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 *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;
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_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 if (security_inode_alloc(inode))
172 spin_lock_init(&inode->i_lock);
173 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
175 init_rwsem(&inode->i_rwsem);
176 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
178 atomic_set(&inode->i_dio_count, 0);
180 mapping->a_ops = &empty_aops;
181 mapping->host = inode;
183 if (sb->s_type->fs_flags & FS_THP_SUPPORT)
184 __set_bit(AS_THP_SUPPORT, &mapping->flags);
186 atomic_set(&mapping->i_mmap_writable, 0);
187 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
188 atomic_set(&mapping->nr_thps, 0);
190 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
191 mapping->private_data = NULL;
192 mapping->writeback_index = 0;
193 inode->i_private = NULL;
194 inode->i_mapping = mapping;
195 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
196 #ifdef CONFIG_FS_POSIX_ACL
197 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
200 #ifdef CONFIG_FSNOTIFY
201 inode->i_fsnotify_mask = 0;
203 inode->i_flctx = NULL;
204 this_cpu_inc(nr_inodes);
210 EXPORT_SYMBOL(inode_init_always);
212 void free_inode_nonrcu(struct inode *inode)
214 kmem_cache_free(inode_cachep, inode);
216 EXPORT_SYMBOL(free_inode_nonrcu);
218 static void i_callback(struct rcu_head *head)
220 struct inode *inode = container_of(head, struct inode, i_rcu);
221 if (inode->free_inode)
222 inode->free_inode(inode);
224 free_inode_nonrcu(inode);
227 static struct inode *alloc_inode(struct super_block *sb)
229 const struct super_operations *ops = sb->s_op;
232 if (ops->alloc_inode)
233 inode = ops->alloc_inode(sb);
235 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
240 if (unlikely(inode_init_always(sb, inode))) {
241 if (ops->destroy_inode) {
242 ops->destroy_inode(inode);
243 if (!ops->free_inode)
246 inode->free_inode = ops->free_inode;
247 i_callback(&inode->i_rcu);
254 void __destroy_inode(struct inode *inode)
256 BUG_ON(inode_has_buffers(inode));
257 inode_detach_wb(inode);
258 security_inode_free(inode);
259 fsnotify_inode_delete(inode);
260 locks_free_lock_context(inode);
261 if (!inode->i_nlink) {
262 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
263 atomic_long_dec(&inode->i_sb->s_remove_count);
266 #ifdef CONFIG_FS_POSIX_ACL
267 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
268 posix_acl_release(inode->i_acl);
269 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
270 posix_acl_release(inode->i_default_acl);
272 this_cpu_dec(nr_inodes);
274 EXPORT_SYMBOL(__destroy_inode);
276 static void destroy_inode(struct inode *inode)
278 const struct super_operations *ops = inode->i_sb->s_op;
280 BUG_ON(!list_empty(&inode->i_lru));
281 __destroy_inode(inode);
282 if (ops->destroy_inode) {
283 ops->destroy_inode(inode);
284 if (!ops->free_inode)
287 inode->free_inode = ops->free_inode;
288 call_rcu(&inode->i_rcu, i_callback);
292 * drop_nlink - directly drop an inode's link count
295 * This is a low-level filesystem helper to replace any
296 * direct filesystem manipulation of i_nlink. In cases
297 * where we are attempting to track writes to the
298 * filesystem, a decrement to zero means an imminent
299 * write when the file is truncated and actually unlinked
302 void drop_nlink(struct inode *inode)
304 WARN_ON(inode->i_nlink == 0);
307 atomic_long_inc(&inode->i_sb->s_remove_count);
309 EXPORT_SYMBOL(drop_nlink);
312 * clear_nlink - directly zero an inode's link count
315 * This is a low-level filesystem helper to replace any
316 * direct filesystem manipulation of i_nlink. See
317 * drop_nlink() for why we care about i_nlink hitting zero.
319 void clear_nlink(struct inode *inode)
321 if (inode->i_nlink) {
322 inode->__i_nlink = 0;
323 atomic_long_inc(&inode->i_sb->s_remove_count);
326 EXPORT_SYMBOL(clear_nlink);
329 * set_nlink - directly set an inode's link count
331 * @nlink: new nlink (should be non-zero)
333 * This is a low-level filesystem helper to replace any
334 * direct filesystem manipulation of i_nlink.
336 void set_nlink(struct inode *inode, unsigned int nlink)
341 /* Yes, some filesystems do change nlink from zero to one */
342 if (inode->i_nlink == 0)
343 atomic_long_dec(&inode->i_sb->s_remove_count);
345 inode->__i_nlink = nlink;
348 EXPORT_SYMBOL(set_nlink);
351 * inc_nlink - directly increment an inode's link count
354 * This is a low-level filesystem helper to replace any
355 * direct filesystem manipulation of i_nlink. Currently,
356 * it is only here for parity with dec_nlink().
358 void inc_nlink(struct inode *inode)
360 if (unlikely(inode->i_nlink == 0)) {
361 WARN_ON(!(inode->i_state & I_LINKABLE));
362 atomic_long_dec(&inode->i_sb->s_remove_count);
367 EXPORT_SYMBOL(inc_nlink);
369 static void __address_space_init_once(struct address_space *mapping)
371 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
372 init_rwsem(&mapping->i_mmap_rwsem);
373 INIT_LIST_HEAD(&mapping->private_list);
374 spin_lock_init(&mapping->private_lock);
375 mapping->i_mmap = RB_ROOT_CACHED;
378 void address_space_init_once(struct address_space *mapping)
380 memset(mapping, 0, sizeof(*mapping));
381 __address_space_init_once(mapping);
383 EXPORT_SYMBOL(address_space_init_once);
386 * These are initializations that only need to be done
387 * once, because the fields are idempotent across use
388 * of the inode, so let the slab aware of that.
390 void inode_init_once(struct inode *inode)
392 memset(inode, 0, sizeof(*inode));
393 INIT_HLIST_NODE(&inode->i_hash);
394 INIT_LIST_HEAD(&inode->i_devices);
395 INIT_LIST_HEAD(&inode->i_io_list);
396 INIT_LIST_HEAD(&inode->i_wb_list);
397 INIT_LIST_HEAD(&inode->i_lru);
398 __address_space_init_once(&inode->i_data);
399 i_size_ordered_init(inode);
401 EXPORT_SYMBOL(inode_init_once);
403 static void init_once(void *foo)
405 struct inode *inode = (struct inode *) foo;
407 inode_init_once(inode);
411 * inode->i_lock must be held
413 void __iget(struct inode *inode)
415 atomic_inc(&inode->i_count);
419 * get additional reference to inode; caller must already hold one.
421 void ihold(struct inode *inode)
423 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
425 EXPORT_SYMBOL(ihold);
427 static void inode_lru_list_add(struct inode *inode)
429 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
430 this_cpu_inc(nr_unused);
432 inode->i_state |= I_REFERENCED;
436 * Add inode to LRU if needed (inode is unused and clean).
438 * Needs inode->i_lock held.
440 void inode_add_lru(struct inode *inode)
442 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
443 I_FREEING | I_WILL_FREE)) &&
444 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
445 inode_lru_list_add(inode);
449 static void inode_lru_list_del(struct inode *inode)
452 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
453 this_cpu_dec(nr_unused);
457 * inode_sb_list_add - add inode to the superblock list of inodes
458 * @inode: inode to add
460 void inode_sb_list_add(struct inode *inode)
462 spin_lock(&inode->i_sb->s_inode_list_lock);
463 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
464 spin_unlock(&inode->i_sb->s_inode_list_lock);
466 EXPORT_SYMBOL_GPL(inode_sb_list_add);
468 static inline void inode_sb_list_del(struct inode *inode)
470 if (!list_empty(&inode->i_sb_list)) {
471 spin_lock(&inode->i_sb->s_inode_list_lock);
472 list_del_init(&inode->i_sb_list);
473 spin_unlock(&inode->i_sb->s_inode_list_lock);
477 static unsigned long hash(struct super_block *sb, unsigned long hashval)
481 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
483 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
484 return tmp & i_hash_mask;
488 * __insert_inode_hash - hash an inode
489 * @inode: unhashed inode
490 * @hashval: unsigned long value used to locate this object in the
493 * Add an inode to the inode hash for this superblock.
495 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
497 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
499 spin_lock(&inode_hash_lock);
500 spin_lock(&inode->i_lock);
501 hlist_add_head_rcu(&inode->i_hash, b);
502 spin_unlock(&inode->i_lock);
503 spin_unlock(&inode_hash_lock);
505 EXPORT_SYMBOL(__insert_inode_hash);
508 * __remove_inode_hash - remove an inode from the hash
509 * @inode: inode to unhash
511 * Remove an inode from the superblock.
513 void __remove_inode_hash(struct inode *inode)
515 spin_lock(&inode_hash_lock);
516 spin_lock(&inode->i_lock);
517 hlist_del_init_rcu(&inode->i_hash);
518 spin_unlock(&inode->i_lock);
519 spin_unlock(&inode_hash_lock);
521 EXPORT_SYMBOL(__remove_inode_hash);
523 void clear_inode(struct inode *inode)
526 * We have to cycle the i_pages lock here because reclaim can be in the
527 * process of removing the last page (in __delete_from_page_cache())
528 * and we must not free the mapping under it.
530 xa_lock_irq(&inode->i_data.i_pages);
531 BUG_ON(inode->i_data.nrpages);
533 * Almost always, mapping_empty(&inode->i_data) here; but there are
534 * two known and long-standing ways in which nodes may get left behind
535 * (when deep radix-tree node allocation failed partway; or when THP
536 * collapse_file() failed). Until those two known cases are cleaned up,
537 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
538 * nor even WARN_ON(!mapping_empty).
540 xa_unlock_irq(&inode->i_data.i_pages);
541 BUG_ON(!list_empty(&inode->i_data.private_list));
542 BUG_ON(!(inode->i_state & I_FREEING));
543 BUG_ON(inode->i_state & I_CLEAR);
544 BUG_ON(!list_empty(&inode->i_wb_list));
545 /* don't need i_lock here, no concurrent mods to i_state */
546 inode->i_state = I_FREEING | I_CLEAR;
548 EXPORT_SYMBOL(clear_inode);
551 * Free the inode passed in, removing it from the lists it is still connected
552 * to. We remove any pages still attached to the inode and wait for any IO that
553 * is still in progress before finally destroying the inode.
555 * An inode must already be marked I_FREEING so that we avoid the inode being
556 * moved back onto lists if we race with other code that manipulates the lists
557 * (e.g. writeback_single_inode). The caller is responsible for setting this.
559 * An inode must already be removed from the LRU list before being evicted from
560 * the cache. This should occur atomically with setting the I_FREEING state
561 * flag, so no inodes here should ever be on the LRU when being evicted.
563 static void evict(struct inode *inode)
565 const struct super_operations *op = inode->i_sb->s_op;
567 BUG_ON(!(inode->i_state & I_FREEING));
568 BUG_ON(!list_empty(&inode->i_lru));
570 if (!list_empty(&inode->i_io_list))
571 inode_io_list_del(inode);
573 inode_sb_list_del(inode);
576 * Wait for flusher thread to be done with the inode so that filesystem
577 * does not start destroying it while writeback is still running. Since
578 * the inode has I_FREEING set, flusher thread won't start new work on
579 * the inode. We just have to wait for running writeback to finish.
581 inode_wait_for_writeback(inode);
583 if (op->evict_inode) {
584 op->evict_inode(inode);
586 truncate_inode_pages_final(&inode->i_data);
589 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
592 remove_inode_hash(inode);
594 spin_lock(&inode->i_lock);
595 wake_up_bit(&inode->i_state, __I_NEW);
596 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
597 spin_unlock(&inode->i_lock);
599 destroy_inode(inode);
603 * dispose_list - dispose of the contents of a local list
604 * @head: the head of the list to free
606 * Dispose-list gets a local list with local inodes in it, so it doesn't
607 * need to worry about list corruption and SMP locks.
609 static void dispose_list(struct list_head *head)
611 while (!list_empty(head)) {
614 inode = list_first_entry(head, struct inode, i_lru);
615 list_del_init(&inode->i_lru);
623 * evict_inodes - evict all evictable inodes for a superblock
624 * @sb: superblock to operate on
626 * Make sure that no inodes with zero refcount are retained. This is
627 * called by superblock shutdown after having SB_ACTIVE flag removed,
628 * so any inode reaching zero refcount during or after that call will
629 * be immediately evicted.
631 void evict_inodes(struct super_block *sb)
633 struct inode *inode, *next;
637 spin_lock(&sb->s_inode_list_lock);
638 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
639 if (atomic_read(&inode->i_count))
642 spin_lock(&inode->i_lock);
643 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
644 spin_unlock(&inode->i_lock);
648 inode->i_state |= I_FREEING;
649 inode_lru_list_del(inode);
650 spin_unlock(&inode->i_lock);
651 list_add(&inode->i_lru, &dispose);
654 * We can have a ton of inodes to evict at unmount time given
655 * enough memory, check to see if we need to go to sleep for a
656 * bit so we don't livelock.
658 if (need_resched()) {
659 spin_unlock(&sb->s_inode_list_lock);
661 dispose_list(&dispose);
665 spin_unlock(&sb->s_inode_list_lock);
667 dispose_list(&dispose);
669 EXPORT_SYMBOL_GPL(evict_inodes);
672 * invalidate_inodes - attempt to free all inodes on a superblock
673 * @sb: superblock to operate on
674 * @kill_dirty: flag to guide handling of dirty inodes
676 * Attempts to free all inodes for a given superblock. If there were any
677 * busy inodes return a non-zero value, else zero.
678 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
681 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
684 struct inode *inode, *next;
688 spin_lock(&sb->s_inode_list_lock);
689 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
690 spin_lock(&inode->i_lock);
691 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
692 spin_unlock(&inode->i_lock);
695 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
696 spin_unlock(&inode->i_lock);
700 if (atomic_read(&inode->i_count)) {
701 spin_unlock(&inode->i_lock);
706 inode->i_state |= I_FREEING;
707 inode_lru_list_del(inode);
708 spin_unlock(&inode->i_lock);
709 list_add(&inode->i_lru, &dispose);
710 if (need_resched()) {
711 spin_unlock(&sb->s_inode_list_lock);
713 dispose_list(&dispose);
717 spin_unlock(&sb->s_inode_list_lock);
719 dispose_list(&dispose);
725 * Isolate the inode from the LRU in preparation for freeing it.
727 * Any inodes which are pinned purely because of attached pagecache have their
728 * pagecache removed. If the inode has metadata buffers attached to
729 * mapping->private_list then try to remove them.
731 * If the inode has the I_REFERENCED flag set, then it means that it has been
732 * used recently - the flag is set in iput_final(). When we encounter such an
733 * inode, clear the flag and move it to the back of the LRU so it gets another
734 * pass through the LRU before it gets reclaimed. This is necessary because of
735 * the fact we are doing lazy LRU updates to minimise lock contention so the
736 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
737 * with this flag set because they are the inodes that are out of order.
739 static enum lru_status inode_lru_isolate(struct list_head *item,
740 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
742 struct list_head *freeable = arg;
743 struct inode *inode = container_of(item, struct inode, i_lru);
746 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
747 * If we fail to get the lock, just skip it.
749 if (!spin_trylock(&inode->i_lock))
753 * Referenced or dirty inodes are still in use. Give them another pass
754 * through the LRU as we canot reclaim them now.
756 if (atomic_read(&inode->i_count) ||
757 (inode->i_state & ~I_REFERENCED)) {
758 list_lru_isolate(lru, &inode->i_lru);
759 spin_unlock(&inode->i_lock);
760 this_cpu_dec(nr_unused);
764 /* recently referenced inodes get one more pass */
765 if (inode->i_state & I_REFERENCED) {
766 inode->i_state &= ~I_REFERENCED;
767 spin_unlock(&inode->i_lock);
771 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
773 spin_unlock(&inode->i_lock);
774 spin_unlock(lru_lock);
775 if (remove_inode_buffers(inode)) {
777 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
778 if (current_is_kswapd())
779 __count_vm_events(KSWAPD_INODESTEAL, reap);
781 __count_vm_events(PGINODESTEAL, reap);
782 if (current->reclaim_state)
783 current->reclaim_state->reclaimed_slab += reap;
790 WARN_ON(inode->i_state & I_NEW);
791 inode->i_state |= I_FREEING;
792 list_lru_isolate_move(lru, &inode->i_lru, freeable);
793 spin_unlock(&inode->i_lock);
795 this_cpu_dec(nr_unused);
800 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
801 * This is called from the superblock shrinker function with a number of inodes
802 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
803 * then are freed outside inode_lock by dispose_list().
805 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
810 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
811 inode_lru_isolate, &freeable);
812 dispose_list(&freeable);
816 static void __wait_on_freeing_inode(struct inode *inode);
818 * Called with the inode lock held.
820 static struct inode *find_inode(struct super_block *sb,
821 struct hlist_head *head,
822 int (*test)(struct inode *, void *),
825 struct inode *inode = NULL;
828 hlist_for_each_entry(inode, head, i_hash) {
829 if (inode->i_sb != sb)
831 if (!test(inode, data))
833 spin_lock(&inode->i_lock);
834 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
835 __wait_on_freeing_inode(inode);
838 if (unlikely(inode->i_state & I_CREATING)) {
839 spin_unlock(&inode->i_lock);
840 return ERR_PTR(-ESTALE);
843 spin_unlock(&inode->i_lock);
850 * find_inode_fast is the fast path version of find_inode, see the comment at
851 * iget_locked for details.
853 static struct inode *find_inode_fast(struct super_block *sb,
854 struct hlist_head *head, unsigned long ino)
856 struct inode *inode = NULL;
859 hlist_for_each_entry(inode, head, i_hash) {
860 if (inode->i_ino != ino)
862 if (inode->i_sb != sb)
864 spin_lock(&inode->i_lock);
865 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
866 __wait_on_freeing_inode(inode);
869 if (unlikely(inode->i_state & I_CREATING)) {
870 spin_unlock(&inode->i_lock);
871 return ERR_PTR(-ESTALE);
874 spin_unlock(&inode->i_lock);
881 * Each cpu owns a range of LAST_INO_BATCH numbers.
882 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
883 * to renew the exhausted range.
885 * This does not significantly increase overflow rate because every CPU can
886 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
887 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
888 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
889 * overflow rate by 2x, which does not seem too significant.
891 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
892 * error if st_ino won't fit in target struct field. Use 32bit counter
893 * here to attempt to avoid that.
895 #define LAST_INO_BATCH 1024
896 static DEFINE_PER_CPU(unsigned int, last_ino);
898 unsigned int get_next_ino(void)
900 unsigned int *p = &get_cpu_var(last_ino);
901 unsigned int res = *p;
904 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
905 static atomic_t shared_last_ino;
906 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
908 res = next - LAST_INO_BATCH;
913 /* get_next_ino should not provide a 0 inode number */
917 put_cpu_var(last_ino);
920 EXPORT_SYMBOL(get_next_ino);
923 * new_inode_pseudo - obtain an inode
926 * Allocates a new inode for given superblock.
927 * Inode wont be chained in superblock s_inodes list
929 * - fs can't be unmount
930 * - quotas, fsnotify, writeback can't work
932 struct inode *new_inode_pseudo(struct super_block *sb)
934 struct inode *inode = alloc_inode(sb);
937 spin_lock(&inode->i_lock);
939 spin_unlock(&inode->i_lock);
940 INIT_LIST_HEAD(&inode->i_sb_list);
946 * new_inode - obtain an inode
949 * Allocates a new inode for given superblock. The default gfp_mask
950 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
951 * If HIGHMEM pages are unsuitable or it is known that pages allocated
952 * for the page cache are not reclaimable or migratable,
953 * mapping_set_gfp_mask() must be called with suitable flags on the
954 * newly created inode's mapping
957 struct inode *new_inode(struct super_block *sb)
961 spin_lock_prefetch(&sb->s_inode_list_lock);
963 inode = new_inode_pseudo(sb);
965 inode_sb_list_add(inode);
968 EXPORT_SYMBOL(new_inode);
970 #ifdef CONFIG_DEBUG_LOCK_ALLOC
971 void lockdep_annotate_inode_mutex_key(struct inode *inode)
973 if (S_ISDIR(inode->i_mode)) {
974 struct file_system_type *type = inode->i_sb->s_type;
976 /* Set new key only if filesystem hasn't already changed it */
977 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
979 * ensure nobody is actually holding i_mutex
981 // mutex_destroy(&inode->i_mutex);
982 init_rwsem(&inode->i_rwsem);
983 lockdep_set_class(&inode->i_rwsem,
984 &type->i_mutex_dir_key);
988 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
992 * unlock_new_inode - clear the I_NEW state and wake up any waiters
993 * @inode: new inode to unlock
995 * Called when the inode is fully initialised to clear the new state of the
996 * inode and wake up anyone waiting for the inode to finish initialisation.
998 void unlock_new_inode(struct inode *inode)
1000 lockdep_annotate_inode_mutex_key(inode);
1001 spin_lock(&inode->i_lock);
1002 WARN_ON(!(inode->i_state & I_NEW));
1003 inode->i_state &= ~I_NEW & ~I_CREATING;
1005 wake_up_bit(&inode->i_state, __I_NEW);
1006 spin_unlock(&inode->i_lock);
1008 EXPORT_SYMBOL(unlock_new_inode);
1010 void discard_new_inode(struct inode *inode)
1012 lockdep_annotate_inode_mutex_key(inode);
1013 spin_lock(&inode->i_lock);
1014 WARN_ON(!(inode->i_state & I_NEW));
1015 inode->i_state &= ~I_NEW;
1017 wake_up_bit(&inode->i_state, __I_NEW);
1018 spin_unlock(&inode->i_lock);
1021 EXPORT_SYMBOL(discard_new_inode);
1024 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1026 * Lock any non-NULL argument that is not a directory.
1027 * Zero, one or two objects may be locked by this function.
1029 * @inode1: first inode to lock
1030 * @inode2: second inode to lock
1032 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1034 if (inode1 > inode2)
1035 swap(inode1, inode2);
1037 if (inode1 && !S_ISDIR(inode1->i_mode))
1039 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1040 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1042 EXPORT_SYMBOL(lock_two_nondirectories);
1045 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1046 * @inode1: first inode to unlock
1047 * @inode2: second inode to unlock
1049 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1051 if (inode1 && !S_ISDIR(inode1->i_mode))
1052 inode_unlock(inode1);
1053 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1054 inode_unlock(inode2);
1056 EXPORT_SYMBOL(unlock_two_nondirectories);
1059 * inode_insert5 - obtain an inode from a mounted file system
1060 * @inode: pre-allocated inode to use for insert to cache
1061 * @hashval: hash value (usually inode number) to get
1062 * @test: callback used for comparisons between inodes
1063 * @set: callback used to initialize a new struct inode
1064 * @data: opaque data pointer to pass to @test and @set
1066 * Search for the inode specified by @hashval and @data in the inode cache,
1067 * and if present it is return it with an increased reference count. This is
1068 * a variant of iget5_locked() for callers that don't want to fail on memory
1069 * allocation of inode.
1071 * If the inode is not in cache, insert the pre-allocated inode to cache and
1072 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1073 * to fill it in before unlocking it via unlock_new_inode().
1075 * Note both @test and @set are called with the inode_hash_lock held, so can't
1078 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1079 int (*test)(struct inode *, void *),
1080 int (*set)(struct inode *, void *), void *data)
1082 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1084 bool creating = inode->i_state & I_CREATING;
1087 spin_lock(&inode_hash_lock);
1088 old = find_inode(inode->i_sb, head, test, data);
1089 if (unlikely(old)) {
1091 * Uhhuh, somebody else created the same inode under us.
1092 * Use the old inode instead of the preallocated one.
1094 spin_unlock(&inode_hash_lock);
1098 if (unlikely(inode_unhashed(old))) {
1105 if (set && unlikely(set(inode, data))) {
1111 * Return the locked inode with I_NEW set, the
1112 * caller is responsible for filling in the contents
1114 spin_lock(&inode->i_lock);
1115 inode->i_state |= I_NEW;
1116 hlist_add_head_rcu(&inode->i_hash, head);
1117 spin_unlock(&inode->i_lock);
1119 inode_sb_list_add(inode);
1121 spin_unlock(&inode_hash_lock);
1125 EXPORT_SYMBOL(inode_insert5);
1128 * iget5_locked - obtain an inode from a mounted file system
1129 * @sb: super block of file system
1130 * @hashval: hash value (usually inode number) to get
1131 * @test: callback used for comparisons between inodes
1132 * @set: callback used to initialize a new struct inode
1133 * @data: opaque data pointer to pass to @test and @set
1135 * Search for the inode specified by @hashval and @data in the inode cache,
1136 * and if present it is return it with an increased reference count. This is
1137 * a generalized version of iget_locked() for file systems where the inode
1138 * number is not sufficient for unique identification of an inode.
1140 * If the inode is not in cache, allocate a new inode and return it locked,
1141 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1142 * before unlocking it via unlock_new_inode().
1144 * Note both @test and @set are called with the inode_hash_lock held, so can't
1147 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1148 int (*test)(struct inode *, void *),
1149 int (*set)(struct inode *, void *), void *data)
1151 struct inode *inode = ilookup5(sb, hashval, test, data);
1154 struct inode *new = alloc_inode(sb);
1158 inode = inode_insert5(new, hashval, test, set, data);
1159 if (unlikely(inode != new))
1165 EXPORT_SYMBOL(iget5_locked);
1168 * iget_locked - obtain an inode from a mounted file system
1169 * @sb: super block of file system
1170 * @ino: inode number to get
1172 * Search for the inode specified by @ino in the inode cache and if present
1173 * return it with an increased reference count. This is for file systems
1174 * where the inode number is sufficient for unique identification of an inode.
1176 * If the inode is not in cache, allocate a new inode and return it locked,
1177 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1178 * before unlocking it via unlock_new_inode().
1180 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1182 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1183 struct inode *inode;
1185 spin_lock(&inode_hash_lock);
1186 inode = find_inode_fast(sb, head, ino);
1187 spin_unlock(&inode_hash_lock);
1191 wait_on_inode(inode);
1192 if (unlikely(inode_unhashed(inode))) {
1199 inode = alloc_inode(sb);
1203 spin_lock(&inode_hash_lock);
1204 /* We released the lock, so.. */
1205 old = find_inode_fast(sb, head, ino);
1208 spin_lock(&inode->i_lock);
1209 inode->i_state = I_NEW;
1210 hlist_add_head_rcu(&inode->i_hash, head);
1211 spin_unlock(&inode->i_lock);
1212 inode_sb_list_add(inode);
1213 spin_unlock(&inode_hash_lock);
1215 /* Return the locked inode with I_NEW set, the
1216 * caller is responsible for filling in the contents
1222 * Uhhuh, somebody else created the same inode under
1223 * us. Use the old inode instead of the one we just
1226 spin_unlock(&inode_hash_lock);
1227 destroy_inode(inode);
1231 wait_on_inode(inode);
1232 if (unlikely(inode_unhashed(inode))) {
1239 EXPORT_SYMBOL(iget_locked);
1242 * search the inode cache for a matching inode number.
1243 * If we find one, then the inode number we are trying to
1244 * allocate is not unique and so we should not use it.
1246 * Returns 1 if the inode number is unique, 0 if it is not.
1248 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1250 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1251 struct inode *inode;
1253 hlist_for_each_entry_rcu(inode, b, i_hash) {
1254 if (inode->i_ino == ino && inode->i_sb == sb)
1261 * iunique - get a unique inode number
1263 * @max_reserved: highest reserved inode number
1265 * Obtain an inode number that is unique on the system for a given
1266 * superblock. This is used by file systems that have no natural
1267 * permanent inode numbering system. An inode number is returned that
1268 * is higher than the reserved limit but unique.
1271 * With a large number of inodes live on the file system this function
1272 * currently becomes quite slow.
1274 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1277 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1278 * error if st_ino won't fit in target struct field. Use 32bit counter
1279 * here to attempt to avoid that.
1281 static DEFINE_SPINLOCK(iunique_lock);
1282 static unsigned int counter;
1286 spin_lock(&iunique_lock);
1288 if (counter <= max_reserved)
1289 counter = max_reserved + 1;
1291 } while (!test_inode_iunique(sb, res));
1292 spin_unlock(&iunique_lock);
1297 EXPORT_SYMBOL(iunique);
1299 struct inode *igrab(struct inode *inode)
1301 spin_lock(&inode->i_lock);
1302 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1304 spin_unlock(&inode->i_lock);
1306 spin_unlock(&inode->i_lock);
1308 * Handle the case where s_op->clear_inode is not been
1309 * called yet, and somebody is calling igrab
1310 * while the inode is getting freed.
1316 EXPORT_SYMBOL(igrab);
1319 * ilookup5_nowait - search for an inode in the inode cache
1320 * @sb: super block of file system to search
1321 * @hashval: hash value (usually inode number) to search for
1322 * @test: callback used for comparisons between inodes
1323 * @data: opaque data pointer to pass to @test
1325 * Search for the inode specified by @hashval and @data in the inode cache.
1326 * If the inode is in the cache, the inode is returned with an incremented
1329 * Note: I_NEW is not waited upon so you have to be very careful what you do
1330 * with the returned inode. You probably should be using ilookup5() instead.
1332 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1334 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1335 int (*test)(struct inode *, void *), void *data)
1337 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1338 struct inode *inode;
1340 spin_lock(&inode_hash_lock);
1341 inode = find_inode(sb, head, test, data);
1342 spin_unlock(&inode_hash_lock);
1344 return IS_ERR(inode) ? NULL : inode;
1346 EXPORT_SYMBOL(ilookup5_nowait);
1349 * ilookup5 - search for an inode in the inode cache
1350 * @sb: super block of file system to search
1351 * @hashval: hash value (usually inode number) to search for
1352 * @test: callback used for comparisons between inodes
1353 * @data: opaque data pointer to pass to @test
1355 * Search for the inode specified by @hashval and @data in the inode cache,
1356 * and if the inode is in the cache, return the inode with an incremented
1357 * reference count. Waits on I_NEW before returning the inode.
1358 * returned with an incremented reference count.
1360 * This is a generalized version of ilookup() for file systems where the
1361 * inode number is not sufficient for unique identification of an inode.
1363 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1365 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1366 int (*test)(struct inode *, void *), void *data)
1368 struct inode *inode;
1370 inode = ilookup5_nowait(sb, hashval, test, data);
1372 wait_on_inode(inode);
1373 if (unlikely(inode_unhashed(inode))) {
1380 EXPORT_SYMBOL(ilookup5);
1383 * ilookup - search for an inode in the inode cache
1384 * @sb: super block of file system to search
1385 * @ino: inode number to search for
1387 * Search for the inode @ino in the inode cache, and if the inode is in the
1388 * cache, the inode is returned with an incremented reference count.
1390 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1392 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1393 struct inode *inode;
1395 spin_lock(&inode_hash_lock);
1396 inode = find_inode_fast(sb, head, ino);
1397 spin_unlock(&inode_hash_lock);
1402 wait_on_inode(inode);
1403 if (unlikely(inode_unhashed(inode))) {
1410 EXPORT_SYMBOL(ilookup);
1413 * find_inode_nowait - find an inode in the inode cache
1414 * @sb: super block of file system to search
1415 * @hashval: hash value (usually inode number) to search for
1416 * @match: callback used for comparisons between inodes
1417 * @data: opaque data pointer to pass to @match
1419 * Search for the inode specified by @hashval and @data in the inode
1420 * cache, where the helper function @match will return 0 if the inode
1421 * does not match, 1 if the inode does match, and -1 if the search
1422 * should be stopped. The @match function must be responsible for
1423 * taking the i_lock spin_lock and checking i_state for an inode being
1424 * freed or being initialized, and incrementing the reference count
1425 * before returning 1. It also must not sleep, since it is called with
1426 * the inode_hash_lock spinlock held.
1428 * This is a even more generalized version of ilookup5() when the
1429 * function must never block --- find_inode() can block in
1430 * __wait_on_freeing_inode() --- or when the caller can not increment
1431 * the reference count because the resulting iput() might cause an
1432 * inode eviction. The tradeoff is that the @match funtion must be
1433 * very carefully implemented.
1435 struct inode *find_inode_nowait(struct super_block *sb,
1436 unsigned long hashval,
1437 int (*match)(struct inode *, unsigned long,
1441 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1442 struct inode *inode, *ret_inode = NULL;
1445 spin_lock(&inode_hash_lock);
1446 hlist_for_each_entry(inode, head, i_hash) {
1447 if (inode->i_sb != sb)
1449 mval = match(inode, hashval, data);
1457 spin_unlock(&inode_hash_lock);
1460 EXPORT_SYMBOL(find_inode_nowait);
1463 * find_inode_rcu - find an inode in the inode cache
1464 * @sb: Super block of file system to search
1465 * @hashval: Key to hash
1466 * @test: Function to test match on an inode
1467 * @data: Data for test function
1469 * Search for the inode specified by @hashval and @data in the inode cache,
1470 * where the helper function @test will return 0 if the inode does not match
1471 * and 1 if it does. The @test function must be responsible for taking the
1472 * i_lock spin_lock and checking i_state for an inode being freed or being
1475 * If successful, this will return the inode for which the @test function
1476 * returned 1 and NULL otherwise.
1478 * The @test function is not permitted to take a ref on any inode presented.
1479 * It is also not permitted to sleep.
1481 * The caller must hold the RCU read lock.
1483 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1484 int (*test)(struct inode *, void *), void *data)
1486 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1487 struct inode *inode;
1489 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1490 "suspicious find_inode_rcu() usage");
1492 hlist_for_each_entry_rcu(inode, head, i_hash) {
1493 if (inode->i_sb == sb &&
1494 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1500 EXPORT_SYMBOL(find_inode_rcu);
1503 * find_inode_by_ino_rcu - Find an inode in the inode cache
1504 * @sb: Super block of file system to search
1505 * @ino: The inode number to match
1507 * Search for the inode specified by @hashval and @data in the inode cache,
1508 * where the helper function @test will return 0 if the inode does not match
1509 * and 1 if it does. The @test function must be responsible for taking the
1510 * i_lock spin_lock and checking i_state for an inode being freed or being
1513 * If successful, this will return the inode for which the @test function
1514 * returned 1 and NULL otherwise.
1516 * The @test function is not permitted to take a ref on any inode presented.
1517 * It is also not permitted to sleep.
1519 * The caller must hold the RCU read lock.
1521 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1524 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1525 struct inode *inode;
1527 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1528 "suspicious find_inode_by_ino_rcu() usage");
1530 hlist_for_each_entry_rcu(inode, head, i_hash) {
1531 if (inode->i_ino == ino &&
1532 inode->i_sb == sb &&
1533 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1538 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1540 int insert_inode_locked(struct inode *inode)
1542 struct super_block *sb = inode->i_sb;
1543 ino_t ino = inode->i_ino;
1544 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1547 struct inode *old = NULL;
1548 spin_lock(&inode_hash_lock);
1549 hlist_for_each_entry(old, head, i_hash) {
1550 if (old->i_ino != ino)
1552 if (old->i_sb != sb)
1554 spin_lock(&old->i_lock);
1555 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1556 spin_unlock(&old->i_lock);
1562 spin_lock(&inode->i_lock);
1563 inode->i_state |= I_NEW | I_CREATING;
1564 hlist_add_head_rcu(&inode->i_hash, head);
1565 spin_unlock(&inode->i_lock);
1566 spin_unlock(&inode_hash_lock);
1569 if (unlikely(old->i_state & I_CREATING)) {
1570 spin_unlock(&old->i_lock);
1571 spin_unlock(&inode_hash_lock);
1575 spin_unlock(&old->i_lock);
1576 spin_unlock(&inode_hash_lock);
1578 if (unlikely(!inode_unhashed(old))) {
1585 EXPORT_SYMBOL(insert_inode_locked);
1587 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1588 int (*test)(struct inode *, void *), void *data)
1592 inode->i_state |= I_CREATING;
1593 old = inode_insert5(inode, hashval, test, NULL, data);
1601 EXPORT_SYMBOL(insert_inode_locked4);
1604 int generic_delete_inode(struct inode *inode)
1608 EXPORT_SYMBOL(generic_delete_inode);
1611 * Called when we're dropping the last reference
1614 * Call the FS "drop_inode()" function, defaulting to
1615 * the legacy UNIX filesystem behaviour. If it tells
1616 * us to evict inode, do so. Otherwise, retain inode
1617 * in cache if fs is alive, sync and evict if fs is
1620 static void iput_final(struct inode *inode)
1622 struct super_block *sb = inode->i_sb;
1623 const struct super_operations *op = inode->i_sb->s_op;
1624 unsigned long state;
1627 WARN_ON(inode->i_state & I_NEW);
1630 drop = op->drop_inode(inode);
1632 drop = generic_drop_inode(inode);
1635 !(inode->i_state & I_DONTCACHE) &&
1636 (sb->s_flags & SB_ACTIVE)) {
1637 inode_add_lru(inode);
1638 spin_unlock(&inode->i_lock);
1642 state = inode->i_state;
1644 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1645 spin_unlock(&inode->i_lock);
1647 write_inode_now(inode, 1);
1649 spin_lock(&inode->i_lock);
1650 state = inode->i_state;
1651 WARN_ON(state & I_NEW);
1652 state &= ~I_WILL_FREE;
1655 WRITE_ONCE(inode->i_state, state | I_FREEING);
1656 if (!list_empty(&inode->i_lru))
1657 inode_lru_list_del(inode);
1658 spin_unlock(&inode->i_lock);
1664 * iput - put an inode
1665 * @inode: inode to put
1667 * Puts an inode, dropping its usage count. If the inode use count hits
1668 * zero, the inode is then freed and may also be destroyed.
1670 * Consequently, iput() can sleep.
1672 void iput(struct inode *inode)
1676 BUG_ON(inode->i_state & I_CLEAR);
1678 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1679 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1680 atomic_inc(&inode->i_count);
1681 spin_unlock(&inode->i_lock);
1682 trace_writeback_lazytime_iput(inode);
1683 mark_inode_dirty_sync(inode);
1689 EXPORT_SYMBOL(iput);
1693 * bmap - find a block number in a file
1694 * @inode: inode owning the block number being requested
1695 * @block: pointer containing the block to find
1697 * Replaces the value in ``*block`` with the block number on the device holding
1698 * corresponding to the requested block number in the file.
1699 * That is, asked for block 4 of inode 1 the function will replace the
1700 * 4 in ``*block``, with disk block relative to the disk start that holds that
1701 * block of the file.
1703 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1704 * hole, returns 0 and ``*block`` is also set to 0.
1706 int bmap(struct inode *inode, sector_t *block)
1708 if (!inode->i_mapping->a_ops->bmap)
1711 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1714 EXPORT_SYMBOL(bmap);
1718 * With relative atime, only update atime if the previous atime is
1719 * earlier than either the ctime or mtime or if at least a day has
1720 * passed since the last atime update.
1722 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1723 struct timespec64 now)
1726 if (!(mnt->mnt_flags & MNT_RELATIME))
1729 * Is mtime younger than atime? If yes, update atime:
1731 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1734 * Is ctime younger than atime? If yes, update atime:
1736 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1740 * Is the previous atime value older than a day? If yes,
1743 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1746 * Good, we can skip the atime update:
1751 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1753 int dirty_flags = 0;
1755 if (flags & (S_ATIME | S_CTIME | S_MTIME)) {
1756 if (flags & S_ATIME)
1757 inode->i_atime = *time;
1758 if (flags & S_CTIME)
1759 inode->i_ctime = *time;
1760 if (flags & S_MTIME)
1761 inode->i_mtime = *time;
1763 if (inode->i_sb->s_flags & SB_LAZYTIME)
1764 dirty_flags |= I_DIRTY_TIME;
1766 dirty_flags |= I_DIRTY_SYNC;
1769 if ((flags & S_VERSION) && inode_maybe_inc_iversion(inode, false))
1770 dirty_flags |= I_DIRTY_SYNC;
1772 __mark_inode_dirty(inode, dirty_flags);
1775 EXPORT_SYMBOL(generic_update_time);
1778 * This does the actual work of updating an inodes time or version. Must have
1779 * had called mnt_want_write() before calling this.
1781 static int update_time(struct inode *inode, struct timespec64 *time, int flags)
1783 if (inode->i_op->update_time)
1784 return inode->i_op->update_time(inode, time, flags);
1785 return generic_update_time(inode, time, flags);
1789 * atime_needs_update - update the access time
1790 * @path: the &struct path to update
1791 * @inode: inode to update
1793 * Update the accessed time on an inode and mark it for writeback.
1794 * This function automatically handles read only file systems and media,
1795 * as well as the "noatime" flag and inode specific "noatime" markers.
1797 bool atime_needs_update(const struct path *path, struct inode *inode)
1799 struct vfsmount *mnt = path->mnt;
1800 struct timespec64 now;
1802 if (inode->i_flags & S_NOATIME)
1805 /* Atime updates will likely cause i_uid and i_gid to be written
1806 * back improprely if their true value is unknown to the vfs.
1808 if (HAS_UNMAPPED_ID(mnt_user_ns(mnt), inode))
1811 if (IS_NOATIME(inode))
1813 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1816 if (mnt->mnt_flags & MNT_NOATIME)
1818 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1821 now = current_time(inode);
1823 if (!relatime_need_update(mnt, inode, now))
1826 if (timespec64_equal(&inode->i_atime, &now))
1832 void touch_atime(const struct path *path)
1834 struct vfsmount *mnt = path->mnt;
1835 struct inode *inode = d_inode(path->dentry);
1836 struct timespec64 now;
1838 if (!atime_needs_update(path, inode))
1841 if (!sb_start_write_trylock(inode->i_sb))
1844 if (__mnt_want_write(mnt) != 0)
1847 * File systems can error out when updating inodes if they need to
1848 * allocate new space to modify an inode (such is the case for
1849 * Btrfs), but since we touch atime while walking down the path we
1850 * really don't care if we failed to update the atime of the file,
1851 * so just ignore the return value.
1852 * We may also fail on filesystems that have the ability to make parts
1853 * of the fs read only, e.g. subvolumes in Btrfs.
1855 now = current_time(inode);
1856 update_time(inode, &now, S_ATIME);
1857 __mnt_drop_write(mnt);
1859 sb_end_write(inode->i_sb);
1861 EXPORT_SYMBOL(touch_atime);
1864 * The logic we want is
1866 * if suid or (sgid and xgrp)
1869 int should_remove_suid(struct dentry *dentry)
1871 umode_t mode = d_inode(dentry)->i_mode;
1874 /* suid always must be killed */
1875 if (unlikely(mode & S_ISUID))
1876 kill = ATTR_KILL_SUID;
1879 * sgid without any exec bits is just a mandatory locking mark; leave
1880 * it alone. If some exec bits are set, it's a real sgid; kill it.
1882 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1883 kill |= ATTR_KILL_SGID;
1885 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1890 EXPORT_SYMBOL(should_remove_suid);
1893 * Return mask of changes for notify_change() that need to be done as a
1894 * response to write or truncate. Return 0 if nothing has to be changed.
1895 * Negative value on error (change should be denied).
1897 int dentry_needs_remove_privs(struct dentry *dentry)
1899 struct inode *inode = d_inode(dentry);
1903 if (IS_NOSEC(inode))
1906 mask = should_remove_suid(dentry);
1907 ret = security_inode_need_killpriv(dentry);
1911 mask |= ATTR_KILL_PRIV;
1915 static int __remove_privs(struct user_namespace *mnt_userns,
1916 struct dentry *dentry, int kill)
1918 struct iattr newattrs;
1920 newattrs.ia_valid = ATTR_FORCE | kill;
1922 * Note we call this on write, so notify_change will not
1923 * encounter any conflicting delegations:
1925 return notify_change(mnt_userns, dentry, &newattrs, NULL);
1929 * Remove special file priviledges (suid, capabilities) when file is written
1932 int file_remove_privs(struct file *file)
1934 struct dentry *dentry = file_dentry(file);
1935 struct inode *inode = file_inode(file);
1940 * Fast path for nothing security related.
1941 * As well for non-regular files, e.g. blkdev inodes.
1942 * For example, blkdev_write_iter() might get here
1943 * trying to remove privs which it is not allowed to.
1945 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1948 kill = dentry_needs_remove_privs(dentry);
1952 error = __remove_privs(file_mnt_user_ns(file), dentry, kill);
1954 inode_has_no_xattr(inode);
1958 EXPORT_SYMBOL(file_remove_privs);
1961 * file_update_time - update mtime and ctime time
1962 * @file: file accessed
1964 * Update the mtime and ctime members of an inode and mark the inode
1965 * for writeback. Note that this function is meant exclusively for
1966 * usage in the file write path of filesystems, and filesystems may
1967 * choose to explicitly ignore update via this function with the
1968 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1969 * timestamps are handled by the server. This can return an error for
1970 * file systems who need to allocate space in order to update an inode.
1973 int file_update_time(struct file *file)
1975 struct inode *inode = file_inode(file);
1976 struct timespec64 now;
1980 /* First try to exhaust all avenues to not sync */
1981 if (IS_NOCMTIME(inode))
1984 now = current_time(inode);
1985 if (!timespec64_equal(&inode->i_mtime, &now))
1988 if (!timespec64_equal(&inode->i_ctime, &now))
1991 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
1992 sync_it |= S_VERSION;
1997 /* Finally allowed to write? Takes lock. */
1998 if (__mnt_want_write_file(file))
2001 ret = update_time(inode, &now, sync_it);
2002 __mnt_drop_write_file(file);
2006 EXPORT_SYMBOL(file_update_time);
2008 /* Caller must hold the file's inode lock */
2009 int file_modified(struct file *file)
2014 * Clear the security bits if the process is not being run by root.
2015 * This keeps people from modifying setuid and setgid binaries.
2017 err = file_remove_privs(file);
2021 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2024 return file_update_time(file);
2026 EXPORT_SYMBOL(file_modified);
2028 int inode_needs_sync(struct inode *inode)
2032 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2036 EXPORT_SYMBOL(inode_needs_sync);
2039 * If we try to find an inode in the inode hash while it is being
2040 * deleted, we have to wait until the filesystem completes its
2041 * deletion before reporting that it isn't found. This function waits
2042 * until the deletion _might_ have completed. Callers are responsible
2043 * to recheck inode state.
2045 * It doesn't matter if I_NEW is not set initially, a call to
2046 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2049 static void __wait_on_freeing_inode(struct inode *inode)
2051 wait_queue_head_t *wq;
2052 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2053 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2054 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2055 spin_unlock(&inode->i_lock);
2056 spin_unlock(&inode_hash_lock);
2058 finish_wait(wq, &wait.wq_entry);
2059 spin_lock(&inode_hash_lock);
2062 static __initdata unsigned long ihash_entries;
2063 static int __init set_ihash_entries(char *str)
2067 ihash_entries = simple_strtoul(str, &str, 0);
2070 __setup("ihash_entries=", set_ihash_entries);
2073 * Initialize the waitqueues and inode hash table.
2075 void __init inode_init_early(void)
2077 /* If hashes are distributed across NUMA nodes, defer
2078 * hash allocation until vmalloc space is available.
2084 alloc_large_system_hash("Inode-cache",
2085 sizeof(struct hlist_head),
2088 HASH_EARLY | HASH_ZERO,
2095 void __init inode_init(void)
2097 /* inode slab cache */
2098 inode_cachep = kmem_cache_create("inode_cache",
2099 sizeof(struct inode),
2101 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2102 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2105 /* Hash may have been set up in inode_init_early */
2110 alloc_large_system_hash("Inode-cache",
2111 sizeof(struct hlist_head),
2121 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2123 inode->i_mode = mode;
2124 if (S_ISCHR(mode)) {
2125 inode->i_fop = &def_chr_fops;
2126 inode->i_rdev = rdev;
2127 } else if (S_ISBLK(mode)) {
2128 inode->i_fop = &def_blk_fops;
2129 inode->i_rdev = rdev;
2130 } else if (S_ISFIFO(mode))
2131 inode->i_fop = &pipefifo_fops;
2132 else if (S_ISSOCK(mode))
2133 ; /* leave it no_open_fops */
2135 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2136 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2139 EXPORT_SYMBOL(init_special_inode);
2142 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2143 * @mnt_userns: User namespace of the mount the inode was created from
2145 * @dir: Directory inode
2146 * @mode: mode of the new inode
2148 * If the inode has been created through an idmapped mount the user namespace of
2149 * the vfsmount must be passed through @mnt_userns. This function will then take
2150 * care to map the inode according to @mnt_userns before checking permissions
2151 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2152 * checking is to be performed on the raw inode simply passs init_user_ns.
2154 void inode_init_owner(struct user_namespace *mnt_userns, struct inode *inode,
2155 const struct inode *dir, umode_t mode)
2157 inode_fsuid_set(inode, mnt_userns);
2158 if (dir && dir->i_mode & S_ISGID) {
2159 inode->i_gid = dir->i_gid;
2161 /* Directories are special, and always inherit S_ISGID */
2164 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2165 !in_group_p(i_gid_into_mnt(mnt_userns, dir)) &&
2166 !capable_wrt_inode_uidgid(mnt_userns, dir, CAP_FSETID))
2169 inode_fsgid_set(inode, mnt_userns);
2170 inode->i_mode = mode;
2172 EXPORT_SYMBOL(inode_init_owner);
2175 * inode_owner_or_capable - check current task permissions to inode
2176 * @mnt_userns: user namespace of the mount the inode was found from
2177 * @inode: inode being checked
2179 * Return true if current either has CAP_FOWNER in a namespace with the
2180 * inode owner uid mapped, or owns the file.
2182 * If the inode has been found through an idmapped mount the user namespace of
2183 * the vfsmount must be passed through @mnt_userns. This function will then take
2184 * care to map the inode according to @mnt_userns before checking permissions.
2185 * On non-idmapped mounts or if permission checking is to be performed on the
2186 * raw inode simply passs init_user_ns.
2188 bool inode_owner_or_capable(struct user_namespace *mnt_userns,
2189 const struct inode *inode)
2192 struct user_namespace *ns;
2194 i_uid = i_uid_into_mnt(mnt_userns, inode);
2195 if (uid_eq(current_fsuid(), i_uid))
2198 ns = current_user_ns();
2199 if (kuid_has_mapping(ns, i_uid) && ns_capable(ns, CAP_FOWNER))
2203 EXPORT_SYMBOL(inode_owner_or_capable);
2206 * Direct i/o helper functions
2208 static void __inode_dio_wait(struct inode *inode)
2210 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2211 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2214 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2215 if (atomic_read(&inode->i_dio_count))
2217 } while (atomic_read(&inode->i_dio_count));
2218 finish_wait(wq, &q.wq_entry);
2222 * inode_dio_wait - wait for outstanding DIO requests to finish
2223 * @inode: inode to wait for
2225 * Waits for all pending direct I/O requests to finish so that we can
2226 * proceed with a truncate or equivalent operation.
2228 * Must be called under a lock that serializes taking new references
2229 * to i_dio_count, usually by inode->i_mutex.
2231 void inode_dio_wait(struct inode *inode)
2233 if (atomic_read(&inode->i_dio_count))
2234 __inode_dio_wait(inode);
2236 EXPORT_SYMBOL(inode_dio_wait);
2239 * inode_set_flags - atomically set some inode flags
2241 * Note: the caller should be holding i_mutex, or else be sure that
2242 * they have exclusive access to the inode structure (i.e., while the
2243 * inode is being instantiated). The reason for the cmpxchg() loop
2244 * --- which wouldn't be necessary if all code paths which modify
2245 * i_flags actually followed this rule, is that there is at least one
2246 * code path which doesn't today so we use cmpxchg() out of an abundance
2249 * In the long run, i_mutex is overkill, and we should probably look
2250 * at using the i_lock spinlock to protect i_flags, and then make sure
2251 * it is so documented in include/linux/fs.h and that all code follows
2252 * the locking convention!!
2254 void inode_set_flags(struct inode *inode, unsigned int flags,
2257 WARN_ON_ONCE(flags & ~mask);
2258 set_mask_bits(&inode->i_flags, mask, flags);
2260 EXPORT_SYMBOL(inode_set_flags);
2262 void inode_nohighmem(struct inode *inode)
2264 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2266 EXPORT_SYMBOL(inode_nohighmem);
2269 * timestamp_truncate - Truncate timespec to a granularity
2271 * @inode: inode being updated
2273 * Truncate a timespec to the granularity supported by the fs
2274 * containing the inode. Always rounds down. gran must
2275 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2277 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2279 struct super_block *sb = inode->i_sb;
2280 unsigned int gran = sb->s_time_gran;
2282 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2283 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2286 /* Avoid division in the common cases 1 ns and 1 s. */
2289 else if (gran == NSEC_PER_SEC)
2291 else if (gran > 1 && gran < NSEC_PER_SEC)
2292 t.tv_nsec -= t.tv_nsec % gran;
2294 WARN(1, "invalid file time granularity: %u", gran);
2297 EXPORT_SYMBOL(timestamp_truncate);
2300 * current_time - Return FS time
2303 * Return the current time truncated to the time granularity supported by
2306 * Note that inode and inode->sb cannot be NULL.
2307 * Otherwise, the function warns and returns time without truncation.
2309 struct timespec64 current_time(struct inode *inode)
2311 struct timespec64 now;
2313 ktime_get_coarse_real_ts64(&now);
2315 if (unlikely(!inode->i_sb)) {
2316 WARN(1, "current_time() called with uninitialized super_block in the inode");
2320 return timestamp_truncate(now, inode);
2322 EXPORT_SYMBOL(current_time);