1 // SPDX-License-Identifier: GPL-2.0-only
3 * (C) 1997 Linus Torvalds
4 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
6 #include <linux/export.h>
9 #include <linux/backing-dev.h>
10 #include <linux/hash.h>
11 #include <linux/swap.h>
12 #include <linux/security.h>
13 #include <linux/cdev.h>
14 #include <linux/memblock.h>
15 #include <linux/fsnotify.h>
16 #include <linux/mount.h>
17 #include <linux/posix_acl.h>
18 #include <linux/prefetch.h>
19 #include <linux/buffer_head.h> /* for inode_has_buffers */
20 #include <linux/ratelimit.h>
21 #include <linux/list_lru.h>
22 #include <linux/iversion.h>
23 #include <trace/events/writeback.h>
27 * Inode locking rules:
29 * inode->i_lock protects:
30 * inode->i_state, inode->i_hash, __iget()
31 * Inode LRU list locks protect:
32 * inode->i_sb->s_inode_lru, inode->i_lru
33 * inode->i_sb->s_inode_list_lock protects:
34 * inode->i_sb->s_inodes, inode->i_sb_list
35 * bdi->wb.list_lock protects:
36 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
37 * inode_hash_lock protects:
38 * inode_hashtable, inode->i_hash
42 * inode->i_sb->s_inode_list_lock
44 * Inode LRU list locks
50 * inode->i_sb->s_inode_list_lock
57 static unsigned int i_hash_mask __read_mostly;
58 static unsigned int i_hash_shift __read_mostly;
59 static struct hlist_head *inode_hashtable __read_mostly;
60 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
63 * Empty aops. Can be used for the cases where the user does not
64 * define any of the address_space operations.
66 const struct address_space_operations empty_aops = {
68 EXPORT_SYMBOL(empty_aops);
71 * Statistics gathering..
73 struct inodes_stat_t inodes_stat;
75 static DEFINE_PER_CPU(unsigned long, nr_inodes);
76 static DEFINE_PER_CPU(unsigned long, nr_unused);
78 static struct kmem_cache *inode_cachep __read_mostly;
80 static long get_nr_inodes(void)
84 for_each_possible_cpu(i)
85 sum += per_cpu(nr_inodes, i);
86 return sum < 0 ? 0 : sum;
89 static inline long get_nr_inodes_unused(void)
93 for_each_possible_cpu(i)
94 sum += per_cpu(nr_unused, i);
95 return sum < 0 ? 0 : sum;
98 long get_nr_dirty_inodes(void)
100 /* not actually dirty inodes, but a wild approximation */
101 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
102 return nr_dirty > 0 ? nr_dirty : 0;
106 * Handle nr_inode sysctl
109 int proc_nr_inodes(struct ctl_table *table, int write,
110 void *buffer, size_t *lenp, loff_t *ppos)
112 inodes_stat.nr_inodes = get_nr_inodes();
113 inodes_stat.nr_unused = get_nr_inodes_unused();
114 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
118 static int no_open(struct inode *inode, struct file *file)
124 * inode_init_always - perform inode structure initialisation
125 * @sb: superblock inode belongs to
126 * @inode: inode to initialise
128 * These are initializations that need to be done on every inode
129 * allocation as the fields are not initialised by slab allocation.
131 int inode_init_always(struct super_block *sb, struct inode *inode)
133 static const struct inode_operations empty_iops;
134 static const struct file_operations no_open_fops = {.open = no_open};
135 struct address_space *const mapping = &inode->i_data;
138 inode->i_blkbits = sb->s_blocksize_bits;
140 atomic64_set(&inode->i_sequence, 0);
141 atomic_set(&inode->i_count, 1);
142 inode->i_op = &empty_iops;
143 inode->i_fop = &no_open_fops;
145 inode->__i_nlink = 1;
146 inode->i_opflags = 0;
148 inode->i_opflags |= IOP_XATTR;
149 i_uid_write(inode, 0);
150 i_gid_write(inode, 0);
151 atomic_set(&inode->i_writecount, 0);
153 inode->i_write_hint = WRITE_LIFE_NOT_SET;
156 inode->i_generation = 0;
157 inode->i_pipe = NULL;
158 inode->i_cdev = NULL;
159 inode->i_link = NULL;
160 inode->i_dir_seq = 0;
162 inode->dirtied_when = 0;
164 #ifdef CONFIG_CGROUP_WRITEBACK
165 inode->i_wb_frn_winner = 0;
166 inode->i_wb_frn_avg_time = 0;
167 inode->i_wb_frn_history = 0;
170 if (security_inode_alloc(inode))
172 spin_lock_init(&inode->i_lock);
173 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
175 init_rwsem(&inode->i_rwsem);
176 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
178 atomic_set(&inode->i_dio_count, 0);
180 mapping->a_ops = &empty_aops;
181 mapping->host = inode;
183 if (sb->s_type->fs_flags & FS_THP_SUPPORT)
184 __set_bit(AS_THP_SUPPORT, &mapping->flags);
186 atomic_set(&mapping->i_mmap_writable, 0);
187 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
188 atomic_set(&mapping->nr_thps, 0);
190 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
191 mapping->private_data = NULL;
192 mapping->writeback_index = 0;
193 __init_rwsem(&mapping->invalidate_lock, "mapping.invalidate_lock",
194 &sb->s_type->invalidate_lock_key);
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_lru_list_add(struct inode *inode)
431 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
432 this_cpu_inc(nr_unused);
434 inode->i_state |= I_REFERENCED;
438 * Add inode to LRU if needed (inode is unused and clean).
440 * Needs inode->i_lock held.
442 void inode_add_lru(struct inode *inode)
444 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
445 I_FREEING | I_WILL_FREE)) &&
446 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
447 inode_lru_list_add(inode);
451 static void inode_lru_list_del(struct inode *inode)
454 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
455 this_cpu_dec(nr_unused);
459 * inode_sb_list_add - add inode to the superblock list of inodes
460 * @inode: inode to add
462 void inode_sb_list_add(struct inode *inode)
464 spin_lock(&inode->i_sb->s_inode_list_lock);
465 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
466 spin_unlock(&inode->i_sb->s_inode_list_lock);
468 EXPORT_SYMBOL_GPL(inode_sb_list_add);
470 static inline void inode_sb_list_del(struct inode *inode)
472 if (!list_empty(&inode->i_sb_list)) {
473 spin_lock(&inode->i_sb->s_inode_list_lock);
474 list_del_init(&inode->i_sb_list);
475 spin_unlock(&inode->i_sb->s_inode_list_lock);
479 static unsigned long hash(struct super_block *sb, unsigned long hashval)
483 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
485 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
486 return tmp & i_hash_mask;
490 * __insert_inode_hash - hash an inode
491 * @inode: unhashed inode
492 * @hashval: unsigned long value used to locate this object in the
495 * Add an inode to the inode hash for this superblock.
497 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
499 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
501 spin_lock(&inode_hash_lock);
502 spin_lock(&inode->i_lock);
503 hlist_add_head_rcu(&inode->i_hash, b);
504 spin_unlock(&inode->i_lock);
505 spin_unlock(&inode_hash_lock);
507 EXPORT_SYMBOL(__insert_inode_hash);
510 * __remove_inode_hash - remove an inode from the hash
511 * @inode: inode to unhash
513 * Remove an inode from the superblock.
515 void __remove_inode_hash(struct inode *inode)
517 spin_lock(&inode_hash_lock);
518 spin_lock(&inode->i_lock);
519 hlist_del_init_rcu(&inode->i_hash);
520 spin_unlock(&inode->i_lock);
521 spin_unlock(&inode_hash_lock);
523 EXPORT_SYMBOL(__remove_inode_hash);
525 void clear_inode(struct inode *inode)
528 * We have to cycle the i_pages lock here because reclaim can be in the
529 * process of removing the last page (in __delete_from_page_cache())
530 * and we must not free the mapping under it.
532 xa_lock_irq(&inode->i_data.i_pages);
533 BUG_ON(inode->i_data.nrpages);
535 * Almost always, mapping_empty(&inode->i_data) here; but there are
536 * two known and long-standing ways in which nodes may get left behind
537 * (when deep radix-tree node allocation failed partway; or when THP
538 * collapse_file() failed). Until those two known cases are cleaned up,
539 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
540 * nor even WARN_ON(!mapping_empty).
542 xa_unlock_irq(&inode->i_data.i_pages);
543 BUG_ON(!list_empty(&inode->i_data.private_list));
544 BUG_ON(!(inode->i_state & I_FREEING));
545 BUG_ON(inode->i_state & I_CLEAR);
546 BUG_ON(!list_empty(&inode->i_wb_list));
547 /* don't need i_lock here, no concurrent mods to i_state */
548 inode->i_state = I_FREEING | I_CLEAR;
550 EXPORT_SYMBOL(clear_inode);
553 * Free the inode passed in, removing it from the lists it is still connected
554 * to. We remove any pages still attached to the inode and wait for any IO that
555 * is still in progress before finally destroying the inode.
557 * An inode must already be marked I_FREEING so that we avoid the inode being
558 * moved back onto lists if we race with other code that manipulates the lists
559 * (e.g. writeback_single_inode). The caller is responsible for setting this.
561 * An inode must already be removed from the LRU list before being evicted from
562 * the cache. This should occur atomically with setting the I_FREEING state
563 * flag, so no inodes here should ever be on the LRU when being evicted.
565 static void evict(struct inode *inode)
567 const struct super_operations *op = inode->i_sb->s_op;
569 BUG_ON(!(inode->i_state & I_FREEING));
570 BUG_ON(!list_empty(&inode->i_lru));
572 if (!list_empty(&inode->i_io_list))
573 inode_io_list_del(inode);
575 inode_sb_list_del(inode);
578 * Wait for flusher thread to be done with the inode so that filesystem
579 * does not start destroying it while writeback is still running. Since
580 * the inode has I_FREEING set, flusher thread won't start new work on
581 * the inode. We just have to wait for running writeback to finish.
583 inode_wait_for_writeback(inode);
585 if (op->evict_inode) {
586 op->evict_inode(inode);
588 truncate_inode_pages_final(&inode->i_data);
591 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
594 remove_inode_hash(inode);
596 spin_lock(&inode->i_lock);
597 wake_up_bit(&inode->i_state, __I_NEW);
598 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
599 spin_unlock(&inode->i_lock);
601 destroy_inode(inode);
605 * dispose_list - dispose of the contents of a local list
606 * @head: the head of the list to free
608 * Dispose-list gets a local list with local inodes in it, so it doesn't
609 * need to worry about list corruption and SMP locks.
611 static void dispose_list(struct list_head *head)
613 while (!list_empty(head)) {
616 inode = list_first_entry(head, struct inode, i_lru);
617 list_del_init(&inode->i_lru);
625 * evict_inodes - evict all evictable inodes for a superblock
626 * @sb: superblock to operate on
628 * Make sure that no inodes with zero refcount are retained. This is
629 * called by superblock shutdown after having SB_ACTIVE flag removed,
630 * so any inode reaching zero refcount during or after that call will
631 * be immediately evicted.
633 void evict_inodes(struct super_block *sb)
635 struct inode *inode, *next;
639 spin_lock(&sb->s_inode_list_lock);
640 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
641 if (atomic_read(&inode->i_count))
644 spin_lock(&inode->i_lock);
645 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
646 spin_unlock(&inode->i_lock);
650 inode->i_state |= I_FREEING;
651 inode_lru_list_del(inode);
652 spin_unlock(&inode->i_lock);
653 list_add(&inode->i_lru, &dispose);
656 * We can have a ton of inodes to evict at unmount time given
657 * enough memory, check to see if we need to go to sleep for a
658 * bit so we don't livelock.
660 if (need_resched()) {
661 spin_unlock(&sb->s_inode_list_lock);
663 dispose_list(&dispose);
667 spin_unlock(&sb->s_inode_list_lock);
669 dispose_list(&dispose);
671 EXPORT_SYMBOL_GPL(evict_inodes);
674 * invalidate_inodes - attempt to free all inodes on a superblock
675 * @sb: superblock to operate on
676 * @kill_dirty: flag to guide handling of dirty inodes
678 * Attempts to free all inodes for a given superblock. If there were any
679 * busy inodes return a non-zero value, else zero.
680 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
683 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
686 struct inode *inode, *next;
690 spin_lock(&sb->s_inode_list_lock);
691 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
692 spin_lock(&inode->i_lock);
693 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
694 spin_unlock(&inode->i_lock);
697 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
698 spin_unlock(&inode->i_lock);
702 if (atomic_read(&inode->i_count)) {
703 spin_unlock(&inode->i_lock);
708 inode->i_state |= I_FREEING;
709 inode_lru_list_del(inode);
710 spin_unlock(&inode->i_lock);
711 list_add(&inode->i_lru, &dispose);
712 if (need_resched()) {
713 spin_unlock(&sb->s_inode_list_lock);
715 dispose_list(&dispose);
719 spin_unlock(&sb->s_inode_list_lock);
721 dispose_list(&dispose);
727 * Isolate the inode from the LRU in preparation for freeing it.
729 * Any inodes which are pinned purely because of attached pagecache have their
730 * pagecache removed. If the inode has metadata buffers attached to
731 * mapping->private_list then try to remove them.
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 trylock.
749 * If we fail to get the lock, just skip it.
751 if (!spin_trylock(&inode->i_lock))
755 * Referenced or dirty inodes are still in use. Give them another pass
756 * through the LRU as we canot reclaim them now.
758 if (atomic_read(&inode->i_count) ||
759 (inode->i_state & ~I_REFERENCED)) {
760 list_lru_isolate(lru, &inode->i_lru);
761 spin_unlock(&inode->i_lock);
762 this_cpu_dec(nr_unused);
766 /* recently referenced inodes get one more pass */
767 if (inode->i_state & I_REFERENCED) {
768 inode->i_state &= ~I_REFERENCED;
769 spin_unlock(&inode->i_lock);
773 if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
775 spin_unlock(&inode->i_lock);
776 spin_unlock(lru_lock);
777 if (remove_inode_buffers(inode)) {
779 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
780 if (current_is_kswapd())
781 __count_vm_events(KSWAPD_INODESTEAL, reap);
783 __count_vm_events(PGINODESTEAL, reap);
784 if (current->reclaim_state)
785 current->reclaim_state->reclaimed_slab += reap;
792 WARN_ON(inode->i_state & I_NEW);
793 inode->i_state |= I_FREEING;
794 list_lru_isolate_move(lru, &inode->i_lru, freeable);
795 spin_unlock(&inode->i_lock);
797 this_cpu_dec(nr_unused);
802 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
803 * This is called from the superblock shrinker function with a number of inodes
804 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
805 * then are freed outside inode_lock by dispose_list().
807 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
812 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
813 inode_lru_isolate, &freeable);
814 dispose_list(&freeable);
818 static void __wait_on_freeing_inode(struct inode *inode);
820 * Called with the inode lock held.
822 static struct inode *find_inode(struct super_block *sb,
823 struct hlist_head *head,
824 int (*test)(struct inode *, void *),
827 struct inode *inode = NULL;
830 hlist_for_each_entry(inode, head, i_hash) {
831 if (inode->i_sb != sb)
833 if (!test(inode, data))
835 spin_lock(&inode->i_lock);
836 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
837 __wait_on_freeing_inode(inode);
840 if (unlikely(inode->i_state & I_CREATING)) {
841 spin_unlock(&inode->i_lock);
842 return ERR_PTR(-ESTALE);
845 spin_unlock(&inode->i_lock);
852 * find_inode_fast is the fast path version of find_inode, see the comment at
853 * iget_locked for details.
855 static struct inode *find_inode_fast(struct super_block *sb,
856 struct hlist_head *head, unsigned long ino)
858 struct inode *inode = NULL;
861 hlist_for_each_entry(inode, head, i_hash) {
862 if (inode->i_ino != ino)
864 if (inode->i_sb != sb)
866 spin_lock(&inode->i_lock);
867 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
868 __wait_on_freeing_inode(inode);
871 if (unlikely(inode->i_state & I_CREATING)) {
872 spin_unlock(&inode->i_lock);
873 return ERR_PTR(-ESTALE);
876 spin_unlock(&inode->i_lock);
883 * Each cpu owns a range of LAST_INO_BATCH numbers.
884 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
885 * to renew the exhausted range.
887 * This does not significantly increase overflow rate because every CPU can
888 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
889 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
890 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
891 * overflow rate by 2x, which does not seem too significant.
893 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
894 * error if st_ino won't fit in target struct field. Use 32bit counter
895 * here to attempt to avoid that.
897 #define LAST_INO_BATCH 1024
898 static DEFINE_PER_CPU(unsigned int, last_ino);
900 unsigned int get_next_ino(void)
902 unsigned int *p = &get_cpu_var(last_ino);
903 unsigned int res = *p;
906 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
907 static atomic_t shared_last_ino;
908 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
910 res = next - LAST_INO_BATCH;
915 /* get_next_ino should not provide a 0 inode number */
919 put_cpu_var(last_ino);
922 EXPORT_SYMBOL(get_next_ino);
925 * new_inode_pseudo - obtain an inode
928 * Allocates a new inode for given superblock.
929 * Inode wont be chained in superblock s_inodes list
931 * - fs can't be unmount
932 * - quotas, fsnotify, writeback can't work
934 struct inode *new_inode_pseudo(struct super_block *sb)
936 struct inode *inode = alloc_inode(sb);
939 spin_lock(&inode->i_lock);
941 spin_unlock(&inode->i_lock);
942 INIT_LIST_HEAD(&inode->i_sb_list);
948 * new_inode - obtain an inode
951 * Allocates a new inode for given superblock. The default gfp_mask
952 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
953 * If HIGHMEM pages are unsuitable or it is known that pages allocated
954 * for the page cache are not reclaimable or migratable,
955 * mapping_set_gfp_mask() must be called with suitable flags on the
956 * newly created inode's mapping
959 struct inode *new_inode(struct super_block *sb)
963 spin_lock_prefetch(&sb->s_inode_list_lock);
965 inode = new_inode_pseudo(sb);
967 inode_sb_list_add(inode);
970 EXPORT_SYMBOL(new_inode);
972 #ifdef CONFIG_DEBUG_LOCK_ALLOC
973 void lockdep_annotate_inode_mutex_key(struct inode *inode)
975 if (S_ISDIR(inode->i_mode)) {
976 struct file_system_type *type = inode->i_sb->s_type;
978 /* Set new key only if filesystem hasn't already changed it */
979 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
981 * ensure nobody is actually holding i_mutex
983 // mutex_destroy(&inode->i_mutex);
984 init_rwsem(&inode->i_rwsem);
985 lockdep_set_class(&inode->i_rwsem,
986 &type->i_mutex_dir_key);
990 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
994 * unlock_new_inode - clear the I_NEW state and wake up any waiters
995 * @inode: new inode to unlock
997 * Called when the inode is fully initialised to clear the new state of the
998 * inode and wake up anyone waiting for the inode to finish initialisation.
1000 void unlock_new_inode(struct inode *inode)
1002 lockdep_annotate_inode_mutex_key(inode);
1003 spin_lock(&inode->i_lock);
1004 WARN_ON(!(inode->i_state & I_NEW));
1005 inode->i_state &= ~I_NEW & ~I_CREATING;
1007 wake_up_bit(&inode->i_state, __I_NEW);
1008 spin_unlock(&inode->i_lock);
1010 EXPORT_SYMBOL(unlock_new_inode);
1012 void discard_new_inode(struct inode *inode)
1014 lockdep_annotate_inode_mutex_key(inode);
1015 spin_lock(&inode->i_lock);
1016 WARN_ON(!(inode->i_state & I_NEW));
1017 inode->i_state &= ~I_NEW;
1019 wake_up_bit(&inode->i_state, __I_NEW);
1020 spin_unlock(&inode->i_lock);
1023 EXPORT_SYMBOL(discard_new_inode);
1026 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1028 * Lock any non-NULL argument that is not a directory.
1029 * Zero, one or two objects may be locked by this function.
1031 * @inode1: first inode to lock
1032 * @inode2: second inode to lock
1034 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1036 if (inode1 > inode2)
1037 swap(inode1, inode2);
1039 if (inode1 && !S_ISDIR(inode1->i_mode))
1041 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1042 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1044 EXPORT_SYMBOL(lock_two_nondirectories);
1047 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1048 * @inode1: first inode to unlock
1049 * @inode2: second inode to unlock
1051 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1053 if (inode1 && !S_ISDIR(inode1->i_mode))
1054 inode_unlock(inode1);
1055 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1056 inode_unlock(inode2);
1058 EXPORT_SYMBOL(unlock_two_nondirectories);
1061 * inode_insert5 - obtain an inode from a mounted file system
1062 * @inode: pre-allocated inode to use for insert to cache
1063 * @hashval: hash value (usually inode number) to get
1064 * @test: callback used for comparisons between inodes
1065 * @set: callback used to initialize a new struct inode
1066 * @data: opaque data pointer to pass to @test and @set
1068 * Search for the inode specified by @hashval and @data in the inode cache,
1069 * and if present it is return it with an increased reference count. This is
1070 * a variant of iget5_locked() for callers that don't want to fail on memory
1071 * allocation of inode.
1073 * If the inode is not in cache, insert the pre-allocated inode to cache and
1074 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1075 * to fill it in before unlocking it via unlock_new_inode().
1077 * Note both @test and @set are called with the inode_hash_lock held, so can't
1080 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1081 int (*test)(struct inode *, void *),
1082 int (*set)(struct inode *, void *), void *data)
1084 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1086 bool creating = inode->i_state & I_CREATING;
1089 spin_lock(&inode_hash_lock);
1090 old = find_inode(inode->i_sb, head, test, data);
1091 if (unlikely(old)) {
1093 * Uhhuh, somebody else created the same inode under us.
1094 * Use the old inode instead of the preallocated one.
1096 spin_unlock(&inode_hash_lock);
1100 if (unlikely(inode_unhashed(old))) {
1107 if (set && unlikely(set(inode, data))) {
1113 * Return the locked inode with I_NEW set, the
1114 * caller is responsible for filling in the contents
1116 spin_lock(&inode->i_lock);
1117 inode->i_state |= I_NEW;
1118 hlist_add_head_rcu(&inode->i_hash, head);
1119 spin_unlock(&inode->i_lock);
1121 inode_sb_list_add(inode);
1123 spin_unlock(&inode_hash_lock);
1127 EXPORT_SYMBOL(inode_insert5);
1130 * iget5_locked - obtain an inode from a mounted file system
1131 * @sb: super block of file system
1132 * @hashval: hash value (usually inode number) to get
1133 * @test: callback used for comparisons between inodes
1134 * @set: callback used to initialize a new struct inode
1135 * @data: opaque data pointer to pass to @test and @set
1137 * Search for the inode specified by @hashval and @data in the inode cache,
1138 * and if present it is return it with an increased reference count. This is
1139 * a generalized version of iget_locked() for file systems where the inode
1140 * number is not sufficient for unique identification of an inode.
1142 * If the inode is not in cache, allocate a new inode and return it locked,
1143 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1144 * before unlocking it via unlock_new_inode().
1146 * Note both @test and @set are called with the inode_hash_lock held, so can't
1149 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1150 int (*test)(struct inode *, void *),
1151 int (*set)(struct inode *, void *), void *data)
1153 struct inode *inode = ilookup5(sb, hashval, test, data);
1156 struct inode *new = alloc_inode(sb);
1160 inode = inode_insert5(new, hashval, test, set, data);
1161 if (unlikely(inode != new))
1167 EXPORT_SYMBOL(iget5_locked);
1170 * iget_locked - obtain an inode from a mounted file system
1171 * @sb: super block of file system
1172 * @ino: inode number to get
1174 * Search for the inode specified by @ino in the inode cache and if present
1175 * return it with an increased reference count. This is for file systems
1176 * where the inode number is sufficient for unique identification of an inode.
1178 * If the inode is not in cache, allocate a new inode and return it locked,
1179 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1180 * before unlocking it via unlock_new_inode().
1182 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1184 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1185 struct inode *inode;
1187 spin_lock(&inode_hash_lock);
1188 inode = find_inode_fast(sb, head, ino);
1189 spin_unlock(&inode_hash_lock);
1193 wait_on_inode(inode);
1194 if (unlikely(inode_unhashed(inode))) {
1201 inode = alloc_inode(sb);
1205 spin_lock(&inode_hash_lock);
1206 /* We released the lock, so.. */
1207 old = find_inode_fast(sb, head, ino);
1210 spin_lock(&inode->i_lock);
1211 inode->i_state = I_NEW;
1212 hlist_add_head_rcu(&inode->i_hash, head);
1213 spin_unlock(&inode->i_lock);
1214 inode_sb_list_add(inode);
1215 spin_unlock(&inode_hash_lock);
1217 /* Return the locked inode with I_NEW set, the
1218 * caller is responsible for filling in the contents
1224 * Uhhuh, somebody else created the same inode under
1225 * us. Use the old inode instead of the one we just
1228 spin_unlock(&inode_hash_lock);
1229 destroy_inode(inode);
1233 wait_on_inode(inode);
1234 if (unlikely(inode_unhashed(inode))) {
1241 EXPORT_SYMBOL(iget_locked);
1244 * search the inode cache for a matching inode number.
1245 * If we find one, then the inode number we are trying to
1246 * allocate is not unique and so we should not use it.
1248 * Returns 1 if the inode number is unique, 0 if it is not.
1250 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1252 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1253 struct inode *inode;
1255 hlist_for_each_entry_rcu(inode, b, i_hash) {
1256 if (inode->i_ino == ino && inode->i_sb == sb)
1263 * iunique - get a unique inode number
1265 * @max_reserved: highest reserved inode number
1267 * Obtain an inode number that is unique on the system for a given
1268 * superblock. This is used by file systems that have no natural
1269 * permanent inode numbering system. An inode number is returned that
1270 * is higher than the reserved limit but unique.
1273 * With a large number of inodes live on the file system this function
1274 * currently becomes quite slow.
1276 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1279 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1280 * error if st_ino won't fit in target struct field. Use 32bit counter
1281 * here to attempt to avoid that.
1283 static DEFINE_SPINLOCK(iunique_lock);
1284 static unsigned int counter;
1288 spin_lock(&iunique_lock);
1290 if (counter <= max_reserved)
1291 counter = max_reserved + 1;
1293 } while (!test_inode_iunique(sb, res));
1294 spin_unlock(&iunique_lock);
1299 EXPORT_SYMBOL(iunique);
1301 struct inode *igrab(struct inode *inode)
1303 spin_lock(&inode->i_lock);
1304 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1306 spin_unlock(&inode->i_lock);
1308 spin_unlock(&inode->i_lock);
1310 * Handle the case where s_op->clear_inode is not been
1311 * called yet, and somebody is calling igrab
1312 * while the inode is getting freed.
1318 EXPORT_SYMBOL(igrab);
1321 * ilookup5_nowait - search for an inode in the inode cache
1322 * @sb: super block of file system to search
1323 * @hashval: hash value (usually inode number) to search for
1324 * @test: callback used for comparisons between inodes
1325 * @data: opaque data pointer to pass to @test
1327 * Search for the inode specified by @hashval and @data in the inode cache.
1328 * If the inode is in the cache, the inode is returned with an incremented
1331 * Note: I_NEW is not waited upon so you have to be very careful what you do
1332 * with the returned inode. You probably should be using ilookup5() instead.
1334 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1336 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1337 int (*test)(struct inode *, void *), void *data)
1339 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1340 struct inode *inode;
1342 spin_lock(&inode_hash_lock);
1343 inode = find_inode(sb, head, test, data);
1344 spin_unlock(&inode_hash_lock);
1346 return IS_ERR(inode) ? NULL : inode;
1348 EXPORT_SYMBOL(ilookup5_nowait);
1351 * ilookup5 - search for an inode in the inode cache
1352 * @sb: super block of file system to search
1353 * @hashval: hash value (usually inode number) to search for
1354 * @test: callback used for comparisons between inodes
1355 * @data: opaque data pointer to pass to @test
1357 * Search for the inode specified by @hashval and @data in the inode cache,
1358 * and if the inode is in the cache, return the inode with an incremented
1359 * reference count. Waits on I_NEW before returning the inode.
1360 * returned with an incremented reference count.
1362 * This is a generalized version of ilookup() for file systems where the
1363 * inode number is not sufficient for unique identification of an inode.
1365 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1367 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1368 int (*test)(struct inode *, void *), void *data)
1370 struct inode *inode;
1372 inode = ilookup5_nowait(sb, hashval, test, data);
1374 wait_on_inode(inode);
1375 if (unlikely(inode_unhashed(inode))) {
1382 EXPORT_SYMBOL(ilookup5);
1385 * ilookup - search for an inode in the inode cache
1386 * @sb: super block of file system to search
1387 * @ino: inode number to search for
1389 * Search for the inode @ino in the inode cache, and if the inode is in the
1390 * cache, the inode is returned with an incremented reference count.
1392 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1394 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1395 struct inode *inode;
1397 spin_lock(&inode_hash_lock);
1398 inode = find_inode_fast(sb, head, ino);
1399 spin_unlock(&inode_hash_lock);
1404 wait_on_inode(inode);
1405 if (unlikely(inode_unhashed(inode))) {
1412 EXPORT_SYMBOL(ilookup);
1415 * find_inode_nowait - find an inode in the inode cache
1416 * @sb: super block of file system to search
1417 * @hashval: hash value (usually inode number) to search for
1418 * @match: callback used for comparisons between inodes
1419 * @data: opaque data pointer to pass to @match
1421 * Search for the inode specified by @hashval and @data in the inode
1422 * cache, where the helper function @match will return 0 if the inode
1423 * does not match, 1 if the inode does match, and -1 if the search
1424 * should be stopped. The @match function must be responsible for
1425 * taking the i_lock spin_lock and checking i_state for an inode being
1426 * freed or being initialized, and incrementing the reference count
1427 * before returning 1. It also must not sleep, since it is called with
1428 * the inode_hash_lock spinlock held.
1430 * This is a even more generalized version of ilookup5() when the
1431 * function must never block --- find_inode() can block in
1432 * __wait_on_freeing_inode() --- or when the caller can not increment
1433 * the reference count because the resulting iput() might cause an
1434 * inode eviction. The tradeoff is that the @match funtion must be
1435 * very carefully implemented.
1437 struct inode *find_inode_nowait(struct super_block *sb,
1438 unsigned long hashval,
1439 int (*match)(struct inode *, unsigned long,
1443 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1444 struct inode *inode, *ret_inode = NULL;
1447 spin_lock(&inode_hash_lock);
1448 hlist_for_each_entry(inode, head, i_hash) {
1449 if (inode->i_sb != sb)
1451 mval = match(inode, hashval, data);
1459 spin_unlock(&inode_hash_lock);
1462 EXPORT_SYMBOL(find_inode_nowait);
1465 * find_inode_rcu - find an inode in the inode cache
1466 * @sb: Super block of file system to search
1467 * @hashval: Key to hash
1468 * @test: Function to test match on an inode
1469 * @data: Data for test function
1471 * Search for the inode specified by @hashval and @data in the inode cache,
1472 * where the helper function @test will return 0 if the inode does not match
1473 * and 1 if it does. The @test function must be responsible for taking the
1474 * i_lock spin_lock and checking i_state for an inode being freed or being
1477 * If successful, this will return the inode for which the @test function
1478 * returned 1 and NULL otherwise.
1480 * The @test function is not permitted to take a ref on any inode presented.
1481 * It is also not permitted to sleep.
1483 * The caller must hold the RCU read lock.
1485 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1486 int (*test)(struct inode *, void *), void *data)
1488 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1489 struct inode *inode;
1491 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1492 "suspicious find_inode_rcu() usage");
1494 hlist_for_each_entry_rcu(inode, head, i_hash) {
1495 if (inode->i_sb == sb &&
1496 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1502 EXPORT_SYMBOL(find_inode_rcu);
1505 * find_inode_by_ino_rcu - Find an inode in the inode cache
1506 * @sb: Super block of file system to search
1507 * @ino: The inode number to match
1509 * Search for the inode specified by @hashval and @data in the inode cache,
1510 * where the helper function @test will return 0 if the inode does not match
1511 * and 1 if it does. The @test function must be responsible for taking the
1512 * i_lock spin_lock and checking i_state for an inode being freed or being
1515 * If successful, this will return the inode for which the @test function
1516 * returned 1 and NULL otherwise.
1518 * The @test function is not permitted to take a ref on any inode presented.
1519 * It is also not permitted to sleep.
1521 * The caller must hold the RCU read lock.
1523 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1526 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1527 struct inode *inode;
1529 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1530 "suspicious find_inode_by_ino_rcu() usage");
1532 hlist_for_each_entry_rcu(inode, head, i_hash) {
1533 if (inode->i_ino == ino &&
1534 inode->i_sb == sb &&
1535 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1540 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1542 int insert_inode_locked(struct inode *inode)
1544 struct super_block *sb = inode->i_sb;
1545 ino_t ino = inode->i_ino;
1546 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1549 struct inode *old = NULL;
1550 spin_lock(&inode_hash_lock);
1551 hlist_for_each_entry(old, head, i_hash) {
1552 if (old->i_ino != ino)
1554 if (old->i_sb != sb)
1556 spin_lock(&old->i_lock);
1557 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1558 spin_unlock(&old->i_lock);
1564 spin_lock(&inode->i_lock);
1565 inode->i_state |= I_NEW | I_CREATING;
1566 hlist_add_head_rcu(&inode->i_hash, head);
1567 spin_unlock(&inode->i_lock);
1568 spin_unlock(&inode_hash_lock);
1571 if (unlikely(old->i_state & I_CREATING)) {
1572 spin_unlock(&old->i_lock);
1573 spin_unlock(&inode_hash_lock);
1577 spin_unlock(&old->i_lock);
1578 spin_unlock(&inode_hash_lock);
1580 if (unlikely(!inode_unhashed(old))) {
1587 EXPORT_SYMBOL(insert_inode_locked);
1589 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1590 int (*test)(struct inode *, void *), void *data)
1594 inode->i_state |= I_CREATING;
1595 old = inode_insert5(inode, hashval, test, NULL, data);
1603 EXPORT_SYMBOL(insert_inode_locked4);
1606 int generic_delete_inode(struct inode *inode)
1610 EXPORT_SYMBOL(generic_delete_inode);
1613 * Called when we're dropping the last reference
1616 * Call the FS "drop_inode()" function, defaulting to
1617 * the legacy UNIX filesystem behaviour. If it tells
1618 * us to evict inode, do so. Otherwise, retain inode
1619 * in cache if fs is alive, sync and evict if fs is
1622 static void iput_final(struct inode *inode)
1624 struct super_block *sb = inode->i_sb;
1625 const struct super_operations *op = inode->i_sb->s_op;
1626 unsigned long state;
1629 WARN_ON(inode->i_state & I_NEW);
1632 drop = op->drop_inode(inode);
1634 drop = generic_drop_inode(inode);
1637 !(inode->i_state & I_DONTCACHE) &&
1638 (sb->s_flags & SB_ACTIVE)) {
1639 inode_add_lru(inode);
1640 spin_unlock(&inode->i_lock);
1644 state = inode->i_state;
1646 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1647 spin_unlock(&inode->i_lock);
1649 write_inode_now(inode, 1);
1651 spin_lock(&inode->i_lock);
1652 state = inode->i_state;
1653 WARN_ON(state & I_NEW);
1654 state &= ~I_WILL_FREE;
1657 WRITE_ONCE(inode->i_state, state | I_FREEING);
1658 if (!list_empty(&inode->i_lru))
1659 inode_lru_list_del(inode);
1660 spin_unlock(&inode->i_lock);
1666 * iput - put an inode
1667 * @inode: inode to put
1669 * Puts an inode, dropping its usage count. If the inode use count hits
1670 * zero, the inode is then freed and may also be destroyed.
1672 * Consequently, iput() can sleep.
1674 void iput(struct inode *inode)
1678 BUG_ON(inode->i_state & I_CLEAR);
1680 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1681 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1682 atomic_inc(&inode->i_count);
1683 spin_unlock(&inode->i_lock);
1684 trace_writeback_lazytime_iput(inode);
1685 mark_inode_dirty_sync(inode);
1691 EXPORT_SYMBOL(iput);
1695 * bmap - find a block number in a file
1696 * @inode: inode owning the block number being requested
1697 * @block: pointer containing the block to find
1699 * Replaces the value in ``*block`` with the block number on the device holding
1700 * corresponding to the requested block number in the file.
1701 * That is, asked for block 4 of inode 1 the function will replace the
1702 * 4 in ``*block``, with disk block relative to the disk start that holds that
1703 * block of the file.
1705 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1706 * hole, returns 0 and ``*block`` is also set to 0.
1708 int bmap(struct inode *inode, sector_t *block)
1710 if (!inode->i_mapping->a_ops->bmap)
1713 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1716 EXPORT_SYMBOL(bmap);
1720 * With relative atime, only update atime if the previous atime is
1721 * earlier than either the ctime or mtime or if at least a day has
1722 * passed since the last atime update.
1724 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1725 struct timespec64 now)
1728 if (!(mnt->mnt_flags & MNT_RELATIME))
1731 * Is mtime younger than atime? If yes, update atime:
1733 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1736 * Is ctime younger than atime? If yes, update atime:
1738 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1742 * Is the previous atime value older than a day? If yes,
1745 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1748 * Good, we can skip the atime update:
1753 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1755 int dirty_flags = 0;
1757 if (flags & (S_ATIME | S_CTIME | S_MTIME)) {
1758 if (flags & S_ATIME)
1759 inode->i_atime = *time;
1760 if (flags & S_CTIME)
1761 inode->i_ctime = *time;
1762 if (flags & S_MTIME)
1763 inode->i_mtime = *time;
1765 if (inode->i_sb->s_flags & SB_LAZYTIME)
1766 dirty_flags |= I_DIRTY_TIME;
1768 dirty_flags |= I_DIRTY_SYNC;
1771 if ((flags & S_VERSION) && inode_maybe_inc_iversion(inode, false))
1772 dirty_flags |= I_DIRTY_SYNC;
1774 __mark_inode_dirty(inode, dirty_flags);
1777 EXPORT_SYMBOL(generic_update_time);
1780 * This does the actual work of updating an inodes time or version. Must have
1781 * had called mnt_want_write() before calling this.
1783 static int update_time(struct inode *inode, struct timespec64 *time, int flags)
1785 if (inode->i_op->update_time)
1786 return inode->i_op->update_time(inode, time, flags);
1787 return generic_update_time(inode, time, flags);
1791 * atime_needs_update - update the access time
1792 * @path: the &struct path to update
1793 * @inode: inode to update
1795 * Update the accessed time on an inode and mark it for writeback.
1796 * This function automatically handles read only file systems and media,
1797 * as well as the "noatime" flag and inode specific "noatime" markers.
1799 bool atime_needs_update(const struct path *path, struct inode *inode)
1801 struct vfsmount *mnt = path->mnt;
1802 struct timespec64 now;
1804 if (inode->i_flags & S_NOATIME)
1807 /* Atime updates will likely cause i_uid and i_gid to be written
1808 * back improprely if their true value is unknown to the vfs.
1810 if (HAS_UNMAPPED_ID(mnt_user_ns(mnt), inode))
1813 if (IS_NOATIME(inode))
1815 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1818 if (mnt->mnt_flags & MNT_NOATIME)
1820 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1823 now = current_time(inode);
1825 if (!relatime_need_update(mnt, inode, now))
1828 if (timespec64_equal(&inode->i_atime, &now))
1834 void touch_atime(const struct path *path)
1836 struct vfsmount *mnt = path->mnt;
1837 struct inode *inode = d_inode(path->dentry);
1838 struct timespec64 now;
1840 if (!atime_needs_update(path, inode))
1843 if (!sb_start_write_trylock(inode->i_sb))
1846 if (__mnt_want_write(mnt) != 0)
1849 * File systems can error out when updating inodes if they need to
1850 * allocate new space to modify an inode (such is the case for
1851 * Btrfs), but since we touch atime while walking down the path we
1852 * really don't care if we failed to update the atime of the file,
1853 * so just ignore the return value.
1854 * We may also fail on filesystems that have the ability to make parts
1855 * of the fs read only, e.g. subvolumes in Btrfs.
1857 now = current_time(inode);
1858 update_time(inode, &now, S_ATIME);
1859 __mnt_drop_write(mnt);
1861 sb_end_write(inode->i_sb);
1863 EXPORT_SYMBOL(touch_atime);
1866 * The logic we want is
1868 * if suid or (sgid and xgrp)
1871 int should_remove_suid(struct dentry *dentry)
1873 umode_t mode = d_inode(dentry)->i_mode;
1876 /* suid always must be killed */
1877 if (unlikely(mode & S_ISUID))
1878 kill = ATTR_KILL_SUID;
1881 * sgid without any exec bits is just a mandatory locking mark; leave
1882 * it alone. If some exec bits are set, it's a real sgid; kill it.
1884 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1885 kill |= ATTR_KILL_SGID;
1887 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1892 EXPORT_SYMBOL(should_remove_suid);
1895 * Return mask of changes for notify_change() that need to be done as a
1896 * response to write or truncate. Return 0 if nothing has to be changed.
1897 * Negative value on error (change should be denied).
1899 int dentry_needs_remove_privs(struct dentry *dentry)
1901 struct inode *inode = d_inode(dentry);
1905 if (IS_NOSEC(inode))
1908 mask = should_remove_suid(dentry);
1909 ret = security_inode_need_killpriv(dentry);
1913 mask |= ATTR_KILL_PRIV;
1917 static int __remove_privs(struct user_namespace *mnt_userns,
1918 struct dentry *dentry, int kill)
1920 struct iattr newattrs;
1922 newattrs.ia_valid = ATTR_FORCE | kill;
1924 * Note we call this on write, so notify_change will not
1925 * encounter any conflicting delegations:
1927 return notify_change(mnt_userns, dentry, &newattrs, NULL);
1931 * Remove special file priviledges (suid, capabilities) when file is written
1934 int file_remove_privs(struct file *file)
1936 struct dentry *dentry = file_dentry(file);
1937 struct inode *inode = file_inode(file);
1942 * Fast path for nothing security related.
1943 * As well for non-regular files, e.g. blkdev inodes.
1944 * For example, blkdev_write_iter() might get here
1945 * trying to remove privs which it is not allowed to.
1947 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1950 kill = dentry_needs_remove_privs(dentry);
1954 error = __remove_privs(file_mnt_user_ns(file), dentry, kill);
1956 inode_has_no_xattr(inode);
1960 EXPORT_SYMBOL(file_remove_privs);
1963 * file_update_time - update mtime and ctime time
1964 * @file: file accessed
1966 * Update the mtime and ctime members of an inode and mark the inode
1967 * for writeback. Note that this function is meant exclusively for
1968 * usage in the file write path of filesystems, and filesystems may
1969 * choose to explicitly ignore update via this function with the
1970 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1971 * timestamps are handled by the server. This can return an error for
1972 * file systems who need to allocate space in order to update an inode.
1975 int file_update_time(struct file *file)
1977 struct inode *inode = file_inode(file);
1978 struct timespec64 now;
1982 /* First try to exhaust all avenues to not sync */
1983 if (IS_NOCMTIME(inode))
1986 now = current_time(inode);
1987 if (!timespec64_equal(&inode->i_mtime, &now))
1990 if (!timespec64_equal(&inode->i_ctime, &now))
1993 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
1994 sync_it |= S_VERSION;
1999 /* Finally allowed to write? Takes lock. */
2000 if (__mnt_want_write_file(file))
2003 ret = update_time(inode, &now, sync_it);
2004 __mnt_drop_write_file(file);
2008 EXPORT_SYMBOL(file_update_time);
2010 /* Caller must hold the file's inode lock */
2011 int file_modified(struct file *file)
2016 * Clear the security bits if the process is not being run by root.
2017 * This keeps people from modifying setuid and setgid binaries.
2019 err = file_remove_privs(file);
2023 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2026 return file_update_time(file);
2028 EXPORT_SYMBOL(file_modified);
2030 int inode_needs_sync(struct inode *inode)
2034 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2038 EXPORT_SYMBOL(inode_needs_sync);
2041 * If we try to find an inode in the inode hash while it is being
2042 * deleted, we have to wait until the filesystem completes its
2043 * deletion before reporting that it isn't found. This function waits
2044 * until the deletion _might_ have completed. Callers are responsible
2045 * to recheck inode state.
2047 * It doesn't matter if I_NEW is not set initially, a call to
2048 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2051 static void __wait_on_freeing_inode(struct inode *inode)
2053 wait_queue_head_t *wq;
2054 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2055 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2056 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2057 spin_unlock(&inode->i_lock);
2058 spin_unlock(&inode_hash_lock);
2060 finish_wait(wq, &wait.wq_entry);
2061 spin_lock(&inode_hash_lock);
2064 static __initdata unsigned long ihash_entries;
2065 static int __init set_ihash_entries(char *str)
2069 ihash_entries = simple_strtoul(str, &str, 0);
2072 __setup("ihash_entries=", set_ihash_entries);
2075 * Initialize the waitqueues and inode hash table.
2077 void __init inode_init_early(void)
2079 /* If hashes are distributed across NUMA nodes, defer
2080 * hash allocation until vmalloc space is available.
2086 alloc_large_system_hash("Inode-cache",
2087 sizeof(struct hlist_head),
2090 HASH_EARLY | HASH_ZERO,
2097 void __init inode_init(void)
2099 /* inode slab cache */
2100 inode_cachep = kmem_cache_create("inode_cache",
2101 sizeof(struct inode),
2103 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2104 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2107 /* Hash may have been set up in inode_init_early */
2112 alloc_large_system_hash("Inode-cache",
2113 sizeof(struct hlist_head),
2123 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2125 inode->i_mode = mode;
2126 if (S_ISCHR(mode)) {
2127 inode->i_fop = &def_chr_fops;
2128 inode->i_rdev = rdev;
2129 } else if (S_ISBLK(mode)) {
2130 inode->i_fop = &def_blk_fops;
2131 inode->i_rdev = rdev;
2132 } else if (S_ISFIFO(mode))
2133 inode->i_fop = &pipefifo_fops;
2134 else if (S_ISSOCK(mode))
2135 ; /* leave it no_open_fops */
2137 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2138 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2141 EXPORT_SYMBOL(init_special_inode);
2144 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2145 * @mnt_userns: User namespace of the mount the inode was created from
2147 * @dir: Directory inode
2148 * @mode: mode of the new inode
2150 * If the inode has been created through an idmapped mount the user namespace of
2151 * the vfsmount must be passed through @mnt_userns. This function will then take
2152 * care to map the inode according to @mnt_userns before checking permissions
2153 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2154 * checking is to be performed on the raw inode simply passs init_user_ns.
2156 void inode_init_owner(struct user_namespace *mnt_userns, struct inode *inode,
2157 const struct inode *dir, umode_t mode)
2159 inode_fsuid_set(inode, mnt_userns);
2160 if (dir && dir->i_mode & S_ISGID) {
2161 inode->i_gid = dir->i_gid;
2163 /* Directories are special, and always inherit S_ISGID */
2166 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2167 !in_group_p(i_gid_into_mnt(mnt_userns, dir)) &&
2168 !capable_wrt_inode_uidgid(mnt_userns, dir, CAP_FSETID))
2171 inode_fsgid_set(inode, mnt_userns);
2172 inode->i_mode = mode;
2174 EXPORT_SYMBOL(inode_init_owner);
2177 * inode_owner_or_capable - check current task permissions to inode
2178 * @mnt_userns: user namespace of the mount the inode was found from
2179 * @inode: inode being checked
2181 * Return true if current either has CAP_FOWNER in a namespace with the
2182 * inode owner uid mapped, or owns the file.
2184 * If the inode has been found through an idmapped mount the user namespace of
2185 * the vfsmount must be passed through @mnt_userns. This function will then take
2186 * care to map the inode according to @mnt_userns before checking permissions.
2187 * On non-idmapped mounts or if permission checking is to be performed on the
2188 * raw inode simply passs init_user_ns.
2190 bool inode_owner_or_capable(struct user_namespace *mnt_userns,
2191 const struct inode *inode)
2194 struct user_namespace *ns;
2196 i_uid = i_uid_into_mnt(mnt_userns, inode);
2197 if (uid_eq(current_fsuid(), i_uid))
2200 ns = current_user_ns();
2201 if (kuid_has_mapping(ns, i_uid) && ns_capable(ns, CAP_FOWNER))
2205 EXPORT_SYMBOL(inode_owner_or_capable);
2208 * Direct i/o helper functions
2210 static void __inode_dio_wait(struct inode *inode)
2212 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2213 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2216 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2217 if (atomic_read(&inode->i_dio_count))
2219 } while (atomic_read(&inode->i_dio_count));
2220 finish_wait(wq, &q.wq_entry);
2224 * inode_dio_wait - wait for outstanding DIO requests to finish
2225 * @inode: inode to wait for
2227 * Waits for all pending direct I/O requests to finish so that we can
2228 * proceed with a truncate or equivalent operation.
2230 * Must be called under a lock that serializes taking new references
2231 * to i_dio_count, usually by inode->i_mutex.
2233 void inode_dio_wait(struct inode *inode)
2235 if (atomic_read(&inode->i_dio_count))
2236 __inode_dio_wait(inode);
2238 EXPORT_SYMBOL(inode_dio_wait);
2241 * inode_set_flags - atomically set some inode flags
2243 * Note: the caller should be holding i_mutex, or else be sure that
2244 * they have exclusive access to the inode structure (i.e., while the
2245 * inode is being instantiated). The reason for the cmpxchg() loop
2246 * --- which wouldn't be necessary if all code paths which modify
2247 * i_flags actually followed this rule, is that there is at least one
2248 * code path which doesn't today so we use cmpxchg() out of an abundance
2251 * In the long run, i_mutex is overkill, and we should probably look
2252 * at using the i_lock spinlock to protect i_flags, and then make sure
2253 * it is so documented in include/linux/fs.h and that all code follows
2254 * the locking convention!!
2256 void inode_set_flags(struct inode *inode, unsigned int flags,
2259 WARN_ON_ONCE(flags & ~mask);
2260 set_mask_bits(&inode->i_flags, mask, flags);
2262 EXPORT_SYMBOL(inode_set_flags);
2264 void inode_nohighmem(struct inode *inode)
2266 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2268 EXPORT_SYMBOL(inode_nohighmem);
2271 * timestamp_truncate - Truncate timespec to a granularity
2273 * @inode: inode being updated
2275 * Truncate a timespec to the granularity supported by the fs
2276 * containing the inode. Always rounds down. gran must
2277 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2279 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2281 struct super_block *sb = inode->i_sb;
2282 unsigned int gran = sb->s_time_gran;
2284 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2285 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2288 /* Avoid division in the common cases 1 ns and 1 s. */
2291 else if (gran == NSEC_PER_SEC)
2293 else if (gran > 1 && gran < NSEC_PER_SEC)
2294 t.tv_nsec -= t.tv_nsec % gran;
2296 WARN(1, "invalid file time granularity: %u", gran);
2299 EXPORT_SYMBOL(timestamp_truncate);
2302 * current_time - Return FS time
2305 * Return the current time truncated to the time granularity supported by
2308 * Note that inode and inode->sb cannot be NULL.
2309 * Otherwise, the function warns and returns time without truncation.
2311 struct timespec64 current_time(struct inode *inode)
2313 struct timespec64 now;
2315 ktime_get_coarse_real_ts64(&now);
2317 if (unlikely(!inode->i_sb)) {
2318 WARN(1, "current_time() called with uninitialized super_block in the inode");
2322 return timestamp_truncate(now, inode);
2324 EXPORT_SYMBOL(current_time);