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>
29 * This is needed for the following functions:
33 * FIXME: remove all knowledge of the buffer layer from this file
35 #include <linux/buffer_head.h>
38 * New inode.c implementation.
40 * This implementation has the basic premise of trying
41 * to be extremely low-overhead and SMP-safe, yet be
42 * simple enough to be "obviously correct".
47 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
49 /* #define INODE_PARANOIA 1 */
50 /* #define INODE_DEBUG 1 */
53 * Inode lookup is no longer as critical as it used to be:
54 * most of the lookups are going to be through the dcache.
56 #define I_HASHBITS i_hash_shift
57 #define I_HASHMASK i_hash_mask
59 static unsigned int i_hash_mask __read_mostly;
60 static unsigned int i_hash_shift __read_mostly;
63 * Each inode can be on two separate lists. One is
64 * the hash list of the inode, used for lookups. The
65 * other linked list is the "type" list:
66 * "in_use" - valid inode, i_count > 0, i_nlink > 0
67 * "dirty" - as "in_use" but also dirty
68 * "unused" - valid inode, i_count = 0
70 * A "dirty" list is maintained for each super block,
71 * allowing for low-overhead inode sync() operations.
74 static LIST_HEAD(inode_lru);
75 static struct hlist_head *inode_hashtable __read_mostly;
78 * A simple spinlock to protect the list manipulations.
80 * NOTE! You also have to own the lock if you change
81 * the i_state of an inode while it is in use..
83 DEFINE_SPINLOCK(inode_lock);
86 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
87 * icache shrinking path, and the umount path. Without this exclusion,
88 * by the time prune_icache calls iput for the inode whose pages it has
89 * been invalidating, or by the time it calls clear_inode & destroy_inode
90 * from its final dispose_list, the struct super_block they refer to
91 * (for inode->i_sb->s_op) may already have been freed and reused.
93 * We make this an rwsem because the fastpath is icache shrinking. In
94 * some cases a filesystem may be doing a significant amount of work in
95 * its inode reclaim code, so this should improve parallelism.
97 static DECLARE_RWSEM(iprune_sem);
100 * Statistics gathering..
102 struct inodes_stat_t inodes_stat;
104 static struct percpu_counter nr_inodes __cacheline_aligned_in_smp;
105 static struct percpu_counter nr_inodes_unused __cacheline_aligned_in_smp;
107 static struct kmem_cache *inode_cachep __read_mostly;
109 static inline int get_nr_inodes(void)
111 return percpu_counter_sum_positive(&nr_inodes);
114 static inline int get_nr_inodes_unused(void)
116 return percpu_counter_sum_positive(&nr_inodes_unused);
119 int get_nr_dirty_inodes(void)
121 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
122 return nr_dirty > 0 ? nr_dirty : 0;
127 * Handle nr_inode sysctl
130 int proc_nr_inodes(ctl_table *table, int write,
131 void __user *buffer, size_t *lenp, loff_t *ppos)
133 inodes_stat.nr_inodes = get_nr_inodes();
134 inodes_stat.nr_unused = get_nr_inodes_unused();
135 return proc_dointvec(table, write, buffer, lenp, ppos);
139 static void wake_up_inode(struct inode *inode)
142 * Prevent speculative execution through spin_unlock(&inode_lock);
145 wake_up_bit(&inode->i_state, __I_NEW);
149 * inode_init_always - perform inode structure intialisation
150 * @sb: superblock inode belongs to
151 * @inode: inode to initialise
153 * These are initializations that need to be done on every inode
154 * allocation as the fields are not initialised by slab allocation.
156 int inode_init_always(struct super_block *sb, struct inode *inode)
158 static const struct address_space_operations empty_aops;
159 static const struct inode_operations empty_iops;
160 static const struct file_operations empty_fops;
161 struct address_space *const mapping = &inode->i_data;
164 inode->i_blkbits = sb->s_blocksize_bits;
166 atomic_set(&inode->i_count, 1);
167 inode->i_op = &empty_iops;
168 inode->i_fop = &empty_fops;
172 atomic_set(&inode->i_writecount, 0);
176 inode->i_generation = 0;
178 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
180 inode->i_pipe = NULL;
181 inode->i_bdev = NULL;
182 inode->i_cdev = NULL;
184 inode->dirtied_when = 0;
186 if (security_inode_alloc(inode))
188 spin_lock_init(&inode->i_lock);
189 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
191 mutex_init(&inode->i_mutex);
192 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
194 init_rwsem(&inode->i_alloc_sem);
195 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
197 mapping->a_ops = &empty_aops;
198 mapping->host = inode;
200 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
201 mapping->assoc_mapping = NULL;
202 mapping->backing_dev_info = &default_backing_dev_info;
203 mapping->writeback_index = 0;
206 * If the block_device provides a backing_dev_info for client
207 * inodes then use that. Otherwise the inode share the bdev's
211 struct backing_dev_info *bdi;
213 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
214 mapping->backing_dev_info = bdi;
216 inode->i_private = NULL;
217 inode->i_mapping = mapping;
218 #ifdef CONFIG_FS_POSIX_ACL
219 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
222 #ifdef CONFIG_FSNOTIFY
223 inode->i_fsnotify_mask = 0;
226 percpu_counter_inc(&nr_inodes);
232 EXPORT_SYMBOL(inode_init_always);
234 static struct inode *alloc_inode(struct super_block *sb)
238 if (sb->s_op->alloc_inode)
239 inode = sb->s_op->alloc_inode(sb);
241 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
246 if (unlikely(inode_init_always(sb, inode))) {
247 if (inode->i_sb->s_op->destroy_inode)
248 inode->i_sb->s_op->destroy_inode(inode);
250 kmem_cache_free(inode_cachep, inode);
257 void __destroy_inode(struct inode *inode)
259 BUG_ON(inode_has_buffers(inode));
260 security_inode_free(inode);
261 fsnotify_inode_delete(inode);
262 #ifdef CONFIG_FS_POSIX_ACL
263 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
264 posix_acl_release(inode->i_acl);
265 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
266 posix_acl_release(inode->i_default_acl);
268 percpu_counter_dec(&nr_inodes);
270 EXPORT_SYMBOL(__destroy_inode);
272 static void destroy_inode(struct inode *inode)
274 BUG_ON(!list_empty(&inode->i_lru));
275 __destroy_inode(inode);
276 if (inode->i_sb->s_op->destroy_inode)
277 inode->i_sb->s_op->destroy_inode(inode);
279 kmem_cache_free(inode_cachep, (inode));
283 * These are initializations that only need to be done
284 * once, because the fields are idempotent across use
285 * of the inode, so let the slab aware of that.
287 void inode_init_once(struct inode *inode)
289 memset(inode, 0, sizeof(*inode));
290 INIT_HLIST_NODE(&inode->i_hash);
291 INIT_LIST_HEAD(&inode->i_dentry);
292 INIT_LIST_HEAD(&inode->i_devices);
293 INIT_LIST_HEAD(&inode->i_wb_list);
294 INIT_LIST_HEAD(&inode->i_lru);
295 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
296 spin_lock_init(&inode->i_data.tree_lock);
297 spin_lock_init(&inode->i_data.i_mmap_lock);
298 INIT_LIST_HEAD(&inode->i_data.private_list);
299 spin_lock_init(&inode->i_data.private_lock);
300 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
301 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
302 i_size_ordered_init(inode);
303 #ifdef CONFIG_FSNOTIFY
304 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
307 EXPORT_SYMBOL(inode_init_once);
309 static void init_once(void *foo)
311 struct inode *inode = (struct inode *) foo;
313 inode_init_once(inode);
317 * inode_lock must be held
319 void __iget(struct inode *inode)
321 atomic_inc(&inode->i_count);
325 * get additional reference to inode; caller must already hold one.
327 void ihold(struct inode *inode)
329 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
331 EXPORT_SYMBOL(ihold);
333 static void inode_lru_list_add(struct inode *inode)
335 if (list_empty(&inode->i_lru)) {
336 list_add(&inode->i_lru, &inode_lru);
337 percpu_counter_inc(&nr_inodes_unused);
341 static void inode_lru_list_del(struct inode *inode)
343 if (!list_empty(&inode->i_lru)) {
344 list_del_init(&inode->i_lru);
345 percpu_counter_dec(&nr_inodes_unused);
349 static inline void __inode_sb_list_add(struct inode *inode)
351 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
355 * inode_sb_list_add - add inode to the superblock list of inodes
356 * @inode: inode to add
358 void inode_sb_list_add(struct inode *inode)
360 spin_lock(&inode_lock);
361 __inode_sb_list_add(inode);
362 spin_unlock(&inode_lock);
364 EXPORT_SYMBOL_GPL(inode_sb_list_add);
366 static inline void __inode_sb_list_del(struct inode *inode)
368 list_del_init(&inode->i_sb_list);
371 static unsigned long hash(struct super_block *sb, unsigned long hashval)
375 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
377 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
378 return tmp & I_HASHMASK;
382 * __insert_inode_hash - hash an inode
383 * @inode: unhashed inode
384 * @hashval: unsigned long value used to locate this object in the
387 * Add an inode to the inode hash for this superblock.
389 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
391 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
393 spin_lock(&inode_lock);
394 hlist_add_head(&inode->i_hash, b);
395 spin_unlock(&inode_lock);
397 EXPORT_SYMBOL(__insert_inode_hash);
400 * __remove_inode_hash - remove an inode from the hash
401 * @inode: inode to unhash
403 * Remove an inode from the superblock.
405 static void __remove_inode_hash(struct inode *inode)
407 hlist_del_init(&inode->i_hash);
411 * remove_inode_hash - remove an inode from the hash
412 * @inode: inode to unhash
414 * Remove an inode from the superblock.
416 void remove_inode_hash(struct inode *inode)
418 spin_lock(&inode_lock);
419 hlist_del_init(&inode->i_hash);
420 spin_unlock(&inode_lock);
422 EXPORT_SYMBOL(remove_inode_hash);
424 void end_writeback(struct inode *inode)
427 BUG_ON(inode->i_data.nrpages);
428 BUG_ON(!list_empty(&inode->i_data.private_list));
429 BUG_ON(!(inode->i_state & I_FREEING));
430 BUG_ON(inode->i_state & I_CLEAR);
431 inode_sync_wait(inode);
432 inode->i_state = I_FREEING | I_CLEAR;
434 EXPORT_SYMBOL(end_writeback);
436 static void evict(struct inode *inode)
438 const struct super_operations *op = inode->i_sb->s_op;
440 if (op->evict_inode) {
441 op->evict_inode(inode);
443 if (inode->i_data.nrpages)
444 truncate_inode_pages(&inode->i_data, 0);
445 end_writeback(inode);
447 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
449 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
454 * dispose_list - dispose of the contents of a local list
455 * @head: the head of the list to free
457 * Dispose-list gets a local list with local inodes in it, so it doesn't
458 * need to worry about list corruption and SMP locks.
460 static void dispose_list(struct list_head *head)
462 while (!list_empty(head)) {
465 inode = list_first_entry(head, struct inode, i_lru);
466 list_del_init(&inode->i_lru);
470 spin_lock(&inode_lock);
471 __remove_inode_hash(inode);
472 __inode_sb_list_del(inode);
473 spin_unlock(&inode_lock);
475 wake_up_inode(inode);
476 destroy_inode(inode);
481 * evict_inodes - evict all evictable inodes for a superblock
482 * @sb: superblock to operate on
484 * Make sure that no inodes with zero refcount are retained. This is
485 * called by superblock shutdown after having MS_ACTIVE flag removed,
486 * so any inode reaching zero refcount during or after that call will
487 * be immediately evicted.
489 void evict_inodes(struct super_block *sb)
491 struct inode *inode, *next;
494 down_write(&iprune_sem);
496 spin_lock(&inode_lock);
497 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
498 if (atomic_read(&inode->i_count))
501 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
506 inode->i_state |= I_FREEING;
509 * Move the inode off the IO lists and LRU once I_FREEING is
510 * set so that it won't get moved back on there if it is dirty.
512 list_move(&inode->i_lru, &dispose);
513 list_del_init(&inode->i_wb_list);
514 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
515 percpu_counter_dec(&nr_inodes_unused);
517 spin_unlock(&inode_lock);
519 dispose_list(&dispose);
520 up_write(&iprune_sem);
524 * invalidate_inodes - attempt to free all inodes on a superblock
525 * @sb: superblock to operate on
527 * Attempts to free all inodes for a given superblock. If there were any
528 * busy inodes return a non-zero value, else zero.
530 int invalidate_inodes(struct super_block *sb)
533 struct inode *inode, *next;
536 down_write(&iprune_sem);
538 spin_lock(&inode_lock);
539 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
540 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE))
542 if (atomic_read(&inode->i_count)) {
547 inode->i_state |= I_FREEING;
550 * Move the inode off the IO lists and LRU once I_FREEING is
551 * set so that it won't get moved back on there if it is dirty.
553 list_move(&inode->i_lru, &dispose);
554 list_del_init(&inode->i_wb_list);
555 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
556 percpu_counter_dec(&nr_inodes_unused);
558 spin_unlock(&inode_lock);
560 dispose_list(&dispose);
561 up_write(&iprune_sem);
566 static int can_unuse(struct inode *inode)
568 if (inode->i_state & ~I_REFERENCED)
570 if (inode_has_buffers(inode))
572 if (atomic_read(&inode->i_count))
574 if (inode->i_data.nrpages)
580 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
581 * temporary list and then are freed outside inode_lock by dispose_list().
583 * Any inodes which are pinned purely because of attached pagecache have their
584 * pagecache removed. If the inode has metadata buffers attached to
585 * mapping->private_list then try to remove them.
587 * If the inode has the I_REFERENCED flag set, then it means that it has been
588 * used recently - the flag is set in iput_final(). When we encounter such an
589 * inode, clear the flag and move it to the back of the LRU so it gets another
590 * pass through the LRU before it gets reclaimed. This is necessary because of
591 * the fact we are doing lazy LRU updates to minimise lock contention so the
592 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
593 * with this flag set because they are the inodes that are out of order.
595 static void prune_icache(int nr_to_scan)
599 unsigned long reap = 0;
601 down_read(&iprune_sem);
602 spin_lock(&inode_lock);
603 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
606 if (list_empty(&inode_lru))
609 inode = list_entry(inode_lru.prev, struct inode, i_lru);
612 * Referenced or dirty inodes are still in use. Give them
613 * another pass through the LRU as we canot reclaim them now.
615 if (atomic_read(&inode->i_count) ||
616 (inode->i_state & ~I_REFERENCED)) {
617 list_del_init(&inode->i_lru);
618 percpu_counter_dec(&nr_inodes_unused);
622 /* recently referenced inodes get one more pass */
623 if (inode->i_state & I_REFERENCED) {
624 list_move(&inode->i_lru, &inode_lru);
625 inode->i_state &= ~I_REFERENCED;
628 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
630 spin_unlock(&inode_lock);
631 if (remove_inode_buffers(inode))
632 reap += invalidate_mapping_pages(&inode->i_data,
635 spin_lock(&inode_lock);
637 if (inode != list_entry(inode_lru.next,
638 struct inode, i_lru))
639 continue; /* wrong inode or list_empty */
640 if (!can_unuse(inode))
643 WARN_ON(inode->i_state & I_NEW);
644 inode->i_state |= I_FREEING;
647 * Move the inode off the IO lists and LRU once I_FREEING is
648 * set so that it won't get moved back on there if it is dirty.
650 list_move(&inode->i_lru, &freeable);
651 list_del_init(&inode->i_wb_list);
652 percpu_counter_dec(&nr_inodes_unused);
654 if (current_is_kswapd())
655 __count_vm_events(KSWAPD_INODESTEAL, reap);
657 __count_vm_events(PGINODESTEAL, reap);
658 spin_unlock(&inode_lock);
660 dispose_list(&freeable);
661 up_read(&iprune_sem);
665 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
666 * "unused" means that no dentries are referring to the inodes: the files are
667 * not open and the dcache references to those inodes have already been
670 * This function is passed the number of inodes to scan, and it returns the
671 * total number of remaining possibly-reclaimable inodes.
673 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
677 * Nasty deadlock avoidance. We may hold various FS locks,
678 * and we don't want to recurse into the FS that called us
679 * in clear_inode() and friends..
681 if (!(gfp_mask & __GFP_FS))
685 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
688 static struct shrinker icache_shrinker = {
689 .shrink = shrink_icache_memory,
690 .seeks = DEFAULT_SEEKS,
693 static void __wait_on_freeing_inode(struct inode *inode);
695 * Called with the inode lock held.
697 static struct inode *find_inode(struct super_block *sb,
698 struct hlist_head *head,
699 int (*test)(struct inode *, void *),
702 struct hlist_node *node;
703 struct inode *inode = NULL;
706 hlist_for_each_entry(inode, node, head, i_hash) {
707 if (inode->i_sb != sb)
709 if (!test(inode, data))
711 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
712 __wait_on_freeing_inode(inode);
722 * find_inode_fast is the fast path version of find_inode, see the comment at
723 * iget_locked for details.
725 static struct inode *find_inode_fast(struct super_block *sb,
726 struct hlist_head *head, unsigned long ino)
728 struct hlist_node *node;
729 struct inode *inode = NULL;
732 hlist_for_each_entry(inode, node, head, i_hash) {
733 if (inode->i_ino != ino)
735 if (inode->i_sb != sb)
737 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
738 __wait_on_freeing_inode(inode);
748 * Each cpu owns a range of LAST_INO_BATCH numbers.
749 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
750 * to renew the exhausted range.
752 * This does not significantly increase overflow rate because every CPU can
753 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
754 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
755 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
756 * overflow rate by 2x, which does not seem too significant.
758 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
759 * error if st_ino won't fit in target struct field. Use 32bit counter
760 * here to attempt to avoid that.
762 #define LAST_INO_BATCH 1024
763 static DEFINE_PER_CPU(unsigned int, last_ino);
765 unsigned int get_next_ino(void)
767 unsigned int *p = &get_cpu_var(last_ino);
768 unsigned int res = *p;
771 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
772 static atomic_t shared_last_ino;
773 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
775 res = next - LAST_INO_BATCH;
780 put_cpu_var(last_ino);
783 EXPORT_SYMBOL(get_next_ino);
786 * new_inode - obtain an inode
789 * Allocates a new inode for given superblock. The default gfp_mask
790 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
791 * If HIGHMEM pages are unsuitable or it is known that pages allocated
792 * for the page cache are not reclaimable or migratable,
793 * mapping_set_gfp_mask() must be called with suitable flags on the
794 * newly created inode's mapping
797 struct inode *new_inode(struct super_block *sb)
801 spin_lock_prefetch(&inode_lock);
803 inode = alloc_inode(sb);
805 spin_lock(&inode_lock);
806 __inode_sb_list_add(inode);
808 spin_unlock(&inode_lock);
812 EXPORT_SYMBOL(new_inode);
814 void unlock_new_inode(struct inode *inode)
816 #ifdef CONFIG_DEBUG_LOCK_ALLOC
817 if (S_ISDIR(inode->i_mode)) {
818 struct file_system_type *type = inode->i_sb->s_type;
820 /* Set new key only if filesystem hasn't already changed it */
821 if (!lockdep_match_class(&inode->i_mutex,
822 &type->i_mutex_key)) {
824 * ensure nobody is actually holding i_mutex
826 mutex_destroy(&inode->i_mutex);
827 mutex_init(&inode->i_mutex);
828 lockdep_set_class(&inode->i_mutex,
829 &type->i_mutex_dir_key);
834 * This is special! We do not need the spinlock when clearing I_NEW,
835 * because we're guaranteed that nobody else tries to do anything about
836 * the state of the inode when it is locked, as we just created it (so
837 * there can be no old holders that haven't tested I_NEW).
838 * However we must emit the memory barrier so that other CPUs reliably
839 * see the clearing of I_NEW after the other inode initialisation has
843 WARN_ON(!(inode->i_state & I_NEW));
844 inode->i_state &= ~I_NEW;
845 wake_up_inode(inode);
847 EXPORT_SYMBOL(unlock_new_inode);
850 * This is called without the inode lock held.. Be careful.
852 * We no longer cache the sb_flags in i_flags - see fs.h
853 * -- rmk@arm.uk.linux.org
855 static struct inode *get_new_inode(struct super_block *sb,
856 struct hlist_head *head,
857 int (*test)(struct inode *, void *),
858 int (*set)(struct inode *, void *),
863 inode = alloc_inode(sb);
867 spin_lock(&inode_lock);
868 /* We released the lock, so.. */
869 old = find_inode(sb, head, test, data);
871 if (set(inode, data))
874 hlist_add_head(&inode->i_hash, head);
875 __inode_sb_list_add(inode);
876 inode->i_state = I_NEW;
877 spin_unlock(&inode_lock);
879 /* Return the locked inode with I_NEW set, the
880 * caller is responsible for filling in the contents
886 * Uhhuh, somebody else created the same inode under
887 * us. Use the old inode instead of the one we just
890 spin_unlock(&inode_lock);
891 destroy_inode(inode);
893 wait_on_inode(inode);
898 spin_unlock(&inode_lock);
899 destroy_inode(inode);
904 * get_new_inode_fast is the fast path version of get_new_inode, see the
905 * comment at iget_locked for details.
907 static struct inode *get_new_inode_fast(struct super_block *sb,
908 struct hlist_head *head, unsigned long ino)
912 inode = alloc_inode(sb);
916 spin_lock(&inode_lock);
917 /* We released the lock, so.. */
918 old = find_inode_fast(sb, head, ino);
921 hlist_add_head(&inode->i_hash, head);
922 __inode_sb_list_add(inode);
923 inode->i_state = I_NEW;
924 spin_unlock(&inode_lock);
926 /* Return the locked inode with I_NEW set, the
927 * caller is responsible for filling in the contents
933 * Uhhuh, somebody else created the same inode under
934 * us. Use the old inode instead of the one we just
937 spin_unlock(&inode_lock);
938 destroy_inode(inode);
940 wait_on_inode(inode);
946 * search the inode cache for a matching inode number.
947 * If we find one, then the inode number we are trying to
948 * allocate is not unique and so we should not use it.
950 * Returns 1 if the inode number is unique, 0 if it is not.
952 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
954 struct hlist_head *b = inode_hashtable + hash(sb, ino);
955 struct hlist_node *node;
958 hlist_for_each_entry(inode, node, b, i_hash) {
959 if (inode->i_ino == ino && inode->i_sb == sb)
967 * iunique - get a unique inode number
969 * @max_reserved: highest reserved inode number
971 * Obtain an inode number that is unique on the system for a given
972 * superblock. This is used by file systems that have no natural
973 * permanent inode numbering system. An inode number is returned that
974 * is higher than the reserved limit but unique.
977 * With a large number of inodes live on the file system this function
978 * currently becomes quite slow.
980 ino_t iunique(struct super_block *sb, ino_t max_reserved)
983 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
984 * error if st_ino won't fit in target struct field. Use 32bit counter
985 * here to attempt to avoid that.
987 static DEFINE_SPINLOCK(iunique_lock);
988 static unsigned int counter;
991 spin_lock(&inode_lock);
992 spin_lock(&iunique_lock);
994 if (counter <= max_reserved)
995 counter = max_reserved + 1;
997 } while (!test_inode_iunique(sb, res));
998 spin_unlock(&iunique_lock);
999 spin_unlock(&inode_lock);
1003 EXPORT_SYMBOL(iunique);
1005 struct inode *igrab(struct inode *inode)
1007 spin_lock(&inode_lock);
1008 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
1012 * Handle the case where s_op->clear_inode is not been
1013 * called yet, and somebody is calling igrab
1014 * while the inode is getting freed.
1017 spin_unlock(&inode_lock);
1020 EXPORT_SYMBOL(igrab);
1023 * ifind - internal function, you want ilookup5() or iget5().
1024 * @sb: super block of file system to search
1025 * @head: the head of the list to search
1026 * @test: callback used for comparisons between inodes
1027 * @data: opaque data pointer to pass to @test
1028 * @wait: if true wait for the inode to be unlocked, if false do not
1030 * ifind() searches for the inode specified by @data in the inode
1031 * cache. This is a generalized version of ifind_fast() for file systems where
1032 * the inode number is not sufficient for unique identification of an inode.
1034 * If the inode is in the cache, the inode is returned with an incremented
1037 * Otherwise NULL is returned.
1039 * Note, @test is called with the inode_lock held, so can't sleep.
1041 static struct inode *ifind(struct super_block *sb,
1042 struct hlist_head *head, int (*test)(struct inode *, void *),
1043 void *data, const int wait)
1045 struct inode *inode;
1047 spin_lock(&inode_lock);
1048 inode = find_inode(sb, head, test, data);
1050 spin_unlock(&inode_lock);
1052 wait_on_inode(inode);
1055 spin_unlock(&inode_lock);
1060 * ifind_fast - internal function, you want ilookup() or iget().
1061 * @sb: super block of file system to search
1062 * @head: head of the list to search
1063 * @ino: inode number to search for
1065 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1066 * file systems where the inode number is sufficient for unique identification
1069 * If the inode is in the cache, the inode is returned with an incremented
1072 * Otherwise NULL is returned.
1074 static struct inode *ifind_fast(struct super_block *sb,
1075 struct hlist_head *head, unsigned long ino)
1077 struct inode *inode;
1079 spin_lock(&inode_lock);
1080 inode = find_inode_fast(sb, head, ino);
1082 spin_unlock(&inode_lock);
1083 wait_on_inode(inode);
1086 spin_unlock(&inode_lock);
1091 * ilookup5_nowait - search for an inode in the inode cache
1092 * @sb: super block of file system to search
1093 * @hashval: hash value (usually inode number) to search for
1094 * @test: callback used for comparisons between inodes
1095 * @data: opaque data pointer to pass to @test
1097 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1098 * @data in the inode cache. This is a generalized version of ilookup() for
1099 * file systems where the inode number is not sufficient for unique
1100 * identification of an inode.
1102 * If the inode is in the cache, the inode is returned with an incremented
1103 * reference count. Note, the inode lock is not waited upon so you have to be
1104 * very careful what you do with the returned inode. You probably should be
1105 * using ilookup5() instead.
1107 * Otherwise NULL is returned.
1109 * Note, @test is called with the inode_lock held, so can't sleep.
1111 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1112 int (*test)(struct inode *, void *), void *data)
1114 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1116 return ifind(sb, head, test, data, 0);
1118 EXPORT_SYMBOL(ilookup5_nowait);
1121 * ilookup5 - search for an inode in the inode cache
1122 * @sb: super block of file system to search
1123 * @hashval: hash value (usually inode number) to search for
1124 * @test: callback used for comparisons between inodes
1125 * @data: opaque data pointer to pass to @test
1127 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1128 * @data in the inode cache. This is a generalized version of ilookup() for
1129 * file systems where the inode number is not sufficient for unique
1130 * identification of an inode.
1132 * If the inode is in the cache, the inode lock is waited upon and the inode is
1133 * returned with an incremented reference count.
1135 * Otherwise NULL is returned.
1137 * Note, @test is called with the inode_lock held, so can't sleep.
1139 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1140 int (*test)(struct inode *, void *), void *data)
1142 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1144 return ifind(sb, head, test, data, 1);
1146 EXPORT_SYMBOL(ilookup5);
1149 * ilookup - search for an inode in the inode cache
1150 * @sb: super block of file system to search
1151 * @ino: inode number to search for
1153 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1154 * This is for file systems where the inode number is sufficient for unique
1155 * identification of an inode.
1157 * If the inode is in the cache, the inode is returned with an incremented
1160 * Otherwise NULL is returned.
1162 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1164 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1166 return ifind_fast(sb, head, ino);
1168 EXPORT_SYMBOL(ilookup);
1171 * iget5_locked - obtain an inode from a mounted file system
1172 * @sb: super block of file system
1173 * @hashval: hash value (usually inode number) to get
1174 * @test: callback used for comparisons between inodes
1175 * @set: callback used to initialize a new struct inode
1176 * @data: opaque data pointer to pass to @test and @set
1178 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1179 * and @data in the inode cache and if present it is returned with an increased
1180 * reference count. This is a generalized version of iget_locked() for file
1181 * systems where the inode number is not sufficient for unique identification
1184 * If the inode is not in cache, get_new_inode() is called to allocate a new
1185 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1186 * file system gets to fill it in before unlocking it via unlock_new_inode().
1188 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1190 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1191 int (*test)(struct inode *, void *),
1192 int (*set)(struct inode *, void *), void *data)
1194 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1195 struct inode *inode;
1197 inode = ifind(sb, head, test, data, 1);
1201 * get_new_inode() will do the right thing, re-trying the search
1202 * in case it had to block at any point.
1204 return get_new_inode(sb, head, test, set, data);
1206 EXPORT_SYMBOL(iget5_locked);
1209 * iget_locked - obtain an inode from a mounted file system
1210 * @sb: super block of file system
1211 * @ino: inode number to get
1213 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1214 * the inode cache and if present it is returned with an increased reference
1215 * count. This is for file systems where the inode number is sufficient for
1216 * unique identification of an inode.
1218 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1219 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1220 * The file system gets to fill it in before unlocking it via
1221 * unlock_new_inode().
1223 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1225 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1226 struct inode *inode;
1228 inode = ifind_fast(sb, head, ino);
1232 * get_new_inode_fast() will do the right thing, re-trying the search
1233 * in case it had to block at any point.
1235 return get_new_inode_fast(sb, head, ino);
1237 EXPORT_SYMBOL(iget_locked);
1239 int insert_inode_locked(struct inode *inode)
1241 struct super_block *sb = inode->i_sb;
1242 ino_t ino = inode->i_ino;
1243 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1245 inode->i_state |= I_NEW;
1247 struct hlist_node *node;
1248 struct inode *old = NULL;
1249 spin_lock(&inode_lock);
1250 hlist_for_each_entry(old, node, head, i_hash) {
1251 if (old->i_ino != ino)
1253 if (old->i_sb != sb)
1255 if (old->i_state & (I_FREEING|I_WILL_FREE))
1259 if (likely(!node)) {
1260 hlist_add_head(&inode->i_hash, head);
1261 spin_unlock(&inode_lock);
1265 spin_unlock(&inode_lock);
1267 if (unlikely(!inode_unhashed(old))) {
1274 EXPORT_SYMBOL(insert_inode_locked);
1276 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1277 int (*test)(struct inode *, void *), void *data)
1279 struct super_block *sb = inode->i_sb;
1280 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1282 inode->i_state |= I_NEW;
1285 struct hlist_node *node;
1286 struct inode *old = NULL;
1288 spin_lock(&inode_lock);
1289 hlist_for_each_entry(old, node, head, i_hash) {
1290 if (old->i_sb != sb)
1292 if (!test(old, data))
1294 if (old->i_state & (I_FREEING|I_WILL_FREE))
1298 if (likely(!node)) {
1299 hlist_add_head(&inode->i_hash, head);
1300 spin_unlock(&inode_lock);
1304 spin_unlock(&inode_lock);
1306 if (unlikely(!inode_unhashed(old))) {
1313 EXPORT_SYMBOL(insert_inode_locked4);
1316 int generic_delete_inode(struct inode *inode)
1320 EXPORT_SYMBOL(generic_delete_inode);
1323 * Normal UNIX filesystem behaviour: delete the
1324 * inode when the usage count drops to zero, and
1327 int generic_drop_inode(struct inode *inode)
1329 return !inode->i_nlink || inode_unhashed(inode);
1331 EXPORT_SYMBOL_GPL(generic_drop_inode);
1334 * Called when we're dropping the last reference
1337 * Call the FS "drop_inode()" function, defaulting to
1338 * the legacy UNIX filesystem behaviour. If it tells
1339 * us to evict inode, do so. Otherwise, retain inode
1340 * in cache if fs is alive, sync and evict if fs is
1343 static void iput_final(struct inode *inode)
1345 struct super_block *sb = inode->i_sb;
1346 const struct super_operations *op = inode->i_sb->s_op;
1349 if (op && op->drop_inode)
1350 drop = op->drop_inode(inode);
1352 drop = generic_drop_inode(inode);
1355 if (sb->s_flags & MS_ACTIVE) {
1356 inode->i_state |= I_REFERENCED;
1357 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1358 inode_lru_list_add(inode);
1360 spin_unlock(&inode_lock);
1363 WARN_ON(inode->i_state & I_NEW);
1364 inode->i_state |= I_WILL_FREE;
1365 spin_unlock(&inode_lock);
1366 write_inode_now(inode, 1);
1367 spin_lock(&inode_lock);
1368 WARN_ON(inode->i_state & I_NEW);
1369 inode->i_state &= ~I_WILL_FREE;
1370 __remove_inode_hash(inode);
1373 WARN_ON(inode->i_state & I_NEW);
1374 inode->i_state |= I_FREEING;
1377 * Move the inode off the IO lists and LRU once I_FREEING is
1378 * set so that it won't get moved back on there if it is dirty.
1380 inode_lru_list_del(inode);
1381 list_del_init(&inode->i_wb_list);
1383 __inode_sb_list_del(inode);
1384 spin_unlock(&inode_lock);
1386 remove_inode_hash(inode);
1387 wake_up_inode(inode);
1388 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1389 destroy_inode(inode);
1393 * iput - put an inode
1394 * @inode: inode to put
1396 * Puts an inode, dropping its usage count. If the inode use count hits
1397 * zero, the inode is then freed and may also be destroyed.
1399 * Consequently, iput() can sleep.
1401 void iput(struct inode *inode)
1404 BUG_ON(inode->i_state & I_CLEAR);
1406 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1410 EXPORT_SYMBOL(iput);
1413 * bmap - find a block number in a file
1414 * @inode: inode of file
1415 * @block: block to find
1417 * Returns the block number on the device holding the inode that
1418 * is the disk block number for the block of the file requested.
1419 * That is, asked for block 4 of inode 1 the function will return the
1420 * disk block relative to the disk start that holds that block of the
1423 sector_t bmap(struct inode *inode, sector_t block)
1426 if (inode->i_mapping->a_ops->bmap)
1427 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1430 EXPORT_SYMBOL(bmap);
1433 * With relative atime, only update atime if the previous atime is
1434 * earlier than either the ctime or mtime or if at least a day has
1435 * passed since the last atime update.
1437 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1438 struct timespec now)
1441 if (!(mnt->mnt_flags & MNT_RELATIME))
1444 * Is mtime younger than atime? If yes, update atime:
1446 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1449 * Is ctime younger than atime? If yes, update atime:
1451 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1455 * Is the previous atime value older than a day? If yes,
1458 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1461 * Good, we can skip the atime update:
1467 * touch_atime - update the access time
1468 * @mnt: mount the inode is accessed on
1469 * @dentry: dentry accessed
1471 * Update the accessed time on an inode and mark it for writeback.
1472 * This function automatically handles read only file systems and media,
1473 * as well as the "noatime" flag and inode specific "noatime" markers.
1475 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1477 struct inode *inode = dentry->d_inode;
1478 struct timespec now;
1480 if (inode->i_flags & S_NOATIME)
1482 if (IS_NOATIME(inode))
1484 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1487 if (mnt->mnt_flags & MNT_NOATIME)
1489 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1492 now = current_fs_time(inode->i_sb);
1494 if (!relatime_need_update(mnt, inode, now))
1497 if (timespec_equal(&inode->i_atime, &now))
1500 if (mnt_want_write(mnt))
1503 inode->i_atime = now;
1504 mark_inode_dirty_sync(inode);
1505 mnt_drop_write(mnt);
1507 EXPORT_SYMBOL(touch_atime);
1510 * file_update_time - update mtime and ctime time
1511 * @file: file accessed
1513 * Update the mtime and ctime members of an inode and mark the inode
1514 * for writeback. Note that this function is meant exclusively for
1515 * usage in the file write path of filesystems, and filesystems may
1516 * choose to explicitly ignore update via this function with the
1517 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1518 * timestamps are handled by the server.
1521 void file_update_time(struct file *file)
1523 struct inode *inode = file->f_path.dentry->d_inode;
1524 struct timespec now;
1525 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1527 /* First try to exhaust all avenues to not sync */
1528 if (IS_NOCMTIME(inode))
1531 now = current_fs_time(inode->i_sb);
1532 if (!timespec_equal(&inode->i_mtime, &now))
1535 if (!timespec_equal(&inode->i_ctime, &now))
1538 if (IS_I_VERSION(inode))
1539 sync_it |= S_VERSION;
1544 /* Finally allowed to write? Takes lock. */
1545 if (mnt_want_write_file(file))
1548 /* Only change inode inside the lock region */
1549 if (sync_it & S_VERSION)
1550 inode_inc_iversion(inode);
1551 if (sync_it & S_CTIME)
1552 inode->i_ctime = now;
1553 if (sync_it & S_MTIME)
1554 inode->i_mtime = now;
1555 mark_inode_dirty_sync(inode);
1556 mnt_drop_write(file->f_path.mnt);
1558 EXPORT_SYMBOL(file_update_time);
1560 int inode_needs_sync(struct inode *inode)
1564 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1568 EXPORT_SYMBOL(inode_needs_sync);
1570 int inode_wait(void *word)
1575 EXPORT_SYMBOL(inode_wait);
1578 * If we try to find an inode in the inode hash while it is being
1579 * deleted, we have to wait until the filesystem completes its
1580 * deletion before reporting that it isn't found. This function waits
1581 * until the deletion _might_ have completed. Callers are responsible
1582 * to recheck inode state.
1584 * It doesn't matter if I_NEW is not set initially, a call to
1585 * wake_up_inode() after removing from the hash list will DTRT.
1587 * This is called with inode_lock held.
1589 static void __wait_on_freeing_inode(struct inode *inode)
1591 wait_queue_head_t *wq;
1592 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1593 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1594 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1595 spin_unlock(&inode_lock);
1597 finish_wait(wq, &wait.wait);
1598 spin_lock(&inode_lock);
1601 static __initdata unsigned long ihash_entries;
1602 static int __init set_ihash_entries(char *str)
1606 ihash_entries = simple_strtoul(str, &str, 0);
1609 __setup("ihash_entries=", set_ihash_entries);
1612 * Initialize the waitqueues and inode hash table.
1614 void __init inode_init_early(void)
1618 /* If hashes are distributed across NUMA nodes, defer
1619 * hash allocation until vmalloc space is available.
1625 alloc_large_system_hash("Inode-cache",
1626 sizeof(struct hlist_head),
1634 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1635 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1638 void __init inode_init(void)
1642 /* inode slab cache */
1643 inode_cachep = kmem_cache_create("inode_cache",
1644 sizeof(struct inode),
1646 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1649 register_shrinker(&icache_shrinker);
1650 percpu_counter_init(&nr_inodes, 0);
1651 percpu_counter_init(&nr_inodes_unused, 0);
1653 /* Hash may have been set up in inode_init_early */
1658 alloc_large_system_hash("Inode-cache",
1659 sizeof(struct hlist_head),
1667 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1668 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1671 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1673 inode->i_mode = mode;
1674 if (S_ISCHR(mode)) {
1675 inode->i_fop = &def_chr_fops;
1676 inode->i_rdev = rdev;
1677 } else if (S_ISBLK(mode)) {
1678 inode->i_fop = &def_blk_fops;
1679 inode->i_rdev = rdev;
1680 } else if (S_ISFIFO(mode))
1681 inode->i_fop = &def_fifo_fops;
1682 else if (S_ISSOCK(mode))
1683 inode->i_fop = &bad_sock_fops;
1685 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1686 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1689 EXPORT_SYMBOL(init_special_inode);
1692 * Init uid,gid,mode for new inode according to posix standards
1694 * @dir: Directory inode
1695 * @mode: mode of the new inode
1697 void inode_init_owner(struct inode *inode, const struct inode *dir,
1700 inode->i_uid = current_fsuid();
1701 if (dir && dir->i_mode & S_ISGID) {
1702 inode->i_gid = dir->i_gid;
1706 inode->i_gid = current_fsgid();
1707 inode->i_mode = mode;
1709 EXPORT_SYMBOL(inode_init_owner);