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/fsnotify.h>
16 #include <linux/mount.h>
17 #include <linux/posix_acl.h>
18 #include <linux/prefetch.h>
19 #include <linux/buffer_head.h> /* for inode_has_buffers */
20 #include <linux/ratelimit.h>
21 #include <linux/list_lru.h>
22 #include <linux/iversion.h>
23 #include <trace/events/writeback.h>
27 * Inode locking rules:
29 * inode->i_lock protects:
30 * inode->i_state, inode->i_hash, __iget()
31 * Inode LRU list locks protect:
32 * inode->i_sb->s_inode_lru, inode->i_lru
33 * inode->i_sb->s_inode_list_lock protects:
34 * inode->i_sb->s_inodes, inode->i_sb_list
35 * bdi->wb.list_lock protects:
36 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
37 * inode_hash_lock protects:
38 * inode_hashtable, inode->i_hash
42 * inode->i_sb->s_inode_list_lock
44 * Inode LRU list locks
50 * inode->i_sb->s_inode_list_lock
57 static unsigned int i_hash_mask __read_mostly;
58 static unsigned int i_hash_shift __read_mostly;
59 static struct hlist_head *inode_hashtable __read_mostly;
60 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
63 * Empty aops. Can be used for the cases where the user does not
64 * define any of the address_space operations.
66 const struct address_space_operations empty_aops = {
68 EXPORT_SYMBOL(empty_aops);
71 * Statistics gathering..
73 struct inodes_stat_t inodes_stat;
75 static DEFINE_PER_CPU(unsigned long, nr_inodes);
76 static DEFINE_PER_CPU(unsigned long, nr_unused);
78 static struct kmem_cache *inode_cachep __read_mostly;
80 static long get_nr_inodes(void)
84 for_each_possible_cpu(i)
85 sum += per_cpu(nr_inodes, i);
86 return sum < 0 ? 0 : sum;
89 static inline long get_nr_inodes_unused(void)
93 for_each_possible_cpu(i)
94 sum += per_cpu(nr_unused, i);
95 return sum < 0 ? 0 : sum;
98 long get_nr_dirty_inodes(void)
100 /* not actually dirty inodes, but a wild approximation */
101 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
102 return nr_dirty > 0 ? nr_dirty : 0;
106 * Handle nr_inode sysctl
109 int proc_nr_inodes(struct ctl_table *table, int write,
110 void __user *buffer, size_t *lenp, loff_t *ppos)
112 inodes_stat.nr_inodes = get_nr_inodes();
113 inodes_stat.nr_unused = get_nr_inodes_unused();
114 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
118 static int no_open(struct inode *inode, struct file *file)
124 * inode_init_always - perform inode structure initialisation
125 * @sb: superblock inode belongs to
126 * @inode: inode to initialise
128 * These are initializations that need to be done on every inode
129 * allocation as the fields are not initialised by slab allocation.
131 int inode_init_always(struct super_block *sb, struct inode *inode)
133 static const struct inode_operations empty_iops;
134 static const struct file_operations no_open_fops = {.open = no_open};
135 struct address_space *const mapping = &inode->i_data;
138 inode->i_blkbits = sb->s_blocksize_bits;
140 atomic64_set(&inode->i_sequence, 0);
141 atomic_set(&inode->i_count, 1);
142 inode->i_op = &empty_iops;
143 inode->i_fop = &no_open_fops;
144 inode->__i_nlink = 1;
145 inode->i_opflags = 0;
147 inode->i_opflags |= IOP_XATTR;
148 i_uid_write(inode, 0);
149 i_gid_write(inode, 0);
150 atomic_set(&inode->i_writecount, 0);
152 inode->i_write_hint = WRITE_LIFE_NOT_SET;
155 inode->i_generation = 0;
156 inode->i_pipe = NULL;
157 inode->i_bdev = NULL;
158 inode->i_cdev = NULL;
159 inode->i_link = NULL;
160 inode->i_dir_seq = 0;
162 inode->dirtied_when = 0;
164 #ifdef CONFIG_CGROUP_WRITEBACK
165 inode->i_wb_frn_winner = 0;
166 inode->i_wb_frn_avg_time = 0;
167 inode->i_wb_frn_history = 0;
170 spin_lock_init(&inode->i_lock);
171 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
173 init_rwsem(&inode->i_rwsem);
174 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
176 atomic_set(&inode->i_dio_count, 0);
178 mapping->a_ops = &empty_aops;
179 mapping->host = inode;
182 atomic_set(&mapping->i_mmap_writable, 0);
183 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
184 atomic_set(&mapping->nr_thps, 0);
186 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
187 mapping->private_data = NULL;
188 mapping->writeback_index = 0;
189 inode->i_private = NULL;
190 inode->i_mapping = mapping;
191 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
192 #ifdef CONFIG_FS_POSIX_ACL
193 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
196 #ifdef CONFIG_FSNOTIFY
197 inode->i_fsnotify_mask = 0;
199 inode->i_flctx = NULL;
201 if (unlikely(security_inode_alloc(inode)))
203 this_cpu_inc(nr_inodes);
207 EXPORT_SYMBOL(inode_init_always);
209 void free_inode_nonrcu(struct inode *inode)
211 kmem_cache_free(inode_cachep, inode);
213 EXPORT_SYMBOL(free_inode_nonrcu);
215 static void i_callback(struct rcu_head *head)
217 struct inode *inode = container_of(head, struct inode, i_rcu);
218 if (inode->free_inode)
219 inode->free_inode(inode);
221 free_inode_nonrcu(inode);
224 static struct inode *alloc_inode(struct super_block *sb)
226 const struct super_operations *ops = sb->s_op;
229 if (ops->alloc_inode)
230 inode = ops->alloc_inode(sb);
232 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
237 if (unlikely(inode_init_always(sb, inode))) {
238 if (ops->destroy_inode) {
239 ops->destroy_inode(inode);
240 if (!ops->free_inode)
243 inode->free_inode = ops->free_inode;
244 i_callback(&inode->i_rcu);
251 void __destroy_inode(struct inode *inode)
253 BUG_ON(inode_has_buffers(inode));
254 inode_detach_wb(inode);
255 security_inode_free(inode);
256 fsnotify_inode_delete(inode);
257 locks_free_lock_context(inode);
258 if (!inode->i_nlink) {
259 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
260 atomic_long_dec(&inode->i_sb->s_remove_count);
263 #ifdef CONFIG_FS_POSIX_ACL
264 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
265 posix_acl_release(inode->i_acl);
266 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
267 posix_acl_release(inode->i_default_acl);
269 this_cpu_dec(nr_inodes);
271 EXPORT_SYMBOL(__destroy_inode);
273 static void destroy_inode(struct inode *inode)
275 const struct super_operations *ops = inode->i_sb->s_op;
277 BUG_ON(!list_empty(&inode->i_lru));
278 __destroy_inode(inode);
279 if (ops->destroy_inode) {
280 ops->destroy_inode(inode);
281 if (!ops->free_inode)
284 inode->free_inode = ops->free_inode;
285 call_rcu(&inode->i_rcu, i_callback);
289 * drop_nlink - directly drop an inode's link count
292 * This is a low-level filesystem helper to replace any
293 * direct filesystem manipulation of i_nlink. In cases
294 * where we are attempting to track writes to the
295 * filesystem, a decrement to zero means an imminent
296 * write when the file is truncated and actually unlinked
299 void drop_nlink(struct inode *inode)
301 WARN_ON(inode->i_nlink == 0);
304 atomic_long_inc(&inode->i_sb->s_remove_count);
306 EXPORT_SYMBOL(drop_nlink);
309 * clear_nlink - directly zero an inode's link count
312 * This is a low-level filesystem helper to replace any
313 * direct filesystem manipulation of i_nlink. See
314 * drop_nlink() for why we care about i_nlink hitting zero.
316 void clear_nlink(struct inode *inode)
318 if (inode->i_nlink) {
319 inode->__i_nlink = 0;
320 atomic_long_inc(&inode->i_sb->s_remove_count);
323 EXPORT_SYMBOL(clear_nlink);
326 * set_nlink - directly set an inode's link count
328 * @nlink: new nlink (should be non-zero)
330 * This is a low-level filesystem helper to replace any
331 * direct filesystem manipulation of i_nlink.
333 void set_nlink(struct inode *inode, unsigned int nlink)
338 /* Yes, some filesystems do change nlink from zero to one */
339 if (inode->i_nlink == 0)
340 atomic_long_dec(&inode->i_sb->s_remove_count);
342 inode->__i_nlink = nlink;
345 EXPORT_SYMBOL(set_nlink);
348 * inc_nlink - directly increment an inode's link count
351 * This is a low-level filesystem helper to replace any
352 * direct filesystem manipulation of i_nlink. Currently,
353 * it is only here for parity with dec_nlink().
355 void inc_nlink(struct inode *inode)
357 if (unlikely(inode->i_nlink == 0)) {
358 WARN_ON(!(inode->i_state & I_LINKABLE));
359 atomic_long_dec(&inode->i_sb->s_remove_count);
364 EXPORT_SYMBOL(inc_nlink);
366 static void __address_space_init_once(struct address_space *mapping)
368 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
369 init_rwsem(&mapping->i_mmap_rwsem);
370 INIT_LIST_HEAD(&mapping->private_list);
371 spin_lock_init(&mapping->private_lock);
372 mapping->i_mmap = RB_ROOT_CACHED;
375 void address_space_init_once(struct address_space *mapping)
377 memset(mapping, 0, sizeof(*mapping));
378 __address_space_init_once(mapping);
380 EXPORT_SYMBOL(address_space_init_once);
383 * These are initializations that only need to be done
384 * once, because the fields are idempotent across use
385 * of the inode, so let the slab aware of that.
387 void inode_init_once(struct inode *inode)
389 memset(inode, 0, sizeof(*inode));
390 INIT_HLIST_NODE(&inode->i_hash);
391 INIT_LIST_HEAD(&inode->i_devices);
392 INIT_LIST_HEAD(&inode->i_io_list);
393 INIT_LIST_HEAD(&inode->i_wb_list);
394 INIT_LIST_HEAD(&inode->i_lru);
395 __address_space_init_once(&inode->i_data);
396 i_size_ordered_init(inode);
398 EXPORT_SYMBOL(inode_init_once);
400 static void init_once(void *foo)
402 struct inode *inode = (struct inode *) foo;
404 inode_init_once(inode);
408 * inode->i_lock must be held
410 void __iget(struct inode *inode)
412 atomic_inc(&inode->i_count);
416 * get additional reference to inode; caller must already hold one.
418 void ihold(struct inode *inode)
420 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
422 EXPORT_SYMBOL(ihold);
424 static void inode_lru_list_add(struct inode *inode)
426 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
427 this_cpu_inc(nr_unused);
429 inode->i_state |= I_REFERENCED;
433 * Add inode to LRU if needed (inode is unused and clean).
435 * Needs inode->i_lock held.
437 void inode_add_lru(struct inode *inode)
439 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
440 I_FREEING | I_WILL_FREE)) &&
441 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
442 inode_lru_list_add(inode);
446 static void inode_lru_list_del(struct inode *inode)
449 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
450 this_cpu_dec(nr_unused);
454 * inode_sb_list_add - add inode to the superblock list of inodes
455 * @inode: inode to add
457 void inode_sb_list_add(struct inode *inode)
459 spin_lock(&inode->i_sb->s_inode_list_lock);
460 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
461 spin_unlock(&inode->i_sb->s_inode_list_lock);
463 EXPORT_SYMBOL_GPL(inode_sb_list_add);
465 static inline void inode_sb_list_del(struct inode *inode)
467 if (!list_empty(&inode->i_sb_list)) {
468 spin_lock(&inode->i_sb->s_inode_list_lock);
469 list_del_init(&inode->i_sb_list);
470 spin_unlock(&inode->i_sb->s_inode_list_lock);
474 static unsigned long hash(struct super_block *sb, unsigned long hashval)
478 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
480 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
481 return tmp & i_hash_mask;
485 * __insert_inode_hash - hash an inode
486 * @inode: unhashed inode
487 * @hashval: unsigned long value used to locate this object in the
490 * Add an inode to the inode hash for this superblock.
492 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
494 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
496 spin_lock(&inode_hash_lock);
497 spin_lock(&inode->i_lock);
498 hlist_add_head(&inode->i_hash, b);
499 spin_unlock(&inode->i_lock);
500 spin_unlock(&inode_hash_lock);
502 EXPORT_SYMBOL(__insert_inode_hash);
505 * __remove_inode_hash - remove an inode from the hash
506 * @inode: inode to unhash
508 * Remove an inode from the superblock.
510 void __remove_inode_hash(struct inode *inode)
512 spin_lock(&inode_hash_lock);
513 spin_lock(&inode->i_lock);
514 hlist_del_init(&inode->i_hash);
515 spin_unlock(&inode->i_lock);
516 spin_unlock(&inode_hash_lock);
518 EXPORT_SYMBOL(__remove_inode_hash);
520 void clear_inode(struct inode *inode)
523 * We have to cycle the i_pages lock here because reclaim can be in the
524 * process of removing the last page (in __delete_from_page_cache())
525 * and we must not free the mapping under it.
527 xa_lock_irq(&inode->i_data.i_pages);
528 BUG_ON(inode->i_data.nrpages);
529 BUG_ON(inode->i_data.nrexceptional);
530 xa_unlock_irq(&inode->i_data.i_pages);
531 BUG_ON(!list_empty(&inode->i_data.private_list));
532 BUG_ON(!(inode->i_state & I_FREEING));
533 BUG_ON(inode->i_state & I_CLEAR);
534 BUG_ON(!list_empty(&inode->i_wb_list));
535 /* don't need i_lock here, no concurrent mods to i_state */
536 inode->i_state = I_FREEING | I_CLEAR;
538 EXPORT_SYMBOL(clear_inode);
541 * Free the inode passed in, removing it from the lists it is still connected
542 * to. We remove any pages still attached to the inode and wait for any IO that
543 * is still in progress before finally destroying the inode.
545 * An inode must already be marked I_FREEING so that we avoid the inode being
546 * moved back onto lists if we race with other code that manipulates the lists
547 * (e.g. writeback_single_inode). The caller is responsible for setting this.
549 * An inode must already be removed from the LRU list before being evicted from
550 * the cache. This should occur atomically with setting the I_FREEING state
551 * flag, so no inodes here should ever be on the LRU when being evicted.
553 static void evict(struct inode *inode)
555 const struct super_operations *op = inode->i_sb->s_op;
557 BUG_ON(!(inode->i_state & I_FREEING));
558 BUG_ON(!list_empty(&inode->i_lru));
560 if (!list_empty(&inode->i_io_list))
561 inode_io_list_del(inode);
563 inode_sb_list_del(inode);
566 * Wait for flusher thread to be done with the inode so that filesystem
567 * does not start destroying it while writeback is still running. Since
568 * the inode has I_FREEING set, flusher thread won't start new work on
569 * the inode. We just have to wait for running writeback to finish.
571 inode_wait_for_writeback(inode);
573 if (op->evict_inode) {
574 op->evict_inode(inode);
576 truncate_inode_pages_final(&inode->i_data);
579 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
581 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
584 remove_inode_hash(inode);
586 spin_lock(&inode->i_lock);
587 wake_up_bit(&inode->i_state, __I_NEW);
588 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
589 spin_unlock(&inode->i_lock);
591 destroy_inode(inode);
595 * dispose_list - dispose of the contents of a local list
596 * @head: the head of the list to free
598 * Dispose-list gets a local list with local inodes in it, so it doesn't
599 * need to worry about list corruption and SMP locks.
601 static void dispose_list(struct list_head *head)
603 while (!list_empty(head)) {
606 inode = list_first_entry(head, struct inode, i_lru);
607 list_del_init(&inode->i_lru);
615 * evict_inodes - evict all evictable inodes for a superblock
616 * @sb: superblock to operate on
618 * Make sure that no inodes with zero refcount are retained. This is
619 * called by superblock shutdown after having SB_ACTIVE flag removed,
620 * so any inode reaching zero refcount during or after that call will
621 * be immediately evicted.
623 void evict_inodes(struct super_block *sb)
625 struct inode *inode, *next;
629 spin_lock(&sb->s_inode_list_lock);
630 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
631 if (atomic_read(&inode->i_count))
634 spin_lock(&inode->i_lock);
635 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
636 spin_unlock(&inode->i_lock);
640 inode->i_state |= I_FREEING;
641 inode_lru_list_del(inode);
642 spin_unlock(&inode->i_lock);
643 list_add(&inode->i_lru, &dispose);
646 * We can have a ton of inodes to evict at unmount time given
647 * enough memory, check to see if we need to go to sleep for a
648 * bit so we don't livelock.
650 if (need_resched()) {
651 spin_unlock(&sb->s_inode_list_lock);
653 dispose_list(&dispose);
657 spin_unlock(&sb->s_inode_list_lock);
659 dispose_list(&dispose);
661 EXPORT_SYMBOL_GPL(evict_inodes);
664 * invalidate_inodes - attempt to free all inodes on a superblock
665 * @sb: superblock to operate on
666 * @kill_dirty: flag to guide handling of dirty inodes
668 * Attempts to free all inodes for a given superblock. If there were any
669 * busy inodes return a non-zero value, else zero.
670 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
673 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
676 struct inode *inode, *next;
680 spin_lock(&sb->s_inode_list_lock);
681 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
682 spin_lock(&inode->i_lock);
683 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
684 spin_unlock(&inode->i_lock);
687 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
688 spin_unlock(&inode->i_lock);
692 if (atomic_read(&inode->i_count)) {
693 spin_unlock(&inode->i_lock);
698 inode->i_state |= I_FREEING;
699 inode_lru_list_del(inode);
700 spin_unlock(&inode->i_lock);
701 list_add(&inode->i_lru, &dispose);
702 if (need_resched()) {
703 spin_unlock(&sb->s_inode_list_lock);
705 dispose_list(&dispose);
709 spin_unlock(&sb->s_inode_list_lock);
711 dispose_list(&dispose);
717 * Isolate the inode from the LRU in preparation for freeing it.
719 * Any inodes which are pinned purely because of attached pagecache have their
720 * pagecache removed. If the inode has metadata buffers attached to
721 * mapping->private_list then try to remove them.
723 * If the inode has the I_REFERENCED flag set, then it means that it has been
724 * used recently - the flag is set in iput_final(). When we encounter such an
725 * inode, clear the flag and move it to the back of the LRU so it gets another
726 * pass through the LRU before it gets reclaimed. This is necessary because of
727 * the fact we are doing lazy LRU updates to minimise lock contention so the
728 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
729 * with this flag set because they are the inodes that are out of order.
731 static enum lru_status inode_lru_isolate(struct list_head *item,
732 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
734 struct list_head *freeable = arg;
735 struct inode *inode = container_of(item, struct inode, i_lru);
738 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
739 * If we fail to get the lock, just skip it.
741 if (!spin_trylock(&inode->i_lock))
745 * Referenced or dirty inodes are still in use. Give them another pass
746 * through the LRU as we canot reclaim them now.
748 if (atomic_read(&inode->i_count) ||
749 (inode->i_state & ~I_REFERENCED)) {
750 list_lru_isolate(lru, &inode->i_lru);
751 spin_unlock(&inode->i_lock);
752 this_cpu_dec(nr_unused);
756 /* recently referenced inodes get one more pass */
757 if (inode->i_state & I_REFERENCED) {
758 inode->i_state &= ~I_REFERENCED;
759 spin_unlock(&inode->i_lock);
763 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
765 spin_unlock(&inode->i_lock);
766 spin_unlock(lru_lock);
767 if (remove_inode_buffers(inode)) {
769 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
770 if (current_is_kswapd())
771 __count_vm_events(KSWAPD_INODESTEAL, reap);
773 __count_vm_events(PGINODESTEAL, reap);
774 if (current->reclaim_state)
775 current->reclaim_state->reclaimed_slab += reap;
782 WARN_ON(inode->i_state & I_NEW);
783 inode->i_state |= I_FREEING;
784 list_lru_isolate_move(lru, &inode->i_lru, freeable);
785 spin_unlock(&inode->i_lock);
787 this_cpu_dec(nr_unused);
792 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
793 * This is called from the superblock shrinker function with a number of inodes
794 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
795 * then are freed outside inode_lock by dispose_list().
797 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
802 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
803 inode_lru_isolate, &freeable);
804 dispose_list(&freeable);
808 static void __wait_on_freeing_inode(struct inode *inode);
810 * Called with the inode lock held.
812 static struct inode *find_inode(struct super_block *sb,
813 struct hlist_head *head,
814 int (*test)(struct inode *, void *),
817 struct inode *inode = NULL;
820 hlist_for_each_entry(inode, head, i_hash) {
821 if (inode->i_sb != sb)
823 if (!test(inode, data))
825 spin_lock(&inode->i_lock);
826 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
827 __wait_on_freeing_inode(inode);
830 if (unlikely(inode->i_state & I_CREATING)) {
831 spin_unlock(&inode->i_lock);
832 return ERR_PTR(-ESTALE);
835 spin_unlock(&inode->i_lock);
842 * find_inode_fast is the fast path version of find_inode, see the comment at
843 * iget_locked for details.
845 static struct inode *find_inode_fast(struct super_block *sb,
846 struct hlist_head *head, unsigned long ino)
848 struct inode *inode = NULL;
851 hlist_for_each_entry(inode, head, i_hash) {
852 if (inode->i_ino != ino)
854 if (inode->i_sb != sb)
856 spin_lock(&inode->i_lock);
857 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
858 __wait_on_freeing_inode(inode);
861 if (unlikely(inode->i_state & I_CREATING)) {
862 spin_unlock(&inode->i_lock);
863 return ERR_PTR(-ESTALE);
866 spin_unlock(&inode->i_lock);
873 * Each cpu owns a range of LAST_INO_BATCH numbers.
874 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
875 * to renew the exhausted range.
877 * This does not significantly increase overflow rate because every CPU can
878 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
879 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
880 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
881 * overflow rate by 2x, which does not seem too significant.
883 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
884 * error if st_ino won't fit in target struct field. Use 32bit counter
885 * here to attempt to avoid that.
887 #define LAST_INO_BATCH 1024
888 static DEFINE_PER_CPU(unsigned int, last_ino);
890 unsigned int get_next_ino(void)
892 unsigned int *p = &get_cpu_var(last_ino);
893 unsigned int res = *p;
896 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
897 static atomic_t shared_last_ino;
898 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
900 res = next - LAST_INO_BATCH;
905 /* get_next_ino should not provide a 0 inode number */
909 put_cpu_var(last_ino);
912 EXPORT_SYMBOL(get_next_ino);
915 * new_inode_pseudo - obtain an inode
918 * Allocates a new inode for given superblock.
919 * Inode wont be chained in superblock s_inodes list
921 * - fs can't be unmount
922 * - quotas, fsnotify, writeback can't work
924 struct inode *new_inode_pseudo(struct super_block *sb)
926 struct inode *inode = alloc_inode(sb);
929 spin_lock(&inode->i_lock);
931 spin_unlock(&inode->i_lock);
932 INIT_LIST_HEAD(&inode->i_sb_list);
938 * new_inode - obtain an inode
941 * Allocates a new inode for given superblock. The default gfp_mask
942 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
943 * If HIGHMEM pages are unsuitable or it is known that pages allocated
944 * for the page cache are not reclaimable or migratable,
945 * mapping_set_gfp_mask() must be called with suitable flags on the
946 * newly created inode's mapping
949 struct inode *new_inode(struct super_block *sb)
953 spin_lock_prefetch(&sb->s_inode_list_lock);
955 inode = new_inode_pseudo(sb);
957 inode_sb_list_add(inode);
960 EXPORT_SYMBOL(new_inode);
962 #ifdef CONFIG_DEBUG_LOCK_ALLOC
963 void lockdep_annotate_inode_mutex_key(struct inode *inode)
965 if (S_ISDIR(inode->i_mode)) {
966 struct file_system_type *type = inode->i_sb->s_type;
968 /* Set new key only if filesystem hasn't already changed it */
969 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
971 * ensure nobody is actually holding i_mutex
973 // mutex_destroy(&inode->i_mutex);
974 init_rwsem(&inode->i_rwsem);
975 lockdep_set_class(&inode->i_rwsem,
976 &type->i_mutex_dir_key);
980 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
984 * unlock_new_inode - clear the I_NEW state and wake up any waiters
985 * @inode: new inode to unlock
987 * Called when the inode is fully initialised to clear the new state of the
988 * inode and wake up anyone waiting for the inode to finish initialisation.
990 void unlock_new_inode(struct inode *inode)
992 lockdep_annotate_inode_mutex_key(inode);
993 spin_lock(&inode->i_lock);
994 WARN_ON(!(inode->i_state & I_NEW));
995 inode->i_state &= ~I_NEW & ~I_CREATING;
997 wake_up_bit(&inode->i_state, __I_NEW);
998 spin_unlock(&inode->i_lock);
1000 EXPORT_SYMBOL(unlock_new_inode);
1002 void discard_new_inode(struct inode *inode)
1004 lockdep_annotate_inode_mutex_key(inode);
1005 spin_lock(&inode->i_lock);
1006 WARN_ON(!(inode->i_state & I_NEW));
1007 inode->i_state &= ~I_NEW;
1009 wake_up_bit(&inode->i_state, __I_NEW);
1010 spin_unlock(&inode->i_lock);
1013 EXPORT_SYMBOL(discard_new_inode);
1016 * lock_two_inodes - lock two inodes (may be regular files but also dirs)
1018 * Lock any non-NULL argument. The caller must make sure that if he is passing
1019 * in two directories, one is not ancestor of the other. Zero, one or two
1020 * objects may be locked by this function.
1022 * @inode1: first inode to lock
1023 * @inode2: second inode to lock
1024 * @subclass1: inode lock subclass for the first lock obtained
1025 * @subclass2: inode lock subclass for the second lock obtained
1027 void lock_two_inodes(struct inode *inode1, struct inode *inode2,
1028 unsigned subclass1, unsigned subclass2)
1030 if (!inode1 || !inode2) {
1032 * Make sure @subclass1 will be used for the acquired lock.
1033 * This is not strictly necessary (no current caller cares) but
1034 * let's keep things consistent.
1037 swap(inode1, inode2);
1042 * If one object is directory and the other is not, we must make sure
1043 * to lock directory first as the other object may be its child.
1045 if (S_ISDIR(inode2->i_mode) == S_ISDIR(inode1->i_mode)) {
1046 if (inode1 > inode2)
1047 swap(inode1, inode2);
1048 } else if (!S_ISDIR(inode1->i_mode))
1049 swap(inode1, inode2);
1052 inode_lock_nested(inode1, subclass1);
1053 if (inode2 && inode2 != inode1)
1054 inode_lock_nested(inode2, subclass2);
1058 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1060 * Lock any non-NULL argument that is not a directory.
1061 * Zero, one or two objects may be locked by this function.
1063 * @inode1: first inode to lock
1064 * @inode2: second inode to lock
1066 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1068 if (inode1 > inode2)
1069 swap(inode1, inode2);
1071 if (inode1 && !S_ISDIR(inode1->i_mode))
1073 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1074 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1076 EXPORT_SYMBOL(lock_two_nondirectories);
1079 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1080 * @inode1: first inode to unlock
1081 * @inode2: second inode to unlock
1083 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1085 if (inode1 && !S_ISDIR(inode1->i_mode))
1086 inode_unlock(inode1);
1087 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1088 inode_unlock(inode2);
1090 EXPORT_SYMBOL(unlock_two_nondirectories);
1093 * inode_insert5 - obtain an inode from a mounted file system
1094 * @inode: pre-allocated inode to use for insert to cache
1095 * @hashval: hash value (usually inode number) to get
1096 * @test: callback used for comparisons between inodes
1097 * @set: callback used to initialize a new struct inode
1098 * @data: opaque data pointer to pass to @test and @set
1100 * Search for the inode specified by @hashval and @data in the inode cache,
1101 * and if present it is return it with an increased reference count. This is
1102 * a variant of iget5_locked() for callers that don't want to fail on memory
1103 * allocation of inode.
1105 * If the inode is not in cache, insert the pre-allocated inode to cache and
1106 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1107 * to fill it in before unlocking it via unlock_new_inode().
1109 * Note both @test and @set are called with the inode_hash_lock held, so can't
1112 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1113 int (*test)(struct inode *, void *),
1114 int (*set)(struct inode *, void *), void *data)
1116 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1118 bool creating = inode->i_state & I_CREATING;
1121 spin_lock(&inode_hash_lock);
1122 old = find_inode(inode->i_sb, head, test, data);
1123 if (unlikely(old)) {
1125 * Uhhuh, somebody else created the same inode under us.
1126 * Use the old inode instead of the preallocated one.
1128 spin_unlock(&inode_hash_lock);
1132 if (unlikely(inode_unhashed(old))) {
1139 if (set && unlikely(set(inode, data))) {
1145 * Return the locked inode with I_NEW set, the
1146 * caller is responsible for filling in the contents
1148 spin_lock(&inode->i_lock);
1149 inode->i_state |= I_NEW;
1150 hlist_add_head(&inode->i_hash, head);
1151 spin_unlock(&inode->i_lock);
1153 inode_sb_list_add(inode);
1155 spin_unlock(&inode_hash_lock);
1159 EXPORT_SYMBOL(inode_insert5);
1162 * iget5_locked - obtain an inode from a mounted file system
1163 * @sb: super block of file system
1164 * @hashval: hash value (usually inode number) to get
1165 * @test: callback used for comparisons between inodes
1166 * @set: callback used to initialize a new struct inode
1167 * @data: opaque data pointer to pass to @test and @set
1169 * Search for the inode specified by @hashval and @data in the inode cache,
1170 * and if present it is return it with an increased reference count. This is
1171 * a generalized version of iget_locked() for file systems where the inode
1172 * number is not sufficient for unique identification of an inode.
1174 * If the inode is not in cache, allocate a new inode and return it locked,
1175 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1176 * before unlocking it via unlock_new_inode().
1178 * Note both @test and @set are called with the inode_hash_lock held, so can't
1181 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1182 int (*test)(struct inode *, void *),
1183 int (*set)(struct inode *, void *), void *data)
1185 struct inode *inode = ilookup5(sb, hashval, test, data);
1188 struct inode *new = alloc_inode(sb);
1192 inode = inode_insert5(new, hashval, test, set, data);
1193 if (unlikely(inode != new))
1199 EXPORT_SYMBOL(iget5_locked);
1202 * iget_locked - obtain an inode from a mounted file system
1203 * @sb: super block of file system
1204 * @ino: inode number to get
1206 * Search for the inode specified by @ino in the inode cache and if present
1207 * return it with an increased reference count. This is for file systems
1208 * where the inode number is sufficient for unique identification of an inode.
1210 * If the inode is not in cache, allocate a new inode and return it locked,
1211 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1212 * before unlocking it via unlock_new_inode().
1214 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1216 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1217 struct inode *inode;
1219 spin_lock(&inode_hash_lock);
1220 inode = find_inode_fast(sb, head, ino);
1221 spin_unlock(&inode_hash_lock);
1225 wait_on_inode(inode);
1226 if (unlikely(inode_unhashed(inode))) {
1233 inode = alloc_inode(sb);
1237 spin_lock(&inode_hash_lock);
1238 /* We released the lock, so.. */
1239 old = find_inode_fast(sb, head, ino);
1242 spin_lock(&inode->i_lock);
1243 inode->i_state = I_NEW;
1244 hlist_add_head(&inode->i_hash, head);
1245 spin_unlock(&inode->i_lock);
1246 inode_sb_list_add(inode);
1247 spin_unlock(&inode_hash_lock);
1249 /* Return the locked inode with I_NEW set, the
1250 * caller is responsible for filling in the contents
1256 * Uhhuh, somebody else created the same inode under
1257 * us. Use the old inode instead of the one we just
1260 spin_unlock(&inode_hash_lock);
1261 destroy_inode(inode);
1265 wait_on_inode(inode);
1266 if (unlikely(inode_unhashed(inode))) {
1273 EXPORT_SYMBOL(iget_locked);
1276 * search the inode cache for a matching inode number.
1277 * If we find one, then the inode number we are trying to
1278 * allocate is not unique and so we should not use it.
1280 * Returns 1 if the inode number is unique, 0 if it is not.
1282 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1284 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1285 struct inode *inode;
1287 spin_lock(&inode_hash_lock);
1288 hlist_for_each_entry(inode, b, i_hash) {
1289 if (inode->i_ino == ino && inode->i_sb == sb) {
1290 spin_unlock(&inode_hash_lock);
1294 spin_unlock(&inode_hash_lock);
1300 * iunique - get a unique inode number
1302 * @max_reserved: highest reserved inode number
1304 * Obtain an inode number that is unique on the system for a given
1305 * superblock. This is used by file systems that have no natural
1306 * permanent inode numbering system. An inode number is returned that
1307 * is higher than the reserved limit but unique.
1310 * With a large number of inodes live on the file system this function
1311 * currently becomes quite slow.
1313 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1316 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1317 * error if st_ino won't fit in target struct field. Use 32bit counter
1318 * here to attempt to avoid that.
1320 static DEFINE_SPINLOCK(iunique_lock);
1321 static unsigned int counter;
1324 spin_lock(&iunique_lock);
1326 if (counter <= max_reserved)
1327 counter = max_reserved + 1;
1329 } while (!test_inode_iunique(sb, res));
1330 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);
1499 int insert_inode_locked(struct inode *inode)
1501 struct super_block *sb = inode->i_sb;
1502 ino_t ino = inode->i_ino;
1503 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1506 struct inode *old = NULL;
1507 spin_lock(&inode_hash_lock);
1508 hlist_for_each_entry(old, head, i_hash) {
1509 if (old->i_ino != ino)
1511 if (old->i_sb != sb)
1513 spin_lock(&old->i_lock);
1514 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1515 spin_unlock(&old->i_lock);
1521 spin_lock(&inode->i_lock);
1522 inode->i_state |= I_NEW | I_CREATING;
1523 hlist_add_head(&inode->i_hash, head);
1524 spin_unlock(&inode->i_lock);
1525 spin_unlock(&inode_hash_lock);
1528 if (unlikely(old->i_state & I_CREATING)) {
1529 spin_unlock(&old->i_lock);
1530 spin_unlock(&inode_hash_lock);
1534 spin_unlock(&old->i_lock);
1535 spin_unlock(&inode_hash_lock);
1537 if (unlikely(!inode_unhashed(old))) {
1544 EXPORT_SYMBOL(insert_inode_locked);
1546 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1547 int (*test)(struct inode *, void *), void *data)
1551 inode->i_state |= I_CREATING;
1552 old = inode_insert5(inode, hashval, test, NULL, data);
1560 EXPORT_SYMBOL(insert_inode_locked4);
1563 int generic_delete_inode(struct inode *inode)
1567 EXPORT_SYMBOL(generic_delete_inode);
1570 * Called when we're dropping the last reference
1573 * Call the FS "drop_inode()" function, defaulting to
1574 * the legacy UNIX filesystem behaviour. If it tells
1575 * us to evict inode, do so. Otherwise, retain inode
1576 * in cache if fs is alive, sync and evict if fs is
1579 static void iput_final(struct inode *inode)
1581 struct super_block *sb = inode->i_sb;
1582 const struct super_operations *op = inode->i_sb->s_op;
1585 WARN_ON(inode->i_state & I_NEW);
1588 drop = op->drop_inode(inode);
1590 drop = generic_drop_inode(inode);
1592 if (!drop && (sb->s_flags & SB_ACTIVE)) {
1593 inode_add_lru(inode);
1594 spin_unlock(&inode->i_lock);
1599 inode->i_state |= I_WILL_FREE;
1600 spin_unlock(&inode->i_lock);
1601 write_inode_now(inode, 1);
1602 spin_lock(&inode->i_lock);
1603 WARN_ON(inode->i_state & I_NEW);
1604 inode->i_state &= ~I_WILL_FREE;
1607 inode->i_state |= I_FREEING;
1608 if (!list_empty(&inode->i_lru))
1609 inode_lru_list_del(inode);
1610 spin_unlock(&inode->i_lock);
1616 * iput - put an inode
1617 * @inode: inode to put
1619 * Puts an inode, dropping its usage count. If the inode use count hits
1620 * zero, the inode is then freed and may also be destroyed.
1622 * Consequently, iput() can sleep.
1624 void iput(struct inode *inode)
1628 BUG_ON(inode->i_state & I_CLEAR);
1630 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1631 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1632 atomic_inc(&inode->i_count);
1633 spin_unlock(&inode->i_lock);
1634 trace_writeback_lazytime_iput(inode);
1635 mark_inode_dirty_sync(inode);
1641 EXPORT_SYMBOL(iput);
1644 * bmap - find a block number in a file
1645 * @inode: inode of file
1646 * @block: block to find
1648 * Returns the block number on the device holding the inode that
1649 * is the disk block number for the block of the file requested.
1650 * That is, asked for block 4 of inode 1 the function will return the
1651 * disk block relative to the disk start that holds that block of the
1654 sector_t bmap(struct inode *inode, sector_t block)
1657 if (inode->i_mapping->a_ops->bmap)
1658 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1661 EXPORT_SYMBOL(bmap);
1664 * With relative atime, only update atime if the previous atime is
1665 * earlier than either the ctime or mtime or if at least a day has
1666 * passed since the last atime update.
1668 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1669 struct timespec64 now)
1672 if (!(mnt->mnt_flags & MNT_RELATIME))
1675 * Is mtime younger than atime? If yes, update atime:
1677 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1680 * Is ctime younger than atime? If yes, update atime:
1682 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1686 * Is the previous atime value older than a day? If yes,
1689 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1692 * Good, we can skip the atime update:
1697 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1699 int iflags = I_DIRTY_TIME;
1702 if (flags & S_ATIME)
1703 inode->i_atime = *time;
1704 if (flags & S_VERSION)
1705 dirty = inode_maybe_inc_iversion(inode, false);
1706 if (flags & S_CTIME)
1707 inode->i_ctime = *time;
1708 if (flags & S_MTIME)
1709 inode->i_mtime = *time;
1710 if ((flags & (S_ATIME | S_CTIME | S_MTIME)) &&
1711 !(inode->i_sb->s_flags & SB_LAZYTIME))
1715 iflags |= I_DIRTY_SYNC;
1716 __mark_inode_dirty(inode, iflags);
1719 EXPORT_SYMBOL(generic_update_time);
1722 * This does the actual work of updating an inodes time or version. Must have
1723 * had called mnt_want_write() before calling this.
1725 static int update_time(struct inode *inode, struct timespec64 *time, int flags)
1727 int (*update_time)(struct inode *, struct timespec64 *, int);
1729 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1730 generic_update_time;
1732 return update_time(inode, time, flags);
1736 * touch_atime - update the access time
1737 * @path: the &struct path to update
1738 * @inode: inode to update
1740 * Update the accessed time on an inode and mark it for writeback.
1741 * This function automatically handles read only file systems and media,
1742 * as well as the "noatime" flag and inode specific "noatime" markers.
1744 bool atime_needs_update(const struct path *path, struct inode *inode)
1746 struct vfsmount *mnt = path->mnt;
1747 struct timespec64 now;
1749 if (inode->i_flags & S_NOATIME)
1752 /* Atime updates will likely cause i_uid and i_gid to be written
1753 * back improprely if their true value is unknown to the vfs.
1755 if (HAS_UNMAPPED_ID(inode))
1758 if (IS_NOATIME(inode))
1760 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1763 if (mnt->mnt_flags & MNT_NOATIME)
1765 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1768 now = current_time(inode);
1770 if (!relatime_need_update(mnt, inode, now))
1773 if (timespec64_equal(&inode->i_atime, &now))
1779 void touch_atime(const struct path *path)
1781 struct vfsmount *mnt = path->mnt;
1782 struct inode *inode = d_inode(path->dentry);
1783 struct timespec64 now;
1785 if (!atime_needs_update(path, inode))
1788 if (!sb_start_write_trylock(inode->i_sb))
1791 if (__mnt_want_write(mnt) != 0)
1794 * File systems can error out when updating inodes if they need to
1795 * allocate new space to modify an inode (such is the case for
1796 * Btrfs), but since we touch atime while walking down the path we
1797 * really don't care if we failed to update the atime of the file,
1798 * so just ignore the return value.
1799 * We may also fail on filesystems that have the ability to make parts
1800 * of the fs read only, e.g. subvolumes in Btrfs.
1802 now = current_time(inode);
1803 update_time(inode, &now, S_ATIME);
1804 __mnt_drop_write(mnt);
1806 sb_end_write(inode->i_sb);
1808 EXPORT_SYMBOL(touch_atime);
1811 * The logic we want is
1813 * if suid or (sgid and xgrp)
1816 int should_remove_suid(struct dentry *dentry)
1818 umode_t mode = d_inode(dentry)->i_mode;
1821 /* suid always must be killed */
1822 if (unlikely(mode & S_ISUID))
1823 kill = ATTR_KILL_SUID;
1826 * sgid without any exec bits is just a mandatory locking mark; leave
1827 * it alone. If some exec bits are set, it's a real sgid; kill it.
1829 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1830 kill |= ATTR_KILL_SGID;
1832 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1837 EXPORT_SYMBOL(should_remove_suid);
1840 * Return mask of changes for notify_change() that need to be done as a
1841 * response to write or truncate. Return 0 if nothing has to be changed.
1842 * Negative value on error (change should be denied).
1844 int dentry_needs_remove_privs(struct dentry *dentry)
1846 struct inode *inode = d_inode(dentry);
1850 if (IS_NOSEC(inode))
1853 mask = should_remove_suid(dentry);
1854 ret = security_inode_need_killpriv(dentry);
1858 mask |= ATTR_KILL_PRIV;
1862 static int __remove_privs(struct dentry *dentry, int kill)
1864 struct iattr newattrs;
1866 newattrs.ia_valid = ATTR_FORCE | kill;
1868 * Note we call this on write, so notify_change will not
1869 * encounter any conflicting delegations:
1871 return notify_change(dentry, &newattrs, NULL);
1875 * Remove special file priviledges (suid, capabilities) when file is written
1878 int file_remove_privs(struct file *file)
1880 struct dentry *dentry = file_dentry(file);
1881 struct inode *inode = file_inode(file);
1886 * Fast path for nothing security related.
1887 * As well for non-regular files, e.g. blkdev inodes.
1888 * For example, blkdev_write_iter() might get here
1889 * trying to remove privs which it is not allowed to.
1891 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1894 kill = dentry_needs_remove_privs(dentry);
1898 error = __remove_privs(dentry, kill);
1900 inode_has_no_xattr(inode);
1904 EXPORT_SYMBOL(file_remove_privs);
1907 * file_update_time - update mtime and ctime time
1908 * @file: file accessed
1910 * Update the mtime and ctime members of an inode and mark the inode
1911 * for writeback. Note that this function is meant exclusively for
1912 * usage in the file write path of filesystems, and filesystems may
1913 * choose to explicitly ignore update via this function with the
1914 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1915 * timestamps are handled by the server. This can return an error for
1916 * file systems who need to allocate space in order to update an inode.
1919 int file_update_time(struct file *file)
1921 struct inode *inode = file_inode(file);
1922 struct timespec64 now;
1926 /* First try to exhaust all avenues to not sync */
1927 if (IS_NOCMTIME(inode))
1930 now = current_time(inode);
1931 if (!timespec64_equal(&inode->i_mtime, &now))
1934 if (!timespec64_equal(&inode->i_ctime, &now))
1937 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
1938 sync_it |= S_VERSION;
1943 /* Finally allowed to write? Takes lock. */
1944 if (__mnt_want_write_file(file))
1947 ret = update_time(inode, &now, sync_it);
1948 __mnt_drop_write_file(file);
1952 EXPORT_SYMBOL(file_update_time);
1954 /* Caller must hold the file's inode lock */
1955 int file_modified(struct file *file)
1960 * Clear the security bits if the process is not being run by root.
1961 * This keeps people from modifying setuid and setgid binaries.
1963 err = file_remove_privs(file);
1967 if (unlikely(file->f_mode & FMODE_NOCMTIME))
1970 return file_update_time(file);
1972 EXPORT_SYMBOL(file_modified);
1974 int inode_needs_sync(struct inode *inode)
1978 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1982 EXPORT_SYMBOL(inode_needs_sync);
1985 * If we try to find an inode in the inode hash while it is being
1986 * deleted, we have to wait until the filesystem completes its
1987 * deletion before reporting that it isn't found. This function waits
1988 * until the deletion _might_ have completed. Callers are responsible
1989 * to recheck inode state.
1991 * It doesn't matter if I_NEW is not set initially, a call to
1992 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1995 static void __wait_on_freeing_inode(struct inode *inode)
1997 wait_queue_head_t *wq;
1998 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1999 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2000 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2001 spin_unlock(&inode->i_lock);
2002 spin_unlock(&inode_hash_lock);
2004 finish_wait(wq, &wait.wq_entry);
2005 spin_lock(&inode_hash_lock);
2008 static __initdata unsigned long ihash_entries;
2009 static int __init set_ihash_entries(char *str)
2013 ihash_entries = simple_strtoul(str, &str, 0);
2016 __setup("ihash_entries=", set_ihash_entries);
2019 * Initialize the waitqueues and inode hash table.
2021 void __init inode_init_early(void)
2023 /* If hashes are distributed across NUMA nodes, defer
2024 * hash allocation until vmalloc space is available.
2030 alloc_large_system_hash("Inode-cache",
2031 sizeof(struct hlist_head),
2034 HASH_EARLY | HASH_ZERO,
2041 void __init inode_init(void)
2043 /* inode slab cache */
2044 inode_cachep = kmem_cache_create("inode_cache",
2045 sizeof(struct inode),
2047 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2048 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2051 /* Hash may have been set up in inode_init_early */
2056 alloc_large_system_hash("Inode-cache",
2057 sizeof(struct hlist_head),
2067 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2069 inode->i_mode = mode;
2070 if (S_ISCHR(mode)) {
2071 inode->i_fop = &def_chr_fops;
2072 inode->i_rdev = rdev;
2073 } else if (S_ISBLK(mode)) {
2074 inode->i_fop = &def_blk_fops;
2075 inode->i_rdev = rdev;
2076 } else if (S_ISFIFO(mode))
2077 inode->i_fop = &pipefifo_fops;
2078 else if (S_ISSOCK(mode))
2079 ; /* leave it no_open_fops */
2081 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2082 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2085 EXPORT_SYMBOL(init_special_inode);
2088 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2090 * @dir: Directory inode
2091 * @mode: mode of the new inode
2093 void inode_init_owner(struct inode *inode, const struct inode *dir,
2096 inode->i_uid = current_fsuid();
2097 if (dir && dir->i_mode & S_ISGID) {
2098 inode->i_gid = dir->i_gid;
2100 /* Directories are special, and always inherit S_ISGID */
2103 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2104 !in_group_p(inode->i_gid) &&
2105 !capable_wrt_inode_uidgid(dir, CAP_FSETID))
2108 inode->i_gid = current_fsgid();
2109 inode->i_mode = mode;
2111 EXPORT_SYMBOL(inode_init_owner);
2114 * inode_owner_or_capable - check current task permissions to inode
2115 * @inode: inode being checked
2117 * Return true if current either has CAP_FOWNER in a namespace with the
2118 * inode owner uid mapped, or owns the file.
2120 bool inode_owner_or_capable(const struct inode *inode)
2122 struct user_namespace *ns;
2124 if (uid_eq(current_fsuid(), inode->i_uid))
2127 ns = current_user_ns();
2128 if (kuid_has_mapping(ns, inode->i_uid) && ns_capable(ns, CAP_FOWNER))
2132 EXPORT_SYMBOL(inode_owner_or_capable);
2135 * Direct i/o helper functions
2137 static void __inode_dio_wait(struct inode *inode)
2139 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2140 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2143 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2144 if (atomic_read(&inode->i_dio_count))
2146 } while (atomic_read(&inode->i_dio_count));
2147 finish_wait(wq, &q.wq_entry);
2151 * inode_dio_wait - wait for outstanding DIO requests to finish
2152 * @inode: inode to wait for
2154 * Waits for all pending direct I/O requests to finish so that we can
2155 * proceed with a truncate or equivalent operation.
2157 * Must be called under a lock that serializes taking new references
2158 * to i_dio_count, usually by inode->i_mutex.
2160 void inode_dio_wait(struct inode *inode)
2162 if (atomic_read(&inode->i_dio_count))
2163 __inode_dio_wait(inode);
2165 EXPORT_SYMBOL(inode_dio_wait);
2168 * inode_set_flags - atomically set some inode flags
2170 * Note: the caller should be holding i_mutex, or else be sure that
2171 * they have exclusive access to the inode structure (i.e., while the
2172 * inode is being instantiated). The reason for the cmpxchg() loop
2173 * --- which wouldn't be necessary if all code paths which modify
2174 * i_flags actually followed this rule, is that there is at least one
2175 * code path which doesn't today so we use cmpxchg() out of an abundance
2178 * In the long run, i_mutex is overkill, and we should probably look
2179 * at using the i_lock spinlock to protect i_flags, and then make sure
2180 * it is so documented in include/linux/fs.h and that all code follows
2181 * the locking convention!!
2183 void inode_set_flags(struct inode *inode, unsigned int flags,
2186 WARN_ON_ONCE(flags & ~mask);
2187 set_mask_bits(&inode->i_flags, mask, flags);
2189 EXPORT_SYMBOL(inode_set_flags);
2191 void inode_nohighmem(struct inode *inode)
2193 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2195 EXPORT_SYMBOL(inode_nohighmem);
2198 * timespec64_trunc - Truncate timespec64 to a granularity
2200 * @gran: Granularity in ns.
2202 * Truncate a timespec64 to a granularity. Always rounds down. gran must
2203 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2205 struct timespec64 timespec64_trunc(struct timespec64 t, unsigned gran)
2207 /* Avoid division in the common cases 1 ns and 1 s. */
2210 } else if (gran == NSEC_PER_SEC) {
2212 } else if (gran > 1 && gran < NSEC_PER_SEC) {
2213 t.tv_nsec -= t.tv_nsec % gran;
2215 WARN(1, "illegal file time granularity: %u", gran);
2219 EXPORT_SYMBOL(timespec64_trunc);
2222 * timestamp_truncate - Truncate timespec to a granularity
2224 * @inode: inode being updated
2226 * Truncate a timespec to the granularity supported by the fs
2227 * containing the inode. Always rounds down. gran must
2228 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2230 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2232 struct super_block *sb = inode->i_sb;
2233 unsigned int gran = sb->s_time_gran;
2235 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2236 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2239 /* Avoid division in the common cases 1 ns and 1 s. */
2242 else if (gran == NSEC_PER_SEC)
2244 else if (gran > 1 && gran < NSEC_PER_SEC)
2245 t.tv_nsec -= t.tv_nsec % gran;
2247 WARN(1, "invalid file time granularity: %u", gran);
2250 EXPORT_SYMBOL(timestamp_truncate);
2253 * current_time - Return FS time
2256 * Return the current time truncated to the time granularity supported by
2259 * Note that inode and inode->sb cannot be NULL.
2260 * Otherwise, the function warns and returns time without truncation.
2262 struct timespec64 current_time(struct inode *inode)
2264 struct timespec64 now;
2266 ktime_get_coarse_real_ts64(&now);
2268 if (unlikely(!inode->i_sb)) {
2269 WARN(1, "current_time() called with uninitialized super_block in the inode");
2273 return timestamp_truncate(now, inode);
2275 EXPORT_SYMBOL(current_time);
2278 * Generic function to check FS_IOC_SETFLAGS values and reject any invalid
2281 * Note: the caller should be holding i_mutex, or else be sure that they have
2282 * exclusive access to the inode structure.
2284 int vfs_ioc_setflags_prepare(struct inode *inode, unsigned int oldflags,
2288 * The IMMUTABLE and APPEND_ONLY flags can only be changed by
2289 * the relevant capability.
2291 * This test looks nicer. Thanks to Pauline Middelink
2293 if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL) &&
2294 !capable(CAP_LINUX_IMMUTABLE))
2299 EXPORT_SYMBOL(vfs_ioc_setflags_prepare);
2302 * Generic function to check FS_IOC_FSSETXATTR values and reject any invalid
2305 * Note: the caller should be holding i_mutex, or else be sure that they have
2306 * exclusive access to the inode structure.
2308 int vfs_ioc_fssetxattr_check(struct inode *inode, const struct fsxattr *old_fa,
2312 * Can't modify an immutable/append-only file unless we have
2313 * appropriate permission.
2315 if ((old_fa->fsx_xflags ^ fa->fsx_xflags) &
2316 (FS_XFLAG_IMMUTABLE | FS_XFLAG_APPEND) &&
2317 !capable(CAP_LINUX_IMMUTABLE))
2321 * Project Quota ID state is only allowed to change from within the init
2322 * namespace. Enforce that restriction only if we are trying to change
2323 * the quota ID state. Everything else is allowed in user namespaces.
2325 if (current_user_ns() != &init_user_ns) {
2326 if (old_fa->fsx_projid != fa->fsx_projid)
2328 if ((old_fa->fsx_xflags ^ fa->fsx_xflags) &
2329 FS_XFLAG_PROJINHERIT)
2333 /* Check extent size hints. */
2334 if ((fa->fsx_xflags & FS_XFLAG_EXTSIZE) && !S_ISREG(inode->i_mode))
2337 if ((fa->fsx_xflags & FS_XFLAG_EXTSZINHERIT) &&
2338 !S_ISDIR(inode->i_mode))
2341 if ((fa->fsx_xflags & FS_XFLAG_COWEXTSIZE) &&
2342 !S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2346 * It is only valid to set the DAX flag on regular files and
2347 * directories on filesystems.
2349 if ((fa->fsx_xflags & FS_XFLAG_DAX) &&
2350 !(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
2353 /* Extent size hints of zero turn off the flags. */
2354 if (fa->fsx_extsize == 0)
2355 fa->fsx_xflags &= ~(FS_XFLAG_EXTSIZE | FS_XFLAG_EXTSZINHERIT);
2356 if (fa->fsx_cowextsize == 0)
2357 fa->fsx_xflags &= ~FS_XFLAG_COWEXTSIZE;
2361 EXPORT_SYMBOL(vfs_ioc_fssetxattr_check);