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;
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 #ifdef CONFIG_FS_POSIX_ACL
195 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
198 #ifdef CONFIG_FSNOTIFY
199 inode->i_fsnotify_mask = 0;
202 this_cpu_inc(nr_inodes);
208 EXPORT_SYMBOL(inode_init_always);
210 static struct inode *alloc_inode(struct super_block *sb)
214 if (sb->s_op->alloc_inode)
215 inode = sb->s_op->alloc_inode(sb);
217 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
222 if (unlikely(inode_init_always(sb, inode))) {
223 if (inode->i_sb->s_op->destroy_inode)
224 inode->i_sb->s_op->destroy_inode(inode);
226 kmem_cache_free(inode_cachep, inode);
233 void free_inode_nonrcu(struct inode *inode)
235 kmem_cache_free(inode_cachep, inode);
237 EXPORT_SYMBOL(free_inode_nonrcu);
239 void __destroy_inode(struct inode *inode)
241 BUG_ON(inode_has_buffers(inode));
242 security_inode_free(inode);
243 fsnotify_inode_delete(inode);
244 #ifdef CONFIG_FS_POSIX_ACL
245 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
246 posix_acl_release(inode->i_acl);
247 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
248 posix_acl_release(inode->i_default_acl);
250 this_cpu_dec(nr_inodes);
252 EXPORT_SYMBOL(__destroy_inode);
254 static void i_callback(struct rcu_head *head)
256 struct inode *inode = container_of(head, struct inode, i_rcu);
257 INIT_LIST_HEAD(&inode->i_dentry);
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_dentry);
294 INIT_LIST_HEAD(&inode->i_devices);
295 INIT_LIST_HEAD(&inode->i_wb_list);
296 INIT_LIST_HEAD(&inode->i_lru);
297 address_space_init_once(&inode->i_data);
298 i_size_ordered_init(inode);
299 #ifdef CONFIG_FSNOTIFY
300 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
303 EXPORT_SYMBOL(inode_init_once);
305 static void init_once(void *foo)
307 struct inode *inode = (struct inode *) foo;
309 inode_init_once(inode);
313 * inode->i_lock must be held
315 void __iget(struct inode *inode)
317 atomic_inc(&inode->i_count);
321 * get additional reference to inode; caller must already hold one.
323 void ihold(struct inode *inode)
325 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
327 EXPORT_SYMBOL(ihold);
329 static void inode_lru_list_add(struct inode *inode)
331 spin_lock(&inode->i_sb->s_inode_lru_lock);
332 if (list_empty(&inode->i_lru)) {
333 list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
334 inode->i_sb->s_nr_inodes_unused++;
335 this_cpu_inc(nr_unused);
337 spin_unlock(&inode->i_sb->s_inode_lru_lock);
340 static void inode_lru_list_del(struct inode *inode)
342 spin_lock(&inode->i_sb->s_inode_lru_lock);
343 if (!list_empty(&inode->i_lru)) {
344 list_del_init(&inode->i_lru);
345 inode->i_sb->s_nr_inodes_unused--;
346 this_cpu_dec(nr_unused);
348 spin_unlock(&inode->i_sb->s_inode_lru_lock);
352 * inode_sb_list_add - add inode to the superblock list of inodes
353 * @inode: inode to add
355 void inode_sb_list_add(struct inode *inode)
357 spin_lock(&inode_sb_list_lock);
358 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
359 spin_unlock(&inode_sb_list_lock);
361 EXPORT_SYMBOL_GPL(inode_sb_list_add);
363 static inline void inode_sb_list_del(struct inode *inode)
365 if (!list_empty(&inode->i_sb_list)) {
366 spin_lock(&inode_sb_list_lock);
367 list_del_init(&inode->i_sb_list);
368 spin_unlock(&inode_sb_list_lock);
372 static unsigned long hash(struct super_block *sb, unsigned long hashval)
376 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
378 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
379 return tmp & i_hash_mask;
383 * __insert_inode_hash - hash an inode
384 * @inode: unhashed inode
385 * @hashval: unsigned long value used to locate this object in the
388 * Add an inode to the inode hash for this superblock.
390 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
392 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
394 spin_lock(&inode_hash_lock);
395 spin_lock(&inode->i_lock);
396 hlist_add_head(&inode->i_hash, b);
397 spin_unlock(&inode->i_lock);
398 spin_unlock(&inode_hash_lock);
400 EXPORT_SYMBOL(__insert_inode_hash);
403 * __remove_inode_hash - remove an inode from the hash
404 * @inode: inode to unhash
406 * Remove an inode from the superblock.
408 void __remove_inode_hash(struct inode *inode)
410 spin_lock(&inode_hash_lock);
411 spin_lock(&inode->i_lock);
412 hlist_del_init(&inode->i_hash);
413 spin_unlock(&inode->i_lock);
414 spin_unlock(&inode_hash_lock);
416 EXPORT_SYMBOL(__remove_inode_hash);
418 void end_writeback(struct inode *inode)
422 * We have to cycle tree_lock here because reclaim can be still in the
423 * process of removing the last page (in __delete_from_page_cache())
424 * and we must not free mapping under it.
426 spin_lock_irq(&inode->i_data.tree_lock);
427 BUG_ON(inode->i_data.nrpages);
428 spin_unlock_irq(&inode->i_data.tree_lock);
429 BUG_ON(!list_empty(&inode->i_data.private_list));
430 BUG_ON(!(inode->i_state & I_FREEING));
431 BUG_ON(inode->i_state & I_CLEAR);
432 inode_sync_wait(inode);
433 /* don't need i_lock here, no concurrent mods to i_state */
434 inode->i_state = I_FREEING | I_CLEAR;
436 EXPORT_SYMBOL(end_writeback);
439 * Free the inode passed in, removing it from the lists it is still connected
440 * to. We remove any pages still attached to the inode and wait for any IO that
441 * is still in progress before finally destroying the inode.
443 * An inode must already be marked I_FREEING so that we avoid the inode being
444 * moved back onto lists if we race with other code that manipulates the lists
445 * (e.g. writeback_single_inode). The caller is responsible for setting this.
447 * An inode must already be removed from the LRU list before being evicted from
448 * the cache. This should occur atomically with setting the I_FREEING state
449 * flag, so no inodes here should ever be on the LRU when being evicted.
451 static void evict(struct inode *inode)
453 const struct super_operations *op = inode->i_sb->s_op;
455 BUG_ON(!(inode->i_state & I_FREEING));
456 BUG_ON(!list_empty(&inode->i_lru));
458 if (!list_empty(&inode->i_wb_list))
459 inode_wb_list_del(inode);
461 inode_sb_list_del(inode);
463 if (op->evict_inode) {
464 op->evict_inode(inode);
466 if (inode->i_data.nrpages)
467 truncate_inode_pages(&inode->i_data, 0);
468 end_writeback(inode);
470 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
472 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
475 remove_inode_hash(inode);
477 spin_lock(&inode->i_lock);
478 wake_up_bit(&inode->i_state, __I_NEW);
479 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
480 spin_unlock(&inode->i_lock);
482 destroy_inode(inode);
486 * dispose_list - dispose of the contents of a local list
487 * @head: the head of the list to free
489 * Dispose-list gets a local list with local inodes in it, so it doesn't
490 * need to worry about list corruption and SMP locks.
492 static void dispose_list(struct list_head *head)
494 while (!list_empty(head)) {
497 inode = list_first_entry(head, struct inode, i_lru);
498 list_del_init(&inode->i_lru);
505 * evict_inodes - evict all evictable inodes for a superblock
506 * @sb: superblock to operate on
508 * Make sure that no inodes with zero refcount are retained. This is
509 * called by superblock shutdown after having MS_ACTIVE flag removed,
510 * so any inode reaching zero refcount during or after that call will
511 * be immediately evicted.
513 void evict_inodes(struct super_block *sb)
515 struct inode *inode, *next;
518 spin_lock(&inode_sb_list_lock);
519 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
520 if (atomic_read(&inode->i_count))
523 spin_lock(&inode->i_lock);
524 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
525 spin_unlock(&inode->i_lock);
529 inode->i_state |= I_FREEING;
530 inode_lru_list_del(inode);
531 spin_unlock(&inode->i_lock);
532 list_add(&inode->i_lru, &dispose);
534 spin_unlock(&inode_sb_list_lock);
536 dispose_list(&dispose);
540 * invalidate_inodes - attempt to free all inodes on a superblock
541 * @sb: superblock to operate on
542 * @kill_dirty: flag to guide handling of dirty inodes
544 * Attempts to free all inodes for a given superblock. If there were any
545 * busy inodes return a non-zero value, else zero.
546 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
549 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
552 struct inode *inode, *next;
555 spin_lock(&inode_sb_list_lock);
556 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
557 spin_lock(&inode->i_lock);
558 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
559 spin_unlock(&inode->i_lock);
562 if (inode->i_state & I_DIRTY && !kill_dirty) {
563 spin_unlock(&inode->i_lock);
567 if (atomic_read(&inode->i_count)) {
568 spin_unlock(&inode->i_lock);
573 inode->i_state |= I_FREEING;
574 inode_lru_list_del(inode);
575 spin_unlock(&inode->i_lock);
576 list_add(&inode->i_lru, &dispose);
578 spin_unlock(&inode_sb_list_lock);
580 dispose_list(&dispose);
585 static int can_unuse(struct inode *inode)
587 if (inode->i_state & ~I_REFERENCED)
589 if (inode_has_buffers(inode))
591 if (atomic_read(&inode->i_count))
593 if (inode->i_data.nrpages)
599 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
600 * This is called from the superblock shrinker function with a number of inodes
601 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
602 * then are freed outside inode_lock by dispose_list().
604 * Any inodes which are pinned purely because of attached pagecache have their
605 * pagecache removed. If the inode has metadata buffers attached to
606 * mapping->private_list then try to remove them.
608 * If the inode has the I_REFERENCED flag set, then it means that it has been
609 * used recently - the flag is set in iput_final(). When we encounter such an
610 * inode, clear the flag and move it to the back of the LRU so it gets another
611 * pass through the LRU before it gets reclaimed. This is necessary because of
612 * the fact we are doing lazy LRU updates to minimise lock contention so the
613 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
614 * with this flag set because they are the inodes that are out of order.
616 void prune_icache_sb(struct super_block *sb, int nr_to_scan)
620 unsigned long reap = 0;
622 spin_lock(&sb->s_inode_lru_lock);
623 for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
626 if (list_empty(&sb->s_inode_lru))
629 inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
632 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
633 * so use a trylock. If we fail to get the lock, just move the
634 * inode to the back of the list so we don't spin on it.
636 if (!spin_trylock(&inode->i_lock)) {
637 list_move_tail(&inode->i_lru, &sb->s_inode_lru);
642 * Referenced or dirty inodes are still in use. Give them
643 * another pass through the LRU as we canot reclaim them now.
645 if (atomic_read(&inode->i_count) ||
646 (inode->i_state & ~I_REFERENCED)) {
647 list_del_init(&inode->i_lru);
648 spin_unlock(&inode->i_lock);
649 sb->s_nr_inodes_unused--;
650 this_cpu_dec(nr_unused);
654 /* recently referenced inodes get one more pass */
655 if (inode->i_state & I_REFERENCED) {
656 inode->i_state &= ~I_REFERENCED;
657 list_move(&inode->i_lru, &sb->s_inode_lru);
658 spin_unlock(&inode->i_lock);
661 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
663 spin_unlock(&inode->i_lock);
664 spin_unlock(&sb->s_inode_lru_lock);
665 if (remove_inode_buffers(inode))
666 reap += invalidate_mapping_pages(&inode->i_data,
669 spin_lock(&sb->s_inode_lru_lock);
671 if (inode != list_entry(sb->s_inode_lru.next,
672 struct inode, i_lru))
673 continue; /* wrong inode or list_empty */
674 /* avoid lock inversions with trylock */
675 if (!spin_trylock(&inode->i_lock))
677 if (!can_unuse(inode)) {
678 spin_unlock(&inode->i_lock);
682 WARN_ON(inode->i_state & I_NEW);
683 inode->i_state |= I_FREEING;
684 spin_unlock(&inode->i_lock);
686 list_move(&inode->i_lru, &freeable);
687 sb->s_nr_inodes_unused--;
688 this_cpu_dec(nr_unused);
690 if (current_is_kswapd())
691 __count_vm_events(KSWAPD_INODESTEAL, reap);
693 __count_vm_events(PGINODESTEAL, reap);
694 spin_unlock(&sb->s_inode_lru_lock);
696 dispose_list(&freeable);
699 static void __wait_on_freeing_inode(struct inode *inode);
701 * Called with the inode lock held.
703 static struct inode *find_inode(struct super_block *sb,
704 struct hlist_head *head,
705 int (*test)(struct inode *, void *),
708 struct hlist_node *node;
709 struct inode *inode = NULL;
712 hlist_for_each_entry(inode, node, head, i_hash) {
713 spin_lock(&inode->i_lock);
714 if (inode->i_sb != sb) {
715 spin_unlock(&inode->i_lock);
718 if (!test(inode, data)) {
719 spin_unlock(&inode->i_lock);
722 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
723 __wait_on_freeing_inode(inode);
727 spin_unlock(&inode->i_lock);
734 * find_inode_fast is the fast path version of find_inode, see the comment at
735 * iget_locked for details.
737 static struct inode *find_inode_fast(struct super_block *sb,
738 struct hlist_head *head, unsigned long ino)
740 struct hlist_node *node;
741 struct inode *inode = NULL;
744 hlist_for_each_entry(inode, node, head, i_hash) {
745 spin_lock(&inode->i_lock);
746 if (inode->i_ino != ino) {
747 spin_unlock(&inode->i_lock);
750 if (inode->i_sb != sb) {
751 spin_unlock(&inode->i_lock);
754 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
755 __wait_on_freeing_inode(inode);
759 spin_unlock(&inode->i_lock);
766 * Each cpu owns a range of LAST_INO_BATCH numbers.
767 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
768 * to renew the exhausted range.
770 * This does not significantly increase overflow rate because every CPU can
771 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
772 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
773 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
774 * overflow rate by 2x, which does not seem too significant.
776 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
777 * error if st_ino won't fit in target struct field. Use 32bit counter
778 * here to attempt to avoid that.
780 #define LAST_INO_BATCH 1024
781 static DEFINE_PER_CPU(unsigned int, last_ino);
783 unsigned int get_next_ino(void)
785 unsigned int *p = &get_cpu_var(last_ino);
786 unsigned int res = *p;
789 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
790 static atomic_t shared_last_ino;
791 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
793 res = next - LAST_INO_BATCH;
798 put_cpu_var(last_ino);
801 EXPORT_SYMBOL(get_next_ino);
804 * new_inode_pseudo - obtain an inode
807 * Allocates a new inode for given superblock.
808 * Inode wont be chained in superblock s_inodes list
810 * - fs can't be unmount
811 * - quotas, fsnotify, writeback can't work
813 struct inode *new_inode_pseudo(struct super_block *sb)
815 struct inode *inode = alloc_inode(sb);
818 spin_lock(&inode->i_lock);
820 spin_unlock(&inode->i_lock);
821 INIT_LIST_HEAD(&inode->i_sb_list);
827 * new_inode - obtain an inode
830 * Allocates a new inode for given superblock. The default gfp_mask
831 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
832 * If HIGHMEM pages are unsuitable or it is known that pages allocated
833 * for the page cache are not reclaimable or migratable,
834 * mapping_set_gfp_mask() must be called with suitable flags on the
835 * newly created inode's mapping
838 struct inode *new_inode(struct super_block *sb)
842 spin_lock_prefetch(&inode_sb_list_lock);
844 inode = new_inode_pseudo(sb);
846 inode_sb_list_add(inode);
849 EXPORT_SYMBOL(new_inode);
851 #ifdef CONFIG_DEBUG_LOCK_ALLOC
852 void lockdep_annotate_inode_mutex_key(struct inode *inode)
854 if (S_ISDIR(inode->i_mode)) {
855 struct file_system_type *type = inode->i_sb->s_type;
857 /* Set new key only if filesystem hasn't already changed it */
858 if (!lockdep_match_class(&inode->i_mutex,
859 &type->i_mutex_key)) {
861 * ensure nobody is actually holding i_mutex
863 mutex_destroy(&inode->i_mutex);
864 mutex_init(&inode->i_mutex);
865 lockdep_set_class(&inode->i_mutex,
866 &type->i_mutex_dir_key);
870 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
874 * unlock_new_inode - clear the I_NEW state and wake up any waiters
875 * @inode: new inode to unlock
877 * Called when the inode is fully initialised to clear the new state of the
878 * inode and wake up anyone waiting for the inode to finish initialisation.
880 void unlock_new_inode(struct inode *inode)
882 lockdep_annotate_inode_mutex_key(inode);
883 spin_lock(&inode->i_lock);
884 WARN_ON(!(inode->i_state & I_NEW));
885 inode->i_state &= ~I_NEW;
886 wake_up_bit(&inode->i_state, __I_NEW);
887 spin_unlock(&inode->i_lock);
889 EXPORT_SYMBOL(unlock_new_inode);
892 * iget5_locked - obtain an inode from a mounted file system
893 * @sb: super block of file system
894 * @hashval: hash value (usually inode number) to get
895 * @test: callback used for comparisons between inodes
896 * @set: callback used to initialize a new struct inode
897 * @data: opaque data pointer to pass to @test and @set
899 * Search for the inode specified by @hashval and @data in the inode cache,
900 * and if present it is return it with an increased reference count. This is
901 * a generalized version of iget_locked() for file systems where the inode
902 * number is not sufficient for unique identification of an inode.
904 * If the inode is not in cache, allocate a new inode and return it locked,
905 * hashed, and with the I_NEW flag set. The file system gets to fill it in
906 * before unlocking it via unlock_new_inode().
908 * Note both @test and @set are called with the inode_hash_lock held, so can't
911 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
912 int (*test)(struct inode *, void *),
913 int (*set)(struct inode *, void *), void *data)
915 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
918 spin_lock(&inode_hash_lock);
919 inode = find_inode(sb, head, test, data);
920 spin_unlock(&inode_hash_lock);
923 wait_on_inode(inode);
927 inode = alloc_inode(sb);
931 spin_lock(&inode_hash_lock);
932 /* We released the lock, so.. */
933 old = find_inode(sb, head, test, data);
935 if (set(inode, data))
938 spin_lock(&inode->i_lock);
939 inode->i_state = I_NEW;
940 hlist_add_head(&inode->i_hash, head);
941 spin_unlock(&inode->i_lock);
942 inode_sb_list_add(inode);
943 spin_unlock(&inode_hash_lock);
945 /* Return the locked inode with I_NEW set, the
946 * caller is responsible for filling in the contents
952 * Uhhuh, somebody else created the same inode under
953 * us. Use the old inode instead of the one we just
956 spin_unlock(&inode_hash_lock);
957 destroy_inode(inode);
959 wait_on_inode(inode);
964 spin_unlock(&inode_hash_lock);
965 destroy_inode(inode);
968 EXPORT_SYMBOL(iget5_locked);
971 * iget_locked - obtain an inode from a mounted file system
972 * @sb: super block of file system
973 * @ino: inode number to get
975 * Search for the inode specified by @ino in the inode cache and if present
976 * return it with an increased reference count. This is for file systems
977 * where the inode number is sufficient for unique identification of an inode.
979 * If the inode is not in cache, allocate a new inode and return it locked,
980 * hashed, and with the I_NEW flag set. The file system gets to fill it in
981 * before unlocking it via unlock_new_inode().
983 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
985 struct hlist_head *head = inode_hashtable + hash(sb, ino);
988 spin_lock(&inode_hash_lock);
989 inode = find_inode_fast(sb, head, ino);
990 spin_unlock(&inode_hash_lock);
992 wait_on_inode(inode);
996 inode = alloc_inode(sb);
1000 spin_lock(&inode_hash_lock);
1001 /* We released the lock, so.. */
1002 old = find_inode_fast(sb, head, ino);
1005 spin_lock(&inode->i_lock);
1006 inode->i_state = I_NEW;
1007 hlist_add_head(&inode->i_hash, head);
1008 spin_unlock(&inode->i_lock);
1009 inode_sb_list_add(inode);
1010 spin_unlock(&inode_hash_lock);
1012 /* Return the locked inode with I_NEW set, the
1013 * caller is responsible for filling in the contents
1019 * Uhhuh, somebody else created the same inode under
1020 * us. Use the old inode instead of the one we just
1023 spin_unlock(&inode_hash_lock);
1024 destroy_inode(inode);
1026 wait_on_inode(inode);
1030 EXPORT_SYMBOL(iget_locked);
1033 * search the inode cache for a matching inode number.
1034 * If we find one, then the inode number we are trying to
1035 * allocate is not unique and so we should not use it.
1037 * Returns 1 if the inode number is unique, 0 if it is not.
1039 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1041 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1042 struct hlist_node *node;
1043 struct inode *inode;
1045 spin_lock(&inode_hash_lock);
1046 hlist_for_each_entry(inode, node, b, i_hash) {
1047 if (inode->i_ino == ino && inode->i_sb == sb) {
1048 spin_unlock(&inode_hash_lock);
1052 spin_unlock(&inode_hash_lock);
1058 * iunique - get a unique inode number
1060 * @max_reserved: highest reserved inode number
1062 * Obtain an inode number that is unique on the system for a given
1063 * superblock. This is used by file systems that have no natural
1064 * permanent inode numbering system. An inode number is returned that
1065 * is higher than the reserved limit but unique.
1068 * With a large number of inodes live on the file system this function
1069 * currently becomes quite slow.
1071 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1074 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1075 * error if st_ino won't fit in target struct field. Use 32bit counter
1076 * here to attempt to avoid that.
1078 static DEFINE_SPINLOCK(iunique_lock);
1079 static unsigned int counter;
1082 spin_lock(&iunique_lock);
1084 if (counter <= max_reserved)
1085 counter = max_reserved + 1;
1087 } while (!test_inode_iunique(sb, res));
1088 spin_unlock(&iunique_lock);
1092 EXPORT_SYMBOL(iunique);
1094 struct inode *igrab(struct inode *inode)
1096 spin_lock(&inode->i_lock);
1097 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1099 spin_unlock(&inode->i_lock);
1101 spin_unlock(&inode->i_lock);
1103 * Handle the case where s_op->clear_inode is not been
1104 * called yet, and somebody is calling igrab
1105 * while the inode is getting freed.
1111 EXPORT_SYMBOL(igrab);
1114 * ilookup5_nowait - search for an inode in the inode cache
1115 * @sb: super block of file system to search
1116 * @hashval: hash value (usually inode number) to search for
1117 * @test: callback used for comparisons between inodes
1118 * @data: opaque data pointer to pass to @test
1120 * Search for the inode specified by @hashval and @data in the inode cache.
1121 * If the inode is in the cache, the inode is returned with an incremented
1124 * Note: I_NEW is not waited upon so you have to be very careful what you do
1125 * with the returned inode. You probably should be using ilookup5() instead.
1127 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1129 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1130 int (*test)(struct inode *, void *), void *data)
1132 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1133 struct inode *inode;
1135 spin_lock(&inode_hash_lock);
1136 inode = find_inode(sb, head, test, data);
1137 spin_unlock(&inode_hash_lock);
1141 EXPORT_SYMBOL(ilookup5_nowait);
1144 * ilookup5 - search for an inode in the inode cache
1145 * @sb: super block of file system to search
1146 * @hashval: hash value (usually inode number) to search for
1147 * @test: callback used for comparisons between inodes
1148 * @data: opaque data pointer to pass to @test
1150 * Search for the inode specified by @hashval and @data in the inode cache,
1151 * and if the inode is in the cache, return the inode with an incremented
1152 * reference count. Waits on I_NEW before returning the inode.
1153 * returned with an incremented reference count.
1155 * This is a generalized version of ilookup() for file systems where the
1156 * inode number is not sufficient for unique identification of an inode.
1158 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1160 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1161 int (*test)(struct inode *, void *), void *data)
1163 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1166 wait_on_inode(inode);
1169 EXPORT_SYMBOL(ilookup5);
1172 * ilookup - search for an inode in the inode cache
1173 * @sb: super block of file system to search
1174 * @ino: inode number to search for
1176 * Search for the inode @ino in the inode cache, and if the inode is in the
1177 * cache, the inode is returned with an incremented reference count.
1179 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1181 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1182 struct inode *inode;
1184 spin_lock(&inode_hash_lock);
1185 inode = find_inode_fast(sb, head, ino);
1186 spin_unlock(&inode_hash_lock);
1189 wait_on_inode(inode);
1192 EXPORT_SYMBOL(ilookup);
1194 int insert_inode_locked(struct inode *inode)
1196 struct super_block *sb = inode->i_sb;
1197 ino_t ino = inode->i_ino;
1198 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1201 struct hlist_node *node;
1202 struct inode *old = NULL;
1203 spin_lock(&inode_hash_lock);
1204 hlist_for_each_entry(old, node, head, i_hash) {
1205 if (old->i_ino != ino)
1207 if (old->i_sb != sb)
1209 spin_lock(&old->i_lock);
1210 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1211 spin_unlock(&old->i_lock);
1216 if (likely(!node)) {
1217 spin_lock(&inode->i_lock);
1218 inode->i_state |= I_NEW;
1219 hlist_add_head(&inode->i_hash, head);
1220 spin_unlock(&inode->i_lock);
1221 spin_unlock(&inode_hash_lock);
1225 spin_unlock(&old->i_lock);
1226 spin_unlock(&inode_hash_lock);
1228 if (unlikely(!inode_unhashed(old))) {
1235 EXPORT_SYMBOL(insert_inode_locked);
1237 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1238 int (*test)(struct inode *, void *), void *data)
1240 struct super_block *sb = inode->i_sb;
1241 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1244 struct hlist_node *node;
1245 struct inode *old = NULL;
1247 spin_lock(&inode_hash_lock);
1248 hlist_for_each_entry(old, node, head, i_hash) {
1249 if (old->i_sb != sb)
1251 if (!test(old, data))
1253 spin_lock(&old->i_lock);
1254 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1255 spin_unlock(&old->i_lock);
1260 if (likely(!node)) {
1261 spin_lock(&inode->i_lock);
1262 inode->i_state |= I_NEW;
1263 hlist_add_head(&inode->i_hash, head);
1264 spin_unlock(&inode->i_lock);
1265 spin_unlock(&inode_hash_lock);
1269 spin_unlock(&old->i_lock);
1270 spin_unlock(&inode_hash_lock);
1272 if (unlikely(!inode_unhashed(old))) {
1279 EXPORT_SYMBOL(insert_inode_locked4);
1282 int generic_delete_inode(struct inode *inode)
1286 EXPORT_SYMBOL(generic_delete_inode);
1289 * Normal UNIX filesystem behaviour: delete the
1290 * inode when the usage count drops to zero, and
1293 int generic_drop_inode(struct inode *inode)
1295 return !inode->i_nlink || inode_unhashed(inode);
1297 EXPORT_SYMBOL_GPL(generic_drop_inode);
1300 * Called when we're dropping the last reference
1303 * Call the FS "drop_inode()" function, defaulting to
1304 * the legacy UNIX filesystem behaviour. If it tells
1305 * us to evict inode, do so. Otherwise, retain inode
1306 * in cache if fs is alive, sync and evict if fs is
1309 static void iput_final(struct inode *inode)
1311 struct super_block *sb = inode->i_sb;
1312 const struct super_operations *op = inode->i_sb->s_op;
1315 WARN_ON(inode->i_state & I_NEW);
1318 drop = op->drop_inode(inode);
1320 drop = generic_drop_inode(inode);
1322 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1323 inode->i_state |= I_REFERENCED;
1324 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1325 inode_lru_list_add(inode);
1326 spin_unlock(&inode->i_lock);
1331 inode->i_state |= I_WILL_FREE;
1332 spin_unlock(&inode->i_lock);
1333 write_inode_now(inode, 1);
1334 spin_lock(&inode->i_lock);
1335 WARN_ON(inode->i_state & I_NEW);
1336 inode->i_state &= ~I_WILL_FREE;
1339 inode->i_state |= I_FREEING;
1340 if (!list_empty(&inode->i_lru))
1341 inode_lru_list_del(inode);
1342 spin_unlock(&inode->i_lock);
1348 * iput - put an inode
1349 * @inode: inode to put
1351 * Puts an inode, dropping its usage count. If the inode use count hits
1352 * zero, the inode is then freed and may also be destroyed.
1354 * Consequently, iput() can sleep.
1356 void iput(struct inode *inode)
1359 BUG_ON(inode->i_state & I_CLEAR);
1361 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1365 EXPORT_SYMBOL(iput);
1368 * bmap - find a block number in a file
1369 * @inode: inode of file
1370 * @block: block to find
1372 * Returns the block number on the device holding the inode that
1373 * is the disk block number for the block of the file requested.
1374 * That is, asked for block 4 of inode 1 the function will return the
1375 * disk block relative to the disk start that holds that block of the
1378 sector_t bmap(struct inode *inode, sector_t block)
1381 if (inode->i_mapping->a_ops->bmap)
1382 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1385 EXPORT_SYMBOL(bmap);
1388 * With relative atime, only update atime if the previous atime is
1389 * earlier than either the ctime or mtime or if at least a day has
1390 * passed since the last atime update.
1392 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1393 struct timespec now)
1396 if (!(mnt->mnt_flags & MNT_RELATIME))
1399 * Is mtime younger than atime? If yes, update atime:
1401 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1404 * Is ctime younger than atime? If yes, update atime:
1406 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1410 * Is the previous atime value older than a day? If yes,
1413 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1416 * Good, we can skip the atime update:
1422 * touch_atime - update the access time
1423 * @mnt: mount the inode is accessed on
1424 * @dentry: dentry accessed
1426 * Update the accessed time on an inode and mark it for writeback.
1427 * This function automatically handles read only file systems and media,
1428 * as well as the "noatime" flag and inode specific "noatime" markers.
1430 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1432 struct inode *inode = dentry->d_inode;
1433 struct timespec now;
1435 if (inode->i_flags & S_NOATIME)
1437 if (IS_NOATIME(inode))
1439 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1442 if (mnt->mnt_flags & MNT_NOATIME)
1444 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1447 now = current_fs_time(inode->i_sb);
1449 if (!relatime_need_update(mnt, inode, now))
1452 if (timespec_equal(&inode->i_atime, &now))
1455 if (mnt_want_write(mnt))
1458 inode->i_atime = now;
1459 mark_inode_dirty_sync(inode);
1460 mnt_drop_write(mnt);
1462 EXPORT_SYMBOL(touch_atime);
1465 * file_update_time - update mtime and ctime time
1466 * @file: file accessed
1468 * Update the mtime and ctime members of an inode and mark the inode
1469 * for writeback. Note that this function is meant exclusively for
1470 * usage in the file write path of filesystems, and filesystems may
1471 * choose to explicitly ignore update via this function with the
1472 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1473 * timestamps are handled by the server.
1476 void file_update_time(struct file *file)
1478 struct inode *inode = file->f_path.dentry->d_inode;
1479 struct timespec now;
1480 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1482 /* First try to exhaust all avenues to not sync */
1483 if (IS_NOCMTIME(inode))
1486 now = current_fs_time(inode->i_sb);
1487 if (!timespec_equal(&inode->i_mtime, &now))
1490 if (!timespec_equal(&inode->i_ctime, &now))
1493 if (IS_I_VERSION(inode))
1494 sync_it |= S_VERSION;
1499 /* Finally allowed to write? Takes lock. */
1500 if (mnt_want_write_file(file))
1503 /* Only change inode inside the lock region */
1504 if (sync_it & S_VERSION)
1505 inode_inc_iversion(inode);
1506 if (sync_it & S_CTIME)
1507 inode->i_ctime = now;
1508 if (sync_it & S_MTIME)
1509 inode->i_mtime = now;
1510 mark_inode_dirty_sync(inode);
1511 mnt_drop_write(file->f_path.mnt);
1513 EXPORT_SYMBOL(file_update_time);
1515 int inode_needs_sync(struct inode *inode)
1519 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1523 EXPORT_SYMBOL(inode_needs_sync);
1525 int inode_wait(void *word)
1530 EXPORT_SYMBOL(inode_wait);
1533 * If we try to find an inode in the inode hash while it is being
1534 * deleted, we have to wait until the filesystem completes its
1535 * deletion before reporting that it isn't found. This function waits
1536 * until the deletion _might_ have completed. Callers are responsible
1537 * to recheck inode state.
1539 * It doesn't matter if I_NEW is not set initially, a call to
1540 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1543 static void __wait_on_freeing_inode(struct inode *inode)
1545 wait_queue_head_t *wq;
1546 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1547 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1548 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1549 spin_unlock(&inode->i_lock);
1550 spin_unlock(&inode_hash_lock);
1552 finish_wait(wq, &wait.wait);
1553 spin_lock(&inode_hash_lock);
1556 static __initdata unsigned long ihash_entries;
1557 static int __init set_ihash_entries(char *str)
1561 ihash_entries = simple_strtoul(str, &str, 0);
1564 __setup("ihash_entries=", set_ihash_entries);
1567 * Initialize the waitqueues and inode hash table.
1569 void __init inode_init_early(void)
1573 /* If hashes are distributed across NUMA nodes, defer
1574 * hash allocation until vmalloc space is available.
1580 alloc_large_system_hash("Inode-cache",
1581 sizeof(struct hlist_head),
1589 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1590 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1593 void __init inode_init(void)
1597 /* inode slab cache */
1598 inode_cachep = kmem_cache_create("inode_cache",
1599 sizeof(struct inode),
1601 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1605 /* Hash may have been set up in inode_init_early */
1610 alloc_large_system_hash("Inode-cache",
1611 sizeof(struct hlist_head),
1619 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1620 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1623 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1625 inode->i_mode = mode;
1626 if (S_ISCHR(mode)) {
1627 inode->i_fop = &def_chr_fops;
1628 inode->i_rdev = rdev;
1629 } else if (S_ISBLK(mode)) {
1630 inode->i_fop = &def_blk_fops;
1631 inode->i_rdev = rdev;
1632 } else if (S_ISFIFO(mode))
1633 inode->i_fop = &def_fifo_fops;
1634 else if (S_ISSOCK(mode))
1635 inode->i_fop = &bad_sock_fops;
1637 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1638 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1641 EXPORT_SYMBOL(init_special_inode);
1644 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1646 * @dir: Directory inode
1647 * @mode: mode of the new inode
1649 void inode_init_owner(struct inode *inode, const struct inode *dir,
1652 inode->i_uid = current_fsuid();
1653 if (dir && dir->i_mode & S_ISGID) {
1654 inode->i_gid = dir->i_gid;
1658 inode->i_gid = current_fsgid();
1659 inode->i_mode = mode;
1661 EXPORT_SYMBOL(inode_init_owner);
1664 * inode_owner_or_capable - check current task permissions to inode
1665 * @inode: inode being checked
1667 * Return true if current either has CAP_FOWNER to the inode, or
1670 bool inode_owner_or_capable(const struct inode *inode)
1672 struct user_namespace *ns = inode_userns(inode);
1674 if (current_user_ns() == ns && current_fsuid() == inode->i_uid)
1676 if (ns_capable(ns, CAP_FOWNER))
1680 EXPORT_SYMBOL(inode_owner_or_capable);