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);
68 EXPORT_SYMBOL(inode_sb_list_lock);
71 * Empty aops. Can be used for the cases where the user does not
72 * define any of the address_space operations.
74 const struct address_space_operations empty_aops = {
76 EXPORT_SYMBOL(empty_aops);
79 * Statistics gathering..
81 struct inodes_stat_t inodes_stat;
83 static DEFINE_PER_CPU(unsigned int, nr_inodes);
84 static DEFINE_PER_CPU(unsigned int, nr_unused);
86 static struct kmem_cache *inode_cachep __read_mostly;
88 static int get_nr_inodes(void)
92 for_each_possible_cpu(i)
93 sum += per_cpu(nr_inodes, i);
94 return sum < 0 ? 0 : sum;
97 static inline int get_nr_inodes_unused(void)
101 for_each_possible_cpu(i)
102 sum += per_cpu(nr_unused, i);
103 return sum < 0 ? 0 : sum;
106 int get_nr_dirty_inodes(void)
108 /* not actually dirty inodes, but a wild approximation */
109 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
110 return nr_dirty > 0 ? nr_dirty : 0;
114 * Handle nr_inode sysctl
117 int proc_nr_inodes(ctl_table *table, int write,
118 void __user *buffer, size_t *lenp, loff_t *ppos)
120 inodes_stat.nr_inodes = get_nr_inodes();
121 inodes_stat.nr_unused = get_nr_inodes_unused();
122 return proc_dointvec(table, write, buffer, lenp, ppos);
127 * inode_init_always - perform inode structure intialisation
128 * @sb: superblock inode belongs to
129 * @inode: inode to initialise
131 * These are initializations that need to be done on every inode
132 * allocation as the fields are not initialised by slab allocation.
134 int inode_init_always(struct super_block *sb, struct inode *inode)
136 static const struct inode_operations empty_iops;
137 static const struct file_operations empty_fops;
138 struct address_space *const mapping = &inode->i_data;
141 inode->i_blkbits = sb->s_blocksize_bits;
143 atomic_set(&inode->i_count, 1);
144 inode->i_op = &empty_iops;
145 inode->i_fop = &empty_fops;
146 inode->__i_nlink = 1;
147 inode->i_opflags = 0;
150 atomic_set(&inode->i_writecount, 0);
154 inode->i_generation = 0;
156 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
158 inode->i_pipe = NULL;
159 inode->i_bdev = NULL;
160 inode->i_cdev = NULL;
162 inode->dirtied_when = 0;
164 if (security_inode_alloc(inode))
166 spin_lock_init(&inode->i_lock);
167 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
169 mutex_init(&inode->i_mutex);
170 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
172 atomic_set(&inode->i_dio_count, 0);
174 mapping->a_ops = &empty_aops;
175 mapping->host = inode;
177 atomic_set(&mapping->i_mmap_writable, 0);
178 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
179 mapping->assoc_mapping = NULL;
180 mapping->backing_dev_info = &default_backing_dev_info;
181 mapping->writeback_index = 0;
184 * If the block_device provides a backing_dev_info for client
185 * inodes then use that. Otherwise the inode share the bdev's
189 struct backing_dev_info *bdi;
191 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
192 mapping->backing_dev_info = bdi;
194 inode->i_private = NULL;
195 inode->i_mapping = mapping;
196 #ifdef CONFIG_FS_POSIX_ACL
197 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
200 #ifdef CONFIG_FSNOTIFY
201 inode->i_fsnotify_mask = 0;
204 this_cpu_inc(nr_inodes);
210 EXPORT_SYMBOL(inode_init_always);
212 static struct inode *alloc_inode(struct super_block *sb)
216 if (sb->s_op->alloc_inode)
217 inode = sb->s_op->alloc_inode(sb);
219 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
224 if (unlikely(inode_init_always(sb, inode))) {
225 if (inode->i_sb->s_op->destroy_inode)
226 inode->i_sb->s_op->destroy_inode(inode);
228 kmem_cache_free(inode_cachep, inode);
235 void free_inode_nonrcu(struct inode *inode)
237 kmem_cache_free(inode_cachep, inode);
239 EXPORT_SYMBOL(free_inode_nonrcu);
241 void __destroy_inode(struct inode *inode)
243 BUG_ON(inode_has_buffers(inode));
244 security_inode_free(inode);
245 fsnotify_inode_delete(inode);
246 #ifdef CONFIG_FS_POSIX_ACL
247 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
248 posix_acl_release(inode->i_acl);
249 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
250 posix_acl_release(inode->i_default_acl);
252 this_cpu_dec(nr_inodes);
254 EXPORT_SYMBOL(__destroy_inode);
256 static void i_callback(struct rcu_head *head)
258 struct inode *inode = container_of(head, struct inode, i_rcu);
259 INIT_LIST_HEAD(&inode->i_dentry);
260 kmem_cache_free(inode_cachep, inode);
263 static void destroy_inode(struct inode *inode)
265 BUG_ON(!list_empty(&inode->i_lru));
266 __destroy_inode(inode);
267 if (inode->i_sb->s_op->destroy_inode)
268 inode->i_sb->s_op->destroy_inode(inode);
270 call_rcu(&inode->i_rcu, i_callback);
273 void address_space_init_once(struct address_space *mapping)
275 memset(mapping, 0, sizeof(*mapping));
276 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
277 spin_lock_init(&mapping->tree_lock);
278 mutex_init(&mapping->i_mmap_mutex);
279 INIT_LIST_HEAD(&mapping->private_list);
280 spin_lock_init(&mapping->private_lock);
281 INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
282 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
284 EXPORT_SYMBOL(address_space_init_once);
287 * These are initializations that only need to be done
288 * once, because the fields are idempotent across use
289 * of the inode, so let the slab aware of that.
291 void inode_init_once(struct inode *inode)
293 memset(inode, 0, sizeof(*inode));
294 INIT_HLIST_NODE(&inode->i_hash);
295 INIT_LIST_HEAD(&inode->i_dentry);
296 INIT_LIST_HEAD(&inode->i_devices);
297 INIT_LIST_HEAD(&inode->i_wb_list);
298 INIT_LIST_HEAD(&inode->i_lru);
299 address_space_init_once(&inode->i_data);
300 i_size_ordered_init(inode);
301 #ifdef CONFIG_FSNOTIFY
302 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
305 EXPORT_SYMBOL(inode_init_once);
307 static void init_once(void *foo)
309 struct inode *inode = (struct inode *) foo;
311 inode_init_once(inode);
315 * inode->i_lock must be held
317 void __iget(struct inode *inode)
319 atomic_inc(&inode->i_count);
323 * get additional reference to inode; caller must already hold one.
325 void ihold(struct inode *inode)
327 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
329 EXPORT_SYMBOL(ihold);
331 static void inode_lru_list_add(struct inode *inode)
333 spin_lock(&inode->i_sb->s_inode_lru_lock);
334 if (list_empty(&inode->i_lru)) {
335 list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
336 inode->i_sb->s_nr_inodes_unused++;
337 this_cpu_inc(nr_unused);
339 spin_unlock(&inode->i_sb->s_inode_lru_lock);
342 static void inode_lru_list_del(struct inode *inode)
344 spin_lock(&inode->i_sb->s_inode_lru_lock);
345 if (!list_empty(&inode->i_lru)) {
346 list_del_init(&inode->i_lru);
347 inode->i_sb->s_nr_inodes_unused--;
348 this_cpu_dec(nr_unused);
350 spin_unlock(&inode->i_sb->s_inode_lru_lock);
354 * inode_sb_list_add - add inode to the superblock list of inodes
355 * @inode: inode to add
357 void inode_sb_list_add(struct inode *inode)
359 spin_lock(&inode_sb_list_lock);
360 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
361 spin_unlock(&inode_sb_list_lock);
363 EXPORT_SYMBOL_GPL(inode_sb_list_add);
365 static inline void inode_sb_list_del(struct inode *inode)
367 if (!list_empty(&inode->i_sb_list)) {
368 spin_lock(&inode_sb_list_lock);
369 list_del_init(&inode->i_sb_list);
370 spin_unlock(&inode_sb_list_lock);
374 static unsigned long hash(struct super_block *sb, unsigned long hashval)
378 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
380 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
381 return tmp & i_hash_mask;
385 * __insert_inode_hash - hash an inode
386 * @inode: unhashed inode
387 * @hashval: unsigned long value used to locate this object in the
390 * Add an inode to the inode hash for this superblock.
392 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
394 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
396 spin_lock(&inode_hash_lock);
397 spin_lock(&inode->i_lock);
398 hlist_add_head(&inode->i_hash, b);
399 spin_unlock(&inode->i_lock);
400 spin_unlock(&inode_hash_lock);
402 EXPORT_SYMBOL(__insert_inode_hash);
405 * __remove_inode_hash - remove an inode from the hash
406 * @inode: inode to unhash
408 * Remove an inode from the superblock.
410 void __remove_inode_hash(struct inode *inode)
412 spin_lock(&inode_hash_lock);
413 spin_lock(&inode->i_lock);
414 hlist_del_init(&inode->i_hash);
415 spin_unlock(&inode->i_lock);
416 spin_unlock(&inode_hash_lock);
418 EXPORT_SYMBOL(__remove_inode_hash);
420 void end_writeback(struct inode *inode)
424 * We have to cycle tree_lock here because reclaim can be still in the
425 * process of removing the last page (in __delete_from_page_cache())
426 * and we must not free mapping under it.
428 spin_lock_irq(&inode->i_data.tree_lock);
429 BUG_ON(inode->i_data.nrpages);
430 spin_unlock_irq(&inode->i_data.tree_lock);
431 BUG_ON(!list_empty(&inode->i_data.private_list));
432 BUG_ON(!(inode->i_state & I_FREEING));
433 BUG_ON(inode->i_state & I_CLEAR);
434 inode_sync_wait(inode);
435 /* don't need i_lock here, no concurrent mods to i_state */
436 inode->i_state = I_FREEING | I_CLEAR;
438 EXPORT_SYMBOL(end_writeback);
441 * Free the inode passed in, removing it from the lists it is still connected
442 * to. We remove any pages still attached to the inode and wait for any IO that
443 * is still in progress before finally destroying the inode.
445 * An inode must already be marked I_FREEING so that we avoid the inode being
446 * moved back onto lists if we race with other code that manipulates the lists
447 * (e.g. writeback_single_inode). The caller is responsible for setting this.
449 * An inode must already be removed from the LRU list before being evicted from
450 * the cache. This should occur atomically with setting the I_FREEING state
451 * flag, so no inodes here should ever be on the LRU when being evicted.
453 static void evict(struct inode *inode)
455 const struct super_operations *op = inode->i_sb->s_op;
457 BUG_ON(!(inode->i_state & I_FREEING));
458 BUG_ON(!list_empty(&inode->i_lru));
460 if (!list_empty(&inode->i_wb_list))
461 inode_wb_list_del(inode);
463 inode_sb_list_del(inode);
465 if (op->evict_inode) {
466 op->evict_inode(inode);
468 if (inode->i_data.nrpages)
469 truncate_inode_pages(&inode->i_data, 0);
470 end_writeback(inode);
472 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
474 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
477 remove_inode_hash(inode);
479 spin_lock(&inode->i_lock);
480 wake_up_bit(&inode->i_state, __I_NEW);
481 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
482 spin_unlock(&inode->i_lock);
484 destroy_inode(inode);
488 * dispose_list - dispose of the contents of a local list
489 * @head: the head of the list to free
491 * Dispose-list gets a local list with local inodes in it, so it doesn't
492 * need to worry about list corruption and SMP locks.
494 static void dispose_list(struct list_head *head)
496 while (!list_empty(head)) {
499 inode = list_first_entry(head, struct inode, i_lru);
500 list_del_init(&inode->i_lru);
507 * evict_inodes - evict all evictable inodes for a superblock
508 * @sb: superblock to operate on
510 * Make sure that no inodes with zero refcount are retained. This is
511 * called by superblock shutdown after having MS_ACTIVE flag removed,
512 * so any inode reaching zero refcount during or after that call will
513 * be immediately evicted.
515 void evict_inodes(struct super_block *sb)
517 struct inode *inode, *next;
520 spin_lock(&inode_sb_list_lock);
521 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
522 if (atomic_read(&inode->i_count))
525 spin_lock(&inode->i_lock);
526 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
527 spin_unlock(&inode->i_lock);
531 inode->i_state |= I_FREEING;
532 inode_lru_list_del(inode);
533 spin_unlock(&inode->i_lock);
534 list_add(&inode->i_lru, &dispose);
536 spin_unlock(&inode_sb_list_lock);
538 dispose_list(&dispose);
542 * invalidate_inodes - attempt to free all inodes on a superblock
543 * @sb: superblock to operate on
544 * @kill_dirty: flag to guide handling of dirty inodes
546 * Attempts to free all inodes for a given superblock. If there were any
547 * busy inodes return a non-zero value, else zero.
548 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
551 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
554 struct inode *inode, *next;
557 spin_lock(&inode_sb_list_lock);
558 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
559 spin_lock(&inode->i_lock);
560 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
561 spin_unlock(&inode->i_lock);
564 if (inode->i_state & I_DIRTY && !kill_dirty) {
565 spin_unlock(&inode->i_lock);
569 if (atomic_read(&inode->i_count)) {
570 spin_unlock(&inode->i_lock);
575 inode->i_state |= I_FREEING;
576 inode_lru_list_del(inode);
577 spin_unlock(&inode->i_lock);
578 list_add(&inode->i_lru, &dispose);
580 spin_unlock(&inode_sb_list_lock);
582 dispose_list(&dispose);
587 static int can_unuse(struct inode *inode)
589 if (inode->i_state & ~I_REFERENCED)
591 if (inode_has_buffers(inode))
593 if (atomic_read(&inode->i_count))
595 if (inode->i_data.nrpages)
601 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
602 * This is called from the superblock shrinker function with a number of inodes
603 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
604 * then are freed outside inode_lock by dispose_list().
606 * Any inodes which are pinned purely because of attached pagecache have their
607 * pagecache removed. If the inode has metadata buffers attached to
608 * mapping->private_list then try to remove them.
610 * If the inode has the I_REFERENCED flag set, then it means that it has been
611 * used recently - the flag is set in iput_final(). When we encounter such an
612 * inode, clear the flag and move it to the back of the LRU so it gets another
613 * pass through the LRU before it gets reclaimed. This is necessary because of
614 * the fact we are doing lazy LRU updates to minimise lock contention so the
615 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
616 * with this flag set because they are the inodes that are out of order.
618 void prune_icache_sb(struct super_block *sb, int nr_to_scan)
622 unsigned long reap = 0;
624 spin_lock(&sb->s_inode_lru_lock);
625 for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
628 if (list_empty(&sb->s_inode_lru))
631 inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
634 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
635 * so use a trylock. If we fail to get the lock, just move the
636 * inode to the back of the list so we don't spin on it.
638 if (!spin_trylock(&inode->i_lock)) {
639 list_move(&inode->i_lru, &sb->s_inode_lru);
644 * Referenced or dirty inodes are still in use. Give them
645 * another pass through the LRU as we canot reclaim them now.
647 if (atomic_read(&inode->i_count) ||
648 (inode->i_state & ~I_REFERENCED)) {
649 list_del_init(&inode->i_lru);
650 spin_unlock(&inode->i_lock);
651 sb->s_nr_inodes_unused--;
652 this_cpu_dec(nr_unused);
656 /* recently referenced inodes get one more pass */
657 if (inode->i_state & I_REFERENCED) {
658 inode->i_state &= ~I_REFERENCED;
659 list_move(&inode->i_lru, &sb->s_inode_lru);
660 spin_unlock(&inode->i_lock);
663 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
665 spin_unlock(&inode->i_lock);
666 spin_unlock(&sb->s_inode_lru_lock);
667 if (remove_inode_buffers(inode))
668 reap += invalidate_mapping_pages(&inode->i_data,
671 spin_lock(&sb->s_inode_lru_lock);
673 if (inode != list_entry(sb->s_inode_lru.next,
674 struct inode, i_lru))
675 continue; /* wrong inode or list_empty */
676 /* avoid lock inversions with trylock */
677 if (!spin_trylock(&inode->i_lock))
679 if (!can_unuse(inode)) {
680 spin_unlock(&inode->i_lock);
684 WARN_ON(inode->i_state & I_NEW);
685 inode->i_state |= I_FREEING;
686 spin_unlock(&inode->i_lock);
688 list_move(&inode->i_lru, &freeable);
689 sb->s_nr_inodes_unused--;
690 this_cpu_dec(nr_unused);
692 if (current_is_kswapd())
693 __count_vm_events(KSWAPD_INODESTEAL, reap);
695 __count_vm_events(PGINODESTEAL, reap);
696 spin_unlock(&sb->s_inode_lru_lock);
698 dispose_list(&freeable);
701 static void __wait_on_freeing_inode(struct inode *inode);
703 * Called with the inode lock held.
705 static struct inode *find_inode(struct super_block *sb,
706 struct hlist_head *head,
707 int (*test)(struct inode *, void *),
710 struct hlist_node *node;
711 struct inode *inode = NULL;
714 hlist_for_each_entry(inode, node, head, i_hash) {
715 spin_lock(&inode->i_lock);
716 if (inode->i_sb != sb) {
717 spin_unlock(&inode->i_lock);
720 if (!test(inode, data)) {
721 spin_unlock(&inode->i_lock);
724 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
725 __wait_on_freeing_inode(inode);
729 spin_unlock(&inode->i_lock);
736 * find_inode_fast is the fast path version of find_inode, see the comment at
737 * iget_locked for details.
739 static struct inode *find_inode_fast(struct super_block *sb,
740 struct hlist_head *head, unsigned long ino)
742 struct hlist_node *node;
743 struct inode *inode = NULL;
746 hlist_for_each_entry(inode, node, head, i_hash) {
747 spin_lock(&inode->i_lock);
748 if (inode->i_ino != ino) {
749 spin_unlock(&inode->i_lock);
752 if (inode->i_sb != sb) {
753 spin_unlock(&inode->i_lock);
756 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
757 __wait_on_freeing_inode(inode);
761 spin_unlock(&inode->i_lock);
768 * Each cpu owns a range of LAST_INO_BATCH numbers.
769 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
770 * to renew the exhausted range.
772 * This does not significantly increase overflow rate because every CPU can
773 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
774 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
775 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
776 * overflow rate by 2x, which does not seem too significant.
778 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
779 * error if st_ino won't fit in target struct field. Use 32bit counter
780 * here to attempt to avoid that.
782 #define LAST_INO_BATCH 1024
783 static DEFINE_PER_CPU(unsigned int, last_ino);
785 unsigned int get_next_ino(void)
787 unsigned int *p = &get_cpu_var(last_ino);
788 unsigned int res = *p;
791 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
792 static atomic_t shared_last_ino;
793 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
795 res = next - LAST_INO_BATCH;
800 put_cpu_var(last_ino);
803 EXPORT_SYMBOL(get_next_ino);
806 * new_inode_pseudo - obtain an inode
809 * Allocates a new inode for given superblock.
810 * Inode wont be chained in superblock s_inodes list
812 * - fs can't be unmount
813 * - quotas, fsnotify, writeback can't work
815 struct inode *new_inode_pseudo(struct super_block *sb)
817 struct inode *inode = alloc_inode(sb);
820 spin_lock(&inode->i_lock);
822 spin_unlock(&inode->i_lock);
823 INIT_LIST_HEAD(&inode->i_sb_list);
829 * new_inode - obtain an inode
832 * Allocates a new inode for given superblock. The default gfp_mask
833 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
834 * If HIGHMEM pages are unsuitable or it is known that pages allocated
835 * for the page cache are not reclaimable or migratable,
836 * mapping_set_gfp_mask() must be called with suitable flags on the
837 * newly created inode's mapping
840 struct inode *new_inode(struct super_block *sb)
844 spin_lock_prefetch(&inode_sb_list_lock);
846 inode = new_inode_pseudo(sb);
848 inode_sb_list_add(inode);
851 EXPORT_SYMBOL(new_inode);
853 #ifdef CONFIG_DEBUG_LOCK_ALLOC
854 void lockdep_annotate_inode_mutex_key(struct inode *inode)
856 if (S_ISDIR(inode->i_mode)) {
857 struct file_system_type *type = inode->i_sb->s_type;
859 /* Set new key only if filesystem hasn't already changed it */
860 if (!lockdep_match_class(&inode->i_mutex,
861 &type->i_mutex_key)) {
863 * ensure nobody is actually holding i_mutex
865 mutex_destroy(&inode->i_mutex);
866 mutex_init(&inode->i_mutex);
867 lockdep_set_class(&inode->i_mutex,
868 &type->i_mutex_dir_key);
872 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
876 * unlock_new_inode - clear the I_NEW state and wake up any waiters
877 * @inode: new inode to unlock
879 * Called when the inode is fully initialised to clear the new state of the
880 * inode and wake up anyone waiting for the inode to finish initialisation.
882 void unlock_new_inode(struct inode *inode)
884 lockdep_annotate_inode_mutex_key(inode);
885 spin_lock(&inode->i_lock);
886 WARN_ON(!(inode->i_state & I_NEW));
887 inode->i_state &= ~I_NEW;
888 wake_up_bit(&inode->i_state, __I_NEW);
889 spin_unlock(&inode->i_lock);
891 EXPORT_SYMBOL(unlock_new_inode);
894 * iget5_locked - obtain an inode from a mounted file system
895 * @sb: super block of file system
896 * @hashval: hash value (usually inode number) to get
897 * @test: callback used for comparisons between inodes
898 * @set: callback used to initialize a new struct inode
899 * @data: opaque data pointer to pass to @test and @set
901 * Search for the inode specified by @hashval and @data in the inode cache,
902 * and if present it is return it with an increased reference count. This is
903 * a generalized version of iget_locked() for file systems where the inode
904 * number is not sufficient for unique identification of an inode.
906 * If the inode is not in cache, allocate a new inode and return it locked,
907 * hashed, and with the I_NEW flag set. The file system gets to fill it in
908 * before unlocking it via unlock_new_inode().
910 * Note both @test and @set are called with the inode_hash_lock held, so can't
913 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
914 int (*test)(struct inode *, void *),
915 int (*set)(struct inode *, void *), void *data)
917 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
920 spin_lock(&inode_hash_lock);
921 inode = find_inode(sb, head, test, data);
922 spin_unlock(&inode_hash_lock);
925 wait_on_inode(inode);
929 inode = alloc_inode(sb);
933 spin_lock(&inode_hash_lock);
934 /* We released the lock, so.. */
935 old = find_inode(sb, head, test, data);
937 if (set(inode, data))
940 spin_lock(&inode->i_lock);
941 inode->i_state = I_NEW;
942 hlist_add_head(&inode->i_hash, head);
943 spin_unlock(&inode->i_lock);
944 inode_sb_list_add(inode);
945 spin_unlock(&inode_hash_lock);
947 /* Return the locked inode with I_NEW set, the
948 * caller is responsible for filling in the contents
954 * Uhhuh, somebody else created the same inode under
955 * us. Use the old inode instead of the one we just
958 spin_unlock(&inode_hash_lock);
959 destroy_inode(inode);
961 wait_on_inode(inode);
966 spin_unlock(&inode_hash_lock);
967 destroy_inode(inode);
970 EXPORT_SYMBOL(iget5_locked);
973 * iget_locked - obtain an inode from a mounted file system
974 * @sb: super block of file system
975 * @ino: inode number to get
977 * Search for the inode specified by @ino in the inode cache and if present
978 * return it with an increased reference count. This is for file systems
979 * where the inode number is sufficient for unique identification of an inode.
981 * If the inode is not in cache, allocate a new inode and return it locked,
982 * hashed, and with the I_NEW flag set. The file system gets to fill it in
983 * before unlocking it via unlock_new_inode().
985 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
987 struct hlist_head *head = inode_hashtable + hash(sb, ino);
990 spin_lock(&inode_hash_lock);
991 inode = find_inode_fast(sb, head, ino);
992 spin_unlock(&inode_hash_lock);
994 wait_on_inode(inode);
998 inode = alloc_inode(sb);
1002 spin_lock(&inode_hash_lock);
1003 /* We released the lock, so.. */
1004 old = find_inode_fast(sb, head, ino);
1007 spin_lock(&inode->i_lock);
1008 inode->i_state = I_NEW;
1009 hlist_add_head(&inode->i_hash, head);
1010 spin_unlock(&inode->i_lock);
1011 inode_sb_list_add(inode);
1012 spin_unlock(&inode_hash_lock);
1014 /* Return the locked inode with I_NEW set, the
1015 * caller is responsible for filling in the contents
1021 * Uhhuh, somebody else created the same inode under
1022 * us. Use the old inode instead of the one we just
1025 spin_unlock(&inode_hash_lock);
1026 destroy_inode(inode);
1028 wait_on_inode(inode);
1032 EXPORT_SYMBOL(iget_locked);
1035 * search the inode cache for a matching inode number.
1036 * If we find one, then the inode number we are trying to
1037 * allocate is not unique and so we should not use it.
1039 * Returns 1 if the inode number is unique, 0 if it is not.
1041 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1043 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1044 struct hlist_node *node;
1045 struct inode *inode;
1047 spin_lock(&inode_hash_lock);
1048 hlist_for_each_entry(inode, node, b, i_hash) {
1049 if (inode->i_ino == ino && inode->i_sb == sb) {
1050 spin_unlock(&inode_hash_lock);
1054 spin_unlock(&inode_hash_lock);
1060 * iunique - get a unique inode number
1062 * @max_reserved: highest reserved inode number
1064 * Obtain an inode number that is unique on the system for a given
1065 * superblock. This is used by file systems that have no natural
1066 * permanent inode numbering system. An inode number is returned that
1067 * is higher than the reserved limit but unique.
1070 * With a large number of inodes live on the file system this function
1071 * currently becomes quite slow.
1073 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1076 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1077 * error if st_ino won't fit in target struct field. Use 32bit counter
1078 * here to attempt to avoid that.
1080 static DEFINE_SPINLOCK(iunique_lock);
1081 static unsigned int counter;
1084 spin_lock(&iunique_lock);
1086 if (counter <= max_reserved)
1087 counter = max_reserved + 1;
1089 } while (!test_inode_iunique(sb, res));
1090 spin_unlock(&iunique_lock);
1094 EXPORT_SYMBOL(iunique);
1096 struct inode *igrab(struct inode *inode)
1098 spin_lock(&inode->i_lock);
1099 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1101 spin_unlock(&inode->i_lock);
1103 spin_unlock(&inode->i_lock);
1105 * Handle the case where s_op->clear_inode is not been
1106 * called yet, and somebody is calling igrab
1107 * while the inode is getting freed.
1113 EXPORT_SYMBOL(igrab);
1116 * ilookup5_nowait - search for an inode in the inode cache
1117 * @sb: super block of file system to search
1118 * @hashval: hash value (usually inode number) to search for
1119 * @test: callback used for comparisons between inodes
1120 * @data: opaque data pointer to pass to @test
1122 * Search for the inode specified by @hashval and @data in the inode cache.
1123 * If the inode is in the cache, the inode is returned with an incremented
1126 * Note: I_NEW is not waited upon so you have to be very careful what you do
1127 * with the returned inode. You probably should be using ilookup5() instead.
1129 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1131 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1132 int (*test)(struct inode *, void *), void *data)
1134 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1135 struct inode *inode;
1137 spin_lock(&inode_hash_lock);
1138 inode = find_inode(sb, head, test, data);
1139 spin_unlock(&inode_hash_lock);
1143 EXPORT_SYMBOL(ilookup5_nowait);
1146 * ilookup5 - search for an inode in the inode cache
1147 * @sb: super block of file system to search
1148 * @hashval: hash value (usually inode number) to search for
1149 * @test: callback used for comparisons between inodes
1150 * @data: opaque data pointer to pass to @test
1152 * Search for the inode specified by @hashval and @data in the inode cache,
1153 * and if the inode is in the cache, return the inode with an incremented
1154 * reference count. Waits on I_NEW before returning the inode.
1155 * returned with an incremented reference count.
1157 * This is a generalized version of ilookup() for file systems where the
1158 * inode number is not sufficient for unique identification of an inode.
1160 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1162 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1163 int (*test)(struct inode *, void *), void *data)
1165 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1168 wait_on_inode(inode);
1171 EXPORT_SYMBOL(ilookup5);
1174 * ilookup - search for an inode in the inode cache
1175 * @sb: super block of file system to search
1176 * @ino: inode number to search for
1178 * Search for the inode @ino in the inode cache, and if the inode is in the
1179 * cache, the inode is returned with an incremented reference count.
1181 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1183 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1184 struct inode *inode;
1186 spin_lock(&inode_hash_lock);
1187 inode = find_inode_fast(sb, head, ino);
1188 spin_unlock(&inode_hash_lock);
1191 wait_on_inode(inode);
1194 EXPORT_SYMBOL(ilookup);
1196 int insert_inode_locked(struct inode *inode)
1198 struct super_block *sb = inode->i_sb;
1199 ino_t ino = inode->i_ino;
1200 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1203 struct hlist_node *node;
1204 struct inode *old = NULL;
1205 spin_lock(&inode_hash_lock);
1206 hlist_for_each_entry(old, node, head, i_hash) {
1207 if (old->i_ino != ino)
1209 if (old->i_sb != sb)
1211 spin_lock(&old->i_lock);
1212 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1213 spin_unlock(&old->i_lock);
1218 if (likely(!node)) {
1219 spin_lock(&inode->i_lock);
1220 inode->i_state |= I_NEW;
1221 hlist_add_head(&inode->i_hash, head);
1222 spin_unlock(&inode->i_lock);
1223 spin_unlock(&inode_hash_lock);
1227 spin_unlock(&old->i_lock);
1228 spin_unlock(&inode_hash_lock);
1230 if (unlikely(!inode_unhashed(old))) {
1237 EXPORT_SYMBOL(insert_inode_locked);
1239 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1240 int (*test)(struct inode *, void *), void *data)
1242 struct super_block *sb = inode->i_sb;
1243 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1246 struct hlist_node *node;
1247 struct inode *old = NULL;
1249 spin_lock(&inode_hash_lock);
1250 hlist_for_each_entry(old, node, head, i_hash) {
1251 if (old->i_sb != sb)
1253 if (!test(old, data))
1255 spin_lock(&old->i_lock);
1256 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1257 spin_unlock(&old->i_lock);
1262 if (likely(!node)) {
1263 spin_lock(&inode->i_lock);
1264 inode->i_state |= I_NEW;
1265 hlist_add_head(&inode->i_hash, head);
1266 spin_unlock(&inode->i_lock);
1267 spin_unlock(&inode_hash_lock);
1271 spin_unlock(&old->i_lock);
1272 spin_unlock(&inode_hash_lock);
1274 if (unlikely(!inode_unhashed(old))) {
1281 EXPORT_SYMBOL(insert_inode_locked4);
1284 int generic_delete_inode(struct inode *inode)
1288 EXPORT_SYMBOL(generic_delete_inode);
1291 * Normal UNIX filesystem behaviour: delete the
1292 * inode when the usage count drops to zero, and
1295 int generic_drop_inode(struct inode *inode)
1297 return !inode->i_nlink || inode_unhashed(inode);
1299 EXPORT_SYMBOL_GPL(generic_drop_inode);
1302 * Called when we're dropping the last reference
1305 * Call the FS "drop_inode()" function, defaulting to
1306 * the legacy UNIX filesystem behaviour. If it tells
1307 * us to evict inode, do so. Otherwise, retain inode
1308 * in cache if fs is alive, sync and evict if fs is
1311 static void iput_final(struct inode *inode)
1313 struct super_block *sb = inode->i_sb;
1314 const struct super_operations *op = inode->i_sb->s_op;
1317 WARN_ON(inode->i_state & I_NEW);
1320 drop = op->drop_inode(inode);
1322 drop = generic_drop_inode(inode);
1324 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1325 inode->i_state |= I_REFERENCED;
1326 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1327 inode_lru_list_add(inode);
1328 spin_unlock(&inode->i_lock);
1333 inode->i_state |= I_WILL_FREE;
1334 spin_unlock(&inode->i_lock);
1335 write_inode_now(inode, 1);
1336 spin_lock(&inode->i_lock);
1337 WARN_ON(inode->i_state & I_NEW);
1338 inode->i_state &= ~I_WILL_FREE;
1341 inode->i_state |= I_FREEING;
1342 if (!list_empty(&inode->i_lru))
1343 inode_lru_list_del(inode);
1344 spin_unlock(&inode->i_lock);
1350 * iput - put an inode
1351 * @inode: inode to put
1353 * Puts an inode, dropping its usage count. If the inode use count hits
1354 * zero, the inode is then freed and may also be destroyed.
1356 * Consequently, iput() can sleep.
1358 void iput(struct inode *inode)
1361 BUG_ON(inode->i_state & I_CLEAR);
1363 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1367 EXPORT_SYMBOL(iput);
1370 * bmap - find a block number in a file
1371 * @inode: inode of file
1372 * @block: block to find
1374 * Returns the block number on the device holding the inode that
1375 * is the disk block number for the block of the file requested.
1376 * That is, asked for block 4 of inode 1 the function will return the
1377 * disk block relative to the disk start that holds that block of the
1380 sector_t bmap(struct inode *inode, sector_t block)
1383 if (inode->i_mapping->a_ops->bmap)
1384 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1387 EXPORT_SYMBOL(bmap);
1390 * With relative atime, only update atime if the previous atime is
1391 * earlier than either the ctime or mtime or if at least a day has
1392 * passed since the last atime update.
1394 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1395 struct timespec now)
1398 if (!(mnt->mnt_flags & MNT_RELATIME))
1401 * Is mtime younger than atime? If yes, update atime:
1403 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1406 * Is ctime younger than atime? If yes, update atime:
1408 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1412 * Is the previous atime value older than a day? If yes,
1415 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1418 * Good, we can skip the atime update:
1424 * touch_atime - update the access time
1425 * @mnt: mount the inode is accessed on
1426 * @dentry: dentry accessed
1428 * Update the accessed time on an inode and mark it for writeback.
1429 * This function automatically handles read only file systems and media,
1430 * as well as the "noatime" flag and inode specific "noatime" markers.
1432 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1434 struct inode *inode = dentry->d_inode;
1435 struct timespec now;
1437 if (inode->i_flags & S_NOATIME)
1439 if (IS_NOATIME(inode))
1441 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1444 if (mnt->mnt_flags & MNT_NOATIME)
1446 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1449 now = current_fs_time(inode->i_sb);
1451 if (!relatime_need_update(mnt, inode, now))
1454 if (timespec_equal(&inode->i_atime, &now))
1457 if (mnt_want_write(mnt))
1460 inode->i_atime = now;
1461 mark_inode_dirty_sync(inode);
1462 mnt_drop_write(mnt);
1464 EXPORT_SYMBOL(touch_atime);
1467 * file_update_time - update mtime and ctime time
1468 * @file: file accessed
1470 * Update the mtime and ctime members of an inode and mark the inode
1471 * for writeback. Note that this function is meant exclusively for
1472 * usage in the file write path of filesystems, and filesystems may
1473 * choose to explicitly ignore update via this function with the
1474 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1475 * timestamps are handled by the server.
1478 void file_update_time(struct file *file)
1480 struct inode *inode = file->f_path.dentry->d_inode;
1481 struct timespec now;
1482 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1484 /* First try to exhaust all avenues to not sync */
1485 if (IS_NOCMTIME(inode))
1488 now = current_fs_time(inode->i_sb);
1489 if (!timespec_equal(&inode->i_mtime, &now))
1492 if (!timespec_equal(&inode->i_ctime, &now))
1495 if (IS_I_VERSION(inode))
1496 sync_it |= S_VERSION;
1501 /* Finally allowed to write? Takes lock. */
1502 if (mnt_want_write_file(file))
1505 /* Only change inode inside the lock region */
1506 if (sync_it & S_VERSION)
1507 inode_inc_iversion(inode);
1508 if (sync_it & S_CTIME)
1509 inode->i_ctime = now;
1510 if (sync_it & S_MTIME)
1511 inode->i_mtime = now;
1512 mark_inode_dirty_sync(inode);
1513 mnt_drop_write(file->f_path.mnt);
1515 EXPORT_SYMBOL(file_update_time);
1517 int inode_needs_sync(struct inode *inode)
1521 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1525 EXPORT_SYMBOL(inode_needs_sync);
1527 int inode_wait(void *word)
1532 EXPORT_SYMBOL(inode_wait);
1535 * If we try to find an inode in the inode hash while it is being
1536 * deleted, we have to wait until the filesystem completes its
1537 * deletion before reporting that it isn't found. This function waits
1538 * until the deletion _might_ have completed. Callers are responsible
1539 * to recheck inode state.
1541 * It doesn't matter if I_NEW is not set initially, a call to
1542 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1545 static void __wait_on_freeing_inode(struct inode *inode)
1547 wait_queue_head_t *wq;
1548 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1549 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1550 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1551 spin_unlock(&inode->i_lock);
1552 spin_unlock(&inode_hash_lock);
1554 finish_wait(wq, &wait.wait);
1555 spin_lock(&inode_hash_lock);
1558 static __initdata unsigned long ihash_entries;
1559 static int __init set_ihash_entries(char *str)
1563 ihash_entries = simple_strtoul(str, &str, 0);
1566 __setup("ihash_entries=", set_ihash_entries);
1569 * Initialize the waitqueues and inode hash table.
1571 void __init inode_init_early(void)
1575 /* If hashes are distributed across NUMA nodes, defer
1576 * hash allocation until vmalloc space is available.
1582 alloc_large_system_hash("Inode-cache",
1583 sizeof(struct hlist_head),
1591 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1592 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1595 void __init inode_init(void)
1599 /* inode slab cache */
1600 inode_cachep = kmem_cache_create("inode_cache",
1601 sizeof(struct inode),
1603 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1607 /* Hash may have been set up in inode_init_early */
1612 alloc_large_system_hash("Inode-cache",
1613 sizeof(struct hlist_head),
1621 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1622 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1625 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1627 inode->i_mode = mode;
1628 if (S_ISCHR(mode)) {
1629 inode->i_fop = &def_chr_fops;
1630 inode->i_rdev = rdev;
1631 } else if (S_ISBLK(mode)) {
1632 inode->i_fop = &def_blk_fops;
1633 inode->i_rdev = rdev;
1634 } else if (S_ISFIFO(mode))
1635 inode->i_fop = &def_fifo_fops;
1636 else if (S_ISSOCK(mode))
1637 inode->i_fop = &bad_sock_fops;
1639 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1640 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1643 EXPORT_SYMBOL(init_special_inode);
1646 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1648 * @dir: Directory inode
1649 * @mode: mode of the new inode
1651 void inode_init_owner(struct inode *inode, const struct inode *dir,
1654 inode->i_uid = current_fsuid();
1655 if (dir && dir->i_mode & S_ISGID) {
1656 inode->i_gid = dir->i_gid;
1660 inode->i_gid = current_fsgid();
1661 inode->i_mode = mode;
1663 EXPORT_SYMBOL(inode_init_owner);
1666 * inode_owner_or_capable - check current task permissions to inode
1667 * @inode: inode being checked
1669 * Return true if current either has CAP_FOWNER to the inode, or
1672 bool inode_owner_or_capable(const struct inode *inode)
1674 struct user_namespace *ns = inode_userns(inode);
1676 if (current_user_ns() == ns && current_fsuid() == inode->i_uid)
1678 if (ns_capable(ns, CAP_FOWNER))
1682 EXPORT_SYMBOL(inode_owner_or_capable);