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 init_rwsem(&mapping->invalidate_lock);
194 lockdep_set_class_and_name(&mapping->invalidate_lock,
195 &sb->s_type->invalidate_lock_key,
196 "mapping.invalidate_lock");
197 inode->i_private = NULL;
198 inode->i_mapping = mapping;
199 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
200 #ifdef CONFIG_FS_POSIX_ACL
201 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
204 #ifdef CONFIG_FSNOTIFY
205 inode->i_fsnotify_mask = 0;
207 inode->i_flctx = NULL;
208 this_cpu_inc(nr_inodes);
214 EXPORT_SYMBOL(inode_init_always);
216 void free_inode_nonrcu(struct inode *inode)
218 kmem_cache_free(inode_cachep, inode);
220 EXPORT_SYMBOL(free_inode_nonrcu);
222 static void i_callback(struct rcu_head *head)
224 struct inode *inode = container_of(head, struct inode, i_rcu);
225 if (inode->free_inode)
226 inode->free_inode(inode);
228 free_inode_nonrcu(inode);
231 static struct inode *alloc_inode(struct super_block *sb)
233 const struct super_operations *ops = sb->s_op;
236 if (ops->alloc_inode)
237 inode = ops->alloc_inode(sb);
239 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
244 if (unlikely(inode_init_always(sb, inode))) {
245 if (ops->destroy_inode) {
246 ops->destroy_inode(inode);
247 if (!ops->free_inode)
250 inode->free_inode = ops->free_inode;
251 i_callback(&inode->i_rcu);
258 void __destroy_inode(struct inode *inode)
260 BUG_ON(inode_has_buffers(inode));
261 inode_detach_wb(inode);
262 security_inode_free(inode);
263 fsnotify_inode_delete(inode);
264 locks_free_lock_context(inode);
265 if (!inode->i_nlink) {
266 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
267 atomic_long_dec(&inode->i_sb->s_remove_count);
270 #ifdef CONFIG_FS_POSIX_ACL
271 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
272 posix_acl_release(inode->i_acl);
273 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
274 posix_acl_release(inode->i_default_acl);
276 this_cpu_dec(nr_inodes);
278 EXPORT_SYMBOL(__destroy_inode);
280 static void destroy_inode(struct inode *inode)
282 const struct super_operations *ops = inode->i_sb->s_op;
284 BUG_ON(!list_empty(&inode->i_lru));
285 __destroy_inode(inode);
286 if (ops->destroy_inode) {
287 ops->destroy_inode(inode);
288 if (!ops->free_inode)
291 inode->free_inode = ops->free_inode;
292 call_rcu(&inode->i_rcu, i_callback);
296 * drop_nlink - directly drop an inode's link count
299 * This is a low-level filesystem helper to replace any
300 * direct filesystem manipulation of i_nlink. In cases
301 * where we are attempting to track writes to the
302 * filesystem, a decrement to zero means an imminent
303 * write when the file is truncated and actually unlinked
306 void drop_nlink(struct inode *inode)
308 WARN_ON(inode->i_nlink == 0);
311 atomic_long_inc(&inode->i_sb->s_remove_count);
313 EXPORT_SYMBOL(drop_nlink);
316 * clear_nlink - directly zero an inode's link count
319 * This is a low-level filesystem helper to replace any
320 * direct filesystem manipulation of i_nlink. See
321 * drop_nlink() for why we care about i_nlink hitting zero.
323 void clear_nlink(struct inode *inode)
325 if (inode->i_nlink) {
326 inode->__i_nlink = 0;
327 atomic_long_inc(&inode->i_sb->s_remove_count);
330 EXPORT_SYMBOL(clear_nlink);
333 * set_nlink - directly set an inode's link count
335 * @nlink: new nlink (should be non-zero)
337 * This is a low-level filesystem helper to replace any
338 * direct filesystem manipulation of i_nlink.
340 void set_nlink(struct inode *inode, unsigned int nlink)
345 /* Yes, some filesystems do change nlink from zero to one */
346 if (inode->i_nlink == 0)
347 atomic_long_dec(&inode->i_sb->s_remove_count);
349 inode->__i_nlink = nlink;
352 EXPORT_SYMBOL(set_nlink);
355 * inc_nlink - directly increment an inode's link count
358 * This is a low-level filesystem helper to replace any
359 * direct filesystem manipulation of i_nlink. Currently,
360 * it is only here for parity with dec_nlink().
362 void inc_nlink(struct inode *inode)
364 if (unlikely(inode->i_nlink == 0)) {
365 WARN_ON(!(inode->i_state & I_LINKABLE));
366 atomic_long_dec(&inode->i_sb->s_remove_count);
371 EXPORT_SYMBOL(inc_nlink);
373 static void __address_space_init_once(struct address_space *mapping)
375 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
376 init_rwsem(&mapping->i_mmap_rwsem);
377 INIT_LIST_HEAD(&mapping->private_list);
378 spin_lock_init(&mapping->private_lock);
379 mapping->i_mmap = RB_ROOT_CACHED;
382 void address_space_init_once(struct address_space *mapping)
384 memset(mapping, 0, sizeof(*mapping));
385 __address_space_init_once(mapping);
387 EXPORT_SYMBOL(address_space_init_once);
390 * These are initializations that only need to be done
391 * once, because the fields are idempotent across use
392 * of the inode, so let the slab aware of that.
394 void inode_init_once(struct inode *inode)
396 memset(inode, 0, sizeof(*inode));
397 INIT_HLIST_NODE(&inode->i_hash);
398 INIT_LIST_HEAD(&inode->i_devices);
399 INIT_LIST_HEAD(&inode->i_io_list);
400 INIT_LIST_HEAD(&inode->i_wb_list);
401 INIT_LIST_HEAD(&inode->i_lru);
402 __address_space_init_once(&inode->i_data);
403 i_size_ordered_init(inode);
405 EXPORT_SYMBOL(inode_init_once);
407 static void init_once(void *foo)
409 struct inode *inode = (struct inode *) foo;
411 inode_init_once(inode);
415 * inode->i_lock must be held
417 void __iget(struct inode *inode)
419 atomic_inc(&inode->i_count);
423 * get additional reference to inode; caller must already hold one.
425 void ihold(struct inode *inode)
427 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
429 EXPORT_SYMBOL(ihold);
431 static void inode_lru_list_add(struct inode *inode)
433 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
434 this_cpu_inc(nr_unused);
436 inode->i_state |= I_REFERENCED;
440 * Add inode to LRU if needed (inode is unused and clean).
442 * Needs inode->i_lock held.
444 void inode_add_lru(struct inode *inode)
446 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
447 I_FREEING | I_WILL_FREE)) &&
448 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
449 inode_lru_list_add(inode);
453 static void inode_lru_list_del(struct inode *inode)
456 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
457 this_cpu_dec(nr_unused);
461 * inode_sb_list_add - add inode to the superblock list of inodes
462 * @inode: inode to add
464 void inode_sb_list_add(struct inode *inode)
466 spin_lock(&inode->i_sb->s_inode_list_lock);
467 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
468 spin_unlock(&inode->i_sb->s_inode_list_lock);
470 EXPORT_SYMBOL_GPL(inode_sb_list_add);
472 static inline void inode_sb_list_del(struct inode *inode)
474 if (!list_empty(&inode->i_sb_list)) {
475 spin_lock(&inode->i_sb->s_inode_list_lock);
476 list_del_init(&inode->i_sb_list);
477 spin_unlock(&inode->i_sb->s_inode_list_lock);
481 static unsigned long hash(struct super_block *sb, unsigned long hashval)
485 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
487 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
488 return tmp & i_hash_mask;
492 * __insert_inode_hash - hash an inode
493 * @inode: unhashed inode
494 * @hashval: unsigned long value used to locate this object in the
497 * Add an inode to the inode hash for this superblock.
499 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
501 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
503 spin_lock(&inode_hash_lock);
504 spin_lock(&inode->i_lock);
505 hlist_add_head_rcu(&inode->i_hash, b);
506 spin_unlock(&inode->i_lock);
507 spin_unlock(&inode_hash_lock);
509 EXPORT_SYMBOL(__insert_inode_hash);
512 * __remove_inode_hash - remove an inode from the hash
513 * @inode: inode to unhash
515 * Remove an inode from the superblock.
517 void __remove_inode_hash(struct inode *inode)
519 spin_lock(&inode_hash_lock);
520 spin_lock(&inode->i_lock);
521 hlist_del_init_rcu(&inode->i_hash);
522 spin_unlock(&inode->i_lock);
523 spin_unlock(&inode_hash_lock);
525 EXPORT_SYMBOL(__remove_inode_hash);
527 void clear_inode(struct inode *inode)
530 * We have to cycle the i_pages lock here because reclaim can be in the
531 * process of removing the last page (in __delete_from_page_cache())
532 * and we must not free the mapping under it.
534 xa_lock_irq(&inode->i_data.i_pages);
535 BUG_ON(inode->i_data.nrpages);
537 * Almost always, mapping_empty(&inode->i_data) here; but there are
538 * two known and long-standing ways in which nodes may get left behind
539 * (when deep radix-tree node allocation failed partway; or when THP
540 * collapse_file() failed). Until those two known cases are cleaned up,
541 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
542 * nor even WARN_ON(!mapping_empty).
544 xa_unlock_irq(&inode->i_data.i_pages);
545 BUG_ON(!list_empty(&inode->i_data.private_list));
546 BUG_ON(!(inode->i_state & I_FREEING));
547 BUG_ON(inode->i_state & I_CLEAR);
548 BUG_ON(!list_empty(&inode->i_wb_list));
549 /* don't need i_lock here, no concurrent mods to i_state */
550 inode->i_state = I_FREEING | I_CLEAR;
552 EXPORT_SYMBOL(clear_inode);
555 * Free the inode passed in, removing it from the lists it is still connected
556 * to. We remove any pages still attached to the inode and wait for any IO that
557 * is still in progress before finally destroying the inode.
559 * An inode must already be marked I_FREEING so that we avoid the inode being
560 * moved back onto lists if we race with other code that manipulates the lists
561 * (e.g. writeback_single_inode). The caller is responsible for setting this.
563 * An inode must already be removed from the LRU list before being evicted from
564 * the cache. This should occur atomically with setting the I_FREEING state
565 * flag, so no inodes here should ever be on the LRU when being evicted.
567 static void evict(struct inode *inode)
569 const struct super_operations *op = inode->i_sb->s_op;
571 BUG_ON(!(inode->i_state & I_FREEING));
572 BUG_ON(!list_empty(&inode->i_lru));
574 if (!list_empty(&inode->i_io_list))
575 inode_io_list_del(inode);
577 inode_sb_list_del(inode);
580 * Wait for flusher thread to be done with the inode so that filesystem
581 * does not start destroying it while writeback is still running. Since
582 * the inode has I_FREEING set, flusher thread won't start new work on
583 * the inode. We just have to wait for running writeback to finish.
585 inode_wait_for_writeback(inode);
587 if (op->evict_inode) {
588 op->evict_inode(inode);
590 truncate_inode_pages_final(&inode->i_data);
593 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
596 remove_inode_hash(inode);
598 spin_lock(&inode->i_lock);
599 wake_up_bit(&inode->i_state, __I_NEW);
600 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
601 spin_unlock(&inode->i_lock);
603 destroy_inode(inode);
607 * dispose_list - dispose of the contents of a local list
608 * @head: the head of the list to free
610 * Dispose-list gets a local list with local inodes in it, so it doesn't
611 * need to worry about list corruption and SMP locks.
613 static void dispose_list(struct list_head *head)
615 while (!list_empty(head)) {
618 inode = list_first_entry(head, struct inode, i_lru);
619 list_del_init(&inode->i_lru);
627 * evict_inodes - evict all evictable inodes for a superblock
628 * @sb: superblock to operate on
630 * Make sure that no inodes with zero refcount are retained. This is
631 * called by superblock shutdown after having SB_ACTIVE flag removed,
632 * so any inode reaching zero refcount during or after that call will
633 * be immediately evicted.
635 void evict_inodes(struct super_block *sb)
637 struct inode *inode, *next;
641 spin_lock(&sb->s_inode_list_lock);
642 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
643 if (atomic_read(&inode->i_count))
646 spin_lock(&inode->i_lock);
647 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
648 spin_unlock(&inode->i_lock);
652 inode->i_state |= I_FREEING;
653 inode_lru_list_del(inode);
654 spin_unlock(&inode->i_lock);
655 list_add(&inode->i_lru, &dispose);
658 * We can have a ton of inodes to evict at unmount time given
659 * enough memory, check to see if we need to go to sleep for a
660 * bit so we don't livelock.
662 if (need_resched()) {
663 spin_unlock(&sb->s_inode_list_lock);
665 dispose_list(&dispose);
669 spin_unlock(&sb->s_inode_list_lock);
671 dispose_list(&dispose);
673 EXPORT_SYMBOL_GPL(evict_inodes);
676 * invalidate_inodes - attempt to free all inodes on a superblock
677 * @sb: superblock to operate on
678 * @kill_dirty: flag to guide handling of dirty inodes
680 * Attempts to free all inodes for a given superblock. If there were any
681 * busy inodes return a non-zero value, else zero.
682 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
685 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
688 struct inode *inode, *next;
692 spin_lock(&sb->s_inode_list_lock);
693 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
694 spin_lock(&inode->i_lock);
695 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
696 spin_unlock(&inode->i_lock);
699 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
700 spin_unlock(&inode->i_lock);
704 if (atomic_read(&inode->i_count)) {
705 spin_unlock(&inode->i_lock);
710 inode->i_state |= I_FREEING;
711 inode_lru_list_del(inode);
712 spin_unlock(&inode->i_lock);
713 list_add(&inode->i_lru, &dispose);
714 if (need_resched()) {
715 spin_unlock(&sb->s_inode_list_lock);
717 dispose_list(&dispose);
721 spin_unlock(&sb->s_inode_list_lock);
723 dispose_list(&dispose);
729 * Isolate the inode from the LRU in preparation for freeing it.
731 * Any inodes which are pinned purely because of attached pagecache have their
732 * pagecache removed. If the inode has metadata buffers attached to
733 * mapping->private_list then try to remove them.
735 * If the inode has the I_REFERENCED flag set, then it means that it has been
736 * used recently - the flag is set in iput_final(). When we encounter such an
737 * inode, clear the flag and move it to the back of the LRU so it gets another
738 * pass through the LRU before it gets reclaimed. This is necessary because of
739 * the fact we are doing lazy LRU updates to minimise lock contention so the
740 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
741 * with this flag set because they are the inodes that are out of order.
743 static enum lru_status inode_lru_isolate(struct list_head *item,
744 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
746 struct list_head *freeable = arg;
747 struct inode *inode = container_of(item, struct inode, i_lru);
750 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
751 * If we fail to get the lock, just skip it.
753 if (!spin_trylock(&inode->i_lock))
757 * Referenced or dirty inodes are still in use. Give them another pass
758 * through the LRU as we canot reclaim them now.
760 if (atomic_read(&inode->i_count) ||
761 (inode->i_state & ~I_REFERENCED)) {
762 list_lru_isolate(lru, &inode->i_lru);
763 spin_unlock(&inode->i_lock);
764 this_cpu_dec(nr_unused);
768 /* recently referenced inodes get one more pass */
769 if (inode->i_state & I_REFERENCED) {
770 inode->i_state &= ~I_REFERENCED;
771 spin_unlock(&inode->i_lock);
775 if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
777 spin_unlock(&inode->i_lock);
778 spin_unlock(lru_lock);
779 if (remove_inode_buffers(inode)) {
781 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
782 if (current_is_kswapd())
783 __count_vm_events(KSWAPD_INODESTEAL, reap);
785 __count_vm_events(PGINODESTEAL, reap);
786 if (current->reclaim_state)
787 current->reclaim_state->reclaimed_slab += reap;
794 WARN_ON(inode->i_state & I_NEW);
795 inode->i_state |= I_FREEING;
796 list_lru_isolate_move(lru, &inode->i_lru, freeable);
797 spin_unlock(&inode->i_lock);
799 this_cpu_dec(nr_unused);
804 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
805 * This is called from the superblock shrinker function with a number of inodes
806 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
807 * then are freed outside inode_lock by dispose_list().
809 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
814 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
815 inode_lru_isolate, &freeable);
816 dispose_list(&freeable);
820 static void __wait_on_freeing_inode(struct inode *inode);
822 * Called with the inode lock held.
824 static struct inode *find_inode(struct super_block *sb,
825 struct hlist_head *head,
826 int (*test)(struct inode *, void *),
829 struct inode *inode = NULL;
832 hlist_for_each_entry(inode, head, i_hash) {
833 if (inode->i_sb != sb)
835 if (!test(inode, data))
837 spin_lock(&inode->i_lock);
838 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
839 __wait_on_freeing_inode(inode);
842 if (unlikely(inode->i_state & I_CREATING)) {
843 spin_unlock(&inode->i_lock);
844 return ERR_PTR(-ESTALE);
847 spin_unlock(&inode->i_lock);
854 * find_inode_fast is the fast path version of find_inode, see the comment at
855 * iget_locked for details.
857 static struct inode *find_inode_fast(struct super_block *sb,
858 struct hlist_head *head, unsigned long ino)
860 struct inode *inode = NULL;
863 hlist_for_each_entry(inode, head, i_hash) {
864 if (inode->i_ino != ino)
866 if (inode->i_sb != sb)
868 spin_lock(&inode->i_lock);
869 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
870 __wait_on_freeing_inode(inode);
873 if (unlikely(inode->i_state & I_CREATING)) {
874 spin_unlock(&inode->i_lock);
875 return ERR_PTR(-ESTALE);
878 spin_unlock(&inode->i_lock);
885 * Each cpu owns a range of LAST_INO_BATCH numbers.
886 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
887 * to renew the exhausted range.
889 * This does not significantly increase overflow rate because every CPU can
890 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
891 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
892 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
893 * overflow rate by 2x, which does not seem too significant.
895 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
896 * error if st_ino won't fit in target struct field. Use 32bit counter
897 * here to attempt to avoid that.
899 #define LAST_INO_BATCH 1024
900 static DEFINE_PER_CPU(unsigned int, last_ino);
902 unsigned int get_next_ino(void)
904 unsigned int *p = &get_cpu_var(last_ino);
905 unsigned int res = *p;
908 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
909 static atomic_t shared_last_ino;
910 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
912 res = next - LAST_INO_BATCH;
917 /* get_next_ino should not provide a 0 inode number */
921 put_cpu_var(last_ino);
924 EXPORT_SYMBOL(get_next_ino);
927 * new_inode_pseudo - obtain an inode
930 * Allocates a new inode for given superblock.
931 * Inode wont be chained in superblock s_inodes list
933 * - fs can't be unmount
934 * - quotas, fsnotify, writeback can't work
936 struct inode *new_inode_pseudo(struct super_block *sb)
938 struct inode *inode = alloc_inode(sb);
941 spin_lock(&inode->i_lock);
943 spin_unlock(&inode->i_lock);
944 INIT_LIST_HEAD(&inode->i_sb_list);
950 * new_inode - obtain an inode
953 * Allocates a new inode for given superblock. The default gfp_mask
954 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
955 * If HIGHMEM pages are unsuitable or it is known that pages allocated
956 * for the page cache are not reclaimable or migratable,
957 * mapping_set_gfp_mask() must be called with suitable flags on the
958 * newly created inode's mapping
961 struct inode *new_inode(struct super_block *sb)
965 spin_lock_prefetch(&sb->s_inode_list_lock);
967 inode = new_inode_pseudo(sb);
969 inode_sb_list_add(inode);
972 EXPORT_SYMBOL(new_inode);
974 #ifdef CONFIG_DEBUG_LOCK_ALLOC
975 void lockdep_annotate_inode_mutex_key(struct inode *inode)
977 if (S_ISDIR(inode->i_mode)) {
978 struct file_system_type *type = inode->i_sb->s_type;
980 /* Set new key only if filesystem hasn't already changed it */
981 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
983 * ensure nobody is actually holding i_mutex
985 // mutex_destroy(&inode->i_mutex);
986 init_rwsem(&inode->i_rwsem);
987 lockdep_set_class(&inode->i_rwsem,
988 &type->i_mutex_dir_key);
992 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
996 * unlock_new_inode - clear the I_NEW state and wake up any waiters
997 * @inode: new inode to unlock
999 * Called when the inode is fully initialised to clear the new state of the
1000 * inode and wake up anyone waiting for the inode to finish initialisation.
1002 void unlock_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 & ~I_CREATING;
1009 wake_up_bit(&inode->i_state, __I_NEW);
1010 spin_unlock(&inode->i_lock);
1012 EXPORT_SYMBOL(unlock_new_inode);
1014 void discard_new_inode(struct inode *inode)
1016 lockdep_annotate_inode_mutex_key(inode);
1017 spin_lock(&inode->i_lock);
1018 WARN_ON(!(inode->i_state & I_NEW));
1019 inode->i_state &= ~I_NEW;
1021 wake_up_bit(&inode->i_state, __I_NEW);
1022 spin_unlock(&inode->i_lock);
1025 EXPORT_SYMBOL(discard_new_inode);
1028 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1030 * Lock any non-NULL argument that is not a directory.
1031 * Zero, one or two objects may be locked by this function.
1033 * @inode1: first inode to lock
1034 * @inode2: second inode to lock
1036 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1038 if (inode1 > inode2)
1039 swap(inode1, inode2);
1041 if (inode1 && !S_ISDIR(inode1->i_mode))
1043 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1044 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1046 EXPORT_SYMBOL(lock_two_nondirectories);
1049 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1050 * @inode1: first inode to unlock
1051 * @inode2: second inode to unlock
1053 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1055 if (inode1 && !S_ISDIR(inode1->i_mode))
1056 inode_unlock(inode1);
1057 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1058 inode_unlock(inode2);
1060 EXPORT_SYMBOL(unlock_two_nondirectories);
1063 * inode_insert5 - obtain an inode from a mounted file system
1064 * @inode: pre-allocated inode to use for insert to cache
1065 * @hashval: hash value (usually inode number) to get
1066 * @test: callback used for comparisons between inodes
1067 * @set: callback used to initialize a new struct inode
1068 * @data: opaque data pointer to pass to @test and @set
1070 * Search for the inode specified by @hashval and @data in the inode cache,
1071 * and if present it is return it with an increased reference count. This is
1072 * a variant of iget5_locked() for callers that don't want to fail on memory
1073 * allocation of inode.
1075 * If the inode is not in cache, insert the pre-allocated inode to cache and
1076 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1077 * to fill it in before unlocking it via unlock_new_inode().
1079 * Note both @test and @set are called with the inode_hash_lock held, so can't
1082 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1083 int (*test)(struct inode *, void *),
1084 int (*set)(struct inode *, void *), void *data)
1086 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1088 bool creating = inode->i_state & I_CREATING;
1091 spin_lock(&inode_hash_lock);
1092 old = find_inode(inode->i_sb, head, test, data);
1093 if (unlikely(old)) {
1095 * Uhhuh, somebody else created the same inode under us.
1096 * Use the old inode instead of the preallocated one.
1098 spin_unlock(&inode_hash_lock);
1102 if (unlikely(inode_unhashed(old))) {
1109 if (set && unlikely(set(inode, data))) {
1115 * Return the locked inode with I_NEW set, the
1116 * caller is responsible for filling in the contents
1118 spin_lock(&inode->i_lock);
1119 inode->i_state |= I_NEW;
1120 hlist_add_head_rcu(&inode->i_hash, head);
1121 spin_unlock(&inode->i_lock);
1123 inode_sb_list_add(inode);
1125 spin_unlock(&inode_hash_lock);
1129 EXPORT_SYMBOL(inode_insert5);
1132 * iget5_locked - obtain an inode from a mounted file system
1133 * @sb: super block of file system
1134 * @hashval: hash value (usually inode number) to get
1135 * @test: callback used for comparisons between inodes
1136 * @set: callback used to initialize a new struct inode
1137 * @data: opaque data pointer to pass to @test and @set
1139 * Search for the inode specified by @hashval and @data in the inode cache,
1140 * and if present it is return it with an increased reference count. This is
1141 * a generalized version of iget_locked() for file systems where the inode
1142 * number is not sufficient for unique identification of an inode.
1144 * If the inode is not in cache, allocate a new inode and return it locked,
1145 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1146 * before unlocking it via unlock_new_inode().
1148 * Note both @test and @set are called with the inode_hash_lock held, so can't
1151 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1152 int (*test)(struct inode *, void *),
1153 int (*set)(struct inode *, void *), void *data)
1155 struct inode *inode = ilookup5(sb, hashval, test, data);
1158 struct inode *new = alloc_inode(sb);
1162 inode = inode_insert5(new, hashval, test, set, data);
1163 if (unlikely(inode != new))
1169 EXPORT_SYMBOL(iget5_locked);
1172 * iget_locked - obtain an inode from a mounted file system
1173 * @sb: super block of file system
1174 * @ino: inode number to get
1176 * Search for the inode specified by @ino in the inode cache and if present
1177 * return it with an increased reference count. This is for file systems
1178 * where the inode number is sufficient for unique identification of an inode.
1180 * If the inode is not in cache, allocate a new inode and return it locked,
1181 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1182 * before unlocking it via unlock_new_inode().
1184 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1186 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1187 struct inode *inode;
1189 spin_lock(&inode_hash_lock);
1190 inode = find_inode_fast(sb, head, ino);
1191 spin_unlock(&inode_hash_lock);
1195 wait_on_inode(inode);
1196 if (unlikely(inode_unhashed(inode))) {
1203 inode = alloc_inode(sb);
1207 spin_lock(&inode_hash_lock);
1208 /* We released the lock, so.. */
1209 old = find_inode_fast(sb, head, ino);
1212 spin_lock(&inode->i_lock);
1213 inode->i_state = I_NEW;
1214 hlist_add_head_rcu(&inode->i_hash, head);
1215 spin_unlock(&inode->i_lock);
1216 inode_sb_list_add(inode);
1217 spin_unlock(&inode_hash_lock);
1219 /* Return the locked inode with I_NEW set, the
1220 * caller is responsible for filling in the contents
1226 * Uhhuh, somebody else created the same inode under
1227 * us. Use the old inode instead of the one we just
1230 spin_unlock(&inode_hash_lock);
1231 destroy_inode(inode);
1235 wait_on_inode(inode);
1236 if (unlikely(inode_unhashed(inode))) {
1243 EXPORT_SYMBOL(iget_locked);
1246 * search the inode cache for a matching inode number.
1247 * If we find one, then the inode number we are trying to
1248 * allocate is not unique and so we should not use it.
1250 * Returns 1 if the inode number is unique, 0 if it is not.
1252 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1254 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1255 struct inode *inode;
1257 hlist_for_each_entry_rcu(inode, b, i_hash) {
1258 if (inode->i_ino == ino && inode->i_sb == sb)
1265 * iunique - get a unique inode number
1267 * @max_reserved: highest reserved inode number
1269 * Obtain an inode number that is unique on the system for a given
1270 * superblock. This is used by file systems that have no natural
1271 * permanent inode numbering system. An inode number is returned that
1272 * is higher than the reserved limit but unique.
1275 * With a large number of inodes live on the file system this function
1276 * currently becomes quite slow.
1278 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1281 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1282 * error if st_ino won't fit in target struct field. Use 32bit counter
1283 * here to attempt to avoid that.
1285 static DEFINE_SPINLOCK(iunique_lock);
1286 static unsigned int counter;
1290 spin_lock(&iunique_lock);
1292 if (counter <= max_reserved)
1293 counter = max_reserved + 1;
1295 } while (!test_inode_iunique(sb, res));
1296 spin_unlock(&iunique_lock);
1301 EXPORT_SYMBOL(iunique);
1303 struct inode *igrab(struct inode *inode)
1305 spin_lock(&inode->i_lock);
1306 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1308 spin_unlock(&inode->i_lock);
1310 spin_unlock(&inode->i_lock);
1312 * Handle the case where s_op->clear_inode is not been
1313 * called yet, and somebody is calling igrab
1314 * while the inode is getting freed.
1320 EXPORT_SYMBOL(igrab);
1323 * ilookup5_nowait - search for an inode in the inode cache
1324 * @sb: super block of file system to search
1325 * @hashval: hash value (usually inode number) to search for
1326 * @test: callback used for comparisons between inodes
1327 * @data: opaque data pointer to pass to @test
1329 * Search for the inode specified by @hashval and @data in the inode cache.
1330 * If the inode is in the cache, the inode is returned with an incremented
1333 * Note: I_NEW is not waited upon so you have to be very careful what you do
1334 * with the returned inode. You probably should be using ilookup5() instead.
1336 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1338 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1339 int (*test)(struct inode *, void *), void *data)
1341 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1342 struct inode *inode;
1344 spin_lock(&inode_hash_lock);
1345 inode = find_inode(sb, head, test, data);
1346 spin_unlock(&inode_hash_lock);
1348 return IS_ERR(inode) ? NULL : inode;
1350 EXPORT_SYMBOL(ilookup5_nowait);
1353 * ilookup5 - search for an inode in the inode cache
1354 * @sb: super block of file system to search
1355 * @hashval: hash value (usually inode number) to search for
1356 * @test: callback used for comparisons between inodes
1357 * @data: opaque data pointer to pass to @test
1359 * Search for the inode specified by @hashval and @data in the inode cache,
1360 * and if the inode is in the cache, return the inode with an incremented
1361 * reference count. Waits on I_NEW before returning the inode.
1362 * returned with an incremented reference count.
1364 * This is a generalized version of ilookup() for file systems where the
1365 * inode number is not sufficient for unique identification of an inode.
1367 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1369 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1370 int (*test)(struct inode *, void *), void *data)
1372 struct inode *inode;
1374 inode = ilookup5_nowait(sb, hashval, test, data);
1376 wait_on_inode(inode);
1377 if (unlikely(inode_unhashed(inode))) {
1384 EXPORT_SYMBOL(ilookup5);
1387 * ilookup - search for an inode in the inode cache
1388 * @sb: super block of file system to search
1389 * @ino: inode number to search for
1391 * Search for the inode @ino in the inode cache, and if the inode is in the
1392 * cache, the inode is returned with an incremented reference count.
1394 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1396 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1397 struct inode *inode;
1399 spin_lock(&inode_hash_lock);
1400 inode = find_inode_fast(sb, head, ino);
1401 spin_unlock(&inode_hash_lock);
1406 wait_on_inode(inode);
1407 if (unlikely(inode_unhashed(inode))) {
1414 EXPORT_SYMBOL(ilookup);
1417 * find_inode_nowait - find an inode in the inode cache
1418 * @sb: super block of file system to search
1419 * @hashval: hash value (usually inode number) to search for
1420 * @match: callback used for comparisons between inodes
1421 * @data: opaque data pointer to pass to @match
1423 * Search for the inode specified by @hashval and @data in the inode
1424 * cache, where the helper function @match will return 0 if the inode
1425 * does not match, 1 if the inode does match, and -1 if the search
1426 * should be stopped. The @match function must be responsible for
1427 * taking the i_lock spin_lock and checking i_state for an inode being
1428 * freed or being initialized, and incrementing the reference count
1429 * before returning 1. It also must not sleep, since it is called with
1430 * the inode_hash_lock spinlock held.
1432 * This is a even more generalized version of ilookup5() when the
1433 * function must never block --- find_inode() can block in
1434 * __wait_on_freeing_inode() --- or when the caller can not increment
1435 * the reference count because the resulting iput() might cause an
1436 * inode eviction. The tradeoff is that the @match funtion must be
1437 * very carefully implemented.
1439 struct inode *find_inode_nowait(struct super_block *sb,
1440 unsigned long hashval,
1441 int (*match)(struct inode *, unsigned long,
1445 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1446 struct inode *inode, *ret_inode = NULL;
1449 spin_lock(&inode_hash_lock);
1450 hlist_for_each_entry(inode, head, i_hash) {
1451 if (inode->i_sb != sb)
1453 mval = match(inode, hashval, data);
1461 spin_unlock(&inode_hash_lock);
1464 EXPORT_SYMBOL(find_inode_nowait);
1467 * find_inode_rcu - find an inode in the inode cache
1468 * @sb: Super block of file system to search
1469 * @hashval: Key to hash
1470 * @test: Function to test match on an inode
1471 * @data: Data for test function
1473 * Search for the inode specified by @hashval and @data in the inode cache,
1474 * where the helper function @test will return 0 if the inode does not match
1475 * and 1 if it does. The @test function must be responsible for taking the
1476 * i_lock spin_lock and checking i_state for an inode being freed or being
1479 * If successful, this will return the inode for which the @test function
1480 * returned 1 and NULL otherwise.
1482 * The @test function is not permitted to take a ref on any inode presented.
1483 * It is also not permitted to sleep.
1485 * The caller must hold the RCU read lock.
1487 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1488 int (*test)(struct inode *, void *), void *data)
1490 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1491 struct inode *inode;
1493 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1494 "suspicious find_inode_rcu() usage");
1496 hlist_for_each_entry_rcu(inode, head, i_hash) {
1497 if (inode->i_sb == sb &&
1498 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1504 EXPORT_SYMBOL(find_inode_rcu);
1507 * find_inode_by_ino_rcu - Find an inode in the inode cache
1508 * @sb: Super block of file system to search
1509 * @ino: The inode number to match
1511 * Search for the inode specified by @hashval and @data in the inode cache,
1512 * where the helper function @test will return 0 if the inode does not match
1513 * and 1 if it does. The @test function must be responsible for taking the
1514 * i_lock spin_lock and checking i_state for an inode being freed or being
1517 * If successful, this will return the inode for which the @test function
1518 * returned 1 and NULL otherwise.
1520 * The @test function is not permitted to take a ref on any inode presented.
1521 * It is also not permitted to sleep.
1523 * The caller must hold the RCU read lock.
1525 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1528 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1529 struct inode *inode;
1531 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1532 "suspicious find_inode_by_ino_rcu() usage");
1534 hlist_for_each_entry_rcu(inode, head, i_hash) {
1535 if (inode->i_ino == ino &&
1536 inode->i_sb == sb &&
1537 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1542 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1544 int insert_inode_locked(struct inode *inode)
1546 struct super_block *sb = inode->i_sb;
1547 ino_t ino = inode->i_ino;
1548 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1551 struct inode *old = NULL;
1552 spin_lock(&inode_hash_lock);
1553 hlist_for_each_entry(old, head, i_hash) {
1554 if (old->i_ino != ino)
1556 if (old->i_sb != sb)
1558 spin_lock(&old->i_lock);
1559 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1560 spin_unlock(&old->i_lock);
1566 spin_lock(&inode->i_lock);
1567 inode->i_state |= I_NEW | I_CREATING;
1568 hlist_add_head_rcu(&inode->i_hash, head);
1569 spin_unlock(&inode->i_lock);
1570 spin_unlock(&inode_hash_lock);
1573 if (unlikely(old->i_state & I_CREATING)) {
1574 spin_unlock(&old->i_lock);
1575 spin_unlock(&inode_hash_lock);
1579 spin_unlock(&old->i_lock);
1580 spin_unlock(&inode_hash_lock);
1582 if (unlikely(!inode_unhashed(old))) {
1589 EXPORT_SYMBOL(insert_inode_locked);
1591 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1592 int (*test)(struct inode *, void *), void *data)
1596 inode->i_state |= I_CREATING;
1597 old = inode_insert5(inode, hashval, test, NULL, data);
1605 EXPORT_SYMBOL(insert_inode_locked4);
1608 int generic_delete_inode(struct inode *inode)
1612 EXPORT_SYMBOL(generic_delete_inode);
1615 * Called when we're dropping the last reference
1618 * Call the FS "drop_inode()" function, defaulting to
1619 * the legacy UNIX filesystem behaviour. If it tells
1620 * us to evict inode, do so. Otherwise, retain inode
1621 * in cache if fs is alive, sync and evict if fs is
1624 static void iput_final(struct inode *inode)
1626 struct super_block *sb = inode->i_sb;
1627 const struct super_operations *op = inode->i_sb->s_op;
1628 unsigned long state;
1631 WARN_ON(inode->i_state & I_NEW);
1634 drop = op->drop_inode(inode);
1636 drop = generic_drop_inode(inode);
1639 !(inode->i_state & I_DONTCACHE) &&
1640 (sb->s_flags & SB_ACTIVE)) {
1641 inode_add_lru(inode);
1642 spin_unlock(&inode->i_lock);
1646 state = inode->i_state;
1648 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1649 spin_unlock(&inode->i_lock);
1651 write_inode_now(inode, 1);
1653 spin_lock(&inode->i_lock);
1654 state = inode->i_state;
1655 WARN_ON(state & I_NEW);
1656 state &= ~I_WILL_FREE;
1659 WRITE_ONCE(inode->i_state, state | I_FREEING);
1660 if (!list_empty(&inode->i_lru))
1661 inode_lru_list_del(inode);
1662 spin_unlock(&inode->i_lock);
1668 * iput - put an inode
1669 * @inode: inode to put
1671 * Puts an inode, dropping its usage count. If the inode use count hits
1672 * zero, the inode is then freed and may also be destroyed.
1674 * Consequently, iput() can sleep.
1676 void iput(struct inode *inode)
1680 BUG_ON(inode->i_state & I_CLEAR);
1682 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1683 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1684 atomic_inc(&inode->i_count);
1685 spin_unlock(&inode->i_lock);
1686 trace_writeback_lazytime_iput(inode);
1687 mark_inode_dirty_sync(inode);
1693 EXPORT_SYMBOL(iput);
1697 * bmap - find a block number in a file
1698 * @inode: inode owning the block number being requested
1699 * @block: pointer containing the block to find
1701 * Replaces the value in ``*block`` with the block number on the device holding
1702 * corresponding to the requested block number in the file.
1703 * That is, asked for block 4 of inode 1 the function will replace the
1704 * 4 in ``*block``, with disk block relative to the disk start that holds that
1705 * block of the file.
1707 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1708 * hole, returns 0 and ``*block`` is also set to 0.
1710 int bmap(struct inode *inode, sector_t *block)
1712 if (!inode->i_mapping->a_ops->bmap)
1715 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1718 EXPORT_SYMBOL(bmap);
1722 * With relative atime, only update atime if the previous atime is
1723 * earlier than either the ctime or mtime or if at least a day has
1724 * passed since the last atime update.
1726 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1727 struct timespec64 now)
1730 if (!(mnt->mnt_flags & MNT_RELATIME))
1733 * Is mtime younger than atime? If yes, update atime:
1735 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1738 * Is ctime younger than atime? If yes, update atime:
1740 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1744 * Is the previous atime value older than a day? If yes,
1747 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1750 * Good, we can skip the atime update:
1755 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1757 int dirty_flags = 0;
1759 if (flags & (S_ATIME | S_CTIME | S_MTIME)) {
1760 if (flags & S_ATIME)
1761 inode->i_atime = *time;
1762 if (flags & S_CTIME)
1763 inode->i_ctime = *time;
1764 if (flags & S_MTIME)
1765 inode->i_mtime = *time;
1767 if (inode->i_sb->s_flags & SB_LAZYTIME)
1768 dirty_flags |= I_DIRTY_TIME;
1770 dirty_flags |= I_DIRTY_SYNC;
1773 if ((flags & S_VERSION) && inode_maybe_inc_iversion(inode, false))
1774 dirty_flags |= I_DIRTY_SYNC;
1776 __mark_inode_dirty(inode, dirty_flags);
1779 EXPORT_SYMBOL(generic_update_time);
1782 * This does the actual work of updating an inodes time or version. Must have
1783 * had called mnt_want_write() before calling this.
1785 static int update_time(struct inode *inode, struct timespec64 *time, int flags)
1787 if (inode->i_op->update_time)
1788 return inode->i_op->update_time(inode, time, flags);
1789 return generic_update_time(inode, time, flags);
1793 * atime_needs_update - update the access time
1794 * @path: the &struct path to update
1795 * @inode: inode to update
1797 * Update the accessed time on an inode and mark it for writeback.
1798 * This function automatically handles read only file systems and media,
1799 * as well as the "noatime" flag and inode specific "noatime" markers.
1801 bool atime_needs_update(const struct path *path, struct inode *inode)
1803 struct vfsmount *mnt = path->mnt;
1804 struct timespec64 now;
1806 if (inode->i_flags & S_NOATIME)
1809 /* Atime updates will likely cause i_uid and i_gid to be written
1810 * back improprely if their true value is unknown to the vfs.
1812 if (HAS_UNMAPPED_ID(mnt_user_ns(mnt), inode))
1815 if (IS_NOATIME(inode))
1817 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1820 if (mnt->mnt_flags & MNT_NOATIME)
1822 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1825 now = current_time(inode);
1827 if (!relatime_need_update(mnt, inode, now))
1830 if (timespec64_equal(&inode->i_atime, &now))
1836 void touch_atime(const struct path *path)
1838 struct vfsmount *mnt = path->mnt;
1839 struct inode *inode = d_inode(path->dentry);
1840 struct timespec64 now;
1842 if (!atime_needs_update(path, inode))
1845 if (!sb_start_write_trylock(inode->i_sb))
1848 if (__mnt_want_write(mnt) != 0)
1851 * File systems can error out when updating inodes if they need to
1852 * allocate new space to modify an inode (such is the case for
1853 * Btrfs), but since we touch atime while walking down the path we
1854 * really don't care if we failed to update the atime of the file,
1855 * so just ignore the return value.
1856 * We may also fail on filesystems that have the ability to make parts
1857 * of the fs read only, e.g. subvolumes in Btrfs.
1859 now = current_time(inode);
1860 update_time(inode, &now, S_ATIME);
1861 __mnt_drop_write(mnt);
1863 sb_end_write(inode->i_sb);
1865 EXPORT_SYMBOL(touch_atime);
1868 * The logic we want is
1870 * if suid or (sgid and xgrp)
1873 int should_remove_suid(struct dentry *dentry)
1875 umode_t mode = d_inode(dentry)->i_mode;
1878 /* suid always must be killed */
1879 if (unlikely(mode & S_ISUID))
1880 kill = ATTR_KILL_SUID;
1883 * sgid without any exec bits is just a mandatory locking mark; leave
1884 * it alone. If some exec bits are set, it's a real sgid; kill it.
1886 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1887 kill |= ATTR_KILL_SGID;
1889 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1894 EXPORT_SYMBOL(should_remove_suid);
1897 * Return mask of changes for notify_change() that need to be done as a
1898 * response to write or truncate. Return 0 if nothing has to be changed.
1899 * Negative value on error (change should be denied).
1901 int dentry_needs_remove_privs(struct dentry *dentry)
1903 struct inode *inode = d_inode(dentry);
1907 if (IS_NOSEC(inode))
1910 mask = should_remove_suid(dentry);
1911 ret = security_inode_need_killpriv(dentry);
1915 mask |= ATTR_KILL_PRIV;
1919 static int __remove_privs(struct user_namespace *mnt_userns,
1920 struct dentry *dentry, int kill)
1922 struct iattr newattrs;
1924 newattrs.ia_valid = ATTR_FORCE | kill;
1926 * Note we call this on write, so notify_change will not
1927 * encounter any conflicting delegations:
1929 return notify_change(mnt_userns, dentry, &newattrs, NULL);
1933 * Remove special file priviledges (suid, capabilities) when file is written
1936 int file_remove_privs(struct file *file)
1938 struct dentry *dentry = file_dentry(file);
1939 struct inode *inode = file_inode(file);
1944 * Fast path for nothing security related.
1945 * As well for non-regular files, e.g. blkdev inodes.
1946 * For example, blkdev_write_iter() might get here
1947 * trying to remove privs which it is not allowed to.
1949 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1952 kill = dentry_needs_remove_privs(dentry);
1956 error = __remove_privs(file_mnt_user_ns(file), dentry, kill);
1958 inode_has_no_xattr(inode);
1962 EXPORT_SYMBOL(file_remove_privs);
1965 * file_update_time - update mtime and ctime time
1966 * @file: file accessed
1968 * Update the mtime and ctime members of an inode and mark the inode
1969 * for writeback. Note that this function is meant exclusively for
1970 * usage in the file write path of filesystems, and filesystems may
1971 * choose to explicitly ignore update via this function with the
1972 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1973 * timestamps are handled by the server. This can return an error for
1974 * file systems who need to allocate space in order to update an inode.
1977 int file_update_time(struct file *file)
1979 struct inode *inode = file_inode(file);
1980 struct timespec64 now;
1984 /* First try to exhaust all avenues to not sync */
1985 if (IS_NOCMTIME(inode))
1988 now = current_time(inode);
1989 if (!timespec64_equal(&inode->i_mtime, &now))
1992 if (!timespec64_equal(&inode->i_ctime, &now))
1995 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
1996 sync_it |= S_VERSION;
2001 /* Finally allowed to write? Takes lock. */
2002 if (__mnt_want_write_file(file))
2005 ret = update_time(inode, &now, sync_it);
2006 __mnt_drop_write_file(file);
2010 EXPORT_SYMBOL(file_update_time);
2012 /* Caller must hold the file's inode lock */
2013 int file_modified(struct file *file)
2018 * Clear the security bits if the process is not being run by root.
2019 * This keeps people from modifying setuid and setgid binaries.
2021 err = file_remove_privs(file);
2025 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2028 return file_update_time(file);
2030 EXPORT_SYMBOL(file_modified);
2032 int inode_needs_sync(struct inode *inode)
2036 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2040 EXPORT_SYMBOL(inode_needs_sync);
2043 * If we try to find an inode in the inode hash while it is being
2044 * deleted, we have to wait until the filesystem completes its
2045 * deletion before reporting that it isn't found. This function waits
2046 * until the deletion _might_ have completed. Callers are responsible
2047 * to recheck inode state.
2049 * It doesn't matter if I_NEW is not set initially, a call to
2050 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2053 static void __wait_on_freeing_inode(struct inode *inode)
2055 wait_queue_head_t *wq;
2056 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2057 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2058 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2059 spin_unlock(&inode->i_lock);
2060 spin_unlock(&inode_hash_lock);
2062 finish_wait(wq, &wait.wq_entry);
2063 spin_lock(&inode_hash_lock);
2066 static __initdata unsigned long ihash_entries;
2067 static int __init set_ihash_entries(char *str)
2071 ihash_entries = simple_strtoul(str, &str, 0);
2074 __setup("ihash_entries=", set_ihash_entries);
2077 * Initialize the waitqueues and inode hash table.
2079 void __init inode_init_early(void)
2081 /* If hashes are distributed across NUMA nodes, defer
2082 * hash allocation until vmalloc space is available.
2088 alloc_large_system_hash("Inode-cache",
2089 sizeof(struct hlist_head),
2092 HASH_EARLY | HASH_ZERO,
2099 void __init inode_init(void)
2101 /* inode slab cache */
2102 inode_cachep = kmem_cache_create("inode_cache",
2103 sizeof(struct inode),
2105 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2106 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2109 /* Hash may have been set up in inode_init_early */
2114 alloc_large_system_hash("Inode-cache",
2115 sizeof(struct hlist_head),
2125 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2127 inode->i_mode = mode;
2128 if (S_ISCHR(mode)) {
2129 inode->i_fop = &def_chr_fops;
2130 inode->i_rdev = rdev;
2131 } else if (S_ISBLK(mode)) {
2132 inode->i_fop = &def_blk_fops;
2133 inode->i_rdev = rdev;
2134 } else if (S_ISFIFO(mode))
2135 inode->i_fop = &pipefifo_fops;
2136 else if (S_ISSOCK(mode))
2137 ; /* leave it no_open_fops */
2139 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2140 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2143 EXPORT_SYMBOL(init_special_inode);
2146 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2147 * @mnt_userns: User namespace of the mount the inode was created from
2149 * @dir: Directory inode
2150 * @mode: mode of the new inode
2152 * If the inode has been created through an idmapped mount the user namespace of
2153 * the vfsmount must be passed through @mnt_userns. This function will then take
2154 * care to map the inode according to @mnt_userns before checking permissions
2155 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2156 * checking is to be performed on the raw inode simply passs init_user_ns.
2158 void inode_init_owner(struct user_namespace *mnt_userns, struct inode *inode,
2159 const struct inode *dir, umode_t mode)
2161 inode_fsuid_set(inode, mnt_userns);
2162 if (dir && dir->i_mode & S_ISGID) {
2163 inode->i_gid = dir->i_gid;
2165 /* Directories are special, and always inherit S_ISGID */
2168 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2169 !in_group_p(i_gid_into_mnt(mnt_userns, dir)) &&
2170 !capable_wrt_inode_uidgid(mnt_userns, dir, CAP_FSETID))
2173 inode_fsgid_set(inode, mnt_userns);
2174 inode->i_mode = mode;
2176 EXPORT_SYMBOL(inode_init_owner);
2179 * inode_owner_or_capable - check current task permissions to inode
2180 * @mnt_userns: user namespace of the mount the inode was found from
2181 * @inode: inode being checked
2183 * Return true if current either has CAP_FOWNER in a namespace with the
2184 * inode owner uid mapped, or owns the file.
2186 * If the inode has been found through an idmapped mount the user namespace of
2187 * the vfsmount must be passed through @mnt_userns. This function will then take
2188 * care to map the inode according to @mnt_userns before checking permissions.
2189 * On non-idmapped mounts or if permission checking is to be performed on the
2190 * raw inode simply passs init_user_ns.
2192 bool inode_owner_or_capable(struct user_namespace *mnt_userns,
2193 const struct inode *inode)
2196 struct user_namespace *ns;
2198 i_uid = i_uid_into_mnt(mnt_userns, inode);
2199 if (uid_eq(current_fsuid(), i_uid))
2202 ns = current_user_ns();
2203 if (kuid_has_mapping(ns, i_uid) && ns_capable(ns, CAP_FOWNER))
2207 EXPORT_SYMBOL(inode_owner_or_capable);
2210 * Direct i/o helper functions
2212 static void __inode_dio_wait(struct inode *inode)
2214 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2215 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2218 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2219 if (atomic_read(&inode->i_dio_count))
2221 } while (atomic_read(&inode->i_dio_count));
2222 finish_wait(wq, &q.wq_entry);
2226 * inode_dio_wait - wait for outstanding DIO requests to finish
2227 * @inode: inode to wait for
2229 * Waits for all pending direct I/O requests to finish so that we can
2230 * proceed with a truncate or equivalent operation.
2232 * Must be called under a lock that serializes taking new references
2233 * to i_dio_count, usually by inode->i_mutex.
2235 void inode_dio_wait(struct inode *inode)
2237 if (atomic_read(&inode->i_dio_count))
2238 __inode_dio_wait(inode);
2240 EXPORT_SYMBOL(inode_dio_wait);
2243 * inode_set_flags - atomically set some inode flags
2245 * Note: the caller should be holding i_mutex, or else be sure that
2246 * they have exclusive access to the inode structure (i.e., while the
2247 * inode is being instantiated). The reason for the cmpxchg() loop
2248 * --- which wouldn't be necessary if all code paths which modify
2249 * i_flags actually followed this rule, is that there is at least one
2250 * code path which doesn't today so we use cmpxchg() out of an abundance
2253 * In the long run, i_mutex is overkill, and we should probably look
2254 * at using the i_lock spinlock to protect i_flags, and then make sure
2255 * it is so documented in include/linux/fs.h and that all code follows
2256 * the locking convention!!
2258 void inode_set_flags(struct inode *inode, unsigned int flags,
2261 WARN_ON_ONCE(flags & ~mask);
2262 set_mask_bits(&inode->i_flags, mask, flags);
2264 EXPORT_SYMBOL(inode_set_flags);
2266 void inode_nohighmem(struct inode *inode)
2268 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2270 EXPORT_SYMBOL(inode_nohighmem);
2273 * timestamp_truncate - Truncate timespec to a granularity
2275 * @inode: inode being updated
2277 * Truncate a timespec to the granularity supported by the fs
2278 * containing the inode. Always rounds down. gran must
2279 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2281 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2283 struct super_block *sb = inode->i_sb;
2284 unsigned int gran = sb->s_time_gran;
2286 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2287 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2290 /* Avoid division in the common cases 1 ns and 1 s. */
2293 else if (gran == NSEC_PER_SEC)
2295 else if (gran > 1 && gran < NSEC_PER_SEC)
2296 t.tv_nsec -= t.tv_nsec % gran;
2298 WARN(1, "invalid file time granularity: %u", gran);
2301 EXPORT_SYMBOL(timestamp_truncate);
2304 * current_time - Return FS time
2307 * Return the current time truncated to the time granularity supported by
2310 * Note that inode and inode->sb cannot be NULL.
2311 * Otherwise, the function warns and returns time without truncation.
2313 struct timespec64 current_time(struct inode *inode)
2315 struct timespec64 now;
2317 ktime_get_coarse_real_ts64(&now);
2319 if (unlikely(!inode->i_sb)) {
2320 WARN(1, "current_time() called with uninitialized super_block in the inode");
2324 return timestamp_truncate(now, inode);
2326 EXPORT_SYMBOL(current_time);