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>
25 * This is needed for the following functions:
27 * - invalidate_inode_buffers
31 * FIXME: remove all knowledge of the buffer layer from this file
33 #include <linux/buffer_head.h>
36 * New inode.c implementation.
38 * This implementation has the basic premise of trying
39 * to be extremely low-overhead and SMP-safe, yet be
40 * simple enough to be "obviously correct".
45 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
47 /* #define INODE_PARANOIA 1 */
48 /* #define INODE_DEBUG 1 */
51 * Inode lookup is no longer as critical as it used to be:
52 * most of the lookups are going to be through the dcache.
54 #define I_HASHBITS i_hash_shift
55 #define I_HASHMASK i_hash_mask
57 static unsigned int i_hash_mask;
58 static unsigned int i_hash_shift;
61 * Each inode can be on two separate lists. One is
62 * the hash list of the inode, used for lookups. The
63 * other linked list is the "type" list:
64 * "in_use" - valid inode, i_count > 0, i_nlink > 0
65 * "dirty" - as "in_use" but also dirty
66 * "unused" - valid inode, i_count = 0
68 * A "dirty" list is maintained for each super block,
69 * allowing for low-overhead inode sync() operations.
72 LIST_HEAD(inode_in_use);
73 LIST_HEAD(inode_unused);
74 static struct hlist_head *inode_hashtable;
77 * A simple spinlock to protect the list manipulations.
79 * NOTE! You also have to own the lock if you change
80 * the i_state of an inode while it is in use..
82 spinlock_t inode_lock = SPIN_LOCK_UNLOCKED;
85 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
86 * icache shrinking path, and the umount path. Without this exclusion,
87 * by the time prune_icache calls iput for the inode whose pages it has
88 * been invalidating, or by the time it calls clear_inode & destroy_inode
89 * from its final dispose_list, the struct super_block they refer to
90 * (for inode->i_sb->s_op) may already have been freed and reused.
92 static DECLARE_MUTEX(iprune_sem);
95 * Statistics gathering..
97 struct inodes_stat_t inodes_stat;
99 static kmem_cache_t * inode_cachep;
101 static struct inode *alloc_inode(struct super_block *sb)
103 static struct address_space_operations empty_aops;
104 static struct inode_operations empty_iops;
105 static struct file_operations empty_fops;
108 if (sb->s_op->alloc_inode)
109 inode = sb->s_op->alloc_inode(sb);
111 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
114 struct address_space * const mapping = &inode->i_data;
117 if (sb->s_flags & MS_TAGXID)
118 inode->i_xid = current->xid;
120 inode->i_xid = 0; /* maybe xid -1 would be better? */
121 // inode->i_dqh = dqhget(sb->s_dqh);
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_xid = 0; /* maybe not too wise ... */
142 inode->i_security = NULL;
143 inode->dirtied_when = 0;
144 if (security_inode_alloc(inode)) {
145 if (inode->i_sb->s_op->destroy_inode)
146 inode->i_sb->s_op->destroy_inode(inode);
148 kmem_cache_free(inode_cachep, (inode));
152 mapping->a_ops = &empty_aops;
153 mapping->host = inode;
155 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
156 mapping->assoc_mapping = NULL;
157 mapping->backing_dev_info = &default_backing_dev_info;
159 mapping->backing_dev_info = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
160 memset(&inode->u, 0, sizeof(inode->u));
161 inode->i_mapping = mapping;
166 void destroy_inode(struct inode *inode)
168 if (inode_has_buffers(inode))
170 security_inode_free(inode);
171 if (inode->i_sb->s_op->destroy_inode)
172 inode->i_sb->s_op->destroy_inode(inode);
174 kmem_cache_free(inode_cachep, (inode));
179 * These are initializations that only need to be done
180 * once, because the fields are idempotent across use
181 * of the inode, so let the slab aware of that.
183 void inode_init_once(struct inode *inode)
185 memset(inode, 0, sizeof(*inode));
186 INIT_HLIST_NODE(&inode->i_hash);
187 INIT_LIST_HEAD(&inode->i_dentry);
188 INIT_LIST_HEAD(&inode->i_devices);
189 sema_init(&inode->i_sem, 1);
190 init_rwsem(&inode->i_alloc_sem);
191 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
192 spin_lock_init(&inode->i_data.tree_lock);
193 init_MUTEX(&inode->i_data.i_shared_sem);
194 atomic_set(&inode->i_data.truncate_count, 0);
195 INIT_LIST_HEAD(&inode->i_data.private_list);
196 spin_lock_init(&inode->i_data.private_lock);
197 INIT_LIST_HEAD(&inode->i_data.i_mmap);
198 INIT_LIST_HEAD(&inode->i_data.i_mmap_shared);
199 spin_lock_init(&inode->i_lock);
200 i_size_ordered_init(inode);
203 EXPORT_SYMBOL(inode_init_once);
205 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
207 struct inode * inode = (struct inode *) foo;
209 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
210 SLAB_CTOR_CONSTRUCTOR)
211 inode_init_once(inode);
215 * inode_lock must be held
217 void __iget(struct inode * inode)
219 if (atomic_read(&inode->i_count)) {
220 atomic_inc(&inode->i_count);
223 atomic_inc(&inode->i_count);
224 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
225 list_move(&inode->i_list, &inode_in_use);
226 inodes_stat.nr_unused--;
230 * clear_inode - clear an inode
231 * @inode: inode to clear
233 * This is called by the filesystem to tell us
234 * that the inode is no longer useful. We just
235 * terminate it with extreme prejudice.
237 void clear_inode(struct inode *inode)
239 invalidate_inode_buffers(inode);
241 if (inode->i_data.nrpages)
243 if (!(inode->i_state & I_FREEING))
245 if (inode->i_state & I_CLEAR)
247 wait_on_inode(inode);
249 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
250 inode->i_sb->s_op->clear_inode(inode);
255 inode->i_state = I_CLEAR;
258 EXPORT_SYMBOL(clear_inode);
261 * dispose_list - dispose of the contents of a local list
262 * @head: the head of the list to free
264 * Dispose-list gets a local list with local inodes in it, so it doesn't
265 * need to worry about list corruption and SMP locks.
267 static void dispose_list(struct list_head *head)
271 while (!list_empty(head)) {
274 inode = list_entry(head->next, struct inode, i_list);
275 list_del(&inode->i_list);
277 if (inode->i_data.nrpages)
278 truncate_inode_pages(&inode->i_data, 0);
280 destroy_inode(inode);
283 spin_lock(&inode_lock);
284 inodes_stat.nr_inodes -= nr_disposed;
285 spin_unlock(&inode_lock);
289 * Invalidate all inodes for a device.
291 static int invalidate_list(struct list_head *head, struct super_block * sb, struct list_head * dispose)
293 struct list_head *next;
294 int busy = 0, count = 0;
298 struct list_head * tmp = next;
299 struct inode * inode;
304 inode = list_entry(tmp, struct inode, i_list);
305 if (inode->i_sb != sb)
307 invalidate_inode_buffers(inode);
308 if (!atomic_read(&inode->i_count)) {
309 hlist_del_init(&inode->i_hash);
310 list_move(&inode->i_list, dispose);
311 inode->i_state |= I_FREEING;
317 /* only unused inodes may be cached with i_count zero */
318 inodes_stat.nr_unused -= count;
323 * This is a two-stage process. First we collect all
324 * offending inodes onto the throw-away list, and in
325 * the second stage we actually dispose of them. This
326 * is because we don't want to sleep while messing
327 * with the global lists..
331 * invalidate_inodes - discard the inodes on a device
334 * Discard all of the inodes for a given superblock. If the discard
335 * fails because there are busy inodes then a non zero value is returned.
336 * If the discard is successful all the inodes have been discarded.
338 int invalidate_inodes(struct super_block * sb)
341 LIST_HEAD(throw_away);
344 spin_lock(&inode_lock);
345 busy = invalidate_list(&inode_in_use, sb, &throw_away);
346 busy |= invalidate_list(&inode_unused, sb, &throw_away);
347 busy |= invalidate_list(&sb->s_dirty, sb, &throw_away);
348 busy |= invalidate_list(&sb->s_io, sb, &throw_away);
349 spin_unlock(&inode_lock);
351 dispose_list(&throw_away);
357 EXPORT_SYMBOL(invalidate_inodes);
359 int __invalidate_device(struct block_device *bdev, int do_sync)
361 struct super_block *sb;
368 sb = get_super(bdev);
371 * no need to lock the super, get_super holds the
372 * read semaphore so the filesystem cannot go away
373 * under us (->put_super runs with the write lock
376 shrink_dcache_sb(sb);
377 res = invalidate_inodes(sb);
380 invalidate_bdev(bdev, 0);
384 EXPORT_SYMBOL(__invalidate_device);
386 static int can_unuse(struct inode *inode)
390 if (inode_has_buffers(inode))
392 if (atomic_read(&inode->i_count))
394 if (inode->i_data.nrpages)
400 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
401 * a temporary list and then are freed outside inode_lock by dispose_list().
403 * Any inodes which are pinned purely because of attached pagecache have their
404 * pagecache removed. We expect the final iput() on that inode to add it to
405 * the front of the inode_unused list. So look for it there and if the
406 * inode is still freeable, proceed. The right inode is found 99.9% of the
407 * time in testing on a 4-way.
409 * If the inode has metadata buffers attached to mapping->private_list then
410 * try to remove them.
412 static void prune_icache(int nr_to_scan)
417 unsigned long reap = 0;
420 spin_lock(&inode_lock);
421 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
424 if (list_empty(&inode_unused))
427 inode = list_entry(inode_unused.prev, struct inode, i_list);
429 if (inode->i_state || atomic_read(&inode->i_count)) {
430 list_move(&inode->i_list, &inode_unused);
433 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
435 spin_unlock(&inode_lock);
436 if (remove_inode_buffers(inode))
437 reap += invalidate_inode_pages(&inode->i_data);
439 spin_lock(&inode_lock);
441 if (inode != list_entry(inode_unused.next,
442 struct inode, i_list))
443 continue; /* wrong inode or list_empty */
444 if (!can_unuse(inode))
447 hlist_del_init(&inode->i_hash);
448 list_move(&inode->i_list, &freeable);
449 inode->i_state |= I_FREEING;
452 inodes_stat.nr_unused -= nr_pruned;
453 spin_unlock(&inode_lock);
455 dispose_list(&freeable);
458 if (current_is_kswapd())
459 mod_page_state(kswapd_inodesteal, reap);
461 mod_page_state(pginodesteal, reap);
465 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
466 * "unused" means that no dentries are referring to the inodes: the files are
467 * not open and the dcache references to those inodes have already been
470 * This function is passed the number of inodes to scan, and it returns the
471 * total number of remaining possibly-reclaimable inodes.
473 static int shrink_icache_memory(int nr, unsigned int gfp_mask)
477 * Nasty deadlock avoidance. We may hold various FS locks,
478 * and we don't want to recurse into the FS that called us
479 * in clear_inode() and friends..
481 if (gfp_mask & __GFP_FS)
484 return inodes_stat.nr_unused;
487 static void __wait_on_freeing_inode(struct inode *inode);
489 * Called with the inode lock held.
490 * NOTE: we are not increasing the inode-refcount, you must call __iget()
491 * by hand after calling find_inode now! This simplifies iunique and won't
492 * add any additional branch in the common code.
494 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
496 struct hlist_node *node;
497 struct inode * inode = NULL;
500 hlist_for_each (node, head) {
501 inode = hlist_entry(node, struct inode, i_hash);
502 if (inode->i_sb != sb)
504 if (!test(inode, data))
506 if (inode->i_state & (I_FREEING|I_CLEAR)) {
507 __wait_on_freeing_inode(inode);
512 return node ? inode : NULL;
516 * find_inode_fast is the fast path version of find_inode, see the comment at
517 * iget_locked for details.
519 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
521 struct hlist_node *node;
522 struct inode * inode = NULL;
525 hlist_for_each (node, head) {
526 inode = hlist_entry(node, struct inode, i_hash);
527 if (inode->i_ino != ino)
529 if (inode->i_sb != sb)
531 if (inode->i_state & (I_FREEING|I_CLEAR)) {
532 __wait_on_freeing_inode(inode);
537 return node ? inode : NULL;
541 * new_inode - obtain an inode
544 * Allocates a new inode for given superblock.
546 struct inode *new_inode(struct super_block *sb)
548 static unsigned long last_ino;
549 struct inode * inode;
551 spin_lock_prefetch(&inode_lock);
553 inode = alloc_inode(sb);
555 spin_lock(&inode_lock);
556 inodes_stat.nr_inodes++;
557 list_add(&inode->i_list, &inode_in_use);
558 inode->i_ino = ++last_ino;
560 spin_unlock(&inode_lock);
565 EXPORT_SYMBOL(new_inode);
567 void unlock_new_inode(struct inode *inode)
570 * This is special! We do not need the spinlock
571 * when clearing I_LOCK, because we're guaranteed
572 * that nobody else tries to do anything about the
573 * state of the inode when it is locked, as we
574 * just created it (so there can be no old holders
575 * that haven't tested I_LOCK).
577 inode->i_state &= ~(I_LOCK|I_NEW);
578 wake_up_inode(inode);
581 EXPORT_SYMBOL(unlock_new_inode);
584 * This is called without the inode lock held.. Be careful.
586 * We no longer cache the sb_flags in i_flags - see fs.h
587 * -- rmk@arm.uk.linux.org
589 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)
591 struct inode * inode;
593 inode = alloc_inode(sb);
597 spin_lock(&inode_lock);
598 /* We released the lock, so.. */
599 old = find_inode(sb, head, test, data);
601 if (set(inode, data))
604 inodes_stat.nr_inodes++;
605 list_add(&inode->i_list, &inode_in_use);
606 hlist_add_head(&inode->i_hash, head);
607 inode->i_state = I_LOCK|I_NEW;
608 spin_unlock(&inode_lock);
610 /* Return the locked inode with I_NEW set, the
611 * caller is responsible for filling in the contents
617 * Uhhuh, somebody else created the same inode under
618 * us. Use the old inode instead of the one we just
622 spin_unlock(&inode_lock);
623 destroy_inode(inode);
625 wait_on_inode(inode);
630 spin_unlock(&inode_lock);
631 destroy_inode(inode);
636 * get_new_inode_fast is the fast path version of get_new_inode, see the
637 * comment at iget_locked for details.
639 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
641 struct inode * inode;
643 inode = alloc_inode(sb);
647 spin_lock(&inode_lock);
648 /* We released the lock, so.. */
649 old = find_inode_fast(sb, head, ino);
652 inodes_stat.nr_inodes++;
653 list_add(&inode->i_list, &inode_in_use);
654 hlist_add_head(&inode->i_hash, head);
655 inode->i_state = I_LOCK|I_NEW;
656 spin_unlock(&inode_lock);
658 /* Return the locked inode with I_NEW set, the
659 * caller is responsible for filling in the contents
665 * Uhhuh, somebody else created the same inode under
666 * us. Use the old inode instead of the one we just
670 spin_unlock(&inode_lock);
671 destroy_inode(inode);
673 wait_on_inode(inode);
678 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
680 unsigned long tmp = hashval + ((unsigned long) sb / L1_CACHE_BYTES);
681 tmp = tmp + (tmp >> I_HASHBITS);
682 return tmp & I_HASHMASK;
685 /* Yeah, I know about quadratic hash. Maybe, later. */
688 * iunique - get a unique inode number
690 * @max_reserved: highest reserved inode number
692 * Obtain an inode number that is unique on the system for a given
693 * superblock. This is used by file systems that have no natural
694 * permanent inode numbering system. An inode number is returned that
695 * is higher than the reserved limit but unique.
698 * With a large number of inodes live on the file system this function
699 * currently becomes quite slow.
701 ino_t iunique(struct super_block *sb, ino_t max_reserved)
703 static ino_t counter;
705 struct hlist_head * head;
707 spin_lock(&inode_lock);
709 if (counter > max_reserved) {
710 head = inode_hashtable + hash(sb,counter);
712 inode = find_inode_fast(sb, head, res);
714 spin_unlock(&inode_lock);
718 counter = max_reserved + 1;
724 EXPORT_SYMBOL(iunique);
726 struct inode *igrab(struct inode *inode)
728 spin_lock(&inode_lock);
729 if (!(inode->i_state & I_FREEING))
733 * Handle the case where s_op->clear_inode is not been
734 * called yet, and somebody is calling igrab
735 * while the inode is getting freed.
738 spin_unlock(&inode_lock);
742 EXPORT_SYMBOL(igrab);
745 * ifind - internal function, you want ilookup5() or iget5().
746 * @sb: super block of file system to search
747 * @head: the head of the list to search
748 * @test: callback used for comparisons between inodes
749 * @data: opaque data pointer to pass to @test
751 * ifind() searches for the inode specified by @data in the inode
752 * cache. This is a generalized version of ifind_fast() for file systems where
753 * the inode number is not sufficient for unique identification of an inode.
755 * If the inode is in the cache, the inode is returned with an incremented
758 * Otherwise NULL is returned.
760 * Note, @test is called with the inode_lock held, so can't sleep.
762 static inline struct inode *ifind(struct super_block *sb,
763 struct hlist_head *head, int (*test)(struct inode *, void *),
768 spin_lock(&inode_lock);
769 inode = find_inode(sb, head, test, data);
772 spin_unlock(&inode_lock);
773 wait_on_inode(inode);
776 spin_unlock(&inode_lock);
781 * ifind_fast - internal function, you want ilookup() or iget().
782 * @sb: super block of file system to search
783 * @head: head of the list to search
784 * @ino: inode number to search for
786 * ifind_fast() searches for the inode @ino in the inode cache. This is for
787 * file systems where the inode number is sufficient for unique identification
790 * If the inode is in the cache, the inode is returned with an incremented
793 * Otherwise NULL is returned.
795 static inline struct inode *ifind_fast(struct super_block *sb,
796 struct hlist_head *head, unsigned long ino)
800 spin_lock(&inode_lock);
801 inode = find_inode_fast(sb, head, ino);
804 spin_unlock(&inode_lock);
805 wait_on_inode(inode);
808 spin_unlock(&inode_lock);
813 * ilookup5 - search for an inode in the inode cache
814 * @sb: super block of file system to search
815 * @hashval: hash value (usually inode number) to search for
816 * @test: callback used for comparisons between inodes
817 * @data: opaque data pointer to pass to @test
819 * ilookup5() uses ifind() to search for the inode specified by @hashval and
820 * @data in the inode cache. This is a generalized version of ilookup() for
821 * file systems where the inode number is not sufficient for unique
822 * identification of an inode.
824 * If the inode is in the cache, the inode is returned with an incremented
827 * Otherwise NULL is returned.
829 * Note, @test is called with the inode_lock held, so can't sleep.
831 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
832 int (*test)(struct inode *, void *), void *data)
834 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
836 return ifind(sb, head, test, data);
839 EXPORT_SYMBOL(ilookup5);
842 * ilookup - search for an inode in the inode cache
843 * @sb: super block of file system to search
844 * @ino: inode number to search for
846 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
847 * This is for file systems where the inode number is sufficient for unique
848 * identification of an inode.
850 * If the inode is in the cache, the inode is returned with an incremented
853 * Otherwise NULL is returned.
855 struct inode *ilookup(struct super_block *sb, unsigned long ino)
857 struct hlist_head *head = inode_hashtable + hash(sb, ino);
859 return ifind_fast(sb, head, ino);
862 EXPORT_SYMBOL(ilookup);
865 * iget5_locked - obtain an inode from a mounted file system
866 * @sb: super block of file system
867 * @hashval: hash value (usually inode number) to get
868 * @test: callback used for comparisons between inodes
869 * @set: callback used to initialize a new struct inode
870 * @data: opaque data pointer to pass to @test and @set
872 * This is iget() without the read_inode() portion of get_new_inode().
874 * iget5_locked() uses ifind() to search for the inode specified by @hashval
875 * and @data in the inode cache and if present it is returned with an increased
876 * reference count. This is a generalized version of iget_locked() for file
877 * systems where the inode number is not sufficient for unique identification
880 * If the inode is not in cache, get_new_inode() is called to allocate a new
881 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
882 * file system gets to fill it in before unlocking it via unlock_new_inode().
884 * Note both @test and @set are called with the inode_lock held, so can't sleep.
886 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
887 int (*test)(struct inode *, void *),
888 int (*set)(struct inode *, void *), void *data)
890 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
893 inode = ifind(sb, head, test, data);
897 * get_new_inode() will do the right thing, re-trying the search
898 * in case it had to block at any point.
900 return get_new_inode(sb, head, test, set, data);
903 EXPORT_SYMBOL(iget5_locked);
906 * iget_locked - obtain an inode from a mounted file system
907 * @sb: super block of file system
908 * @ino: inode number to get
910 * This is iget() without the read_inode() portion of get_new_inode_fast().
912 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
913 * the inode cache and if present it is returned with an increased reference
914 * count. This is for file systems where the inode number is sufficient for
915 * unique identification of an inode.
917 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
918 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
919 * The file system gets to fill it in before unlocking it via
920 * unlock_new_inode().
922 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
924 struct hlist_head *head = inode_hashtable + hash(sb, ino);
927 inode = ifind_fast(sb, head, ino);
931 * get_new_inode_fast() will do the right thing, re-trying the search
932 * in case it had to block at any point.
934 return get_new_inode_fast(sb, head, ino);
937 EXPORT_SYMBOL(iget_locked);
940 * __insert_inode_hash - hash an inode
941 * @inode: unhashed inode
942 * @hashval: unsigned long value used to locate this object in the
945 * Add an inode to the inode hash for this superblock.
947 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
949 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
950 spin_lock(&inode_lock);
951 hlist_add_head(&inode->i_hash, head);
952 spin_unlock(&inode_lock);
955 EXPORT_SYMBOL(__insert_inode_hash);
958 * remove_inode_hash - remove an inode from the hash
959 * @inode: inode to unhash
961 * Remove an inode from the superblock.
963 void remove_inode_hash(struct inode *inode)
965 spin_lock(&inode_lock);
966 hlist_del_init(&inode->i_hash);
967 spin_unlock(&inode_lock);
970 EXPORT_SYMBOL(remove_inode_hash);
973 * Tell the filesystem that this inode is no longer of any interest and should
974 * be completely destroyed.
976 * We leave the inode in the inode hash table until *after* the filesystem's
977 * ->delete_inode completes. This ensures that an iget (such as nfsd might
978 * instigate) will always find up-to-date information either in the hash or on
981 * I_FREEING is set so that no-one will take a new reference to the inode while
982 * it is being deleted.
984 void generic_delete_inode(struct inode *inode)
986 struct super_operations *op = inode->i_sb->s_op;
988 list_del_init(&inode->i_list);
989 inode->i_state|=I_FREEING;
990 inodes_stat.nr_inodes--;
991 spin_unlock(&inode_lock);
993 if (inode->i_data.nrpages)
994 truncate_inode_pages(&inode->i_data, 0);
996 security_inode_delete(inode);
998 if (op->delete_inode) {
999 void (*delete)(struct inode *) = op->delete_inode;
1000 if (!is_bad_inode(inode))
1002 /* s_op->delete_inode internally recalls clear_inode() */
1006 spin_lock(&inode_lock);
1007 hlist_del_init(&inode->i_hash);
1008 spin_unlock(&inode_lock);
1009 wake_up_inode(inode);
1010 if (inode->i_state != I_CLEAR)
1012 destroy_inode(inode);
1015 EXPORT_SYMBOL(generic_delete_inode);
1017 static void generic_forget_inode(struct inode *inode)
1019 struct super_block *sb = inode->i_sb;
1021 if (!hlist_unhashed(&inode->i_hash)) {
1022 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1023 list_move(&inode->i_list, &inode_unused);
1024 inodes_stat.nr_unused++;
1025 spin_unlock(&inode_lock);
1026 if (!sb || (sb->s_flags & MS_ACTIVE))
1028 write_inode_now(inode, 1);
1029 spin_lock(&inode_lock);
1030 inodes_stat.nr_unused--;
1031 hlist_del_init(&inode->i_hash);
1033 list_del_init(&inode->i_list);
1034 inode->i_state|=I_FREEING;
1035 inodes_stat.nr_inodes--;
1036 spin_unlock(&inode_lock);
1037 if (inode->i_data.nrpages)
1038 truncate_inode_pages(&inode->i_data, 0);
1040 destroy_inode(inode);
1044 * Normal UNIX filesystem behaviour: delete the
1045 * inode when the usage count drops to zero, and
1048 static void generic_drop_inode(struct inode *inode)
1050 if (!inode->i_nlink)
1051 generic_delete_inode(inode);
1053 generic_forget_inode(inode);
1057 * Called when we're dropping the last reference
1060 * Call the FS "drop()" function, defaulting to
1061 * the legacy UNIX filesystem behaviour..
1063 * NOTE! NOTE! NOTE! We're called with the inode lock
1064 * held, and the drop function is supposed to release
1067 static inline void iput_final(struct inode *inode)
1069 struct super_operations *op = inode->i_sb->s_op;
1070 void (*drop)(struct inode *) = generic_drop_inode;
1072 if (op && op->drop_inode)
1073 drop = op->drop_inode;
1078 * iput - put an inode
1079 * @inode: inode to put
1081 * Puts an inode, dropping its usage count. If the inode use count hits
1082 * zero the inode is also then freed and may be destroyed.
1084 void iput(struct inode *inode)
1087 struct super_operations *op = inode->i_sb->s_op;
1089 if (inode->i_state == I_CLEAR)
1092 if (op && op->put_inode)
1093 op->put_inode(inode);
1095 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1100 EXPORT_SYMBOL(iput);
1103 * bmap - find a block number in a file
1104 * @inode: inode of file
1105 * @block: block to find
1107 * Returns the block number on the device holding the inode that
1108 * is the disk block number for the block of the file requested.
1109 * That is, asked for block 4 of inode 1 the function will return the
1110 * disk block relative to the disk start that holds that block of the
1113 sector_t bmap(struct inode * inode, sector_t block)
1116 if (inode->i_mapping->a_ops->bmap)
1117 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1121 EXPORT_SYMBOL(bmap);
1124 * Return true if the filesystem which backs this inode considers the two
1125 * passed timespecs to be sufficiently different to warrant flushing the
1126 * altered time out to disk.
1128 static int inode_times_differ(struct inode *inode,
1129 struct timespec *old, struct timespec *new)
1131 if (IS_ONE_SECOND(inode))
1132 return old->tv_sec != new->tv_sec;
1133 return !timespec_equal(old, new);
1137 * update_atime - update the access time
1138 * @inode: inode accessed
1140 * Update the accessed time on an inode and mark it for writeback.
1141 * This function automatically handles read only file systems and media,
1142 * as well as the "noatime" flag and inode specific "noatime" markers.
1144 void update_atime(struct inode *inode)
1146 struct timespec now;
1148 if (IS_NOATIME(inode))
1150 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1152 if (IS_RDONLY(inode))
1155 now = current_kernel_time();
1156 if (inode_times_differ(inode, &inode->i_atime, &now)) {
1157 inode->i_atime = now;
1158 mark_inode_dirty_sync(inode);
1160 if (!timespec_equal(&inode->i_atime, &now))
1161 inode->i_atime = now;
1165 EXPORT_SYMBOL(update_atime);
1168 * inode_update_time - update mtime and ctime time
1169 * @inode: inode accessed
1170 * @ctime_too: update ctime too
1172 * Update the mtime time on an inode and mark it for writeback.
1173 * When ctime_too is specified update the ctime too.
1176 void inode_update_time(struct inode *inode, int ctime_too)
1178 struct timespec now;
1181 if (IS_NOCMTIME(inode))
1183 if (IS_RDONLY(inode))
1186 now = current_kernel_time();
1188 if (inode_times_differ(inode, &inode->i_mtime, &now))
1190 inode->i_mtime = now;
1193 if (inode_times_differ(inode, &inode->i_ctime, &now))
1195 inode->i_ctime = now;
1198 mark_inode_dirty_sync(inode);
1201 EXPORT_SYMBOL(inode_update_time);
1203 int inode_needs_sync(struct inode *inode)
1207 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1212 EXPORT_SYMBOL(inode_needs_sync);
1215 * Quota functions that want to walk the inode lists..
1219 /* Function back in dquot.c */
1220 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1222 void remove_dquot_ref(struct super_block *sb, int type, struct list_head *tofree_head)
1224 struct inode *inode;
1225 struct list_head *act_head;
1228 return; /* nothing to do */
1229 spin_lock(&inode_lock); /* This lock is for inodes code */
1230 /* We don't have to lock against quota code - test IS_QUOTAINIT is just for speedup... */
1232 list_for_each(act_head, &inode_in_use) {
1233 inode = list_entry(act_head, struct inode, i_list);
1234 if (inode->i_sb == sb && IS_QUOTAINIT(inode))
1235 remove_inode_dquot_ref(inode, type, tofree_head);
1237 list_for_each(act_head, &inode_unused) {
1238 inode = list_entry(act_head, struct inode, i_list);
1239 if (inode->i_sb == sb && IS_QUOTAINIT(inode))
1240 remove_inode_dquot_ref(inode, type, tofree_head);
1242 list_for_each(act_head, &sb->s_dirty) {
1243 inode = list_entry(act_head, struct inode, i_list);
1244 if (IS_QUOTAINIT(inode))
1245 remove_inode_dquot_ref(inode, type, tofree_head);
1247 list_for_each(act_head, &sb->s_io) {
1248 inode = list_entry(act_head, struct inode, i_list);
1249 if (IS_QUOTAINIT(inode))
1250 remove_inode_dquot_ref(inode, type, tofree_head);
1252 spin_unlock(&inode_lock);
1258 * Hashed waitqueues for wait_on_inode(). The table is pretty small - the
1259 * kernel doesn't lock many inodes at the same time.
1261 #define I_WAIT_TABLE_ORDER 3
1262 static struct i_wait_queue_head {
1263 wait_queue_head_t wqh;
1264 } ____cacheline_aligned_in_smp i_wait_queue_heads[1<<I_WAIT_TABLE_ORDER];
1267 * Return the address of the waitqueue_head to be used for this inode
1269 static wait_queue_head_t *i_waitq_head(struct inode *inode)
1271 return &i_wait_queue_heads[hash_ptr(inode, I_WAIT_TABLE_ORDER)].wqh;
1274 void __wait_on_inode(struct inode *inode)
1276 DECLARE_WAITQUEUE(wait, current);
1277 wait_queue_head_t *wq = i_waitq_head(inode);
1279 add_wait_queue(wq, &wait);
1281 set_current_state(TASK_UNINTERRUPTIBLE);
1282 if (inode->i_state & I_LOCK) {
1286 remove_wait_queue(wq, &wait);
1287 __set_current_state(TASK_RUNNING);
1291 * If we try to find an inode in the inode hash while it is being deleted, we
1292 * have to wait until the filesystem completes its deletion before reporting
1293 * that it isn't found. This is because iget will immediately call
1294 * ->read_inode, and we want to be sure that evidence of the deletion is found
1297 * This call might return early if an inode which shares the waitq is woken up.
1298 * This is most easily handled by the caller which will loop around again
1299 * looking for the inode.
1301 * This is called with inode_lock held.
1303 static void __wait_on_freeing_inode(struct inode *inode)
1305 DECLARE_WAITQUEUE(wait, current);
1306 wait_queue_head_t *wq = i_waitq_head(inode);
1308 add_wait_queue(wq, &wait);
1309 set_current_state(TASK_UNINTERRUPTIBLE);
1310 spin_unlock(&inode_lock);
1312 remove_wait_queue(wq, &wait);
1313 spin_lock(&inode_lock);
1316 void wake_up_inode(struct inode *inode)
1318 wait_queue_head_t *wq = i_waitq_head(inode);
1321 * Prevent speculative execution through spin_unlock(&inode_lock);
1324 if (waitqueue_active(wq))
1328 static __initdata unsigned long ihash_entries;
1329 static int __init set_ihash_entries(char *str)
1333 ihash_entries = simple_strtoul(str, &str, 0);
1336 __setup("ihash_entries=", set_ihash_entries);
1339 * Initialize the waitqueues and inode hash table.
1341 void __init inode_init(unsigned long mempages)
1343 struct hlist_head *head;
1344 unsigned long order;
1345 unsigned int nr_hash;
1348 for (i = 0; i < ARRAY_SIZE(i_wait_queue_heads); i++)
1349 init_waitqueue_head(&i_wait_queue_heads[i].wqh);
1352 ihash_entries = PAGE_SHIFT < 14 ?
1353 mempages >> (14 - PAGE_SHIFT) :
1354 mempages << (PAGE_SHIFT - 14);
1356 ihash_entries *= sizeof(struct hlist_head);
1357 for (order = 0; ((1UL << order) << PAGE_SHIFT) < ihash_entries; order++)
1363 nr_hash = (1UL << order) * PAGE_SIZE /
1364 sizeof(struct hlist_head);
1365 i_hash_mask = (nr_hash - 1);
1369 while ((tmp >>= 1UL) != 0UL)
1372 inode_hashtable = (struct hlist_head *)
1373 __get_free_pages(GFP_ATOMIC, order);
1374 } while (inode_hashtable == NULL && --order >= 0);
1376 printk("Inode-cache hash table entries: %d (order: %ld, %ld bytes)\n",
1377 nr_hash, order, (PAGE_SIZE << order));
1379 if (!inode_hashtable)
1380 panic("Failed to allocate inode hash table\n");
1382 head = inode_hashtable;
1385 INIT_HLIST_HEAD(head);
1390 /* inode slab cache */
1391 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1392 0, SLAB_HWCACHE_ALIGN, init_once,
1395 panic("cannot create inode slab cache");
1397 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1400 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1402 inode->i_mode = mode;
1403 if (S_ISCHR(mode)) {
1404 inode->i_fop = &def_chr_fops;
1405 inode->i_rdev = rdev;
1406 } else if (S_ISBLK(mode)) {
1407 inode->i_fop = &def_blk_fops;
1408 inode->i_rdev = rdev;
1409 } else if (S_ISFIFO(mode))
1410 inode->i_fop = &def_fifo_fops;
1411 else if (S_ISSOCK(mode))
1412 inode->i_fop = &bad_sock_fops;
1414 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1418 EXPORT_SYMBOL(init_special_inode);