1 // SPDX-License-Identifier: GPL-2.0-only
3 * (C) 1997 Linus Torvalds
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;
184 atomic_set(&mapping->i_mmap_writable, 0);
185 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
186 atomic_set(&mapping->nr_thps, 0);
188 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
189 mapping->private_data = NULL;
190 mapping->writeback_index = 0;
191 init_rwsem(&mapping->invalidate_lock);
192 lockdep_set_class_and_name(&mapping->invalidate_lock,
193 &sb->s_type->invalidate_lock_key,
194 "mapping.invalidate_lock");
195 inode->i_private = NULL;
196 inode->i_mapping = mapping;
197 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
198 #ifdef CONFIG_FS_POSIX_ACL
199 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
202 #ifdef CONFIG_FSNOTIFY
203 inode->i_fsnotify_mask = 0;
205 inode->i_flctx = NULL;
206 this_cpu_inc(nr_inodes);
212 EXPORT_SYMBOL(inode_init_always);
214 void free_inode_nonrcu(struct inode *inode)
216 kmem_cache_free(inode_cachep, inode);
218 EXPORT_SYMBOL(free_inode_nonrcu);
220 static void i_callback(struct rcu_head *head)
222 struct inode *inode = container_of(head, struct inode, i_rcu);
223 if (inode->free_inode)
224 inode->free_inode(inode);
226 free_inode_nonrcu(inode);
229 static struct inode *alloc_inode(struct super_block *sb)
231 const struct super_operations *ops = sb->s_op;
234 if (ops->alloc_inode)
235 inode = ops->alloc_inode(sb);
237 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
242 if (unlikely(inode_init_always(sb, inode))) {
243 if (ops->destroy_inode) {
244 ops->destroy_inode(inode);
245 if (!ops->free_inode)
248 inode->free_inode = ops->free_inode;
249 i_callback(&inode->i_rcu);
256 void __destroy_inode(struct inode *inode)
258 BUG_ON(inode_has_buffers(inode));
259 inode_detach_wb(inode);
260 security_inode_free(inode);
261 fsnotify_inode_delete(inode);
262 locks_free_lock_context(inode);
263 if (!inode->i_nlink) {
264 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
265 atomic_long_dec(&inode->i_sb->s_remove_count);
268 #ifdef CONFIG_FS_POSIX_ACL
269 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
270 posix_acl_release(inode->i_acl);
271 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
272 posix_acl_release(inode->i_default_acl);
274 this_cpu_dec(nr_inodes);
276 EXPORT_SYMBOL(__destroy_inode);
278 static void destroy_inode(struct inode *inode)
280 const struct super_operations *ops = inode->i_sb->s_op;
282 BUG_ON(!list_empty(&inode->i_lru));
283 __destroy_inode(inode);
284 if (ops->destroy_inode) {
285 ops->destroy_inode(inode);
286 if (!ops->free_inode)
289 inode->free_inode = ops->free_inode;
290 call_rcu(&inode->i_rcu, i_callback);
294 * drop_nlink - directly drop an inode's link count
297 * This is a low-level filesystem helper to replace any
298 * direct filesystem manipulation of i_nlink. In cases
299 * where we are attempting to track writes to the
300 * filesystem, a decrement to zero means an imminent
301 * write when the file is truncated and actually unlinked
304 void drop_nlink(struct inode *inode)
306 WARN_ON(inode->i_nlink == 0);
309 atomic_long_inc(&inode->i_sb->s_remove_count);
311 EXPORT_SYMBOL(drop_nlink);
314 * clear_nlink - directly zero an inode's link count
317 * This is a low-level filesystem helper to replace any
318 * direct filesystem manipulation of i_nlink. See
319 * drop_nlink() for why we care about i_nlink hitting zero.
321 void clear_nlink(struct inode *inode)
323 if (inode->i_nlink) {
324 inode->__i_nlink = 0;
325 atomic_long_inc(&inode->i_sb->s_remove_count);
328 EXPORT_SYMBOL(clear_nlink);
331 * set_nlink - directly set an inode's link count
333 * @nlink: new nlink (should be non-zero)
335 * This is a low-level filesystem helper to replace any
336 * direct filesystem manipulation of i_nlink.
338 void set_nlink(struct inode *inode, unsigned int nlink)
343 /* Yes, some filesystems do change nlink from zero to one */
344 if (inode->i_nlink == 0)
345 atomic_long_dec(&inode->i_sb->s_remove_count);
347 inode->__i_nlink = nlink;
350 EXPORT_SYMBOL(set_nlink);
353 * inc_nlink - directly increment an inode's link count
356 * This is a low-level filesystem helper to replace any
357 * direct filesystem manipulation of i_nlink. Currently,
358 * it is only here for parity with dec_nlink().
360 void inc_nlink(struct inode *inode)
362 if (unlikely(inode->i_nlink == 0)) {
363 WARN_ON(!(inode->i_state & I_LINKABLE));
364 atomic_long_dec(&inode->i_sb->s_remove_count);
369 EXPORT_SYMBOL(inc_nlink);
371 static void __address_space_init_once(struct address_space *mapping)
373 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
374 init_rwsem(&mapping->i_mmap_rwsem);
375 INIT_LIST_HEAD(&mapping->private_list);
376 spin_lock_init(&mapping->private_lock);
377 mapping->i_mmap = RB_ROOT_CACHED;
380 void address_space_init_once(struct address_space *mapping)
382 memset(mapping, 0, sizeof(*mapping));
383 __address_space_init_once(mapping);
385 EXPORT_SYMBOL(address_space_init_once);
388 * These are initializations that only need to be done
389 * once, because the fields are idempotent across use
390 * of the inode, so let the slab aware of that.
392 void inode_init_once(struct inode *inode)
394 memset(inode, 0, sizeof(*inode));
395 INIT_HLIST_NODE(&inode->i_hash);
396 INIT_LIST_HEAD(&inode->i_devices);
397 INIT_LIST_HEAD(&inode->i_io_list);
398 INIT_LIST_HEAD(&inode->i_wb_list);
399 INIT_LIST_HEAD(&inode->i_lru);
400 __address_space_init_once(&inode->i_data);
401 i_size_ordered_init(inode);
403 EXPORT_SYMBOL(inode_init_once);
405 static void init_once(void *foo)
407 struct inode *inode = (struct inode *) foo;
409 inode_init_once(inode);
413 * inode->i_lock must be held
415 void __iget(struct inode *inode)
417 atomic_inc(&inode->i_count);
421 * get additional reference to inode; caller must already hold one.
423 void ihold(struct inode *inode)
425 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
427 EXPORT_SYMBOL(ihold);
429 static void __inode_add_lru(struct inode *inode, bool rotate)
431 if (inode->i_state & (I_DIRTY_ALL | I_SYNC | I_FREEING | I_WILL_FREE))
433 if (atomic_read(&inode->i_count))
435 if (!(inode->i_sb->s_flags & SB_ACTIVE))
437 if (!mapping_shrinkable(&inode->i_data))
440 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
441 this_cpu_inc(nr_unused);
443 inode->i_state |= I_REFERENCED;
447 * Add inode to LRU if needed (inode is unused and clean).
449 * Needs inode->i_lock held.
451 void inode_add_lru(struct inode *inode)
453 __inode_add_lru(inode, false);
456 static void inode_lru_list_del(struct inode *inode)
458 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
459 this_cpu_dec(nr_unused);
463 * inode_sb_list_add - add inode to the superblock list of inodes
464 * @inode: inode to add
466 void inode_sb_list_add(struct inode *inode)
468 spin_lock(&inode->i_sb->s_inode_list_lock);
469 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
470 spin_unlock(&inode->i_sb->s_inode_list_lock);
472 EXPORT_SYMBOL_GPL(inode_sb_list_add);
474 static inline void inode_sb_list_del(struct inode *inode)
476 if (!list_empty(&inode->i_sb_list)) {
477 spin_lock(&inode->i_sb->s_inode_list_lock);
478 list_del_init(&inode->i_sb_list);
479 spin_unlock(&inode->i_sb->s_inode_list_lock);
483 static unsigned long hash(struct super_block *sb, unsigned long hashval)
487 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
489 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
490 return tmp & i_hash_mask;
494 * __insert_inode_hash - hash an inode
495 * @inode: unhashed inode
496 * @hashval: unsigned long value used to locate this object in the
499 * Add an inode to the inode hash for this superblock.
501 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
503 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
505 spin_lock(&inode_hash_lock);
506 spin_lock(&inode->i_lock);
507 hlist_add_head_rcu(&inode->i_hash, b);
508 spin_unlock(&inode->i_lock);
509 spin_unlock(&inode_hash_lock);
511 EXPORT_SYMBOL(__insert_inode_hash);
514 * __remove_inode_hash - remove an inode from the hash
515 * @inode: inode to unhash
517 * Remove an inode from the superblock.
519 void __remove_inode_hash(struct inode *inode)
521 spin_lock(&inode_hash_lock);
522 spin_lock(&inode->i_lock);
523 hlist_del_init_rcu(&inode->i_hash);
524 spin_unlock(&inode->i_lock);
525 spin_unlock(&inode_hash_lock);
527 EXPORT_SYMBOL(__remove_inode_hash);
529 void clear_inode(struct inode *inode)
532 * We have to cycle the i_pages lock here because reclaim can be in the
533 * process of removing the last page (in __delete_from_page_cache())
534 * and we must not free the mapping under it.
536 xa_lock_irq(&inode->i_data.i_pages);
537 BUG_ON(inode->i_data.nrpages);
539 * Almost always, mapping_empty(&inode->i_data) here; but there are
540 * two known and long-standing ways in which nodes may get left behind
541 * (when deep radix-tree node allocation failed partway; or when THP
542 * collapse_file() failed). Until those two known cases are cleaned up,
543 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
544 * nor even WARN_ON(!mapping_empty).
546 xa_unlock_irq(&inode->i_data.i_pages);
547 BUG_ON(!list_empty(&inode->i_data.private_list));
548 BUG_ON(!(inode->i_state & I_FREEING));
549 BUG_ON(inode->i_state & I_CLEAR);
550 BUG_ON(!list_empty(&inode->i_wb_list));
551 /* don't need i_lock here, no concurrent mods to i_state */
552 inode->i_state = I_FREEING | I_CLEAR;
554 EXPORT_SYMBOL(clear_inode);
557 * Free the inode passed in, removing it from the lists it is still connected
558 * to. We remove any pages still attached to the inode and wait for any IO that
559 * is still in progress before finally destroying the inode.
561 * An inode must already be marked I_FREEING so that we avoid the inode being
562 * moved back onto lists if we race with other code that manipulates the lists
563 * (e.g. writeback_single_inode). The caller is responsible for setting this.
565 * An inode must already be removed from the LRU list before being evicted from
566 * the cache. This should occur atomically with setting the I_FREEING state
567 * flag, so no inodes here should ever be on the LRU when being evicted.
569 static void evict(struct inode *inode)
571 const struct super_operations *op = inode->i_sb->s_op;
573 BUG_ON(!(inode->i_state & I_FREEING));
574 BUG_ON(!list_empty(&inode->i_lru));
576 if (!list_empty(&inode->i_io_list))
577 inode_io_list_del(inode);
579 inode_sb_list_del(inode);
582 * Wait for flusher thread to be done with the inode so that filesystem
583 * does not start destroying it while writeback is still running. Since
584 * the inode has I_FREEING set, flusher thread won't start new work on
585 * the inode. We just have to wait for running writeback to finish.
587 inode_wait_for_writeback(inode);
589 if (op->evict_inode) {
590 op->evict_inode(inode);
592 truncate_inode_pages_final(&inode->i_data);
595 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
598 remove_inode_hash(inode);
600 spin_lock(&inode->i_lock);
601 wake_up_bit(&inode->i_state, __I_NEW);
602 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
603 spin_unlock(&inode->i_lock);
605 destroy_inode(inode);
609 * dispose_list - dispose of the contents of a local list
610 * @head: the head of the list to free
612 * Dispose-list gets a local list with local inodes in it, so it doesn't
613 * need to worry about list corruption and SMP locks.
615 static void dispose_list(struct list_head *head)
617 while (!list_empty(head)) {
620 inode = list_first_entry(head, struct inode, i_lru);
621 list_del_init(&inode->i_lru);
629 * evict_inodes - evict all evictable inodes for a superblock
630 * @sb: superblock to operate on
632 * Make sure that no inodes with zero refcount are retained. This is
633 * called by superblock shutdown after having SB_ACTIVE flag removed,
634 * so any inode reaching zero refcount during or after that call will
635 * be immediately evicted.
637 void evict_inodes(struct super_block *sb)
639 struct inode *inode, *next;
643 spin_lock(&sb->s_inode_list_lock);
644 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
645 if (atomic_read(&inode->i_count))
648 spin_lock(&inode->i_lock);
649 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
650 spin_unlock(&inode->i_lock);
654 inode->i_state |= I_FREEING;
655 inode_lru_list_del(inode);
656 spin_unlock(&inode->i_lock);
657 list_add(&inode->i_lru, &dispose);
660 * We can have a ton of inodes to evict at unmount time given
661 * enough memory, check to see if we need to go to sleep for a
662 * bit so we don't livelock.
664 if (need_resched()) {
665 spin_unlock(&sb->s_inode_list_lock);
667 dispose_list(&dispose);
671 spin_unlock(&sb->s_inode_list_lock);
673 dispose_list(&dispose);
675 EXPORT_SYMBOL_GPL(evict_inodes);
678 * invalidate_inodes - attempt to free all inodes on a superblock
679 * @sb: superblock to operate on
680 * @kill_dirty: flag to guide handling of dirty inodes
682 * Attempts to free all inodes for a given superblock. If there were any
683 * busy inodes return a non-zero value, else zero.
684 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
687 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
690 struct inode *inode, *next;
694 spin_lock(&sb->s_inode_list_lock);
695 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
696 spin_lock(&inode->i_lock);
697 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
698 spin_unlock(&inode->i_lock);
701 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
702 spin_unlock(&inode->i_lock);
706 if (atomic_read(&inode->i_count)) {
707 spin_unlock(&inode->i_lock);
712 inode->i_state |= I_FREEING;
713 inode_lru_list_del(inode);
714 spin_unlock(&inode->i_lock);
715 list_add(&inode->i_lru, &dispose);
716 if (need_resched()) {
717 spin_unlock(&sb->s_inode_list_lock);
719 dispose_list(&dispose);
723 spin_unlock(&sb->s_inode_list_lock);
725 dispose_list(&dispose);
731 * Isolate the inode from the LRU in preparation for freeing it.
733 * If the inode has the I_REFERENCED flag set, then it means that it has been
734 * used recently - the flag is set in iput_final(). When we encounter such an
735 * inode, clear the flag and move it to the back of the LRU so it gets another
736 * pass through the LRU before it gets reclaimed. This is necessary because of
737 * the fact we are doing lazy LRU updates to minimise lock contention so the
738 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
739 * with this flag set because they are the inodes that are out of order.
741 static enum lru_status inode_lru_isolate(struct list_head *item,
742 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
744 struct list_head *freeable = arg;
745 struct inode *inode = container_of(item, struct inode, i_lru);
748 * We are inverting the lru lock/inode->i_lock here, so use a
749 * trylock. If we fail to get the lock, just skip it.
751 if (!spin_trylock(&inode->i_lock))
755 * Inodes can get referenced, redirtied, or repopulated while
756 * they're already on the LRU, and this can make them
757 * unreclaimable for a while. Remove them lazily here; iput,
758 * sync, or the last page cache deletion will requeue them.
760 if (atomic_read(&inode->i_count) ||
761 (inode->i_state & ~I_REFERENCED) ||
762 !mapping_shrinkable(&inode->i_data)) {
763 list_lru_isolate(lru, &inode->i_lru);
764 spin_unlock(&inode->i_lock);
765 this_cpu_dec(nr_unused);
769 /* Recently referenced inodes get one more pass */
770 if (inode->i_state & I_REFERENCED) {
771 inode->i_state &= ~I_REFERENCED;
772 spin_unlock(&inode->i_lock);
777 * On highmem systems, mapping_shrinkable() permits dropping
778 * page cache in order to free up struct inodes: lowmem might
779 * be under pressure before the cache inside the highmem zone.
781 if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
783 spin_unlock(&inode->i_lock);
784 spin_unlock(lru_lock);
785 if (remove_inode_buffers(inode)) {
787 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
788 if (current_is_kswapd())
789 __count_vm_events(KSWAPD_INODESTEAL, reap);
791 __count_vm_events(PGINODESTEAL, reap);
792 if (current->reclaim_state)
793 current->reclaim_state->reclaimed_slab += reap;
800 WARN_ON(inode->i_state & I_NEW);
801 inode->i_state |= I_FREEING;
802 list_lru_isolate_move(lru, &inode->i_lru, freeable);
803 spin_unlock(&inode->i_lock);
805 this_cpu_dec(nr_unused);
810 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
811 * This is called from the superblock shrinker function with a number of inodes
812 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
813 * then are freed outside inode_lock by dispose_list().
815 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
820 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
821 inode_lru_isolate, &freeable);
822 dispose_list(&freeable);
826 static void __wait_on_freeing_inode(struct inode *inode);
828 * Called with the inode lock held.
830 static struct inode *find_inode(struct super_block *sb,
831 struct hlist_head *head,
832 int (*test)(struct inode *, void *),
835 struct inode *inode = NULL;
838 hlist_for_each_entry(inode, head, i_hash) {
839 if (inode->i_sb != sb)
841 if (!test(inode, data))
843 spin_lock(&inode->i_lock);
844 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
845 __wait_on_freeing_inode(inode);
848 if (unlikely(inode->i_state & I_CREATING)) {
849 spin_unlock(&inode->i_lock);
850 return ERR_PTR(-ESTALE);
853 spin_unlock(&inode->i_lock);
860 * find_inode_fast is the fast path version of find_inode, see the comment at
861 * iget_locked for details.
863 static struct inode *find_inode_fast(struct super_block *sb,
864 struct hlist_head *head, unsigned long ino)
866 struct inode *inode = NULL;
869 hlist_for_each_entry(inode, head, i_hash) {
870 if (inode->i_ino != ino)
872 if (inode->i_sb != sb)
874 spin_lock(&inode->i_lock);
875 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
876 __wait_on_freeing_inode(inode);
879 if (unlikely(inode->i_state & I_CREATING)) {
880 spin_unlock(&inode->i_lock);
881 return ERR_PTR(-ESTALE);
884 spin_unlock(&inode->i_lock);
891 * Each cpu owns a range of LAST_INO_BATCH numbers.
892 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
893 * to renew the exhausted range.
895 * This does not significantly increase overflow rate because every CPU can
896 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
897 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
898 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
899 * overflow rate by 2x, which does not seem too significant.
901 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
902 * error if st_ino won't fit in target struct field. Use 32bit counter
903 * here to attempt to avoid that.
905 #define LAST_INO_BATCH 1024
906 static DEFINE_PER_CPU(unsigned int, last_ino);
908 unsigned int get_next_ino(void)
910 unsigned int *p = &get_cpu_var(last_ino);
911 unsigned int res = *p;
914 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
915 static atomic_t shared_last_ino;
916 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
918 res = next - LAST_INO_BATCH;
923 /* get_next_ino should not provide a 0 inode number */
927 put_cpu_var(last_ino);
930 EXPORT_SYMBOL(get_next_ino);
933 * new_inode_pseudo - obtain an inode
936 * Allocates a new inode for given superblock.
937 * Inode wont be chained in superblock s_inodes list
939 * - fs can't be unmount
940 * - quotas, fsnotify, writeback can't work
942 struct inode *new_inode_pseudo(struct super_block *sb)
944 struct inode *inode = alloc_inode(sb);
947 spin_lock(&inode->i_lock);
949 spin_unlock(&inode->i_lock);
950 INIT_LIST_HEAD(&inode->i_sb_list);
956 * new_inode - obtain an inode
959 * Allocates a new inode for given superblock. The default gfp_mask
960 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
961 * If HIGHMEM pages are unsuitable or it is known that pages allocated
962 * for the page cache are not reclaimable or migratable,
963 * mapping_set_gfp_mask() must be called with suitable flags on the
964 * newly created inode's mapping
967 struct inode *new_inode(struct super_block *sb)
971 spin_lock_prefetch(&sb->s_inode_list_lock);
973 inode = new_inode_pseudo(sb);
975 inode_sb_list_add(inode);
978 EXPORT_SYMBOL(new_inode);
980 #ifdef CONFIG_DEBUG_LOCK_ALLOC
981 void lockdep_annotate_inode_mutex_key(struct inode *inode)
983 if (S_ISDIR(inode->i_mode)) {
984 struct file_system_type *type = inode->i_sb->s_type;
986 /* Set new key only if filesystem hasn't already changed it */
987 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
989 * ensure nobody is actually holding i_mutex
991 // mutex_destroy(&inode->i_mutex);
992 init_rwsem(&inode->i_rwsem);
993 lockdep_set_class(&inode->i_rwsem,
994 &type->i_mutex_dir_key);
998 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1002 * unlock_new_inode - clear the I_NEW state and wake up any waiters
1003 * @inode: new inode to unlock
1005 * Called when the inode is fully initialised to clear the new state of the
1006 * inode and wake up anyone waiting for the inode to finish initialisation.
1008 void unlock_new_inode(struct inode *inode)
1010 lockdep_annotate_inode_mutex_key(inode);
1011 spin_lock(&inode->i_lock);
1012 WARN_ON(!(inode->i_state & I_NEW));
1013 inode->i_state &= ~I_NEW & ~I_CREATING;
1015 wake_up_bit(&inode->i_state, __I_NEW);
1016 spin_unlock(&inode->i_lock);
1018 EXPORT_SYMBOL(unlock_new_inode);
1020 void discard_new_inode(struct inode *inode)
1022 lockdep_annotate_inode_mutex_key(inode);
1023 spin_lock(&inode->i_lock);
1024 WARN_ON(!(inode->i_state & I_NEW));
1025 inode->i_state &= ~I_NEW;
1027 wake_up_bit(&inode->i_state, __I_NEW);
1028 spin_unlock(&inode->i_lock);
1031 EXPORT_SYMBOL(discard_new_inode);
1034 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1036 * Lock any non-NULL argument that is not a directory.
1037 * Zero, one or two objects may be locked by this function.
1039 * @inode1: first inode to lock
1040 * @inode2: second inode to lock
1042 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1044 if (inode1 > inode2)
1045 swap(inode1, inode2);
1047 if (inode1 && !S_ISDIR(inode1->i_mode))
1049 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1050 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1052 EXPORT_SYMBOL(lock_two_nondirectories);
1055 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1056 * @inode1: first inode to unlock
1057 * @inode2: second inode to unlock
1059 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1061 if (inode1 && !S_ISDIR(inode1->i_mode))
1062 inode_unlock(inode1);
1063 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1064 inode_unlock(inode2);
1066 EXPORT_SYMBOL(unlock_two_nondirectories);
1069 * inode_insert5 - obtain an inode from a mounted file system
1070 * @inode: pre-allocated inode to use for insert to cache
1071 * @hashval: hash value (usually inode number) to get
1072 * @test: callback used for comparisons between inodes
1073 * @set: callback used to initialize a new struct inode
1074 * @data: opaque data pointer to pass to @test and @set
1076 * Search for the inode specified by @hashval and @data in the inode cache,
1077 * and if present it is return it with an increased reference count. This is
1078 * a variant of iget5_locked() for callers that don't want to fail on memory
1079 * allocation of inode.
1081 * If the inode is not in cache, insert the pre-allocated inode to cache and
1082 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1083 * to fill it in before unlocking it via unlock_new_inode().
1085 * Note both @test and @set are called with the inode_hash_lock held, so can't
1088 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1089 int (*test)(struct inode *, void *),
1090 int (*set)(struct inode *, void *), void *data)
1092 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1094 bool creating = inode->i_state & I_CREATING;
1097 spin_lock(&inode_hash_lock);
1098 old = find_inode(inode->i_sb, head, test, data);
1099 if (unlikely(old)) {
1101 * Uhhuh, somebody else created the same inode under us.
1102 * Use the old inode instead of the preallocated one.
1104 spin_unlock(&inode_hash_lock);
1108 if (unlikely(inode_unhashed(old))) {
1115 if (set && unlikely(set(inode, data))) {
1121 * Return the locked inode with I_NEW set, the
1122 * caller is responsible for filling in the contents
1124 spin_lock(&inode->i_lock);
1125 inode->i_state |= I_NEW;
1126 hlist_add_head_rcu(&inode->i_hash, head);
1127 spin_unlock(&inode->i_lock);
1129 inode_sb_list_add(inode);
1131 spin_unlock(&inode_hash_lock);
1135 EXPORT_SYMBOL(inode_insert5);
1138 * iget5_locked - obtain an inode from a mounted file system
1139 * @sb: super block of file system
1140 * @hashval: hash value (usually inode number) to get
1141 * @test: callback used for comparisons between inodes
1142 * @set: callback used to initialize a new struct inode
1143 * @data: opaque data pointer to pass to @test and @set
1145 * Search for the inode specified by @hashval and @data in the inode cache,
1146 * and if present it is return it with an increased reference count. This is
1147 * a generalized version of iget_locked() for file systems where the inode
1148 * number is not sufficient for unique identification of an inode.
1150 * If the inode is not in cache, allocate a new inode and return it locked,
1151 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1152 * before unlocking it via unlock_new_inode().
1154 * Note both @test and @set are called with the inode_hash_lock held, so can't
1157 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1158 int (*test)(struct inode *, void *),
1159 int (*set)(struct inode *, void *), void *data)
1161 struct inode *inode = ilookup5(sb, hashval, test, data);
1164 struct inode *new = alloc_inode(sb);
1168 inode = inode_insert5(new, hashval, test, set, data);
1169 if (unlikely(inode != new))
1175 EXPORT_SYMBOL(iget5_locked);
1178 * iget_locked - obtain an inode from a mounted file system
1179 * @sb: super block of file system
1180 * @ino: inode number to get
1182 * Search for the inode specified by @ino in the inode cache and if present
1183 * return it with an increased reference count. This is for file systems
1184 * where the inode number is sufficient for unique identification of an inode.
1186 * If the inode is not in cache, allocate a new inode and return it locked,
1187 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1188 * before unlocking it via unlock_new_inode().
1190 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1192 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1193 struct inode *inode;
1195 spin_lock(&inode_hash_lock);
1196 inode = find_inode_fast(sb, head, ino);
1197 spin_unlock(&inode_hash_lock);
1201 wait_on_inode(inode);
1202 if (unlikely(inode_unhashed(inode))) {
1209 inode = alloc_inode(sb);
1213 spin_lock(&inode_hash_lock);
1214 /* We released the lock, so.. */
1215 old = find_inode_fast(sb, head, ino);
1218 spin_lock(&inode->i_lock);
1219 inode->i_state = I_NEW;
1220 hlist_add_head_rcu(&inode->i_hash, head);
1221 spin_unlock(&inode->i_lock);
1222 inode_sb_list_add(inode);
1223 spin_unlock(&inode_hash_lock);
1225 /* Return the locked inode with I_NEW set, the
1226 * caller is responsible for filling in the contents
1232 * Uhhuh, somebody else created the same inode under
1233 * us. Use the old inode instead of the one we just
1236 spin_unlock(&inode_hash_lock);
1237 destroy_inode(inode);
1241 wait_on_inode(inode);
1242 if (unlikely(inode_unhashed(inode))) {
1249 EXPORT_SYMBOL(iget_locked);
1252 * search the inode cache for a matching inode number.
1253 * If we find one, then the inode number we are trying to
1254 * allocate is not unique and so we should not use it.
1256 * Returns 1 if the inode number is unique, 0 if it is not.
1258 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1260 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1261 struct inode *inode;
1263 hlist_for_each_entry_rcu(inode, b, i_hash) {
1264 if (inode->i_ino == ino && inode->i_sb == sb)
1271 * iunique - get a unique inode number
1273 * @max_reserved: highest reserved inode number
1275 * Obtain an inode number that is unique on the system for a given
1276 * superblock. This is used by file systems that have no natural
1277 * permanent inode numbering system. An inode number is returned that
1278 * is higher than the reserved limit but unique.
1281 * With a large number of inodes live on the file system this function
1282 * currently becomes quite slow.
1284 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1287 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1288 * error if st_ino won't fit in target struct field. Use 32bit counter
1289 * here to attempt to avoid that.
1291 static DEFINE_SPINLOCK(iunique_lock);
1292 static unsigned int counter;
1296 spin_lock(&iunique_lock);
1298 if (counter <= max_reserved)
1299 counter = max_reserved + 1;
1301 } while (!test_inode_iunique(sb, res));
1302 spin_unlock(&iunique_lock);
1307 EXPORT_SYMBOL(iunique);
1309 struct inode *igrab(struct inode *inode)
1311 spin_lock(&inode->i_lock);
1312 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1314 spin_unlock(&inode->i_lock);
1316 spin_unlock(&inode->i_lock);
1318 * Handle the case where s_op->clear_inode is not been
1319 * called yet, and somebody is calling igrab
1320 * while the inode is getting freed.
1326 EXPORT_SYMBOL(igrab);
1329 * ilookup5_nowait - search for an inode in the inode cache
1330 * @sb: super block of file system to search
1331 * @hashval: hash value (usually inode number) to search for
1332 * @test: callback used for comparisons between inodes
1333 * @data: opaque data pointer to pass to @test
1335 * Search for the inode specified by @hashval and @data in the inode cache.
1336 * If the inode is in the cache, the inode is returned with an incremented
1339 * Note: I_NEW is not waited upon so you have to be very careful what you do
1340 * with the returned inode. You probably should be using ilookup5() instead.
1342 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1344 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1345 int (*test)(struct inode *, void *), void *data)
1347 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1348 struct inode *inode;
1350 spin_lock(&inode_hash_lock);
1351 inode = find_inode(sb, head, test, data);
1352 spin_unlock(&inode_hash_lock);
1354 return IS_ERR(inode) ? NULL : inode;
1356 EXPORT_SYMBOL(ilookup5_nowait);
1359 * ilookup5 - search for an inode in the inode cache
1360 * @sb: super block of file system to search
1361 * @hashval: hash value (usually inode number) to search for
1362 * @test: callback used for comparisons between inodes
1363 * @data: opaque data pointer to pass to @test
1365 * Search for the inode specified by @hashval and @data in the inode cache,
1366 * and if the inode is in the cache, return the inode with an incremented
1367 * reference count. Waits on I_NEW before returning the inode.
1368 * returned with an incremented reference count.
1370 * This is a generalized version of ilookup() for file systems where the
1371 * inode number is not sufficient for unique identification of an inode.
1373 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1375 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1376 int (*test)(struct inode *, void *), void *data)
1378 struct inode *inode;
1380 inode = ilookup5_nowait(sb, hashval, test, data);
1382 wait_on_inode(inode);
1383 if (unlikely(inode_unhashed(inode))) {
1390 EXPORT_SYMBOL(ilookup5);
1393 * ilookup - search for an inode in the inode cache
1394 * @sb: super block of file system to search
1395 * @ino: inode number to search for
1397 * Search for the inode @ino in the inode cache, and if the inode is in the
1398 * cache, the inode is returned with an incremented reference count.
1400 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1402 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1403 struct inode *inode;
1405 spin_lock(&inode_hash_lock);
1406 inode = find_inode_fast(sb, head, ino);
1407 spin_unlock(&inode_hash_lock);
1412 wait_on_inode(inode);
1413 if (unlikely(inode_unhashed(inode))) {
1420 EXPORT_SYMBOL(ilookup);
1423 * find_inode_nowait - find an inode in the inode cache
1424 * @sb: super block of file system to search
1425 * @hashval: hash value (usually inode number) to search for
1426 * @match: callback used for comparisons between inodes
1427 * @data: opaque data pointer to pass to @match
1429 * Search for the inode specified by @hashval and @data in the inode
1430 * cache, where the helper function @match will return 0 if the inode
1431 * does not match, 1 if the inode does match, and -1 if the search
1432 * should be stopped. The @match function must be responsible for
1433 * taking the i_lock spin_lock and checking i_state for an inode being
1434 * freed or being initialized, and incrementing the reference count
1435 * before returning 1. It also must not sleep, since it is called with
1436 * the inode_hash_lock spinlock held.
1438 * This is a even more generalized version of ilookup5() when the
1439 * function must never block --- find_inode() can block in
1440 * __wait_on_freeing_inode() --- or when the caller can not increment
1441 * the reference count because the resulting iput() might cause an
1442 * inode eviction. The tradeoff is that the @match funtion must be
1443 * very carefully implemented.
1445 struct inode *find_inode_nowait(struct super_block *sb,
1446 unsigned long hashval,
1447 int (*match)(struct inode *, unsigned long,
1451 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1452 struct inode *inode, *ret_inode = NULL;
1455 spin_lock(&inode_hash_lock);
1456 hlist_for_each_entry(inode, head, i_hash) {
1457 if (inode->i_sb != sb)
1459 mval = match(inode, hashval, data);
1467 spin_unlock(&inode_hash_lock);
1470 EXPORT_SYMBOL(find_inode_nowait);
1473 * find_inode_rcu - find an inode in the inode cache
1474 * @sb: Super block of file system to search
1475 * @hashval: Key to hash
1476 * @test: Function to test match on an inode
1477 * @data: Data for test function
1479 * Search for the inode specified by @hashval and @data in the inode cache,
1480 * where the helper function @test will return 0 if the inode does not match
1481 * and 1 if it does. The @test function must be responsible for taking the
1482 * i_lock spin_lock and checking i_state for an inode being freed or being
1485 * If successful, this will return the inode for which the @test function
1486 * returned 1 and NULL otherwise.
1488 * The @test function is not permitted to take a ref on any inode presented.
1489 * It is also not permitted to sleep.
1491 * The caller must hold the RCU read lock.
1493 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1494 int (*test)(struct inode *, void *), void *data)
1496 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1497 struct inode *inode;
1499 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1500 "suspicious find_inode_rcu() usage");
1502 hlist_for_each_entry_rcu(inode, head, i_hash) {
1503 if (inode->i_sb == sb &&
1504 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1510 EXPORT_SYMBOL(find_inode_rcu);
1513 * find_inode_by_ino_rcu - Find an inode in the inode cache
1514 * @sb: Super block of file system to search
1515 * @ino: The inode number to match
1517 * Search for the inode specified by @hashval and @data in the inode cache,
1518 * where the helper function @test will return 0 if the inode does not match
1519 * and 1 if it does. The @test function must be responsible for taking the
1520 * i_lock spin_lock and checking i_state for an inode being freed or being
1523 * If successful, this will return the inode for which the @test function
1524 * returned 1 and NULL otherwise.
1526 * The @test function is not permitted to take a ref on any inode presented.
1527 * It is also not permitted to sleep.
1529 * The caller must hold the RCU read lock.
1531 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1534 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1535 struct inode *inode;
1537 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1538 "suspicious find_inode_by_ino_rcu() usage");
1540 hlist_for_each_entry_rcu(inode, head, i_hash) {
1541 if (inode->i_ino == ino &&
1542 inode->i_sb == sb &&
1543 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1548 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1550 int insert_inode_locked(struct inode *inode)
1552 struct super_block *sb = inode->i_sb;
1553 ino_t ino = inode->i_ino;
1554 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1557 struct inode *old = NULL;
1558 spin_lock(&inode_hash_lock);
1559 hlist_for_each_entry(old, head, i_hash) {
1560 if (old->i_ino != ino)
1562 if (old->i_sb != sb)
1564 spin_lock(&old->i_lock);
1565 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1566 spin_unlock(&old->i_lock);
1572 spin_lock(&inode->i_lock);
1573 inode->i_state |= I_NEW | I_CREATING;
1574 hlist_add_head_rcu(&inode->i_hash, head);
1575 spin_unlock(&inode->i_lock);
1576 spin_unlock(&inode_hash_lock);
1579 if (unlikely(old->i_state & I_CREATING)) {
1580 spin_unlock(&old->i_lock);
1581 spin_unlock(&inode_hash_lock);
1585 spin_unlock(&old->i_lock);
1586 spin_unlock(&inode_hash_lock);
1588 if (unlikely(!inode_unhashed(old))) {
1595 EXPORT_SYMBOL(insert_inode_locked);
1597 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1598 int (*test)(struct inode *, void *), void *data)
1602 inode->i_state |= I_CREATING;
1603 old = inode_insert5(inode, hashval, test, NULL, data);
1611 EXPORT_SYMBOL(insert_inode_locked4);
1614 int generic_delete_inode(struct inode *inode)
1618 EXPORT_SYMBOL(generic_delete_inode);
1621 * Called when we're dropping the last reference
1624 * Call the FS "drop_inode()" function, defaulting to
1625 * the legacy UNIX filesystem behaviour. If it tells
1626 * us to evict inode, do so. Otherwise, retain inode
1627 * in cache if fs is alive, sync and evict if fs is
1630 static void iput_final(struct inode *inode)
1632 struct super_block *sb = inode->i_sb;
1633 const struct super_operations *op = inode->i_sb->s_op;
1634 unsigned long state;
1637 WARN_ON(inode->i_state & I_NEW);
1640 drop = op->drop_inode(inode);
1642 drop = generic_drop_inode(inode);
1645 !(inode->i_state & I_DONTCACHE) &&
1646 (sb->s_flags & SB_ACTIVE)) {
1647 __inode_add_lru(inode, true);
1648 spin_unlock(&inode->i_lock);
1652 state = inode->i_state;
1654 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1655 spin_unlock(&inode->i_lock);
1657 write_inode_now(inode, 1);
1659 spin_lock(&inode->i_lock);
1660 state = inode->i_state;
1661 WARN_ON(state & I_NEW);
1662 state &= ~I_WILL_FREE;
1665 WRITE_ONCE(inode->i_state, state | I_FREEING);
1666 if (!list_empty(&inode->i_lru))
1667 inode_lru_list_del(inode);
1668 spin_unlock(&inode->i_lock);
1674 * iput - put an inode
1675 * @inode: inode to put
1677 * Puts an inode, dropping its usage count. If the inode use count hits
1678 * zero, the inode is then freed and may also be destroyed.
1680 * Consequently, iput() can sleep.
1682 void iput(struct inode *inode)
1686 BUG_ON(inode->i_state & I_CLEAR);
1688 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1689 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1690 atomic_inc(&inode->i_count);
1691 spin_unlock(&inode->i_lock);
1692 trace_writeback_lazytime_iput(inode);
1693 mark_inode_dirty_sync(inode);
1699 EXPORT_SYMBOL(iput);
1703 * bmap - find a block number in a file
1704 * @inode: inode owning the block number being requested
1705 * @block: pointer containing the block to find
1707 * Replaces the value in ``*block`` with the block number on the device holding
1708 * corresponding to the requested block number in the file.
1709 * That is, asked for block 4 of inode 1 the function will replace the
1710 * 4 in ``*block``, with disk block relative to the disk start that holds that
1711 * block of the file.
1713 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1714 * hole, returns 0 and ``*block`` is also set to 0.
1716 int bmap(struct inode *inode, sector_t *block)
1718 if (!inode->i_mapping->a_ops->bmap)
1721 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1724 EXPORT_SYMBOL(bmap);
1728 * With relative atime, only update atime if the previous atime is
1729 * earlier than either the ctime or mtime or if at least a day has
1730 * passed since the last atime update.
1732 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1733 struct timespec64 now)
1736 if (!(mnt->mnt_flags & MNT_RELATIME))
1739 * Is mtime younger than atime? If yes, update atime:
1741 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1744 * Is ctime younger than atime? If yes, update atime:
1746 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1750 * Is the previous atime value older than a day? If yes,
1753 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1756 * Good, we can skip the atime update:
1761 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1763 int dirty_flags = 0;
1765 if (flags & (S_ATIME | S_CTIME | S_MTIME)) {
1766 if (flags & S_ATIME)
1767 inode->i_atime = *time;
1768 if (flags & S_CTIME)
1769 inode->i_ctime = *time;
1770 if (flags & S_MTIME)
1771 inode->i_mtime = *time;
1773 if (inode->i_sb->s_flags & SB_LAZYTIME)
1774 dirty_flags |= I_DIRTY_TIME;
1776 dirty_flags |= I_DIRTY_SYNC;
1779 if ((flags & S_VERSION) && inode_maybe_inc_iversion(inode, false))
1780 dirty_flags |= I_DIRTY_SYNC;
1782 __mark_inode_dirty(inode, dirty_flags);
1785 EXPORT_SYMBOL(generic_update_time);
1788 * This does the actual work of updating an inodes time or version. Must have
1789 * had called mnt_want_write() before calling this.
1791 int inode_update_time(struct inode *inode, struct timespec64 *time, int flags)
1793 if (inode->i_op->update_time)
1794 return inode->i_op->update_time(inode, time, flags);
1795 return generic_update_time(inode, time, flags);
1797 EXPORT_SYMBOL(inode_update_time);
1800 * atime_needs_update - update the access time
1801 * @path: the &struct path to update
1802 * @inode: inode to update
1804 * Update the accessed time on an inode and mark it for writeback.
1805 * This function automatically handles read only file systems and media,
1806 * as well as the "noatime" flag and inode specific "noatime" markers.
1808 bool atime_needs_update(const struct path *path, struct inode *inode)
1810 struct vfsmount *mnt = path->mnt;
1811 struct timespec64 now;
1813 if (inode->i_flags & S_NOATIME)
1816 /* Atime updates will likely cause i_uid and i_gid to be written
1817 * back improprely if their true value is unknown to the vfs.
1819 if (HAS_UNMAPPED_ID(mnt_user_ns(mnt), inode))
1822 if (IS_NOATIME(inode))
1824 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1827 if (mnt->mnt_flags & MNT_NOATIME)
1829 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1832 now = current_time(inode);
1834 if (!relatime_need_update(mnt, inode, now))
1837 if (timespec64_equal(&inode->i_atime, &now))
1843 void touch_atime(const struct path *path)
1845 struct vfsmount *mnt = path->mnt;
1846 struct inode *inode = d_inode(path->dentry);
1847 struct timespec64 now;
1849 if (!atime_needs_update(path, inode))
1852 if (!sb_start_write_trylock(inode->i_sb))
1855 if (__mnt_want_write(mnt) != 0)
1858 * File systems can error out when updating inodes if they need to
1859 * allocate new space to modify an inode (such is the case for
1860 * Btrfs), but since we touch atime while walking down the path we
1861 * really don't care if we failed to update the atime of the file,
1862 * so just ignore the return value.
1863 * We may also fail on filesystems that have the ability to make parts
1864 * of the fs read only, e.g. subvolumes in Btrfs.
1866 now = current_time(inode);
1867 inode_update_time(inode, &now, S_ATIME);
1868 __mnt_drop_write(mnt);
1870 sb_end_write(inode->i_sb);
1872 EXPORT_SYMBOL(touch_atime);
1875 * The logic we want is
1877 * if suid or (sgid and xgrp)
1880 int should_remove_suid(struct dentry *dentry)
1882 umode_t mode = d_inode(dentry)->i_mode;
1885 /* suid always must be killed */
1886 if (unlikely(mode & S_ISUID))
1887 kill = ATTR_KILL_SUID;
1890 * sgid without any exec bits is just a mandatory locking mark; leave
1891 * it alone. If some exec bits are set, it's a real sgid; kill it.
1893 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1894 kill |= ATTR_KILL_SGID;
1896 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1901 EXPORT_SYMBOL(should_remove_suid);
1904 * Return mask of changes for notify_change() that need to be done as a
1905 * response to write or truncate. Return 0 if nothing has to be changed.
1906 * Negative value on error (change should be denied).
1908 int dentry_needs_remove_privs(struct dentry *dentry)
1910 struct inode *inode = d_inode(dentry);
1914 if (IS_NOSEC(inode))
1917 mask = should_remove_suid(dentry);
1918 ret = security_inode_need_killpriv(dentry);
1922 mask |= ATTR_KILL_PRIV;
1926 static int __remove_privs(struct user_namespace *mnt_userns,
1927 struct dentry *dentry, int kill)
1929 struct iattr newattrs;
1931 newattrs.ia_valid = ATTR_FORCE | kill;
1933 * Note we call this on write, so notify_change will not
1934 * encounter any conflicting delegations:
1936 return notify_change(mnt_userns, dentry, &newattrs, NULL);
1940 * Remove special file priviledges (suid, capabilities) when file is written
1943 int file_remove_privs(struct file *file)
1945 struct dentry *dentry = file_dentry(file);
1946 struct inode *inode = file_inode(file);
1951 * Fast path for nothing security related.
1952 * As well for non-regular files, e.g. blkdev inodes.
1953 * For example, blkdev_write_iter() might get here
1954 * trying to remove privs which it is not allowed to.
1956 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1959 kill = dentry_needs_remove_privs(dentry);
1963 error = __remove_privs(file_mnt_user_ns(file), dentry, kill);
1965 inode_has_no_xattr(inode);
1969 EXPORT_SYMBOL(file_remove_privs);
1972 * file_update_time - update mtime and ctime time
1973 * @file: file accessed
1975 * Update the mtime and ctime members of an inode and mark the inode
1976 * for writeback. Note that this function is meant exclusively for
1977 * usage in the file write path of filesystems, and filesystems may
1978 * choose to explicitly ignore update via this function with the
1979 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1980 * timestamps are handled by the server. This can return an error for
1981 * file systems who need to allocate space in order to update an inode.
1984 int file_update_time(struct file *file)
1986 struct inode *inode = file_inode(file);
1987 struct timespec64 now;
1991 /* First try to exhaust all avenues to not sync */
1992 if (IS_NOCMTIME(inode))
1995 now = current_time(inode);
1996 if (!timespec64_equal(&inode->i_mtime, &now))
1999 if (!timespec64_equal(&inode->i_ctime, &now))
2002 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2003 sync_it |= S_VERSION;
2008 /* Finally allowed to write? Takes lock. */
2009 if (__mnt_want_write_file(file))
2012 ret = inode_update_time(inode, &now, sync_it);
2013 __mnt_drop_write_file(file);
2017 EXPORT_SYMBOL(file_update_time);
2019 /* Caller must hold the file's inode lock */
2020 int file_modified(struct file *file)
2025 * Clear the security bits if the process is not being run by root.
2026 * This keeps people from modifying setuid and setgid binaries.
2028 err = file_remove_privs(file);
2032 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2035 return file_update_time(file);
2037 EXPORT_SYMBOL(file_modified);
2039 int inode_needs_sync(struct inode *inode)
2043 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2047 EXPORT_SYMBOL(inode_needs_sync);
2050 * If we try to find an inode in the inode hash while it is being
2051 * deleted, we have to wait until the filesystem completes its
2052 * deletion before reporting that it isn't found. This function waits
2053 * until the deletion _might_ have completed. Callers are responsible
2054 * to recheck inode state.
2056 * It doesn't matter if I_NEW is not set initially, a call to
2057 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2060 static void __wait_on_freeing_inode(struct inode *inode)
2062 wait_queue_head_t *wq;
2063 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2064 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2065 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2066 spin_unlock(&inode->i_lock);
2067 spin_unlock(&inode_hash_lock);
2069 finish_wait(wq, &wait.wq_entry);
2070 spin_lock(&inode_hash_lock);
2073 static __initdata unsigned long ihash_entries;
2074 static int __init set_ihash_entries(char *str)
2078 ihash_entries = simple_strtoul(str, &str, 0);
2081 __setup("ihash_entries=", set_ihash_entries);
2084 * Initialize the waitqueues and inode hash table.
2086 void __init inode_init_early(void)
2088 /* If hashes are distributed across NUMA nodes, defer
2089 * hash allocation until vmalloc space is available.
2095 alloc_large_system_hash("Inode-cache",
2096 sizeof(struct hlist_head),
2099 HASH_EARLY | HASH_ZERO,
2106 void __init inode_init(void)
2108 /* inode slab cache */
2109 inode_cachep = kmem_cache_create("inode_cache",
2110 sizeof(struct inode),
2112 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2113 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2116 /* Hash may have been set up in inode_init_early */
2121 alloc_large_system_hash("Inode-cache",
2122 sizeof(struct hlist_head),
2132 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2134 inode->i_mode = mode;
2135 if (S_ISCHR(mode)) {
2136 inode->i_fop = &def_chr_fops;
2137 inode->i_rdev = rdev;
2138 } else if (S_ISBLK(mode)) {
2139 inode->i_fop = &def_blk_fops;
2140 inode->i_rdev = rdev;
2141 } else if (S_ISFIFO(mode))
2142 inode->i_fop = &pipefifo_fops;
2143 else if (S_ISSOCK(mode))
2144 ; /* leave it no_open_fops */
2146 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2147 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2150 EXPORT_SYMBOL(init_special_inode);
2153 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2154 * @mnt_userns: User namespace of the mount the inode was created from
2156 * @dir: Directory inode
2157 * @mode: mode of the new inode
2159 * If the inode has been created through an idmapped mount the user namespace of
2160 * the vfsmount must be passed through @mnt_userns. This function will then take
2161 * care to map the inode according to @mnt_userns before checking permissions
2162 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2163 * checking is to be performed on the raw inode simply passs init_user_ns.
2165 void inode_init_owner(struct user_namespace *mnt_userns, struct inode *inode,
2166 const struct inode *dir, umode_t mode)
2168 inode_fsuid_set(inode, mnt_userns);
2169 if (dir && dir->i_mode & S_ISGID) {
2170 inode->i_gid = dir->i_gid;
2172 /* Directories are special, and always inherit S_ISGID */
2175 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2176 !in_group_p(i_gid_into_mnt(mnt_userns, dir)) &&
2177 !capable_wrt_inode_uidgid(mnt_userns, dir, CAP_FSETID))
2180 inode_fsgid_set(inode, mnt_userns);
2181 inode->i_mode = mode;
2183 EXPORT_SYMBOL(inode_init_owner);
2186 * inode_owner_or_capable - check current task permissions to inode
2187 * @mnt_userns: user namespace of the mount the inode was found from
2188 * @inode: inode being checked
2190 * Return true if current either has CAP_FOWNER in a namespace with the
2191 * inode owner uid mapped, or owns the file.
2193 * If the inode has been found through an idmapped mount the user namespace of
2194 * the vfsmount must be passed through @mnt_userns. This function will then take
2195 * care to map the inode according to @mnt_userns before checking permissions.
2196 * On non-idmapped mounts or if permission checking is to be performed on the
2197 * raw inode simply passs init_user_ns.
2199 bool inode_owner_or_capable(struct user_namespace *mnt_userns,
2200 const struct inode *inode)
2203 struct user_namespace *ns;
2205 i_uid = i_uid_into_mnt(mnt_userns, inode);
2206 if (uid_eq(current_fsuid(), i_uid))
2209 ns = current_user_ns();
2210 if (kuid_has_mapping(ns, i_uid) && ns_capable(ns, CAP_FOWNER))
2214 EXPORT_SYMBOL(inode_owner_or_capable);
2217 * Direct i/o helper functions
2219 static void __inode_dio_wait(struct inode *inode)
2221 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2222 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2225 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2226 if (atomic_read(&inode->i_dio_count))
2228 } while (atomic_read(&inode->i_dio_count));
2229 finish_wait(wq, &q.wq_entry);
2233 * inode_dio_wait - wait for outstanding DIO requests to finish
2234 * @inode: inode to wait for
2236 * Waits for all pending direct I/O requests to finish so that we can
2237 * proceed with a truncate or equivalent operation.
2239 * Must be called under a lock that serializes taking new references
2240 * to i_dio_count, usually by inode->i_mutex.
2242 void inode_dio_wait(struct inode *inode)
2244 if (atomic_read(&inode->i_dio_count))
2245 __inode_dio_wait(inode);
2247 EXPORT_SYMBOL(inode_dio_wait);
2250 * inode_set_flags - atomically set some inode flags
2252 * Note: the caller should be holding i_mutex, or else be sure that
2253 * they have exclusive access to the inode structure (i.e., while the
2254 * inode is being instantiated). The reason for the cmpxchg() loop
2255 * --- which wouldn't be necessary if all code paths which modify
2256 * i_flags actually followed this rule, is that there is at least one
2257 * code path which doesn't today so we use cmpxchg() out of an abundance
2260 * In the long run, i_mutex is overkill, and we should probably look
2261 * at using the i_lock spinlock to protect i_flags, and then make sure
2262 * it is so documented in include/linux/fs.h and that all code follows
2263 * the locking convention!!
2265 void inode_set_flags(struct inode *inode, unsigned int flags,
2268 WARN_ON_ONCE(flags & ~mask);
2269 set_mask_bits(&inode->i_flags, mask, flags);
2271 EXPORT_SYMBOL(inode_set_flags);
2273 void inode_nohighmem(struct inode *inode)
2275 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2277 EXPORT_SYMBOL(inode_nohighmem);
2280 * timestamp_truncate - Truncate timespec to a granularity
2282 * @inode: inode being updated
2284 * Truncate a timespec to the granularity supported by the fs
2285 * containing the inode. Always rounds down. gran must
2286 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2288 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2290 struct super_block *sb = inode->i_sb;
2291 unsigned int gran = sb->s_time_gran;
2293 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2294 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2297 /* Avoid division in the common cases 1 ns and 1 s. */
2300 else if (gran == NSEC_PER_SEC)
2302 else if (gran > 1 && gran < NSEC_PER_SEC)
2303 t.tv_nsec -= t.tv_nsec % gran;
2305 WARN(1, "invalid file time granularity: %u", gran);
2308 EXPORT_SYMBOL(timestamp_truncate);
2311 * current_time - Return FS time
2314 * Return the current time truncated to the time granularity supported by
2317 * Note that inode and inode->sb cannot be NULL.
2318 * Otherwise, the function warns and returns time without truncation.
2320 struct timespec64 current_time(struct inode *inode)
2322 struct timespec64 now;
2324 ktime_get_coarse_real_ts64(&now);
2326 if (unlikely(!inode->i_sb)) {
2327 WARN(1, "current_time() called with uninitialized super_block in the inode");
2331 return timestamp_truncate(now, inode);
2333 EXPORT_SYMBOL(current_time);