you want ilookup5() or iget5().
767 * @sb: super block of file system to search
768 * @head: the head of the list to search
769 * @test: callback used for comparisons between inodes
770 * @data: opaque data pointer to pass to @test
772 * ifind() searches for the inode specified by @data in the inode
773 * cache. This is a generalized version of ifind_fast() for file systems where
774 * the inode number is not sufficient for unique identification of an inode.
776 * If the inode is in the cache, the inode is returned with an incremented
779 * Otherwise NULL is returned.
781 * Note, @test is called with the inode_lock held, so can't sleep.
783 static inline struct inode *ifind(struct super_block *sb,
784 struct hlist_head *head, int (*test)(struct inode *, void *),
789 spin_lock(&inode_lock);
790 inode = find_inode(sb, head, test, data);
793 spin_unlock(&inode_lock);
794 wait_on_inode(inode);
797 spin_unlock(&inode_lock);
802 * ifind_fast - internal function, you want ilookup() or iget().
803 * @sb: super block of file system to search
804 * @head: head of the list to search
805 * @ino: inode number to search for
807 * ifind_fast() searches for the inode @ino in the inode cache. This is for
808 * file systems where the inode number is sufficient for unique identification
811 * If the inode is in the cache, the inode is returned with an incremented
814 * Otherwise NULL is returned.
816 static inline struct inode *ifind_fast(struct super_block *sb,
817 struct hlist_head *head, unsigned long ino)
821 spin_lock(&inode_lock);
822 inode = find_inode_fast(sb, head, ino);
825 spin_unlock(&inode_lock);
826 wait_on_inode(inode);
829 spin_unlock(&inode_lock);
834 * ilookup5 - search for an inode in the inode cache
835 * @sb: super block of file system to search
836 * @hashval: hash value (usually inode number) to search for
837 * @test: callback used for comparisons between inodes
838 * @data: opaque data pointer to pass to @test
840 * ilookup5() uses ifind() to search for the inode specified by @hashval and
841 * @data in the inode cache. This is a generalized version of ilookup() for
842 * file systems where the inode number is not sufficient for unique
843 * identification of an inode.
845 * If the inode is in the cache, the inode is returned with an incremented
848 * Otherwise NULL is returned.
850 * Note, @test is called with the inode_lock held, so can't sleep.
852 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
853 int (*test)(struct inode *, void *), void *data)
855 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
857 return ifind(sb, head, test, data);
860 EXPORT_SYMBOL(ilookup5);
863 * ilookup - search for an inode in the inode cache
864 * @sb: super block of file system to search
865 * @ino: inode number to search for
867 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
868 * This is for file systems where the inode number is sufficient for unique
869 * identification of an inode.
871 * If the inode is in the cache, the inode is returned with an incremented
874 * Otherwise NULL is returned.
876 struct inode *ilookup(struct super_block *sb, unsigned long ino)
878 struct hlist_head *head = inode_hashtable + hash(sb, ino);
880 return ifind_fast(sb, head, ino);
883 EXPORT_SYMBOL(ilookup);
886 * iget5_locked - obtain an inode from a mounted file system
887 * @sb: super block of file system
888 * @hashval: hash value (usually inode number) to get
889 * @test: callback used for comparisons between inodes
890 * @set: callback used to initialize a new struct inode
891 * @data: opaque data pointer to pass to @test and @set
893 * This is iget() without the read_inode() portion of get_new_inode().
895 * iget5_locked() uses ifind() to search for the inode specified by @hashval
896 * and @data in the inode cache and if present it is returned with an increased
897 * reference count. This is a generalized version of iget_locked() for file
898 * systems where the inode number is not sufficient for unique identification
901 * If the inode is not in cache, get_new_inode() is called to allocate a new
902 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
903 * file system gets to fill it in before unlocking it via unlock_new_inode().
905 * Note both @test and @set are called with the inode_lock held, so can't sleep.
907 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
908 int (*test)(struct inode *, void *),
909 int (*set)(struct inode *, void *), void *data)
911 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
914 inode = ifind(sb, head, test, data);
918 * get_new_inode() will do the right thing, re-trying the search
919 * in case it had to block at any point.
921 return get_new_inode(sb, head, test, set, data);
924 EXPORT_SYMBOL(iget5_locked);
927 * iget_locked - obtain an inode from a mounted file system
928 * @sb: super block of file system
929 * @ino: inode number to get
931 * This is iget() without the read_inode() portion of get_new_inode_fast().
933 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
934 * the inode cache and if present it is returned with an increased reference
935 * count. This is for file systems where the inode number is sufficient for
936 * unique identification of an inode.
938 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
939 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
940 * The file system gets to fill it in before unlocking it via
941 * unlock_new_inode().
943 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
945 struct hlist_head *head = inode_hashtable + hash(sb, ino);
948 inode = ifind_fast(sb, head, ino);
952 * get_new_inode_fast() will do the right thing, re-trying the search
953 * in case it had to block at any point.
955 return get_new_inode_fast(sb, head, ino);
958 EXPORT_SYMBOL(iget_locked);
961 * __insert_inode_hash - hash an inode
962 * @inode: unhashed inode
963 * @hashval: unsigned long value used to locate this object in the
966 * Add an inode to the inode hash for this superblock.
968 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
970 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
971 spin_lock(&inode_lock);
972 hlist_add_head(&inode->i_hash, head);
973 spin_unlock(&inode_lock);
976 EXPORT_SYMBOL(__insert_inode_hash);
979 * remove_inode_hash - remove an inode from the hash
980 * @inode: inode to unhash
982 * Remove an inode from the superblock.
984 void remove_inode_hash(struct inode *inode)
986 spin_lock(&inode_lock);
987 hlist_del_init(&inode->i_hash);
988 spin_unlock(&inode_lock);
991 EXPORT_SYMBOL(remove_inode_hash);
994 * Tell the filesystem that this inode is no longer of any interest and should
995 * be completely destroyed.
997 * We leave the inode in the inode hash table until *after* the filesystem's
998 * ->delete_inode completes. This ensures that an iget (such as nfsd might
999 * instigate) will always find up-to-date information either in the hash or on
1002 * I_FREEING is set so that no-one will take a new reference to the inode while
1003 * it is being deleted.
1005 void generic_delete_inode(struct inode *inode)
1007 struct super_operations *op = inode->i_sb->s_op;
1009 list_del_init(&inode->i_list);
1010 list_del_init(&inode->i_sb_list);
1011 inode->i_state|=I_FREEING;
1012 inodes_stat.nr_inodes--;
1013 spin_unlock(&inode_lock);
1015 if (inode->i_data.nrpages)
1016 truncate_inode_pages(&inode->i_data, 0);
1018 security_inode_delete(inode);
1020 if (op->delete_inode) {
1021 void (*delete)(struct inode *) = op->delete_inode;
1022 if (!is_bad_inode(inode))
1024 /* s_op->delete_inode internally recalls clear_inode() */
1028 spin_lock(&inode_lock);
1029 hlist_del_init(&inode->i_hash);
1030 spin_unlock(&inode_lock);
1031 wake_up_inode(inode);
1032 if (inode->i_state != I_CLEAR)
1034 destroy_inode(inode);
1037 EXPORT_SYMBOL(generic_delete_inode);
1039 static void generic_forget_inode(struct inode *inode)
1041 struct super_block *sb = inode->i_sb;
1043 if (!hlist_unhashed(&inode->i_hash)) {
1044 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1045 list_move(&inode->i_list, &inode_unused);
1046 inodes_stat.nr_unused++;
1047 spin_unlock(&inode_lock);
1048 if (!sb || (sb->s_flags & MS_ACTIVE))
1050 write_inode_now(inode, 1);
1051 spin_lock(&inode_lock);
1052 inodes_stat.nr_unused--;
1053 hlist_del_init(&inode->i_hash);
1055 list_del_init(&inode->i_list);
1056 list_del_init(&inode->i_sb_list);
1057 inode->i_state|=I_FREEING;
1058 inodes_stat.nr_inodes--;
1059 spin_unlock(&inode_lock);
1060 if (inode->i_data.nrpages)
1061 truncate_inode_pages(&inode->i_data, 0);
1063 destroy_inode(inode);
1067 * Normal UNIX filesystem behaviour: delete the
1068 * inode when the usage count drops to zero, and
1071 static void generic_drop_inode(struct inode *inode)
1073 if (!inode->i_nlink)
1074 generic_delete_inode(inode);
1076 generic_forget_inode(inode);
1080 * Called when we're dropping the last reference
1083 * Call the FS "drop()" function, defaulting to
1084 * the legacy UNIX filesystem behaviour..
1086 * NOTE! NOTE! NOTE! We're called with the inode lock
1087 * held, and the drop function is supposed to release
1090 static inline void iput_final(struct inode *inode)
1092 struct super_operations *op = inode->i_sb->s_op;
1093 void (*drop)(struct inode *) = generic_drop_inode;
1095 if (op && op->drop_inode)
1096 drop = op->drop_inode;
1101 * iput - put an inode
1102 * @inode: inode to put
1104 * Puts an inode, dropping its usage count. If the inode use count hits
1105 * zero the inode is also then freed and may be destroyed.
1107 void iput(struct inode *inode)
1110 struct super_operations *op = inode->i_sb->s_op;
1112 if (inode->i_state == I_CLEAR)
1115 if (op && op->put_inode)
1116 op->put_inode(inode);
1118 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1123 EXPORT_SYMBOL(iput);
1126 * bmap - find a block number in a file
1127 * @inode: inode of file
1128 * @block: block to find
1130 * Returns the block number on the device holding the inode that
1131 * is the disk block number for the block of the file requested.
1132 * That is, asked for block 4 of inode 1 the function will return the
1133 * disk block relative to the disk start that holds that block of the
1136 sector_t bmap(struct inode * inode, sector_t block)
1139 if (inode->i_mapping->a_ops->bmap)
1140 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1144 EXPORT_SYMBOL(bmap);
1147 * update_atime - update the access time
1148 * @inode: inode accessed
1150 * Update the accessed time on an inode and mark it for writeback.
1151 * This function automatically handles read only file systems and media,
1152 * as well as the "noatime" flag and inode specific "noatime" markers.
1154 void update_atime(struct inode *inode)
1156 struct timespec now;
1158 if (IS_NOATIME(inode))
1160 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1162 if (IS_RDONLY(inode))
1165 now = current_fs_time(inode->i_sb);
1166 if (!timespec_equal(&inode->i_atime, &now)) {
1167 inode->i_atime = now;
1168 mark_inode_dirty_sync(inode);
1170 if (!timespec_equal(&inode->i_atime, &now))
1171 inode->i_atime = now;
1175 EXPORT_SYMBOL(update_atime);
1178 * inode_update_time - update mtime and ctime time
1179 * @inode: inode accessed
1180 * @ctime_too: update ctime too
1182 * Update the mtime time on an inode and mark it for writeback.
1183 * When ctime_too is specified update the ctime too.
1186 void inode_update_time(struct inode *inode, int ctime_too)
1188 struct timespec now;
1191 if (IS_NOCMTIME(inode))
1193 if (IS_RDONLY(inode))
1196 now = current_fs_time(inode->i_sb);
1197 if (!timespec_equal(&inode->i_mtime, &now))
1199 inode->i_mtime = now;
1202 if (!timespec_equal(&inode->i_ctime, &now))
1204 inode->i_ctime = now;
1207 mark_inode_dirty_sync(inode);
1210 EXPORT_SYMBOL(inode_update_time);
1212 int inode_needs_sync(struct inode *inode)
1216 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1221 EXPORT_SYMBOL(inode_needs_sync);
1224 * Quota functions that want to walk the inode lists..
1228 /* Function back in dquot.c */
1229 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1231 void remove_dquot_ref(struct super_block *sb, int type,
1232 struct list_head *tofree_head)
1234 struct inode *inode;
1237 return; /* nothing to do */
1238 spin_lock(&inode_lock); /* This lock is for inodes code */
1241 * We don't have to lock against quota code - test IS_QUOTAINIT is
1242 * just for speedup...
1244 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1245 if (!IS_NOQUOTA(inode))
1246 remove_inode_dquot_ref(inode, type, tofree_head);
1248 spin_unlock(&inode_lock);
1253 int inode_wait(void *word)
1260 * If we try to find an inode in the inode hash while it is being deleted, we
1261 * have to wait until the filesystem completes its deletion before reporting
1262 * that it isn't found. This is because iget will immediately call
1263 * ->read_inode, and we want to be sure that evidence of the deletion is found
1265 * This is called with inode_lock held.
1267 static void __wait_on_freeing_inode(struct inode *inode)
1269 wait_queue_head_t *wq;
1270 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1273 * I_FREEING and I_CLEAR are cleared in process context under
1274 * inode_lock, so we have to give the tasks who would clear them
1275 * a chance to run and acquire inode_lock.
1277 if (!(inode->i_state & I_LOCK)) {
1278 spin_unlock(&inode_lock);
1280 spin_lock(&inode_lock);
1283 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1284 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1285 spin_unlock(&inode_lock);
1287 finish_wait(wq, &wait.wait);
1288 spin_lock(&inode_lock);
1291 void wake_up_inode(struct inode *inode)
1294 * Prevent speculative execution through spin_unlock(&inode_lock);
1297 wake_up_bit(&inode->i_state, __I_LOCK);
1300 static __initdata unsigned long ihash_entries;
1301 static int __init set_ihash_entries(char *str)
1305 ihash_entries = simple_strtoul(str, &str, 0);
1308 __setup("ihash_entries=", set_ihash_entries);
1311 * Initialize the waitqueues and inode hash table.
1313 void __init inode_init_early(void)
1317 /* If hashes are distributed across NUMA nodes, defer
1318 * hash allocation until vmalloc space is available.
1324 alloc_large_system_hash("Inode-cache",
1325 sizeof(struct hlist_head),
1333 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1334 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1337 void __init inode_init(unsigned long mempages)
1341 /* inode slab cache */
1342 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1343 0, SLAB_PANIC, init_once, NULL);
1344 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1346 /* Hash may have been set up in inode_init_early */
1351 alloc_large_system_hash("Inode-cache",
1352 sizeof(struct hlist_head),
1360 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1361 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1364 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1366 inode->i_mode = mode;
1367 if (S_ISCHR(mode)) {
1368 inode->i_fop = &def_chr_fops;
1369 inode->i_rdev = rdev;
1370 } else if (S_ISBLK(mode)) {
1371 inode->i_fop = &def_blk_fops;
1372 inode->i_rdev = rdev;
1373 } else if (S_ISFIFO(mode))
1374 inode->i_fop = &def_fifo_fops;
1375 else if (S_ISSOCK(mode))
1376 inode->i_fop = &bad_sock_fops;
1378 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1381 EXPORT_SYMBOL(init_special_inode);