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/vs_base.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 static 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 if (sb->s_flags & MS_TAGXID)
119 inode->i_xid = current->xid;
121 inode->i_xid = 0; /* maybe xid -1 would be better? */
122 // inode->i_dqh = dqhget(sb->s_dqh);
123 inode->i_blkbits = sb->s_blocksize_bits;
125 atomic_set(&inode->i_count, 1);
127 inode->i_op = &empty_iops;
128 inode->i_fop = &empty_fops;
130 atomic_set(&inode->i_writecount, 0);
134 inode->i_generation = 0;
136 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
138 inode->i_pipe = NULL;
139 inode->i_bdev = NULL;
140 inode->i_cdev = NULL;
142 // inode->i_xid = 0; /* maybe not too wise ... */
143 inode->i_security = NULL;
144 inode->dirtied_when = 0;
145 if (security_inode_alloc(inode)) {
146 if (inode->i_sb->s_op->destroy_inode)
147 inode->i_sb->s_op->destroy_inode(inode);
149 kmem_cache_free(inode_cachep, (inode));
153 mapping->a_ops = &empty_aops;
154 mapping->host = inode;
156 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
157 mapping->assoc_mapping = NULL;
158 mapping->backing_dev_info = &default_backing_dev_info;
161 * If the block_device provides a backing_dev_info for client
162 * inodes then use that. Otherwise the inode share the bdev's
166 struct backing_dev_info *bdi;
168 bdi = sb->s_bdev->bd_inode_backing_dev_info;
170 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
171 mapping->backing_dev_info = bdi;
173 memset(&inode->u, 0, sizeof(inode->u));
174 inode->i_mapping = mapping;
179 void destroy_inode(struct inode *inode)
181 if (inode_has_buffers(inode))
183 security_inode_free(inode);
184 if (inode->i_sb->s_op->destroy_inode)
185 inode->i_sb->s_op->destroy_inode(inode);
187 kmem_cache_free(inode_cachep, (inode));
192 * These are initializations that only need to be done
193 * once, because the fields are idempotent across use
194 * of the inode, so let the slab aware of that.
196 void inode_init_once(struct inode *inode)
198 memset(inode, 0, sizeof(*inode));
199 INIT_HLIST_NODE(&inode->i_hash);
200 INIT_LIST_HEAD(&inode->i_dentry);
201 INIT_LIST_HEAD(&inode->i_devices);
202 sema_init(&inode->i_sem, 1);
203 init_rwsem(&inode->i_alloc_sem);
204 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
205 spin_lock_init(&inode->i_data.tree_lock);
206 spin_lock_init(&inode->i_data.i_mmap_lock);
207 atomic_set(&inode->i_data.truncate_count, 0);
208 INIT_LIST_HEAD(&inode->i_data.private_list);
209 spin_lock_init(&inode->i_data.private_lock);
210 INIT_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
211 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
212 spin_lock_init(&inode->i_lock);
213 i_size_ordered_init(inode);
216 EXPORT_SYMBOL(inode_init_once);
218 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
220 struct inode * inode = (struct inode *) foo;
222 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
223 SLAB_CTOR_CONSTRUCTOR)
224 inode_init_once(inode);
228 * inode_lock must be held
230 void __iget(struct inode * inode)
232 if (atomic_read(&inode->i_count)) {
233 atomic_inc(&inode->i_count);
236 atomic_inc(&inode->i_count);
237 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
238 list_move(&inode->i_list, &inode_in_use);
239 inodes_stat.nr_unused--;
243 * clear_inode - clear an inode
244 * @inode: inode to clear
246 * This is called by the filesystem to tell us
247 * that the inode is no longer useful. We just
248 * terminate it with extreme prejudice.
250 void clear_inode(struct inode *inode)
252 invalidate_inode_buffers(inode);
254 if (inode->i_data.nrpages)
256 if (!(inode->i_state & I_FREEING))
258 if (inode->i_state & I_CLEAR)
260 wait_on_inode(inode);
262 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
263 inode->i_sb->s_op->clear_inode(inode);
268 inode->i_state = I_CLEAR;
271 EXPORT_SYMBOL(clear_inode);
274 * dispose_list - dispose of the contents of a local list
275 * @head: the head of the list to free
277 * Dispose-list gets a local list with local inodes in it, so it doesn't
278 * need to worry about list corruption and SMP locks.
280 static void dispose_list(struct list_head *head)
284 while (!list_empty(head)) {
287 inode = list_entry(head->next, struct inode, i_list);
288 list_del(&inode->i_list);
290 if (inode->i_data.nrpages)
291 truncate_inode_pages(&inode->i_data, 0);
293 destroy_inode(inode);
296 spin_lock(&inode_lock);
297 inodes_stat.nr_inodes -= nr_disposed;
298 spin_unlock(&inode_lock);
302 * Invalidate all inodes for a device.
304 static int invalidate_list(struct list_head *head, struct super_block * sb, struct list_head * dispose)
306 struct list_head *next;
307 int busy = 0, count = 0;
311 struct list_head * tmp = next;
312 struct inode * inode;
317 inode = list_entry(tmp, struct inode, i_list);
318 if (inode->i_sb != sb)
320 invalidate_inode_buffers(inode);
321 if (!atomic_read(&inode->i_count)) {
322 hlist_del_init(&inode->i_hash);
323 list_move(&inode->i_list, dispose);
324 inode->i_state |= I_FREEING;
330 /* only unused inodes may be cached with i_count zero */
331 inodes_stat.nr_unused -= count;
336 * This is a two-stage process. First we collect all
337 * offending inodes onto the throw-away list, and in
338 * the second stage we actually dispose of them. This
339 * is because we don't want to sleep while messing
340 * with the global lists..
344 * invalidate_inodes - discard the inodes on a device
347 * Discard all of the inodes for a given superblock. If the discard
348 * fails because there are busy inodes then a non zero value is returned.
349 * If the discard is successful all the inodes have been discarded.
351 int invalidate_inodes(struct super_block * sb)
354 LIST_HEAD(throw_away);
357 spin_lock(&inode_lock);
358 busy = invalidate_list(&inode_in_use, sb, &throw_away);
359 busy |= invalidate_list(&inode_unused, sb, &throw_away);
360 busy |= invalidate_list(&sb->s_dirty, sb, &throw_away);
361 busy |= invalidate_list(&sb->s_io, sb, &throw_away);
362 spin_unlock(&inode_lock);
364 dispose_list(&throw_away);
370 EXPORT_SYMBOL(invalidate_inodes);
372 int __invalidate_device(struct block_device *bdev, int do_sync)
374 struct super_block *sb;
381 sb = get_super(bdev);
384 * no need to lock the super, get_super holds the
385 * read semaphore so the filesystem cannot go away
386 * under us (->put_super runs with the write lock
389 shrink_dcache_sb(sb);
390 res = invalidate_inodes(sb);
393 invalidate_bdev(bdev, 0);
397 EXPORT_SYMBOL(__invalidate_device);
399 static int can_unuse(struct inode *inode)
403 if (inode_has_buffers(inode))
405 if (atomic_read(&inode->i_count))
407 if (inode->i_data.nrpages)
413 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
414 * a temporary list and then are freed outside inode_lock by dispose_list().
416 * Any inodes which are pinned purely because of attached pagecache have their
417 * pagecache removed. We expect the final iput() on that inode to add it to
418 * the front of the inode_unused list. So look for it there and if the
419 * inode is still freeable, proceed. The right inode is found 99.9% of the
420 * time in testing on a 4-way.
422 * If the inode has metadata buffers attached to mapping->private_list then
423 * try to remove them.
425 static void prune_icache(int nr_to_scan)
430 unsigned long reap = 0;
433 spin_lock(&inode_lock);
434 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
437 if (list_empty(&inode_unused))
440 inode = list_entry(inode_unused.prev, struct inode, i_list);
442 if (inode->i_state || atomic_read(&inode->i_count)) {
443 list_move(&inode->i_list, &inode_unused);
446 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
448 spin_unlock(&inode_lock);
449 if (remove_inode_buffers(inode))
450 reap += invalidate_inode_pages(&inode->i_data);
452 spin_lock(&inode_lock);
454 if (inode != list_entry(inode_unused.next,
455 struct inode, i_list))
456 continue; /* wrong inode or list_empty */
457 if (!can_unuse(inode))
460 hlist_del_init(&inode->i_hash);
461 list_move(&inode->i_list, &freeable);
462 inode->i_state |= I_FREEING;
465 inodes_stat.nr_unused -= nr_pruned;
466 spin_unlock(&inode_lock);
468 dispose_list(&freeable);
471 if (current_is_kswapd())
472 mod_page_state(kswapd_inodesteal, reap);
474 mod_page_state(pginodesteal, reap);
478 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
479 * "unused" means that no dentries are referring to the inodes: the files are
480 * not open and the dcache references to those inodes have already been
483 * This function is passed the number of inodes to scan, and it returns the
484 * total number of remaining possibly-reclaimable inodes.
486 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
490 * Nasty deadlock avoidance. We may hold various FS locks,
491 * and we don't want to recurse into the FS that called us
492 * in clear_inode() and friends..
494 if (gfp_mask & __GFP_FS)
497 return inodes_stat.nr_unused;
500 static void __wait_on_freeing_inode(struct inode *inode);
502 * Called with the inode lock held.
503 * NOTE: we are not increasing the inode-refcount, you must call __iget()
504 * by hand after calling find_inode now! This simplifies iunique and won't
505 * add any additional branch in the common code.
507 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
509 struct hlist_node *node;
510 struct inode * inode = NULL;
513 hlist_for_each (node, head) {
514 inode = hlist_entry(node, struct inode, i_hash);
515 if (inode->i_sb != sb)
517 if (!test(inode, data))
519 if (inode->i_state & (I_FREEING|I_CLEAR)) {
520 __wait_on_freeing_inode(inode);
525 return node ? inode : NULL;
529 * find_inode_fast is the fast path version of find_inode, see the comment at
530 * iget_locked for details.
532 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
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_ino != ino)
542 if (inode->i_sb != sb)
544 if (inode->i_state & (I_FREEING|I_CLEAR)) {
545 __wait_on_freeing_inode(inode);
550 return node ? inode : NULL;
554 * new_inode - obtain an inode
557 * Allocates a new inode for given superblock.
559 struct inode *new_inode(struct super_block *sb)
561 static unsigned long last_ino;
562 struct inode * inode;
564 spin_lock_prefetch(&inode_lock);
566 inode = alloc_inode(sb);
568 spin_lock(&inode_lock);
569 inodes_stat.nr_inodes++;
570 list_add(&inode->i_list, &inode_in_use);
571 inode->i_ino = ++last_ino;
573 inode->i_xid = vx_current_xid();
574 spin_unlock(&inode_lock);
579 EXPORT_SYMBOL(new_inode);
581 void unlock_new_inode(struct inode *inode)
584 * This is special! We do not need the spinlock
585 * when clearing I_LOCK, because we're guaranteed
586 * that nobody else tries to do anything about the
587 * state of the inode when it is locked, as we
588 * just created it (so there can be no old holders
589 * that haven't tested I_LOCK).
591 inode->i_state &= ~(I_LOCK|I_NEW);
592 wake_up_inode(inode);
595 EXPORT_SYMBOL(unlock_new_inode);
598 * This is called without the inode lock held.. Be careful.
600 * We no longer cache the sb_flags in i_flags - see fs.h
601 * -- rmk@arm.uk.linux.org
603 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)
605 struct inode * inode;
607 inode = alloc_inode(sb);
611 spin_lock(&inode_lock);
612 /* We released the lock, so.. */
613 old = find_inode(sb, head, test, data);
615 if (set(inode, data))
618 inodes_stat.nr_inodes++;
619 list_add(&inode->i_list, &inode_in_use);
620 hlist_add_head(&inode->i_hash, head);
621 inode->i_state = I_LOCK|I_NEW;
622 spin_unlock(&inode_lock);
624 /* Return the locked inode with I_NEW set, the
625 * caller is responsible for filling in the contents
631 * Uhhuh, somebody else created the same inode under
632 * us. Use the old inode instead of the one we just
636 spin_unlock(&inode_lock);
637 destroy_inode(inode);
639 wait_on_inode(inode);
644 spin_unlock(&inode_lock);
645 destroy_inode(inode);
650 * get_new_inode_fast is the fast path version of get_new_inode, see the
651 * comment at iget_locked for details.
653 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
655 struct inode * inode;
657 inode = alloc_inode(sb);
661 spin_lock(&inode_lock);
662 /* We released the lock, so.. */
663 old = find_inode_fast(sb, head, ino);
666 inodes_stat.nr_inodes++;
667 list_add(&inode->i_list, &inode_in_use);
668 hlist_add_head(&inode->i_hash, head);
669 inode->i_state = I_LOCK|I_NEW;
670 spin_unlock(&inode_lock);
672 /* Return the locked inode with I_NEW set, the
673 * caller is responsible for filling in the contents
679 * Uhhuh, somebody else created the same inode under
680 * us. Use the old inode instead of the one we just
684 spin_unlock(&inode_lock);
685 destroy_inode(inode);
687 wait_on_inode(inode);
692 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
696 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
698 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
699 return tmp & I_HASHMASK;
703 * iunique - get a unique inode number
705 * @max_reserved: highest reserved inode number
707 * Obtain an inode number that is unique on the system for a given
708 * superblock. This is used by file systems that have no natural
709 * permanent inode numbering system. An inode number is returned that
710 * is higher than the reserved limit but unique.
713 * With a large number of inodes live on the file system this function
714 * currently becomes quite slow.
716 ino_t iunique(struct super_block *sb, ino_t max_reserved)
718 static ino_t counter;
720 struct hlist_head * head;
722 spin_lock(&inode_lock);
724 if (counter > max_reserved) {
725 head = inode_hashtable + hash(sb,counter);
727 inode = find_inode_fast(sb, head, res);
729 spin_unlock(&inode_lock);
733 counter = max_reserved + 1;
739 EXPORT_SYMBOL(iunique);
741 struct inode *igrab(struct inode *inode)
743 spin_lock(&inode_lock);
744 if (!(inode->i_state & I_FREEING))
748 * Handle the case where s_op->clear_inode is not been
749 * called yet, and somebody is calling igrab
750 * while the inode is getting freed.
753 spin_unlock(&inode_lock);
757 EXPORT_SYMBOL(igrab);
760 * ifind - internal function, you want ilookup5() or iget5().
761 * @sb: super block of file system to search
762 * @head: the head of the list to search
763 * @test: callback used for comparisons between inodes
764 * @data: opaque data pointer to pass to @test
766 * ifind() searches for the inode specified by @data in the inode
767 * cache. This is a generalized version of ifind_fast() for file systems where
768 * the inode number is not sufficient for unique identification of an inode.
770 * If the inode is in the cache, the inode is returned with an incremented
773 * Otherwise NULL is returned.
775 * Note, @test is called with the inode_lock held, so can't sleep.
777 static inline struct inode *ifind(struct super_block *sb,
778 struct hlist_head *head, int (*test)(struct inode *, void *),
783 spin_lock(&inode_lock);
784 inode = find_inode(sb, head, test, data);
787 spin_unlock(&inode_lock);
788 wait_on_inode(inode);
791 spin_unlock(&inode_lock);
796 * ifind_fast - internal function, you want ilookup() or iget().
797 * @sb: super block of file system to search
798 * @head: head of the list to search
799 * @ino: inode number to search for
801 * ifind_fast() searches for the inode @ino in the inode cache. This is for
802 * file systems where the inode number is sufficient for unique identification
805 * If the inode is in the cache, the inode is returned with an incremented
808 * Otherwise NULL is returned.
810 static inline struct inode *ifind_fast(struct super_block *sb,
811 struct hlist_head *head, unsigned long ino)
815 spin_lock(&inode_lock);
816 inode = find_inode_fast(sb, head, ino);
819 spin_unlock(&inode_lock);
820 wait_on_inode(inode);
823 spin_unlock(&inode_lock);
828 * ilookup5 - search for an inode in the inode cache
829 * @sb: super block of file system to search
830 * @hashval: hash value (usually inode number) to search for
831 * @test: callback used for comparisons between inodes
832 * @data: opaque data pointer to pass to @test
834 * ilookup5() uses ifind() to search for the inode specified by @hashval and
835 * @data in the inode cache. This is a generalized version of ilookup() for
836 * file systems where the inode number is not sufficient for unique
837 * identification of an inode.
839 * If the inode is in the cache, the inode is returned with an incremented
842 * Otherwise NULL is returned.
844 * Note, @test is called with the inode_lock held, so can't sleep.
846 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
847 int (*test)(struct inode *, void *), void *data)
849 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
851 return ifind(sb, head, test, data);
854 EXPORT_SYMBOL(ilookup5);
857 * ilookup - search for an inode in the inode cache
858 * @sb: super block of file system to search
859 * @ino: inode number to search for
861 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
862 * This is for file systems where the inode number is sufficient for unique
863 * identification of an inode.
865 * If the inode is in the cache, the inode is returned with an incremented
868 * Otherwise NULL is returned.
870 struct inode *ilookup(struct super_block *sb, unsigned long ino)
872 struct hlist_head *head = inode_hashtable + hash(sb, ino);
874 return ifind_fast(sb, head, ino);
877 EXPORT_SYMBOL(ilookup);
880 * iget5_locked - obtain an inode from a mounted file system
881 * @sb: super block of file system
882 * @hashval: hash value (usually inode number) to get
883 * @test: callback used for comparisons between inodes
884 * @set: callback used to initialize a new struct inode
885 * @data: opaque data pointer to pass to @test and @set
887 * This is iget() without the read_inode() portion of get_new_inode().
889 * iget5_locked() uses ifind() to search for the inode specified by @hashval
890 * and @data in the inode cache and if present it is returned with an increased
891 * reference count. This is a generalized version of iget_locked() for file
892 * systems where the inode number is not sufficient for unique identification
895 * If the inode is not in cache, get_new_inode() is called to allocate a new
896 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
897 * file system gets to fill it in before unlocking it via unlock_new_inode().
899 * Note both @test and @set are called with the inode_lock held, so can't sleep.
901 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
902 int (*test)(struct inode *, void *),
903 int (*set)(struct inode *, void *), void *data)
905 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
908 inode = ifind(sb, head, test, data);
912 * get_new_inode() will do the right thing, re-trying the search
913 * in case it had to block at any point.
915 return get_new_inode(sb, head, test, set, data);
918 EXPORT_SYMBOL(iget5_locked);
921 * iget_locked - obtain an inode from a mounted file system
922 * @sb: super block of file system
923 * @ino: inode number to get
925 * This is iget() without the read_inode() portion of get_new_inode_fast().
927 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
928 * the inode cache and if present it is returned with an increased reference
929 * count. This is for file systems where the inode number is sufficient for
930 * unique identification of an inode.
932 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
933 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
934 * The file system gets to fill it in before unlocking it via
935 * unlock_new_inode().
937 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
939 struct hlist_head *head = inode_hashtable + hash(sb, ino);
942 inode = ifind_fast(sb, head, ino);
946 * get_new_inode_fast() will do the right thing, re-trying the search
947 * in case it had to block at any point.
949 return get_new_inode_fast(sb, head, ino);
952 EXPORT_SYMBOL(iget_locked);
955 * __insert_inode_hash - hash an inode
956 * @inode: unhashed inode
957 * @hashval: unsigned long value used to locate this object in the
960 * Add an inode to the inode hash for this superblock.
962 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
964 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
965 spin_lock(&inode_lock);
966 hlist_add_head(&inode->i_hash, head);
967 spin_unlock(&inode_lock);
970 EXPORT_SYMBOL(__insert_inode_hash);
973 * remove_inode_hash - remove an inode from the hash
974 * @inode: inode to unhash
976 * Remove an inode from the superblock.
978 void remove_inode_hash(struct inode *inode)
980 spin_lock(&inode_lock);
981 hlist_del_init(&inode->i_hash);
982 spin_unlock(&inode_lock);
985 EXPORT_SYMBOL(remove_inode_hash);
988 * Tell the filesystem that this inode is no longer of any interest and should
989 * be completely destroyed.
991 * We leave the inode in the inode hash table until *after* the filesystem's
992 * ->delete_inode completes. This ensures that an iget (such as nfsd might
993 * instigate) will always find up-to-date information either in the hash or on
996 * I_FREEING is set so that no-one will take a new reference to the inode while
997 * it is being deleted.
999 void generic_delete_inode(struct inode *inode)
1001 struct super_operations *op = inode->i_sb->s_op;
1003 list_del_init(&inode->i_list);
1004 inode->i_state|=I_FREEING;
1005 inodes_stat.nr_inodes--;
1006 spin_unlock(&inode_lock);
1008 if (inode->i_data.nrpages)
1009 truncate_inode_pages(&inode->i_data, 0);
1011 security_inode_delete(inode);
1013 if (op->delete_inode) {
1014 void (*delete)(struct inode *) = op->delete_inode;
1015 if (!is_bad_inode(inode))
1017 /* s_op->delete_inode internally recalls clear_inode() */
1021 spin_lock(&inode_lock);
1022 hlist_del_init(&inode->i_hash);
1023 spin_unlock(&inode_lock);
1024 wake_up_inode(inode);
1025 if (inode->i_state != I_CLEAR)
1027 destroy_inode(inode);
1030 EXPORT_SYMBOL(generic_delete_inode);
1032 static void generic_forget_inode(struct inode *inode)
1034 struct super_block *sb = inode->i_sb;
1036 if (!hlist_unhashed(&inode->i_hash)) {
1037 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1038 list_move(&inode->i_list, &inode_unused);
1039 inodes_stat.nr_unused++;
1040 spin_unlock(&inode_lock);
1041 if (!sb || (sb->s_flags & MS_ACTIVE))
1043 write_inode_now(inode, 1);
1044 spin_lock(&inode_lock);
1045 inodes_stat.nr_unused--;
1046 hlist_del_init(&inode->i_hash);
1048 list_del_init(&inode->i_list);
1049 inode->i_state|=I_FREEING;
1050 inodes_stat.nr_inodes--;
1051 spin_unlock(&inode_lock);
1052 if (inode->i_data.nrpages)
1053 truncate_inode_pages(&inode->i_data, 0);
1055 destroy_inode(inode);
1059 * Normal UNIX filesystem behaviour: delete the
1060 * inode when the usage count drops to zero, and
1063 static void generic_drop_inode(struct inode *inode)
1065 if (!inode->i_nlink)
1066 generic_delete_inode(inode);
1068 generic_forget_inode(inode);
1072 * Called when we're dropping the last reference
1075 * Call the FS "drop()" function, defaulting to
1076 * the legacy UNIX filesystem behaviour..
1078 * NOTE! NOTE! NOTE! We're called with the inode lock
1079 * held, and the drop function is supposed to release
1082 static inline void iput_final(struct inode *inode)
1084 struct super_operations *op = inode->i_sb->s_op;
1085 void (*drop)(struct inode *) = generic_drop_inode;
1087 if (op && op->drop_inode)
1088 drop = op->drop_inode;
1093 * iput - put an inode
1094 * @inode: inode to put
1096 * Puts an inode, dropping its usage count. If the inode use count hits
1097 * zero the inode is also then freed and may be destroyed.
1099 void iput(struct inode *inode)
1102 struct super_operations *op = inode->i_sb->s_op;
1104 if (inode->i_state == I_CLEAR)
1107 if (op && op->put_inode)
1108 op->put_inode(inode);
1110 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1115 EXPORT_SYMBOL(iput);
1118 * bmap - find a block number in a file
1119 * @inode: inode of file
1120 * @block: block to find
1122 * Returns the block number on the device holding the inode that
1123 * is the disk block number for the block of the file requested.
1124 * That is, asked for block 4 of inode 1 the function will return the
1125 * disk block relative to the disk start that holds that block of the
1128 sector_t bmap(struct inode * inode, sector_t block)
1131 if (inode->i_mapping->a_ops->bmap)
1132 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1136 EXPORT_SYMBOL(bmap);
1139 * Return true if the filesystem which backs this inode considers the two
1140 * passed timespecs to be sufficiently different to warrant flushing the
1141 * altered time out to disk.
1143 static int inode_times_differ(struct inode *inode,
1144 struct timespec *old, struct timespec *new)
1146 if (IS_ONE_SECOND(inode))
1147 return old->tv_sec != new->tv_sec;
1148 return !timespec_equal(old, new);
1152 * update_atime - update the access time
1153 * @inode: inode accessed
1155 * Update the accessed time on an inode and mark it for writeback.
1156 * This function automatically handles read only file systems and media,
1157 * as well as the "noatime" flag and inode specific "noatime" markers.
1159 void update_atime(struct inode *inode)
1161 struct timespec now;
1163 if (IS_NOATIME(inode))
1165 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1167 if (IS_RDONLY(inode))
1170 now = current_kernel_time();
1171 if (inode_times_differ(inode, &inode->i_atime, &now)) {
1172 inode->i_atime = now;
1173 mark_inode_dirty_sync(inode);
1175 if (!timespec_equal(&inode->i_atime, &now))
1176 inode->i_atime = now;
1180 EXPORT_SYMBOL(update_atime);
1183 * inode_update_time - update mtime and ctime time
1184 * @inode: inode accessed
1185 * @ctime_too: update ctime too
1187 * Update the mtime time on an inode and mark it for writeback.
1188 * When ctime_too is specified update the ctime too.
1191 void inode_update_time(struct inode *inode, int ctime_too)
1193 struct timespec now;
1196 if (IS_NOCMTIME(inode))
1198 if (IS_RDONLY(inode))
1201 now = current_kernel_time();
1203 if (inode_times_differ(inode, &inode->i_mtime, &now))
1205 inode->i_mtime = now;
1208 if (inode_times_differ(inode, &inode->i_ctime, &now))
1210 inode->i_ctime = now;
1213 mark_inode_dirty_sync(inode);
1216 EXPORT_SYMBOL(inode_update_time);
1218 int inode_needs_sync(struct inode *inode)
1222 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1227 EXPORT_SYMBOL(inode_needs_sync);
1230 * Quota functions that want to walk the inode lists..
1234 /* Function back in dquot.c */
1235 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1237 void remove_dquot_ref(struct super_block *sb, int type, struct list_head *tofree_head)
1239 struct inode *inode;
1240 struct list_head *act_head;
1243 return; /* nothing to do */
1244 spin_lock(&inode_lock); /* This lock is for inodes code */
1245 /* We don't have to lock against quota code - test IS_QUOTAINIT is just for speedup... */
1247 list_for_each(act_head, &inode_in_use) {
1248 inode = list_entry(act_head, struct inode, i_list);
1249 if (inode->i_sb == sb && IS_QUOTAINIT(inode))
1250 remove_inode_dquot_ref(inode, type, tofree_head);
1252 list_for_each(act_head, &inode_unused) {
1253 inode = list_entry(act_head, struct inode, i_list);
1254 if (inode->i_sb == sb && IS_QUOTAINIT(inode))
1255 remove_inode_dquot_ref(inode, type, tofree_head);
1257 list_for_each(act_head, &sb->s_dirty) {
1258 inode = list_entry(act_head, struct inode, i_list);
1259 if (IS_QUOTAINIT(inode))
1260 remove_inode_dquot_ref(inode, type, tofree_head);
1262 list_for_each(act_head, &sb->s_io) {
1263 inode = list_entry(act_head, struct inode, i_list);
1264 if (IS_QUOTAINIT(inode))
1265 remove_inode_dquot_ref(inode, type, tofree_head);
1267 spin_unlock(&inode_lock);
1273 * Hashed waitqueues for wait_on_inode(). The table is pretty small - the
1274 * kernel doesn't lock many inodes at the same time.
1276 #define I_WAIT_TABLE_ORDER 3
1277 static struct i_wait_queue_head {
1278 wait_queue_head_t wqh;
1279 } ____cacheline_aligned_in_smp i_wait_queue_heads[1<<I_WAIT_TABLE_ORDER];
1282 * Return the address of the waitqueue_head to be used for this inode
1284 static wait_queue_head_t *i_waitq_head(struct inode *inode)
1286 return &i_wait_queue_heads[hash_ptr(inode, I_WAIT_TABLE_ORDER)].wqh;
1289 void __wait_on_inode(struct inode *inode)
1291 DECLARE_WAITQUEUE(wait, current);
1292 wait_queue_head_t *wq = i_waitq_head(inode);
1294 add_wait_queue(wq, &wait);
1296 set_current_state(TASK_UNINTERRUPTIBLE);
1297 if (inode->i_state & I_LOCK) {
1301 remove_wait_queue(wq, &wait);
1302 __set_current_state(TASK_RUNNING);
1306 * If we try to find an inode in the inode hash while it is being deleted, we
1307 * have to wait until the filesystem completes its deletion before reporting
1308 * that it isn't found. This is because iget will immediately call
1309 * ->read_inode, and we want to be sure that evidence of the deletion is found
1312 * This call might return early if an inode which shares the waitq is woken up.
1313 * This is most easily handled by the caller which will loop around again
1314 * looking for the inode.
1316 * This is called with inode_lock held.
1318 static void __wait_on_freeing_inode(struct inode *inode)
1320 DECLARE_WAITQUEUE(wait, current);
1321 wait_queue_head_t *wq = i_waitq_head(inode);
1323 add_wait_queue(wq, &wait);
1324 set_current_state(TASK_UNINTERRUPTIBLE);
1325 spin_unlock(&inode_lock);
1327 remove_wait_queue(wq, &wait);
1328 spin_lock(&inode_lock);
1331 void wake_up_inode(struct inode *inode)
1333 wait_queue_head_t *wq = i_waitq_head(inode);
1336 * Prevent speculative execution through spin_unlock(&inode_lock);
1339 if (waitqueue_active(wq))
1343 static __initdata unsigned long ihash_entries;
1344 static int __init set_ihash_entries(char *str)
1348 ihash_entries = simple_strtoul(str, &str, 0);
1351 __setup("ihash_entries=", set_ihash_entries);
1354 * Initialize the waitqueues and inode hash table.
1356 void __init inode_init(unsigned long mempages)
1358 struct hlist_head *head;
1359 unsigned long order;
1360 unsigned int nr_hash;
1363 for (i = 0; i < ARRAY_SIZE(i_wait_queue_heads); i++)
1364 init_waitqueue_head(&i_wait_queue_heads[i].wqh);
1367 ihash_entries = PAGE_SHIFT < 14 ?
1368 mempages >> (14 - PAGE_SHIFT) :
1369 mempages << (PAGE_SHIFT - 14);
1371 ihash_entries *= sizeof(struct hlist_head);
1372 for (order = 0; ((1UL << order) << PAGE_SHIFT) < ihash_entries; order++)
1378 nr_hash = (1UL << order) * PAGE_SIZE /
1379 sizeof(struct hlist_head);
1380 i_hash_mask = (nr_hash - 1);
1384 while ((tmp >>= 1UL) != 0UL)
1387 inode_hashtable = (struct hlist_head *)
1388 __get_free_pages(GFP_ATOMIC, order);
1389 } while (inode_hashtable == NULL && --order >= 0);
1391 printk("Inode-cache hash table entries: %d (order: %ld, %ld bytes)\n",
1392 nr_hash, order, (PAGE_SIZE << order));
1394 if (!inode_hashtable)
1395 panic("Failed to allocate inode hash table\n");
1397 head = inode_hashtable;
1400 INIT_HLIST_HEAD(head);
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);