1 // SPDX-License-Identifier: GPL-2.0-only
3 * (C) 1997 Linus Torvalds
4 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
6 #include <linux/export.h>
9 #include <linux/backing-dev.h>
10 #include <linux/hash.h>
11 #include <linux/swap.h>
12 #include <linux/security.h>
13 #include <linux/cdev.h>
14 #include <linux/memblock.h>
15 #include <linux/fscrypt.h>
16 #include <linux/fsnotify.h>
17 #include <linux/mount.h>
18 #include <linux/posix_acl.h>
19 #include <linux/prefetch.h>
20 #include <linux/buffer_head.h> /* for inode_has_buffers */
21 #include <linux/ratelimit.h>
22 #include <linux/list_lru.h>
23 #include <linux/iversion.h>
24 #include <trace/events/writeback.h>
28 * Inode locking rules:
30 * inode->i_lock protects:
31 * inode->i_state, inode->i_hash, __iget()
32 * Inode LRU list locks protect:
33 * inode->i_sb->s_inode_lru, inode->i_lru
34 * inode->i_sb->s_inode_list_lock protects:
35 * inode->i_sb->s_inodes, inode->i_sb_list
36 * bdi->wb.list_lock protects:
37 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
38 * inode_hash_lock protects:
39 * inode_hashtable, inode->i_hash
43 * inode->i_sb->s_inode_list_lock
45 * Inode LRU list locks
51 * inode->i_sb->s_inode_list_lock
58 static unsigned int i_hash_mask __read_mostly;
59 static unsigned int i_hash_shift __read_mostly;
60 static struct hlist_head *inode_hashtable __read_mostly;
61 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
64 * Empty aops. Can be used for the cases where the user does not
65 * define any of the address_space operations.
67 const struct address_space_operations empty_aops = {
69 EXPORT_SYMBOL(empty_aops);
72 * Statistics gathering..
74 struct inodes_stat_t inodes_stat;
76 static DEFINE_PER_CPU(unsigned long, nr_inodes);
77 static DEFINE_PER_CPU(unsigned long, nr_unused);
79 static struct kmem_cache *inode_cachep __read_mostly;
81 static long get_nr_inodes(void)
85 for_each_possible_cpu(i)
86 sum += per_cpu(nr_inodes, i);
87 return sum < 0 ? 0 : sum;
90 static inline long get_nr_inodes_unused(void)
94 for_each_possible_cpu(i)
95 sum += per_cpu(nr_unused, i);
96 return sum < 0 ? 0 : sum;
99 long get_nr_dirty_inodes(void)
101 /* not actually dirty inodes, but a wild approximation */
102 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
103 return nr_dirty > 0 ? nr_dirty : 0;
107 * Handle nr_inode sysctl
110 int proc_nr_inodes(struct ctl_table *table, int write,
111 void *buffer, size_t *lenp, loff_t *ppos)
113 inodes_stat.nr_inodes = get_nr_inodes();
114 inodes_stat.nr_unused = get_nr_inodes_unused();
115 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
119 static int no_open(struct inode *inode, struct file *file)
125 * inode_init_always - perform inode structure initialisation
126 * @sb: superblock inode belongs to
127 * @inode: inode to initialise
129 * These are initializations that need to be done on every inode
130 * allocation as the fields are not initialised by slab allocation.
132 int inode_init_always(struct super_block *sb, struct inode *inode)
134 static const struct inode_operations empty_iops;
135 static const struct file_operations no_open_fops = {.open = no_open};
136 struct address_space *const mapping = &inode->i_data;
139 inode->i_blkbits = sb->s_blocksize_bits;
141 atomic64_set(&inode->i_sequence, 0);
142 atomic_set(&inode->i_count, 1);
143 inode->i_op = &empty_iops;
144 inode->i_fop = &no_open_fops;
145 inode->__i_nlink = 1;
146 inode->i_opflags = 0;
148 inode->i_opflags |= IOP_XATTR;
149 i_uid_write(inode, 0);
150 i_gid_write(inode, 0);
151 atomic_set(&inode->i_writecount, 0);
153 inode->i_write_hint = WRITE_LIFE_NOT_SET;
156 inode->i_generation = 0;
157 inode->i_pipe = NULL;
158 inode->i_bdev = NULL;
159 inode->i_cdev = NULL;
160 inode->i_link = NULL;
161 inode->i_dir_seq = 0;
163 inode->dirtied_when = 0;
165 #ifdef CONFIG_CGROUP_WRITEBACK
166 inode->i_wb_frn_winner = 0;
167 inode->i_wb_frn_avg_time = 0;
168 inode->i_wb_frn_history = 0;
171 spin_lock_init(&inode->i_lock);
172 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
174 init_rwsem(&inode->i_rwsem);
175 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
177 atomic_set(&inode->i_dio_count, 0);
179 mapping->a_ops = &empty_aops;
180 mapping->host = inode;
182 if (sb->s_type->fs_flags & FS_THP_SUPPORT)
183 __set_bit(AS_THP_SUPPORT, &mapping->flags);
185 atomic_set(&mapping->i_mmap_writable, 0);
186 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
187 atomic_set(&mapping->nr_thps, 0);
189 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
190 mapping->private_data = NULL;
191 mapping->writeback_index = 0;
192 inode->i_private = NULL;
193 inode->i_mapping = mapping;
194 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
195 #ifdef CONFIG_FS_POSIX_ACL
196 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
199 #ifdef CONFIG_FSNOTIFY
200 inode->i_fsnotify_mask = 0;
202 inode->i_flctx = NULL;
204 if (unlikely(security_inode_alloc(inode)))
206 this_cpu_inc(nr_inodes);
210 EXPORT_SYMBOL(inode_init_always);
212 void free_inode_nonrcu(struct inode *inode)
214 kmem_cache_free(inode_cachep, inode);
216 EXPORT_SYMBOL(free_inode_nonrcu);
218 static void i_callback(struct rcu_head *head)
220 struct inode *inode = container_of(head, struct inode, i_rcu);
221 if (inode->free_inode)
222 inode->free_inode(inode);
224 free_inode_nonrcu(inode);
227 static struct inode *alloc_inode(struct super_block *sb)
229 const struct super_operations *ops = sb->s_op;
232 if (ops->alloc_inode)
233 inode = ops->alloc_inode(sb);
235 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
240 if (unlikely(inode_init_always(sb, inode))) {
241 if (ops->destroy_inode) {
242 ops->destroy_inode(inode);
243 if (!ops->free_inode)
246 inode->free_inode = ops->free_inode;
247 i_callback(&inode->i_rcu);
254 void __destroy_inode(struct inode *inode)
256 BUG_ON(inode_has_buffers(inode));
257 inode_detach_wb(inode);
258 security_inode_free(inode);
259 fsnotify_inode_delete(inode);
260 locks_free_lock_context(inode);
261 if (!inode->i_nlink) {
262 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
263 atomic_long_dec(&inode->i_sb->s_remove_count);
266 #ifdef CONFIG_FS_POSIX_ACL
267 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
268 posix_acl_release(inode->i_acl);
269 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
270 posix_acl_release(inode->i_default_acl);
272 this_cpu_dec(nr_inodes);
274 EXPORT_SYMBOL(__destroy_inode);
276 static void destroy_inode(struct inode *inode)
278 const struct super_operations *ops = inode->i_sb->s_op;
280 BUG_ON(!list_empty(&inode->i_lru));
281 __destroy_inode(inode);
282 if (ops->destroy_inode) {
283 ops->destroy_inode(inode);
284 if (!ops->free_inode)
287 inode->free_inode = ops->free_inode;
288 call_rcu(&inode->i_rcu, i_callback);
292 * drop_nlink - directly drop an inode's link count
295 * This is a low-level filesystem helper to replace any
296 * direct filesystem manipulation of i_nlink. In cases
297 * where we are attempting to track writes to the
298 * filesystem, a decrement to zero means an imminent
299 * write when the file is truncated and actually unlinked
302 void drop_nlink(struct inode *inode)
304 WARN_ON(inode->i_nlink == 0);
307 atomic_long_inc(&inode->i_sb->s_remove_count);
309 EXPORT_SYMBOL(drop_nlink);
312 * clear_nlink - directly zero an inode's link count
315 * This is a low-level filesystem helper to replace any
316 * direct filesystem manipulation of i_nlink. See
317 * drop_nlink() for why we care about i_nlink hitting zero.
319 void clear_nlink(struct inode *inode)
321 if (inode->i_nlink) {
322 inode->__i_nlink = 0;
323 atomic_long_inc(&inode->i_sb->s_remove_count);
326 EXPORT_SYMBOL(clear_nlink);
329 * set_nlink - directly set an inode's link count
331 * @nlink: new nlink (should be non-zero)
333 * This is a low-level filesystem helper to replace any
334 * direct filesystem manipulation of i_nlink.
336 void set_nlink(struct inode *inode, unsigned int nlink)
341 /* Yes, some filesystems do change nlink from zero to one */
342 if (inode->i_nlink == 0)
343 atomic_long_dec(&inode->i_sb->s_remove_count);
345 inode->__i_nlink = nlink;
348 EXPORT_SYMBOL(set_nlink);
351 * inc_nlink - directly increment an inode's link count
354 * This is a low-level filesystem helper to replace any
355 * direct filesystem manipulation of i_nlink. Currently,
356 * it is only here for parity with dec_nlink().
358 void inc_nlink(struct inode *inode)
360 if (unlikely(inode->i_nlink == 0)) {
361 WARN_ON(!(inode->i_state & I_LINKABLE));
362 atomic_long_dec(&inode->i_sb->s_remove_count);
367 EXPORT_SYMBOL(inc_nlink);
369 static void __address_space_init_once(struct address_space *mapping)
371 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
372 init_rwsem(&mapping->i_mmap_rwsem);
373 INIT_LIST_HEAD(&mapping->private_list);
374 spin_lock_init(&mapping->private_lock);
375 mapping->i_mmap = RB_ROOT_CACHED;
378 void address_space_init_once(struct address_space *mapping)
380 memset(mapping, 0, sizeof(*mapping));
381 __address_space_init_once(mapping);
383 EXPORT_SYMBOL(address_space_init_once);
386 * These are initializations that only need to be done
387 * once, because the fields are idempotent across use
388 * of the inode, so let the slab aware of that.
390 void inode_init_once(struct inode *inode)
392 memset(inode, 0, sizeof(*inode));
393 INIT_HLIST_NODE(&inode->i_hash);
394 INIT_LIST_HEAD(&inode->i_devices);
395 INIT_LIST_HEAD(&inode->i_io_list);
396 INIT_LIST_HEAD(&inode->i_wb_list);
397 INIT_LIST_HEAD(&inode->i_lru);
398 __address_space_init_once(&inode->i_data);
399 i_size_ordered_init(inode);
401 EXPORT_SYMBOL(inode_init_once);
403 static void init_once(void *foo)
405 struct inode *inode = (struct inode *) foo;
407 inode_init_once(inode);
411 * inode->i_lock must be held
413 void __iget(struct inode *inode)
415 atomic_inc(&inode->i_count);
419 * get additional reference to inode; caller must already hold one.
421 void ihold(struct inode *inode)
423 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
425 EXPORT_SYMBOL(ihold);
427 static void inode_lru_list_add(struct inode *inode)
429 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
430 this_cpu_inc(nr_unused);
432 inode->i_state |= I_REFERENCED;
436 * Add inode to LRU if needed (inode is unused and clean).
438 * Needs inode->i_lock held.
440 void inode_add_lru(struct inode *inode)
442 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
443 I_FREEING | I_WILL_FREE)) &&
444 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
445 inode_lru_list_add(inode);
449 static void inode_lru_list_del(struct inode *inode)
452 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
453 this_cpu_dec(nr_unused);
457 * inode_sb_list_add - add inode to the superblock list of inodes
458 * @inode: inode to add
460 void inode_sb_list_add(struct inode *inode)
462 spin_lock(&inode->i_sb->s_inode_list_lock);
463 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
464 spin_unlock(&inode->i_sb->s_inode_list_lock);
466 EXPORT_SYMBOL_GPL(inode_sb_list_add);
468 static inline void inode_sb_list_del(struct inode *inode)
470 if (!list_empty(&inode->i_sb_list)) {
471 spin_lock(&inode->i_sb->s_inode_list_lock);
472 list_del_init(&inode->i_sb_list);
473 spin_unlock(&inode->i_sb->s_inode_list_lock);
477 static unsigned long hash(struct super_block *sb, unsigned long hashval)
481 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
483 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
484 return tmp & i_hash_mask;
488 * __insert_inode_hash - hash an inode
489 * @inode: unhashed inode
490 * @hashval: unsigned long value used to locate this object in the
493 * Add an inode to the inode hash for this superblock.
495 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
497 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
499 spin_lock(&inode_hash_lock);
500 spin_lock(&inode->i_lock);
501 hlist_add_head_rcu(&inode->i_hash, b);
502 spin_unlock(&inode->i_lock);
503 spin_unlock(&inode_hash_lock);
505 EXPORT_SYMBOL(__insert_inode_hash);
508 * __remove_inode_hash - remove an inode from the hash
509 * @inode: inode to unhash
511 * Remove an inode from the superblock.
513 void __remove_inode_hash(struct inode *inode)
515 spin_lock(&inode_hash_lock);
516 spin_lock(&inode->i_lock);
517 hlist_del_init_rcu(&inode->i_hash);
518 spin_unlock(&inode->i_lock);
519 spin_unlock(&inode_hash_lock);
521 EXPORT_SYMBOL(__remove_inode_hash);
523 void clear_inode(struct inode *inode)
526 * We have to cycle the i_pages lock here because reclaim can be in the
527 * process of removing the last page (in __delete_from_page_cache())
528 * and we must not free the mapping under it.
530 xa_lock_irq(&inode->i_data.i_pages);
531 BUG_ON(inode->i_data.nrpages);
532 BUG_ON(inode->i_data.nrexceptional);
533 xa_unlock_irq(&inode->i_data.i_pages);
534 BUG_ON(!list_empty(&inode->i_data.private_list));
535 BUG_ON(!(inode->i_state & I_FREEING));
536 BUG_ON(inode->i_state & I_CLEAR);
537 BUG_ON(!list_empty(&inode->i_wb_list));
538 /* don't need i_lock here, no concurrent mods to i_state */
539 inode->i_state = I_FREEING | I_CLEAR;
541 EXPORT_SYMBOL(clear_inode);
544 * Free the inode passed in, removing it from the lists it is still connected
545 * to. We remove any pages still attached to the inode and wait for any IO that
546 * is still in progress before finally destroying the inode.
548 * An inode must already be marked I_FREEING so that we avoid the inode being
549 * moved back onto lists if we race with other code that manipulates the lists
550 * (e.g. writeback_single_inode). The caller is responsible for setting this.
552 * An inode must already be removed from the LRU list before being evicted from
553 * the cache. This should occur atomically with setting the I_FREEING state
554 * flag, so no inodes here should ever be on the LRU when being evicted.
556 static void evict(struct inode *inode)
558 const struct super_operations *op = inode->i_sb->s_op;
560 BUG_ON(!(inode->i_state & I_FREEING));
561 BUG_ON(!list_empty(&inode->i_lru));
563 if (!list_empty(&inode->i_io_list))
564 inode_io_list_del(inode);
566 inode_sb_list_del(inode);
569 * Wait for flusher thread to be done with the inode so that filesystem
570 * does not start destroying it while writeback is still running. Since
571 * the inode has I_FREEING set, flusher thread won't start new work on
572 * the inode. We just have to wait for running writeback to finish.
574 inode_wait_for_writeback(inode);
576 if (op->evict_inode) {
577 op->evict_inode(inode);
579 truncate_inode_pages_final(&inode->i_data);
582 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
584 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
587 remove_inode_hash(inode);
589 spin_lock(&inode->i_lock);
590 wake_up_bit(&inode->i_state, __I_NEW);
591 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
592 spin_unlock(&inode->i_lock);
594 destroy_inode(inode);
598 * dispose_list - dispose of the contents of a local list
599 * @head: the head of the list to free
601 * Dispose-list gets a local list with local inodes in it, so it doesn't
602 * need to worry about list corruption and SMP locks.
604 static void dispose_list(struct list_head *head)
606 while (!list_empty(head)) {
609 inode = list_first_entry(head, struct inode, i_lru);
610 list_del_init(&inode->i_lru);
618 * evict_inodes - evict all evictable inodes for a superblock
619 * @sb: superblock to operate on
621 * Make sure that no inodes with zero refcount are retained. This is
622 * called by superblock shutdown after having SB_ACTIVE flag removed,
623 * so any inode reaching zero refcount during or after that call will
624 * be immediately evicted.
626 void evict_inodes(struct super_block *sb)
628 struct inode *inode, *next;
632 spin_lock(&sb->s_inode_list_lock);
633 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
634 if (atomic_read(&inode->i_count))
637 spin_lock(&inode->i_lock);
638 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
639 spin_unlock(&inode->i_lock);
643 inode->i_state |= I_FREEING;
644 inode_lru_list_del(inode);
645 spin_unlock(&inode->i_lock);
646 list_add(&inode->i_lru, &dispose);
649 * We can have a ton of inodes to evict at unmount time given
650 * enough memory, check to see if we need to go to sleep for a
651 * bit so we don't livelock.
653 if (need_resched()) {
654 spin_unlock(&sb->s_inode_list_lock);
656 dispose_list(&dispose);
660 spin_unlock(&sb->s_inode_list_lock);
662 dispose_list(&dispose);
664 EXPORT_SYMBOL_GPL(evict_inodes);
667 * invalidate_inodes - attempt to free all inodes on a superblock
668 * @sb: superblock to operate on
669 * @kill_dirty: flag to guide handling of dirty inodes
671 * Attempts to free all inodes for a given superblock. If there were any
672 * busy inodes return a non-zero value, else zero.
673 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
676 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
679 struct inode *inode, *next;
683 spin_lock(&sb->s_inode_list_lock);
684 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
685 spin_lock(&inode->i_lock);
686 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
687 spin_unlock(&inode->i_lock);
690 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
691 spin_unlock(&inode->i_lock);
695 if (atomic_read(&inode->i_count)) {
696 spin_unlock(&inode->i_lock);
701 inode->i_state |= I_FREEING;
702 inode_lru_list_del(inode);
703 spin_unlock(&inode->i_lock);
704 list_add(&inode->i_lru, &dispose);
705 if (need_resched()) {
706 spin_unlock(&sb->s_inode_list_lock);
708 dispose_list(&dispose);
712 spin_unlock(&sb->s_inode_list_lock);
714 dispose_list(&dispose);
720 * Isolate the inode from the LRU in preparation for freeing it.
722 * Any inodes which are pinned purely because of attached pagecache have their
723 * pagecache removed. If the inode has metadata buffers attached to
724 * mapping->private_list then try to remove them.
726 * If the inode has the I_REFERENCED flag set, then it means that it has been
727 * used recently - the flag is set in iput_final(). When we encounter such an
728 * inode, clear the flag and move it to the back of the LRU so it gets another
729 * pass through the LRU before it gets reclaimed. This is necessary because of
730 * the fact we are doing lazy LRU updates to minimise lock contention so the
731 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
732 * with this flag set because they are the inodes that are out of order.
734 static enum lru_status inode_lru_isolate(struct list_head *item,
735 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
737 struct list_head *freeable = arg;
738 struct inode *inode = container_of(item, struct inode, i_lru);
741 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
742 * If we fail to get the lock, just skip it.
744 if (!spin_trylock(&inode->i_lock))
748 * Referenced or dirty inodes are still in use. Give them another pass
749 * through the LRU as we canot reclaim them now.
751 if (atomic_read(&inode->i_count) ||
752 (inode->i_state & ~I_REFERENCED)) {
753 list_lru_isolate(lru, &inode->i_lru);
754 spin_unlock(&inode->i_lock);
755 this_cpu_dec(nr_unused);
759 /* recently referenced inodes get one more pass */
760 if (inode->i_state & I_REFERENCED) {
761 inode->i_state &= ~I_REFERENCED;
762 spin_unlock(&inode->i_lock);
766 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
768 spin_unlock(&inode->i_lock);
769 spin_unlock(lru_lock);
770 if (remove_inode_buffers(inode)) {
772 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
773 if (current_is_kswapd())
774 __count_vm_events(KSWAPD_INODESTEAL, reap);
776 __count_vm_events(PGINODESTEAL, reap);
777 if (current->reclaim_state)
778 current->reclaim_state->reclaimed_slab += reap;
785 WARN_ON(inode->i_state & I_NEW);
786 inode->i_state |= I_FREEING;
787 list_lru_isolate_move(lru, &inode->i_lru, freeable);
788 spin_unlock(&inode->i_lock);
790 this_cpu_dec(nr_unused);
795 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
796 * This is called from the superblock shrinker function with a number of inodes
797 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
798 * then are freed outside inode_lock by dispose_list().
800 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
805 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
806 inode_lru_isolate, &freeable);
807 dispose_list(&freeable);
811 static void __wait_on_freeing_inode(struct inode *inode);
813 * Called with the inode lock held.
815 static struct inode *find_inode(struct super_block *sb,
816 struct hlist_head *head,
817 int (*test)(struct inode *, void *),
820 struct inode *inode = NULL;
823 hlist_for_each_entry(inode, head, i_hash) {
824 if (inode->i_sb != sb)
826 if (!test(inode, data))
828 spin_lock(&inode->i_lock);
829 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
830 __wait_on_freeing_inode(inode);
833 if (unlikely(inode->i_state & I_CREATING)) {
834 spin_unlock(&inode->i_lock);
835 return ERR_PTR(-ESTALE);
838 spin_unlock(&inode->i_lock);
845 * find_inode_fast is the fast path version of find_inode, see the comment at
846 * iget_locked for details.
848 static struct inode *find_inode_fast(struct super_block *sb,
849 struct hlist_head *head, unsigned long ino)
851 struct inode *inode = NULL;
854 hlist_for_each_entry(inode, head, i_hash) {
855 if (inode->i_ino != ino)
857 if (inode->i_sb != sb)
859 spin_lock(&inode->i_lock);
860 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
861 __wait_on_freeing_inode(inode);
864 if (unlikely(inode->i_state & I_CREATING)) {
865 spin_unlock(&inode->i_lock);
866 return ERR_PTR(-ESTALE);
869 spin_unlock(&inode->i_lock);
876 * Each cpu owns a range of LAST_INO_BATCH numbers.
877 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
878 * to renew the exhausted range.
880 * This does not significantly increase overflow rate because every CPU can
881 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
882 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
883 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
884 * overflow rate by 2x, which does not seem too significant.
886 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
887 * error if st_ino won't fit in target struct field. Use 32bit counter
888 * here to attempt to avoid that.
890 #define LAST_INO_BATCH 1024
891 static DEFINE_PER_CPU(unsigned int, last_ino);
893 unsigned int get_next_ino(void)
895 unsigned int *p = &get_cpu_var(last_ino);
896 unsigned int res = *p;
899 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
900 static atomic_t shared_last_ino;
901 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
903 res = next - LAST_INO_BATCH;
908 /* get_next_ino should not provide a 0 inode number */
912 put_cpu_var(last_ino);
915 EXPORT_SYMBOL(get_next_ino);
918 * new_inode_pseudo - obtain an inode
921 * Allocates a new inode for given superblock.
922 * Inode wont be chained in superblock s_inodes list
924 * - fs can't be unmount
925 * - quotas, fsnotify, writeback can't work
927 struct inode *new_inode_pseudo(struct super_block *sb)
929 struct inode *inode = alloc_inode(sb);
932 spin_lock(&inode->i_lock);
934 spin_unlock(&inode->i_lock);
935 INIT_LIST_HEAD(&inode->i_sb_list);
941 * new_inode - obtain an inode
944 * Allocates a new inode for given superblock. The default gfp_mask
945 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
946 * If HIGHMEM pages are unsuitable or it is known that pages allocated
947 * for the page cache are not reclaimable or migratable,
948 * mapping_set_gfp_mask() must be called with suitable flags on the
949 * newly created inode's mapping
952 struct inode *new_inode(struct super_block *sb)
956 spin_lock_prefetch(&sb->s_inode_list_lock);
958 inode = new_inode_pseudo(sb);
960 inode_sb_list_add(inode);
963 EXPORT_SYMBOL(new_inode);
965 #ifdef CONFIG_DEBUG_LOCK_ALLOC
966 void lockdep_annotate_inode_mutex_key(struct inode *inode)
968 if (S_ISDIR(inode->i_mode)) {
969 struct file_system_type *type = inode->i_sb->s_type;
971 /* Set new key only if filesystem hasn't already changed it */
972 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
974 * ensure nobody is actually holding i_mutex
976 // mutex_destroy(&inode->i_mutex);
977 init_rwsem(&inode->i_rwsem);
978 lockdep_set_class(&inode->i_rwsem,
979 &type->i_mutex_dir_key);
983 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
987 * unlock_new_inode - clear the I_NEW state and wake up any waiters
988 * @inode: new inode to unlock
990 * Called when the inode is fully initialised to clear the new state of the
991 * inode and wake up anyone waiting for the inode to finish initialisation.
993 void unlock_new_inode(struct inode *inode)
995 lockdep_annotate_inode_mutex_key(inode);
996 spin_lock(&inode->i_lock);
997 WARN_ON(!(inode->i_state & I_NEW));
998 inode->i_state &= ~I_NEW & ~I_CREATING;
1000 wake_up_bit(&inode->i_state, __I_NEW);
1001 spin_unlock(&inode->i_lock);
1003 EXPORT_SYMBOL(unlock_new_inode);
1005 void discard_new_inode(struct inode *inode)
1007 lockdep_annotate_inode_mutex_key(inode);
1008 spin_lock(&inode->i_lock);
1009 WARN_ON(!(inode->i_state & I_NEW));
1010 inode->i_state &= ~I_NEW;
1012 wake_up_bit(&inode->i_state, __I_NEW);
1013 spin_unlock(&inode->i_lock);
1016 EXPORT_SYMBOL(discard_new_inode);
1019 * lock_two_inodes - lock two inodes (may be regular files but also dirs)
1021 * Lock any non-NULL argument. The caller must make sure that if he is passing
1022 * in two directories, one is not ancestor of the other. Zero, one or two
1023 * objects may be locked by this function.
1025 * @inode1: first inode to lock
1026 * @inode2: second inode to lock
1027 * @subclass1: inode lock subclass for the first lock obtained
1028 * @subclass2: inode lock subclass for the second lock obtained
1030 void lock_two_inodes(struct inode *inode1, struct inode *inode2,
1031 unsigned subclass1, unsigned subclass2)
1033 if (!inode1 || !inode2) {
1035 * Make sure @subclass1 will be used for the acquired lock.
1036 * This is not strictly necessary (no current caller cares) but
1037 * let's keep things consistent.
1040 swap(inode1, inode2);
1045 * If one object is directory and the other is not, we must make sure
1046 * to lock directory first as the other object may be its child.
1048 if (S_ISDIR(inode2->i_mode) == S_ISDIR(inode1->i_mode)) {
1049 if (inode1 > inode2)
1050 swap(inode1, inode2);
1051 } else if (!S_ISDIR(inode1->i_mode))
1052 swap(inode1, inode2);
1055 inode_lock_nested(inode1, subclass1);
1056 if (inode2 && inode2 != inode1)
1057 inode_lock_nested(inode2, subclass2);
1061 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1063 * Lock any non-NULL argument that is not a directory.
1064 * Zero, one or two objects may be locked by this function.
1066 * @inode1: first inode to lock
1067 * @inode2: second inode to lock
1069 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1071 if (inode1 > inode2)
1072 swap(inode1, inode2);
1074 if (inode1 && !S_ISDIR(inode1->i_mode))
1076 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1077 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1079 EXPORT_SYMBOL(lock_two_nondirectories);
1082 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1083 * @inode1: first inode to unlock
1084 * @inode2: second inode to unlock
1086 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1088 if (inode1 && !S_ISDIR(inode1->i_mode))
1089 inode_unlock(inode1);
1090 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1091 inode_unlock(inode2);
1093 EXPORT_SYMBOL(unlock_two_nondirectories);
1096 * inode_insert5 - obtain an inode from a mounted file system
1097 * @inode: pre-allocated inode to use for insert to cache
1098 * @hashval: hash value (usually inode number) to get
1099 * @test: callback used for comparisons between inodes
1100 * @set: callback used to initialize a new struct inode
1101 * @data: opaque data pointer to pass to @test and @set
1103 * Search for the inode specified by @hashval and @data in the inode cache,
1104 * and if present it is return it with an increased reference count. This is
1105 * a variant of iget5_locked() for callers that don't want to fail on memory
1106 * allocation of inode.
1108 * If the inode is not in cache, insert the pre-allocated inode to cache and
1109 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1110 * to fill it in before unlocking it via unlock_new_inode().
1112 * Note both @test and @set are called with the inode_hash_lock held, so can't
1115 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1116 int (*test)(struct inode *, void *),
1117 int (*set)(struct inode *, void *), void *data)
1119 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1121 bool creating = inode->i_state & I_CREATING;
1124 spin_lock(&inode_hash_lock);
1125 old = find_inode(inode->i_sb, head, test, data);
1126 if (unlikely(old)) {
1128 * Uhhuh, somebody else created the same inode under us.
1129 * Use the old inode instead of the preallocated one.
1131 spin_unlock(&inode_hash_lock);
1135 if (unlikely(inode_unhashed(old))) {
1142 if (set && unlikely(set(inode, data))) {
1148 * Return the locked inode with I_NEW set, the
1149 * caller is responsible for filling in the contents
1151 spin_lock(&inode->i_lock);
1152 inode->i_state |= I_NEW;
1153 hlist_add_head_rcu(&inode->i_hash, head);
1154 spin_unlock(&inode->i_lock);
1156 inode_sb_list_add(inode);
1158 spin_unlock(&inode_hash_lock);
1162 EXPORT_SYMBOL(inode_insert5);
1165 * iget5_locked - obtain an inode from a mounted file system
1166 * @sb: super block of file system
1167 * @hashval: hash value (usually inode number) to get
1168 * @test: callback used for comparisons between inodes
1169 * @set: callback used to initialize a new struct inode
1170 * @data: opaque data pointer to pass to @test and @set
1172 * Search for the inode specified by @hashval and @data in the inode cache,
1173 * and if present it is return it with an increased reference count. This is
1174 * a generalized version of iget_locked() for file systems where the inode
1175 * number is not sufficient for unique identification of an inode.
1177 * If the inode is not in cache, allocate a new inode and return it locked,
1178 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1179 * before unlocking it via unlock_new_inode().
1181 * Note both @test and @set are called with the inode_hash_lock held, so can't
1184 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1185 int (*test)(struct inode *, void *),
1186 int (*set)(struct inode *, void *), void *data)
1188 struct inode *inode = ilookup5(sb, hashval, test, data);
1191 struct inode *new = alloc_inode(sb);
1195 inode = inode_insert5(new, hashval, test, set, data);
1196 if (unlikely(inode != new))
1202 EXPORT_SYMBOL(iget5_locked);
1205 * iget_locked - obtain an inode from a mounted file system
1206 * @sb: super block of file system
1207 * @ino: inode number to get
1209 * Search for the inode specified by @ino in the inode cache and if present
1210 * return it with an increased reference count. This is for file systems
1211 * where the inode number is sufficient for unique identification of an inode.
1213 * If the inode is not in cache, allocate a new inode and return it locked,
1214 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1215 * before unlocking it via unlock_new_inode().
1217 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1219 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1220 struct inode *inode;
1222 spin_lock(&inode_hash_lock);
1223 inode = find_inode_fast(sb, head, ino);
1224 spin_unlock(&inode_hash_lock);
1228 wait_on_inode(inode);
1229 if (unlikely(inode_unhashed(inode))) {
1236 inode = alloc_inode(sb);
1240 spin_lock(&inode_hash_lock);
1241 /* We released the lock, so.. */
1242 old = find_inode_fast(sb, head, ino);
1245 spin_lock(&inode->i_lock);
1246 inode->i_state = I_NEW;
1247 hlist_add_head_rcu(&inode->i_hash, head);
1248 spin_unlock(&inode->i_lock);
1249 inode_sb_list_add(inode);
1250 spin_unlock(&inode_hash_lock);
1252 /* Return the locked inode with I_NEW set, the
1253 * caller is responsible for filling in the contents
1259 * Uhhuh, somebody else created the same inode under
1260 * us. Use the old inode instead of the one we just
1263 spin_unlock(&inode_hash_lock);
1264 destroy_inode(inode);
1268 wait_on_inode(inode);
1269 if (unlikely(inode_unhashed(inode))) {
1276 EXPORT_SYMBOL(iget_locked);
1279 * search the inode cache for a matching inode number.
1280 * If we find one, then the inode number we are trying to
1281 * allocate is not unique and so we should not use it.
1283 * Returns 1 if the inode number is unique, 0 if it is not.
1285 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1287 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1288 struct inode *inode;
1290 hlist_for_each_entry_rcu(inode, b, i_hash) {
1291 if (inode->i_ino == ino && inode->i_sb == sb)
1298 * iunique - get a unique inode number
1300 * @max_reserved: highest reserved inode number
1302 * Obtain an inode number that is unique on the system for a given
1303 * superblock. This is used by file systems that have no natural
1304 * permanent inode numbering system. An inode number is returned that
1305 * is higher than the reserved limit but unique.
1308 * With a large number of inodes live on the file system this function
1309 * currently becomes quite slow.
1311 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1314 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1315 * error if st_ino won't fit in target struct field. Use 32bit counter
1316 * here to attempt to avoid that.
1318 static DEFINE_SPINLOCK(iunique_lock);
1319 static unsigned int counter;
1323 spin_lock(&iunique_lock);
1325 if (counter <= max_reserved)
1326 counter = max_reserved + 1;
1328 } while (!test_inode_iunique(sb, res));
1329 spin_unlock(&iunique_lock);
1334 EXPORT_SYMBOL(iunique);
1336 struct inode *igrab(struct inode *inode)
1338 spin_lock(&inode->i_lock);
1339 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1341 spin_unlock(&inode->i_lock);
1343 spin_unlock(&inode->i_lock);
1345 * Handle the case where s_op->clear_inode is not been
1346 * called yet, and somebody is calling igrab
1347 * while the inode is getting freed.
1353 EXPORT_SYMBOL(igrab);
1356 * ilookup5_nowait - search for an inode in the inode cache
1357 * @sb: super block of file system to search
1358 * @hashval: hash value (usually inode number) to search for
1359 * @test: callback used for comparisons between inodes
1360 * @data: opaque data pointer to pass to @test
1362 * Search for the inode specified by @hashval and @data in the inode cache.
1363 * If the inode is in the cache, the inode is returned with an incremented
1366 * Note: I_NEW is not waited upon so you have to be very careful what you do
1367 * with the returned inode. You probably should be using ilookup5() instead.
1369 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1371 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1372 int (*test)(struct inode *, void *), void *data)
1374 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1375 struct inode *inode;
1377 spin_lock(&inode_hash_lock);
1378 inode = find_inode(sb, head, test, data);
1379 spin_unlock(&inode_hash_lock);
1381 return IS_ERR(inode) ? NULL : inode;
1383 EXPORT_SYMBOL(ilookup5_nowait);
1386 * ilookup5 - search for an inode in the inode cache
1387 * @sb: super block of file system to search
1388 * @hashval: hash value (usually inode number) to search for
1389 * @test: callback used for comparisons between inodes
1390 * @data: opaque data pointer to pass to @test
1392 * Search for the inode specified by @hashval and @data in the inode cache,
1393 * and if the inode is in the cache, return the inode with an incremented
1394 * reference count. Waits on I_NEW before returning the inode.
1395 * returned with an incremented reference count.
1397 * This is a generalized version of ilookup() for file systems where the
1398 * inode number is not sufficient for unique identification of an inode.
1400 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1402 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1403 int (*test)(struct inode *, void *), void *data)
1405 struct inode *inode;
1407 inode = ilookup5_nowait(sb, hashval, test, data);
1409 wait_on_inode(inode);
1410 if (unlikely(inode_unhashed(inode))) {
1417 EXPORT_SYMBOL(ilookup5);
1420 * ilookup - search for an inode in the inode cache
1421 * @sb: super block of file system to search
1422 * @ino: inode number to search for
1424 * Search for the inode @ino in the inode cache, and if the inode is in the
1425 * cache, the inode is returned with an incremented reference count.
1427 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1429 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1430 struct inode *inode;
1432 spin_lock(&inode_hash_lock);
1433 inode = find_inode_fast(sb, head, ino);
1434 spin_unlock(&inode_hash_lock);
1439 wait_on_inode(inode);
1440 if (unlikely(inode_unhashed(inode))) {
1447 EXPORT_SYMBOL(ilookup);
1450 * find_inode_nowait - find an inode in the inode cache
1451 * @sb: super block of file system to search
1452 * @hashval: hash value (usually inode number) to search for
1453 * @match: callback used for comparisons between inodes
1454 * @data: opaque data pointer to pass to @match
1456 * Search for the inode specified by @hashval and @data in the inode
1457 * cache, where the helper function @match will return 0 if the inode
1458 * does not match, 1 if the inode does match, and -1 if the search
1459 * should be stopped. The @match function must be responsible for
1460 * taking the i_lock spin_lock and checking i_state for an inode being
1461 * freed or being initialized, and incrementing the reference count
1462 * before returning 1. It also must not sleep, since it is called with
1463 * the inode_hash_lock spinlock held.
1465 * This is a even more generalized version of ilookup5() when the
1466 * function must never block --- find_inode() can block in
1467 * __wait_on_freeing_inode() --- or when the caller can not increment
1468 * the reference count because the resulting iput() might cause an
1469 * inode eviction. The tradeoff is that the @match funtion must be
1470 * very carefully implemented.
1472 struct inode *find_inode_nowait(struct super_block *sb,
1473 unsigned long hashval,
1474 int (*match)(struct inode *, unsigned long,
1478 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1479 struct inode *inode, *ret_inode = NULL;
1482 spin_lock(&inode_hash_lock);
1483 hlist_for_each_entry(inode, head, i_hash) {
1484 if (inode->i_sb != sb)
1486 mval = match(inode, hashval, data);
1494 spin_unlock(&inode_hash_lock);
1497 EXPORT_SYMBOL(find_inode_nowait);
1500 * find_inode_rcu - find an inode in the inode cache
1501 * @sb: Super block of file system to search
1502 * @hashval: Key to hash
1503 * @test: Function to test match on an inode
1504 * @data: Data for test function
1506 * Search for the inode specified by @hashval and @data in the inode cache,
1507 * where the helper function @test will return 0 if the inode does not match
1508 * and 1 if it does. The @test function must be responsible for taking the
1509 * i_lock spin_lock and checking i_state for an inode being freed or being
1512 * If successful, this will return the inode for which the @test function
1513 * returned 1 and NULL otherwise.
1515 * The @test function is not permitted to take a ref on any inode presented.
1516 * It is also not permitted to sleep.
1518 * The caller must hold the RCU read lock.
1520 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1521 int (*test)(struct inode *, void *), void *data)
1523 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1524 struct inode *inode;
1526 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1527 "suspicious find_inode_rcu() usage");
1529 hlist_for_each_entry_rcu(inode, head, i_hash) {
1530 if (inode->i_sb == sb &&
1531 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1537 EXPORT_SYMBOL(find_inode_rcu);
1540 * find_inode_by_rcu - Find an inode in the inode cache
1541 * @sb: Super block of file system to search
1542 * @ino: The inode number to match
1544 * Search for the inode specified by @hashval and @data in the inode cache,
1545 * where the helper function @test will return 0 if the inode does not match
1546 * and 1 if it does. The @test function must be responsible for taking the
1547 * i_lock spin_lock and checking i_state for an inode being freed or being
1550 * If successful, this will return the inode for which the @test function
1551 * returned 1 and NULL otherwise.
1553 * The @test function is not permitted to take a ref on any inode presented.
1554 * It is also not permitted to sleep.
1556 * The caller must hold the RCU read lock.
1558 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1561 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1562 struct inode *inode;
1564 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1565 "suspicious find_inode_by_ino_rcu() usage");
1567 hlist_for_each_entry_rcu(inode, head, i_hash) {
1568 if (inode->i_ino == ino &&
1569 inode->i_sb == sb &&
1570 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1575 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1577 int insert_inode_locked(struct inode *inode)
1579 struct super_block *sb = inode->i_sb;
1580 ino_t ino = inode->i_ino;
1581 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1584 struct inode *old = NULL;
1585 spin_lock(&inode_hash_lock);
1586 hlist_for_each_entry(old, head, i_hash) {
1587 if (old->i_ino != ino)
1589 if (old->i_sb != sb)
1591 spin_lock(&old->i_lock);
1592 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1593 spin_unlock(&old->i_lock);
1599 spin_lock(&inode->i_lock);
1600 inode->i_state |= I_NEW | I_CREATING;
1601 hlist_add_head_rcu(&inode->i_hash, head);
1602 spin_unlock(&inode->i_lock);
1603 spin_unlock(&inode_hash_lock);
1606 if (unlikely(old->i_state & I_CREATING)) {
1607 spin_unlock(&old->i_lock);
1608 spin_unlock(&inode_hash_lock);
1612 spin_unlock(&old->i_lock);
1613 spin_unlock(&inode_hash_lock);
1615 if (unlikely(!inode_unhashed(old))) {
1622 EXPORT_SYMBOL(insert_inode_locked);
1624 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1625 int (*test)(struct inode *, void *), void *data)
1629 inode->i_state |= I_CREATING;
1630 old = inode_insert5(inode, hashval, test, NULL, data);
1638 EXPORT_SYMBOL(insert_inode_locked4);
1641 int generic_delete_inode(struct inode *inode)
1645 EXPORT_SYMBOL(generic_delete_inode);
1648 * Called when we're dropping the last reference
1651 * Call the FS "drop_inode()" function, defaulting to
1652 * the legacy UNIX filesystem behaviour. If it tells
1653 * us to evict inode, do so. Otherwise, retain inode
1654 * in cache if fs is alive, sync and evict if fs is
1657 static void iput_final(struct inode *inode)
1659 struct super_block *sb = inode->i_sb;
1660 const struct super_operations *op = inode->i_sb->s_op;
1661 unsigned long state;
1664 WARN_ON(inode->i_state & I_NEW);
1667 drop = op->drop_inode(inode);
1669 drop = generic_drop_inode(inode);
1672 !(inode->i_state & I_DONTCACHE) &&
1673 (sb->s_flags & SB_ACTIVE)) {
1674 inode_add_lru(inode);
1675 spin_unlock(&inode->i_lock);
1679 state = inode->i_state;
1681 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1682 spin_unlock(&inode->i_lock);
1684 write_inode_now(inode, 1);
1686 spin_lock(&inode->i_lock);
1687 state = inode->i_state;
1688 WARN_ON(state & I_NEW);
1689 state &= ~I_WILL_FREE;
1692 WRITE_ONCE(inode->i_state, state | I_FREEING);
1693 if (!list_empty(&inode->i_lru))
1694 inode_lru_list_del(inode);
1695 spin_unlock(&inode->i_lock);
1701 * iput - put an inode
1702 * @inode: inode to put
1704 * Puts an inode, dropping its usage count. If the inode use count hits
1705 * zero, the inode is then freed and may also be destroyed.
1707 * Consequently, iput() can sleep.
1709 void iput(struct inode *inode)
1713 BUG_ON(inode->i_state & I_CLEAR);
1715 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1716 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1717 atomic_inc(&inode->i_count);
1718 spin_unlock(&inode->i_lock);
1719 trace_writeback_lazytime_iput(inode);
1720 mark_inode_dirty_sync(inode);
1726 EXPORT_SYMBOL(iput);
1730 * bmap - find a block number in a file
1731 * @inode: inode owning the block number being requested
1732 * @block: pointer containing the block to find
1734 * Replaces the value in ``*block`` with the block number on the device holding
1735 * corresponding to the requested block number in the file.
1736 * That is, asked for block 4 of inode 1 the function will replace the
1737 * 4 in ``*block``, with disk block relative to the disk start that holds that
1738 * block of the file.
1740 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1741 * hole, returns 0 and ``*block`` is also set to 0.
1743 int bmap(struct inode *inode, sector_t *block)
1745 if (!inode->i_mapping->a_ops->bmap)
1748 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1751 EXPORT_SYMBOL(bmap);
1755 * With relative atime, only update atime if the previous atime is
1756 * earlier than either the ctime or mtime or if at least a day has
1757 * passed since the last atime update.
1759 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1760 struct timespec64 now)
1763 if (!(mnt->mnt_flags & MNT_RELATIME))
1766 * Is mtime younger than atime? If yes, update atime:
1768 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1771 * Is ctime younger than atime? If yes, update atime:
1773 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1777 * Is the previous atime value older than a day? If yes,
1780 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1783 * Good, we can skip the atime update:
1788 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1790 int iflags = I_DIRTY_TIME;
1793 if (flags & S_ATIME)
1794 inode->i_atime = *time;
1795 if (flags & S_VERSION)
1796 dirty = inode_maybe_inc_iversion(inode, false);
1797 if (flags & S_CTIME)
1798 inode->i_ctime = *time;
1799 if (flags & S_MTIME)
1800 inode->i_mtime = *time;
1801 if ((flags & (S_ATIME | S_CTIME | S_MTIME)) &&
1802 !(inode->i_sb->s_flags & SB_LAZYTIME))
1806 iflags |= I_DIRTY_SYNC;
1807 __mark_inode_dirty(inode, iflags);
1810 EXPORT_SYMBOL(generic_update_time);
1813 * This does the actual work of updating an inodes time or version. Must have
1814 * had called mnt_want_write() before calling this.
1816 int inode_update_time(struct inode *inode, struct timespec64 *time, int flags)
1818 if (inode->i_op->update_time)
1819 return inode->i_op->update_time(inode, time, flags);
1820 return generic_update_time(inode, time, flags);
1822 EXPORT_SYMBOL(inode_update_time);
1825 * touch_atime - update the access time
1826 * @path: the &struct path to update
1827 * @inode: inode to update
1829 * Update the accessed time on an inode and mark it for writeback.
1830 * This function automatically handles read only file systems and media,
1831 * as well as the "noatime" flag and inode specific "noatime" markers.
1833 bool atime_needs_update(const struct path *path, struct inode *inode)
1835 struct vfsmount *mnt = path->mnt;
1836 struct timespec64 now;
1838 if (inode->i_flags & S_NOATIME)
1841 /* Atime updates will likely cause i_uid and i_gid to be written
1842 * back improprely if their true value is unknown to the vfs.
1844 if (HAS_UNMAPPED_ID(inode))
1847 if (IS_NOATIME(inode))
1849 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1852 if (mnt->mnt_flags & MNT_NOATIME)
1854 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1857 now = current_time(inode);
1859 if (!relatime_need_update(mnt, inode, now))
1862 if (timespec64_equal(&inode->i_atime, &now))
1868 void touch_atime(const struct path *path)
1870 struct vfsmount *mnt = path->mnt;
1871 struct inode *inode = d_inode(path->dentry);
1872 struct timespec64 now;
1874 if (!atime_needs_update(path, inode))
1877 if (!sb_start_write_trylock(inode->i_sb))
1880 if (__mnt_want_write(mnt) != 0)
1883 * File systems can error out when updating inodes if they need to
1884 * allocate new space to modify an inode (such is the case for
1885 * Btrfs), but since we touch atime while walking down the path we
1886 * really don't care if we failed to update the atime of the file,
1887 * so just ignore the return value.
1888 * We may also fail on filesystems that have the ability to make parts
1889 * of the fs read only, e.g. subvolumes in Btrfs.
1891 now = current_time(inode);
1892 inode_update_time(inode, &now, S_ATIME);
1893 __mnt_drop_write(mnt);
1895 sb_end_write(inode->i_sb);
1897 EXPORT_SYMBOL(touch_atime);
1900 * Return mask of changes for notify_change() that need to be done as a
1901 * response to write or truncate. Return 0 if nothing has to be changed.
1902 * Negative value on error (change should be denied).
1904 int dentry_needs_remove_privs(struct dentry *dentry)
1906 struct inode *inode = d_inode(dentry);
1910 if (IS_NOSEC(inode))
1913 mask = setattr_should_drop_suidgid(inode);
1914 ret = security_inode_need_killpriv(dentry);
1918 mask |= ATTR_KILL_PRIV;
1922 static int __remove_privs(struct dentry *dentry, int kill)
1924 struct iattr newattrs;
1926 newattrs.ia_valid = ATTR_FORCE | kill;
1928 * Note we call this on write, so notify_change will not
1929 * encounter any conflicting delegations:
1931 return notify_change(dentry, &newattrs, NULL);
1935 * Remove special file priviledges (suid, capabilities) when file is written
1938 int file_remove_privs(struct file *file)
1940 struct dentry *dentry = file_dentry(file);
1941 struct inode *inode = file_inode(file);
1946 * Fast path for nothing security related.
1947 * As well for non-regular files, e.g. blkdev inodes.
1948 * For example, blkdev_write_iter() might get here
1949 * trying to remove privs which it is not allowed to.
1951 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1954 kill = dentry_needs_remove_privs(dentry);
1958 error = __remove_privs(dentry, kill);
1960 inode_has_no_xattr(inode);
1964 EXPORT_SYMBOL(file_remove_privs);
1967 * file_update_time - update mtime and ctime time
1968 * @file: file accessed
1970 * Update the mtime and ctime members of an inode and mark the inode
1971 * for writeback. Note that this function is meant exclusively for
1972 * usage in the file write path of filesystems, and filesystems may
1973 * choose to explicitly ignore update via this function with the
1974 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1975 * timestamps are handled by the server. This can return an error for
1976 * file systems who need to allocate space in order to update an inode.
1979 int file_update_time(struct file *file)
1981 struct inode *inode = file_inode(file);
1982 struct timespec64 now;
1986 /* First try to exhaust all avenues to not sync */
1987 if (IS_NOCMTIME(inode))
1990 now = current_time(inode);
1991 if (!timespec64_equal(&inode->i_mtime, &now))
1994 if (!timespec64_equal(&inode->i_ctime, &now))
1997 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
1998 sync_it |= S_VERSION;
2003 /* Finally allowed to write? Takes lock. */
2004 if (__mnt_want_write_file(file))
2007 ret = inode_update_time(inode, &now, sync_it);
2008 __mnt_drop_write_file(file);
2012 EXPORT_SYMBOL(file_update_time);
2014 /* Caller must hold the file's inode lock */
2015 int file_modified(struct file *file)
2020 * Clear the security bits if the process is not being run by root.
2021 * This keeps people from modifying setuid and setgid binaries.
2023 err = file_remove_privs(file);
2027 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2030 return file_update_time(file);
2032 EXPORT_SYMBOL(file_modified);
2034 int inode_needs_sync(struct inode *inode)
2038 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2042 EXPORT_SYMBOL(inode_needs_sync);
2045 * If we try to find an inode in the inode hash while it is being
2046 * deleted, we have to wait until the filesystem completes its
2047 * deletion before reporting that it isn't found. This function waits
2048 * until the deletion _might_ have completed. Callers are responsible
2049 * to recheck inode state.
2051 * It doesn't matter if I_NEW is not set initially, a call to
2052 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2055 static void __wait_on_freeing_inode(struct inode *inode)
2057 wait_queue_head_t *wq;
2058 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2059 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2060 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2061 spin_unlock(&inode->i_lock);
2062 spin_unlock(&inode_hash_lock);
2064 finish_wait(wq, &wait.wq_entry);
2065 spin_lock(&inode_hash_lock);
2068 static __initdata unsigned long ihash_entries;
2069 static int __init set_ihash_entries(char *str)
2073 ihash_entries = simple_strtoul(str, &str, 0);
2076 __setup("ihash_entries=", set_ihash_entries);
2079 * Initialize the waitqueues and inode hash table.
2081 void __init inode_init_early(void)
2083 /* If hashes are distributed across NUMA nodes, defer
2084 * hash allocation until vmalloc space is available.
2090 alloc_large_system_hash("Inode-cache",
2091 sizeof(struct hlist_head),
2094 HASH_EARLY | HASH_ZERO,
2101 void __init inode_init(void)
2103 /* inode slab cache */
2104 inode_cachep = kmem_cache_create("inode_cache",
2105 sizeof(struct inode),
2107 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2108 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2111 /* Hash may have been set up in inode_init_early */
2116 alloc_large_system_hash("Inode-cache",
2117 sizeof(struct hlist_head),
2127 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2129 inode->i_mode = mode;
2130 if (S_ISCHR(mode)) {
2131 inode->i_fop = &def_chr_fops;
2132 inode->i_rdev = rdev;
2133 } else if (S_ISBLK(mode)) {
2134 inode->i_fop = &def_blk_fops;
2135 inode->i_rdev = rdev;
2136 } else if (S_ISFIFO(mode))
2137 inode->i_fop = &pipefifo_fops;
2138 else if (S_ISSOCK(mode))
2139 ; /* leave it no_open_fops */
2141 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2142 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2145 EXPORT_SYMBOL(init_special_inode);
2148 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2150 * @dir: Directory inode
2151 * @mode: mode of the new inode
2153 void inode_init_owner(struct inode *inode, const struct inode *dir,
2156 inode->i_uid = current_fsuid();
2157 if (dir && dir->i_mode & S_ISGID) {
2158 inode->i_gid = dir->i_gid;
2160 /* Directories are special, and always inherit S_ISGID */
2164 inode->i_gid = current_fsgid();
2165 inode->i_mode = mode;
2167 EXPORT_SYMBOL(inode_init_owner);
2170 * inode_owner_or_capable - check current task permissions to inode
2171 * @inode: inode being checked
2173 * Return true if current either has CAP_FOWNER in a namespace with the
2174 * inode owner uid mapped, or owns the file.
2176 bool inode_owner_or_capable(const struct inode *inode)
2178 struct user_namespace *ns;
2180 if (uid_eq(current_fsuid(), inode->i_uid))
2183 ns = current_user_ns();
2184 if (kuid_has_mapping(ns, inode->i_uid) && ns_capable(ns, CAP_FOWNER))
2188 EXPORT_SYMBOL(inode_owner_or_capable);
2191 * Direct i/o helper functions
2193 static void __inode_dio_wait(struct inode *inode)
2195 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2196 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2199 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2200 if (atomic_read(&inode->i_dio_count))
2202 } while (atomic_read(&inode->i_dio_count));
2203 finish_wait(wq, &q.wq_entry);
2207 * inode_dio_wait - wait for outstanding DIO requests to finish
2208 * @inode: inode to wait for
2210 * Waits for all pending direct I/O requests to finish so that we can
2211 * proceed with a truncate or equivalent operation.
2213 * Must be called under a lock that serializes taking new references
2214 * to i_dio_count, usually by inode->i_mutex.
2216 void inode_dio_wait(struct inode *inode)
2218 if (atomic_read(&inode->i_dio_count))
2219 __inode_dio_wait(inode);
2221 EXPORT_SYMBOL(inode_dio_wait);
2224 * inode_set_flags - atomically set some inode flags
2226 * Note: the caller should be holding i_mutex, or else be sure that
2227 * they have exclusive access to the inode structure (i.e., while the
2228 * inode is being instantiated). The reason for the cmpxchg() loop
2229 * --- which wouldn't be necessary if all code paths which modify
2230 * i_flags actually followed this rule, is that there is at least one
2231 * code path which doesn't today so we use cmpxchg() out of an abundance
2234 * In the long run, i_mutex is overkill, and we should probably look
2235 * at using the i_lock spinlock to protect i_flags, and then make sure
2236 * it is so documented in include/linux/fs.h and that all code follows
2237 * the locking convention!!
2239 void inode_set_flags(struct inode *inode, unsigned int flags,
2242 WARN_ON_ONCE(flags & ~mask);
2243 set_mask_bits(&inode->i_flags, mask, flags);
2245 EXPORT_SYMBOL(inode_set_flags);
2247 void inode_nohighmem(struct inode *inode)
2249 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2251 EXPORT_SYMBOL(inode_nohighmem);
2254 * timestamp_truncate - Truncate timespec to a granularity
2256 * @inode: inode being updated
2258 * Truncate a timespec to the granularity supported by the fs
2259 * containing the inode. Always rounds down. gran must
2260 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2262 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2264 struct super_block *sb = inode->i_sb;
2265 unsigned int gran = sb->s_time_gran;
2267 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2268 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2271 /* Avoid division in the common cases 1 ns and 1 s. */
2274 else if (gran == NSEC_PER_SEC)
2276 else if (gran > 1 && gran < NSEC_PER_SEC)
2277 t.tv_nsec -= t.tv_nsec % gran;
2279 WARN(1, "invalid file time granularity: %u", gran);
2282 EXPORT_SYMBOL(timestamp_truncate);
2285 * current_time - Return FS time
2288 * Return the current time truncated to the time granularity supported by
2291 * Note that inode and inode->sb cannot be NULL.
2292 * Otherwise, the function warns and returns time without truncation.
2294 struct timespec64 current_time(struct inode *inode)
2296 struct timespec64 now;
2298 ktime_get_coarse_real_ts64(&now);
2300 if (unlikely(!inode->i_sb)) {
2301 WARN(1, "current_time() called with uninitialized super_block in the inode");
2305 return timestamp_truncate(now, inode);
2307 EXPORT_SYMBOL(current_time);
2310 * Generic function to check FS_IOC_SETFLAGS values and reject any invalid
2313 * Note: the caller should be holding i_mutex, or else be sure that they have
2314 * exclusive access to the inode structure.
2316 int vfs_ioc_setflags_prepare(struct inode *inode, unsigned int oldflags,
2320 * The IMMUTABLE and APPEND_ONLY flags can only be changed by
2321 * the relevant capability.
2323 * This test looks nicer. Thanks to Pauline Middelink
2325 if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL) &&
2326 !capable(CAP_LINUX_IMMUTABLE))
2329 return fscrypt_prepare_setflags(inode, oldflags, flags);
2331 EXPORT_SYMBOL(vfs_ioc_setflags_prepare);
2334 * Generic function to check FS_IOC_FSSETXATTR values and reject any invalid
2337 * Note: the caller should be holding i_mutex, or else be sure that they have
2338 * exclusive access to the inode structure.
2340 int vfs_ioc_fssetxattr_check(struct inode *inode, const struct fsxattr *old_fa,
2344 * Can't modify an immutable/append-only file unless we have
2345 * appropriate permission.
2347 if ((old_fa->fsx_xflags ^ fa->fsx_xflags) &
2348 (FS_XFLAG_IMMUTABLE | FS_XFLAG_APPEND) &&
2349 !capable(CAP_LINUX_IMMUTABLE))
2353 * Project Quota ID state is only allowed to change from within the init
2354 * namespace. Enforce that restriction only if we are trying to change
2355 * the quota ID state. Everything else is allowed in user namespaces.
2357 if (current_user_ns() != &init_user_ns) {
2358 if (old_fa->fsx_projid != fa->fsx_projid)
2360 if ((old_fa->fsx_xflags ^ fa->fsx_xflags) &
2361 FS_XFLAG_PROJINHERIT)
2365 /* Check extent size hints. */
2366 if ((fa->fsx_xflags & FS_XFLAG_EXTSIZE) && !S_ISREG(inode->i_mode))
2369 if ((fa->fsx_xflags & FS_XFLAG_EXTSZINHERIT) &&
2370 !S_ISDIR(inode->i_mode))
2373 if ((fa->fsx_xflags & FS_XFLAG_COWEXTSIZE) &&
2374 !S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2378 * It is only valid to set the DAX flag on regular files and
2379 * directories on filesystems.
2381 if ((fa->fsx_xflags & FS_XFLAG_DAX) &&
2382 !(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
2385 /* Extent size hints of zero turn off the flags. */
2386 if (fa->fsx_extsize == 0)
2387 fa->fsx_xflags &= ~(FS_XFLAG_EXTSIZE | FS_XFLAG_EXTSZINHERIT);
2388 if (fa->fsx_cowextsize == 0)
2389 fa->fsx_xflags &= ~FS_XFLAG_COWEXTSIZE;
2393 EXPORT_SYMBOL(vfs_ioc_fssetxattr_check);
2396 * inode_set_ctime_current - set the ctime to current_time
2399 * Set the inode->i_ctime to the current value for the inode. Returns
2400 * the current value that was assigned to i_ctime.
2402 struct timespec64 inode_set_ctime_current(struct inode *inode)
2404 struct timespec64 now = current_time(inode);
2406 inode_set_ctime(inode, now.tv_sec, now.tv_nsec);
2409 EXPORT_SYMBOL(inode_set_ctime_current);
2412 * in_group_or_capable - check whether caller is CAP_FSETID privileged
2413 * @inode: inode to check
2414 * @gid: the new/current gid of @inode
2416 * Check wether @gid is in the caller's group list or if the caller is
2417 * privileged with CAP_FSETID over @inode. This can be used to determine
2418 * whether the setgid bit can be kept or must be dropped.
2420 * Return: true if the caller is sufficiently privileged, false if not.
2422 bool in_group_or_capable(const struct inode *inode, kgid_t gid)
2424 if (in_group_p(gid))
2426 if (capable_wrt_inode_uidgid(inode, CAP_FSETID))
2432 * mode_strip_sgid - handle the sgid bit for non-directories
2433 * @dir: parent directory inode
2434 * @mode: mode of the file to be created in @dir
2436 * If the @mode of the new file has both the S_ISGID and S_IXGRP bit
2437 * raised and @dir has the S_ISGID bit raised ensure that the caller is
2438 * either in the group of the parent directory or they have CAP_FSETID
2439 * in their user namespace and are privileged over the parent directory.
2440 * In all other cases, strip the S_ISGID bit from @mode.
2442 * Return: the new mode to use for the file
2444 umode_t mode_strip_sgid(const struct inode *dir, umode_t mode)
2446 if ((mode & (S_ISGID | S_IXGRP)) != (S_ISGID | S_IXGRP))
2448 if (S_ISDIR(mode) || !dir || !(dir->i_mode & S_ISGID))
2450 if (in_group_or_capable(dir, dir->i_gid))
2452 return mode & ~S_ISGID;
2454 EXPORT_SYMBOL(mode_strip_sgid);