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;
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_add_lru(struct inode *inode, bool rotate)
433 if (inode->i_state & (I_DIRTY_ALL | I_SYNC | I_FREEING | I_WILL_FREE))
435 if (atomic_read(&inode->i_count))
437 if (!(inode->i_sb->s_flags & SB_ACTIVE))
439 if (!mapping_shrinkable(&inode->i_data))
442 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
443 this_cpu_inc(nr_unused);
445 inode->i_state |= I_REFERENCED;
449 * Add inode to LRU if needed (inode is unused and clean).
451 * Needs inode->i_lock held.
453 void inode_add_lru(struct inode *inode)
455 __inode_add_lru(inode, false);
458 static void inode_lru_list_del(struct inode *inode)
460 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
461 this_cpu_dec(nr_unused);
465 * inode_sb_list_add - add inode to the superblock list of inodes
466 * @inode: inode to add
468 void inode_sb_list_add(struct inode *inode)
470 spin_lock(&inode->i_sb->s_inode_list_lock);
471 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
472 spin_unlock(&inode->i_sb->s_inode_list_lock);
474 EXPORT_SYMBOL_GPL(inode_sb_list_add);
476 static inline void inode_sb_list_del(struct inode *inode)
478 if (!list_empty(&inode->i_sb_list)) {
479 spin_lock(&inode->i_sb->s_inode_list_lock);
480 list_del_init(&inode->i_sb_list);
481 spin_unlock(&inode->i_sb->s_inode_list_lock);
485 static unsigned long hash(struct super_block *sb, unsigned long hashval)
489 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
491 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
492 return tmp & i_hash_mask;
496 * __insert_inode_hash - hash an inode
497 * @inode: unhashed inode
498 * @hashval: unsigned long value used to locate this object in the
501 * Add an inode to the inode hash for this superblock.
503 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
505 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
507 spin_lock(&inode_hash_lock);
508 spin_lock(&inode->i_lock);
509 hlist_add_head_rcu(&inode->i_hash, b);
510 spin_unlock(&inode->i_lock);
511 spin_unlock(&inode_hash_lock);
513 EXPORT_SYMBOL(__insert_inode_hash);
516 * __remove_inode_hash - remove an inode from the hash
517 * @inode: inode to unhash
519 * Remove an inode from the superblock.
521 void __remove_inode_hash(struct inode *inode)
523 spin_lock(&inode_hash_lock);
524 spin_lock(&inode->i_lock);
525 hlist_del_init_rcu(&inode->i_hash);
526 spin_unlock(&inode->i_lock);
527 spin_unlock(&inode_hash_lock);
529 EXPORT_SYMBOL(__remove_inode_hash);
531 void clear_inode(struct inode *inode)
534 * We have to cycle the i_pages lock here because reclaim can be in the
535 * process of removing the last page (in __delete_from_page_cache())
536 * and we must not free the mapping under it.
538 xa_lock_irq(&inode->i_data.i_pages);
539 BUG_ON(inode->i_data.nrpages);
541 * Almost always, mapping_empty(&inode->i_data) here; but there are
542 * two known and long-standing ways in which nodes may get left behind
543 * (when deep radix-tree node allocation failed partway; or when THP
544 * collapse_file() failed). Until those two known cases are cleaned up,
545 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
546 * nor even WARN_ON(!mapping_empty).
548 xa_unlock_irq(&inode->i_data.i_pages);
549 BUG_ON(!list_empty(&inode->i_data.private_list));
550 BUG_ON(!(inode->i_state & I_FREEING));
551 BUG_ON(inode->i_state & I_CLEAR);
552 BUG_ON(!list_empty(&inode->i_wb_list));
553 /* don't need i_lock here, no concurrent mods to i_state */
554 inode->i_state = I_FREEING | I_CLEAR;
556 EXPORT_SYMBOL(clear_inode);
559 * Free the inode passed in, removing it from the lists it is still connected
560 * to. We remove any pages still attached to the inode and wait for any IO that
561 * is still in progress before finally destroying the inode.
563 * An inode must already be marked I_FREEING so that we avoid the inode being
564 * moved back onto lists if we race with other code that manipulates the lists
565 * (e.g. writeback_single_inode). The caller is responsible for setting this.
567 * An inode must already be removed from the LRU list before being evicted from
568 * the cache. This should occur atomically with setting the I_FREEING state
569 * flag, so no inodes here should ever be on the LRU when being evicted.
571 static void evict(struct inode *inode)
573 const struct super_operations *op = inode->i_sb->s_op;
575 BUG_ON(!(inode->i_state & I_FREEING));
576 BUG_ON(!list_empty(&inode->i_lru));
578 if (!list_empty(&inode->i_io_list))
579 inode_io_list_del(inode);
581 inode_sb_list_del(inode);
584 * Wait for flusher thread to be done with the inode so that filesystem
585 * does not start destroying it while writeback is still running. Since
586 * the inode has I_FREEING set, flusher thread won't start new work on
587 * the inode. We just have to wait for running writeback to finish.
589 inode_wait_for_writeback(inode);
591 if (op->evict_inode) {
592 op->evict_inode(inode);
594 truncate_inode_pages_final(&inode->i_data);
597 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
600 remove_inode_hash(inode);
602 spin_lock(&inode->i_lock);
603 wake_up_bit(&inode->i_state, __I_NEW);
604 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
605 spin_unlock(&inode->i_lock);
607 destroy_inode(inode);
611 * dispose_list - dispose of the contents of a local list
612 * @head: the head of the list to free
614 * Dispose-list gets a local list with local inodes in it, so it doesn't
615 * need to worry about list corruption and SMP locks.
617 static void dispose_list(struct list_head *head)
619 while (!list_empty(head)) {
622 inode = list_first_entry(head, struct inode, i_lru);
623 list_del_init(&inode->i_lru);
631 * evict_inodes - evict all evictable inodes for a superblock
632 * @sb: superblock to operate on
634 * Make sure that no inodes with zero refcount are retained. This is
635 * called by superblock shutdown after having SB_ACTIVE flag removed,
636 * so any inode reaching zero refcount during or after that call will
637 * be immediately evicted.
639 void evict_inodes(struct super_block *sb)
641 struct inode *inode, *next;
645 spin_lock(&sb->s_inode_list_lock);
646 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
647 if (atomic_read(&inode->i_count))
650 spin_lock(&inode->i_lock);
651 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
652 spin_unlock(&inode->i_lock);
656 inode->i_state |= I_FREEING;
657 inode_lru_list_del(inode);
658 spin_unlock(&inode->i_lock);
659 list_add(&inode->i_lru, &dispose);
662 * We can have a ton of inodes to evict at unmount time given
663 * enough memory, check to see if we need to go to sleep for a
664 * bit so we don't livelock.
666 if (need_resched()) {
667 spin_unlock(&sb->s_inode_list_lock);
669 dispose_list(&dispose);
673 spin_unlock(&sb->s_inode_list_lock);
675 dispose_list(&dispose);
677 EXPORT_SYMBOL_GPL(evict_inodes);
680 * invalidate_inodes - attempt to free all inodes on a superblock
681 * @sb: superblock to operate on
682 * @kill_dirty: flag to guide handling of dirty inodes
684 * Attempts to free all inodes for a given superblock. If there were any
685 * busy inodes return a non-zero value, else zero.
686 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
689 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
692 struct inode *inode, *next;
696 spin_lock(&sb->s_inode_list_lock);
697 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
698 spin_lock(&inode->i_lock);
699 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
700 spin_unlock(&inode->i_lock);
703 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
704 spin_unlock(&inode->i_lock);
708 if (atomic_read(&inode->i_count)) {
709 spin_unlock(&inode->i_lock);
714 inode->i_state |= I_FREEING;
715 inode_lru_list_del(inode);
716 spin_unlock(&inode->i_lock);
717 list_add(&inode->i_lru, &dispose);
718 if (need_resched()) {
719 spin_unlock(&sb->s_inode_list_lock);
721 dispose_list(&dispose);
725 spin_unlock(&sb->s_inode_list_lock);
727 dispose_list(&dispose);
733 * Isolate the inode from the LRU in preparation for freeing it.
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
751 * trylock. If we fail to get the lock, just skip it.
753 if (!spin_trylock(&inode->i_lock))
757 * Inodes can get referenced, redirtied, or repopulated while
758 * they're already on the LRU, and this can make them
759 * unreclaimable for a while. Remove them lazily here; iput,
760 * sync, or the last page cache deletion will requeue them.
762 if (atomic_read(&inode->i_count) ||
763 (inode->i_state & ~I_REFERENCED) ||
764 !mapping_shrinkable(&inode->i_data)) {
765 list_lru_isolate(lru, &inode->i_lru);
766 spin_unlock(&inode->i_lock);
767 this_cpu_dec(nr_unused);
771 /* Recently referenced inodes get one more pass */
772 if (inode->i_state & I_REFERENCED) {
773 inode->i_state &= ~I_REFERENCED;
774 spin_unlock(&inode->i_lock);
779 * On highmem systems, mapping_shrinkable() permits dropping
780 * page cache in order to free up struct inodes: lowmem might
781 * be under pressure before the cache inside the highmem zone.
783 if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
785 spin_unlock(&inode->i_lock);
786 spin_unlock(lru_lock);
787 if (remove_inode_buffers(inode)) {
789 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
790 if (current_is_kswapd())
791 __count_vm_events(KSWAPD_INODESTEAL, reap);
793 __count_vm_events(PGINODESTEAL, reap);
794 if (current->reclaim_state)
795 current->reclaim_state->reclaimed_slab += reap;
802 WARN_ON(inode->i_state & I_NEW);
803 inode->i_state |= I_FREEING;
804 list_lru_isolate_move(lru, &inode->i_lru, freeable);
805 spin_unlock(&inode->i_lock);
807 this_cpu_dec(nr_unused);
812 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
813 * This is called from the superblock shrinker function with a number of inodes
814 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
815 * then are freed outside inode_lock by dispose_list().
817 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
822 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
823 inode_lru_isolate, &freeable);
824 dispose_list(&freeable);
828 static void __wait_on_freeing_inode(struct inode *inode);
830 * Called with the inode lock held.
832 static struct inode *find_inode(struct super_block *sb,
833 struct hlist_head *head,
834 int (*test)(struct inode *, void *),
837 struct inode *inode = NULL;
840 hlist_for_each_entry(inode, head, i_hash) {
841 if (inode->i_sb != sb)
843 if (!test(inode, data))
845 spin_lock(&inode->i_lock);
846 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
847 __wait_on_freeing_inode(inode);
850 if (unlikely(inode->i_state & I_CREATING)) {
851 spin_unlock(&inode->i_lock);
852 return ERR_PTR(-ESTALE);
855 spin_unlock(&inode->i_lock);
862 * find_inode_fast is the fast path version of find_inode, see the comment at
863 * iget_locked for details.
865 static struct inode *find_inode_fast(struct super_block *sb,
866 struct hlist_head *head, unsigned long ino)
868 struct inode *inode = NULL;
871 hlist_for_each_entry(inode, head, i_hash) {
872 if (inode->i_ino != ino)
874 if (inode->i_sb != sb)
876 spin_lock(&inode->i_lock);
877 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
878 __wait_on_freeing_inode(inode);
881 if (unlikely(inode->i_state & I_CREATING)) {
882 spin_unlock(&inode->i_lock);
883 return ERR_PTR(-ESTALE);
886 spin_unlock(&inode->i_lock);
893 * Each cpu owns a range of LAST_INO_BATCH numbers.
894 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
895 * to renew the exhausted range.
897 * This does not significantly increase overflow rate because every CPU can
898 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
899 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
900 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
901 * overflow rate by 2x, which does not seem too significant.
903 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
904 * error if st_ino won't fit in target struct field. Use 32bit counter
905 * here to attempt to avoid that.
907 #define LAST_INO_BATCH 1024
908 static DEFINE_PER_CPU(unsigned int, last_ino);
910 unsigned int get_next_ino(void)
912 unsigned int *p = &get_cpu_var(last_ino);
913 unsigned int res = *p;
916 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
917 static atomic_t shared_last_ino;
918 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
920 res = next - LAST_INO_BATCH;
925 /* get_next_ino should not provide a 0 inode number */
929 put_cpu_var(last_ino);
932 EXPORT_SYMBOL(get_next_ino);
935 * new_inode_pseudo - obtain an inode
938 * Allocates a new inode for given superblock.
939 * Inode wont be chained in superblock s_inodes list
941 * - fs can't be unmount
942 * - quotas, fsnotify, writeback can't work
944 struct inode *new_inode_pseudo(struct super_block *sb)
946 struct inode *inode = alloc_inode(sb);
949 spin_lock(&inode->i_lock);
951 spin_unlock(&inode->i_lock);
952 INIT_LIST_HEAD(&inode->i_sb_list);
958 * new_inode - obtain an inode
961 * Allocates a new inode for given superblock. The default gfp_mask
962 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
963 * If HIGHMEM pages are unsuitable or it is known that pages allocated
964 * for the page cache are not reclaimable or migratable,
965 * mapping_set_gfp_mask() must be called with suitable flags on the
966 * newly created inode's mapping
969 struct inode *new_inode(struct super_block *sb)
973 spin_lock_prefetch(&sb->s_inode_list_lock);
975 inode = new_inode_pseudo(sb);
977 inode_sb_list_add(inode);
980 EXPORT_SYMBOL(new_inode);
982 #ifdef CONFIG_DEBUG_LOCK_ALLOC
983 void lockdep_annotate_inode_mutex_key(struct inode *inode)
985 if (S_ISDIR(inode->i_mode)) {
986 struct file_system_type *type = inode->i_sb->s_type;
988 /* Set new key only if filesystem hasn't already changed it */
989 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
991 * ensure nobody is actually holding i_mutex
993 // mutex_destroy(&inode->i_mutex);
994 init_rwsem(&inode->i_rwsem);
995 lockdep_set_class(&inode->i_rwsem,
996 &type->i_mutex_dir_key);
1000 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1004 * unlock_new_inode - clear the I_NEW state and wake up any waiters
1005 * @inode: new inode to unlock
1007 * Called when the inode is fully initialised to clear the new state of the
1008 * inode and wake up anyone waiting for the inode to finish initialisation.
1010 void unlock_new_inode(struct inode *inode)
1012 lockdep_annotate_inode_mutex_key(inode);
1013 spin_lock(&inode->i_lock);
1014 WARN_ON(!(inode->i_state & I_NEW));
1015 inode->i_state &= ~I_NEW & ~I_CREATING;
1017 wake_up_bit(&inode->i_state, __I_NEW);
1018 spin_unlock(&inode->i_lock);
1020 EXPORT_SYMBOL(unlock_new_inode);
1022 void discard_new_inode(struct inode *inode)
1024 lockdep_annotate_inode_mutex_key(inode);
1025 spin_lock(&inode->i_lock);
1026 WARN_ON(!(inode->i_state & I_NEW));
1027 inode->i_state &= ~I_NEW;
1029 wake_up_bit(&inode->i_state, __I_NEW);
1030 spin_unlock(&inode->i_lock);
1033 EXPORT_SYMBOL(discard_new_inode);
1036 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1038 * Lock any non-NULL argument that is not a directory.
1039 * Zero, one or two objects may be locked by this function.
1041 * @inode1: first inode to lock
1042 * @inode2: second inode to lock
1044 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1046 if (inode1 > inode2)
1047 swap(inode1, inode2);
1049 if (inode1 && !S_ISDIR(inode1->i_mode))
1051 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1052 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1054 EXPORT_SYMBOL(lock_two_nondirectories);
1057 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1058 * @inode1: first inode to unlock
1059 * @inode2: second inode to unlock
1061 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1063 if (inode1 && !S_ISDIR(inode1->i_mode))
1064 inode_unlock(inode1);
1065 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1066 inode_unlock(inode2);
1068 EXPORT_SYMBOL(unlock_two_nondirectories);
1071 * inode_insert5 - obtain an inode from a mounted file system
1072 * @inode: pre-allocated inode to use for insert to cache
1073 * @hashval: hash value (usually inode number) to get
1074 * @test: callback used for comparisons between inodes
1075 * @set: callback used to initialize a new struct inode
1076 * @data: opaque data pointer to pass to @test and @set
1078 * Search for the inode specified by @hashval and @data in the inode cache,
1079 * and if present it is return it with an increased reference count. This is
1080 * a variant of iget5_locked() for callers that don't want to fail on memory
1081 * allocation of inode.
1083 * If the inode is not in cache, insert the pre-allocated inode to cache and
1084 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1085 * to fill it in before unlocking it via unlock_new_inode().
1087 * Note both @test and @set are called with the inode_hash_lock held, so can't
1090 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1091 int (*test)(struct inode *, void *),
1092 int (*set)(struct inode *, void *), void *data)
1094 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1096 bool creating = inode->i_state & I_CREATING;
1099 spin_lock(&inode_hash_lock);
1100 old = find_inode(inode->i_sb, head, test, data);
1101 if (unlikely(old)) {
1103 * Uhhuh, somebody else created the same inode under us.
1104 * Use the old inode instead of the preallocated one.
1106 spin_unlock(&inode_hash_lock);
1110 if (unlikely(inode_unhashed(old))) {
1117 if (set && unlikely(set(inode, data))) {
1123 * Return the locked inode with I_NEW set, the
1124 * caller is responsible for filling in the contents
1126 spin_lock(&inode->i_lock);
1127 inode->i_state |= I_NEW;
1128 hlist_add_head_rcu(&inode->i_hash, head);
1129 spin_unlock(&inode->i_lock);
1131 inode_sb_list_add(inode);
1133 spin_unlock(&inode_hash_lock);
1137 EXPORT_SYMBOL(inode_insert5);
1140 * iget5_locked - obtain an inode from a mounted file system
1141 * @sb: super block of file system
1142 * @hashval: hash value (usually inode number) to get
1143 * @test: callback used for comparisons between inodes
1144 * @set: callback used to initialize a new struct inode
1145 * @data: opaque data pointer to pass to @test and @set
1147 * Search for the inode specified by @hashval and @data in the inode cache,
1148 * and if present it is return it with an increased reference count. This is
1149 * a generalized version of iget_locked() for file systems where the inode
1150 * number is not sufficient for unique identification of an inode.
1152 * If the inode is not in cache, allocate a new inode and return it locked,
1153 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1154 * before unlocking it via unlock_new_inode().
1156 * Note both @test and @set are called with the inode_hash_lock held, so can't
1159 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1160 int (*test)(struct inode *, void *),
1161 int (*set)(struct inode *, void *), void *data)
1163 struct inode *inode = ilookup5(sb, hashval, test, data);
1166 struct inode *new = alloc_inode(sb);
1170 inode = inode_insert5(new, hashval, test, set, data);
1171 if (unlikely(inode != new))
1177 EXPORT_SYMBOL(iget5_locked);
1180 * iget_locked - obtain an inode from a mounted file system
1181 * @sb: super block of file system
1182 * @ino: inode number to get
1184 * Search for the inode specified by @ino in the inode cache and if present
1185 * return it with an increased reference count. This is for file systems
1186 * where the inode number is sufficient for unique identification of an inode.
1188 * If the inode is not in cache, allocate a new inode and return it locked,
1189 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1190 * before unlocking it via unlock_new_inode().
1192 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1194 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1195 struct inode *inode;
1197 spin_lock(&inode_hash_lock);
1198 inode = find_inode_fast(sb, head, ino);
1199 spin_unlock(&inode_hash_lock);
1203 wait_on_inode(inode);
1204 if (unlikely(inode_unhashed(inode))) {
1211 inode = alloc_inode(sb);
1215 spin_lock(&inode_hash_lock);
1216 /* We released the lock, so.. */
1217 old = find_inode_fast(sb, head, ino);
1220 spin_lock(&inode->i_lock);
1221 inode->i_state = I_NEW;
1222 hlist_add_head_rcu(&inode->i_hash, head);
1223 spin_unlock(&inode->i_lock);
1224 inode_sb_list_add(inode);
1225 spin_unlock(&inode_hash_lock);
1227 /* Return the locked inode with I_NEW set, the
1228 * caller is responsible for filling in the contents
1234 * Uhhuh, somebody else created the same inode under
1235 * us. Use the old inode instead of the one we just
1238 spin_unlock(&inode_hash_lock);
1239 destroy_inode(inode);
1243 wait_on_inode(inode);
1244 if (unlikely(inode_unhashed(inode))) {
1251 EXPORT_SYMBOL(iget_locked);
1254 * search the inode cache for a matching inode number.
1255 * If we find one, then the inode number we are trying to
1256 * allocate is not unique and so we should not use it.
1258 * Returns 1 if the inode number is unique, 0 if it is not.
1260 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1262 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1263 struct inode *inode;
1265 hlist_for_each_entry_rcu(inode, b, i_hash) {
1266 if (inode->i_ino == ino && inode->i_sb == sb)
1273 * iunique - get a unique inode number
1275 * @max_reserved: highest reserved inode number
1277 * Obtain an inode number that is unique on the system for a given
1278 * superblock. This is used by file systems that have no natural
1279 * permanent inode numbering system. An inode number is returned that
1280 * is higher than the reserved limit but unique.
1283 * With a large number of inodes live on the file system this function
1284 * currently becomes quite slow.
1286 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1289 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1290 * error if st_ino won't fit in target struct field. Use 32bit counter
1291 * here to attempt to avoid that.
1293 static DEFINE_SPINLOCK(iunique_lock);
1294 static unsigned int counter;
1298 spin_lock(&iunique_lock);
1300 if (counter <= max_reserved)
1301 counter = max_reserved + 1;
1303 } while (!test_inode_iunique(sb, res));
1304 spin_unlock(&iunique_lock);
1309 EXPORT_SYMBOL(iunique);
1311 struct inode *igrab(struct inode *inode)
1313 spin_lock(&inode->i_lock);
1314 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1316 spin_unlock(&inode->i_lock);
1318 spin_unlock(&inode->i_lock);
1320 * Handle the case where s_op->clear_inode is not been
1321 * called yet, and somebody is calling igrab
1322 * while the inode is getting freed.
1328 EXPORT_SYMBOL(igrab);
1331 * ilookup5_nowait - search for an inode in the inode cache
1332 * @sb: super block of file system to search
1333 * @hashval: hash value (usually inode number) to search for
1334 * @test: callback used for comparisons between inodes
1335 * @data: opaque data pointer to pass to @test
1337 * Search for the inode specified by @hashval and @data in the inode cache.
1338 * If the inode is in the cache, the inode is returned with an incremented
1341 * Note: I_NEW is not waited upon so you have to be very careful what you do
1342 * with the returned inode. You probably should be using ilookup5() instead.
1344 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1346 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1347 int (*test)(struct inode *, void *), void *data)
1349 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1350 struct inode *inode;
1352 spin_lock(&inode_hash_lock);
1353 inode = find_inode(sb, head, test, data);
1354 spin_unlock(&inode_hash_lock);
1356 return IS_ERR(inode) ? NULL : inode;
1358 EXPORT_SYMBOL(ilookup5_nowait);
1361 * ilookup5 - search for an inode in the inode cache
1362 * @sb: super block of file system to search
1363 * @hashval: hash value (usually inode number) to search for
1364 * @test: callback used for comparisons between inodes
1365 * @data: opaque data pointer to pass to @test
1367 * Search for the inode specified by @hashval and @data in the inode cache,
1368 * and if the inode is in the cache, return the inode with an incremented
1369 * reference count. Waits on I_NEW before returning the inode.
1370 * returned with an incremented reference count.
1372 * This is a generalized version of ilookup() for file systems where the
1373 * inode number is not sufficient for unique identification of an inode.
1375 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1377 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1378 int (*test)(struct inode *, void *), void *data)
1380 struct inode *inode;
1382 inode = ilookup5_nowait(sb, hashval, test, data);
1384 wait_on_inode(inode);
1385 if (unlikely(inode_unhashed(inode))) {
1392 EXPORT_SYMBOL(ilookup5);
1395 * ilookup - search for an inode in the inode cache
1396 * @sb: super block of file system to search
1397 * @ino: inode number to search for
1399 * Search for the inode @ino in the inode cache, and if the inode is in the
1400 * cache, the inode is returned with an incremented reference count.
1402 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1404 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1405 struct inode *inode;
1407 spin_lock(&inode_hash_lock);
1408 inode = find_inode_fast(sb, head, ino);
1409 spin_unlock(&inode_hash_lock);
1414 wait_on_inode(inode);
1415 if (unlikely(inode_unhashed(inode))) {
1422 EXPORT_SYMBOL(ilookup);
1425 * find_inode_nowait - find an inode in the inode cache
1426 * @sb: super block of file system to search
1427 * @hashval: hash value (usually inode number) to search for
1428 * @match: callback used for comparisons between inodes
1429 * @data: opaque data pointer to pass to @match
1431 * Search for the inode specified by @hashval and @data in the inode
1432 * cache, where the helper function @match will return 0 if the inode
1433 * does not match, 1 if the inode does match, and -1 if the search
1434 * should be stopped. The @match function must be responsible for
1435 * taking the i_lock spin_lock and checking i_state for an inode being
1436 * freed or being initialized, and incrementing the reference count
1437 * before returning 1. It also must not sleep, since it is called with
1438 * the inode_hash_lock spinlock held.
1440 * This is a even more generalized version of ilookup5() when the
1441 * function must never block --- find_inode() can block in
1442 * __wait_on_freeing_inode() --- or when the caller can not increment
1443 * the reference count because the resulting iput() might cause an
1444 * inode eviction. The tradeoff is that the @match funtion must be
1445 * very carefully implemented.
1447 struct inode *find_inode_nowait(struct super_block *sb,
1448 unsigned long hashval,
1449 int (*match)(struct inode *, unsigned long,
1453 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1454 struct inode *inode, *ret_inode = NULL;
1457 spin_lock(&inode_hash_lock);
1458 hlist_for_each_entry(inode, head, i_hash) {
1459 if (inode->i_sb != sb)
1461 mval = match(inode, hashval, data);
1469 spin_unlock(&inode_hash_lock);
1472 EXPORT_SYMBOL(find_inode_nowait);
1475 * find_inode_rcu - find an inode in the inode cache
1476 * @sb: Super block of file system to search
1477 * @hashval: Key to hash
1478 * @test: Function to test match on an inode
1479 * @data: Data for test function
1481 * Search for the inode specified by @hashval and @data in the inode cache,
1482 * where the helper function @test will return 0 if the inode does not match
1483 * and 1 if it does. The @test function must be responsible for taking the
1484 * i_lock spin_lock and checking i_state for an inode being freed or being
1487 * If successful, this will return the inode for which the @test function
1488 * returned 1 and NULL otherwise.
1490 * The @test function is not permitted to take a ref on any inode presented.
1491 * It is also not permitted to sleep.
1493 * The caller must hold the RCU read lock.
1495 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1496 int (*test)(struct inode *, void *), void *data)
1498 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1499 struct inode *inode;
1501 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1502 "suspicious find_inode_rcu() usage");
1504 hlist_for_each_entry_rcu(inode, head, i_hash) {
1505 if (inode->i_sb == sb &&
1506 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1512 EXPORT_SYMBOL(find_inode_rcu);
1515 * find_inode_by_ino_rcu - Find an inode in the inode cache
1516 * @sb: Super block of file system to search
1517 * @ino: The inode number to match
1519 * Search for the inode specified by @hashval and @data in the inode cache,
1520 * where the helper function @test will return 0 if the inode does not match
1521 * and 1 if it does. The @test function must be responsible for taking the
1522 * i_lock spin_lock and checking i_state for an inode being freed or being
1525 * If successful, this will return the inode for which the @test function
1526 * returned 1 and NULL otherwise.
1528 * The @test function is not permitted to take a ref on any inode presented.
1529 * It is also not permitted to sleep.
1531 * The caller must hold the RCU read lock.
1533 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1536 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1537 struct inode *inode;
1539 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1540 "suspicious find_inode_by_ino_rcu() usage");
1542 hlist_for_each_entry_rcu(inode, head, i_hash) {
1543 if (inode->i_ino == ino &&
1544 inode->i_sb == sb &&
1545 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1550 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1552 int insert_inode_locked(struct inode *inode)
1554 struct super_block *sb = inode->i_sb;
1555 ino_t ino = inode->i_ino;
1556 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1559 struct inode *old = NULL;
1560 spin_lock(&inode_hash_lock);
1561 hlist_for_each_entry(old, head, i_hash) {
1562 if (old->i_ino != ino)
1564 if (old->i_sb != sb)
1566 spin_lock(&old->i_lock);
1567 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1568 spin_unlock(&old->i_lock);
1574 spin_lock(&inode->i_lock);
1575 inode->i_state |= I_NEW | I_CREATING;
1576 hlist_add_head_rcu(&inode->i_hash, head);
1577 spin_unlock(&inode->i_lock);
1578 spin_unlock(&inode_hash_lock);
1581 if (unlikely(old->i_state & I_CREATING)) {
1582 spin_unlock(&old->i_lock);
1583 spin_unlock(&inode_hash_lock);
1587 spin_unlock(&old->i_lock);
1588 spin_unlock(&inode_hash_lock);
1590 if (unlikely(!inode_unhashed(old))) {
1597 EXPORT_SYMBOL(insert_inode_locked);
1599 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1600 int (*test)(struct inode *, void *), void *data)
1604 inode->i_state |= I_CREATING;
1605 old = inode_insert5(inode, hashval, test, NULL, data);
1613 EXPORT_SYMBOL(insert_inode_locked4);
1616 int generic_delete_inode(struct inode *inode)
1620 EXPORT_SYMBOL(generic_delete_inode);
1623 * Called when we're dropping the last reference
1626 * Call the FS "drop_inode()" function, defaulting to
1627 * the legacy UNIX filesystem behaviour. If it tells
1628 * us to evict inode, do so. Otherwise, retain inode
1629 * in cache if fs is alive, sync and evict if fs is
1632 static void iput_final(struct inode *inode)
1634 struct super_block *sb = inode->i_sb;
1635 const struct super_operations *op = inode->i_sb->s_op;
1636 unsigned long state;
1639 WARN_ON(inode->i_state & I_NEW);
1642 drop = op->drop_inode(inode);
1644 drop = generic_drop_inode(inode);
1647 !(inode->i_state & I_DONTCACHE) &&
1648 (sb->s_flags & SB_ACTIVE)) {
1649 __inode_add_lru(inode, true);
1650 spin_unlock(&inode->i_lock);
1654 state = inode->i_state;
1656 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1657 spin_unlock(&inode->i_lock);
1659 write_inode_now(inode, 1);
1661 spin_lock(&inode->i_lock);
1662 state = inode->i_state;
1663 WARN_ON(state & I_NEW);
1664 state &= ~I_WILL_FREE;
1667 WRITE_ONCE(inode->i_state, state | I_FREEING);
1668 if (!list_empty(&inode->i_lru))
1669 inode_lru_list_del(inode);
1670 spin_unlock(&inode->i_lock);
1676 * iput - put an inode
1677 * @inode: inode to put
1679 * Puts an inode, dropping its usage count. If the inode use count hits
1680 * zero, the inode is then freed and may also be destroyed.
1682 * Consequently, iput() can sleep.
1684 void iput(struct inode *inode)
1688 BUG_ON(inode->i_state & I_CLEAR);
1690 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1691 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1692 atomic_inc(&inode->i_count);
1693 spin_unlock(&inode->i_lock);
1694 trace_writeback_lazytime_iput(inode);
1695 mark_inode_dirty_sync(inode);
1701 EXPORT_SYMBOL(iput);
1705 * bmap - find a block number in a file
1706 * @inode: inode owning the block number being requested
1707 * @block: pointer containing the block to find
1709 * Replaces the value in ``*block`` with the block number on the device holding
1710 * corresponding to the requested block number in the file.
1711 * That is, asked for block 4 of inode 1 the function will replace the
1712 * 4 in ``*block``, with disk block relative to the disk start that holds that
1713 * block of the file.
1715 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1716 * hole, returns 0 and ``*block`` is also set to 0.
1718 int bmap(struct inode *inode, sector_t *block)
1720 if (!inode->i_mapping->a_ops->bmap)
1723 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1726 EXPORT_SYMBOL(bmap);
1730 * With relative atime, only update atime if the previous atime is
1731 * earlier than either the ctime or mtime or if at least a day has
1732 * passed since the last atime update.
1734 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1735 struct timespec64 now)
1738 if (!(mnt->mnt_flags & MNT_RELATIME))
1741 * Is mtime younger than atime? If yes, update atime:
1743 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1746 * Is ctime younger than atime? If yes, update atime:
1748 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1752 * Is the previous atime value older than a day? If yes,
1755 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1758 * Good, we can skip the atime update:
1763 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1765 int dirty_flags = 0;
1767 if (flags & (S_ATIME | S_CTIME | S_MTIME)) {
1768 if (flags & S_ATIME)
1769 inode->i_atime = *time;
1770 if (flags & S_CTIME)
1771 inode->i_ctime = *time;
1772 if (flags & S_MTIME)
1773 inode->i_mtime = *time;
1775 if (inode->i_sb->s_flags & SB_LAZYTIME)
1776 dirty_flags |= I_DIRTY_TIME;
1778 dirty_flags |= I_DIRTY_SYNC;
1781 if ((flags & S_VERSION) && inode_maybe_inc_iversion(inode, false))
1782 dirty_flags |= I_DIRTY_SYNC;
1784 __mark_inode_dirty(inode, dirty_flags);
1787 EXPORT_SYMBOL(generic_update_time);
1790 * This does the actual work of updating an inodes time or version. Must have
1791 * had called mnt_want_write() before calling this.
1793 int inode_update_time(struct inode *inode, struct timespec64 *time, int flags)
1795 if (inode->i_op->update_time)
1796 return inode->i_op->update_time(inode, time, flags);
1797 return generic_update_time(inode, time, flags);
1799 EXPORT_SYMBOL(inode_update_time);
1802 * atime_needs_update - update the access time
1803 * @path: the &struct path to update
1804 * @inode: inode to update
1806 * Update the accessed time on an inode and mark it for writeback.
1807 * This function automatically handles read only file systems and media,
1808 * as well as the "noatime" flag and inode specific "noatime" markers.
1810 bool atime_needs_update(const struct path *path, struct inode *inode)
1812 struct vfsmount *mnt = path->mnt;
1813 struct timespec64 now;
1815 if (inode->i_flags & S_NOATIME)
1818 /* Atime updates will likely cause i_uid and i_gid to be written
1819 * back improprely if their true value is unknown to the vfs.
1821 if (HAS_UNMAPPED_ID(mnt_user_ns(mnt), inode))
1824 if (IS_NOATIME(inode))
1826 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1829 if (mnt->mnt_flags & MNT_NOATIME)
1831 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1834 now = current_time(inode);
1836 if (!relatime_need_update(mnt, inode, now))
1839 if (timespec64_equal(&inode->i_atime, &now))
1845 void touch_atime(const struct path *path)
1847 struct vfsmount *mnt = path->mnt;
1848 struct inode *inode = d_inode(path->dentry);
1849 struct timespec64 now;
1851 if (!atime_needs_update(path, inode))
1854 if (!sb_start_write_trylock(inode->i_sb))
1857 if (__mnt_want_write(mnt) != 0)
1860 * File systems can error out when updating inodes if they need to
1861 * allocate new space to modify an inode (such is the case for
1862 * Btrfs), but since we touch atime while walking down the path we
1863 * really don't care if we failed to update the atime of the file,
1864 * so just ignore the return value.
1865 * We may also fail on filesystems that have the ability to make parts
1866 * of the fs read only, e.g. subvolumes in Btrfs.
1868 now = current_time(inode);
1869 inode_update_time(inode, &now, S_ATIME);
1870 __mnt_drop_write(mnt);
1872 sb_end_write(inode->i_sb);
1874 EXPORT_SYMBOL(touch_atime);
1877 * The logic we want is
1879 * if suid or (sgid and xgrp)
1882 int should_remove_suid(struct dentry *dentry)
1884 umode_t mode = d_inode(dentry)->i_mode;
1887 /* suid always must be killed */
1888 if (unlikely(mode & S_ISUID))
1889 kill = ATTR_KILL_SUID;
1892 * sgid without any exec bits is just a mandatory locking mark; leave
1893 * it alone. If some exec bits are set, it's a real sgid; kill it.
1895 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1896 kill |= ATTR_KILL_SGID;
1898 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1903 EXPORT_SYMBOL(should_remove_suid);
1906 * Return mask of changes for notify_change() that need to be done as a
1907 * response to write or truncate. Return 0 if nothing has to be changed.
1908 * Negative value on error (change should be denied).
1910 int dentry_needs_remove_privs(struct dentry *dentry)
1912 struct inode *inode = d_inode(dentry);
1916 if (IS_NOSEC(inode))
1919 mask = should_remove_suid(dentry);
1920 ret = security_inode_need_killpriv(dentry);
1924 mask |= ATTR_KILL_PRIV;
1928 static int __remove_privs(struct user_namespace *mnt_userns,
1929 struct dentry *dentry, int kill)
1931 struct iattr newattrs;
1933 newattrs.ia_valid = ATTR_FORCE | kill;
1935 * Note we call this on write, so notify_change will not
1936 * encounter any conflicting delegations:
1938 return notify_change(mnt_userns, dentry, &newattrs, NULL);
1942 * Remove special file priviledges (suid, capabilities) when file is written
1945 int file_remove_privs(struct file *file)
1947 struct dentry *dentry = file_dentry(file);
1948 struct inode *inode = file_inode(file);
1953 * Fast path for nothing security related.
1954 * As well for non-regular files, e.g. blkdev inodes.
1955 * For example, blkdev_write_iter() might get here
1956 * trying to remove privs which it is not allowed to.
1958 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1961 kill = dentry_needs_remove_privs(dentry);
1965 error = __remove_privs(file_mnt_user_ns(file), dentry, kill);
1967 inode_has_no_xattr(inode);
1971 EXPORT_SYMBOL(file_remove_privs);
1974 * file_update_time - update mtime and ctime time
1975 * @file: file accessed
1977 * Update the mtime and ctime members of an inode and mark the inode
1978 * for writeback. Note that this function is meant exclusively for
1979 * usage in the file write path of filesystems, and filesystems may
1980 * choose to explicitly ignore update via this function with the
1981 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1982 * timestamps are handled by the server. This can return an error for
1983 * file systems who need to allocate space in order to update an inode.
1986 int file_update_time(struct file *file)
1988 struct inode *inode = file_inode(file);
1989 struct timespec64 now;
1993 /* First try to exhaust all avenues to not sync */
1994 if (IS_NOCMTIME(inode))
1997 now = current_time(inode);
1998 if (!timespec64_equal(&inode->i_mtime, &now))
2001 if (!timespec64_equal(&inode->i_ctime, &now))
2004 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2005 sync_it |= S_VERSION;
2010 /* Finally allowed to write? Takes lock. */
2011 if (__mnt_want_write_file(file))
2014 ret = inode_update_time(inode, &now, sync_it);
2015 __mnt_drop_write_file(file);
2019 EXPORT_SYMBOL(file_update_time);
2021 /* Caller must hold the file's inode lock */
2022 int file_modified(struct file *file)
2027 * Clear the security bits if the process is not being run by root.
2028 * This keeps people from modifying setuid and setgid binaries.
2030 err = file_remove_privs(file);
2034 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2037 return file_update_time(file);
2039 EXPORT_SYMBOL(file_modified);
2041 int inode_needs_sync(struct inode *inode)
2045 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2049 EXPORT_SYMBOL(inode_needs_sync);
2052 * If we try to find an inode in the inode hash while it is being
2053 * deleted, we have to wait until the filesystem completes its
2054 * deletion before reporting that it isn't found. This function waits
2055 * until the deletion _might_ have completed. Callers are responsible
2056 * to recheck inode state.
2058 * It doesn't matter if I_NEW is not set initially, a call to
2059 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2062 static void __wait_on_freeing_inode(struct inode *inode)
2064 wait_queue_head_t *wq;
2065 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2066 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2067 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2068 spin_unlock(&inode->i_lock);
2069 spin_unlock(&inode_hash_lock);
2071 finish_wait(wq, &wait.wq_entry);
2072 spin_lock(&inode_hash_lock);
2075 static __initdata unsigned long ihash_entries;
2076 static int __init set_ihash_entries(char *str)
2080 ihash_entries = simple_strtoul(str, &str, 0);
2083 __setup("ihash_entries=", set_ihash_entries);
2086 * Initialize the waitqueues and inode hash table.
2088 void __init inode_init_early(void)
2090 /* If hashes are distributed across NUMA nodes, defer
2091 * hash allocation until vmalloc space is available.
2097 alloc_large_system_hash("Inode-cache",
2098 sizeof(struct hlist_head),
2101 HASH_EARLY | HASH_ZERO,
2108 void __init inode_init(void)
2110 /* inode slab cache */
2111 inode_cachep = kmem_cache_create("inode_cache",
2112 sizeof(struct inode),
2114 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2115 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2118 /* Hash may have been set up in inode_init_early */
2123 alloc_large_system_hash("Inode-cache",
2124 sizeof(struct hlist_head),
2134 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2136 inode->i_mode = mode;
2137 if (S_ISCHR(mode)) {
2138 inode->i_fop = &def_chr_fops;
2139 inode->i_rdev = rdev;
2140 } else if (S_ISBLK(mode)) {
2141 inode->i_fop = &def_blk_fops;
2142 inode->i_rdev = rdev;
2143 } else if (S_ISFIFO(mode))
2144 inode->i_fop = &pipefifo_fops;
2145 else if (S_ISSOCK(mode))
2146 ; /* leave it no_open_fops */
2148 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2149 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2152 EXPORT_SYMBOL(init_special_inode);
2155 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2156 * @mnt_userns: User namespace of the mount the inode was created from
2158 * @dir: Directory inode
2159 * @mode: mode of the new inode
2161 * If the inode has been created through an idmapped mount the user namespace of
2162 * the vfsmount must be passed through @mnt_userns. This function will then take
2163 * care to map the inode according to @mnt_userns before checking permissions
2164 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2165 * checking is to be performed on the raw inode simply passs init_user_ns.
2167 void inode_init_owner(struct user_namespace *mnt_userns, struct inode *inode,
2168 const struct inode *dir, umode_t mode)
2170 inode_fsuid_set(inode, mnt_userns);
2171 if (dir && dir->i_mode & S_ISGID) {
2172 inode->i_gid = dir->i_gid;
2174 /* Directories are special, and always inherit S_ISGID */
2177 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2178 !in_group_p(i_gid_into_mnt(mnt_userns, dir)) &&
2179 !capable_wrt_inode_uidgid(mnt_userns, dir, CAP_FSETID))
2182 inode_fsgid_set(inode, mnt_userns);
2183 inode->i_mode = mode;
2185 EXPORT_SYMBOL(inode_init_owner);
2188 * inode_owner_or_capable - check current task permissions to inode
2189 * @mnt_userns: user namespace of the mount the inode was found from
2190 * @inode: inode being checked
2192 * Return true if current either has CAP_FOWNER in a namespace with the
2193 * inode owner uid mapped, or owns the file.
2195 * If the inode has been found through an idmapped mount the user namespace of
2196 * the vfsmount must be passed through @mnt_userns. This function will then take
2197 * care to map the inode according to @mnt_userns before checking permissions.
2198 * On non-idmapped mounts or if permission checking is to be performed on the
2199 * raw inode simply passs init_user_ns.
2201 bool inode_owner_or_capable(struct user_namespace *mnt_userns,
2202 const struct inode *inode)
2205 struct user_namespace *ns;
2207 i_uid = i_uid_into_mnt(mnt_userns, inode);
2208 if (uid_eq(current_fsuid(), i_uid))
2211 ns = current_user_ns();
2212 if (kuid_has_mapping(ns, i_uid) && ns_capable(ns, CAP_FOWNER))
2216 EXPORT_SYMBOL(inode_owner_or_capable);
2219 * Direct i/o helper functions
2221 static void __inode_dio_wait(struct inode *inode)
2223 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2224 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2227 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2228 if (atomic_read(&inode->i_dio_count))
2230 } while (atomic_read(&inode->i_dio_count));
2231 finish_wait(wq, &q.wq_entry);
2235 * inode_dio_wait - wait for outstanding DIO requests to finish
2236 * @inode: inode to wait for
2238 * Waits for all pending direct I/O requests to finish so that we can
2239 * proceed with a truncate or equivalent operation.
2241 * Must be called under a lock that serializes taking new references
2242 * to i_dio_count, usually by inode->i_mutex.
2244 void inode_dio_wait(struct inode *inode)
2246 if (atomic_read(&inode->i_dio_count))
2247 __inode_dio_wait(inode);
2249 EXPORT_SYMBOL(inode_dio_wait);
2252 * inode_set_flags - atomically set some inode flags
2254 * Note: the caller should be holding i_mutex, or else be sure that
2255 * they have exclusive access to the inode structure (i.e., while the
2256 * inode is being instantiated). The reason for the cmpxchg() loop
2257 * --- which wouldn't be necessary if all code paths which modify
2258 * i_flags actually followed this rule, is that there is at least one
2259 * code path which doesn't today so we use cmpxchg() out of an abundance
2262 * In the long run, i_mutex is overkill, and we should probably look
2263 * at using the i_lock spinlock to protect i_flags, and then make sure
2264 * it is so documented in include/linux/fs.h and that all code follows
2265 * the locking convention!!
2267 void inode_set_flags(struct inode *inode, unsigned int flags,
2270 WARN_ON_ONCE(flags & ~mask);
2271 set_mask_bits(&inode->i_flags, mask, flags);
2273 EXPORT_SYMBOL(inode_set_flags);
2275 void inode_nohighmem(struct inode *inode)
2277 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2279 EXPORT_SYMBOL(inode_nohighmem);
2282 * timestamp_truncate - Truncate timespec to a granularity
2284 * @inode: inode being updated
2286 * Truncate a timespec to the granularity supported by the fs
2287 * containing the inode. Always rounds down. gran must
2288 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2290 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2292 struct super_block *sb = inode->i_sb;
2293 unsigned int gran = sb->s_time_gran;
2295 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2296 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2299 /* Avoid division in the common cases 1 ns and 1 s. */
2302 else if (gran == NSEC_PER_SEC)
2304 else if (gran > 1 && gran < NSEC_PER_SEC)
2305 t.tv_nsec -= t.tv_nsec % gran;
2307 WARN(1, "invalid file time granularity: %u", gran);
2310 EXPORT_SYMBOL(timestamp_truncate);
2313 * current_time - Return FS time
2316 * Return the current time truncated to the time granularity supported by
2319 * Note that inode and inode->sb cannot be NULL.
2320 * Otherwise, the function warns and returns time without truncation.
2322 struct timespec64 current_time(struct inode *inode)
2324 struct timespec64 now;
2326 ktime_get_coarse_real_ts64(&now);
2328 if (unlikely(!inode->i_sb)) {
2329 WARN(1, "current_time() called with uninitialized super_block in the inode");
2333 return timestamp_truncate(now, inode);
2335 EXPORT_SYMBOL(current_time);