4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/slab.h>
12 #include <linux/writeback.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/wait.h>
16 #include <linux/rwsem.h>
17 #include <linux/hash.h>
18 #include <linux/swap.h>
19 #include <linux/security.h>
20 #include <linux/pagemap.h>
21 #include <linux/cdev.h>
22 #include <linux/bootmem.h>
23 #include <linux/fsnotify.h>
24 #include <linux/mount.h>
25 #include <linux/async.h>
26 #include <linux/posix_acl.h>
27 #include <linux/ima.h>
30 * This is needed for the following functions:
34 * FIXME: remove all knowledge of the buffer layer from this file
36 #include <linux/buffer_head.h>
39 * New inode.c implementation.
41 * This implementation has the basic premise of trying
42 * to be extremely low-overhead and SMP-safe, yet be
43 * simple enough to be "obviously correct".
48 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
50 /* #define INODE_PARANOIA 1 */
51 /* #define INODE_DEBUG 1 */
54 * Inode lookup is no longer as critical as it used to be:
55 * most of the lookups are going to be through the dcache.
57 #define I_HASHBITS i_hash_shift
58 #define I_HASHMASK i_hash_mask
60 static unsigned int i_hash_mask __read_mostly;
61 static unsigned int i_hash_shift __read_mostly;
64 * Each inode can be on two separate lists. One is
65 * the hash list of the inode, used for lookups. The
66 * other linked list is the "type" list:
67 * "in_use" - valid inode, i_count > 0, i_nlink > 0
68 * "dirty" - as "in_use" but also dirty
69 * "unused" - valid inode, i_count = 0
71 * A "dirty" list is maintained for each super block,
72 * allowing for low-overhead inode sync() operations.
75 static LIST_HEAD(inode_lru);
76 static struct hlist_head *inode_hashtable __read_mostly;
79 * A simple spinlock to protect the list manipulations.
81 * NOTE! You also have to own the lock if you change
82 * the i_state of an inode while it is in use..
84 DEFINE_SPINLOCK(inode_lock);
87 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
88 * icache shrinking path, and the umount path. Without this exclusion,
89 * by the time prune_icache calls iput for the inode whose pages it has
90 * been invalidating, or by the time it calls clear_inode & destroy_inode
91 * from its final dispose_list, the struct super_block they refer to
92 * (for inode->i_sb->s_op) may already have been freed and reused.
94 * We make this an rwsem because the fastpath is icache shrinking. In
95 * some cases a filesystem may be doing a significant amount of work in
96 * its inode reclaim code, so this should improve parallelism.
98 static DECLARE_RWSEM(iprune_sem);
101 * Statistics gathering..
103 struct inodes_stat_t inodes_stat;
105 static DEFINE_PER_CPU(unsigned int, nr_inodes);
107 static struct kmem_cache *inode_cachep __read_mostly;
109 static int get_nr_inodes(void)
113 for_each_possible_cpu(i)
114 sum += per_cpu(nr_inodes, i);
115 return sum < 0 ? 0 : sum;
118 static inline int get_nr_inodes_unused(void)
120 return inodes_stat.nr_unused;
123 int get_nr_dirty_inodes(void)
125 /* not actually dirty inodes, but a wild approximation */
126 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
127 return nr_dirty > 0 ? nr_dirty : 0;
131 * Handle nr_inode sysctl
134 int proc_nr_inodes(ctl_table *table, int write,
135 void __user *buffer, size_t *lenp, loff_t *ppos)
137 inodes_stat.nr_inodes = get_nr_inodes();
138 return proc_dointvec(table, write, buffer, lenp, ppos);
142 static void wake_up_inode(struct inode *inode)
145 * Prevent speculative execution through spin_unlock(&inode_lock);
148 wake_up_bit(&inode->i_state, __I_NEW);
152 * inode_init_always - perform inode structure intialisation
153 * @sb: superblock inode belongs to
154 * @inode: inode to initialise
156 * These are initializations that need to be done on every inode
157 * allocation as the fields are not initialised by slab allocation.
159 int inode_init_always(struct super_block *sb, struct inode *inode)
161 static const struct address_space_operations empty_aops;
162 static const struct inode_operations empty_iops;
163 static const struct file_operations empty_fops;
164 struct address_space *const mapping = &inode->i_data;
167 inode->i_blkbits = sb->s_blocksize_bits;
169 atomic_set(&inode->i_count, 1);
170 inode->i_op = &empty_iops;
171 inode->i_fop = &empty_fops;
175 atomic_set(&inode->i_writecount, 0);
179 inode->i_generation = 0;
181 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
183 inode->i_pipe = NULL;
184 inode->i_bdev = NULL;
185 inode->i_cdev = NULL;
187 inode->dirtied_when = 0;
189 if (security_inode_alloc(inode))
191 spin_lock_init(&inode->i_lock);
192 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
194 mutex_init(&inode->i_mutex);
195 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
197 init_rwsem(&inode->i_alloc_sem);
198 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
200 mapping->a_ops = &empty_aops;
201 mapping->host = inode;
203 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
204 mapping->assoc_mapping = NULL;
205 mapping->backing_dev_info = &default_backing_dev_info;
206 mapping->writeback_index = 0;
209 * If the block_device provides a backing_dev_info for client
210 * inodes then use that. Otherwise the inode share the bdev's
214 struct backing_dev_info *bdi;
216 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
217 mapping->backing_dev_info = bdi;
219 inode->i_private = NULL;
220 inode->i_mapping = mapping;
221 #ifdef CONFIG_FS_POSIX_ACL
222 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
225 #ifdef CONFIG_FSNOTIFY
226 inode->i_fsnotify_mask = 0;
229 this_cpu_inc(nr_inodes);
235 EXPORT_SYMBOL(inode_init_always);
237 static struct inode *alloc_inode(struct super_block *sb)
241 if (sb->s_op->alloc_inode)
242 inode = sb->s_op->alloc_inode(sb);
244 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
249 if (unlikely(inode_init_always(sb, inode))) {
250 if (inode->i_sb->s_op->destroy_inode)
251 inode->i_sb->s_op->destroy_inode(inode);
253 kmem_cache_free(inode_cachep, inode);
260 void __destroy_inode(struct inode *inode)
262 BUG_ON(inode_has_buffers(inode));
263 security_inode_free(inode);
264 fsnotify_inode_delete(inode);
265 #ifdef CONFIG_FS_POSIX_ACL
266 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
267 posix_acl_release(inode->i_acl);
268 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
269 posix_acl_release(inode->i_default_acl);
271 this_cpu_dec(nr_inodes);
273 EXPORT_SYMBOL(__destroy_inode);
275 static void i_callback(struct rcu_head *head)
277 struct inode *inode = container_of(head, struct inode, i_rcu);
278 INIT_LIST_HEAD(&inode->i_dentry);
279 kmem_cache_free(inode_cachep, inode);
282 static void destroy_inode(struct inode *inode)
284 BUG_ON(!list_empty(&inode->i_lru));
285 __destroy_inode(inode);
286 if (inode->i_sb->s_op->destroy_inode)
287 inode->i_sb->s_op->destroy_inode(inode);
289 call_rcu(&inode->i_rcu, i_callback);
293 * These are initializations that only need to be done
294 * once, because the fields are idempotent across use
295 * of the inode, so let the slab aware of that.
297 void inode_init_once(struct inode *inode)
299 memset(inode, 0, sizeof(*inode));
300 INIT_HLIST_NODE(&inode->i_hash);
301 INIT_LIST_HEAD(&inode->i_dentry);
302 INIT_LIST_HEAD(&inode->i_devices);
303 INIT_LIST_HEAD(&inode->i_wb_list);
304 INIT_LIST_HEAD(&inode->i_lru);
305 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
306 spin_lock_init(&inode->i_data.tree_lock);
307 spin_lock_init(&inode->i_data.i_mmap_lock);
308 INIT_LIST_HEAD(&inode->i_data.private_list);
309 spin_lock_init(&inode->i_data.private_lock);
310 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
311 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
312 i_size_ordered_init(inode);
313 #ifdef CONFIG_FSNOTIFY
314 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
317 EXPORT_SYMBOL(inode_init_once);
319 static void init_once(void *foo)
321 struct inode *inode = (struct inode *) foo;
323 inode_init_once(inode);
327 * inode_lock must be held
329 void __iget(struct inode *inode)
331 atomic_inc(&inode->i_count);
335 * get additional reference to inode; caller must already hold one.
337 void ihold(struct inode *inode)
339 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
341 EXPORT_SYMBOL(ihold);
343 static void inode_lru_list_add(struct inode *inode)
345 if (list_empty(&inode->i_lru)) {
346 list_add(&inode->i_lru, &inode_lru);
347 inodes_stat.nr_unused++;
351 static void inode_lru_list_del(struct inode *inode)
353 if (!list_empty(&inode->i_lru)) {
354 list_del_init(&inode->i_lru);
355 inodes_stat.nr_unused--;
359 static inline void __inode_sb_list_add(struct inode *inode)
361 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
365 * inode_sb_list_add - add inode to the superblock list of inodes
366 * @inode: inode to add
368 void inode_sb_list_add(struct inode *inode)
370 spin_lock(&inode_lock);
371 __inode_sb_list_add(inode);
372 spin_unlock(&inode_lock);
374 EXPORT_SYMBOL_GPL(inode_sb_list_add);
376 static inline void __inode_sb_list_del(struct inode *inode)
378 list_del_init(&inode->i_sb_list);
381 static unsigned long hash(struct super_block *sb, unsigned long hashval)
385 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
387 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
388 return tmp & I_HASHMASK;
392 * __insert_inode_hash - hash an inode
393 * @inode: unhashed inode
394 * @hashval: unsigned long value used to locate this object in the
397 * Add an inode to the inode hash for this superblock.
399 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
401 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
403 spin_lock(&inode_lock);
404 hlist_add_head(&inode->i_hash, b);
405 spin_unlock(&inode_lock);
407 EXPORT_SYMBOL(__insert_inode_hash);
410 * __remove_inode_hash - remove an inode from the hash
411 * @inode: inode to unhash
413 * Remove an inode from the superblock.
415 static void __remove_inode_hash(struct inode *inode)
417 hlist_del_init(&inode->i_hash);
421 * remove_inode_hash - remove an inode from the hash
422 * @inode: inode to unhash
424 * Remove an inode from the superblock.
426 void remove_inode_hash(struct inode *inode)
428 spin_lock(&inode_lock);
429 hlist_del_init(&inode->i_hash);
430 spin_unlock(&inode_lock);
432 EXPORT_SYMBOL(remove_inode_hash);
434 void end_writeback(struct inode *inode)
437 BUG_ON(inode->i_data.nrpages);
438 BUG_ON(!list_empty(&inode->i_data.private_list));
439 BUG_ON(!(inode->i_state & I_FREEING));
440 BUG_ON(inode->i_state & I_CLEAR);
441 inode_sync_wait(inode);
442 /* don't need i_lock here, no concurrent mods to i_state */
443 inode->i_state = I_FREEING | I_CLEAR;
445 EXPORT_SYMBOL(end_writeback);
447 static void evict(struct inode *inode)
449 const struct super_operations *op = inode->i_sb->s_op;
451 if (op->evict_inode) {
452 op->evict_inode(inode);
454 if (inode->i_data.nrpages)
455 truncate_inode_pages(&inode->i_data, 0);
456 end_writeback(inode);
458 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
460 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
465 * dispose_list - dispose of the contents of a local list
466 * @head: the head of the list to free
468 * Dispose-list gets a local list with local inodes in it, so it doesn't
469 * need to worry about list corruption and SMP locks.
471 static void dispose_list(struct list_head *head)
473 while (!list_empty(head)) {
476 inode = list_first_entry(head, struct inode, i_lru);
477 list_del_init(&inode->i_lru);
481 spin_lock(&inode_lock);
482 __remove_inode_hash(inode);
483 __inode_sb_list_del(inode);
484 spin_unlock(&inode_lock);
486 wake_up_inode(inode);
487 destroy_inode(inode);
492 * evict_inodes - evict all evictable inodes for a superblock
493 * @sb: superblock to operate on
495 * Make sure that no inodes with zero refcount are retained. This is
496 * called by superblock shutdown after having MS_ACTIVE flag removed,
497 * so any inode reaching zero refcount during or after that call will
498 * be immediately evicted.
500 void evict_inodes(struct super_block *sb)
502 struct inode *inode, *next;
505 down_write(&iprune_sem);
507 spin_lock(&inode_lock);
508 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
509 if (atomic_read(&inode->i_count))
512 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
517 inode->i_state |= I_FREEING;
520 * Move the inode off the IO lists and LRU once I_FREEING is
521 * set so that it won't get moved back on there if it is dirty.
523 list_move(&inode->i_lru, &dispose);
524 list_del_init(&inode->i_wb_list);
525 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
526 inodes_stat.nr_unused--;
528 spin_unlock(&inode_lock);
530 dispose_list(&dispose);
531 up_write(&iprune_sem);
535 * invalidate_inodes - attempt to free all inodes on a superblock
536 * @sb: superblock to operate on
538 * Attempts to free all inodes for a given superblock. If there were any
539 * busy inodes return a non-zero value, else zero.
541 int invalidate_inodes(struct super_block *sb)
544 struct inode *inode, *next;
547 down_write(&iprune_sem);
549 spin_lock(&inode_lock);
550 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
551 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE))
553 if (atomic_read(&inode->i_count)) {
558 inode->i_state |= I_FREEING;
561 * Move the inode off the IO lists and LRU once I_FREEING is
562 * set so that it won't get moved back on there if it is dirty.
564 list_move(&inode->i_lru, &dispose);
565 list_del_init(&inode->i_wb_list);
566 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
567 inodes_stat.nr_unused--;
569 spin_unlock(&inode_lock);
571 dispose_list(&dispose);
572 up_write(&iprune_sem);
577 static int can_unuse(struct inode *inode)
579 if (inode->i_state & ~I_REFERENCED)
581 if (inode_has_buffers(inode))
583 if (atomic_read(&inode->i_count))
585 if (inode->i_data.nrpages)
591 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
592 * temporary list and then are freed outside inode_lock by dispose_list().
594 * Any inodes which are pinned purely because of attached pagecache have their
595 * pagecache removed. If the inode has metadata buffers attached to
596 * mapping->private_list then try to remove them.
598 * If the inode has the I_REFERENCED flag set, then it means that it has been
599 * used recently - the flag is set in iput_final(). When we encounter such an
600 * inode, clear the flag and move it to the back of the LRU so it gets another
601 * pass through the LRU before it gets reclaimed. This is necessary because of
602 * the fact we are doing lazy LRU updates to minimise lock contention so the
603 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
604 * with this flag set because they are the inodes that are out of order.
606 static void prune_icache(int nr_to_scan)
610 unsigned long reap = 0;
612 down_read(&iprune_sem);
613 spin_lock(&inode_lock);
614 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
617 if (list_empty(&inode_lru))
620 inode = list_entry(inode_lru.prev, struct inode, i_lru);
623 * Referenced or dirty inodes are still in use. Give them
624 * another pass through the LRU as we canot reclaim them now.
626 if (atomic_read(&inode->i_count) ||
627 (inode->i_state & ~I_REFERENCED)) {
628 list_del_init(&inode->i_lru);
629 inodes_stat.nr_unused--;
633 /* recently referenced inodes get one more pass */
634 if (inode->i_state & I_REFERENCED) {
635 list_move(&inode->i_lru, &inode_lru);
636 inode->i_state &= ~I_REFERENCED;
639 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
641 spin_unlock(&inode_lock);
642 if (remove_inode_buffers(inode))
643 reap += invalidate_mapping_pages(&inode->i_data,
646 spin_lock(&inode_lock);
648 if (inode != list_entry(inode_lru.next,
649 struct inode, i_lru))
650 continue; /* wrong inode or list_empty */
651 if (!can_unuse(inode))
654 WARN_ON(inode->i_state & I_NEW);
655 inode->i_state |= I_FREEING;
658 * Move the inode off the IO lists and LRU once I_FREEING is
659 * set so that it won't get moved back on there if it is dirty.
661 list_move(&inode->i_lru, &freeable);
662 list_del_init(&inode->i_wb_list);
663 inodes_stat.nr_unused--;
665 if (current_is_kswapd())
666 __count_vm_events(KSWAPD_INODESTEAL, reap);
668 __count_vm_events(PGINODESTEAL, reap);
669 spin_unlock(&inode_lock);
671 dispose_list(&freeable);
672 up_read(&iprune_sem);
676 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
677 * "unused" means that no dentries are referring to the inodes: the files are
678 * not open and the dcache references to those inodes have already been
681 * This function is passed the number of inodes to scan, and it returns the
682 * total number of remaining possibly-reclaimable inodes.
684 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
688 * Nasty deadlock avoidance. We may hold various FS locks,
689 * and we don't want to recurse into the FS that called us
690 * in clear_inode() and friends..
692 if (!(gfp_mask & __GFP_FS))
696 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
699 static struct shrinker icache_shrinker = {
700 .shrink = shrink_icache_memory,
701 .seeks = DEFAULT_SEEKS,
704 static void __wait_on_freeing_inode(struct inode *inode);
706 * Called with the inode lock held.
708 static struct inode *find_inode(struct super_block *sb,
709 struct hlist_head *head,
710 int (*test)(struct inode *, void *),
713 struct hlist_node *node;
714 struct inode *inode = NULL;
717 hlist_for_each_entry(inode, node, head, i_hash) {
718 if (inode->i_sb != sb)
720 if (!test(inode, data))
722 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
723 __wait_on_freeing_inode(inode);
733 * find_inode_fast is the fast path version of find_inode, see the comment at
734 * iget_locked for details.
736 static struct inode *find_inode_fast(struct super_block *sb,
737 struct hlist_head *head, unsigned long ino)
739 struct hlist_node *node;
740 struct inode *inode = NULL;
743 hlist_for_each_entry(inode, node, head, i_hash) {
744 if (inode->i_ino != ino)
746 if (inode->i_sb != sb)
748 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
749 __wait_on_freeing_inode(inode);
759 * Each cpu owns a range of LAST_INO_BATCH numbers.
760 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
761 * to renew the exhausted range.
763 * This does not significantly increase overflow rate because every CPU can
764 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
765 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
766 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
767 * overflow rate by 2x, which does not seem too significant.
769 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
770 * error if st_ino won't fit in target struct field. Use 32bit counter
771 * here to attempt to avoid that.
773 #define LAST_INO_BATCH 1024
774 static DEFINE_PER_CPU(unsigned int, last_ino);
776 unsigned int get_next_ino(void)
778 unsigned int *p = &get_cpu_var(last_ino);
779 unsigned int res = *p;
782 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
783 static atomic_t shared_last_ino;
784 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
786 res = next - LAST_INO_BATCH;
791 put_cpu_var(last_ino);
794 EXPORT_SYMBOL(get_next_ino);
797 * new_inode - obtain an inode
800 * Allocates a new inode for given superblock. The default gfp_mask
801 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
802 * If HIGHMEM pages are unsuitable or it is known that pages allocated
803 * for the page cache are not reclaimable or migratable,
804 * mapping_set_gfp_mask() must be called with suitable flags on the
805 * newly created inode's mapping
808 struct inode *new_inode(struct super_block *sb)
812 spin_lock_prefetch(&inode_lock);
814 inode = alloc_inode(sb);
816 spin_lock(&inode_lock);
817 __inode_sb_list_add(inode);
819 spin_unlock(&inode_lock);
823 EXPORT_SYMBOL(new_inode);
825 void unlock_new_inode(struct inode *inode)
827 #ifdef CONFIG_DEBUG_LOCK_ALLOC
828 if (S_ISDIR(inode->i_mode)) {
829 struct file_system_type *type = inode->i_sb->s_type;
831 /* Set new key only if filesystem hasn't already changed it */
832 if (!lockdep_match_class(&inode->i_mutex,
833 &type->i_mutex_key)) {
835 * ensure nobody is actually holding i_mutex
837 mutex_destroy(&inode->i_mutex);
838 mutex_init(&inode->i_mutex);
839 lockdep_set_class(&inode->i_mutex,
840 &type->i_mutex_dir_key);
845 * This is special! We do not need the spinlock when clearing I_NEW,
846 * because we're guaranteed that nobody else tries to do anything about
847 * the state of the inode when it is locked, as we just created it (so
848 * there can be no old holders that haven't tested I_NEW).
849 * However we must emit the memory barrier so that other CPUs reliably
850 * see the clearing of I_NEW after the other inode initialisation has
854 WARN_ON(!(inode->i_state & I_NEW));
855 inode->i_state &= ~I_NEW;
856 wake_up_inode(inode);
858 EXPORT_SYMBOL(unlock_new_inode);
861 * This is called without the inode lock held.. Be careful.
863 * We no longer cache the sb_flags in i_flags - see fs.h
864 * -- rmk@arm.uk.linux.org
866 static struct inode *get_new_inode(struct super_block *sb,
867 struct hlist_head *head,
868 int (*test)(struct inode *, void *),
869 int (*set)(struct inode *, void *),
874 inode = alloc_inode(sb);
878 spin_lock(&inode_lock);
879 /* We released the lock, so.. */
880 old = find_inode(sb, head, test, data);
882 if (set(inode, data))
885 hlist_add_head(&inode->i_hash, head);
886 __inode_sb_list_add(inode);
887 inode->i_state = I_NEW;
888 spin_unlock(&inode_lock);
890 /* Return the locked inode with I_NEW set, the
891 * caller is responsible for filling in the contents
897 * Uhhuh, somebody else created the same inode under
898 * us. Use the old inode instead of the one we just
901 spin_unlock(&inode_lock);
902 destroy_inode(inode);
904 wait_on_inode(inode);
909 spin_unlock(&inode_lock);
910 destroy_inode(inode);
915 * get_new_inode_fast is the fast path version of get_new_inode, see the
916 * comment at iget_locked for details.
918 static struct inode *get_new_inode_fast(struct super_block *sb,
919 struct hlist_head *head, unsigned long ino)
923 inode = alloc_inode(sb);
927 spin_lock(&inode_lock);
928 /* We released the lock, so.. */
929 old = find_inode_fast(sb, head, ino);
932 hlist_add_head(&inode->i_hash, head);
933 __inode_sb_list_add(inode);
934 inode->i_state = I_NEW;
935 spin_unlock(&inode_lock);
937 /* Return the locked inode with I_NEW set, the
938 * caller is responsible for filling in the contents
944 * Uhhuh, somebody else created the same inode under
945 * us. Use the old inode instead of the one we just
948 spin_unlock(&inode_lock);
949 destroy_inode(inode);
951 wait_on_inode(inode);
957 * search the inode cache for a matching inode number.
958 * If we find one, then the inode number we are trying to
959 * allocate is not unique and so we should not use it.
961 * Returns 1 if the inode number is unique, 0 if it is not.
963 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
965 struct hlist_head *b = inode_hashtable + hash(sb, ino);
966 struct hlist_node *node;
969 hlist_for_each_entry(inode, node, b, i_hash) {
970 if (inode->i_ino == ino && inode->i_sb == sb)
978 * iunique - get a unique inode number
980 * @max_reserved: highest reserved inode number
982 * Obtain an inode number that is unique on the system for a given
983 * superblock. This is used by file systems that have no natural
984 * permanent inode numbering system. An inode number is returned that
985 * is higher than the reserved limit but unique.
988 * With a large number of inodes live on the file system this function
989 * currently becomes quite slow.
991 ino_t iunique(struct super_block *sb, ino_t max_reserved)
994 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
995 * error if st_ino won't fit in target struct field. Use 32bit counter
996 * here to attempt to avoid that.
998 static DEFINE_SPINLOCK(iunique_lock);
999 static unsigned int counter;
1002 spin_lock(&inode_lock);
1003 spin_lock(&iunique_lock);
1005 if (counter <= max_reserved)
1006 counter = max_reserved + 1;
1008 } while (!test_inode_iunique(sb, res));
1009 spin_unlock(&iunique_lock);
1010 spin_unlock(&inode_lock);
1014 EXPORT_SYMBOL(iunique);
1016 struct inode *igrab(struct inode *inode)
1018 spin_lock(&inode_lock);
1019 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
1023 * Handle the case where s_op->clear_inode is not been
1024 * called yet, and somebody is calling igrab
1025 * while the inode is getting freed.
1028 spin_unlock(&inode_lock);
1031 EXPORT_SYMBOL(igrab);
1034 * ifind - internal function, you want ilookup5() or iget5().
1035 * @sb: super block of file system to search
1036 * @head: the head of the list to search
1037 * @test: callback used for comparisons between inodes
1038 * @data: opaque data pointer to pass to @test
1039 * @wait: if true wait for the inode to be unlocked, if false do not
1041 * ifind() searches for the inode specified by @data in the inode
1042 * cache. This is a generalized version of ifind_fast() for file systems where
1043 * the inode number is not sufficient for unique identification of an inode.
1045 * If the inode is in the cache, the inode is returned with an incremented
1048 * Otherwise NULL is returned.
1050 * Note, @test is called with the inode_lock held, so can't sleep.
1052 static struct inode *ifind(struct super_block *sb,
1053 struct hlist_head *head, int (*test)(struct inode *, void *),
1054 void *data, const int wait)
1056 struct inode *inode;
1058 spin_lock(&inode_lock);
1059 inode = find_inode(sb, head, test, data);
1061 spin_unlock(&inode_lock);
1063 wait_on_inode(inode);
1066 spin_unlock(&inode_lock);
1071 * ifind_fast - internal function, you want ilookup() or iget().
1072 * @sb: super block of file system to search
1073 * @head: head of the list to search
1074 * @ino: inode number to search for
1076 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1077 * file systems where the inode number is sufficient for unique identification
1080 * If the inode is in the cache, the inode is returned with an incremented
1083 * Otherwise NULL is returned.
1085 static struct inode *ifind_fast(struct super_block *sb,
1086 struct hlist_head *head, unsigned long ino)
1088 struct inode *inode;
1090 spin_lock(&inode_lock);
1091 inode = find_inode_fast(sb, head, ino);
1093 spin_unlock(&inode_lock);
1094 wait_on_inode(inode);
1097 spin_unlock(&inode_lock);
1102 * ilookup5_nowait - search for an inode in the inode cache
1103 * @sb: super block of file system to search
1104 * @hashval: hash value (usually inode number) to search for
1105 * @test: callback used for comparisons between inodes
1106 * @data: opaque data pointer to pass to @test
1108 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1109 * @data in the inode cache. This is a generalized version of ilookup() for
1110 * file systems where the inode number is not sufficient for unique
1111 * identification of an inode.
1113 * If the inode is in the cache, the inode is returned with an incremented
1114 * reference count. Note, the inode lock is not waited upon so you have to be
1115 * very careful what you do with the returned inode. You probably should be
1116 * using ilookup5() instead.
1118 * Otherwise NULL is returned.
1120 * Note, @test is called with the inode_lock held, so can't sleep.
1122 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1123 int (*test)(struct inode *, void *), void *data)
1125 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1127 return ifind(sb, head, test, data, 0);
1129 EXPORT_SYMBOL(ilookup5_nowait);
1132 * ilookup5 - search for an inode in the inode cache
1133 * @sb: super block of file system to search
1134 * @hashval: hash value (usually inode number) to search for
1135 * @test: callback used for comparisons between inodes
1136 * @data: opaque data pointer to pass to @test
1138 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1139 * @data in the inode cache. This is a generalized version of ilookup() for
1140 * file systems where the inode number is not sufficient for unique
1141 * identification of an inode.
1143 * If the inode is in the cache, the inode lock is waited upon and the inode is
1144 * returned with an incremented reference count.
1146 * Otherwise NULL is returned.
1148 * Note, @test is called with the inode_lock held, so can't sleep.
1150 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1151 int (*test)(struct inode *, void *), void *data)
1153 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1155 return ifind(sb, head, test, data, 1);
1157 EXPORT_SYMBOL(ilookup5);
1160 * ilookup - search for an inode in the inode cache
1161 * @sb: super block of file system to search
1162 * @ino: inode number to search for
1164 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1165 * This is for file systems where the inode number is sufficient for unique
1166 * identification of an inode.
1168 * If the inode is in the cache, the inode is returned with an incremented
1171 * Otherwise NULL is returned.
1173 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1175 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1177 return ifind_fast(sb, head, ino);
1179 EXPORT_SYMBOL(ilookup);
1182 * iget5_locked - obtain an inode from a mounted file system
1183 * @sb: super block of file system
1184 * @hashval: hash value (usually inode number) to get
1185 * @test: callback used for comparisons between inodes
1186 * @set: callback used to initialize a new struct inode
1187 * @data: opaque data pointer to pass to @test and @set
1189 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1190 * and @data in the inode cache and if present it is returned with an increased
1191 * reference count. This is a generalized version of iget_locked() for file
1192 * systems where the inode number is not sufficient for unique identification
1195 * If the inode is not in cache, get_new_inode() is called to allocate a new
1196 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1197 * file system gets to fill it in before unlocking it via unlock_new_inode().
1199 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1201 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1202 int (*test)(struct inode *, void *),
1203 int (*set)(struct inode *, void *), void *data)
1205 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1206 struct inode *inode;
1208 inode = ifind(sb, head, test, data, 1);
1212 * get_new_inode() will do the right thing, re-trying the search
1213 * in case it had to block at any point.
1215 return get_new_inode(sb, head, test, set, data);
1217 EXPORT_SYMBOL(iget5_locked);
1220 * iget_locked - obtain an inode from a mounted file system
1221 * @sb: super block of file system
1222 * @ino: inode number to get
1224 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1225 * the inode cache and if present it is returned with an increased reference
1226 * count. This is for file systems where the inode number is sufficient for
1227 * unique identification of an inode.
1229 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1230 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1231 * The file system gets to fill it in before unlocking it via
1232 * unlock_new_inode().
1234 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1236 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1237 struct inode *inode;
1239 inode = ifind_fast(sb, head, ino);
1243 * get_new_inode_fast() will do the right thing, re-trying the search
1244 * in case it had to block at any point.
1246 return get_new_inode_fast(sb, head, ino);
1248 EXPORT_SYMBOL(iget_locked);
1250 int insert_inode_locked(struct inode *inode)
1252 struct super_block *sb = inode->i_sb;
1253 ino_t ino = inode->i_ino;
1254 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1256 inode->i_state |= I_NEW;
1258 struct hlist_node *node;
1259 struct inode *old = NULL;
1260 spin_lock(&inode_lock);
1261 hlist_for_each_entry(old, node, head, i_hash) {
1262 if (old->i_ino != ino)
1264 if (old->i_sb != sb)
1266 if (old->i_state & (I_FREEING|I_WILL_FREE))
1270 if (likely(!node)) {
1271 hlist_add_head(&inode->i_hash, head);
1272 spin_unlock(&inode_lock);
1276 spin_unlock(&inode_lock);
1278 if (unlikely(!inode_unhashed(old))) {
1285 EXPORT_SYMBOL(insert_inode_locked);
1287 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1288 int (*test)(struct inode *, void *), void *data)
1290 struct super_block *sb = inode->i_sb;
1291 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1293 inode->i_state |= I_NEW;
1296 struct hlist_node *node;
1297 struct inode *old = NULL;
1299 spin_lock(&inode_lock);
1300 hlist_for_each_entry(old, node, head, i_hash) {
1301 if (old->i_sb != sb)
1303 if (!test(old, data))
1305 if (old->i_state & (I_FREEING|I_WILL_FREE))
1309 if (likely(!node)) {
1310 hlist_add_head(&inode->i_hash, head);
1311 spin_unlock(&inode_lock);
1315 spin_unlock(&inode_lock);
1317 if (unlikely(!inode_unhashed(old))) {
1324 EXPORT_SYMBOL(insert_inode_locked4);
1327 int generic_delete_inode(struct inode *inode)
1331 EXPORT_SYMBOL(generic_delete_inode);
1334 * Normal UNIX filesystem behaviour: delete the
1335 * inode when the usage count drops to zero, and
1338 int generic_drop_inode(struct inode *inode)
1340 return !inode->i_nlink || inode_unhashed(inode);
1342 EXPORT_SYMBOL_GPL(generic_drop_inode);
1345 * Called when we're dropping the last reference
1348 * Call the FS "drop_inode()" function, defaulting to
1349 * the legacy UNIX filesystem behaviour. If it tells
1350 * us to evict inode, do so. Otherwise, retain inode
1351 * in cache if fs is alive, sync and evict if fs is
1354 static void iput_final(struct inode *inode)
1356 struct super_block *sb = inode->i_sb;
1357 const struct super_operations *op = inode->i_sb->s_op;
1360 if (op && op->drop_inode)
1361 drop = op->drop_inode(inode);
1363 drop = generic_drop_inode(inode);
1366 if (sb->s_flags & MS_ACTIVE) {
1367 inode->i_state |= I_REFERENCED;
1368 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1369 inode_lru_list_add(inode);
1371 spin_unlock(&inode_lock);
1374 WARN_ON(inode->i_state & I_NEW);
1375 inode->i_state |= I_WILL_FREE;
1376 spin_unlock(&inode_lock);
1377 write_inode_now(inode, 1);
1378 spin_lock(&inode_lock);
1379 WARN_ON(inode->i_state & I_NEW);
1380 inode->i_state &= ~I_WILL_FREE;
1381 __remove_inode_hash(inode);
1384 WARN_ON(inode->i_state & I_NEW);
1385 inode->i_state |= I_FREEING;
1388 * Move the inode off the IO lists and LRU once I_FREEING is
1389 * set so that it won't get moved back on there if it is dirty.
1391 inode_lru_list_del(inode);
1392 list_del_init(&inode->i_wb_list);
1394 __inode_sb_list_del(inode);
1395 spin_unlock(&inode_lock);
1397 remove_inode_hash(inode);
1398 wake_up_inode(inode);
1399 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1400 destroy_inode(inode);
1404 * iput - put an inode
1405 * @inode: inode to put
1407 * Puts an inode, dropping its usage count. If the inode use count hits
1408 * zero, the inode is then freed and may also be destroyed.
1410 * Consequently, iput() can sleep.
1412 void iput(struct inode *inode)
1415 BUG_ON(inode->i_state & I_CLEAR);
1417 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1421 EXPORT_SYMBOL(iput);
1424 * bmap - find a block number in a file
1425 * @inode: inode of file
1426 * @block: block to find
1428 * Returns the block number on the device holding the inode that
1429 * is the disk block number for the block of the file requested.
1430 * That is, asked for block 4 of inode 1 the function will return the
1431 * disk block relative to the disk start that holds that block of the
1434 sector_t bmap(struct inode *inode, sector_t block)
1437 if (inode->i_mapping->a_ops->bmap)
1438 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1441 EXPORT_SYMBOL(bmap);
1444 * With relative atime, only update atime if the previous atime is
1445 * earlier than either the ctime or mtime or if at least a day has
1446 * passed since the last atime update.
1448 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1449 struct timespec now)
1452 if (!(mnt->mnt_flags & MNT_RELATIME))
1455 * Is mtime younger than atime? If yes, update atime:
1457 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1460 * Is ctime younger than atime? If yes, update atime:
1462 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1466 * Is the previous atime value older than a day? If yes,
1469 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1472 * Good, we can skip the atime update:
1478 * touch_atime - update the access time
1479 * @mnt: mount the inode is accessed on
1480 * @dentry: dentry accessed
1482 * Update the accessed time on an inode and mark it for writeback.
1483 * This function automatically handles read only file systems and media,
1484 * as well as the "noatime" flag and inode specific "noatime" markers.
1486 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1488 struct inode *inode = dentry->d_inode;
1489 struct timespec now;
1491 if (inode->i_flags & S_NOATIME)
1493 if (IS_NOATIME(inode))
1495 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1498 if (mnt->mnt_flags & MNT_NOATIME)
1500 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1503 now = current_fs_time(inode->i_sb);
1505 if (!relatime_need_update(mnt, inode, now))
1508 if (timespec_equal(&inode->i_atime, &now))
1511 if (mnt_want_write(mnt))
1514 inode->i_atime = now;
1515 mark_inode_dirty_sync(inode);
1516 mnt_drop_write(mnt);
1518 EXPORT_SYMBOL(touch_atime);
1521 * file_update_time - update mtime and ctime time
1522 * @file: file accessed
1524 * Update the mtime and ctime members of an inode and mark the inode
1525 * for writeback. Note that this function is meant exclusively for
1526 * usage in the file write path of filesystems, and filesystems may
1527 * choose to explicitly ignore update via this function with the
1528 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1529 * timestamps are handled by the server.
1532 void file_update_time(struct file *file)
1534 struct inode *inode = file->f_path.dentry->d_inode;
1535 struct timespec now;
1536 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1538 /* First try to exhaust all avenues to not sync */
1539 if (IS_NOCMTIME(inode))
1542 now = current_fs_time(inode->i_sb);
1543 if (!timespec_equal(&inode->i_mtime, &now))
1546 if (!timespec_equal(&inode->i_ctime, &now))
1549 if (IS_I_VERSION(inode))
1550 sync_it |= S_VERSION;
1555 /* Finally allowed to write? Takes lock. */
1556 if (mnt_want_write_file(file))
1559 /* Only change inode inside the lock region */
1560 if (sync_it & S_VERSION)
1561 inode_inc_iversion(inode);
1562 if (sync_it & S_CTIME)
1563 inode->i_ctime = now;
1564 if (sync_it & S_MTIME)
1565 inode->i_mtime = now;
1566 mark_inode_dirty_sync(inode);
1567 mnt_drop_write(file->f_path.mnt);
1569 EXPORT_SYMBOL(file_update_time);
1571 int inode_needs_sync(struct inode *inode)
1575 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1579 EXPORT_SYMBOL(inode_needs_sync);
1581 int inode_wait(void *word)
1586 EXPORT_SYMBOL(inode_wait);
1589 * If we try to find an inode in the inode hash while it is being
1590 * deleted, we have to wait until the filesystem completes its
1591 * deletion before reporting that it isn't found. This function waits
1592 * until the deletion _might_ have completed. Callers are responsible
1593 * to recheck inode state.
1595 * It doesn't matter if I_NEW is not set initially, a call to
1596 * wake_up_inode() after removing from the hash list will DTRT.
1598 * This is called with inode_lock held.
1600 static void __wait_on_freeing_inode(struct inode *inode)
1602 wait_queue_head_t *wq;
1603 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1604 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1605 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1606 spin_unlock(&inode_lock);
1608 finish_wait(wq, &wait.wait);
1609 spin_lock(&inode_lock);
1612 static __initdata unsigned long ihash_entries;
1613 static int __init set_ihash_entries(char *str)
1617 ihash_entries = simple_strtoul(str, &str, 0);
1620 __setup("ihash_entries=", set_ihash_entries);
1623 * Initialize the waitqueues and inode hash table.
1625 void __init inode_init_early(void)
1629 /* If hashes are distributed across NUMA nodes, defer
1630 * hash allocation until vmalloc space is available.
1636 alloc_large_system_hash("Inode-cache",
1637 sizeof(struct hlist_head),
1645 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1646 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1649 void __init inode_init(void)
1653 /* inode slab cache */
1654 inode_cachep = kmem_cache_create("inode_cache",
1655 sizeof(struct inode),
1657 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1660 register_shrinker(&icache_shrinker);
1662 /* Hash may have been set up in inode_init_early */
1667 alloc_large_system_hash("Inode-cache",
1668 sizeof(struct hlist_head),
1676 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1677 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1680 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1682 inode->i_mode = mode;
1683 if (S_ISCHR(mode)) {
1684 inode->i_fop = &def_chr_fops;
1685 inode->i_rdev = rdev;
1686 } else if (S_ISBLK(mode)) {
1687 inode->i_fop = &def_blk_fops;
1688 inode->i_rdev = rdev;
1689 } else if (S_ISFIFO(mode))
1690 inode->i_fop = &def_fifo_fops;
1691 else if (S_ISSOCK(mode))
1692 inode->i_fop = &bad_sock_fops;
1694 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1695 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1698 EXPORT_SYMBOL(init_special_inode);
1701 * Init uid,gid,mode for new inode according to posix standards
1703 * @dir: Directory inode
1704 * @mode: mode of the new inode
1706 void inode_init_owner(struct inode *inode, const struct inode *dir,
1709 inode->i_uid = current_fsuid();
1710 if (dir && dir->i_mode & S_ISGID) {
1711 inode->i_gid = dir->i_gid;
1715 inode->i_gid = current_fsgid();
1716 inode->i_mode = mode;
1718 EXPORT_SYMBOL(inode_init_owner);