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/fscrypt.h>
16 #include <linux/fsnotify.h>
17 #include <linux/mount.h>
18 #include <linux/posix_acl.h>
19 #include <linux/prefetch.h>
20 #include <linux/buffer_head.h> /* for inode_has_buffers */
21 #include <linux/ratelimit.h>
22 #include <linux/list_lru.h>
23 #include <linux/iversion.h>
24 #include <trace/events/writeback.h>
28 * Inode locking rules:
30 * inode->i_lock protects:
31 * inode->i_state, inode->i_hash, __iget()
32 * Inode LRU list locks protect:
33 * inode->i_sb->s_inode_lru, inode->i_lru
34 * inode->i_sb->s_inode_list_lock protects:
35 * inode->i_sb->s_inodes, inode->i_sb_list
36 * bdi->wb.list_lock protects:
37 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
38 * inode_hash_lock protects:
39 * inode_hashtable, inode->i_hash
43 * inode->i_sb->s_inode_list_lock
45 * Inode LRU list locks
51 * inode->i_sb->s_inode_list_lock
58 static unsigned int i_hash_mask __read_mostly;
59 static unsigned int i_hash_shift __read_mostly;
60 static struct hlist_head *inode_hashtable __read_mostly;
61 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
64 * Empty aops. Can be used for the cases where the user does not
65 * define any of the address_space operations.
67 const struct address_space_operations empty_aops = {
69 EXPORT_SYMBOL(empty_aops);
72 * Statistics gathering..
74 struct inodes_stat_t inodes_stat;
76 static DEFINE_PER_CPU(unsigned long, nr_inodes);
77 static DEFINE_PER_CPU(unsigned long, nr_unused);
79 static struct kmem_cache *inode_cachep __read_mostly;
81 static long get_nr_inodes(void)
85 for_each_possible_cpu(i)
86 sum += per_cpu(nr_inodes, i);
87 return sum < 0 ? 0 : sum;
90 static inline long get_nr_inodes_unused(void)
94 for_each_possible_cpu(i)
95 sum += per_cpu(nr_unused, i);
96 return sum < 0 ? 0 : sum;
99 long get_nr_dirty_inodes(void)
101 /* not actually dirty inodes, but a wild approximation */
102 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
103 return nr_dirty > 0 ? nr_dirty : 0;
107 * Handle nr_inode sysctl
110 int proc_nr_inodes(struct ctl_table *table, int write,
111 void __user *buffer, size_t *lenp, loff_t *ppos)
113 inodes_stat.nr_inodes = get_nr_inodes();
114 inodes_stat.nr_unused = get_nr_inodes_unused();
115 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
119 static int no_open(struct inode *inode, struct file *file)
125 * inode_init_always - perform inode structure initialisation
126 * @sb: superblock inode belongs to
127 * @inode: inode to initialise
129 * These are initializations that need to be done on every inode
130 * allocation as the fields are not initialised by slab allocation.
132 int inode_init_always(struct super_block *sb, struct inode *inode)
134 static const struct inode_operations empty_iops;
135 static const struct file_operations no_open_fops = {.open = no_open};
136 struct address_space *const mapping = &inode->i_data;
139 inode->i_blkbits = sb->s_blocksize_bits;
141 atomic64_set(&inode->i_sequence, 0);
142 atomic_set(&inode->i_count, 1);
143 inode->i_op = &empty_iops;
144 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_bdev = NULL;
159 inode->i_cdev = NULL;
160 inode->i_link = NULL;
161 inode->i_dir_seq = 0;
163 inode->dirtied_when = 0;
165 #ifdef CONFIG_CGROUP_WRITEBACK
166 inode->i_wb_frn_winner = 0;
167 inode->i_wb_frn_avg_time = 0;
168 inode->i_wb_frn_history = 0;
171 if (security_inode_alloc(inode))
173 spin_lock_init(&inode->i_lock);
174 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
176 init_rwsem(&inode->i_rwsem);
177 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
179 atomic_set(&inode->i_dio_count, 0);
181 mapping->a_ops = &empty_aops;
182 mapping->host = inode;
185 atomic_set(&mapping->i_mmap_writable, 0);
186 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
187 atomic_set(&mapping->nr_thps, 0);
189 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
190 mapping->private_data = NULL;
191 mapping->writeback_index = 0;
192 inode->i_private = NULL;
193 inode->i_mapping = mapping;
194 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
195 #ifdef CONFIG_FS_POSIX_ACL
196 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
199 #ifdef CONFIG_FSNOTIFY
200 inode->i_fsnotify_mask = 0;
202 inode->i_flctx = NULL;
203 this_cpu_inc(nr_inodes);
209 EXPORT_SYMBOL(inode_init_always);
211 void free_inode_nonrcu(struct inode *inode)
213 kmem_cache_free(inode_cachep, inode);
215 EXPORT_SYMBOL(free_inode_nonrcu);
217 static void i_callback(struct rcu_head *head)
219 struct inode *inode = container_of(head, struct inode, i_rcu);
220 if (inode->free_inode)
221 inode->free_inode(inode);
223 free_inode_nonrcu(inode);
226 static struct inode *alloc_inode(struct super_block *sb)
228 const struct super_operations *ops = sb->s_op;
231 if (ops->alloc_inode)
232 inode = ops->alloc_inode(sb);
234 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
239 if (unlikely(inode_init_always(sb, inode))) {
240 if (ops->destroy_inode) {
241 ops->destroy_inode(inode);
242 if (!ops->free_inode)
245 inode->free_inode = ops->free_inode;
246 i_callback(&inode->i_rcu);
253 void __destroy_inode(struct inode *inode)
255 BUG_ON(inode_has_buffers(inode));
256 inode_detach_wb(inode);
257 security_inode_free(inode);
258 fsnotify_inode_delete(inode);
259 locks_free_lock_context(inode);
260 if (!inode->i_nlink) {
261 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
262 atomic_long_dec(&inode->i_sb->s_remove_count);
265 #ifdef CONFIG_FS_POSIX_ACL
266 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
267 posix_acl_release(inode->i_acl);
268 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
269 posix_acl_release(inode->i_default_acl);
271 this_cpu_dec(nr_inodes);
273 EXPORT_SYMBOL(__destroy_inode);
275 static void destroy_inode(struct inode *inode)
277 const struct super_operations *ops = inode->i_sb->s_op;
279 BUG_ON(!list_empty(&inode->i_lru));
280 __destroy_inode(inode);
281 if (ops->destroy_inode) {
282 ops->destroy_inode(inode);
283 if (!ops->free_inode)
286 inode->free_inode = ops->free_inode;
287 call_rcu(&inode->i_rcu, i_callback);
291 * drop_nlink - directly drop an inode's link count
294 * This is a low-level filesystem helper to replace any
295 * direct filesystem manipulation of i_nlink. In cases
296 * where we are attempting to track writes to the
297 * filesystem, a decrement to zero means an imminent
298 * write when the file is truncated and actually unlinked
301 void drop_nlink(struct inode *inode)
303 WARN_ON(inode->i_nlink == 0);
306 atomic_long_inc(&inode->i_sb->s_remove_count);
308 EXPORT_SYMBOL(drop_nlink);
311 * clear_nlink - directly zero an inode's link count
314 * This is a low-level filesystem helper to replace any
315 * direct filesystem manipulation of i_nlink. See
316 * drop_nlink() for why we care about i_nlink hitting zero.
318 void clear_nlink(struct inode *inode)
320 if (inode->i_nlink) {
321 inode->__i_nlink = 0;
322 atomic_long_inc(&inode->i_sb->s_remove_count);
325 EXPORT_SYMBOL(clear_nlink);
328 * set_nlink - directly set an inode's link count
330 * @nlink: new nlink (should be non-zero)
332 * This is a low-level filesystem helper to replace any
333 * direct filesystem manipulation of i_nlink.
335 void set_nlink(struct inode *inode, unsigned int nlink)
340 /* Yes, some filesystems do change nlink from zero to one */
341 if (inode->i_nlink == 0)
342 atomic_long_dec(&inode->i_sb->s_remove_count);
344 inode->__i_nlink = nlink;
347 EXPORT_SYMBOL(set_nlink);
350 * inc_nlink - directly increment an inode's link count
353 * This is a low-level filesystem helper to replace any
354 * direct filesystem manipulation of i_nlink. Currently,
355 * it is only here for parity with dec_nlink().
357 void inc_nlink(struct inode *inode)
359 if (unlikely(inode->i_nlink == 0)) {
360 WARN_ON(!(inode->i_state & I_LINKABLE));
361 atomic_long_dec(&inode->i_sb->s_remove_count);
366 EXPORT_SYMBOL(inc_nlink);
368 static void __address_space_init_once(struct address_space *mapping)
370 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
371 init_rwsem(&mapping->i_mmap_rwsem);
372 INIT_LIST_HEAD(&mapping->private_list);
373 spin_lock_init(&mapping->private_lock);
374 mapping->i_mmap = RB_ROOT_CACHED;
377 void address_space_init_once(struct address_space *mapping)
379 memset(mapping, 0, sizeof(*mapping));
380 __address_space_init_once(mapping);
382 EXPORT_SYMBOL(address_space_init_once);
385 * These are initializations that only need to be done
386 * once, because the fields are idempotent across use
387 * of the inode, so let the slab aware of that.
389 void inode_init_once(struct inode *inode)
391 memset(inode, 0, sizeof(*inode));
392 INIT_HLIST_NODE(&inode->i_hash);
393 INIT_LIST_HEAD(&inode->i_devices);
394 INIT_LIST_HEAD(&inode->i_io_list);
395 INIT_LIST_HEAD(&inode->i_wb_list);
396 INIT_LIST_HEAD(&inode->i_lru);
397 __address_space_init_once(&inode->i_data);
398 i_size_ordered_init(inode);
400 EXPORT_SYMBOL(inode_init_once);
402 static void init_once(void *foo)
404 struct inode *inode = (struct inode *) foo;
406 inode_init_once(inode);
410 * inode->i_lock must be held
412 void __iget(struct inode *inode)
414 atomic_inc(&inode->i_count);
418 * get additional reference to inode; caller must already hold one.
420 void ihold(struct inode *inode)
422 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
424 EXPORT_SYMBOL(ihold);
426 static void inode_lru_list_add(struct inode *inode)
428 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
429 this_cpu_inc(nr_unused);
431 inode->i_state |= I_REFERENCED;
435 * Add inode to LRU if needed (inode is unused and clean).
437 * Needs inode->i_lock held.
439 void inode_add_lru(struct inode *inode)
441 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
442 I_FREEING | I_WILL_FREE)) &&
443 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
444 inode_lru_list_add(inode);
448 static void inode_lru_list_del(struct inode *inode)
451 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
452 this_cpu_dec(nr_unused);
456 * inode_sb_list_add - add inode to the superblock list of inodes
457 * @inode: inode to add
459 void inode_sb_list_add(struct inode *inode)
461 spin_lock(&inode->i_sb->s_inode_list_lock);
462 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
463 spin_unlock(&inode->i_sb->s_inode_list_lock);
465 EXPORT_SYMBOL_GPL(inode_sb_list_add);
467 static inline void inode_sb_list_del(struct inode *inode)
469 if (!list_empty(&inode->i_sb_list)) {
470 spin_lock(&inode->i_sb->s_inode_list_lock);
471 list_del_init(&inode->i_sb_list);
472 spin_unlock(&inode->i_sb->s_inode_list_lock);
476 static unsigned long hash(struct super_block *sb, unsigned long hashval)
480 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
482 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
483 return tmp & i_hash_mask;
487 * __insert_inode_hash - hash an inode
488 * @inode: unhashed inode
489 * @hashval: unsigned long value used to locate this object in the
492 * Add an inode to the inode hash for this superblock.
494 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
496 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
498 spin_lock(&inode_hash_lock);
499 spin_lock(&inode->i_lock);
500 hlist_add_head_rcu(&inode->i_hash, b);
501 spin_unlock(&inode->i_lock);
502 spin_unlock(&inode_hash_lock);
504 EXPORT_SYMBOL(__insert_inode_hash);
507 * __remove_inode_hash - remove an inode from the hash
508 * @inode: inode to unhash
510 * Remove an inode from the superblock.
512 void __remove_inode_hash(struct inode *inode)
514 spin_lock(&inode_hash_lock);
515 spin_lock(&inode->i_lock);
516 hlist_del_init_rcu(&inode->i_hash);
517 spin_unlock(&inode->i_lock);
518 spin_unlock(&inode_hash_lock);
520 EXPORT_SYMBOL(__remove_inode_hash);
522 void clear_inode(struct inode *inode)
525 * We have to cycle the i_pages lock here because reclaim can be in the
526 * process of removing the last page (in __delete_from_page_cache())
527 * and we must not free the mapping under it.
529 xa_lock_irq(&inode->i_data.i_pages);
530 BUG_ON(inode->i_data.nrpages);
531 BUG_ON(inode->i_data.nrexceptional);
532 xa_unlock_irq(&inode->i_data.i_pages);
533 BUG_ON(!list_empty(&inode->i_data.private_list));
534 BUG_ON(!(inode->i_state & I_FREEING));
535 BUG_ON(inode->i_state & I_CLEAR);
536 BUG_ON(!list_empty(&inode->i_wb_list));
537 /* don't need i_lock here, no concurrent mods to i_state */
538 inode->i_state = I_FREEING | I_CLEAR;
540 EXPORT_SYMBOL(clear_inode);
543 * Free the inode passed in, removing it from the lists it is still connected
544 * to. We remove any pages still attached to the inode and wait for any IO that
545 * is still in progress before finally destroying the inode.
547 * An inode must already be marked I_FREEING so that we avoid the inode being
548 * moved back onto lists if we race with other code that manipulates the lists
549 * (e.g. writeback_single_inode). The caller is responsible for setting this.
551 * An inode must already be removed from the LRU list before being evicted from
552 * the cache. This should occur atomically with setting the I_FREEING state
553 * flag, so no inodes here should ever be on the LRU when being evicted.
555 static void evict(struct inode *inode)
557 const struct super_operations *op = inode->i_sb->s_op;
559 BUG_ON(!(inode->i_state & I_FREEING));
560 BUG_ON(!list_empty(&inode->i_lru));
562 if (!list_empty(&inode->i_io_list))
563 inode_io_list_del(inode);
565 inode_sb_list_del(inode);
568 * Wait for flusher thread to be done with the inode so that filesystem
569 * does not start destroying it while writeback is still running. Since
570 * the inode has I_FREEING set, flusher thread won't start new work on
571 * the inode. We just have to wait for running writeback to finish.
573 inode_wait_for_writeback(inode);
575 if (op->evict_inode) {
576 op->evict_inode(inode);
578 truncate_inode_pages_final(&inode->i_data);
581 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
583 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
586 remove_inode_hash(inode);
588 spin_lock(&inode->i_lock);
589 wake_up_bit(&inode->i_state, __I_NEW);
590 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
591 spin_unlock(&inode->i_lock);
593 destroy_inode(inode);
597 * dispose_list - dispose of the contents of a local list
598 * @head: the head of the list to free
600 * Dispose-list gets a local list with local inodes in it, so it doesn't
601 * need to worry about list corruption and SMP locks.
603 static void dispose_list(struct list_head *head)
605 while (!list_empty(head)) {
608 inode = list_first_entry(head, struct inode, i_lru);
609 list_del_init(&inode->i_lru);
617 * evict_inodes - evict all evictable inodes for a superblock
618 * @sb: superblock to operate on
620 * Make sure that no inodes with zero refcount are retained. This is
621 * called by superblock shutdown after having SB_ACTIVE flag removed,
622 * so any inode reaching zero refcount during or after that call will
623 * be immediately evicted.
625 void evict_inodes(struct super_block *sb)
627 struct inode *inode, *next;
631 spin_lock(&sb->s_inode_list_lock);
632 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
633 if (atomic_read(&inode->i_count))
636 spin_lock(&inode->i_lock);
637 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
638 spin_unlock(&inode->i_lock);
642 inode->i_state |= I_FREEING;
643 inode_lru_list_del(inode);
644 spin_unlock(&inode->i_lock);
645 list_add(&inode->i_lru, &dispose);
648 * We can have a ton of inodes to evict at unmount time given
649 * enough memory, check to see if we need to go to sleep for a
650 * bit so we don't livelock.
652 if (need_resched()) {
653 spin_unlock(&sb->s_inode_list_lock);
655 dispose_list(&dispose);
659 spin_unlock(&sb->s_inode_list_lock);
661 dispose_list(&dispose);
663 EXPORT_SYMBOL_GPL(evict_inodes);
666 * invalidate_inodes - attempt to free all inodes on a superblock
667 * @sb: superblock to operate on
668 * @kill_dirty: flag to guide handling of dirty inodes
670 * Attempts to free all inodes for a given superblock. If there were any
671 * busy inodes return a non-zero value, else zero.
672 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
675 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
678 struct inode *inode, *next;
682 spin_lock(&sb->s_inode_list_lock);
683 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
684 spin_lock(&inode->i_lock);
685 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
686 spin_unlock(&inode->i_lock);
689 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
690 spin_unlock(&inode->i_lock);
694 if (atomic_read(&inode->i_count)) {
695 spin_unlock(&inode->i_lock);
700 inode->i_state |= I_FREEING;
701 inode_lru_list_del(inode);
702 spin_unlock(&inode->i_lock);
703 list_add(&inode->i_lru, &dispose);
704 if (need_resched()) {
705 spin_unlock(&sb->s_inode_list_lock);
707 dispose_list(&dispose);
711 spin_unlock(&sb->s_inode_list_lock);
713 dispose_list(&dispose);
719 * Isolate the inode from the LRU in preparation for freeing it.
721 * Any inodes which are pinned purely because of attached pagecache have their
722 * pagecache removed. If the inode has metadata buffers attached to
723 * mapping->private_list then try to remove them.
725 * If the inode has the I_REFERENCED flag set, then it means that it has been
726 * used recently - the flag is set in iput_final(). When we encounter such an
727 * inode, clear the flag and move it to the back of the LRU so it gets another
728 * pass through the LRU before it gets reclaimed. This is necessary because of
729 * the fact we are doing lazy LRU updates to minimise lock contention so the
730 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
731 * with this flag set because they are the inodes that are out of order.
733 static enum lru_status inode_lru_isolate(struct list_head *item,
734 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
736 struct list_head *freeable = arg;
737 struct inode *inode = container_of(item, struct inode, i_lru);
740 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
741 * If we fail to get the lock, just skip it.
743 if (!spin_trylock(&inode->i_lock))
747 * Referenced or dirty inodes are still in use. Give them another pass
748 * through the LRU as we canot reclaim them now.
750 if (atomic_read(&inode->i_count) ||
751 (inode->i_state & ~I_REFERENCED)) {
752 list_lru_isolate(lru, &inode->i_lru);
753 spin_unlock(&inode->i_lock);
754 this_cpu_dec(nr_unused);
758 /* recently referenced inodes get one more pass */
759 if (inode->i_state & I_REFERENCED) {
760 inode->i_state &= ~I_REFERENCED;
761 spin_unlock(&inode->i_lock);
765 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
767 spin_unlock(&inode->i_lock);
768 spin_unlock(lru_lock);
769 if (remove_inode_buffers(inode)) {
771 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
772 if (current_is_kswapd())
773 __count_vm_events(KSWAPD_INODESTEAL, reap);
775 __count_vm_events(PGINODESTEAL, reap);
776 if (current->reclaim_state)
777 current->reclaim_state->reclaimed_slab += reap;
784 WARN_ON(inode->i_state & I_NEW);
785 inode->i_state |= I_FREEING;
786 list_lru_isolate_move(lru, &inode->i_lru, freeable);
787 spin_unlock(&inode->i_lock);
789 this_cpu_dec(nr_unused);
794 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
795 * This is called from the superblock shrinker function with a number of inodes
796 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
797 * then are freed outside inode_lock by dispose_list().
799 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
804 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
805 inode_lru_isolate, &freeable);
806 dispose_list(&freeable);
810 static void __wait_on_freeing_inode(struct inode *inode);
812 * Called with the inode lock held.
814 static struct inode *find_inode(struct super_block *sb,
815 struct hlist_head *head,
816 int (*test)(struct inode *, void *),
819 struct inode *inode = NULL;
822 hlist_for_each_entry(inode, head, i_hash) {
823 if (inode->i_sb != sb)
825 if (!test(inode, data))
827 spin_lock(&inode->i_lock);
828 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
829 __wait_on_freeing_inode(inode);
832 if (unlikely(inode->i_state & I_CREATING)) {
833 spin_unlock(&inode->i_lock);
834 return ERR_PTR(-ESTALE);
837 spin_unlock(&inode->i_lock);
844 * find_inode_fast is the fast path version of find_inode, see the comment at
845 * iget_locked for details.
847 static struct inode *find_inode_fast(struct super_block *sb,
848 struct hlist_head *head, unsigned long ino)
850 struct inode *inode = NULL;
853 hlist_for_each_entry(inode, head, i_hash) {
854 if (inode->i_ino != ino)
856 if (inode->i_sb != sb)
858 spin_lock(&inode->i_lock);
859 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
860 __wait_on_freeing_inode(inode);
863 if (unlikely(inode->i_state & I_CREATING)) {
864 spin_unlock(&inode->i_lock);
865 return ERR_PTR(-ESTALE);
868 spin_unlock(&inode->i_lock);
875 * Each cpu owns a range of LAST_INO_BATCH numbers.
876 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
877 * to renew the exhausted range.
879 * This does not significantly increase overflow rate because every CPU can
880 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
881 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
882 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
883 * overflow rate by 2x, which does not seem too significant.
885 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
886 * error if st_ino won't fit in target struct field. Use 32bit counter
887 * here to attempt to avoid that.
889 #define LAST_INO_BATCH 1024
890 static DEFINE_PER_CPU(unsigned int, last_ino);
892 unsigned int get_next_ino(void)
894 unsigned int *p = &get_cpu_var(last_ino);
895 unsigned int res = *p;
898 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
899 static atomic_t shared_last_ino;
900 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
902 res = next - LAST_INO_BATCH;
907 /* get_next_ino should not provide a 0 inode number */
911 put_cpu_var(last_ino);
914 EXPORT_SYMBOL(get_next_ino);
917 * new_inode_pseudo - obtain an inode
920 * Allocates a new inode for given superblock.
921 * Inode wont be chained in superblock s_inodes list
923 * - fs can't be unmount
924 * - quotas, fsnotify, writeback can't work
926 struct inode *new_inode_pseudo(struct super_block *sb)
928 struct inode *inode = alloc_inode(sb);
931 spin_lock(&inode->i_lock);
933 spin_unlock(&inode->i_lock);
934 INIT_LIST_HEAD(&inode->i_sb_list);
940 * new_inode - obtain an inode
943 * Allocates a new inode for given superblock. The default gfp_mask
944 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
945 * If HIGHMEM pages are unsuitable or it is known that pages allocated
946 * for the page cache are not reclaimable or migratable,
947 * mapping_set_gfp_mask() must be called with suitable flags on the
948 * newly created inode's mapping
951 struct inode *new_inode(struct super_block *sb)
955 spin_lock_prefetch(&sb->s_inode_list_lock);
957 inode = new_inode_pseudo(sb);
959 inode_sb_list_add(inode);
962 EXPORT_SYMBOL(new_inode);
964 #ifdef CONFIG_DEBUG_LOCK_ALLOC
965 void lockdep_annotate_inode_mutex_key(struct inode *inode)
967 if (S_ISDIR(inode->i_mode)) {
968 struct file_system_type *type = inode->i_sb->s_type;
970 /* Set new key only if filesystem hasn't already changed it */
971 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
973 * ensure nobody is actually holding i_mutex
975 // mutex_destroy(&inode->i_mutex);
976 init_rwsem(&inode->i_rwsem);
977 lockdep_set_class(&inode->i_rwsem,
978 &type->i_mutex_dir_key);
982 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
986 * unlock_new_inode - clear the I_NEW state and wake up any waiters
987 * @inode: new inode to unlock
989 * Called when the inode is fully initialised to clear the new state of the
990 * inode and wake up anyone waiting for the inode to finish initialisation.
992 void unlock_new_inode(struct inode *inode)
994 lockdep_annotate_inode_mutex_key(inode);
995 spin_lock(&inode->i_lock);
996 WARN_ON(!(inode->i_state & I_NEW));
997 inode->i_state &= ~I_NEW & ~I_CREATING;
999 wake_up_bit(&inode->i_state, __I_NEW);
1000 spin_unlock(&inode->i_lock);
1002 EXPORT_SYMBOL(unlock_new_inode);
1004 void discard_new_inode(struct inode *inode)
1006 lockdep_annotate_inode_mutex_key(inode);
1007 spin_lock(&inode->i_lock);
1008 WARN_ON(!(inode->i_state & I_NEW));
1009 inode->i_state &= ~I_NEW;
1011 wake_up_bit(&inode->i_state, __I_NEW);
1012 spin_unlock(&inode->i_lock);
1015 EXPORT_SYMBOL(discard_new_inode);
1018 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1020 * Lock any non-NULL argument that is not a directory.
1021 * Zero, one or two objects may be locked by this function.
1023 * @inode1: first inode to lock
1024 * @inode2: second inode to lock
1026 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1028 if (inode1 > inode2)
1029 swap(inode1, inode2);
1031 if (inode1 && !S_ISDIR(inode1->i_mode))
1033 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1034 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1036 EXPORT_SYMBOL(lock_two_nondirectories);
1039 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1040 * @inode1: first inode to unlock
1041 * @inode2: second inode to unlock
1043 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1045 if (inode1 && !S_ISDIR(inode1->i_mode))
1046 inode_unlock(inode1);
1047 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1048 inode_unlock(inode2);
1050 EXPORT_SYMBOL(unlock_two_nondirectories);
1053 * inode_insert5 - obtain an inode from a mounted file system
1054 * @inode: pre-allocated inode to use for insert to cache
1055 * @hashval: hash value (usually inode number) to get
1056 * @test: callback used for comparisons between inodes
1057 * @set: callback used to initialize a new struct inode
1058 * @data: opaque data pointer to pass to @test and @set
1060 * Search for the inode specified by @hashval and @data in the inode cache,
1061 * and if present it is return it with an increased reference count. This is
1062 * a variant of iget5_locked() for callers that don't want to fail on memory
1063 * allocation of inode.
1065 * If the inode is not in cache, insert the pre-allocated inode to cache and
1066 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1067 * to fill it in before unlocking it via unlock_new_inode().
1069 * Note both @test and @set are called with the inode_hash_lock held, so can't
1072 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1073 int (*test)(struct inode *, void *),
1074 int (*set)(struct inode *, void *), void *data)
1076 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1078 bool creating = inode->i_state & I_CREATING;
1081 spin_lock(&inode_hash_lock);
1082 old = find_inode(inode->i_sb, head, test, data);
1083 if (unlikely(old)) {
1085 * Uhhuh, somebody else created the same inode under us.
1086 * Use the old inode instead of the preallocated one.
1088 spin_unlock(&inode_hash_lock);
1092 if (unlikely(inode_unhashed(old))) {
1099 if (set && unlikely(set(inode, data))) {
1105 * Return the locked inode with I_NEW set, the
1106 * caller is responsible for filling in the contents
1108 spin_lock(&inode->i_lock);
1109 inode->i_state |= I_NEW;
1110 hlist_add_head_rcu(&inode->i_hash, head);
1111 spin_unlock(&inode->i_lock);
1113 inode_sb_list_add(inode);
1115 spin_unlock(&inode_hash_lock);
1119 EXPORT_SYMBOL(inode_insert5);
1122 * iget5_locked - obtain an inode from a mounted file system
1123 * @sb: super block of file system
1124 * @hashval: hash value (usually inode number) to get
1125 * @test: callback used for comparisons between inodes
1126 * @set: callback used to initialize a new struct inode
1127 * @data: opaque data pointer to pass to @test and @set
1129 * Search for the inode specified by @hashval and @data in the inode cache,
1130 * and if present it is return it with an increased reference count. This is
1131 * a generalized version of iget_locked() for file systems where the inode
1132 * number is not sufficient for unique identification of an inode.
1134 * If the inode is not in cache, allocate a new inode and return it locked,
1135 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1136 * before unlocking it via unlock_new_inode().
1138 * Note both @test and @set are called with the inode_hash_lock held, so can't
1141 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1142 int (*test)(struct inode *, void *),
1143 int (*set)(struct inode *, void *), void *data)
1145 struct inode *inode = ilookup5(sb, hashval, test, data);
1148 struct inode *new = alloc_inode(sb);
1152 inode = inode_insert5(new, hashval, test, set, data);
1153 if (unlikely(inode != new))
1159 EXPORT_SYMBOL(iget5_locked);
1162 * iget_locked - obtain an inode from a mounted file system
1163 * @sb: super block of file system
1164 * @ino: inode number to get
1166 * Search for the inode specified by @ino in the inode cache and if present
1167 * return it with an increased reference count. This is for file systems
1168 * where the inode number is sufficient for unique identification of an inode.
1170 * If the inode is not in cache, allocate a new inode and return it locked,
1171 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1172 * before unlocking it via unlock_new_inode().
1174 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1176 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1177 struct inode *inode;
1179 spin_lock(&inode_hash_lock);
1180 inode = find_inode_fast(sb, head, ino);
1181 spin_unlock(&inode_hash_lock);
1185 wait_on_inode(inode);
1186 if (unlikely(inode_unhashed(inode))) {
1193 inode = alloc_inode(sb);
1197 spin_lock(&inode_hash_lock);
1198 /* We released the lock, so.. */
1199 old = find_inode_fast(sb, head, ino);
1202 spin_lock(&inode->i_lock);
1203 inode->i_state = I_NEW;
1204 hlist_add_head_rcu(&inode->i_hash, head);
1205 spin_unlock(&inode->i_lock);
1206 inode_sb_list_add(inode);
1207 spin_unlock(&inode_hash_lock);
1209 /* Return the locked inode with I_NEW set, the
1210 * caller is responsible for filling in the contents
1216 * Uhhuh, somebody else created the same inode under
1217 * us. Use the old inode instead of the one we just
1220 spin_unlock(&inode_hash_lock);
1221 destroy_inode(inode);
1225 wait_on_inode(inode);
1226 if (unlikely(inode_unhashed(inode))) {
1233 EXPORT_SYMBOL(iget_locked);
1236 * search the inode cache for a matching inode number.
1237 * If we find one, then the inode number we are trying to
1238 * allocate is not unique and so we should not use it.
1240 * Returns 1 if the inode number is unique, 0 if it is not.
1242 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1244 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1245 struct inode *inode;
1247 hlist_for_each_entry_rcu(inode, b, i_hash) {
1248 if (inode->i_ino == ino && inode->i_sb == sb)
1255 * iunique - get a unique inode number
1257 * @max_reserved: highest reserved inode number
1259 * Obtain an inode number that is unique on the system for a given
1260 * superblock. This is used by file systems that have no natural
1261 * permanent inode numbering system. An inode number is returned that
1262 * is higher than the reserved limit but unique.
1265 * With a large number of inodes live on the file system this function
1266 * currently becomes quite slow.
1268 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1271 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1272 * error if st_ino won't fit in target struct field. Use 32bit counter
1273 * here to attempt to avoid that.
1275 static DEFINE_SPINLOCK(iunique_lock);
1276 static unsigned int counter;
1280 spin_lock(&iunique_lock);
1282 if (counter <= max_reserved)
1283 counter = max_reserved + 1;
1285 } while (!test_inode_iunique(sb, res));
1286 spin_unlock(&iunique_lock);
1291 EXPORT_SYMBOL(iunique);
1293 struct inode *igrab(struct inode *inode)
1295 spin_lock(&inode->i_lock);
1296 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1298 spin_unlock(&inode->i_lock);
1300 spin_unlock(&inode->i_lock);
1302 * Handle the case where s_op->clear_inode is not been
1303 * called yet, and somebody is calling igrab
1304 * while the inode is getting freed.
1310 EXPORT_SYMBOL(igrab);
1313 * ilookup5_nowait - search for an inode in the inode cache
1314 * @sb: super block of file system to search
1315 * @hashval: hash value (usually inode number) to search for
1316 * @test: callback used for comparisons between inodes
1317 * @data: opaque data pointer to pass to @test
1319 * Search for the inode specified by @hashval and @data in the inode cache.
1320 * If the inode is in the cache, the inode is returned with an incremented
1323 * Note: I_NEW is not waited upon so you have to be very careful what you do
1324 * with the returned inode. You probably should be using ilookup5() instead.
1326 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1328 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1329 int (*test)(struct inode *, void *), void *data)
1331 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1332 struct inode *inode;
1334 spin_lock(&inode_hash_lock);
1335 inode = find_inode(sb, head, test, data);
1336 spin_unlock(&inode_hash_lock);
1338 return IS_ERR(inode) ? NULL : inode;
1340 EXPORT_SYMBOL(ilookup5_nowait);
1343 * ilookup5 - search for an inode in the inode cache
1344 * @sb: super block of file system to search
1345 * @hashval: hash value (usually inode number) to search for
1346 * @test: callback used for comparisons between inodes
1347 * @data: opaque data pointer to pass to @test
1349 * Search for the inode specified by @hashval and @data in the inode cache,
1350 * and if the inode is in the cache, return the inode with an incremented
1351 * reference count. Waits on I_NEW before returning the inode.
1352 * returned with an incremented reference count.
1354 * This is a generalized version of ilookup() for file systems where the
1355 * inode number is not sufficient for unique identification of an inode.
1357 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1359 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1360 int (*test)(struct inode *, void *), void *data)
1362 struct inode *inode;
1364 inode = ilookup5_nowait(sb, hashval, test, data);
1366 wait_on_inode(inode);
1367 if (unlikely(inode_unhashed(inode))) {
1374 EXPORT_SYMBOL(ilookup5);
1377 * ilookup - search for an inode in the inode cache
1378 * @sb: super block of file system to search
1379 * @ino: inode number to search for
1381 * Search for the inode @ino in the inode cache, and if the inode is in the
1382 * cache, the inode is returned with an incremented reference count.
1384 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1386 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1387 struct inode *inode;
1389 spin_lock(&inode_hash_lock);
1390 inode = find_inode_fast(sb, head, ino);
1391 spin_unlock(&inode_hash_lock);
1396 wait_on_inode(inode);
1397 if (unlikely(inode_unhashed(inode))) {
1404 EXPORT_SYMBOL(ilookup);
1407 * find_inode_nowait - find an inode in the inode cache
1408 * @sb: super block of file system to search
1409 * @hashval: hash value (usually inode number) to search for
1410 * @match: callback used for comparisons between inodes
1411 * @data: opaque data pointer to pass to @match
1413 * Search for the inode specified by @hashval and @data in the inode
1414 * cache, where the helper function @match will return 0 if the inode
1415 * does not match, 1 if the inode does match, and -1 if the search
1416 * should be stopped. The @match function must be responsible for
1417 * taking the i_lock spin_lock and checking i_state for an inode being
1418 * freed or being initialized, and incrementing the reference count
1419 * before returning 1. It also must not sleep, since it is called with
1420 * the inode_hash_lock spinlock held.
1422 * This is a even more generalized version of ilookup5() when the
1423 * function must never block --- find_inode() can block in
1424 * __wait_on_freeing_inode() --- or when the caller can not increment
1425 * the reference count because the resulting iput() might cause an
1426 * inode eviction. The tradeoff is that the @match funtion must be
1427 * very carefully implemented.
1429 struct inode *find_inode_nowait(struct super_block *sb,
1430 unsigned long hashval,
1431 int (*match)(struct inode *, unsigned long,
1435 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1436 struct inode *inode, *ret_inode = NULL;
1439 spin_lock(&inode_hash_lock);
1440 hlist_for_each_entry(inode, head, i_hash) {
1441 if (inode->i_sb != sb)
1443 mval = match(inode, hashval, data);
1451 spin_unlock(&inode_hash_lock);
1454 EXPORT_SYMBOL(find_inode_nowait);
1457 * find_inode_rcu - find an inode in the inode cache
1458 * @sb: Super block of file system to search
1459 * @hashval: Key to hash
1460 * @test: Function to test match on an inode
1461 * @data: Data for test function
1463 * Search for the inode specified by @hashval and @data in the inode cache,
1464 * where the helper function @test will return 0 if the inode does not match
1465 * and 1 if it does. The @test function must be responsible for taking the
1466 * i_lock spin_lock and checking i_state for an inode being freed or being
1469 * If successful, this will return the inode for which the @test function
1470 * returned 1 and NULL otherwise.
1472 * The @test function is not permitted to take a ref on any inode presented.
1473 * It is also not permitted to sleep.
1475 * The caller must hold the RCU read lock.
1477 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1478 int (*test)(struct inode *, void *), void *data)
1480 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1481 struct inode *inode;
1483 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1484 "suspicious find_inode_rcu() usage");
1486 hlist_for_each_entry_rcu(inode, head, i_hash) {
1487 if (inode->i_sb == sb &&
1488 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1494 EXPORT_SYMBOL(find_inode_rcu);
1497 * find_inode_by_rcu - Find an inode in the inode cache
1498 * @sb: Super block of file system to search
1499 * @ino: The inode number to match
1501 * Search for the inode specified by @hashval and @data in the inode cache,
1502 * where the helper function @test will return 0 if the inode does not match
1503 * and 1 if it does. The @test function must be responsible for taking the
1504 * i_lock spin_lock and checking i_state for an inode being freed or being
1507 * If successful, this will return the inode for which the @test function
1508 * returned 1 and NULL otherwise.
1510 * The @test function is not permitted to take a ref on any inode presented.
1511 * It is also not permitted to sleep.
1513 * The caller must hold the RCU read lock.
1515 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1518 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1519 struct inode *inode;
1521 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1522 "suspicious find_inode_by_ino_rcu() usage");
1524 hlist_for_each_entry_rcu(inode, head, i_hash) {
1525 if (inode->i_ino == ino &&
1526 inode->i_sb == sb &&
1527 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1532 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1534 int insert_inode_locked(struct inode *inode)
1536 struct super_block *sb = inode->i_sb;
1537 ino_t ino = inode->i_ino;
1538 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1541 struct inode *old = NULL;
1542 spin_lock(&inode_hash_lock);
1543 hlist_for_each_entry(old, head, i_hash) {
1544 if (old->i_ino != ino)
1546 if (old->i_sb != sb)
1548 spin_lock(&old->i_lock);
1549 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1550 spin_unlock(&old->i_lock);
1556 spin_lock(&inode->i_lock);
1557 inode->i_state |= I_NEW | I_CREATING;
1558 hlist_add_head_rcu(&inode->i_hash, head);
1559 spin_unlock(&inode->i_lock);
1560 spin_unlock(&inode_hash_lock);
1563 if (unlikely(old->i_state & I_CREATING)) {
1564 spin_unlock(&old->i_lock);
1565 spin_unlock(&inode_hash_lock);
1569 spin_unlock(&old->i_lock);
1570 spin_unlock(&inode_hash_lock);
1572 if (unlikely(!inode_unhashed(old))) {
1579 EXPORT_SYMBOL(insert_inode_locked);
1581 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1582 int (*test)(struct inode *, void *), void *data)
1586 inode->i_state |= I_CREATING;
1587 old = inode_insert5(inode, hashval, test, NULL, data);
1595 EXPORT_SYMBOL(insert_inode_locked4);
1598 int generic_delete_inode(struct inode *inode)
1602 EXPORT_SYMBOL(generic_delete_inode);
1605 * Called when we're dropping the last reference
1608 * Call the FS "drop_inode()" function, defaulting to
1609 * the legacy UNIX filesystem behaviour. If it tells
1610 * us to evict inode, do so. Otherwise, retain inode
1611 * in cache if fs is alive, sync and evict if fs is
1614 static void iput_final(struct inode *inode)
1616 struct super_block *sb = inode->i_sb;
1617 const struct super_operations *op = inode->i_sb->s_op;
1618 unsigned long state;
1621 WARN_ON(inode->i_state & I_NEW);
1624 drop = op->drop_inode(inode);
1626 drop = generic_drop_inode(inode);
1628 if (!drop && (sb->s_flags & SB_ACTIVE)) {
1629 inode_add_lru(inode);
1630 spin_unlock(&inode->i_lock);
1634 state = inode->i_state;
1636 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1637 spin_unlock(&inode->i_lock);
1639 write_inode_now(inode, 1);
1641 spin_lock(&inode->i_lock);
1642 state = inode->i_state;
1643 WARN_ON(state & I_NEW);
1644 state &= ~I_WILL_FREE;
1647 WRITE_ONCE(inode->i_state, state | I_FREEING);
1648 if (!list_empty(&inode->i_lru))
1649 inode_lru_list_del(inode);
1650 spin_unlock(&inode->i_lock);
1656 * iput - put an inode
1657 * @inode: inode to put
1659 * Puts an inode, dropping its usage count. If the inode use count hits
1660 * zero, the inode is then freed and may also be destroyed.
1662 * Consequently, iput() can sleep.
1664 void iput(struct inode *inode)
1668 BUG_ON(inode->i_state & I_CLEAR);
1670 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1671 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1672 atomic_inc(&inode->i_count);
1673 spin_unlock(&inode->i_lock);
1674 trace_writeback_lazytime_iput(inode);
1675 mark_inode_dirty_sync(inode);
1681 EXPORT_SYMBOL(iput);
1685 * bmap - find a block number in a file
1686 * @inode: inode owning the block number being requested
1687 * @block: pointer containing the block to find
1689 * Replaces the value in *block with the block number on the device holding
1690 * corresponding to the requested block number in the file.
1691 * That is, asked for block 4 of inode 1 the function will replace the
1692 * 4 in *block, with disk block relative to the disk start that holds that
1693 * block of the file.
1695 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1696 * hole, returns 0 and *block is also set to 0.
1698 int bmap(struct inode *inode, sector_t *block)
1700 if (!inode->i_mapping->a_ops->bmap)
1703 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1706 EXPORT_SYMBOL(bmap);
1710 * With relative atime, only update atime if the previous atime is
1711 * earlier than either the ctime or mtime or if at least a day has
1712 * passed since the last atime update.
1714 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1715 struct timespec64 now)
1718 if (!(mnt->mnt_flags & MNT_RELATIME))
1721 * Is mtime younger than atime? If yes, update atime:
1723 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1726 * Is ctime younger than atime? If yes, update atime:
1728 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1732 * Is the previous atime value older than a day? If yes,
1735 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1738 * Good, we can skip the atime update:
1743 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1745 int iflags = I_DIRTY_TIME;
1748 if (flags & S_ATIME)
1749 inode->i_atime = *time;
1750 if (flags & S_VERSION)
1751 dirty = inode_maybe_inc_iversion(inode, false);
1752 if (flags & S_CTIME)
1753 inode->i_ctime = *time;
1754 if (flags & S_MTIME)
1755 inode->i_mtime = *time;
1756 if ((flags & (S_ATIME | S_CTIME | S_MTIME)) &&
1757 !(inode->i_sb->s_flags & SB_LAZYTIME))
1761 iflags |= I_DIRTY_SYNC;
1762 __mark_inode_dirty(inode, iflags);
1765 EXPORT_SYMBOL(generic_update_time);
1768 * This does the actual work of updating an inodes time or version. Must have
1769 * had called mnt_want_write() before calling this.
1771 static int update_time(struct inode *inode, struct timespec64 *time, int flags)
1773 if (inode->i_op->update_time)
1774 return inode->i_op->update_time(inode, time, flags);
1775 return generic_update_time(inode, time, flags);
1779 * touch_atime - update the access time
1780 * @path: the &struct path to update
1781 * @inode: inode to update
1783 * Update the accessed time on an inode and mark it for writeback.
1784 * This function automatically handles read only file systems and media,
1785 * as well as the "noatime" flag and inode specific "noatime" markers.
1787 bool atime_needs_update(const struct path *path, struct inode *inode)
1789 struct vfsmount *mnt = path->mnt;
1790 struct timespec64 now;
1792 if (inode->i_flags & S_NOATIME)
1795 /* Atime updates will likely cause i_uid and i_gid to be written
1796 * back improprely if their true value is unknown to the vfs.
1798 if (HAS_UNMAPPED_ID(inode))
1801 if (IS_NOATIME(inode))
1803 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1806 if (mnt->mnt_flags & MNT_NOATIME)
1808 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1811 now = current_time(inode);
1813 if (!relatime_need_update(mnt, inode, now))
1816 if (timespec64_equal(&inode->i_atime, &now))
1822 void touch_atime(const struct path *path)
1824 struct vfsmount *mnt = path->mnt;
1825 struct inode *inode = d_inode(path->dentry);
1826 struct timespec64 now;
1828 if (!atime_needs_update(path, inode))
1831 if (!sb_start_write_trylock(inode->i_sb))
1834 if (__mnt_want_write(mnt) != 0)
1837 * File systems can error out when updating inodes if they need to
1838 * allocate new space to modify an inode (such is the case for
1839 * Btrfs), but since we touch atime while walking down the path we
1840 * really don't care if we failed to update the atime of the file,
1841 * so just ignore the return value.
1842 * We may also fail on filesystems that have the ability to make parts
1843 * of the fs read only, e.g. subvolumes in Btrfs.
1845 now = current_time(inode);
1846 update_time(inode, &now, S_ATIME);
1847 __mnt_drop_write(mnt);
1849 sb_end_write(inode->i_sb);
1851 EXPORT_SYMBOL(touch_atime);
1854 * The logic we want is
1856 * if suid or (sgid and xgrp)
1859 int should_remove_suid(struct dentry *dentry)
1861 umode_t mode = d_inode(dentry)->i_mode;
1864 /* suid always must be killed */
1865 if (unlikely(mode & S_ISUID))
1866 kill = ATTR_KILL_SUID;
1869 * sgid without any exec bits is just a mandatory locking mark; leave
1870 * it alone. If some exec bits are set, it's a real sgid; kill it.
1872 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1873 kill |= ATTR_KILL_SGID;
1875 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1880 EXPORT_SYMBOL(should_remove_suid);
1883 * Return mask of changes for notify_change() that need to be done as a
1884 * response to write or truncate. Return 0 if nothing has to be changed.
1885 * Negative value on error (change should be denied).
1887 int dentry_needs_remove_privs(struct dentry *dentry)
1889 struct inode *inode = d_inode(dentry);
1893 if (IS_NOSEC(inode))
1896 mask = should_remove_suid(dentry);
1897 ret = security_inode_need_killpriv(dentry);
1901 mask |= ATTR_KILL_PRIV;
1905 static int __remove_privs(struct dentry *dentry, int kill)
1907 struct iattr newattrs;
1909 newattrs.ia_valid = ATTR_FORCE | kill;
1911 * Note we call this on write, so notify_change will not
1912 * encounter any conflicting delegations:
1914 return notify_change(dentry, &newattrs, NULL);
1918 * Remove special file priviledges (suid, capabilities) when file is written
1921 int file_remove_privs(struct file *file)
1923 struct dentry *dentry = file_dentry(file);
1924 struct inode *inode = file_inode(file);
1929 * Fast path for nothing security related.
1930 * As well for non-regular files, e.g. blkdev inodes.
1931 * For example, blkdev_write_iter() might get here
1932 * trying to remove privs which it is not allowed to.
1934 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1937 kill = dentry_needs_remove_privs(dentry);
1941 error = __remove_privs(dentry, kill);
1943 inode_has_no_xattr(inode);
1947 EXPORT_SYMBOL(file_remove_privs);
1950 * file_update_time - update mtime and ctime time
1951 * @file: file accessed
1953 * Update the mtime and ctime members of an inode and mark the inode
1954 * for writeback. Note that this function is meant exclusively for
1955 * usage in the file write path of filesystems, and filesystems may
1956 * choose to explicitly ignore update via this function with the
1957 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1958 * timestamps are handled by the server. This can return an error for
1959 * file systems who need to allocate space in order to update an inode.
1962 int file_update_time(struct file *file)
1964 struct inode *inode = file_inode(file);
1965 struct timespec64 now;
1969 /* First try to exhaust all avenues to not sync */
1970 if (IS_NOCMTIME(inode))
1973 now = current_time(inode);
1974 if (!timespec64_equal(&inode->i_mtime, &now))
1977 if (!timespec64_equal(&inode->i_ctime, &now))
1980 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
1981 sync_it |= S_VERSION;
1986 /* Finally allowed to write? Takes lock. */
1987 if (__mnt_want_write_file(file))
1990 ret = update_time(inode, &now, sync_it);
1991 __mnt_drop_write_file(file);
1995 EXPORT_SYMBOL(file_update_time);
1997 /* Caller must hold the file's inode lock */
1998 int file_modified(struct file *file)
2003 * Clear the security bits if the process is not being run by root.
2004 * This keeps people from modifying setuid and setgid binaries.
2006 err = file_remove_privs(file);
2010 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2013 return file_update_time(file);
2015 EXPORT_SYMBOL(file_modified);
2017 int inode_needs_sync(struct inode *inode)
2021 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2025 EXPORT_SYMBOL(inode_needs_sync);
2028 * If we try to find an inode in the inode hash while it is being
2029 * deleted, we have to wait until the filesystem completes its
2030 * deletion before reporting that it isn't found. This function waits
2031 * until the deletion _might_ have completed. Callers are responsible
2032 * to recheck inode state.
2034 * It doesn't matter if I_NEW is not set initially, a call to
2035 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2038 static void __wait_on_freeing_inode(struct inode *inode)
2040 wait_queue_head_t *wq;
2041 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2042 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2043 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2044 spin_unlock(&inode->i_lock);
2045 spin_unlock(&inode_hash_lock);
2047 finish_wait(wq, &wait.wq_entry);
2048 spin_lock(&inode_hash_lock);
2051 static __initdata unsigned long ihash_entries;
2052 static int __init set_ihash_entries(char *str)
2056 ihash_entries = simple_strtoul(str, &str, 0);
2059 __setup("ihash_entries=", set_ihash_entries);
2062 * Initialize the waitqueues and inode hash table.
2064 void __init inode_init_early(void)
2066 /* If hashes are distributed across NUMA nodes, defer
2067 * hash allocation until vmalloc space is available.
2073 alloc_large_system_hash("Inode-cache",
2074 sizeof(struct hlist_head),
2077 HASH_EARLY | HASH_ZERO,
2084 void __init inode_init(void)
2086 /* inode slab cache */
2087 inode_cachep = kmem_cache_create("inode_cache",
2088 sizeof(struct inode),
2090 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2091 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2094 /* Hash may have been set up in inode_init_early */
2099 alloc_large_system_hash("Inode-cache",
2100 sizeof(struct hlist_head),
2110 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2112 inode->i_mode = mode;
2113 if (S_ISCHR(mode)) {
2114 inode->i_fop = &def_chr_fops;
2115 inode->i_rdev = rdev;
2116 } else if (S_ISBLK(mode)) {
2117 inode->i_fop = &def_blk_fops;
2118 inode->i_rdev = rdev;
2119 } else if (S_ISFIFO(mode))
2120 inode->i_fop = &pipefifo_fops;
2121 else if (S_ISSOCK(mode))
2122 ; /* leave it no_open_fops */
2124 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2125 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2128 EXPORT_SYMBOL(init_special_inode);
2131 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2133 * @dir: Directory inode
2134 * @mode: mode of the new inode
2136 void inode_init_owner(struct inode *inode, const struct inode *dir,
2139 inode->i_uid = current_fsuid();
2140 if (dir && dir->i_mode & S_ISGID) {
2141 inode->i_gid = dir->i_gid;
2143 /* Directories are special, and always inherit S_ISGID */
2146 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2147 !in_group_p(inode->i_gid) &&
2148 !capable_wrt_inode_uidgid(dir, CAP_FSETID))
2151 inode->i_gid = current_fsgid();
2152 inode->i_mode = mode;
2154 EXPORT_SYMBOL(inode_init_owner);
2157 * inode_owner_or_capable - check current task permissions to inode
2158 * @inode: inode being checked
2160 * Return true if current either has CAP_FOWNER in a namespace with the
2161 * inode owner uid mapped, or owns the file.
2163 bool inode_owner_or_capable(const struct inode *inode)
2165 struct user_namespace *ns;
2167 if (uid_eq(current_fsuid(), inode->i_uid))
2170 ns = current_user_ns();
2171 if (kuid_has_mapping(ns, inode->i_uid) && ns_capable(ns, CAP_FOWNER))
2175 EXPORT_SYMBOL(inode_owner_or_capable);
2178 * Direct i/o helper functions
2180 static void __inode_dio_wait(struct inode *inode)
2182 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2183 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2186 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2187 if (atomic_read(&inode->i_dio_count))
2189 } while (atomic_read(&inode->i_dio_count));
2190 finish_wait(wq, &q.wq_entry);
2194 * inode_dio_wait - wait for outstanding DIO requests to finish
2195 * @inode: inode to wait for
2197 * Waits for all pending direct I/O requests to finish so that we can
2198 * proceed with a truncate or equivalent operation.
2200 * Must be called under a lock that serializes taking new references
2201 * to i_dio_count, usually by inode->i_mutex.
2203 void inode_dio_wait(struct inode *inode)
2205 if (atomic_read(&inode->i_dio_count))
2206 __inode_dio_wait(inode);
2208 EXPORT_SYMBOL(inode_dio_wait);
2211 * inode_set_flags - atomically set some inode flags
2213 * Note: the caller should be holding i_mutex, or else be sure that
2214 * they have exclusive access to the inode structure (i.e., while the
2215 * inode is being instantiated). The reason for the cmpxchg() loop
2216 * --- which wouldn't be necessary if all code paths which modify
2217 * i_flags actually followed this rule, is that there is at least one
2218 * code path which doesn't today so we use cmpxchg() out of an abundance
2221 * In the long run, i_mutex is overkill, and we should probably look
2222 * at using the i_lock spinlock to protect i_flags, and then make sure
2223 * it is so documented in include/linux/fs.h and that all code follows
2224 * the locking convention!!
2226 void inode_set_flags(struct inode *inode, unsigned int flags,
2229 WARN_ON_ONCE(flags & ~mask);
2230 set_mask_bits(&inode->i_flags, mask, flags);
2232 EXPORT_SYMBOL(inode_set_flags);
2234 void inode_nohighmem(struct inode *inode)
2236 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2238 EXPORT_SYMBOL(inode_nohighmem);
2241 * timestamp_truncate - Truncate timespec to a granularity
2243 * @inode: inode being updated
2245 * Truncate a timespec to the granularity supported by the fs
2246 * containing the inode. Always rounds down. gran must
2247 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2249 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2251 struct super_block *sb = inode->i_sb;
2252 unsigned int gran = sb->s_time_gran;
2254 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2255 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2258 /* Avoid division in the common cases 1 ns and 1 s. */
2261 else if (gran == NSEC_PER_SEC)
2263 else if (gran > 1 && gran < NSEC_PER_SEC)
2264 t.tv_nsec -= t.tv_nsec % gran;
2266 WARN(1, "invalid file time granularity: %u", gran);
2269 EXPORT_SYMBOL(timestamp_truncate);
2272 * current_time - Return FS time
2275 * Return the current time truncated to the time granularity supported by
2278 * Note that inode and inode->sb cannot be NULL.
2279 * Otherwise, the function warns and returns time without truncation.
2281 struct timespec64 current_time(struct inode *inode)
2283 struct timespec64 now;
2285 ktime_get_coarse_real_ts64(&now);
2287 if (unlikely(!inode->i_sb)) {
2288 WARN(1, "current_time() called with uninitialized super_block in the inode");
2292 return timestamp_truncate(now, inode);
2294 EXPORT_SYMBOL(current_time);
2297 * Generic function to check FS_IOC_SETFLAGS values and reject any invalid
2300 * Note: the caller should be holding i_mutex, or else be sure that they have
2301 * exclusive access to the inode structure.
2303 int vfs_ioc_setflags_prepare(struct inode *inode, unsigned int oldflags,
2307 * The IMMUTABLE and APPEND_ONLY flags can only be changed by
2308 * the relevant capability.
2310 * This test looks nicer. Thanks to Pauline Middelink
2312 if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL) &&
2313 !capable(CAP_LINUX_IMMUTABLE))
2316 return fscrypt_prepare_setflags(inode, oldflags, flags);
2318 EXPORT_SYMBOL(vfs_ioc_setflags_prepare);
2321 * Generic function to check FS_IOC_FSSETXATTR values and reject any invalid
2324 * Note: the caller should be holding i_mutex, or else be sure that they have
2325 * exclusive access to the inode structure.
2327 int vfs_ioc_fssetxattr_check(struct inode *inode, const struct fsxattr *old_fa,
2331 * Can't modify an immutable/append-only file unless we have
2332 * appropriate permission.
2334 if ((old_fa->fsx_xflags ^ fa->fsx_xflags) &
2335 (FS_XFLAG_IMMUTABLE | FS_XFLAG_APPEND) &&
2336 !capable(CAP_LINUX_IMMUTABLE))
2340 * Project Quota ID state is only allowed to change from within the init
2341 * namespace. Enforce that restriction only if we are trying to change
2342 * the quota ID state. Everything else is allowed in user namespaces.
2344 if (current_user_ns() != &init_user_ns) {
2345 if (old_fa->fsx_projid != fa->fsx_projid)
2347 if ((old_fa->fsx_xflags ^ fa->fsx_xflags) &
2348 FS_XFLAG_PROJINHERIT)
2352 /* Check extent size hints. */
2353 if ((fa->fsx_xflags & FS_XFLAG_EXTSIZE) && !S_ISREG(inode->i_mode))
2356 if ((fa->fsx_xflags & FS_XFLAG_EXTSZINHERIT) &&
2357 !S_ISDIR(inode->i_mode))
2360 if ((fa->fsx_xflags & FS_XFLAG_COWEXTSIZE) &&
2361 !S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2365 * It is only valid to set the DAX flag on regular files and
2366 * directories on filesystems.
2368 if ((fa->fsx_xflags & FS_XFLAG_DAX) &&
2369 !(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
2372 /* Extent size hints of zero turn off the flags. */
2373 if (fa->fsx_extsize == 0)
2374 fa->fsx_xflags &= ~(FS_XFLAG_EXTSIZE | FS_XFLAG_EXTSZINHERIT);
2375 if (fa->fsx_cowextsize == 0)
2376 fa->fsx_xflags &= ~FS_XFLAG_COWEXTSIZE;
2380 EXPORT_SYMBOL(vfs_ioc_fssetxattr_check);