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>
21 #include <trace/events/writeback.h>
25 * Inode locking rules:
27 * inode->i_lock protects:
28 * inode->i_state, inode->i_hash, __iget()
29 * Inode LRU list locks protect:
30 * inode->i_sb->s_inode_lru, inode->i_lru
31 * inode->i_sb->s_inode_list_lock protects:
32 * inode->i_sb->s_inodes, inode->i_sb_list
33 * bdi->wb.list_lock protects:
34 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
35 * inode_hash_lock protects:
36 * inode_hashtable, inode->i_hash
40 * inode->i_sb->s_inode_list_lock
42 * Inode LRU list locks
48 * inode->i_sb->s_inode_list_lock
55 static unsigned int i_hash_mask __read_mostly;
56 static unsigned int i_hash_shift __read_mostly;
57 static struct hlist_head *inode_hashtable __read_mostly;
58 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
61 * Empty aops. Can be used for the cases where the user does not
62 * define any of the address_space operations.
64 const struct address_space_operations empty_aops = {
66 EXPORT_SYMBOL(empty_aops);
69 * Statistics gathering..
71 struct inodes_stat_t inodes_stat;
73 static DEFINE_PER_CPU(unsigned long, nr_inodes);
74 static DEFINE_PER_CPU(unsigned long, nr_unused);
76 static struct kmem_cache *inode_cachep __read_mostly;
78 static long get_nr_inodes(void)
82 for_each_possible_cpu(i)
83 sum += per_cpu(nr_inodes, i);
84 return sum < 0 ? 0 : sum;
87 static inline long get_nr_inodes_unused(void)
91 for_each_possible_cpu(i)
92 sum += per_cpu(nr_unused, i);
93 return sum < 0 ? 0 : sum;
96 long get_nr_dirty_inodes(void)
98 /* not actually dirty inodes, but a wild approximation */
99 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
100 return nr_dirty > 0 ? nr_dirty : 0;
104 * Handle nr_inode sysctl
107 int proc_nr_inodes(struct ctl_table *table, int write,
108 void __user *buffer, size_t *lenp, loff_t *ppos)
110 inodes_stat.nr_inodes = get_nr_inodes();
111 inodes_stat.nr_unused = get_nr_inodes_unused();
112 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
116 static int no_open(struct inode *inode, struct file *file)
122 * inode_init_always - perform inode structure initialisation
123 * @sb: superblock inode belongs to
124 * @inode: inode to initialise
126 * These are initializations that need to be done on every inode
127 * allocation as the fields are not initialised by slab allocation.
129 int inode_init_always(struct super_block *sb, struct inode *inode)
131 static const struct inode_operations empty_iops;
132 static const struct file_operations no_open_fops = {.open = no_open};
133 struct address_space *const mapping = &inode->i_data;
136 inode->i_blkbits = sb->s_blocksize_bits;
138 atomic64_set(&inode->i_sequence, 0);
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;
145 inode->i_opflags |= IOP_XATTR;
146 i_uid_write(inode, 0);
147 i_gid_write(inode, 0);
148 atomic_set(&inode->i_writecount, 0);
150 inode->i_write_hint = WRITE_LIFE_NOT_SET;
153 inode->i_generation = 0;
154 inode->i_pipe = NULL;
155 inode->i_bdev = NULL;
156 inode->i_cdev = NULL;
157 inode->i_link = NULL;
158 inode->i_dir_seq = 0;
160 inode->dirtied_when = 0;
162 #ifdef CONFIG_CGROUP_WRITEBACK
163 inode->i_wb_frn_winner = 0;
164 inode->i_wb_frn_avg_time = 0;
165 inode->i_wb_frn_history = 0;
168 spin_lock_init(&inode->i_lock);
169 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
171 init_rwsem(&inode->i_rwsem);
172 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
174 atomic_set(&inode->i_dio_count, 0);
176 mapping->a_ops = &empty_aops;
177 mapping->host = inode;
180 atomic_set(&mapping->i_mmap_writable, 0);
181 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
182 mapping->private_data = NULL;
183 mapping->writeback_index = 0;
184 inode->i_private = NULL;
185 inode->i_mapping = mapping;
186 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
187 #ifdef CONFIG_FS_POSIX_ACL
188 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
191 #ifdef CONFIG_FSNOTIFY
192 inode->i_fsnotify_mask = 0;
194 inode->i_flctx = NULL;
196 if (unlikely(security_inode_alloc(inode)))
198 this_cpu_inc(nr_inodes);
202 EXPORT_SYMBOL(inode_init_always);
204 static struct inode *alloc_inode(struct super_block *sb)
208 if (sb->s_op->alloc_inode)
209 inode = sb->s_op->alloc_inode(sb);
211 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
216 if (unlikely(inode_init_always(sb, inode))) {
217 if (inode->i_sb->s_op->destroy_inode)
218 inode->i_sb->s_op->destroy_inode(inode);
220 kmem_cache_free(inode_cachep, inode);
227 void free_inode_nonrcu(struct inode *inode)
229 kmem_cache_free(inode_cachep, inode);
231 EXPORT_SYMBOL(free_inode_nonrcu);
233 void __destroy_inode(struct inode *inode)
235 BUG_ON(inode_has_buffers(inode));
236 inode_detach_wb(inode);
237 security_inode_free(inode);
238 fsnotify_inode_delete(inode);
239 locks_free_lock_context(inode);
240 if (!inode->i_nlink) {
241 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
242 atomic_long_dec(&inode->i_sb->s_remove_count);
245 #ifdef CONFIG_FS_POSIX_ACL
246 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
247 posix_acl_release(inode->i_acl);
248 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
249 posix_acl_release(inode->i_default_acl);
251 this_cpu_dec(nr_inodes);
253 EXPORT_SYMBOL(__destroy_inode);
255 static void i_callback(struct rcu_head *head)
257 struct inode *inode = container_of(head, struct inode, i_rcu);
258 kmem_cache_free(inode_cachep, inode);
261 static void destroy_inode(struct inode *inode)
263 BUG_ON(!list_empty(&inode->i_lru));
264 __destroy_inode(inode);
265 if (inode->i_sb->s_op->destroy_inode)
266 inode->i_sb->s_op->destroy_inode(inode);
268 call_rcu(&inode->i_rcu, i_callback);
272 * drop_nlink - directly drop an inode's link count
275 * This is a low-level filesystem helper to replace any
276 * direct filesystem manipulation of i_nlink. In cases
277 * where we are attempting to track writes to the
278 * filesystem, a decrement to zero means an imminent
279 * write when the file is truncated and actually unlinked
282 void drop_nlink(struct inode *inode)
284 WARN_ON(inode->i_nlink == 0);
287 atomic_long_inc(&inode->i_sb->s_remove_count);
289 EXPORT_SYMBOL(drop_nlink);
292 * clear_nlink - directly zero an inode's link count
295 * This is a low-level filesystem helper to replace any
296 * direct filesystem manipulation of i_nlink. See
297 * drop_nlink() for why we care about i_nlink hitting zero.
299 void clear_nlink(struct inode *inode)
301 if (inode->i_nlink) {
302 inode->__i_nlink = 0;
303 atomic_long_inc(&inode->i_sb->s_remove_count);
306 EXPORT_SYMBOL(clear_nlink);
309 * set_nlink - directly set an inode's link count
311 * @nlink: new nlink (should be non-zero)
313 * This is a low-level filesystem helper to replace any
314 * direct filesystem manipulation of i_nlink.
316 void set_nlink(struct inode *inode, unsigned int nlink)
321 /* Yes, some filesystems do change nlink from zero to one */
322 if (inode->i_nlink == 0)
323 atomic_long_dec(&inode->i_sb->s_remove_count);
325 inode->__i_nlink = nlink;
328 EXPORT_SYMBOL(set_nlink);
331 * inc_nlink - directly increment an inode's link count
334 * This is a low-level filesystem helper to replace any
335 * direct filesystem manipulation of i_nlink. Currently,
336 * it is only here for parity with dec_nlink().
338 void inc_nlink(struct inode *inode)
340 if (unlikely(inode->i_nlink == 0)) {
341 WARN_ON(!(inode->i_state & I_LINKABLE));
342 atomic_long_dec(&inode->i_sb->s_remove_count);
347 EXPORT_SYMBOL(inc_nlink);
349 void address_space_init_once(struct address_space *mapping)
351 memset(mapping, 0, sizeof(*mapping));
352 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC | __GFP_ACCOUNT);
353 spin_lock_init(&mapping->tree_lock);
354 init_rwsem(&mapping->i_mmap_rwsem);
355 INIT_LIST_HEAD(&mapping->private_list);
356 spin_lock_init(&mapping->private_lock);
357 mapping->i_mmap = RB_ROOT_CACHED;
359 EXPORT_SYMBOL(address_space_init_once);
362 * These are initializations that only need to be done
363 * once, because the fields are idempotent across use
364 * of the inode, so let the slab aware of that.
366 void inode_init_once(struct inode *inode)
368 memset(inode, 0, sizeof(*inode));
369 INIT_HLIST_NODE(&inode->i_hash);
370 INIT_LIST_HEAD(&inode->i_devices);
371 INIT_LIST_HEAD(&inode->i_io_list);
372 INIT_LIST_HEAD(&inode->i_wb_list);
373 INIT_LIST_HEAD(&inode->i_lru);
374 address_space_init_once(&inode->i_data);
375 i_size_ordered_init(inode);
377 EXPORT_SYMBOL(inode_init_once);
379 static void init_once(void *foo)
381 struct inode *inode = (struct inode *) foo;
383 inode_init_once(inode);
387 * inode->i_lock must be held
389 void __iget(struct inode *inode)
391 atomic_inc(&inode->i_count);
395 * get additional reference to inode; caller must already hold one.
397 void ihold(struct inode *inode)
399 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
401 EXPORT_SYMBOL(ihold);
403 static void inode_lru_list_add(struct inode *inode)
405 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
406 this_cpu_inc(nr_unused);
408 inode->i_state |= I_REFERENCED;
412 * Add inode to LRU if needed (inode is unused and clean).
414 * Needs inode->i_lock held.
416 void inode_add_lru(struct inode *inode)
418 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
419 I_FREEING | I_WILL_FREE)) &&
420 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
421 inode_lru_list_add(inode);
425 static void inode_lru_list_del(struct inode *inode)
428 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
429 this_cpu_dec(nr_unused);
433 * inode_sb_list_add - add inode to the superblock list of inodes
434 * @inode: inode to add
436 void inode_sb_list_add(struct inode *inode)
438 spin_lock(&inode->i_sb->s_inode_list_lock);
439 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
440 spin_unlock(&inode->i_sb->s_inode_list_lock);
442 EXPORT_SYMBOL_GPL(inode_sb_list_add);
444 static inline void inode_sb_list_del(struct inode *inode)
446 if (!list_empty(&inode->i_sb_list)) {
447 spin_lock(&inode->i_sb->s_inode_list_lock);
448 list_del_init(&inode->i_sb_list);
449 spin_unlock(&inode->i_sb->s_inode_list_lock);
453 static unsigned long hash(struct super_block *sb, unsigned long hashval)
457 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
459 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
460 return tmp & i_hash_mask;
464 * __insert_inode_hash - hash an inode
465 * @inode: unhashed inode
466 * @hashval: unsigned long value used to locate this object in the
469 * Add an inode to the inode hash for this superblock.
471 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
473 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
475 spin_lock(&inode_hash_lock);
476 spin_lock(&inode->i_lock);
477 hlist_add_head(&inode->i_hash, b);
478 spin_unlock(&inode->i_lock);
479 spin_unlock(&inode_hash_lock);
481 EXPORT_SYMBOL(__insert_inode_hash);
484 * __remove_inode_hash - remove an inode from the hash
485 * @inode: inode to unhash
487 * Remove an inode from the superblock.
489 void __remove_inode_hash(struct inode *inode)
491 spin_lock(&inode_hash_lock);
492 spin_lock(&inode->i_lock);
493 hlist_del_init(&inode->i_hash);
494 spin_unlock(&inode->i_lock);
495 spin_unlock(&inode_hash_lock);
497 EXPORT_SYMBOL(__remove_inode_hash);
499 void clear_inode(struct inode *inode)
503 * We have to cycle tree_lock here because reclaim can be still in the
504 * process of removing the last page (in __delete_from_page_cache())
505 * and we must not free mapping under it.
507 spin_lock_irq(&inode->i_data.tree_lock);
508 BUG_ON(inode->i_data.nrpages);
509 BUG_ON(inode->i_data.nrexceptional);
510 spin_unlock_irq(&inode->i_data.tree_lock);
511 BUG_ON(!list_empty(&inode->i_data.private_list));
512 BUG_ON(!(inode->i_state & I_FREEING));
513 BUG_ON(inode->i_state & I_CLEAR);
514 BUG_ON(!list_empty(&inode->i_wb_list));
515 /* don't need i_lock here, no concurrent mods to i_state */
516 inode->i_state = I_FREEING | I_CLEAR;
518 EXPORT_SYMBOL(clear_inode);
521 * Free the inode passed in, removing it from the lists it is still connected
522 * to. We remove any pages still attached to the inode and wait for any IO that
523 * is still in progress before finally destroying the inode.
525 * An inode must already be marked I_FREEING so that we avoid the inode being
526 * moved back onto lists if we race with other code that manipulates the lists
527 * (e.g. writeback_single_inode). The caller is responsible for setting this.
529 * An inode must already be removed from the LRU list before being evicted from
530 * the cache. This should occur atomically with setting the I_FREEING state
531 * flag, so no inodes here should ever be on the LRU when being evicted.
533 static void evict(struct inode *inode)
535 const struct super_operations *op = inode->i_sb->s_op;
537 BUG_ON(!(inode->i_state & I_FREEING));
538 BUG_ON(!list_empty(&inode->i_lru));
540 if (!list_empty(&inode->i_io_list))
541 inode_io_list_del(inode);
543 inode_sb_list_del(inode);
546 * Wait for flusher thread to be done with the inode so that filesystem
547 * does not start destroying it while writeback is still running. Since
548 * the inode has I_FREEING set, flusher thread won't start new work on
549 * the inode. We just have to wait for running writeback to finish.
551 inode_wait_for_writeback(inode);
553 if (op->evict_inode) {
554 op->evict_inode(inode);
556 truncate_inode_pages_final(&inode->i_data);
559 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
561 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
564 remove_inode_hash(inode);
566 spin_lock(&inode->i_lock);
567 wake_up_bit(&inode->i_state, __I_NEW);
568 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
569 spin_unlock(&inode->i_lock);
571 destroy_inode(inode);
575 * dispose_list - dispose of the contents of a local list
576 * @head: the head of the list to free
578 * Dispose-list gets a local list with local inodes in it, so it doesn't
579 * need to worry about list corruption and SMP locks.
581 static void dispose_list(struct list_head *head)
583 while (!list_empty(head)) {
586 inode = list_first_entry(head, struct inode, i_lru);
587 list_del_init(&inode->i_lru);
595 * evict_inodes - evict all evictable inodes for a superblock
596 * @sb: superblock to operate on
598 * Make sure that no inodes with zero refcount are retained. This is
599 * called by superblock shutdown after having MS_ACTIVE flag removed,
600 * so any inode reaching zero refcount during or after that call will
601 * be immediately evicted.
603 void evict_inodes(struct super_block *sb)
605 struct inode *inode, *next;
609 spin_lock(&sb->s_inode_list_lock);
610 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
611 if (atomic_read(&inode->i_count))
614 spin_lock(&inode->i_lock);
615 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
616 spin_unlock(&inode->i_lock);
620 inode->i_state |= I_FREEING;
621 inode_lru_list_del(inode);
622 spin_unlock(&inode->i_lock);
623 list_add(&inode->i_lru, &dispose);
626 * We can have a ton of inodes to evict at unmount time given
627 * enough memory, check to see if we need to go to sleep for a
628 * bit so we don't livelock.
630 if (need_resched()) {
631 spin_unlock(&sb->s_inode_list_lock);
633 dispose_list(&dispose);
637 spin_unlock(&sb->s_inode_list_lock);
639 dispose_list(&dispose);
641 EXPORT_SYMBOL_GPL(evict_inodes);
644 * invalidate_inodes - attempt to free all inodes on a superblock
645 * @sb: superblock to operate on
646 * @kill_dirty: flag to guide handling of dirty inodes
648 * Attempts to free all inodes for a given superblock. If there were any
649 * busy inodes return a non-zero value, else zero.
650 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
653 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
656 struct inode *inode, *next;
660 spin_lock(&sb->s_inode_list_lock);
661 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
662 spin_lock(&inode->i_lock);
663 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
664 spin_unlock(&inode->i_lock);
667 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
668 spin_unlock(&inode->i_lock);
672 if (atomic_read(&inode->i_count)) {
673 spin_unlock(&inode->i_lock);
678 inode->i_state |= I_FREEING;
679 inode_lru_list_del(inode);
680 spin_unlock(&inode->i_lock);
681 list_add(&inode->i_lru, &dispose);
682 if (need_resched()) {
683 spin_unlock(&sb->s_inode_list_lock);
685 dispose_list(&dispose);
689 spin_unlock(&sb->s_inode_list_lock);
691 dispose_list(&dispose);
697 * Isolate the inode from the LRU in preparation for freeing it.
699 * Any inodes which are pinned purely because of attached pagecache have their
700 * pagecache removed. If the inode has metadata buffers attached to
701 * mapping->private_list then try to remove them.
703 * If the inode has the I_REFERENCED flag set, then it means that it has been
704 * used recently - the flag is set in iput_final(). When we encounter such an
705 * inode, clear the flag and move it to the back of the LRU so it gets another
706 * pass through the LRU before it gets reclaimed. This is necessary because of
707 * the fact we are doing lazy LRU updates to minimise lock contention so the
708 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
709 * with this flag set because they are the inodes that are out of order.
711 static enum lru_status inode_lru_isolate(struct list_head *item,
712 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
714 struct list_head *freeable = arg;
715 struct inode *inode = container_of(item, struct inode, i_lru);
718 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
719 * If we fail to get the lock, just skip it.
721 if (!spin_trylock(&inode->i_lock))
725 * Referenced or dirty inodes are still in use. Give them another pass
726 * through the LRU as we canot reclaim them now.
728 if (atomic_read(&inode->i_count) ||
729 (inode->i_state & ~I_REFERENCED)) {
730 list_lru_isolate(lru, &inode->i_lru);
731 spin_unlock(&inode->i_lock);
732 this_cpu_dec(nr_unused);
736 /* recently referenced inodes get one more pass */
737 if (inode->i_state & I_REFERENCED) {
738 inode->i_state &= ~I_REFERENCED;
739 spin_unlock(&inode->i_lock);
743 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
745 spin_unlock(&inode->i_lock);
746 spin_unlock(lru_lock);
747 if (remove_inode_buffers(inode)) {
749 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
750 if (current_is_kswapd())
751 __count_vm_events(KSWAPD_INODESTEAL, reap);
753 __count_vm_events(PGINODESTEAL, reap);
754 if (current->reclaim_state)
755 current->reclaim_state->reclaimed_slab += reap;
762 WARN_ON(inode->i_state & I_NEW);
763 inode->i_state |= I_FREEING;
764 list_lru_isolate_move(lru, &inode->i_lru, freeable);
765 spin_unlock(&inode->i_lock);
767 this_cpu_dec(nr_unused);
772 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
773 * This is called from the superblock shrinker function with a number of inodes
774 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
775 * then are freed outside inode_lock by dispose_list().
777 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
782 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
783 inode_lru_isolate, &freeable);
784 dispose_list(&freeable);
788 static void __wait_on_freeing_inode(struct inode *inode);
790 * Called with the inode lock held.
792 static struct inode *find_inode(struct super_block *sb,
793 struct hlist_head *head,
794 int (*test)(struct inode *, void *),
797 struct inode *inode = NULL;
800 hlist_for_each_entry(inode, head, i_hash) {
801 if (inode->i_sb != sb)
803 if (!test(inode, data))
805 spin_lock(&inode->i_lock);
806 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
807 __wait_on_freeing_inode(inode);
811 spin_unlock(&inode->i_lock);
818 * find_inode_fast is the fast path version of find_inode, see the comment at
819 * iget_locked for details.
821 static struct inode *find_inode_fast(struct super_block *sb,
822 struct hlist_head *head, unsigned long ino)
824 struct inode *inode = NULL;
827 hlist_for_each_entry(inode, head, i_hash) {
828 if (inode->i_ino != ino)
830 if (inode->i_sb != sb)
832 spin_lock(&inode->i_lock);
833 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
834 __wait_on_freeing_inode(inode);
838 spin_unlock(&inode->i_lock);
845 * Each cpu owns a range of LAST_INO_BATCH numbers.
846 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
847 * to renew the exhausted range.
849 * This does not significantly increase overflow rate because every CPU can
850 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
851 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
852 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
853 * overflow rate by 2x, which does not seem too significant.
855 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
856 * error if st_ino won't fit in target struct field. Use 32bit counter
857 * here to attempt to avoid that.
859 #define LAST_INO_BATCH 1024
860 static DEFINE_PER_CPU(unsigned int, last_ino);
862 unsigned int get_next_ino(void)
864 unsigned int *p = &get_cpu_var(last_ino);
865 unsigned int res = *p;
868 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
869 static atomic_t shared_last_ino;
870 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
872 res = next - LAST_INO_BATCH;
877 /* get_next_ino should not provide a 0 inode number */
881 put_cpu_var(last_ino);
884 EXPORT_SYMBOL(get_next_ino);
887 * new_inode_pseudo - obtain an inode
890 * Allocates a new inode for given superblock.
891 * Inode wont be chained in superblock s_inodes list
893 * - fs can't be unmount
894 * - quotas, fsnotify, writeback can't work
896 struct inode *new_inode_pseudo(struct super_block *sb)
898 struct inode *inode = alloc_inode(sb);
901 spin_lock(&inode->i_lock);
903 spin_unlock(&inode->i_lock);
904 INIT_LIST_HEAD(&inode->i_sb_list);
910 * new_inode - obtain an inode
913 * Allocates a new inode for given superblock. The default gfp_mask
914 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
915 * If HIGHMEM pages are unsuitable or it is known that pages allocated
916 * for the page cache are not reclaimable or migratable,
917 * mapping_set_gfp_mask() must be called with suitable flags on the
918 * newly created inode's mapping
921 struct inode *new_inode(struct super_block *sb)
925 spin_lock_prefetch(&sb->s_inode_list_lock);
927 inode = new_inode_pseudo(sb);
929 inode_sb_list_add(inode);
932 EXPORT_SYMBOL(new_inode);
934 #ifdef CONFIG_DEBUG_LOCK_ALLOC
935 void lockdep_annotate_inode_mutex_key(struct inode *inode)
937 if (S_ISDIR(inode->i_mode)) {
938 struct file_system_type *type = inode->i_sb->s_type;
940 /* Set new key only if filesystem hasn't already changed it */
941 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
943 * ensure nobody is actually holding i_mutex
945 // mutex_destroy(&inode->i_mutex);
946 init_rwsem(&inode->i_rwsem);
947 lockdep_set_class(&inode->i_rwsem,
948 &type->i_mutex_dir_key);
952 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
956 * unlock_new_inode - clear the I_NEW state and wake up any waiters
957 * @inode: new inode to unlock
959 * Called when the inode is fully initialised to clear the new state of the
960 * inode and wake up anyone waiting for the inode to finish initialisation.
962 void unlock_new_inode(struct inode *inode)
964 lockdep_annotate_inode_mutex_key(inode);
965 spin_lock(&inode->i_lock);
966 WARN_ON(!(inode->i_state & I_NEW));
967 inode->i_state &= ~I_NEW;
969 wake_up_bit(&inode->i_state, __I_NEW);
970 spin_unlock(&inode->i_lock);
972 EXPORT_SYMBOL(unlock_new_inode);
975 * lock_two_nondirectories - take two i_mutexes on non-directory objects
977 * Lock any non-NULL argument that is not a directory.
978 * Zero, one or two objects may be locked by this function.
980 * @inode1: first inode to lock
981 * @inode2: second inode to lock
983 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
986 swap(inode1, inode2);
988 if (inode1 && !S_ISDIR(inode1->i_mode))
990 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
991 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
993 EXPORT_SYMBOL(lock_two_nondirectories);
996 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
997 * @inode1: first inode to unlock
998 * @inode2: second inode to unlock
1000 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1002 if (inode1 && !S_ISDIR(inode1->i_mode))
1003 inode_unlock(inode1);
1004 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1005 inode_unlock(inode2);
1007 EXPORT_SYMBOL(unlock_two_nondirectories);
1010 * iget5_locked - obtain an inode from a mounted file system
1011 * @sb: super block of file system
1012 * @hashval: hash value (usually inode number) to get
1013 * @test: callback used for comparisons between inodes
1014 * @set: callback used to initialize a new struct inode
1015 * @data: opaque data pointer to pass to @test and @set
1017 * Search for the inode specified by @hashval and @data in the inode cache,
1018 * and if present it is return it with an increased reference count. This is
1019 * a generalized version of iget_locked() for file systems where the inode
1020 * number is not sufficient for unique identification of an inode.
1022 * If the inode is not in cache, allocate a new inode and return it locked,
1023 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1024 * before unlocking it via unlock_new_inode().
1026 * Note both @test and @set are called with the inode_hash_lock held, so can't
1029 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1030 int (*test)(struct inode *, void *),
1031 int (*set)(struct inode *, void *), void *data)
1033 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1034 struct inode *inode;
1036 spin_lock(&inode_hash_lock);
1037 inode = find_inode(sb, head, test, data);
1038 spin_unlock(&inode_hash_lock);
1041 wait_on_inode(inode);
1042 if (unlikely(inode_unhashed(inode))) {
1049 inode = alloc_inode(sb);
1053 spin_lock(&inode_hash_lock);
1054 /* We released the lock, so.. */
1055 old = find_inode(sb, head, test, data);
1057 if (set(inode, data))
1060 spin_lock(&inode->i_lock);
1061 inode->i_state = I_NEW;
1062 hlist_add_head(&inode->i_hash, head);
1063 spin_unlock(&inode->i_lock);
1064 inode_sb_list_add(inode);
1065 spin_unlock(&inode_hash_lock);
1067 /* Return the locked inode with I_NEW set, the
1068 * caller is responsible for filling in the contents
1074 * Uhhuh, somebody else created the same inode under
1075 * us. Use the old inode instead of the one we just
1078 spin_unlock(&inode_hash_lock);
1079 destroy_inode(inode);
1081 wait_on_inode(inode);
1082 if (unlikely(inode_unhashed(inode))) {
1090 spin_unlock(&inode_hash_lock);
1091 destroy_inode(inode);
1094 EXPORT_SYMBOL(iget5_locked);
1097 * iget_locked - obtain an inode from a mounted file system
1098 * @sb: super block of file system
1099 * @ino: inode number to get
1101 * Search for the inode specified by @ino in the inode cache and if present
1102 * return it with an increased reference count. This is for file systems
1103 * where the inode number is sufficient for unique identification of an inode.
1105 * If the inode is not in cache, allocate a new inode and return it locked,
1106 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1107 * before unlocking it via unlock_new_inode().
1109 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1111 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1112 struct inode *inode;
1114 spin_lock(&inode_hash_lock);
1115 inode = find_inode_fast(sb, head, ino);
1116 spin_unlock(&inode_hash_lock);
1118 wait_on_inode(inode);
1119 if (unlikely(inode_unhashed(inode))) {
1126 inode = alloc_inode(sb);
1130 spin_lock(&inode_hash_lock);
1131 /* We released the lock, so.. */
1132 old = find_inode_fast(sb, head, ino);
1135 spin_lock(&inode->i_lock);
1136 inode->i_state = I_NEW;
1137 hlist_add_head(&inode->i_hash, head);
1138 spin_unlock(&inode->i_lock);
1139 inode_sb_list_add(inode);
1140 spin_unlock(&inode_hash_lock);
1142 /* Return the locked inode with I_NEW set, the
1143 * caller is responsible for filling in the contents
1149 * Uhhuh, somebody else created the same inode under
1150 * us. Use the old inode instead of the one we just
1153 spin_unlock(&inode_hash_lock);
1154 destroy_inode(inode);
1156 wait_on_inode(inode);
1157 if (unlikely(inode_unhashed(inode))) {
1164 EXPORT_SYMBOL(iget_locked);
1167 * search the inode cache for a matching inode number.
1168 * If we find one, then the inode number we are trying to
1169 * allocate is not unique and so we should not use it.
1171 * Returns 1 if the inode number is unique, 0 if it is not.
1173 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1175 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1176 struct inode *inode;
1178 spin_lock(&inode_hash_lock);
1179 hlist_for_each_entry(inode, b, i_hash) {
1180 if (inode->i_ino == ino && inode->i_sb == sb) {
1181 spin_unlock(&inode_hash_lock);
1185 spin_unlock(&inode_hash_lock);
1191 * iunique - get a unique inode number
1193 * @max_reserved: highest reserved inode number
1195 * Obtain an inode number that is unique on the system for a given
1196 * superblock. This is used by file systems that have no natural
1197 * permanent inode numbering system. An inode number is returned that
1198 * is higher than the reserved limit but unique.
1201 * With a large number of inodes live on the file system this function
1202 * currently becomes quite slow.
1204 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1207 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1208 * error if st_ino won't fit in target struct field. Use 32bit counter
1209 * here to attempt to avoid that.
1211 static DEFINE_SPINLOCK(iunique_lock);
1212 static unsigned int counter;
1215 spin_lock(&iunique_lock);
1217 if (counter <= max_reserved)
1218 counter = max_reserved + 1;
1220 } while (!test_inode_iunique(sb, res));
1221 spin_unlock(&iunique_lock);
1225 EXPORT_SYMBOL(iunique);
1227 struct inode *igrab(struct inode *inode)
1229 spin_lock(&inode->i_lock);
1230 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1232 spin_unlock(&inode->i_lock);
1234 spin_unlock(&inode->i_lock);
1236 * Handle the case where s_op->clear_inode is not been
1237 * called yet, and somebody is calling igrab
1238 * while the inode is getting freed.
1244 EXPORT_SYMBOL(igrab);
1247 * ilookup5_nowait - search for an inode in the inode cache
1248 * @sb: super block of file system to search
1249 * @hashval: hash value (usually inode number) to search for
1250 * @test: callback used for comparisons between inodes
1251 * @data: opaque data pointer to pass to @test
1253 * Search for the inode specified by @hashval and @data in the inode cache.
1254 * If the inode is in the cache, the inode is returned with an incremented
1257 * Note: I_NEW is not waited upon so you have to be very careful what you do
1258 * with the returned inode. You probably should be using ilookup5() instead.
1260 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1262 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1263 int (*test)(struct inode *, void *), void *data)
1265 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1266 struct inode *inode;
1268 spin_lock(&inode_hash_lock);
1269 inode = find_inode(sb, head, test, data);
1270 spin_unlock(&inode_hash_lock);
1274 EXPORT_SYMBOL(ilookup5_nowait);
1277 * ilookup5 - search for an inode in the inode cache
1278 * @sb: super block of file system to search
1279 * @hashval: hash value (usually inode number) to search for
1280 * @test: callback used for comparisons between inodes
1281 * @data: opaque data pointer to pass to @test
1283 * Search for the inode specified by @hashval and @data in the inode cache,
1284 * and if the inode is in the cache, return the inode with an incremented
1285 * reference count. Waits on I_NEW before returning the inode.
1286 * returned with an incremented reference count.
1288 * This is a generalized version of ilookup() for file systems where the
1289 * inode number is not sufficient for unique identification of an inode.
1291 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1293 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1294 int (*test)(struct inode *, void *), void *data)
1296 struct inode *inode;
1298 inode = ilookup5_nowait(sb, hashval, test, data);
1300 wait_on_inode(inode);
1301 if (unlikely(inode_unhashed(inode))) {
1308 EXPORT_SYMBOL(ilookup5);
1311 * ilookup - search for an inode in the inode cache
1312 * @sb: super block of file system to search
1313 * @ino: inode number to search for
1315 * Search for the inode @ino in the inode cache, and if the inode is in the
1316 * cache, the inode is returned with an incremented reference count.
1318 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1320 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1321 struct inode *inode;
1323 spin_lock(&inode_hash_lock);
1324 inode = find_inode_fast(sb, head, ino);
1325 spin_unlock(&inode_hash_lock);
1328 wait_on_inode(inode);
1329 if (unlikely(inode_unhashed(inode))) {
1336 EXPORT_SYMBOL(ilookup);
1339 * find_inode_nowait - find an inode in the inode cache
1340 * @sb: super block of file system to search
1341 * @hashval: hash value (usually inode number) to search for
1342 * @match: callback used for comparisons between inodes
1343 * @data: opaque data pointer to pass to @match
1345 * Search for the inode specified by @hashval and @data in the inode
1346 * cache, where the helper function @match will return 0 if the inode
1347 * does not match, 1 if the inode does match, and -1 if the search
1348 * should be stopped. The @match function must be responsible for
1349 * taking the i_lock spin_lock and checking i_state for an inode being
1350 * freed or being initialized, and incrementing the reference count
1351 * before returning 1. It also must not sleep, since it is called with
1352 * the inode_hash_lock spinlock held.
1354 * This is a even more generalized version of ilookup5() when the
1355 * function must never block --- find_inode() can block in
1356 * __wait_on_freeing_inode() --- or when the caller can not increment
1357 * the reference count because the resulting iput() might cause an
1358 * inode eviction. The tradeoff is that the @match funtion must be
1359 * very carefully implemented.
1361 struct inode *find_inode_nowait(struct super_block *sb,
1362 unsigned long hashval,
1363 int (*match)(struct inode *, unsigned long,
1367 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1368 struct inode *inode, *ret_inode = NULL;
1371 spin_lock(&inode_hash_lock);
1372 hlist_for_each_entry(inode, head, i_hash) {
1373 if (inode->i_sb != sb)
1375 mval = match(inode, hashval, data);
1383 spin_unlock(&inode_hash_lock);
1386 EXPORT_SYMBOL(find_inode_nowait);
1388 int insert_inode_locked(struct inode *inode)
1390 struct super_block *sb = inode->i_sb;
1391 ino_t ino = inode->i_ino;
1392 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1395 struct inode *old = NULL;
1396 spin_lock(&inode_hash_lock);
1397 hlist_for_each_entry(old, head, i_hash) {
1398 if (old->i_ino != ino)
1400 if (old->i_sb != sb)
1402 spin_lock(&old->i_lock);
1403 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1404 spin_unlock(&old->i_lock);
1410 spin_lock(&inode->i_lock);
1411 inode->i_state |= I_NEW;
1412 hlist_add_head(&inode->i_hash, head);
1413 spin_unlock(&inode->i_lock);
1414 spin_unlock(&inode_hash_lock);
1418 spin_unlock(&old->i_lock);
1419 spin_unlock(&inode_hash_lock);
1421 if (unlikely(!inode_unhashed(old))) {
1428 EXPORT_SYMBOL(insert_inode_locked);
1430 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1431 int (*test)(struct inode *, void *), void *data)
1433 struct super_block *sb = inode->i_sb;
1434 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1437 struct inode *old = NULL;
1439 spin_lock(&inode_hash_lock);
1440 hlist_for_each_entry(old, head, i_hash) {
1441 if (old->i_sb != sb)
1443 if (!test(old, data))
1445 spin_lock(&old->i_lock);
1446 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1447 spin_unlock(&old->i_lock);
1453 spin_lock(&inode->i_lock);
1454 inode->i_state |= I_NEW;
1455 hlist_add_head(&inode->i_hash, head);
1456 spin_unlock(&inode->i_lock);
1457 spin_unlock(&inode_hash_lock);
1461 spin_unlock(&old->i_lock);
1462 spin_unlock(&inode_hash_lock);
1464 if (unlikely(!inode_unhashed(old))) {
1471 EXPORT_SYMBOL(insert_inode_locked4);
1474 int generic_delete_inode(struct inode *inode)
1478 EXPORT_SYMBOL(generic_delete_inode);
1481 * Called when we're dropping the last reference
1484 * Call the FS "drop_inode()" function, defaulting to
1485 * the legacy UNIX filesystem behaviour. If it tells
1486 * us to evict inode, do so. Otherwise, retain inode
1487 * in cache if fs is alive, sync and evict if fs is
1490 static void iput_final(struct inode *inode)
1492 struct super_block *sb = inode->i_sb;
1493 const struct super_operations *op = inode->i_sb->s_op;
1496 WARN_ON(inode->i_state & I_NEW);
1499 drop = op->drop_inode(inode);
1501 drop = generic_drop_inode(inode);
1503 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1504 inode_add_lru(inode);
1505 spin_unlock(&inode->i_lock);
1510 inode->i_state |= I_WILL_FREE;
1511 spin_unlock(&inode->i_lock);
1512 write_inode_now(inode, 1);
1513 spin_lock(&inode->i_lock);
1514 WARN_ON(inode->i_state & I_NEW);
1515 inode->i_state &= ~I_WILL_FREE;
1518 inode->i_state |= I_FREEING;
1519 if (!list_empty(&inode->i_lru))
1520 inode_lru_list_del(inode);
1521 spin_unlock(&inode->i_lock);
1527 * iput - put an inode
1528 * @inode: inode to put
1530 * Puts an inode, dropping its usage count. If the inode use count hits
1531 * zero, the inode is then freed and may also be destroyed.
1533 * Consequently, iput() can sleep.
1535 void iput(struct inode *inode)
1539 BUG_ON(inode->i_state & I_CLEAR);
1541 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1542 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1543 atomic_inc(&inode->i_count);
1544 inode->i_state &= ~I_DIRTY_TIME;
1545 spin_unlock(&inode->i_lock);
1546 trace_writeback_lazytime_iput(inode);
1547 mark_inode_dirty_sync(inode);
1553 EXPORT_SYMBOL(iput);
1556 * bmap - find a block number in a file
1557 * @inode: inode of file
1558 * @block: block to find
1560 * Returns the block number on the device holding the inode that
1561 * is the disk block number for the block of the file requested.
1562 * That is, asked for block 4 of inode 1 the function will return the
1563 * disk block relative to the disk start that holds that block of the
1566 sector_t bmap(struct inode *inode, sector_t block)
1569 if (inode->i_mapping->a_ops->bmap)
1570 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1573 EXPORT_SYMBOL(bmap);
1576 * Update times in overlayed inode from underlying real inode
1578 static void update_ovl_inode_times(struct dentry *dentry, struct inode *inode,
1581 struct dentry *upperdentry;
1584 * Nothing to do if in rcu or if non-overlayfs
1586 if (rcu || likely(!(dentry->d_flags & DCACHE_OP_REAL)))
1589 upperdentry = d_real(dentry, NULL, 0, D_REAL_UPPER);
1592 * If file is on lower then we can't update atime, so no worries about
1593 * stale mtime/ctime.
1596 struct inode *realinode = d_inode(upperdentry);
1598 if ((!timespec_equal(&inode->i_mtime, &realinode->i_mtime) ||
1599 !timespec_equal(&inode->i_ctime, &realinode->i_ctime))) {
1600 inode->i_mtime = realinode->i_mtime;
1601 inode->i_ctime = realinode->i_ctime;
1607 * With relative atime, only update atime if the previous atime is
1608 * earlier than either the ctime or mtime or if at least a day has
1609 * passed since the last atime update.
1611 static int relatime_need_update(const struct path *path, struct inode *inode,
1612 struct timespec now, bool rcu)
1615 if (!(path->mnt->mnt_flags & MNT_RELATIME))
1618 update_ovl_inode_times(path->dentry, inode, rcu);
1620 * Is mtime younger than atime? If yes, update atime:
1622 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1625 * Is ctime younger than atime? If yes, update atime:
1627 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1631 * Is the previous atime value older than a day? If yes,
1634 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1637 * Good, we can skip the atime update:
1642 int generic_update_time(struct inode *inode, struct timespec *time, int flags)
1644 int iflags = I_DIRTY_TIME;
1646 if (flags & S_ATIME)
1647 inode->i_atime = *time;
1648 if (flags & S_VERSION)
1649 inode_inc_iversion(inode);
1650 if (flags & S_CTIME)
1651 inode->i_ctime = *time;
1652 if (flags & S_MTIME)
1653 inode->i_mtime = *time;
1655 if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION))
1656 iflags |= I_DIRTY_SYNC;
1657 __mark_inode_dirty(inode, iflags);
1660 EXPORT_SYMBOL(generic_update_time);
1663 * This does the actual work of updating an inodes time or version. Must have
1664 * had called mnt_want_write() before calling this.
1666 static int update_time(struct inode *inode, struct timespec *time, int flags)
1668 int (*update_time)(struct inode *, struct timespec *, int);
1670 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1671 generic_update_time;
1673 return update_time(inode, time, flags);
1677 * touch_atime - update the access time
1678 * @path: the &struct path to update
1679 * @inode: inode to update
1681 * Update the accessed time on an inode and mark it for writeback.
1682 * This function automatically handles read only file systems and media,
1683 * as well as the "noatime" flag and inode specific "noatime" markers.
1685 bool __atime_needs_update(const struct path *path, struct inode *inode,
1688 struct vfsmount *mnt = path->mnt;
1689 struct timespec now;
1691 if (inode->i_flags & S_NOATIME)
1694 /* Atime updates will likely cause i_uid and i_gid to be written
1695 * back improprely if their true value is unknown to the vfs.
1697 if (HAS_UNMAPPED_ID(inode))
1700 if (IS_NOATIME(inode))
1702 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1705 if (mnt->mnt_flags & MNT_NOATIME)
1707 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1710 now = current_time(inode);
1712 if (!relatime_need_update(path, inode, now, rcu))
1715 if (timespec_equal(&inode->i_atime, &now))
1721 void touch_atime(const struct path *path)
1723 struct vfsmount *mnt = path->mnt;
1724 struct inode *inode = d_inode(path->dentry);
1725 struct timespec now;
1727 if (!__atime_needs_update(path, inode, false))
1730 if (!sb_start_write_trylock(inode->i_sb))
1733 if (__mnt_want_write(mnt) != 0)
1736 * File systems can error out when updating inodes if they need to
1737 * allocate new space to modify an inode (such is the case for
1738 * Btrfs), but since we touch atime while walking down the path we
1739 * really don't care if we failed to update the atime of the file,
1740 * so just ignore the return value.
1741 * We may also fail on filesystems that have the ability to make parts
1742 * of the fs read only, e.g. subvolumes in Btrfs.
1744 now = current_time(inode);
1745 update_time(inode, &now, S_ATIME);
1746 __mnt_drop_write(mnt);
1748 sb_end_write(inode->i_sb);
1750 EXPORT_SYMBOL(touch_atime);
1753 * The logic we want is
1755 * if suid or (sgid and xgrp)
1758 int should_remove_suid(struct dentry *dentry)
1760 umode_t mode = d_inode(dentry)->i_mode;
1763 /* suid always must be killed */
1764 if (unlikely(mode & S_ISUID))
1765 kill = ATTR_KILL_SUID;
1768 * sgid without any exec bits is just a mandatory locking mark; leave
1769 * it alone. If some exec bits are set, it's a real sgid; kill it.
1771 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1772 kill |= ATTR_KILL_SGID;
1774 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1779 EXPORT_SYMBOL(should_remove_suid);
1782 * Return mask of changes for notify_change() that need to be done as a
1783 * response to write or truncate. Return 0 if nothing has to be changed.
1784 * Negative value on error (change should be denied).
1786 int dentry_needs_remove_privs(struct dentry *dentry)
1788 struct inode *inode = d_inode(dentry);
1792 if (IS_NOSEC(inode))
1795 mask = should_remove_suid(dentry);
1796 ret = security_inode_need_killpriv(dentry);
1800 mask |= ATTR_KILL_PRIV;
1804 static int __remove_privs(struct dentry *dentry, int kill)
1806 struct iattr newattrs;
1808 newattrs.ia_valid = ATTR_FORCE | kill;
1810 * Note we call this on write, so notify_change will not
1811 * encounter any conflicting delegations:
1813 return notify_change(dentry, &newattrs, NULL);
1817 * Remove special file priviledges (suid, capabilities) when file is written
1820 int file_remove_privs(struct file *file)
1822 struct dentry *dentry = file_dentry(file);
1823 struct inode *inode = file_inode(file);
1828 * Fast path for nothing security related.
1829 * As well for non-regular files, e.g. blkdev inodes.
1830 * For example, blkdev_write_iter() might get here
1831 * trying to remove privs which it is not allowed to.
1833 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1836 kill = dentry_needs_remove_privs(dentry);
1840 error = __remove_privs(dentry, kill);
1842 inode_has_no_xattr(inode);
1846 EXPORT_SYMBOL(file_remove_privs);
1849 * file_update_time - update mtime and ctime time
1850 * @file: file accessed
1852 * Update the mtime and ctime members of an inode and mark the inode
1853 * for writeback. Note that this function is meant exclusively for
1854 * usage in the file write path of filesystems, and filesystems may
1855 * choose to explicitly ignore update via this function with the
1856 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1857 * timestamps are handled by the server. This can return an error for
1858 * file systems who need to allocate space in order to update an inode.
1861 int file_update_time(struct file *file)
1863 struct inode *inode = file_inode(file);
1864 struct timespec now;
1868 /* First try to exhaust all avenues to not sync */
1869 if (IS_NOCMTIME(inode))
1872 now = current_time(inode);
1873 if (!timespec_equal(&inode->i_mtime, &now))
1876 if (!timespec_equal(&inode->i_ctime, &now))
1879 if (IS_I_VERSION(inode))
1880 sync_it |= S_VERSION;
1885 /* Finally allowed to write? Takes lock. */
1886 if (__mnt_want_write_file(file))
1889 ret = update_time(inode, &now, sync_it);
1890 __mnt_drop_write_file(file);
1894 EXPORT_SYMBOL(file_update_time);
1896 int inode_needs_sync(struct inode *inode)
1900 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1904 EXPORT_SYMBOL(inode_needs_sync);
1907 * If we try to find an inode in the inode hash while it is being
1908 * deleted, we have to wait until the filesystem completes its
1909 * deletion before reporting that it isn't found. This function waits
1910 * until the deletion _might_ have completed. Callers are responsible
1911 * to recheck inode state.
1913 * It doesn't matter if I_NEW is not set initially, a call to
1914 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1917 static void __wait_on_freeing_inode(struct inode *inode)
1919 wait_queue_head_t *wq;
1920 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1921 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1922 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1923 spin_unlock(&inode->i_lock);
1924 spin_unlock(&inode_hash_lock);
1926 finish_wait(wq, &wait.wq_entry);
1927 spin_lock(&inode_hash_lock);
1930 static __initdata unsigned long ihash_entries;
1931 static int __init set_ihash_entries(char *str)
1935 ihash_entries = simple_strtoul(str, &str, 0);
1938 __setup("ihash_entries=", set_ihash_entries);
1941 * Initialize the waitqueues and inode hash table.
1943 void __init inode_init_early(void)
1945 /* If hashes are distributed across NUMA nodes, defer
1946 * hash allocation until vmalloc space is available.
1952 alloc_large_system_hash("Inode-cache",
1953 sizeof(struct hlist_head),
1956 HASH_EARLY | HASH_ZERO,
1963 void __init inode_init(void)
1965 /* inode slab cache */
1966 inode_cachep = kmem_cache_create("inode_cache",
1967 sizeof(struct inode),
1969 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1970 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1973 /* Hash may have been set up in inode_init_early */
1978 alloc_large_system_hash("Inode-cache",
1979 sizeof(struct hlist_head),
1989 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1991 inode->i_mode = mode;
1992 if (S_ISCHR(mode)) {
1993 inode->i_fop = &def_chr_fops;
1994 inode->i_rdev = rdev;
1995 } else if (S_ISBLK(mode)) {
1996 inode->i_fop = &def_blk_fops;
1997 inode->i_rdev = rdev;
1998 } else if (S_ISFIFO(mode))
1999 inode->i_fop = &pipefifo_fops;
2000 else if (S_ISSOCK(mode))
2001 ; /* leave it no_open_fops */
2003 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2004 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2007 EXPORT_SYMBOL(init_special_inode);
2010 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2012 * @dir: Directory inode
2013 * @mode: mode of the new inode
2015 void inode_init_owner(struct inode *inode, const struct inode *dir,
2018 inode->i_uid = current_fsuid();
2019 if (dir && dir->i_mode & S_ISGID) {
2020 inode->i_gid = dir->i_gid;
2022 /* Directories are special, and always inherit S_ISGID */
2025 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2026 !in_group_p(inode->i_gid) &&
2027 !capable_wrt_inode_uidgid(dir, CAP_FSETID))
2030 inode->i_gid = current_fsgid();
2031 inode->i_mode = mode;
2033 EXPORT_SYMBOL(inode_init_owner);
2036 * inode_owner_or_capable - check current task permissions to inode
2037 * @inode: inode being checked
2039 * Return true if current either has CAP_FOWNER in a namespace with the
2040 * inode owner uid mapped, or owns the file.
2042 bool inode_owner_or_capable(const struct inode *inode)
2044 struct user_namespace *ns;
2046 if (uid_eq(current_fsuid(), inode->i_uid))
2049 ns = current_user_ns();
2050 if (kuid_has_mapping(ns, inode->i_uid) && ns_capable(ns, CAP_FOWNER))
2054 EXPORT_SYMBOL(inode_owner_or_capable);
2057 * Direct i/o helper functions
2059 static void __inode_dio_wait(struct inode *inode)
2061 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2062 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2065 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2066 if (atomic_read(&inode->i_dio_count))
2068 } while (atomic_read(&inode->i_dio_count));
2069 finish_wait(wq, &q.wq_entry);
2073 * inode_dio_wait - wait for outstanding DIO requests to finish
2074 * @inode: inode to wait for
2076 * Waits for all pending direct I/O requests to finish so that we can
2077 * proceed with a truncate or equivalent operation.
2079 * Must be called under a lock that serializes taking new references
2080 * to i_dio_count, usually by inode->i_mutex.
2082 void inode_dio_wait(struct inode *inode)
2084 if (atomic_read(&inode->i_dio_count))
2085 __inode_dio_wait(inode);
2087 EXPORT_SYMBOL(inode_dio_wait);
2090 * inode_set_flags - atomically set some inode flags
2092 * Note: the caller should be holding i_mutex, or else be sure that
2093 * they have exclusive access to the inode structure (i.e., while the
2094 * inode is being instantiated). The reason for the cmpxchg() loop
2095 * --- which wouldn't be necessary if all code paths which modify
2096 * i_flags actually followed this rule, is that there is at least one
2097 * code path which doesn't today so we use cmpxchg() out of an abundance
2100 * In the long run, i_mutex is overkill, and we should probably look
2101 * at using the i_lock spinlock to protect i_flags, and then make sure
2102 * it is so documented in include/linux/fs.h and that all code follows
2103 * the locking convention!!
2105 void inode_set_flags(struct inode *inode, unsigned int flags,
2108 unsigned int old_flags, new_flags;
2110 WARN_ON_ONCE(flags & ~mask);
2112 old_flags = ACCESS_ONCE(inode->i_flags);
2113 new_flags = (old_flags & ~mask) | flags;
2114 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2115 new_flags) != old_flags));
2117 EXPORT_SYMBOL(inode_set_flags);
2119 void inode_nohighmem(struct inode *inode)
2121 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2123 EXPORT_SYMBOL(inode_nohighmem);
2126 * current_time - Return FS time
2129 * Return the current time truncated to the time granularity supported by
2132 * Note that inode and inode->sb cannot be NULL.
2133 * Otherwise, the function warns and returns time without truncation.
2135 struct timespec current_time(struct inode *inode)
2137 struct timespec now = current_kernel_time();
2139 if (unlikely(!inode->i_sb)) {
2140 WARN(1, "current_time() called with uninitialized super_block in the inode");
2144 return timespec_trunc(now, inode->i_sb->s_time_gran);
2146 EXPORT_SYMBOL(current_time);