2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
5 #include <linux/export.h>
8 #include <linux/backing-dev.h>
9 #include <linux/hash.h>
10 #include <linux/swap.h>
11 #include <linux/security.h>
12 #include <linux/cdev.h>
13 #include <linux/bootmem.h>
14 #include <linux/fsnotify.h>
15 #include <linux/mount.h>
16 #include <linux/posix_acl.h>
17 #include <linux/prefetch.h>
18 #include <linux/buffer_head.h> /* for inode_has_buffers */
19 #include <linux/ratelimit.h>
20 #include <linux/list_lru.h>
24 * Inode locking rules:
26 * inode->i_lock protects:
27 * inode->i_state, inode->i_hash, __iget()
28 * Inode LRU list locks protect:
29 * inode->i_sb->s_inode_lru, inode->i_lru
30 * inode_sb_list_lock protects:
31 * sb->s_inodes, inode->i_sb_list
32 * bdi->wb.list_lock protects:
33 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
34 * inode_hash_lock protects:
35 * inode_hashtable, inode->i_hash
41 * Inode LRU list locks
54 static unsigned int i_hash_mask __read_mostly;
55 static unsigned int i_hash_shift __read_mostly;
56 static struct hlist_head *inode_hashtable __read_mostly;
57 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
59 __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
62 * Empty aops. Can be used for the cases where the user does not
63 * define any of the address_space operations.
65 const struct address_space_operations empty_aops = {
67 EXPORT_SYMBOL(empty_aops);
70 * Statistics gathering..
72 struct inodes_stat_t inodes_stat;
74 static DEFINE_PER_CPU(unsigned long, nr_inodes);
75 static DEFINE_PER_CPU(unsigned long, nr_unused);
77 static struct kmem_cache *inode_cachep __read_mostly;
79 static long get_nr_inodes(void)
83 for_each_possible_cpu(i)
84 sum += per_cpu(nr_inodes, i);
85 return sum < 0 ? 0 : sum;
88 static inline long get_nr_inodes_unused(void)
92 for_each_possible_cpu(i)
93 sum += per_cpu(nr_unused, i);
94 return sum < 0 ? 0 : sum;
97 long get_nr_dirty_inodes(void)
99 /* not actually dirty inodes, but a wild approximation */
100 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
101 return nr_dirty > 0 ? nr_dirty : 0;
105 * Handle nr_inode sysctl
108 int proc_nr_inodes(struct ctl_table *table, int write,
109 void __user *buffer, size_t *lenp, loff_t *ppos)
111 inodes_stat.nr_inodes = get_nr_inodes();
112 inodes_stat.nr_unused = get_nr_inodes_unused();
113 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
117 static int no_open(struct inode *inode, struct file *file)
123 * inode_init_always - perform inode structure intialisation
124 * @sb: superblock inode belongs to
125 * @inode: inode to initialise
127 * These are initializations that need to be done on every inode
128 * allocation as the fields are not initialised by slab allocation.
130 int inode_init_always(struct super_block *sb, struct inode *inode)
132 static const struct inode_operations empty_iops;
133 static const struct file_operations no_open_fops = {.open = no_open};
134 struct address_space *const mapping = &inode->i_data;
137 inode->i_blkbits = sb->s_blocksize_bits;
139 atomic_set(&inode->i_count, 1);
140 inode->i_op = &empty_iops;
141 inode->i_fop = &no_open_fops;
142 inode->__i_nlink = 1;
143 inode->i_opflags = 0;
144 i_uid_write(inode, 0);
145 i_gid_write(inode, 0);
146 atomic_set(&inode->i_writecount, 0);
150 inode->i_generation = 0;
151 inode->i_pipe = NULL;
152 inode->i_bdev = NULL;
153 inode->i_cdev = NULL;
155 inode->dirtied_when = 0;
157 if (security_inode_alloc(inode))
159 spin_lock_init(&inode->i_lock);
160 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
162 mutex_init(&inode->i_mutex);
163 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
165 atomic_set(&inode->i_dio_count, 0);
167 mapping->a_ops = &empty_aops;
168 mapping->host = inode;
170 atomic_set(&mapping->i_mmap_writable, 0);
171 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
172 mapping->private_data = NULL;
173 mapping->writeback_index = 0;
174 inode->i_private = NULL;
175 inode->i_mapping = mapping;
176 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
177 #ifdef CONFIG_FS_POSIX_ACL
178 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
181 #ifdef CONFIG_FSNOTIFY
182 inode->i_fsnotify_mask = 0;
184 inode->i_flctx = NULL;
185 this_cpu_inc(nr_inodes);
191 EXPORT_SYMBOL(inode_init_always);
193 static struct inode *alloc_inode(struct super_block *sb)
197 if (sb->s_op->alloc_inode)
198 inode = sb->s_op->alloc_inode(sb);
200 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
205 if (unlikely(inode_init_always(sb, inode))) {
206 if (inode->i_sb->s_op->destroy_inode)
207 inode->i_sb->s_op->destroy_inode(inode);
209 kmem_cache_free(inode_cachep, inode);
216 void free_inode_nonrcu(struct inode *inode)
218 kmem_cache_free(inode_cachep, inode);
220 EXPORT_SYMBOL(free_inode_nonrcu);
222 void __destroy_inode(struct inode *inode)
224 BUG_ON(inode_has_buffers(inode));
225 security_inode_free(inode);
226 fsnotify_inode_delete(inode);
227 locks_free_lock_context(inode->i_flctx);
228 if (!inode->i_nlink) {
229 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
230 atomic_long_dec(&inode->i_sb->s_remove_count);
233 #ifdef CONFIG_FS_POSIX_ACL
234 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
235 posix_acl_release(inode->i_acl);
236 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
237 posix_acl_release(inode->i_default_acl);
239 this_cpu_dec(nr_inodes);
241 EXPORT_SYMBOL(__destroy_inode);
243 static void i_callback(struct rcu_head *head)
245 struct inode *inode = container_of(head, struct inode, i_rcu);
246 kmem_cache_free(inode_cachep, inode);
249 static void destroy_inode(struct inode *inode)
251 BUG_ON(!list_empty(&inode->i_lru));
252 __destroy_inode(inode);
253 if (inode->i_sb->s_op->destroy_inode)
254 inode->i_sb->s_op->destroy_inode(inode);
256 call_rcu(&inode->i_rcu, i_callback);
260 * drop_nlink - directly drop an inode's link count
263 * This is a low-level filesystem helper to replace any
264 * direct filesystem manipulation of i_nlink. In cases
265 * where we are attempting to track writes to the
266 * filesystem, a decrement to zero means an imminent
267 * write when the file is truncated and actually unlinked
270 void drop_nlink(struct inode *inode)
272 WARN_ON(inode->i_nlink == 0);
275 atomic_long_inc(&inode->i_sb->s_remove_count);
277 EXPORT_SYMBOL(drop_nlink);
280 * clear_nlink - directly zero an inode's link count
283 * This is a low-level filesystem helper to replace any
284 * direct filesystem manipulation of i_nlink. See
285 * drop_nlink() for why we care about i_nlink hitting zero.
287 void clear_nlink(struct inode *inode)
289 if (inode->i_nlink) {
290 inode->__i_nlink = 0;
291 atomic_long_inc(&inode->i_sb->s_remove_count);
294 EXPORT_SYMBOL(clear_nlink);
297 * set_nlink - directly set an inode's link count
299 * @nlink: new nlink (should be non-zero)
301 * This is a low-level filesystem helper to replace any
302 * direct filesystem manipulation of i_nlink.
304 void set_nlink(struct inode *inode, unsigned int nlink)
309 /* Yes, some filesystems do change nlink from zero to one */
310 if (inode->i_nlink == 0)
311 atomic_long_dec(&inode->i_sb->s_remove_count);
313 inode->__i_nlink = nlink;
316 EXPORT_SYMBOL(set_nlink);
319 * inc_nlink - directly increment an inode's link count
322 * This is a low-level filesystem helper to replace any
323 * direct filesystem manipulation of i_nlink. Currently,
324 * it is only here for parity with dec_nlink().
326 void inc_nlink(struct inode *inode)
328 if (unlikely(inode->i_nlink == 0)) {
329 WARN_ON(!(inode->i_state & I_LINKABLE));
330 atomic_long_dec(&inode->i_sb->s_remove_count);
335 EXPORT_SYMBOL(inc_nlink);
337 void address_space_init_once(struct address_space *mapping)
339 memset(mapping, 0, sizeof(*mapping));
340 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
341 spin_lock_init(&mapping->tree_lock);
342 init_rwsem(&mapping->i_mmap_rwsem);
343 INIT_LIST_HEAD(&mapping->private_list);
344 spin_lock_init(&mapping->private_lock);
345 mapping->i_mmap = RB_ROOT;
347 EXPORT_SYMBOL(address_space_init_once);
350 * These are initializations that only need to be done
351 * once, because the fields are idempotent across use
352 * of the inode, so let the slab aware of that.
354 void inode_init_once(struct inode *inode)
356 memset(inode, 0, sizeof(*inode));
357 INIT_HLIST_NODE(&inode->i_hash);
358 INIT_LIST_HEAD(&inode->i_devices);
359 INIT_LIST_HEAD(&inode->i_wb_list);
360 INIT_LIST_HEAD(&inode->i_lru);
361 address_space_init_once(&inode->i_data);
362 i_size_ordered_init(inode);
363 #ifdef CONFIG_FSNOTIFY
364 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
367 EXPORT_SYMBOL(inode_init_once);
369 static void init_once(void *foo)
371 struct inode *inode = (struct inode *) foo;
373 inode_init_once(inode);
377 * inode->i_lock must be held
379 void __iget(struct inode *inode)
381 atomic_inc(&inode->i_count);
385 * get additional reference to inode; caller must already hold one.
387 void ihold(struct inode *inode)
389 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
391 EXPORT_SYMBOL(ihold);
393 static void inode_lru_list_add(struct inode *inode)
395 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
396 this_cpu_inc(nr_unused);
400 * Add inode to LRU if needed (inode is unused and clean).
402 * Needs inode->i_lock held.
404 void inode_add_lru(struct inode *inode)
406 if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) &&
407 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
408 inode_lru_list_add(inode);
412 static void inode_lru_list_del(struct inode *inode)
415 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
416 this_cpu_dec(nr_unused);
420 * inode_sb_list_add - add inode to the superblock list of inodes
421 * @inode: inode to add
423 void inode_sb_list_add(struct inode *inode)
425 spin_lock(&inode_sb_list_lock);
426 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
427 spin_unlock(&inode_sb_list_lock);
429 EXPORT_SYMBOL_GPL(inode_sb_list_add);
431 static inline void inode_sb_list_del(struct inode *inode)
433 if (!list_empty(&inode->i_sb_list)) {
434 spin_lock(&inode_sb_list_lock);
435 list_del_init(&inode->i_sb_list);
436 spin_unlock(&inode_sb_list_lock);
440 static unsigned long hash(struct super_block *sb, unsigned long hashval)
444 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
446 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
447 return tmp & i_hash_mask;
451 * __insert_inode_hash - hash an inode
452 * @inode: unhashed inode
453 * @hashval: unsigned long value used to locate this object in the
456 * Add an inode to the inode hash for this superblock.
458 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
460 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
462 spin_lock(&inode_hash_lock);
463 spin_lock(&inode->i_lock);
464 hlist_add_head(&inode->i_hash, b);
465 spin_unlock(&inode->i_lock);
466 spin_unlock(&inode_hash_lock);
468 EXPORT_SYMBOL(__insert_inode_hash);
471 * __remove_inode_hash - remove an inode from the hash
472 * @inode: inode to unhash
474 * Remove an inode from the superblock.
476 void __remove_inode_hash(struct inode *inode)
478 spin_lock(&inode_hash_lock);
479 spin_lock(&inode->i_lock);
480 hlist_del_init(&inode->i_hash);
481 spin_unlock(&inode->i_lock);
482 spin_unlock(&inode_hash_lock);
484 EXPORT_SYMBOL(__remove_inode_hash);
486 void clear_inode(struct inode *inode)
490 * We have to cycle tree_lock here because reclaim can be still in the
491 * process of removing the last page (in __delete_from_page_cache())
492 * and we must not free mapping under it.
494 spin_lock_irq(&inode->i_data.tree_lock);
495 BUG_ON(inode->i_data.nrpages);
496 BUG_ON(inode->i_data.nrshadows);
497 spin_unlock_irq(&inode->i_data.tree_lock);
498 BUG_ON(!list_empty(&inode->i_data.private_list));
499 BUG_ON(!(inode->i_state & I_FREEING));
500 BUG_ON(inode->i_state & I_CLEAR);
501 /* don't need i_lock here, no concurrent mods to i_state */
502 inode->i_state = I_FREEING | I_CLEAR;
504 EXPORT_SYMBOL(clear_inode);
507 * Free the inode passed in, removing it from the lists it is still connected
508 * to. We remove any pages still attached to the inode and wait for any IO that
509 * is still in progress before finally destroying the inode.
511 * An inode must already be marked I_FREEING so that we avoid the inode being
512 * moved back onto lists if we race with other code that manipulates the lists
513 * (e.g. writeback_single_inode). The caller is responsible for setting this.
515 * An inode must already be removed from the LRU list before being evicted from
516 * the cache. This should occur atomically with setting the I_FREEING state
517 * flag, so no inodes here should ever be on the LRU when being evicted.
519 static void evict(struct inode *inode)
521 const struct super_operations *op = inode->i_sb->s_op;
523 BUG_ON(!(inode->i_state & I_FREEING));
524 BUG_ON(!list_empty(&inode->i_lru));
526 if (!list_empty(&inode->i_wb_list))
527 inode_wb_list_del(inode);
529 inode_sb_list_del(inode);
532 * Wait for flusher thread to be done with the inode so that filesystem
533 * does not start destroying it while writeback is still running. Since
534 * the inode has I_FREEING set, flusher thread won't start new work on
535 * the inode. We just have to wait for running writeback to finish.
537 inode_wait_for_writeback(inode);
539 if (op->evict_inode) {
540 op->evict_inode(inode);
542 truncate_inode_pages_final(&inode->i_data);
545 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
547 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
550 remove_inode_hash(inode);
552 spin_lock(&inode->i_lock);
553 wake_up_bit(&inode->i_state, __I_NEW);
554 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
555 spin_unlock(&inode->i_lock);
557 destroy_inode(inode);
561 * dispose_list - dispose of the contents of a local list
562 * @head: the head of the list to free
564 * Dispose-list gets a local list with local inodes in it, so it doesn't
565 * need to worry about list corruption and SMP locks.
567 static void dispose_list(struct list_head *head)
569 while (!list_empty(head)) {
572 inode = list_first_entry(head, struct inode, i_lru);
573 list_del_init(&inode->i_lru);
580 * evict_inodes - evict all evictable inodes for a superblock
581 * @sb: superblock to operate on
583 * Make sure that no inodes with zero refcount are retained. This is
584 * called by superblock shutdown after having MS_ACTIVE flag removed,
585 * so any inode reaching zero refcount during or after that call will
586 * be immediately evicted.
588 void evict_inodes(struct super_block *sb)
590 struct inode *inode, *next;
593 spin_lock(&inode_sb_list_lock);
594 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
595 if (atomic_read(&inode->i_count))
598 spin_lock(&inode->i_lock);
599 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
600 spin_unlock(&inode->i_lock);
604 inode->i_state |= I_FREEING;
605 inode_lru_list_del(inode);
606 spin_unlock(&inode->i_lock);
607 list_add(&inode->i_lru, &dispose);
609 spin_unlock(&inode_sb_list_lock);
611 dispose_list(&dispose);
615 * invalidate_inodes - attempt to free all inodes on a superblock
616 * @sb: superblock to operate on
617 * @kill_dirty: flag to guide handling of dirty inodes
619 * Attempts to free all inodes for a given superblock. If there were any
620 * busy inodes return a non-zero value, else zero.
621 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
624 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
627 struct inode *inode, *next;
630 spin_lock(&inode_sb_list_lock);
631 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
632 spin_lock(&inode->i_lock);
633 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
634 spin_unlock(&inode->i_lock);
637 if (inode->i_state & I_DIRTY && !kill_dirty) {
638 spin_unlock(&inode->i_lock);
642 if (atomic_read(&inode->i_count)) {
643 spin_unlock(&inode->i_lock);
648 inode->i_state |= I_FREEING;
649 inode_lru_list_del(inode);
650 spin_unlock(&inode->i_lock);
651 list_add(&inode->i_lru, &dispose);
653 spin_unlock(&inode_sb_list_lock);
655 dispose_list(&dispose);
661 * Isolate the inode from the LRU in preparation for freeing it.
663 * Any inodes which are pinned purely because of attached pagecache have their
664 * pagecache removed. If the inode has metadata buffers attached to
665 * mapping->private_list then try to remove them.
667 * If the inode has the I_REFERENCED flag set, then it means that it has been
668 * used recently - the flag is set in iput_final(). When we encounter such an
669 * inode, clear the flag and move it to the back of the LRU so it gets another
670 * pass through the LRU before it gets reclaimed. This is necessary because of
671 * the fact we are doing lazy LRU updates to minimise lock contention so the
672 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
673 * with this flag set because they are the inodes that are out of order.
675 static enum lru_status inode_lru_isolate(struct list_head *item,
676 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
678 struct list_head *freeable = arg;
679 struct inode *inode = container_of(item, struct inode, i_lru);
682 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
683 * If we fail to get the lock, just skip it.
685 if (!spin_trylock(&inode->i_lock))
689 * Referenced or dirty inodes are still in use. Give them another pass
690 * through the LRU as we canot reclaim them now.
692 if (atomic_read(&inode->i_count) ||
693 (inode->i_state & ~I_REFERENCED)) {
694 list_lru_isolate(lru, &inode->i_lru);
695 spin_unlock(&inode->i_lock);
696 this_cpu_dec(nr_unused);
700 /* recently referenced inodes get one more pass */
701 if (inode->i_state & I_REFERENCED) {
702 inode->i_state &= ~I_REFERENCED;
703 spin_unlock(&inode->i_lock);
707 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
709 spin_unlock(&inode->i_lock);
710 spin_unlock(lru_lock);
711 if (remove_inode_buffers(inode)) {
713 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
714 if (current_is_kswapd())
715 __count_vm_events(KSWAPD_INODESTEAL, reap);
717 __count_vm_events(PGINODESTEAL, reap);
718 if (current->reclaim_state)
719 current->reclaim_state->reclaimed_slab += reap;
726 WARN_ON(inode->i_state & I_NEW);
727 inode->i_state |= I_FREEING;
728 list_lru_isolate_move(lru, &inode->i_lru, freeable);
729 spin_unlock(&inode->i_lock);
731 this_cpu_dec(nr_unused);
736 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
737 * This is called from the superblock shrinker function with a number of inodes
738 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
739 * then are freed outside inode_lock by dispose_list().
741 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
746 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
747 inode_lru_isolate, &freeable);
748 dispose_list(&freeable);
752 static void __wait_on_freeing_inode(struct inode *inode);
754 * Called with the inode lock held.
756 static struct inode *find_inode(struct super_block *sb,
757 struct hlist_head *head,
758 int (*test)(struct inode *, void *),
761 struct inode *inode = NULL;
764 hlist_for_each_entry(inode, head, i_hash) {
765 if (inode->i_sb != sb)
767 if (!test(inode, data))
769 spin_lock(&inode->i_lock);
770 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
771 __wait_on_freeing_inode(inode);
775 spin_unlock(&inode->i_lock);
782 * find_inode_fast is the fast path version of find_inode, see the comment at
783 * iget_locked for details.
785 static struct inode *find_inode_fast(struct super_block *sb,
786 struct hlist_head *head, unsigned long ino)
788 struct inode *inode = NULL;
791 hlist_for_each_entry(inode, head, i_hash) {
792 if (inode->i_ino != ino)
794 if (inode->i_sb != sb)
796 spin_lock(&inode->i_lock);
797 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
798 __wait_on_freeing_inode(inode);
802 spin_unlock(&inode->i_lock);
809 * Each cpu owns a range of LAST_INO_BATCH numbers.
810 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
811 * to renew the exhausted range.
813 * This does not significantly increase overflow rate because every CPU can
814 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
815 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
816 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
817 * overflow rate by 2x, which does not seem too significant.
819 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
820 * error if st_ino won't fit in target struct field. Use 32bit counter
821 * here to attempt to avoid that.
823 #define LAST_INO_BATCH 1024
824 static DEFINE_PER_CPU(unsigned int, last_ino);
826 unsigned int get_next_ino(void)
828 unsigned int *p = &get_cpu_var(last_ino);
829 unsigned int res = *p;
832 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
833 static atomic_t shared_last_ino;
834 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
836 res = next - LAST_INO_BATCH;
841 put_cpu_var(last_ino);
844 EXPORT_SYMBOL(get_next_ino);
847 * new_inode_pseudo - obtain an inode
850 * Allocates a new inode for given superblock.
851 * Inode wont be chained in superblock s_inodes list
853 * - fs can't be unmount
854 * - quotas, fsnotify, writeback can't work
856 struct inode *new_inode_pseudo(struct super_block *sb)
858 struct inode *inode = alloc_inode(sb);
861 spin_lock(&inode->i_lock);
863 spin_unlock(&inode->i_lock);
864 INIT_LIST_HEAD(&inode->i_sb_list);
870 * new_inode - obtain an inode
873 * Allocates a new inode for given superblock. The default gfp_mask
874 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
875 * If HIGHMEM pages are unsuitable or it is known that pages allocated
876 * for the page cache are not reclaimable or migratable,
877 * mapping_set_gfp_mask() must be called with suitable flags on the
878 * newly created inode's mapping
881 struct inode *new_inode(struct super_block *sb)
885 spin_lock_prefetch(&inode_sb_list_lock);
887 inode = new_inode_pseudo(sb);
889 inode_sb_list_add(inode);
892 EXPORT_SYMBOL(new_inode);
894 #ifdef CONFIG_DEBUG_LOCK_ALLOC
895 void lockdep_annotate_inode_mutex_key(struct inode *inode)
897 if (S_ISDIR(inode->i_mode)) {
898 struct file_system_type *type = inode->i_sb->s_type;
900 /* Set new key only if filesystem hasn't already changed it */
901 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
903 * ensure nobody is actually holding i_mutex
905 mutex_destroy(&inode->i_mutex);
906 mutex_init(&inode->i_mutex);
907 lockdep_set_class(&inode->i_mutex,
908 &type->i_mutex_dir_key);
912 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
916 * unlock_new_inode - clear the I_NEW state and wake up any waiters
917 * @inode: new inode to unlock
919 * Called when the inode is fully initialised to clear the new state of the
920 * inode and wake up anyone waiting for the inode to finish initialisation.
922 void unlock_new_inode(struct inode *inode)
924 lockdep_annotate_inode_mutex_key(inode);
925 spin_lock(&inode->i_lock);
926 WARN_ON(!(inode->i_state & I_NEW));
927 inode->i_state &= ~I_NEW;
929 wake_up_bit(&inode->i_state, __I_NEW);
930 spin_unlock(&inode->i_lock);
932 EXPORT_SYMBOL(unlock_new_inode);
935 * lock_two_nondirectories - take two i_mutexes on non-directory objects
937 * Lock any non-NULL argument that is not a directory.
938 * Zero, one or two objects may be locked by this function.
940 * @inode1: first inode to lock
941 * @inode2: second inode to lock
943 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
946 swap(inode1, inode2);
948 if (inode1 && !S_ISDIR(inode1->i_mode))
949 mutex_lock(&inode1->i_mutex);
950 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
951 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_NONDIR2);
953 EXPORT_SYMBOL(lock_two_nondirectories);
956 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
957 * @inode1: first inode to unlock
958 * @inode2: second inode to unlock
960 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
962 if (inode1 && !S_ISDIR(inode1->i_mode))
963 mutex_unlock(&inode1->i_mutex);
964 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
965 mutex_unlock(&inode2->i_mutex);
967 EXPORT_SYMBOL(unlock_two_nondirectories);
970 * iget5_locked - obtain an inode from a mounted file system
971 * @sb: super block of file system
972 * @hashval: hash value (usually inode number) to get
973 * @test: callback used for comparisons between inodes
974 * @set: callback used to initialize a new struct inode
975 * @data: opaque data pointer to pass to @test and @set
977 * Search for the inode specified by @hashval and @data in the inode cache,
978 * and if present it is return it with an increased reference count. This is
979 * a generalized version of iget_locked() for file systems where the inode
980 * number is not sufficient for unique identification of an inode.
982 * If the inode is not in cache, allocate a new inode and return it locked,
983 * hashed, and with the I_NEW flag set. The file system gets to fill it in
984 * before unlocking it via unlock_new_inode().
986 * Note both @test and @set are called with the inode_hash_lock held, so can't
989 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
990 int (*test)(struct inode *, void *),
991 int (*set)(struct inode *, void *), void *data)
993 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
996 spin_lock(&inode_hash_lock);
997 inode = find_inode(sb, head, test, data);
998 spin_unlock(&inode_hash_lock);
1001 wait_on_inode(inode);
1005 inode = alloc_inode(sb);
1009 spin_lock(&inode_hash_lock);
1010 /* We released the lock, so.. */
1011 old = find_inode(sb, head, test, data);
1013 if (set(inode, data))
1016 spin_lock(&inode->i_lock);
1017 inode->i_state = I_NEW;
1018 hlist_add_head(&inode->i_hash, head);
1019 spin_unlock(&inode->i_lock);
1020 inode_sb_list_add(inode);
1021 spin_unlock(&inode_hash_lock);
1023 /* Return the locked inode with I_NEW set, the
1024 * caller is responsible for filling in the contents
1030 * Uhhuh, somebody else created the same inode under
1031 * us. Use the old inode instead of the one we just
1034 spin_unlock(&inode_hash_lock);
1035 destroy_inode(inode);
1037 wait_on_inode(inode);
1042 spin_unlock(&inode_hash_lock);
1043 destroy_inode(inode);
1046 EXPORT_SYMBOL(iget5_locked);
1049 * iget_locked - obtain an inode from a mounted file system
1050 * @sb: super block of file system
1051 * @ino: inode number to get
1053 * Search for the inode specified by @ino in the inode cache and if present
1054 * return it with an increased reference count. This is for file systems
1055 * where the inode number is sufficient for unique identification of an inode.
1057 * If the inode is not in cache, allocate a new inode and return it locked,
1058 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1059 * before unlocking it via unlock_new_inode().
1061 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1063 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1064 struct inode *inode;
1066 spin_lock(&inode_hash_lock);
1067 inode = find_inode_fast(sb, head, ino);
1068 spin_unlock(&inode_hash_lock);
1070 wait_on_inode(inode);
1074 inode = alloc_inode(sb);
1078 spin_lock(&inode_hash_lock);
1079 /* We released the lock, so.. */
1080 old = find_inode_fast(sb, head, ino);
1083 spin_lock(&inode->i_lock);
1084 inode->i_state = I_NEW;
1085 hlist_add_head(&inode->i_hash, head);
1086 spin_unlock(&inode->i_lock);
1087 inode_sb_list_add(inode);
1088 spin_unlock(&inode_hash_lock);
1090 /* Return the locked inode with I_NEW set, the
1091 * caller is responsible for filling in the contents
1097 * Uhhuh, somebody else created the same inode under
1098 * us. Use the old inode instead of the one we just
1101 spin_unlock(&inode_hash_lock);
1102 destroy_inode(inode);
1104 wait_on_inode(inode);
1108 EXPORT_SYMBOL(iget_locked);
1111 * search the inode cache for a matching inode number.
1112 * If we find one, then the inode number we are trying to
1113 * allocate is not unique and so we should not use it.
1115 * Returns 1 if the inode number is unique, 0 if it is not.
1117 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1119 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1120 struct inode *inode;
1122 spin_lock(&inode_hash_lock);
1123 hlist_for_each_entry(inode, b, i_hash) {
1124 if (inode->i_ino == ino && inode->i_sb == sb) {
1125 spin_unlock(&inode_hash_lock);
1129 spin_unlock(&inode_hash_lock);
1135 * iunique - get a unique inode number
1137 * @max_reserved: highest reserved inode number
1139 * Obtain an inode number that is unique on the system for a given
1140 * superblock. This is used by file systems that have no natural
1141 * permanent inode numbering system. An inode number is returned that
1142 * is higher than the reserved limit but unique.
1145 * With a large number of inodes live on the file system this function
1146 * currently becomes quite slow.
1148 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1151 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1152 * error if st_ino won't fit in target struct field. Use 32bit counter
1153 * here to attempt to avoid that.
1155 static DEFINE_SPINLOCK(iunique_lock);
1156 static unsigned int counter;
1159 spin_lock(&iunique_lock);
1161 if (counter <= max_reserved)
1162 counter = max_reserved + 1;
1164 } while (!test_inode_iunique(sb, res));
1165 spin_unlock(&iunique_lock);
1169 EXPORT_SYMBOL(iunique);
1171 struct inode *igrab(struct inode *inode)
1173 spin_lock(&inode->i_lock);
1174 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1176 spin_unlock(&inode->i_lock);
1178 spin_unlock(&inode->i_lock);
1180 * Handle the case where s_op->clear_inode is not been
1181 * called yet, and somebody is calling igrab
1182 * while the inode is getting freed.
1188 EXPORT_SYMBOL(igrab);
1191 * ilookup5_nowait - search for an inode in the inode cache
1192 * @sb: super block of file system to search
1193 * @hashval: hash value (usually inode number) to search for
1194 * @test: callback used for comparisons between inodes
1195 * @data: opaque data pointer to pass to @test
1197 * Search for the inode specified by @hashval and @data in the inode cache.
1198 * If the inode is in the cache, the inode is returned with an incremented
1201 * Note: I_NEW is not waited upon so you have to be very careful what you do
1202 * with the returned inode. You probably should be using ilookup5() instead.
1204 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1206 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1207 int (*test)(struct inode *, void *), void *data)
1209 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1210 struct inode *inode;
1212 spin_lock(&inode_hash_lock);
1213 inode = find_inode(sb, head, test, data);
1214 spin_unlock(&inode_hash_lock);
1218 EXPORT_SYMBOL(ilookup5_nowait);
1221 * ilookup5 - search for an inode in the inode cache
1222 * @sb: super block of file system to search
1223 * @hashval: hash value (usually inode number) to search for
1224 * @test: callback used for comparisons between inodes
1225 * @data: opaque data pointer to pass to @test
1227 * Search for the inode specified by @hashval and @data in the inode cache,
1228 * and if the inode is in the cache, return the inode with an incremented
1229 * reference count. Waits on I_NEW before returning the inode.
1230 * returned with an incremented reference count.
1232 * This is a generalized version of ilookup() for file systems where the
1233 * inode number is not sufficient for unique identification of an inode.
1235 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1237 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1238 int (*test)(struct inode *, void *), void *data)
1240 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1243 wait_on_inode(inode);
1246 EXPORT_SYMBOL(ilookup5);
1249 * ilookup - search for an inode in the inode cache
1250 * @sb: super block of file system to search
1251 * @ino: inode number to search for
1253 * Search for the inode @ino in the inode cache, and if the inode is in the
1254 * cache, the inode is returned with an incremented reference count.
1256 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1258 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1259 struct inode *inode;
1261 spin_lock(&inode_hash_lock);
1262 inode = find_inode_fast(sb, head, ino);
1263 spin_unlock(&inode_hash_lock);
1266 wait_on_inode(inode);
1269 EXPORT_SYMBOL(ilookup);
1271 int insert_inode_locked(struct inode *inode)
1273 struct super_block *sb = inode->i_sb;
1274 ino_t ino = inode->i_ino;
1275 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1278 struct inode *old = NULL;
1279 spin_lock(&inode_hash_lock);
1280 hlist_for_each_entry(old, head, i_hash) {
1281 if (old->i_ino != ino)
1283 if (old->i_sb != sb)
1285 spin_lock(&old->i_lock);
1286 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1287 spin_unlock(&old->i_lock);
1293 spin_lock(&inode->i_lock);
1294 inode->i_state |= I_NEW;
1295 hlist_add_head(&inode->i_hash, head);
1296 spin_unlock(&inode->i_lock);
1297 spin_unlock(&inode_hash_lock);
1301 spin_unlock(&old->i_lock);
1302 spin_unlock(&inode_hash_lock);
1304 if (unlikely(!inode_unhashed(old))) {
1311 EXPORT_SYMBOL(insert_inode_locked);
1313 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1314 int (*test)(struct inode *, void *), void *data)
1316 struct super_block *sb = inode->i_sb;
1317 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1320 struct inode *old = NULL;
1322 spin_lock(&inode_hash_lock);
1323 hlist_for_each_entry(old, head, i_hash) {
1324 if (old->i_sb != sb)
1326 if (!test(old, data))
1328 spin_lock(&old->i_lock);
1329 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1330 spin_unlock(&old->i_lock);
1336 spin_lock(&inode->i_lock);
1337 inode->i_state |= I_NEW;
1338 hlist_add_head(&inode->i_hash, head);
1339 spin_unlock(&inode->i_lock);
1340 spin_unlock(&inode_hash_lock);
1344 spin_unlock(&old->i_lock);
1345 spin_unlock(&inode_hash_lock);
1347 if (unlikely(!inode_unhashed(old))) {
1354 EXPORT_SYMBOL(insert_inode_locked4);
1357 int generic_delete_inode(struct inode *inode)
1361 EXPORT_SYMBOL(generic_delete_inode);
1364 * Called when we're dropping the last reference
1367 * Call the FS "drop_inode()" function, defaulting to
1368 * the legacy UNIX filesystem behaviour. If it tells
1369 * us to evict inode, do so. Otherwise, retain inode
1370 * in cache if fs is alive, sync and evict if fs is
1373 static void iput_final(struct inode *inode)
1375 struct super_block *sb = inode->i_sb;
1376 const struct super_operations *op = inode->i_sb->s_op;
1379 WARN_ON(inode->i_state & I_NEW);
1382 drop = op->drop_inode(inode);
1384 drop = generic_drop_inode(inode);
1386 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1387 inode->i_state |= I_REFERENCED;
1388 inode_add_lru(inode);
1389 spin_unlock(&inode->i_lock);
1394 inode->i_state |= I_WILL_FREE;
1395 spin_unlock(&inode->i_lock);
1396 write_inode_now(inode, 1);
1397 spin_lock(&inode->i_lock);
1398 WARN_ON(inode->i_state & I_NEW);
1399 inode->i_state &= ~I_WILL_FREE;
1402 inode->i_state |= I_FREEING;
1403 if (!list_empty(&inode->i_lru))
1404 inode_lru_list_del(inode);
1405 spin_unlock(&inode->i_lock);
1411 * iput - put an inode
1412 * @inode: inode to put
1414 * Puts an inode, dropping its usage count. If the inode use count hits
1415 * zero, the inode is then freed and may also be destroyed.
1417 * Consequently, iput() can sleep.
1419 void iput(struct inode *inode)
1422 BUG_ON(inode->i_state & I_CLEAR);
1424 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1428 EXPORT_SYMBOL(iput);
1431 * bmap - find a block number in a file
1432 * @inode: inode of file
1433 * @block: block to find
1435 * Returns the block number on the device holding the inode that
1436 * is the disk block number for the block of the file requested.
1437 * That is, asked for block 4 of inode 1 the function will return the
1438 * disk block relative to the disk start that holds that block of the
1441 sector_t bmap(struct inode *inode, sector_t block)
1444 if (inode->i_mapping->a_ops->bmap)
1445 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1448 EXPORT_SYMBOL(bmap);
1451 * With relative atime, only update atime if the previous atime is
1452 * earlier than either the ctime or mtime or if at least a day has
1453 * passed since the last atime update.
1455 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1456 struct timespec now)
1459 if (!(mnt->mnt_flags & MNT_RELATIME))
1462 * Is mtime younger than atime? If yes, update atime:
1464 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1467 * Is ctime younger than atime? If yes, update atime:
1469 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1473 * Is the previous atime value older than a day? If yes,
1476 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1479 * Good, we can skip the atime update:
1485 * This does the actual work of updating an inodes time or version. Must have
1486 * had called mnt_want_write() before calling this.
1488 static int update_time(struct inode *inode, struct timespec *time, int flags)
1490 if (inode->i_op->update_time)
1491 return inode->i_op->update_time(inode, time, flags);
1493 if (flags & S_ATIME)
1494 inode->i_atime = *time;
1495 if (flags & S_VERSION)
1496 inode_inc_iversion(inode);
1497 if (flags & S_CTIME)
1498 inode->i_ctime = *time;
1499 if (flags & S_MTIME)
1500 inode->i_mtime = *time;
1501 mark_inode_dirty_sync(inode);
1506 * touch_atime - update the access time
1507 * @path: the &struct path to update
1509 * Update the accessed time on an inode and mark it for writeback.
1510 * This function automatically handles read only file systems and media,
1511 * as well as the "noatime" flag and inode specific "noatime" markers.
1513 void touch_atime(const struct path *path)
1515 struct vfsmount *mnt = path->mnt;
1516 struct inode *inode = path->dentry->d_inode;
1517 struct timespec now;
1519 if (inode->i_flags & S_NOATIME)
1521 if (IS_NOATIME(inode))
1523 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1526 if (mnt->mnt_flags & MNT_NOATIME)
1528 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1531 now = current_fs_time(inode->i_sb);
1533 if (!relatime_need_update(mnt, inode, now))
1536 if (timespec_equal(&inode->i_atime, &now))
1539 if (!sb_start_write_trylock(inode->i_sb))
1542 if (__mnt_want_write(mnt))
1545 * File systems can error out when updating inodes if they need to
1546 * allocate new space to modify an inode (such is the case for
1547 * Btrfs), but since we touch atime while walking down the path we
1548 * really don't care if we failed to update the atime of the file,
1549 * so just ignore the return value.
1550 * We may also fail on filesystems that have the ability to make parts
1551 * of the fs read only, e.g. subvolumes in Btrfs.
1553 update_time(inode, &now, S_ATIME);
1554 __mnt_drop_write(mnt);
1556 sb_end_write(inode->i_sb);
1558 EXPORT_SYMBOL(touch_atime);
1561 * The logic we want is
1563 * if suid or (sgid and xgrp)
1566 int should_remove_suid(struct dentry *dentry)
1568 umode_t mode = dentry->d_inode->i_mode;
1571 /* suid always must be killed */
1572 if (unlikely(mode & S_ISUID))
1573 kill = ATTR_KILL_SUID;
1576 * sgid without any exec bits is just a mandatory locking mark; leave
1577 * it alone. If some exec bits are set, it's a real sgid; kill it.
1579 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1580 kill |= ATTR_KILL_SGID;
1582 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1587 EXPORT_SYMBOL(should_remove_suid);
1589 static int __remove_suid(struct dentry *dentry, int kill)
1591 struct iattr newattrs;
1593 newattrs.ia_valid = ATTR_FORCE | kill;
1595 * Note we call this on write, so notify_change will not
1596 * encounter any conflicting delegations:
1598 return notify_change(dentry, &newattrs, NULL);
1601 int file_remove_suid(struct file *file)
1603 struct dentry *dentry = file->f_path.dentry;
1604 struct inode *inode = dentry->d_inode;
1609 /* Fast path for nothing security related */
1610 if (IS_NOSEC(inode))
1613 killsuid = should_remove_suid(dentry);
1614 killpriv = security_inode_need_killpriv(dentry);
1619 error = security_inode_killpriv(dentry);
1620 if (!error && killsuid)
1621 error = __remove_suid(dentry, killsuid);
1622 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1623 inode->i_flags |= S_NOSEC;
1627 EXPORT_SYMBOL(file_remove_suid);
1630 * file_update_time - update mtime and ctime time
1631 * @file: file accessed
1633 * Update the mtime and ctime members of an inode and mark the inode
1634 * for writeback. Note that this function is meant exclusively for
1635 * usage in the file write path of filesystems, and filesystems may
1636 * choose to explicitly ignore update via this function with the
1637 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1638 * timestamps are handled by the server. This can return an error for
1639 * file systems who need to allocate space in order to update an inode.
1642 int file_update_time(struct file *file)
1644 struct inode *inode = file_inode(file);
1645 struct timespec now;
1649 /* First try to exhaust all avenues to not sync */
1650 if (IS_NOCMTIME(inode))
1653 now = current_fs_time(inode->i_sb);
1654 if (!timespec_equal(&inode->i_mtime, &now))
1657 if (!timespec_equal(&inode->i_ctime, &now))
1660 if (IS_I_VERSION(inode))
1661 sync_it |= S_VERSION;
1666 /* Finally allowed to write? Takes lock. */
1667 if (__mnt_want_write_file(file))
1670 ret = update_time(inode, &now, sync_it);
1671 __mnt_drop_write_file(file);
1675 EXPORT_SYMBOL(file_update_time);
1677 int inode_needs_sync(struct inode *inode)
1681 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1685 EXPORT_SYMBOL(inode_needs_sync);
1688 * If we try to find an inode in the inode hash while it is being
1689 * deleted, we have to wait until the filesystem completes its
1690 * deletion before reporting that it isn't found. This function waits
1691 * until the deletion _might_ have completed. Callers are responsible
1692 * to recheck inode state.
1694 * It doesn't matter if I_NEW is not set initially, a call to
1695 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1698 static void __wait_on_freeing_inode(struct inode *inode)
1700 wait_queue_head_t *wq;
1701 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1702 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1703 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1704 spin_unlock(&inode->i_lock);
1705 spin_unlock(&inode_hash_lock);
1707 finish_wait(wq, &wait.wait);
1708 spin_lock(&inode_hash_lock);
1711 static __initdata unsigned long ihash_entries;
1712 static int __init set_ihash_entries(char *str)
1716 ihash_entries = simple_strtoul(str, &str, 0);
1719 __setup("ihash_entries=", set_ihash_entries);
1722 * Initialize the waitqueues and inode hash table.
1724 void __init inode_init_early(void)
1728 /* If hashes are distributed across NUMA nodes, defer
1729 * hash allocation until vmalloc space is available.
1735 alloc_large_system_hash("Inode-cache",
1736 sizeof(struct hlist_head),
1745 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1746 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1749 void __init inode_init(void)
1753 /* inode slab cache */
1754 inode_cachep = kmem_cache_create("inode_cache",
1755 sizeof(struct inode),
1757 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1761 /* Hash may have been set up in inode_init_early */
1766 alloc_large_system_hash("Inode-cache",
1767 sizeof(struct hlist_head),
1776 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1777 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1780 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1782 inode->i_mode = mode;
1783 if (S_ISCHR(mode)) {
1784 inode->i_fop = &def_chr_fops;
1785 inode->i_rdev = rdev;
1786 } else if (S_ISBLK(mode)) {
1787 inode->i_fop = &def_blk_fops;
1788 inode->i_rdev = rdev;
1789 } else if (S_ISFIFO(mode))
1790 inode->i_fop = &pipefifo_fops;
1791 else if (S_ISSOCK(mode))
1792 ; /* leave it no_open_fops */
1794 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1795 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1798 EXPORT_SYMBOL(init_special_inode);
1801 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1803 * @dir: Directory inode
1804 * @mode: mode of the new inode
1806 void inode_init_owner(struct inode *inode, const struct inode *dir,
1809 inode->i_uid = current_fsuid();
1810 if (dir && dir->i_mode & S_ISGID) {
1811 inode->i_gid = dir->i_gid;
1815 inode->i_gid = current_fsgid();
1816 inode->i_mode = mode;
1818 EXPORT_SYMBOL(inode_init_owner);
1821 * inode_owner_or_capable - check current task permissions to inode
1822 * @inode: inode being checked
1824 * Return true if current either has CAP_FOWNER in a namespace with the
1825 * inode owner uid mapped, or owns the file.
1827 bool inode_owner_or_capable(const struct inode *inode)
1829 struct user_namespace *ns;
1831 if (uid_eq(current_fsuid(), inode->i_uid))
1834 ns = current_user_ns();
1835 if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid))
1839 EXPORT_SYMBOL(inode_owner_or_capable);
1842 * Direct i/o helper functions
1844 static void __inode_dio_wait(struct inode *inode)
1846 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1847 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1850 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1851 if (atomic_read(&inode->i_dio_count))
1853 } while (atomic_read(&inode->i_dio_count));
1854 finish_wait(wq, &q.wait);
1858 * inode_dio_wait - wait for outstanding DIO requests to finish
1859 * @inode: inode to wait for
1861 * Waits for all pending direct I/O requests to finish so that we can
1862 * proceed with a truncate or equivalent operation.
1864 * Must be called under a lock that serializes taking new references
1865 * to i_dio_count, usually by inode->i_mutex.
1867 void inode_dio_wait(struct inode *inode)
1869 if (atomic_read(&inode->i_dio_count))
1870 __inode_dio_wait(inode);
1872 EXPORT_SYMBOL(inode_dio_wait);
1875 * inode_dio_done - signal finish of a direct I/O requests
1876 * @inode: inode the direct I/O happens on
1878 * This is called once we've finished processing a direct I/O request,
1879 * and is used to wake up callers waiting for direct I/O to be quiesced.
1881 void inode_dio_done(struct inode *inode)
1883 if (atomic_dec_and_test(&inode->i_dio_count))
1884 wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1886 EXPORT_SYMBOL(inode_dio_done);
1889 * inode_set_flags - atomically set some inode flags
1891 * Note: the caller should be holding i_mutex, or else be sure that
1892 * they have exclusive access to the inode structure (i.e., while the
1893 * inode is being instantiated). The reason for the cmpxchg() loop
1894 * --- which wouldn't be necessary if all code paths which modify
1895 * i_flags actually followed this rule, is that there is at least one
1896 * code path which doesn't today --- for example,
1897 * __generic_file_aio_write() calls file_remove_suid() without holding
1898 * i_mutex --- so we use cmpxchg() out of an abundance of caution.
1900 * In the long run, i_mutex is overkill, and we should probably look
1901 * at using the i_lock spinlock to protect i_flags, and then make sure
1902 * it is so documented in include/linux/fs.h and that all code follows
1903 * the locking convention!!
1905 void inode_set_flags(struct inode *inode, unsigned int flags,
1908 unsigned int old_flags, new_flags;
1910 WARN_ON_ONCE(flags & ~mask);
1912 old_flags = ACCESS_ONCE(inode->i_flags);
1913 new_flags = (old_flags & ~mask) | flags;
1914 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
1915 new_flags) != old_flags));
1917 EXPORT_SYMBOL(inode_set_flags);