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 if (security_inode_alloc(inode))
173 spin_lock_init(&inode->i_lock);
174 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
176 init_rwsem(&inode->i_rwsem);
177 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
179 atomic_set(&inode->i_dio_count, 0);
181 mapping->a_ops = &empty_aops;
182 mapping->host = inode;
184 if (sb->s_type->fs_flags & FS_THP_SUPPORT)
185 __set_bit(AS_THP_SUPPORT, &mapping->flags);
187 atomic_set(&mapping->i_mmap_writable, 0);
188 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
189 atomic_set(&mapping->nr_thps, 0);
191 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
192 mapping->private_data = NULL;
193 mapping->writeback_index = 0;
194 inode->i_private = NULL;
195 inode->i_mapping = mapping;
196 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
197 #ifdef CONFIG_FS_POSIX_ACL
198 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
201 #ifdef CONFIG_FSNOTIFY
202 inode->i_fsnotify_mask = 0;
204 inode->i_flctx = NULL;
205 this_cpu_inc(nr_inodes);
211 EXPORT_SYMBOL(inode_init_always);
213 void free_inode_nonrcu(struct inode *inode)
215 kmem_cache_free(inode_cachep, inode);
217 EXPORT_SYMBOL(free_inode_nonrcu);
219 static void i_callback(struct rcu_head *head)
221 struct inode *inode = container_of(head, struct inode, i_rcu);
222 if (inode->free_inode)
223 inode->free_inode(inode);
225 free_inode_nonrcu(inode);
228 static struct inode *alloc_inode(struct super_block *sb)
230 const struct super_operations *ops = sb->s_op;
233 if (ops->alloc_inode)
234 inode = ops->alloc_inode(sb);
236 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
241 if (unlikely(inode_init_always(sb, inode))) {
242 if (ops->destroy_inode) {
243 ops->destroy_inode(inode);
244 if (!ops->free_inode)
247 inode->free_inode = ops->free_inode;
248 i_callback(&inode->i_rcu);
255 void __destroy_inode(struct inode *inode)
257 BUG_ON(inode_has_buffers(inode));
258 inode_detach_wb(inode);
259 security_inode_free(inode);
260 fsnotify_inode_delete(inode);
261 locks_free_lock_context(inode);
262 if (!inode->i_nlink) {
263 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
264 atomic_long_dec(&inode->i_sb->s_remove_count);
267 #ifdef CONFIG_FS_POSIX_ACL
268 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
269 posix_acl_release(inode->i_acl);
270 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
271 posix_acl_release(inode->i_default_acl);
273 this_cpu_dec(nr_inodes);
275 EXPORT_SYMBOL(__destroy_inode);
277 static void destroy_inode(struct inode *inode)
279 const struct super_operations *ops = inode->i_sb->s_op;
281 BUG_ON(!list_empty(&inode->i_lru));
282 __destroy_inode(inode);
283 if (ops->destroy_inode) {
284 ops->destroy_inode(inode);
285 if (!ops->free_inode)
288 inode->free_inode = ops->free_inode;
289 call_rcu(&inode->i_rcu, i_callback);
293 * drop_nlink - directly drop an inode's link count
296 * This is a low-level filesystem helper to replace any
297 * direct filesystem manipulation of i_nlink. In cases
298 * where we are attempting to track writes to the
299 * filesystem, a decrement to zero means an imminent
300 * write when the file is truncated and actually unlinked
303 void drop_nlink(struct inode *inode)
305 WARN_ON(inode->i_nlink == 0);
308 atomic_long_inc(&inode->i_sb->s_remove_count);
310 EXPORT_SYMBOL(drop_nlink);
313 * clear_nlink - directly zero an inode's link count
316 * This is a low-level filesystem helper to replace any
317 * direct filesystem manipulation of i_nlink. See
318 * drop_nlink() for why we care about i_nlink hitting zero.
320 void clear_nlink(struct inode *inode)
322 if (inode->i_nlink) {
323 inode->__i_nlink = 0;
324 atomic_long_inc(&inode->i_sb->s_remove_count);
327 EXPORT_SYMBOL(clear_nlink);
330 * set_nlink - directly set an inode's link count
332 * @nlink: new nlink (should be non-zero)
334 * This is a low-level filesystem helper to replace any
335 * direct filesystem manipulation of i_nlink.
337 void set_nlink(struct inode *inode, unsigned int nlink)
342 /* Yes, some filesystems do change nlink from zero to one */
343 if (inode->i_nlink == 0)
344 atomic_long_dec(&inode->i_sb->s_remove_count);
346 inode->__i_nlink = nlink;
349 EXPORT_SYMBOL(set_nlink);
352 * inc_nlink - directly increment an inode's link count
355 * This is a low-level filesystem helper to replace any
356 * direct filesystem manipulation of i_nlink. Currently,
357 * it is only here for parity with dec_nlink().
359 void inc_nlink(struct inode *inode)
361 if (unlikely(inode->i_nlink == 0)) {
362 WARN_ON(!(inode->i_state & I_LINKABLE));
363 atomic_long_dec(&inode->i_sb->s_remove_count);
368 EXPORT_SYMBOL(inc_nlink);
370 static void __address_space_init_once(struct address_space *mapping)
372 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
373 init_rwsem(&mapping->i_mmap_rwsem);
374 INIT_LIST_HEAD(&mapping->private_list);
375 spin_lock_init(&mapping->private_lock);
376 mapping->i_mmap = RB_ROOT_CACHED;
379 void address_space_init_once(struct address_space *mapping)
381 memset(mapping, 0, sizeof(*mapping));
382 __address_space_init_once(mapping);
384 EXPORT_SYMBOL(address_space_init_once);
387 * These are initializations that only need to be done
388 * once, because the fields are idempotent across use
389 * of the inode, so let the slab aware of that.
391 void inode_init_once(struct inode *inode)
393 memset(inode, 0, sizeof(*inode));
394 INIT_HLIST_NODE(&inode->i_hash);
395 INIT_LIST_HEAD(&inode->i_devices);
396 INIT_LIST_HEAD(&inode->i_io_list);
397 INIT_LIST_HEAD(&inode->i_wb_list);
398 INIT_LIST_HEAD(&inode->i_lru);
399 __address_space_init_once(&inode->i_data);
400 i_size_ordered_init(inode);
402 EXPORT_SYMBOL(inode_init_once);
404 static void init_once(void *foo)
406 struct inode *inode = (struct inode *) foo;
408 inode_init_once(inode);
412 * inode->i_lock must be held
414 void __iget(struct inode *inode)
416 atomic_inc(&inode->i_count);
420 * get additional reference to inode; caller must already hold one.
422 void ihold(struct inode *inode)
424 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
426 EXPORT_SYMBOL(ihold);
428 static void inode_lru_list_add(struct inode *inode)
430 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
431 this_cpu_inc(nr_unused);
433 inode->i_state |= I_REFERENCED;
437 * Add inode to LRU if needed (inode is unused and clean).
439 * Needs inode->i_lock held.
441 void inode_add_lru(struct inode *inode)
443 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
444 I_FREEING | I_WILL_FREE)) &&
445 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
446 inode_lru_list_add(inode);
450 static void inode_lru_list_del(struct inode *inode)
453 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
454 this_cpu_dec(nr_unused);
458 * inode_sb_list_add - add inode to the superblock list of inodes
459 * @inode: inode to add
461 void inode_sb_list_add(struct inode *inode)
463 spin_lock(&inode->i_sb->s_inode_list_lock);
464 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
465 spin_unlock(&inode->i_sb->s_inode_list_lock);
467 EXPORT_SYMBOL_GPL(inode_sb_list_add);
469 static inline void inode_sb_list_del(struct inode *inode)
471 if (!list_empty(&inode->i_sb_list)) {
472 spin_lock(&inode->i_sb->s_inode_list_lock);
473 list_del_init(&inode->i_sb_list);
474 spin_unlock(&inode->i_sb->s_inode_list_lock);
478 static unsigned long hash(struct super_block *sb, unsigned long hashval)
482 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
484 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
485 return tmp & i_hash_mask;
489 * __insert_inode_hash - hash an inode
490 * @inode: unhashed inode
491 * @hashval: unsigned long value used to locate this object in the
494 * Add an inode to the inode hash for this superblock.
496 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
498 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
500 spin_lock(&inode_hash_lock);
501 spin_lock(&inode->i_lock);
502 hlist_add_head_rcu(&inode->i_hash, b);
503 spin_unlock(&inode->i_lock);
504 spin_unlock(&inode_hash_lock);
506 EXPORT_SYMBOL(__insert_inode_hash);
509 * __remove_inode_hash - remove an inode from the hash
510 * @inode: inode to unhash
512 * Remove an inode from the superblock.
514 void __remove_inode_hash(struct inode *inode)
516 spin_lock(&inode_hash_lock);
517 spin_lock(&inode->i_lock);
518 hlist_del_init_rcu(&inode->i_hash);
519 spin_unlock(&inode->i_lock);
520 spin_unlock(&inode_hash_lock);
522 EXPORT_SYMBOL(__remove_inode_hash);
524 void clear_inode(struct inode *inode)
527 * We have to cycle the i_pages lock here because reclaim can be in the
528 * process of removing the last page (in __delete_from_page_cache())
529 * and we must not free the mapping under it.
531 xa_lock_irq(&inode->i_data.i_pages);
532 BUG_ON(inode->i_data.nrpages);
533 BUG_ON(inode->i_data.nrexceptional);
534 xa_unlock_irq(&inode->i_data.i_pages);
535 BUG_ON(!list_empty(&inode->i_data.private_list));
536 BUG_ON(!(inode->i_state & I_FREEING));
537 BUG_ON(inode->i_state & I_CLEAR);
538 BUG_ON(!list_empty(&inode->i_wb_list));
539 /* don't need i_lock here, no concurrent mods to i_state */
540 inode->i_state = I_FREEING | I_CLEAR;
542 EXPORT_SYMBOL(clear_inode);
545 * Free the inode passed in, removing it from the lists it is still connected
546 * to. We remove any pages still attached to the inode and wait for any IO that
547 * is still in progress before finally destroying the inode.
549 * An inode must already be marked I_FREEING so that we avoid the inode being
550 * moved back onto lists if we race with other code that manipulates the lists
551 * (e.g. writeback_single_inode). The caller is responsible for setting this.
553 * An inode must already be removed from the LRU list before being evicted from
554 * the cache. This should occur atomically with setting the I_FREEING state
555 * flag, so no inodes here should ever be on the LRU when being evicted.
557 static void evict(struct inode *inode)
559 const struct super_operations *op = inode->i_sb->s_op;
561 BUG_ON(!(inode->i_state & I_FREEING));
562 BUG_ON(!list_empty(&inode->i_lru));
564 if (!list_empty(&inode->i_io_list))
565 inode_io_list_del(inode);
567 inode_sb_list_del(inode);
570 * Wait for flusher thread to be done with the inode so that filesystem
571 * does not start destroying it while writeback is still running. Since
572 * the inode has I_FREEING set, flusher thread won't start new work on
573 * the inode. We just have to wait for running writeback to finish.
575 inode_wait_for_writeback(inode);
577 if (op->evict_inode) {
578 op->evict_inode(inode);
580 truncate_inode_pages_final(&inode->i_data);
583 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
585 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
588 remove_inode_hash(inode);
590 spin_lock(&inode->i_lock);
591 wake_up_bit(&inode->i_state, __I_NEW);
592 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
593 spin_unlock(&inode->i_lock);
595 destroy_inode(inode);
599 * dispose_list - dispose of the contents of a local list
600 * @head: the head of the list to free
602 * Dispose-list gets a local list with local inodes in it, so it doesn't
603 * need to worry about list corruption and SMP locks.
605 static void dispose_list(struct list_head *head)
607 while (!list_empty(head)) {
610 inode = list_first_entry(head, struct inode, i_lru);
611 list_del_init(&inode->i_lru);
619 * evict_inodes - evict all evictable inodes for a superblock
620 * @sb: superblock to operate on
622 * Make sure that no inodes with zero refcount are retained. This is
623 * called by superblock shutdown after having SB_ACTIVE flag removed,
624 * so any inode reaching zero refcount during or after that call will
625 * be immediately evicted.
627 void evict_inodes(struct super_block *sb)
629 struct inode *inode, *next;
633 spin_lock(&sb->s_inode_list_lock);
634 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
635 if (atomic_read(&inode->i_count))
638 spin_lock(&inode->i_lock);
639 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
640 spin_unlock(&inode->i_lock);
644 inode->i_state |= I_FREEING;
645 inode_lru_list_del(inode);
646 spin_unlock(&inode->i_lock);
647 list_add(&inode->i_lru, &dispose);
650 * We can have a ton of inodes to evict at unmount time given
651 * enough memory, check to see if we need to go to sleep for a
652 * bit so we don't livelock.
654 if (need_resched()) {
655 spin_unlock(&sb->s_inode_list_lock);
657 dispose_list(&dispose);
661 spin_unlock(&sb->s_inode_list_lock);
663 dispose_list(&dispose);
665 EXPORT_SYMBOL_GPL(evict_inodes);
668 * invalidate_inodes - attempt to free all inodes on a superblock
669 * @sb: superblock to operate on
670 * @kill_dirty: flag to guide handling of dirty inodes
672 * Attempts to free all inodes for a given superblock. If there were any
673 * busy inodes return a non-zero value, else zero.
674 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
677 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
680 struct inode *inode, *next;
684 spin_lock(&sb->s_inode_list_lock);
685 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
686 spin_lock(&inode->i_lock);
687 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
688 spin_unlock(&inode->i_lock);
691 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
692 spin_unlock(&inode->i_lock);
696 if (atomic_read(&inode->i_count)) {
697 spin_unlock(&inode->i_lock);
702 inode->i_state |= I_FREEING;
703 inode_lru_list_del(inode);
704 spin_unlock(&inode->i_lock);
705 list_add(&inode->i_lru, &dispose);
706 if (need_resched()) {
707 spin_unlock(&sb->s_inode_list_lock);
709 dispose_list(&dispose);
713 spin_unlock(&sb->s_inode_list_lock);
715 dispose_list(&dispose);
721 * Isolate the inode from the LRU in preparation for freeing it.
723 * Any inodes which are pinned purely because of attached pagecache have their
724 * pagecache removed. If the inode has metadata buffers attached to
725 * mapping->private_list then try to remove them.
727 * If the inode has the I_REFERENCED flag set, then it means that it has been
728 * used recently - the flag is set in iput_final(). When we encounter such an
729 * inode, clear the flag and move it to the back of the LRU so it gets another
730 * pass through the LRU before it gets reclaimed. This is necessary because of
731 * the fact we are doing lazy LRU updates to minimise lock contention so the
732 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
733 * with this flag set because they are the inodes that are out of order.
735 static enum lru_status inode_lru_isolate(struct list_head *item,
736 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
738 struct list_head *freeable = arg;
739 struct inode *inode = container_of(item, struct inode, i_lru);
742 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
743 * If we fail to get the lock, just skip it.
745 if (!spin_trylock(&inode->i_lock))
749 * Referenced or dirty inodes are still in use. Give them another pass
750 * through the LRU as we canot reclaim them now.
752 if (atomic_read(&inode->i_count) ||
753 (inode->i_state & ~I_REFERENCED)) {
754 list_lru_isolate(lru, &inode->i_lru);
755 spin_unlock(&inode->i_lock);
756 this_cpu_dec(nr_unused);
760 /* recently referenced inodes get one more pass */
761 if (inode->i_state & I_REFERENCED) {
762 inode->i_state &= ~I_REFERENCED;
763 spin_unlock(&inode->i_lock);
767 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
769 spin_unlock(&inode->i_lock);
770 spin_unlock(lru_lock);
771 if (remove_inode_buffers(inode)) {
773 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
774 if (current_is_kswapd())
775 __count_vm_events(KSWAPD_INODESTEAL, reap);
777 __count_vm_events(PGINODESTEAL, reap);
778 if (current->reclaim_state)
779 current->reclaim_state->reclaimed_slab += reap;
786 WARN_ON(inode->i_state & I_NEW);
787 inode->i_state |= I_FREEING;
788 list_lru_isolate_move(lru, &inode->i_lru, freeable);
789 spin_unlock(&inode->i_lock);
791 this_cpu_dec(nr_unused);
796 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
797 * This is called from the superblock shrinker function with a number of inodes
798 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
799 * then are freed outside inode_lock by dispose_list().
801 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
806 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
807 inode_lru_isolate, &freeable);
808 dispose_list(&freeable);
812 static void __wait_on_freeing_inode(struct inode *inode);
814 * Called with the inode lock held.
816 static struct inode *find_inode(struct super_block *sb,
817 struct hlist_head *head,
818 int (*test)(struct inode *, void *),
821 struct inode *inode = NULL;
824 hlist_for_each_entry(inode, head, i_hash) {
825 if (inode->i_sb != sb)
827 if (!test(inode, data))
829 spin_lock(&inode->i_lock);
830 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
831 __wait_on_freeing_inode(inode);
834 if (unlikely(inode->i_state & I_CREATING)) {
835 spin_unlock(&inode->i_lock);
836 return ERR_PTR(-ESTALE);
839 spin_unlock(&inode->i_lock);
846 * find_inode_fast is the fast path version of find_inode, see the comment at
847 * iget_locked for details.
849 static struct inode *find_inode_fast(struct super_block *sb,
850 struct hlist_head *head, unsigned long ino)
852 struct inode *inode = NULL;
855 hlist_for_each_entry(inode, head, i_hash) {
856 if (inode->i_ino != ino)
858 if (inode->i_sb != sb)
860 spin_lock(&inode->i_lock);
861 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
862 __wait_on_freeing_inode(inode);
865 if (unlikely(inode->i_state & I_CREATING)) {
866 spin_unlock(&inode->i_lock);
867 return ERR_PTR(-ESTALE);
870 spin_unlock(&inode->i_lock);
877 * Each cpu owns a range of LAST_INO_BATCH numbers.
878 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
879 * to renew the exhausted range.
881 * This does not significantly increase overflow rate because every CPU can
882 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
883 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
884 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
885 * overflow rate by 2x, which does not seem too significant.
887 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
888 * error if st_ino won't fit in target struct field. Use 32bit counter
889 * here to attempt to avoid that.
891 #define LAST_INO_BATCH 1024
892 static DEFINE_PER_CPU(unsigned int, last_ino);
894 unsigned int get_next_ino(void)
896 unsigned int *p = &get_cpu_var(last_ino);
897 unsigned int res = *p;
900 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
901 static atomic_t shared_last_ino;
902 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
904 res = next - LAST_INO_BATCH;
909 /* get_next_ino should not provide a 0 inode number */
913 put_cpu_var(last_ino);
916 EXPORT_SYMBOL(get_next_ino);
919 * new_inode_pseudo - obtain an inode
922 * Allocates a new inode for given superblock.
923 * Inode wont be chained in superblock s_inodes list
925 * - fs can't be unmount
926 * - quotas, fsnotify, writeback can't work
928 struct inode *new_inode_pseudo(struct super_block *sb)
930 struct inode *inode = alloc_inode(sb);
933 spin_lock(&inode->i_lock);
935 spin_unlock(&inode->i_lock);
936 INIT_LIST_HEAD(&inode->i_sb_list);
942 * new_inode - obtain an inode
945 * Allocates a new inode for given superblock. The default gfp_mask
946 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
947 * If HIGHMEM pages are unsuitable or it is known that pages allocated
948 * for the page cache are not reclaimable or migratable,
949 * mapping_set_gfp_mask() must be called with suitable flags on the
950 * newly created inode's mapping
953 struct inode *new_inode(struct super_block *sb)
957 spin_lock_prefetch(&sb->s_inode_list_lock);
959 inode = new_inode_pseudo(sb);
961 inode_sb_list_add(inode);
964 EXPORT_SYMBOL(new_inode);
966 #ifdef CONFIG_DEBUG_LOCK_ALLOC
967 void lockdep_annotate_inode_mutex_key(struct inode *inode)
969 if (S_ISDIR(inode->i_mode)) {
970 struct file_system_type *type = inode->i_sb->s_type;
972 /* Set new key only if filesystem hasn't already changed it */
973 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
975 * ensure nobody is actually holding i_mutex
977 // mutex_destroy(&inode->i_mutex);
978 init_rwsem(&inode->i_rwsem);
979 lockdep_set_class(&inode->i_rwsem,
980 &type->i_mutex_dir_key);
984 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
988 * unlock_new_inode - clear the I_NEW state and wake up any waiters
989 * @inode: new inode to unlock
991 * Called when the inode is fully initialised to clear the new state of the
992 * inode and wake up anyone waiting for the inode to finish initialisation.
994 void unlock_new_inode(struct inode *inode)
996 lockdep_annotate_inode_mutex_key(inode);
997 spin_lock(&inode->i_lock);
998 WARN_ON(!(inode->i_state & I_NEW));
999 inode->i_state &= ~I_NEW & ~I_CREATING;
1001 wake_up_bit(&inode->i_state, __I_NEW);
1002 spin_unlock(&inode->i_lock);
1004 EXPORT_SYMBOL(unlock_new_inode);
1006 void discard_new_inode(struct inode *inode)
1008 lockdep_annotate_inode_mutex_key(inode);
1009 spin_lock(&inode->i_lock);
1010 WARN_ON(!(inode->i_state & I_NEW));
1011 inode->i_state &= ~I_NEW;
1013 wake_up_bit(&inode->i_state, __I_NEW);
1014 spin_unlock(&inode->i_lock);
1017 EXPORT_SYMBOL(discard_new_inode);
1020 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1022 * Lock any non-NULL argument that is not a directory.
1023 * Zero, one or two objects may be locked by this function.
1025 * @inode1: first inode to lock
1026 * @inode2: second inode to lock
1028 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1030 if (inode1 > inode2)
1031 swap(inode1, inode2);
1033 if (inode1 && !S_ISDIR(inode1->i_mode))
1035 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1036 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1038 EXPORT_SYMBOL(lock_two_nondirectories);
1041 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1042 * @inode1: first inode to unlock
1043 * @inode2: second inode to unlock
1045 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1047 if (inode1 && !S_ISDIR(inode1->i_mode))
1048 inode_unlock(inode1);
1049 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1050 inode_unlock(inode2);
1052 EXPORT_SYMBOL(unlock_two_nondirectories);
1055 * inode_insert5 - obtain an inode from a mounted file system
1056 * @inode: pre-allocated inode to use for insert to cache
1057 * @hashval: hash value (usually inode number) to get
1058 * @test: callback used for comparisons between inodes
1059 * @set: callback used to initialize a new struct inode
1060 * @data: opaque data pointer to pass to @test and @set
1062 * Search for the inode specified by @hashval and @data in the inode cache,
1063 * and if present it is return it with an increased reference count. This is
1064 * a variant of iget5_locked() for callers that don't want to fail on memory
1065 * allocation of inode.
1067 * If the inode is not in cache, insert the pre-allocated inode to cache and
1068 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1069 * to fill it in before unlocking it via unlock_new_inode().
1071 * Note both @test and @set are called with the inode_hash_lock held, so can't
1074 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1075 int (*test)(struct inode *, void *),
1076 int (*set)(struct inode *, void *), void *data)
1078 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1080 bool creating = inode->i_state & I_CREATING;
1083 spin_lock(&inode_hash_lock);
1084 old = find_inode(inode->i_sb, head, test, data);
1085 if (unlikely(old)) {
1087 * Uhhuh, somebody else created the same inode under us.
1088 * Use the old inode instead of the preallocated one.
1090 spin_unlock(&inode_hash_lock);
1094 if (unlikely(inode_unhashed(old))) {
1101 if (set && unlikely(set(inode, data))) {
1107 * Return the locked inode with I_NEW set, the
1108 * caller is responsible for filling in the contents
1110 spin_lock(&inode->i_lock);
1111 inode->i_state |= I_NEW;
1112 hlist_add_head_rcu(&inode->i_hash, head);
1113 spin_unlock(&inode->i_lock);
1115 inode_sb_list_add(inode);
1117 spin_unlock(&inode_hash_lock);
1121 EXPORT_SYMBOL(inode_insert5);
1124 * iget5_locked - obtain an inode from a mounted file system
1125 * @sb: super block of file system
1126 * @hashval: hash value (usually inode number) to get
1127 * @test: callback used for comparisons between inodes
1128 * @set: callback used to initialize a new struct inode
1129 * @data: opaque data pointer to pass to @test and @set
1131 * Search for the inode specified by @hashval and @data in the inode cache,
1132 * and if present it is return it with an increased reference count. This is
1133 * a generalized version of iget_locked() for file systems where the inode
1134 * number is not sufficient for unique identification of an inode.
1136 * If the inode is not in cache, allocate a new inode and return it locked,
1137 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1138 * before unlocking it via unlock_new_inode().
1140 * Note both @test and @set are called with the inode_hash_lock held, so can't
1143 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1144 int (*test)(struct inode *, void *),
1145 int (*set)(struct inode *, void *), void *data)
1147 struct inode *inode = ilookup5(sb, hashval, test, data);
1150 struct inode *new = alloc_inode(sb);
1154 inode = inode_insert5(new, hashval, test, set, data);
1155 if (unlikely(inode != new))
1161 EXPORT_SYMBOL(iget5_locked);
1164 * iget_locked - obtain an inode from a mounted file system
1165 * @sb: super block of file system
1166 * @ino: inode number to get
1168 * Search for the inode specified by @ino in the inode cache and if present
1169 * return it with an increased reference count. This is for file systems
1170 * where the inode number is sufficient for unique identification of an inode.
1172 * If the inode is not in cache, allocate a new inode and return it locked,
1173 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1174 * before unlocking it via unlock_new_inode().
1176 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1178 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1179 struct inode *inode;
1181 spin_lock(&inode_hash_lock);
1182 inode = find_inode_fast(sb, head, ino);
1183 spin_unlock(&inode_hash_lock);
1187 wait_on_inode(inode);
1188 if (unlikely(inode_unhashed(inode))) {
1195 inode = alloc_inode(sb);
1199 spin_lock(&inode_hash_lock);
1200 /* We released the lock, so.. */
1201 old = find_inode_fast(sb, head, ino);
1204 spin_lock(&inode->i_lock);
1205 inode->i_state = I_NEW;
1206 hlist_add_head_rcu(&inode->i_hash, head);
1207 spin_unlock(&inode->i_lock);
1208 inode_sb_list_add(inode);
1209 spin_unlock(&inode_hash_lock);
1211 /* Return the locked inode with I_NEW set, the
1212 * caller is responsible for filling in the contents
1218 * Uhhuh, somebody else created the same inode under
1219 * us. Use the old inode instead of the one we just
1222 spin_unlock(&inode_hash_lock);
1223 destroy_inode(inode);
1227 wait_on_inode(inode);
1228 if (unlikely(inode_unhashed(inode))) {
1235 EXPORT_SYMBOL(iget_locked);
1238 * search the inode cache for a matching inode number.
1239 * If we find one, then the inode number we are trying to
1240 * allocate is not unique and so we should not use it.
1242 * Returns 1 if the inode number is unique, 0 if it is not.
1244 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1246 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1247 struct inode *inode;
1249 hlist_for_each_entry_rcu(inode, b, i_hash) {
1250 if (inode->i_ino == ino && inode->i_sb == sb)
1257 * iunique - get a unique inode number
1259 * @max_reserved: highest reserved inode number
1261 * Obtain an inode number that is unique on the system for a given
1262 * superblock. This is used by file systems that have no natural
1263 * permanent inode numbering system. An inode number is returned that
1264 * is higher than the reserved limit but unique.
1267 * With a large number of inodes live on the file system this function
1268 * currently becomes quite slow.
1270 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1273 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1274 * error if st_ino won't fit in target struct field. Use 32bit counter
1275 * here to attempt to avoid that.
1277 static DEFINE_SPINLOCK(iunique_lock);
1278 static unsigned int counter;
1282 spin_lock(&iunique_lock);
1284 if (counter <= max_reserved)
1285 counter = max_reserved + 1;
1287 } while (!test_inode_iunique(sb, res));
1288 spin_unlock(&iunique_lock);
1293 EXPORT_SYMBOL(iunique);
1295 struct inode *igrab(struct inode *inode)
1297 spin_lock(&inode->i_lock);
1298 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1300 spin_unlock(&inode->i_lock);
1302 spin_unlock(&inode->i_lock);
1304 * Handle the case where s_op->clear_inode is not been
1305 * called yet, and somebody is calling igrab
1306 * while the inode is getting freed.
1312 EXPORT_SYMBOL(igrab);
1315 * ilookup5_nowait - search for an inode in the inode cache
1316 * @sb: super block of file system to search
1317 * @hashval: hash value (usually inode number) to search for
1318 * @test: callback used for comparisons between inodes
1319 * @data: opaque data pointer to pass to @test
1321 * Search for the inode specified by @hashval and @data in the inode cache.
1322 * If the inode is in the cache, the inode is returned with an incremented
1325 * Note: I_NEW is not waited upon so you have to be very careful what you do
1326 * with the returned inode. You probably should be using ilookup5() instead.
1328 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1330 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1331 int (*test)(struct inode *, void *), void *data)
1333 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1334 struct inode *inode;
1336 spin_lock(&inode_hash_lock);
1337 inode = find_inode(sb, head, test, data);
1338 spin_unlock(&inode_hash_lock);
1340 return IS_ERR(inode) ? NULL : inode;
1342 EXPORT_SYMBOL(ilookup5_nowait);
1345 * ilookup5 - search for an inode in the inode cache
1346 * @sb: super block of file system to search
1347 * @hashval: hash value (usually inode number) to search for
1348 * @test: callback used for comparisons between inodes
1349 * @data: opaque data pointer to pass to @test
1351 * Search for the inode specified by @hashval and @data in the inode cache,
1352 * and if the inode is in the cache, return the inode with an incremented
1353 * reference count. Waits on I_NEW before returning the inode.
1354 * returned with an incremented reference count.
1356 * This is a generalized version of ilookup() for file systems where the
1357 * inode number is not sufficient for unique identification of an inode.
1359 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1361 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1362 int (*test)(struct inode *, void *), void *data)
1364 struct inode *inode;
1366 inode = ilookup5_nowait(sb, hashval, test, data);
1368 wait_on_inode(inode);
1369 if (unlikely(inode_unhashed(inode))) {
1376 EXPORT_SYMBOL(ilookup5);
1379 * ilookup - search for an inode in the inode cache
1380 * @sb: super block of file system to search
1381 * @ino: inode number to search for
1383 * Search for the inode @ino in the inode cache, and if the inode is in the
1384 * cache, the inode is returned with an incremented reference count.
1386 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1388 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1389 struct inode *inode;
1391 spin_lock(&inode_hash_lock);
1392 inode = find_inode_fast(sb, head, ino);
1393 spin_unlock(&inode_hash_lock);
1398 wait_on_inode(inode);
1399 if (unlikely(inode_unhashed(inode))) {
1406 EXPORT_SYMBOL(ilookup);
1409 * find_inode_nowait - find an inode in the inode cache
1410 * @sb: super block of file system to search
1411 * @hashval: hash value (usually inode number) to search for
1412 * @match: callback used for comparisons between inodes
1413 * @data: opaque data pointer to pass to @match
1415 * Search for the inode specified by @hashval and @data in the inode
1416 * cache, where the helper function @match will return 0 if the inode
1417 * does not match, 1 if the inode does match, and -1 if the search
1418 * should be stopped. The @match function must be responsible for
1419 * taking the i_lock spin_lock and checking i_state for an inode being
1420 * freed or being initialized, and incrementing the reference count
1421 * before returning 1. It also must not sleep, since it is called with
1422 * the inode_hash_lock spinlock held.
1424 * This is a even more generalized version of ilookup5() when the
1425 * function must never block --- find_inode() can block in
1426 * __wait_on_freeing_inode() --- or when the caller can not increment
1427 * the reference count because the resulting iput() might cause an
1428 * inode eviction. The tradeoff is that the @match funtion must be
1429 * very carefully implemented.
1431 struct inode *find_inode_nowait(struct super_block *sb,
1432 unsigned long hashval,
1433 int (*match)(struct inode *, unsigned long,
1437 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1438 struct inode *inode, *ret_inode = NULL;
1441 spin_lock(&inode_hash_lock);
1442 hlist_for_each_entry(inode, head, i_hash) {
1443 if (inode->i_sb != sb)
1445 mval = match(inode, hashval, data);
1453 spin_unlock(&inode_hash_lock);
1456 EXPORT_SYMBOL(find_inode_nowait);
1459 * find_inode_rcu - find an inode in the inode cache
1460 * @sb: Super block of file system to search
1461 * @hashval: Key to hash
1462 * @test: Function to test match on an inode
1463 * @data: Data for test function
1465 * Search for the inode specified by @hashval and @data in the inode cache,
1466 * where the helper function @test will return 0 if the inode does not match
1467 * and 1 if it does. The @test function must be responsible for taking the
1468 * i_lock spin_lock and checking i_state for an inode being freed or being
1471 * If successful, this will return the inode for which the @test function
1472 * returned 1 and NULL otherwise.
1474 * The @test function is not permitted to take a ref on any inode presented.
1475 * It is also not permitted to sleep.
1477 * The caller must hold the RCU read lock.
1479 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1480 int (*test)(struct inode *, void *), void *data)
1482 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1483 struct inode *inode;
1485 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1486 "suspicious find_inode_rcu() usage");
1488 hlist_for_each_entry_rcu(inode, head, i_hash) {
1489 if (inode->i_sb == sb &&
1490 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1496 EXPORT_SYMBOL(find_inode_rcu);
1499 * find_inode_by_rcu - Find an inode in the inode cache
1500 * @sb: Super block of file system to search
1501 * @ino: The inode number to match
1503 * Search for the inode specified by @hashval and @data in the inode cache,
1504 * where the helper function @test will return 0 if the inode does not match
1505 * and 1 if it does. The @test function must be responsible for taking the
1506 * i_lock spin_lock and checking i_state for an inode being freed or being
1509 * If successful, this will return the inode for which the @test function
1510 * returned 1 and NULL otherwise.
1512 * The @test function is not permitted to take a ref on any inode presented.
1513 * It is also not permitted to sleep.
1515 * The caller must hold the RCU read lock.
1517 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1520 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1521 struct inode *inode;
1523 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1524 "suspicious find_inode_by_ino_rcu() usage");
1526 hlist_for_each_entry_rcu(inode, head, i_hash) {
1527 if (inode->i_ino == ino &&
1528 inode->i_sb == sb &&
1529 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1534 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1536 int insert_inode_locked(struct inode *inode)
1538 struct super_block *sb = inode->i_sb;
1539 ino_t ino = inode->i_ino;
1540 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1543 struct inode *old = NULL;
1544 spin_lock(&inode_hash_lock);
1545 hlist_for_each_entry(old, head, i_hash) {
1546 if (old->i_ino != ino)
1548 if (old->i_sb != sb)
1550 spin_lock(&old->i_lock);
1551 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1552 spin_unlock(&old->i_lock);
1558 spin_lock(&inode->i_lock);
1559 inode->i_state |= I_NEW | I_CREATING;
1560 hlist_add_head_rcu(&inode->i_hash, head);
1561 spin_unlock(&inode->i_lock);
1562 spin_unlock(&inode_hash_lock);
1565 if (unlikely(old->i_state & I_CREATING)) {
1566 spin_unlock(&old->i_lock);
1567 spin_unlock(&inode_hash_lock);
1571 spin_unlock(&old->i_lock);
1572 spin_unlock(&inode_hash_lock);
1574 if (unlikely(!inode_unhashed(old))) {
1581 EXPORT_SYMBOL(insert_inode_locked);
1583 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1584 int (*test)(struct inode *, void *), void *data)
1588 inode->i_state |= I_CREATING;
1589 old = inode_insert5(inode, hashval, test, NULL, data);
1597 EXPORT_SYMBOL(insert_inode_locked4);
1600 int generic_delete_inode(struct inode *inode)
1604 EXPORT_SYMBOL(generic_delete_inode);
1607 * Called when we're dropping the last reference
1610 * Call the FS "drop_inode()" function, defaulting to
1611 * the legacy UNIX filesystem behaviour. If it tells
1612 * us to evict inode, do so. Otherwise, retain inode
1613 * in cache if fs is alive, sync and evict if fs is
1616 static void iput_final(struct inode *inode)
1618 struct super_block *sb = inode->i_sb;
1619 const struct super_operations *op = inode->i_sb->s_op;
1620 unsigned long state;
1623 WARN_ON(inode->i_state & I_NEW);
1626 drop = op->drop_inode(inode);
1628 drop = generic_drop_inode(inode);
1630 if (!drop && (sb->s_flags & SB_ACTIVE)) {
1631 inode_add_lru(inode);
1632 spin_unlock(&inode->i_lock);
1636 state = inode->i_state;
1638 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1639 spin_unlock(&inode->i_lock);
1641 write_inode_now(inode, 1);
1643 spin_lock(&inode->i_lock);
1644 state = inode->i_state;
1645 WARN_ON(state & I_NEW);
1646 state &= ~I_WILL_FREE;
1649 WRITE_ONCE(inode->i_state, state | I_FREEING);
1650 if (!list_empty(&inode->i_lru))
1651 inode_lru_list_del(inode);
1652 spin_unlock(&inode->i_lock);
1658 * iput - put an inode
1659 * @inode: inode to put
1661 * Puts an inode, dropping its usage count. If the inode use count hits
1662 * zero, the inode is then freed and may also be destroyed.
1664 * Consequently, iput() can sleep.
1666 void iput(struct inode *inode)
1670 BUG_ON(inode->i_state & I_CLEAR);
1672 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1673 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1674 atomic_inc(&inode->i_count);
1675 spin_unlock(&inode->i_lock);
1676 trace_writeback_lazytime_iput(inode);
1677 mark_inode_dirty_sync(inode);
1683 EXPORT_SYMBOL(iput);
1687 * bmap - find a block number in a file
1688 * @inode: inode owning the block number being requested
1689 * @block: pointer containing the block to find
1691 * Replaces the value in ``*block`` with the block number on the device holding
1692 * corresponding to the requested block number in the file.
1693 * That is, asked for block 4 of inode 1 the function will replace the
1694 * 4 in ``*block``, with disk block relative to the disk start that holds that
1695 * block of the file.
1697 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1698 * hole, returns 0 and ``*block`` is also set to 0.
1700 int bmap(struct inode *inode, sector_t *block)
1702 if (!inode->i_mapping->a_ops->bmap)
1705 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1708 EXPORT_SYMBOL(bmap);
1712 * With relative atime, only update atime if the previous atime is
1713 * earlier than either the ctime or mtime or if at least a day has
1714 * passed since the last atime update.
1716 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1717 struct timespec64 now)
1720 if (!(mnt->mnt_flags & MNT_RELATIME))
1723 * Is mtime younger than atime? If yes, update atime:
1725 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1728 * Is ctime younger than atime? If yes, update atime:
1730 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1734 * Is the previous atime value older than a day? If yes,
1737 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1740 * Good, we can skip the atime update:
1745 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1747 int iflags = I_DIRTY_TIME;
1750 if (flags & S_ATIME)
1751 inode->i_atime = *time;
1752 if (flags & S_VERSION)
1753 dirty = inode_maybe_inc_iversion(inode, false);
1754 if (flags & S_CTIME)
1755 inode->i_ctime = *time;
1756 if (flags & S_MTIME)
1757 inode->i_mtime = *time;
1758 if ((flags & (S_ATIME | S_CTIME | S_MTIME)) &&
1759 !(inode->i_sb->s_flags & SB_LAZYTIME))
1763 iflags |= I_DIRTY_SYNC;
1764 __mark_inode_dirty(inode, iflags);
1767 EXPORT_SYMBOL(generic_update_time);
1770 * This does the actual work of updating an inodes time or version. Must have
1771 * had called mnt_want_write() before calling this.
1773 static int update_time(struct inode *inode, struct timespec64 *time, int flags)
1775 if (inode->i_op->update_time)
1776 return inode->i_op->update_time(inode, time, flags);
1777 return generic_update_time(inode, time, flags);
1781 * touch_atime - update the access time
1782 * @path: the &struct path to update
1783 * @inode: inode to update
1785 * Update the accessed time on an inode and mark it for writeback.
1786 * This function automatically handles read only file systems and media,
1787 * as well as the "noatime" flag and inode specific "noatime" markers.
1789 bool atime_needs_update(const struct path *path, struct inode *inode)
1791 struct vfsmount *mnt = path->mnt;
1792 struct timespec64 now;
1794 if (inode->i_flags & S_NOATIME)
1797 /* Atime updates will likely cause i_uid and i_gid to be written
1798 * back improprely if their true value is unknown to the vfs.
1800 if (HAS_UNMAPPED_ID(inode))
1803 if (IS_NOATIME(inode))
1805 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1808 if (mnt->mnt_flags & MNT_NOATIME)
1810 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1813 now = current_time(inode);
1815 if (!relatime_need_update(mnt, inode, now))
1818 if (timespec64_equal(&inode->i_atime, &now))
1824 void touch_atime(const struct path *path)
1826 struct vfsmount *mnt = path->mnt;
1827 struct inode *inode = d_inode(path->dentry);
1828 struct timespec64 now;
1830 if (!atime_needs_update(path, inode))
1833 if (!sb_start_write_trylock(inode->i_sb))
1836 if (__mnt_want_write(mnt) != 0)
1839 * File systems can error out when updating inodes if they need to
1840 * allocate new space to modify an inode (such is the case for
1841 * Btrfs), but since we touch atime while walking down the path we
1842 * really don't care if we failed to update the atime of the file,
1843 * so just ignore the return value.
1844 * We may also fail on filesystems that have the ability to make parts
1845 * of the fs read only, e.g. subvolumes in Btrfs.
1847 now = current_time(inode);
1848 update_time(inode, &now, S_ATIME);
1849 __mnt_drop_write(mnt);
1851 sb_end_write(inode->i_sb);
1853 EXPORT_SYMBOL(touch_atime);
1856 * The logic we want is
1858 * if suid or (sgid and xgrp)
1861 int should_remove_suid(struct dentry *dentry)
1863 umode_t mode = d_inode(dentry)->i_mode;
1866 /* suid always must be killed */
1867 if (unlikely(mode & S_ISUID))
1868 kill = ATTR_KILL_SUID;
1871 * sgid without any exec bits is just a mandatory locking mark; leave
1872 * it alone. If some exec bits are set, it's a real sgid; kill it.
1874 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1875 kill |= ATTR_KILL_SGID;
1877 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1882 EXPORT_SYMBOL(should_remove_suid);
1885 * Return mask of changes for notify_change() that need to be done as a
1886 * response to write or truncate. Return 0 if nothing has to be changed.
1887 * Negative value on error (change should be denied).
1889 int dentry_needs_remove_privs(struct dentry *dentry)
1891 struct inode *inode = d_inode(dentry);
1895 if (IS_NOSEC(inode))
1898 mask = should_remove_suid(dentry);
1899 ret = security_inode_need_killpriv(dentry);
1903 mask |= ATTR_KILL_PRIV;
1907 static int __remove_privs(struct dentry *dentry, int kill)
1909 struct iattr newattrs;
1911 newattrs.ia_valid = ATTR_FORCE | kill;
1913 * Note we call this on write, so notify_change will not
1914 * encounter any conflicting delegations:
1916 return notify_change(dentry, &newattrs, NULL);
1920 * Remove special file priviledges (suid, capabilities) when file is written
1923 int file_remove_privs(struct file *file)
1925 struct dentry *dentry = file_dentry(file);
1926 struct inode *inode = file_inode(file);
1931 * Fast path for nothing security related.
1932 * As well for non-regular files, e.g. blkdev inodes.
1933 * For example, blkdev_write_iter() might get here
1934 * trying to remove privs which it is not allowed to.
1936 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1939 kill = dentry_needs_remove_privs(dentry);
1943 error = __remove_privs(dentry, kill);
1945 inode_has_no_xattr(inode);
1949 EXPORT_SYMBOL(file_remove_privs);
1952 * file_update_time - update mtime and ctime time
1953 * @file: file accessed
1955 * Update the mtime and ctime members of an inode and mark the inode
1956 * for writeback. Note that this function is meant exclusively for
1957 * usage in the file write path of filesystems, and filesystems may
1958 * choose to explicitly ignore update via this function with the
1959 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1960 * timestamps are handled by the server. This can return an error for
1961 * file systems who need to allocate space in order to update an inode.
1964 int file_update_time(struct file *file)
1966 struct inode *inode = file_inode(file);
1967 struct timespec64 now;
1971 /* First try to exhaust all avenues to not sync */
1972 if (IS_NOCMTIME(inode))
1975 now = current_time(inode);
1976 if (!timespec64_equal(&inode->i_mtime, &now))
1979 if (!timespec64_equal(&inode->i_ctime, &now))
1982 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
1983 sync_it |= S_VERSION;
1988 /* Finally allowed to write? Takes lock. */
1989 if (__mnt_want_write_file(file))
1992 ret = update_time(inode, &now, sync_it);
1993 __mnt_drop_write_file(file);
1997 EXPORT_SYMBOL(file_update_time);
1999 /* Caller must hold the file's inode lock */
2000 int file_modified(struct file *file)
2005 * Clear the security bits if the process is not being run by root.
2006 * This keeps people from modifying setuid and setgid binaries.
2008 err = file_remove_privs(file);
2012 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2015 return file_update_time(file);
2017 EXPORT_SYMBOL(file_modified);
2019 int inode_needs_sync(struct inode *inode)
2023 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2027 EXPORT_SYMBOL(inode_needs_sync);
2030 * If we try to find an inode in the inode hash while it is being
2031 * deleted, we have to wait until the filesystem completes its
2032 * deletion before reporting that it isn't found. This function waits
2033 * until the deletion _might_ have completed. Callers are responsible
2034 * to recheck inode state.
2036 * It doesn't matter if I_NEW is not set initially, a call to
2037 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2040 static void __wait_on_freeing_inode(struct inode *inode)
2042 wait_queue_head_t *wq;
2043 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2044 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2045 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2046 spin_unlock(&inode->i_lock);
2047 spin_unlock(&inode_hash_lock);
2049 finish_wait(wq, &wait.wq_entry);
2050 spin_lock(&inode_hash_lock);
2053 static __initdata unsigned long ihash_entries;
2054 static int __init set_ihash_entries(char *str)
2058 ihash_entries = simple_strtoul(str, &str, 0);
2061 __setup("ihash_entries=", set_ihash_entries);
2064 * Initialize the waitqueues and inode hash table.
2066 void __init inode_init_early(void)
2068 /* If hashes are distributed across NUMA nodes, defer
2069 * hash allocation until vmalloc space is available.
2075 alloc_large_system_hash("Inode-cache",
2076 sizeof(struct hlist_head),
2079 HASH_EARLY | HASH_ZERO,
2086 void __init inode_init(void)
2088 /* inode slab cache */
2089 inode_cachep = kmem_cache_create("inode_cache",
2090 sizeof(struct inode),
2092 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2093 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2096 /* Hash may have been set up in inode_init_early */
2101 alloc_large_system_hash("Inode-cache",
2102 sizeof(struct hlist_head),
2112 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2114 inode->i_mode = mode;
2115 if (S_ISCHR(mode)) {
2116 inode->i_fop = &def_chr_fops;
2117 inode->i_rdev = rdev;
2118 } else if (S_ISBLK(mode)) {
2119 inode->i_fop = &def_blk_fops;
2120 inode->i_rdev = rdev;
2121 } else if (S_ISFIFO(mode))
2122 inode->i_fop = &pipefifo_fops;
2123 else if (S_ISSOCK(mode))
2124 ; /* leave it no_open_fops */
2126 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2127 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2130 EXPORT_SYMBOL(init_special_inode);
2133 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2135 * @dir: Directory inode
2136 * @mode: mode of the new inode
2138 void inode_init_owner(struct inode *inode, const struct inode *dir,
2141 inode->i_uid = current_fsuid();
2142 if (dir && dir->i_mode & S_ISGID) {
2143 inode->i_gid = dir->i_gid;
2145 /* Directories are special, and always inherit S_ISGID */
2148 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2149 !in_group_p(inode->i_gid) &&
2150 !capable_wrt_inode_uidgid(dir, CAP_FSETID))
2153 inode->i_gid = current_fsgid();
2154 inode->i_mode = mode;
2156 EXPORT_SYMBOL(inode_init_owner);
2159 * inode_owner_or_capable - check current task permissions to inode
2160 * @inode: inode being checked
2162 * Return true if current either has CAP_FOWNER in a namespace with the
2163 * inode owner uid mapped, or owns the file.
2165 bool inode_owner_or_capable(const struct inode *inode)
2167 struct user_namespace *ns;
2169 if (uid_eq(current_fsuid(), inode->i_uid))
2172 ns = current_user_ns();
2173 if (kuid_has_mapping(ns, inode->i_uid) && ns_capable(ns, CAP_FOWNER))
2177 EXPORT_SYMBOL(inode_owner_or_capable);
2180 * Direct i/o helper functions
2182 static void __inode_dio_wait(struct inode *inode)
2184 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2185 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2188 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2189 if (atomic_read(&inode->i_dio_count))
2191 } while (atomic_read(&inode->i_dio_count));
2192 finish_wait(wq, &q.wq_entry);
2196 * inode_dio_wait - wait for outstanding DIO requests to finish
2197 * @inode: inode to wait for
2199 * Waits for all pending direct I/O requests to finish so that we can
2200 * proceed with a truncate or equivalent operation.
2202 * Must be called under a lock that serializes taking new references
2203 * to i_dio_count, usually by inode->i_mutex.
2205 void inode_dio_wait(struct inode *inode)
2207 if (atomic_read(&inode->i_dio_count))
2208 __inode_dio_wait(inode);
2210 EXPORT_SYMBOL(inode_dio_wait);
2213 * inode_set_flags - atomically set some inode flags
2215 * Note: the caller should be holding i_mutex, or else be sure that
2216 * they have exclusive access to the inode structure (i.e., while the
2217 * inode is being instantiated). The reason for the cmpxchg() loop
2218 * --- which wouldn't be necessary if all code paths which modify
2219 * i_flags actually followed this rule, is that there is at least one
2220 * code path which doesn't today so we use cmpxchg() out of an abundance
2223 * In the long run, i_mutex is overkill, and we should probably look
2224 * at using the i_lock spinlock to protect i_flags, and then make sure
2225 * it is so documented in include/linux/fs.h and that all code follows
2226 * the locking convention!!
2228 void inode_set_flags(struct inode *inode, unsigned int flags,
2231 WARN_ON_ONCE(flags & ~mask);
2232 set_mask_bits(&inode->i_flags, mask, flags);
2234 EXPORT_SYMBOL(inode_set_flags);
2236 void inode_nohighmem(struct inode *inode)
2238 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2240 EXPORT_SYMBOL(inode_nohighmem);
2243 * timestamp_truncate - Truncate timespec to a granularity
2245 * @inode: inode being updated
2247 * Truncate a timespec to the granularity supported by the fs
2248 * containing the inode. Always rounds down. gran must
2249 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2251 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2253 struct super_block *sb = inode->i_sb;
2254 unsigned int gran = sb->s_time_gran;
2256 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2257 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2260 /* Avoid division in the common cases 1 ns and 1 s. */
2263 else if (gran == NSEC_PER_SEC)
2265 else if (gran > 1 && gran < NSEC_PER_SEC)
2266 t.tv_nsec -= t.tv_nsec % gran;
2268 WARN(1, "invalid file time granularity: %u", gran);
2271 EXPORT_SYMBOL(timestamp_truncate);
2274 * current_time - Return FS time
2277 * Return the current time truncated to the time granularity supported by
2280 * Note that inode and inode->sb cannot be NULL.
2281 * Otherwise, the function warns and returns time without truncation.
2283 struct timespec64 current_time(struct inode *inode)
2285 struct timespec64 now;
2287 ktime_get_coarse_real_ts64(&now);
2289 if (unlikely(!inode->i_sb)) {
2290 WARN(1, "current_time() called with uninitialized super_block in the inode");
2294 return timestamp_truncate(now, inode);
2296 EXPORT_SYMBOL(current_time);
2299 * Generic function to check FS_IOC_SETFLAGS values and reject any invalid
2302 * Note: the caller should be holding i_mutex, or else be sure that they have
2303 * exclusive access to the inode structure.
2305 int vfs_ioc_setflags_prepare(struct inode *inode, unsigned int oldflags,
2309 * The IMMUTABLE and APPEND_ONLY flags can only be changed by
2310 * the relevant capability.
2312 * This test looks nicer. Thanks to Pauline Middelink
2314 if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL) &&
2315 !capable(CAP_LINUX_IMMUTABLE))
2318 return fscrypt_prepare_setflags(inode, oldflags, flags);
2320 EXPORT_SYMBOL(vfs_ioc_setflags_prepare);
2323 * Generic function to check FS_IOC_FSSETXATTR values and reject any invalid
2326 * Note: the caller should be holding i_mutex, or else be sure that they have
2327 * exclusive access to the inode structure.
2329 int vfs_ioc_fssetxattr_check(struct inode *inode, const struct fsxattr *old_fa,
2333 * Can't modify an immutable/append-only file unless we have
2334 * appropriate permission.
2336 if ((old_fa->fsx_xflags ^ fa->fsx_xflags) &
2337 (FS_XFLAG_IMMUTABLE | FS_XFLAG_APPEND) &&
2338 !capable(CAP_LINUX_IMMUTABLE))
2342 * Project Quota ID state is only allowed to change from within the init
2343 * namespace. Enforce that restriction only if we are trying to change
2344 * the quota ID state. Everything else is allowed in user namespaces.
2346 if (current_user_ns() != &init_user_ns) {
2347 if (old_fa->fsx_projid != fa->fsx_projid)
2349 if ((old_fa->fsx_xflags ^ fa->fsx_xflags) &
2350 FS_XFLAG_PROJINHERIT)
2354 /* Check extent size hints. */
2355 if ((fa->fsx_xflags & FS_XFLAG_EXTSIZE) && !S_ISREG(inode->i_mode))
2358 if ((fa->fsx_xflags & FS_XFLAG_EXTSZINHERIT) &&
2359 !S_ISDIR(inode->i_mode))
2362 if ((fa->fsx_xflags & FS_XFLAG_COWEXTSIZE) &&
2363 !S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2367 * It is only valid to set the DAX flag on regular files and
2368 * directories on filesystems.
2370 if ((fa->fsx_xflags & FS_XFLAG_DAX) &&
2371 !(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
2374 /* Extent size hints of zero turn off the flags. */
2375 if (fa->fsx_extsize == 0)
2376 fa->fsx_xflags &= ~(FS_XFLAG_EXTSIZE | FS_XFLAG_EXTSZINHERIT);
2377 if (fa->fsx_cowextsize == 0)
2378 fa->fsx_xflags &= ~FS_XFLAG_COWEXTSIZE;
2382 EXPORT_SYMBOL(vfs_ioc_fssetxattr_check);