2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
7 #include <linux/dcache.h>
8 #include <linux/init.h>
9 #include <linux/slab.h>
10 #include <linux/writeback.h>
11 #include <linux/module.h>
12 #include <linux/backing-dev.h>
13 #include <linux/wait.h>
14 #include <linux/rwsem.h>
15 #include <linux/hash.h>
16 #include <linux/swap.h>
17 #include <linux/security.h>
18 #include <linux/pagemap.h>
19 #include <linux/cdev.h>
20 #include <linux/bootmem.h>
21 #include <linux/fsnotify.h>
22 #include <linux/mount.h>
23 #include <linux/async.h>
24 #include <linux/posix_acl.h>
25 #include <linux/prefetch.h>
26 #include <linux/ima.h>
27 #include <linux/cred.h>
28 #include <linux/buffer_head.h> /* for inode_has_buffers */
32 * Inode locking rules:
34 * inode->i_lock protects:
35 * inode->i_state, inode->i_hash, __iget()
36 * inode->i_sb->s_inode_lru_lock protects:
37 * inode->i_sb->s_inode_lru, inode->i_lru
38 * inode_sb_list_lock protects:
39 * sb->s_inodes, inode->i_sb_list
40 * bdi->wb.list_lock protects:
41 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
42 * inode_hash_lock protects:
43 * inode_hashtable, inode->i_hash
49 * inode->i_sb->s_inode_lru_lock
62 static unsigned int i_hash_mask __read_mostly;
63 static unsigned int i_hash_shift __read_mostly;
64 static struct hlist_head *inode_hashtable __read_mostly;
65 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
67 __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
70 * Empty aops. Can be used for the cases where the user does not
71 * define any of the address_space operations.
73 const struct address_space_operations empty_aops = {
75 EXPORT_SYMBOL(empty_aops);
78 * Statistics gathering..
80 struct inodes_stat_t inodes_stat;
82 static DEFINE_PER_CPU(unsigned int, nr_inodes);
83 static DEFINE_PER_CPU(unsigned int, nr_unused);
85 static struct kmem_cache *inode_cachep __read_mostly;
87 static int get_nr_inodes(void)
91 for_each_possible_cpu(i)
92 sum += per_cpu(nr_inodes, i);
93 return sum < 0 ? 0 : sum;
96 static inline int get_nr_inodes_unused(void)
100 for_each_possible_cpu(i)
101 sum += per_cpu(nr_unused, i);
102 return sum < 0 ? 0 : sum;
105 int get_nr_dirty_inodes(void)
107 /* not actually dirty inodes, but a wild approximation */
108 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
109 return nr_dirty > 0 ? nr_dirty : 0;
113 * Handle nr_inode sysctl
116 int proc_nr_inodes(ctl_table *table, int write,
117 void __user *buffer, size_t *lenp, loff_t *ppos)
119 inodes_stat.nr_inodes = get_nr_inodes();
120 inodes_stat.nr_unused = get_nr_inodes_unused();
121 return proc_dointvec(table, write, buffer, lenp, ppos);
126 * inode_init_always - perform inode structure intialisation
127 * @sb: superblock inode belongs to
128 * @inode: inode to initialise
130 * These are initializations that need to be done on every inode
131 * allocation as the fields are not initialised by slab allocation.
133 int inode_init_always(struct super_block *sb, struct inode *inode)
135 static const struct inode_operations empty_iops;
136 static const struct file_operations empty_fops;
137 struct address_space *const mapping = &inode->i_data;
140 inode->i_blkbits = sb->s_blocksize_bits;
142 atomic_set(&inode->i_count, 1);
143 inode->i_op = &empty_iops;
144 inode->i_fop = &empty_fops;
145 inode->__i_nlink = 1;
146 inode->i_opflags = 0;
149 atomic_set(&inode->i_writecount, 0);
153 inode->i_generation = 0;
155 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
157 inode->i_pipe = NULL;
158 inode->i_bdev = NULL;
159 inode->i_cdev = NULL;
161 inode->dirtied_when = 0;
163 if (security_inode_alloc(inode))
165 spin_lock_init(&inode->i_lock);
166 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
168 mutex_init(&inode->i_mutex);
169 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
171 atomic_set(&inode->i_dio_count, 0);
173 mapping->a_ops = &empty_aops;
174 mapping->host = inode;
176 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
177 mapping->assoc_mapping = NULL;
178 mapping->backing_dev_info = &default_backing_dev_info;
179 mapping->writeback_index = 0;
182 * If the block_device provides a backing_dev_info for client
183 * inodes then use that. Otherwise the inode share the bdev's
187 struct backing_dev_info *bdi;
189 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
190 mapping->backing_dev_info = bdi;
192 inode->i_private = NULL;
193 inode->i_mapping = mapping;
194 INIT_LIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
195 #ifdef CONFIG_FS_POSIX_ACL
196 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
199 #ifdef CONFIG_FSNOTIFY
200 inode->i_fsnotify_mask = 0;
203 this_cpu_inc(nr_inodes);
209 EXPORT_SYMBOL(inode_init_always);
211 static struct inode *alloc_inode(struct super_block *sb)
215 if (sb->s_op->alloc_inode)
216 inode = sb->s_op->alloc_inode(sb);
218 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
223 if (unlikely(inode_init_always(sb, inode))) {
224 if (inode->i_sb->s_op->destroy_inode)
225 inode->i_sb->s_op->destroy_inode(inode);
227 kmem_cache_free(inode_cachep, inode);
234 void free_inode_nonrcu(struct inode *inode)
236 kmem_cache_free(inode_cachep, inode);
238 EXPORT_SYMBOL(free_inode_nonrcu);
240 void __destroy_inode(struct inode *inode)
242 BUG_ON(inode_has_buffers(inode));
243 security_inode_free(inode);
244 fsnotify_inode_delete(inode);
245 #ifdef CONFIG_FS_POSIX_ACL
246 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
247 posix_acl_release(inode->i_acl);
248 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
249 posix_acl_release(inode->i_default_acl);
251 this_cpu_dec(nr_inodes);
253 EXPORT_SYMBOL(__destroy_inode);
255 static void i_callback(struct rcu_head *head)
257 struct inode *inode = container_of(head, struct inode, i_rcu);
258 kmem_cache_free(inode_cachep, inode);
261 static void destroy_inode(struct inode *inode)
263 BUG_ON(!list_empty(&inode->i_lru));
264 __destroy_inode(inode);
265 if (inode->i_sb->s_op->destroy_inode)
266 inode->i_sb->s_op->destroy_inode(inode);
268 call_rcu(&inode->i_rcu, i_callback);
271 void address_space_init_once(struct address_space *mapping)
273 memset(mapping, 0, sizeof(*mapping));
274 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
275 spin_lock_init(&mapping->tree_lock);
276 mutex_init(&mapping->i_mmap_mutex);
277 INIT_LIST_HEAD(&mapping->private_list);
278 spin_lock_init(&mapping->private_lock);
279 INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
280 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
282 EXPORT_SYMBOL(address_space_init_once);
285 * These are initializations that only need to be done
286 * once, because the fields are idempotent across use
287 * of the inode, so let the slab aware of that.
289 void inode_init_once(struct inode *inode)
291 memset(inode, 0, sizeof(*inode));
292 INIT_HLIST_NODE(&inode->i_hash);
293 INIT_LIST_HEAD(&inode->i_devices);
294 INIT_LIST_HEAD(&inode->i_wb_list);
295 INIT_LIST_HEAD(&inode->i_lru);
296 address_space_init_once(&inode->i_data);
297 i_size_ordered_init(inode);
298 #ifdef CONFIG_FSNOTIFY
299 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
302 EXPORT_SYMBOL(inode_init_once);
304 static void init_once(void *foo)
306 struct inode *inode = (struct inode *) foo;
308 inode_init_once(inode);
312 * inode->i_lock must be held
314 void __iget(struct inode *inode)
316 atomic_inc(&inode->i_count);
320 * get additional reference to inode; caller must already hold one.
322 void ihold(struct inode *inode)
324 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
326 EXPORT_SYMBOL(ihold);
328 static void inode_lru_list_add(struct inode *inode)
330 spin_lock(&inode->i_sb->s_inode_lru_lock);
331 if (list_empty(&inode->i_lru)) {
332 list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
333 inode->i_sb->s_nr_inodes_unused++;
334 this_cpu_inc(nr_unused);
336 spin_unlock(&inode->i_sb->s_inode_lru_lock);
339 static void inode_lru_list_del(struct inode *inode)
341 spin_lock(&inode->i_sb->s_inode_lru_lock);
342 if (!list_empty(&inode->i_lru)) {
343 list_del_init(&inode->i_lru);
344 inode->i_sb->s_nr_inodes_unused--;
345 this_cpu_dec(nr_unused);
347 spin_unlock(&inode->i_sb->s_inode_lru_lock);
351 * inode_sb_list_add - add inode to the superblock list of inodes
352 * @inode: inode to add
354 void inode_sb_list_add(struct inode *inode)
356 spin_lock(&inode_sb_list_lock);
357 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
358 spin_unlock(&inode_sb_list_lock);
360 EXPORT_SYMBOL_GPL(inode_sb_list_add);
362 static inline void inode_sb_list_del(struct inode *inode)
364 if (!list_empty(&inode->i_sb_list)) {
365 spin_lock(&inode_sb_list_lock);
366 list_del_init(&inode->i_sb_list);
367 spin_unlock(&inode_sb_list_lock);
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_hash_shift);
378 return tmp & i_hash_mask;
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_hash_lock);
394 spin_lock(&inode->i_lock);
395 hlist_add_head(&inode->i_hash, b);
396 spin_unlock(&inode->i_lock);
397 spin_unlock(&inode_hash_lock);
399 EXPORT_SYMBOL(__insert_inode_hash);
402 * __remove_inode_hash - remove an inode from the hash
403 * @inode: inode to unhash
405 * Remove an inode from the superblock.
407 void __remove_inode_hash(struct inode *inode)
409 spin_lock(&inode_hash_lock);
410 spin_lock(&inode->i_lock);
411 hlist_del_init(&inode->i_hash);
412 spin_unlock(&inode->i_lock);
413 spin_unlock(&inode_hash_lock);
415 EXPORT_SYMBOL(__remove_inode_hash);
417 void end_writeback(struct inode *inode)
421 * We have to cycle tree_lock here because reclaim can be still in the
422 * process of removing the last page (in __delete_from_page_cache())
423 * and we must not free mapping under it.
425 spin_lock_irq(&inode->i_data.tree_lock);
426 BUG_ON(inode->i_data.nrpages);
427 spin_unlock_irq(&inode->i_data.tree_lock);
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 /* don't need i_lock here, no concurrent mods to i_state */
433 inode->i_state = I_FREEING | I_CLEAR;
435 EXPORT_SYMBOL(end_writeback);
438 * Free the inode passed in, removing it from the lists it is still connected
439 * to. We remove any pages still attached to the inode and wait for any IO that
440 * is still in progress before finally destroying the inode.
442 * An inode must already be marked I_FREEING so that we avoid the inode being
443 * moved back onto lists if we race with other code that manipulates the lists
444 * (e.g. writeback_single_inode). The caller is responsible for setting this.
446 * An inode must already be removed from the LRU list before being evicted from
447 * the cache. This should occur atomically with setting the I_FREEING state
448 * flag, so no inodes here should ever be on the LRU when being evicted.
450 static void evict(struct inode *inode)
452 const struct super_operations *op = inode->i_sb->s_op;
454 BUG_ON(!(inode->i_state & I_FREEING));
455 BUG_ON(!list_empty(&inode->i_lru));
457 if (!list_empty(&inode->i_wb_list))
458 inode_wb_list_del(inode);
460 inode_sb_list_del(inode);
462 if (op->evict_inode) {
463 op->evict_inode(inode);
465 if (inode->i_data.nrpages)
466 truncate_inode_pages(&inode->i_data, 0);
467 end_writeback(inode);
469 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
471 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
474 remove_inode_hash(inode);
476 spin_lock(&inode->i_lock);
477 wake_up_bit(&inode->i_state, __I_NEW);
478 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
479 spin_unlock(&inode->i_lock);
481 destroy_inode(inode);
485 * dispose_list - dispose of the contents of a local list
486 * @head: the head of the list to free
488 * Dispose-list gets a local list with local inodes in it, so it doesn't
489 * need to worry about list corruption and SMP locks.
491 static void dispose_list(struct list_head *head)
493 while (!list_empty(head)) {
496 inode = list_first_entry(head, struct inode, i_lru);
497 list_del_init(&inode->i_lru);
504 * evict_inodes - evict all evictable inodes for a superblock
505 * @sb: superblock to operate on
507 * Make sure that no inodes with zero refcount are retained. This is
508 * called by superblock shutdown after having MS_ACTIVE flag removed,
509 * so any inode reaching zero refcount during or after that call will
510 * be immediately evicted.
512 void evict_inodes(struct super_block *sb)
514 struct inode *inode, *next;
517 spin_lock(&inode_sb_list_lock);
518 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
519 if (atomic_read(&inode->i_count))
522 spin_lock(&inode->i_lock);
523 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
524 spin_unlock(&inode->i_lock);
528 inode->i_state |= I_FREEING;
529 inode_lru_list_del(inode);
530 spin_unlock(&inode->i_lock);
531 list_add(&inode->i_lru, &dispose);
533 spin_unlock(&inode_sb_list_lock);
535 dispose_list(&dispose);
539 * invalidate_inodes - attempt to free all inodes on a superblock
540 * @sb: superblock to operate on
541 * @kill_dirty: flag to guide handling of dirty inodes
543 * Attempts to free all inodes for a given superblock. If there were any
544 * busy inodes return a non-zero value, else zero.
545 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
548 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
551 struct inode *inode, *next;
554 spin_lock(&inode_sb_list_lock);
555 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
556 spin_lock(&inode->i_lock);
557 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
558 spin_unlock(&inode->i_lock);
561 if (inode->i_state & I_DIRTY && !kill_dirty) {
562 spin_unlock(&inode->i_lock);
566 if (atomic_read(&inode->i_count)) {
567 spin_unlock(&inode->i_lock);
572 inode->i_state |= I_FREEING;
573 inode_lru_list_del(inode);
574 spin_unlock(&inode->i_lock);
575 list_add(&inode->i_lru, &dispose);
577 spin_unlock(&inode_sb_list_lock);
579 dispose_list(&dispose);
584 static int can_unuse(struct inode *inode)
586 if (inode->i_state & ~I_REFERENCED)
588 if (inode_has_buffers(inode))
590 if (atomic_read(&inode->i_count))
592 if (inode->i_data.nrpages)
598 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
599 * This is called from the superblock shrinker function with a number of inodes
600 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
601 * then are freed outside inode_lock by dispose_list().
603 * Any inodes which are pinned purely because of attached pagecache have their
604 * pagecache removed. If the inode has metadata buffers attached to
605 * mapping->private_list then try to remove them.
607 * If the inode has the I_REFERENCED flag set, then it means that it has been
608 * used recently - the flag is set in iput_final(). When we encounter such an
609 * inode, clear the flag and move it to the back of the LRU so it gets another
610 * pass through the LRU before it gets reclaimed. This is necessary because of
611 * the fact we are doing lazy LRU updates to minimise lock contention so the
612 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
613 * with this flag set because they are the inodes that are out of order.
615 void prune_icache_sb(struct super_block *sb, int nr_to_scan)
619 unsigned long reap = 0;
621 spin_lock(&sb->s_inode_lru_lock);
622 for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
625 if (list_empty(&sb->s_inode_lru))
628 inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
631 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
632 * so use a trylock. If we fail to get the lock, just move the
633 * inode to the back of the list so we don't spin on it.
635 if (!spin_trylock(&inode->i_lock)) {
636 list_move_tail(&inode->i_lru, &sb->s_inode_lru);
641 * Referenced or dirty inodes are still in use. Give them
642 * another pass through the LRU as we canot reclaim them now.
644 if (atomic_read(&inode->i_count) ||
645 (inode->i_state & ~I_REFERENCED)) {
646 list_del_init(&inode->i_lru);
647 spin_unlock(&inode->i_lock);
648 sb->s_nr_inodes_unused--;
649 this_cpu_dec(nr_unused);
653 /* recently referenced inodes get one more pass */
654 if (inode->i_state & I_REFERENCED) {
655 inode->i_state &= ~I_REFERENCED;
656 list_move(&inode->i_lru, &sb->s_inode_lru);
657 spin_unlock(&inode->i_lock);
660 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
662 spin_unlock(&inode->i_lock);
663 spin_unlock(&sb->s_inode_lru_lock);
664 if (remove_inode_buffers(inode))
665 reap += invalidate_mapping_pages(&inode->i_data,
668 spin_lock(&sb->s_inode_lru_lock);
670 if (inode != list_entry(sb->s_inode_lru.next,
671 struct inode, i_lru))
672 continue; /* wrong inode or list_empty */
673 /* avoid lock inversions with trylock */
674 if (!spin_trylock(&inode->i_lock))
676 if (!can_unuse(inode)) {
677 spin_unlock(&inode->i_lock);
681 WARN_ON(inode->i_state & I_NEW);
682 inode->i_state |= I_FREEING;
683 spin_unlock(&inode->i_lock);
685 list_move(&inode->i_lru, &freeable);
686 sb->s_nr_inodes_unused--;
687 this_cpu_dec(nr_unused);
689 if (current_is_kswapd())
690 __count_vm_events(KSWAPD_INODESTEAL, reap);
692 __count_vm_events(PGINODESTEAL, reap);
693 spin_unlock(&sb->s_inode_lru_lock);
695 dispose_list(&freeable);
698 static void __wait_on_freeing_inode(struct inode *inode);
700 * Called with the inode lock held.
702 static struct inode *find_inode(struct super_block *sb,
703 struct hlist_head *head,
704 int (*test)(struct inode *, void *),
707 struct hlist_node *node;
708 struct inode *inode = NULL;
711 hlist_for_each_entry(inode, node, head, i_hash) {
712 spin_lock(&inode->i_lock);
713 if (inode->i_sb != sb) {
714 spin_unlock(&inode->i_lock);
717 if (!test(inode, data)) {
718 spin_unlock(&inode->i_lock);
721 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
722 __wait_on_freeing_inode(inode);
726 spin_unlock(&inode->i_lock);
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 spin_lock(&inode->i_lock);
745 if (inode->i_ino != ino) {
746 spin_unlock(&inode->i_lock);
749 if (inode->i_sb != sb) {
750 spin_unlock(&inode->i_lock);
753 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
754 __wait_on_freeing_inode(inode);
758 spin_unlock(&inode->i_lock);
765 * Each cpu owns a range of LAST_INO_BATCH numbers.
766 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
767 * to renew the exhausted range.
769 * This does not significantly increase overflow rate because every CPU can
770 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
771 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
772 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
773 * overflow rate by 2x, which does not seem too significant.
775 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
776 * error if st_ino won't fit in target struct field. Use 32bit counter
777 * here to attempt to avoid that.
779 #define LAST_INO_BATCH 1024
780 static DEFINE_PER_CPU(unsigned int, last_ino);
782 unsigned int get_next_ino(void)
784 unsigned int *p = &get_cpu_var(last_ino);
785 unsigned int res = *p;
788 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
789 static atomic_t shared_last_ino;
790 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
792 res = next - LAST_INO_BATCH;
797 put_cpu_var(last_ino);
800 EXPORT_SYMBOL(get_next_ino);
803 * new_inode_pseudo - obtain an inode
806 * Allocates a new inode for given superblock.
807 * Inode wont be chained in superblock s_inodes list
809 * - fs can't be unmount
810 * - quotas, fsnotify, writeback can't work
812 struct inode *new_inode_pseudo(struct super_block *sb)
814 struct inode *inode = alloc_inode(sb);
817 spin_lock(&inode->i_lock);
819 spin_unlock(&inode->i_lock);
820 INIT_LIST_HEAD(&inode->i_sb_list);
826 * new_inode - obtain an inode
829 * Allocates a new inode for given superblock. The default gfp_mask
830 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
831 * If HIGHMEM pages are unsuitable or it is known that pages allocated
832 * for the page cache are not reclaimable or migratable,
833 * mapping_set_gfp_mask() must be called with suitable flags on the
834 * newly created inode's mapping
837 struct inode *new_inode(struct super_block *sb)
841 spin_lock_prefetch(&inode_sb_list_lock);
843 inode = new_inode_pseudo(sb);
845 inode_sb_list_add(inode);
848 EXPORT_SYMBOL(new_inode);
850 #ifdef CONFIG_DEBUG_LOCK_ALLOC
851 void lockdep_annotate_inode_mutex_key(struct inode *inode)
853 if (S_ISDIR(inode->i_mode)) {
854 struct file_system_type *type = inode->i_sb->s_type;
856 /* Set new key only if filesystem hasn't already changed it */
857 if (!lockdep_match_class(&inode->i_mutex,
858 &type->i_mutex_key)) {
860 * ensure nobody is actually holding i_mutex
862 mutex_destroy(&inode->i_mutex);
863 mutex_init(&inode->i_mutex);
864 lockdep_set_class(&inode->i_mutex,
865 &type->i_mutex_dir_key);
869 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
873 * unlock_new_inode - clear the I_NEW state and wake up any waiters
874 * @inode: new inode to unlock
876 * Called when the inode is fully initialised to clear the new state of the
877 * inode and wake up anyone waiting for the inode to finish initialisation.
879 void unlock_new_inode(struct inode *inode)
881 lockdep_annotate_inode_mutex_key(inode);
882 spin_lock(&inode->i_lock);
883 WARN_ON(!(inode->i_state & I_NEW));
884 inode->i_state &= ~I_NEW;
885 wake_up_bit(&inode->i_state, __I_NEW);
886 spin_unlock(&inode->i_lock);
888 EXPORT_SYMBOL(unlock_new_inode);
891 * iget5_locked - obtain an inode from a mounted file system
892 * @sb: super block of file system
893 * @hashval: hash value (usually inode number) to get
894 * @test: callback used for comparisons between inodes
895 * @set: callback used to initialize a new struct inode
896 * @data: opaque data pointer to pass to @test and @set
898 * Search for the inode specified by @hashval and @data in the inode cache,
899 * and if present it is return it with an increased reference count. This is
900 * a generalized version of iget_locked() for file systems where the inode
901 * number is not sufficient for unique identification of an inode.
903 * If the inode is not in cache, allocate a new inode and return it locked,
904 * hashed, and with the I_NEW flag set. The file system gets to fill it in
905 * before unlocking it via unlock_new_inode().
907 * Note both @test and @set are called with the inode_hash_lock held, so can't
910 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
911 int (*test)(struct inode *, void *),
912 int (*set)(struct inode *, void *), void *data)
914 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
917 spin_lock(&inode_hash_lock);
918 inode = find_inode(sb, head, test, data);
919 spin_unlock(&inode_hash_lock);
922 wait_on_inode(inode);
926 inode = alloc_inode(sb);
930 spin_lock(&inode_hash_lock);
931 /* We released the lock, so.. */
932 old = find_inode(sb, head, test, data);
934 if (set(inode, data))
937 spin_lock(&inode->i_lock);
938 inode->i_state = I_NEW;
939 hlist_add_head(&inode->i_hash, head);
940 spin_unlock(&inode->i_lock);
941 inode_sb_list_add(inode);
942 spin_unlock(&inode_hash_lock);
944 /* Return the locked inode with I_NEW set, the
945 * caller is responsible for filling in the contents
951 * Uhhuh, somebody else created the same inode under
952 * us. Use the old inode instead of the one we just
955 spin_unlock(&inode_hash_lock);
956 destroy_inode(inode);
958 wait_on_inode(inode);
963 spin_unlock(&inode_hash_lock);
964 destroy_inode(inode);
967 EXPORT_SYMBOL(iget5_locked);
970 * iget_locked - obtain an inode from a mounted file system
971 * @sb: super block of file system
972 * @ino: inode number to get
974 * Search for the inode specified by @ino in the inode cache and if present
975 * return it with an increased reference count. This is for file systems
976 * where the inode number is sufficient for unique identification of an inode.
978 * If the inode is not in cache, allocate a new inode and return it locked,
979 * hashed, and with the I_NEW flag set. The file system gets to fill it in
980 * before unlocking it via unlock_new_inode().
982 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
984 struct hlist_head *head = inode_hashtable + hash(sb, ino);
987 spin_lock(&inode_hash_lock);
988 inode = find_inode_fast(sb, head, ino);
989 spin_unlock(&inode_hash_lock);
991 wait_on_inode(inode);
995 inode = alloc_inode(sb);
999 spin_lock(&inode_hash_lock);
1000 /* We released the lock, so.. */
1001 old = find_inode_fast(sb, head, ino);
1004 spin_lock(&inode->i_lock);
1005 inode->i_state = I_NEW;
1006 hlist_add_head(&inode->i_hash, head);
1007 spin_unlock(&inode->i_lock);
1008 inode_sb_list_add(inode);
1009 spin_unlock(&inode_hash_lock);
1011 /* Return the locked inode with I_NEW set, the
1012 * caller is responsible for filling in the contents
1018 * Uhhuh, somebody else created the same inode under
1019 * us. Use the old inode instead of the one we just
1022 spin_unlock(&inode_hash_lock);
1023 destroy_inode(inode);
1025 wait_on_inode(inode);
1029 EXPORT_SYMBOL(iget_locked);
1032 * search the inode cache for a matching inode number.
1033 * If we find one, then the inode number we are trying to
1034 * allocate is not unique and so we should not use it.
1036 * Returns 1 if the inode number is unique, 0 if it is not.
1038 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1040 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1041 struct hlist_node *node;
1042 struct inode *inode;
1044 spin_lock(&inode_hash_lock);
1045 hlist_for_each_entry(inode, node, b, i_hash) {
1046 if (inode->i_ino == ino && inode->i_sb == sb) {
1047 spin_unlock(&inode_hash_lock);
1051 spin_unlock(&inode_hash_lock);
1057 * iunique - get a unique inode number
1059 * @max_reserved: highest reserved inode number
1061 * Obtain an inode number that is unique on the system for a given
1062 * superblock. This is used by file systems that have no natural
1063 * permanent inode numbering system. An inode number is returned that
1064 * is higher than the reserved limit but unique.
1067 * With a large number of inodes live on the file system this function
1068 * currently becomes quite slow.
1070 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1073 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1074 * error if st_ino won't fit in target struct field. Use 32bit counter
1075 * here to attempt to avoid that.
1077 static DEFINE_SPINLOCK(iunique_lock);
1078 static unsigned int counter;
1081 spin_lock(&iunique_lock);
1083 if (counter <= max_reserved)
1084 counter = max_reserved + 1;
1086 } while (!test_inode_iunique(sb, res));
1087 spin_unlock(&iunique_lock);
1091 EXPORT_SYMBOL(iunique);
1093 struct inode *igrab(struct inode *inode)
1095 spin_lock(&inode->i_lock);
1096 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1098 spin_unlock(&inode->i_lock);
1100 spin_unlock(&inode->i_lock);
1102 * Handle the case where s_op->clear_inode is not been
1103 * called yet, and somebody is calling igrab
1104 * while the inode is getting freed.
1110 EXPORT_SYMBOL(igrab);
1113 * ilookup5_nowait - search for an inode in the inode cache
1114 * @sb: super block of file system to search
1115 * @hashval: hash value (usually inode number) to search for
1116 * @test: callback used for comparisons between inodes
1117 * @data: opaque data pointer to pass to @test
1119 * Search for the inode specified by @hashval and @data in the inode cache.
1120 * If the inode is in the cache, the inode is returned with an incremented
1123 * Note: I_NEW is not waited upon so you have to be very careful what you do
1124 * with the returned inode. You probably should be using ilookup5() instead.
1126 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1128 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1129 int (*test)(struct inode *, void *), void *data)
1131 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1132 struct inode *inode;
1134 spin_lock(&inode_hash_lock);
1135 inode = find_inode(sb, head, test, data);
1136 spin_unlock(&inode_hash_lock);
1140 EXPORT_SYMBOL(ilookup5_nowait);
1143 * ilookup5 - search for an inode in the inode cache
1144 * @sb: super block of file system to search
1145 * @hashval: hash value (usually inode number) to search for
1146 * @test: callback used for comparisons between inodes
1147 * @data: opaque data pointer to pass to @test
1149 * Search for the inode specified by @hashval and @data in the inode cache,
1150 * and if the inode is in the cache, return the inode with an incremented
1151 * reference count. Waits on I_NEW before returning the inode.
1152 * returned with an incremented reference count.
1154 * This is a generalized version of ilookup() for file systems where the
1155 * inode number is not sufficient for unique identification of an inode.
1157 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1159 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1160 int (*test)(struct inode *, void *), void *data)
1162 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1165 wait_on_inode(inode);
1168 EXPORT_SYMBOL(ilookup5);
1171 * ilookup - search for an inode in the inode cache
1172 * @sb: super block of file system to search
1173 * @ino: inode number to search for
1175 * Search for the inode @ino in the inode cache, and if the inode is in the
1176 * cache, the inode is returned with an incremented reference count.
1178 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1180 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1181 struct inode *inode;
1183 spin_lock(&inode_hash_lock);
1184 inode = find_inode_fast(sb, head, ino);
1185 spin_unlock(&inode_hash_lock);
1188 wait_on_inode(inode);
1191 EXPORT_SYMBOL(ilookup);
1193 int insert_inode_locked(struct inode *inode)
1195 struct super_block *sb = inode->i_sb;
1196 ino_t ino = inode->i_ino;
1197 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1200 struct hlist_node *node;
1201 struct inode *old = NULL;
1202 spin_lock(&inode_hash_lock);
1203 hlist_for_each_entry(old, node, head, i_hash) {
1204 if (old->i_ino != ino)
1206 if (old->i_sb != sb)
1208 spin_lock(&old->i_lock);
1209 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1210 spin_unlock(&old->i_lock);
1215 if (likely(!node)) {
1216 spin_lock(&inode->i_lock);
1217 inode->i_state |= I_NEW;
1218 hlist_add_head(&inode->i_hash, head);
1219 spin_unlock(&inode->i_lock);
1220 spin_unlock(&inode_hash_lock);
1224 spin_unlock(&old->i_lock);
1225 spin_unlock(&inode_hash_lock);
1227 if (unlikely(!inode_unhashed(old))) {
1234 EXPORT_SYMBOL(insert_inode_locked);
1236 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1237 int (*test)(struct inode *, void *), void *data)
1239 struct super_block *sb = inode->i_sb;
1240 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1243 struct hlist_node *node;
1244 struct inode *old = NULL;
1246 spin_lock(&inode_hash_lock);
1247 hlist_for_each_entry(old, node, head, i_hash) {
1248 if (old->i_sb != sb)
1250 if (!test(old, data))
1252 spin_lock(&old->i_lock);
1253 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1254 spin_unlock(&old->i_lock);
1259 if (likely(!node)) {
1260 spin_lock(&inode->i_lock);
1261 inode->i_state |= I_NEW;
1262 hlist_add_head(&inode->i_hash, head);
1263 spin_unlock(&inode->i_lock);
1264 spin_unlock(&inode_hash_lock);
1268 spin_unlock(&old->i_lock);
1269 spin_unlock(&inode_hash_lock);
1271 if (unlikely(!inode_unhashed(old))) {
1278 EXPORT_SYMBOL(insert_inode_locked4);
1281 int generic_delete_inode(struct inode *inode)
1285 EXPORT_SYMBOL(generic_delete_inode);
1288 * Normal UNIX filesystem behaviour: delete the
1289 * inode when the usage count drops to zero, and
1292 int generic_drop_inode(struct inode *inode)
1294 return !inode->i_nlink || inode_unhashed(inode);
1296 EXPORT_SYMBOL_GPL(generic_drop_inode);
1299 * Called when we're dropping the last reference
1302 * Call the FS "drop_inode()" function, defaulting to
1303 * the legacy UNIX filesystem behaviour. If it tells
1304 * us to evict inode, do so. Otherwise, retain inode
1305 * in cache if fs is alive, sync and evict if fs is
1308 static void iput_final(struct inode *inode)
1310 struct super_block *sb = inode->i_sb;
1311 const struct super_operations *op = inode->i_sb->s_op;
1314 WARN_ON(inode->i_state & I_NEW);
1317 drop = op->drop_inode(inode);
1319 drop = generic_drop_inode(inode);
1321 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1322 inode->i_state |= I_REFERENCED;
1323 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1324 inode_lru_list_add(inode);
1325 spin_unlock(&inode->i_lock);
1330 inode->i_state |= I_WILL_FREE;
1331 spin_unlock(&inode->i_lock);
1332 write_inode_now(inode, 1);
1333 spin_lock(&inode->i_lock);
1334 WARN_ON(inode->i_state & I_NEW);
1335 inode->i_state &= ~I_WILL_FREE;
1338 inode->i_state |= I_FREEING;
1339 if (!list_empty(&inode->i_lru))
1340 inode_lru_list_del(inode);
1341 spin_unlock(&inode->i_lock);
1347 * iput - put an inode
1348 * @inode: inode to put
1350 * Puts an inode, dropping its usage count. If the inode use count hits
1351 * zero, the inode is then freed and may also be destroyed.
1353 * Consequently, iput() can sleep.
1355 void iput(struct inode *inode)
1358 BUG_ON(inode->i_state & I_CLEAR);
1360 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1364 EXPORT_SYMBOL(iput);
1367 * bmap - find a block number in a file
1368 * @inode: inode of file
1369 * @block: block to find
1371 * Returns the block number on the device holding the inode that
1372 * is the disk block number for the block of the file requested.
1373 * That is, asked for block 4 of inode 1 the function will return the
1374 * disk block relative to the disk start that holds that block of the
1377 sector_t bmap(struct inode *inode, sector_t block)
1380 if (inode->i_mapping->a_ops->bmap)
1381 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1384 EXPORT_SYMBOL(bmap);
1387 * With relative atime, only update atime if the previous atime is
1388 * earlier than either the ctime or mtime or if at least a day has
1389 * passed since the last atime update.
1391 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1392 struct timespec now)
1395 if (!(mnt->mnt_flags & MNT_RELATIME))
1398 * Is mtime younger than atime? If yes, update atime:
1400 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1403 * Is ctime younger than atime? If yes, update atime:
1405 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1409 * Is the previous atime value older than a day? If yes,
1412 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1415 * Good, we can skip the atime update:
1421 * touch_atime - update the access time
1422 * @mnt: mount the inode is accessed on
1423 * @dentry: dentry accessed
1425 * Update the accessed time on an inode and mark it for writeback.
1426 * This function automatically handles read only file systems and media,
1427 * as well as the "noatime" flag and inode specific "noatime" markers.
1429 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1431 struct inode *inode = dentry->d_inode;
1432 struct timespec now;
1434 if (inode->i_flags & S_NOATIME)
1436 if (IS_NOATIME(inode))
1438 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1441 if (mnt->mnt_flags & MNT_NOATIME)
1443 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1446 now = current_fs_time(inode->i_sb);
1448 if (!relatime_need_update(mnt, inode, now))
1451 if (timespec_equal(&inode->i_atime, &now))
1454 if (mnt_want_write(mnt))
1457 inode->i_atime = now;
1458 mark_inode_dirty_sync(inode);
1459 mnt_drop_write(mnt);
1461 EXPORT_SYMBOL(touch_atime);
1464 * file_update_time - update mtime and ctime time
1465 * @file: file accessed
1467 * Update the mtime and ctime members of an inode and mark the inode
1468 * for writeback. Note that this function is meant exclusively for
1469 * usage in the file write path of filesystems, and filesystems may
1470 * choose to explicitly ignore update via this function with the
1471 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1472 * timestamps are handled by the server.
1475 void file_update_time(struct file *file)
1477 struct inode *inode = file->f_path.dentry->d_inode;
1478 struct timespec now;
1479 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1481 /* First try to exhaust all avenues to not sync */
1482 if (IS_NOCMTIME(inode))
1485 now = current_fs_time(inode->i_sb);
1486 if (!timespec_equal(&inode->i_mtime, &now))
1489 if (!timespec_equal(&inode->i_ctime, &now))
1492 if (IS_I_VERSION(inode))
1493 sync_it |= S_VERSION;
1498 /* Finally allowed to write? Takes lock. */
1499 if (mnt_want_write_file(file))
1502 /* Only change inode inside the lock region */
1503 if (sync_it & S_VERSION)
1504 inode_inc_iversion(inode);
1505 if (sync_it & S_CTIME)
1506 inode->i_ctime = now;
1507 if (sync_it & S_MTIME)
1508 inode->i_mtime = now;
1509 mark_inode_dirty_sync(inode);
1510 mnt_drop_write_file(file);
1512 EXPORT_SYMBOL(file_update_time);
1514 int inode_needs_sync(struct inode *inode)
1518 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1522 EXPORT_SYMBOL(inode_needs_sync);
1524 int inode_wait(void *word)
1529 EXPORT_SYMBOL(inode_wait);
1532 * If we try to find an inode in the inode hash while it is being
1533 * deleted, we have to wait until the filesystem completes its
1534 * deletion before reporting that it isn't found. This function waits
1535 * until the deletion _might_ have completed. Callers are responsible
1536 * to recheck inode state.
1538 * It doesn't matter if I_NEW is not set initially, a call to
1539 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1542 static void __wait_on_freeing_inode(struct inode *inode)
1544 wait_queue_head_t *wq;
1545 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1546 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1547 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1548 spin_unlock(&inode->i_lock);
1549 spin_unlock(&inode_hash_lock);
1551 finish_wait(wq, &wait.wait);
1552 spin_lock(&inode_hash_lock);
1555 static __initdata unsigned long ihash_entries;
1556 static int __init set_ihash_entries(char *str)
1560 ihash_entries = simple_strtoul(str, &str, 0);
1563 __setup("ihash_entries=", set_ihash_entries);
1566 * Initialize the waitqueues and inode hash table.
1568 void __init inode_init_early(void)
1572 /* If hashes are distributed across NUMA nodes, defer
1573 * hash allocation until vmalloc space is available.
1579 alloc_large_system_hash("Inode-cache",
1580 sizeof(struct hlist_head),
1588 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1589 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1592 void __init inode_init(void)
1596 /* inode slab cache */
1597 inode_cachep = kmem_cache_create("inode_cache",
1598 sizeof(struct inode),
1600 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1604 /* Hash may have been set up in inode_init_early */
1609 alloc_large_system_hash("Inode-cache",
1610 sizeof(struct hlist_head),
1618 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1619 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1622 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1624 inode->i_mode = mode;
1625 if (S_ISCHR(mode)) {
1626 inode->i_fop = &def_chr_fops;
1627 inode->i_rdev = rdev;
1628 } else if (S_ISBLK(mode)) {
1629 inode->i_fop = &def_blk_fops;
1630 inode->i_rdev = rdev;
1631 } else if (S_ISFIFO(mode))
1632 inode->i_fop = &def_fifo_fops;
1633 else if (S_ISSOCK(mode))
1634 inode->i_fop = &bad_sock_fops;
1636 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1637 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1640 EXPORT_SYMBOL(init_special_inode);
1643 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1645 * @dir: Directory inode
1646 * @mode: mode of the new inode
1648 void inode_init_owner(struct inode *inode, const struct inode *dir,
1651 inode->i_uid = current_fsuid();
1652 if (dir && dir->i_mode & S_ISGID) {
1653 inode->i_gid = dir->i_gid;
1657 inode->i_gid = current_fsgid();
1658 inode->i_mode = mode;
1660 EXPORT_SYMBOL(inode_init_owner);
1663 * inode_owner_or_capable - check current task permissions to inode
1664 * @inode: inode being checked
1666 * Return true if current either has CAP_FOWNER to the inode, or
1669 bool inode_owner_or_capable(const struct inode *inode)
1671 struct user_namespace *ns = inode_userns(inode);
1673 if (current_user_ns() == ns && current_fsuid() == inode->i_uid)
1675 if (ns_capable(ns, CAP_FOWNER))
1679 EXPORT_SYMBOL(inode_owner_or_capable);