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(), inode->i_io_list
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 spin_lock_init(&inode->i_lock);
171 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
173 init_rwsem(&inode->i_rwsem);
174 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
176 atomic_set(&inode->i_dio_count, 0);
178 mapping->a_ops = &empty_aops;
179 mapping->host = inode;
181 if (sb->s_type->fs_flags & FS_THP_SUPPORT)
182 __set_bit(AS_THP_SUPPORT, &mapping->flags);
184 atomic_set(&mapping->i_mmap_writable, 0);
185 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
186 atomic_set(&mapping->nr_thps, 0);
188 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
189 mapping->private_data = NULL;
190 mapping->writeback_index = 0;
191 init_rwsem(&mapping->invalidate_lock);
192 lockdep_set_class_and_name(&mapping->invalidate_lock,
193 &sb->s_type->invalidate_lock_key,
194 "mapping.invalidate_lock");
195 inode->i_private = NULL;
196 inode->i_mapping = mapping;
197 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
198 #ifdef CONFIG_FS_POSIX_ACL
199 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
202 #ifdef CONFIG_FSNOTIFY
203 inode->i_fsnotify_mask = 0;
205 inode->i_flctx = NULL;
207 if (unlikely(security_inode_alloc(inode)))
209 this_cpu_inc(nr_inodes);
213 EXPORT_SYMBOL(inode_init_always);
215 void free_inode_nonrcu(struct inode *inode)
217 kmem_cache_free(inode_cachep, inode);
219 EXPORT_SYMBOL(free_inode_nonrcu);
221 static void i_callback(struct rcu_head *head)
223 struct inode *inode = container_of(head, struct inode, i_rcu);
224 if (inode->free_inode)
225 inode->free_inode(inode);
227 free_inode_nonrcu(inode);
230 static struct inode *alloc_inode(struct super_block *sb)
232 const struct super_operations *ops = sb->s_op;
235 if (ops->alloc_inode)
236 inode = ops->alloc_inode(sb);
238 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
243 if (unlikely(inode_init_always(sb, inode))) {
244 if (ops->destroy_inode) {
245 ops->destroy_inode(inode);
246 if (!ops->free_inode)
249 inode->free_inode = ops->free_inode;
250 i_callback(&inode->i_rcu);
257 void __destroy_inode(struct inode *inode)
259 BUG_ON(inode_has_buffers(inode));
260 inode_detach_wb(inode);
261 security_inode_free(inode);
262 fsnotify_inode_delete(inode);
263 locks_free_lock_context(inode);
264 if (!inode->i_nlink) {
265 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
266 atomic_long_dec(&inode->i_sb->s_remove_count);
269 #ifdef CONFIG_FS_POSIX_ACL
270 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
271 posix_acl_release(inode->i_acl);
272 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
273 posix_acl_release(inode->i_default_acl);
275 this_cpu_dec(nr_inodes);
277 EXPORT_SYMBOL(__destroy_inode);
279 static void destroy_inode(struct inode *inode)
281 const struct super_operations *ops = inode->i_sb->s_op;
283 BUG_ON(!list_empty(&inode->i_lru));
284 __destroy_inode(inode);
285 if (ops->destroy_inode) {
286 ops->destroy_inode(inode);
287 if (!ops->free_inode)
290 inode->free_inode = ops->free_inode;
291 call_rcu(&inode->i_rcu, i_callback);
295 * drop_nlink - directly drop an inode's link count
298 * This is a low-level filesystem helper to replace any
299 * direct filesystem manipulation of i_nlink. In cases
300 * where we are attempting to track writes to the
301 * filesystem, a decrement to zero means an imminent
302 * write when the file is truncated and actually unlinked
305 void drop_nlink(struct inode *inode)
307 WARN_ON(inode->i_nlink == 0);
310 atomic_long_inc(&inode->i_sb->s_remove_count);
312 EXPORT_SYMBOL(drop_nlink);
315 * clear_nlink - directly zero an inode's link count
318 * This is a low-level filesystem helper to replace any
319 * direct filesystem manipulation of i_nlink. See
320 * drop_nlink() for why we care about i_nlink hitting zero.
322 void clear_nlink(struct inode *inode)
324 if (inode->i_nlink) {
325 inode->__i_nlink = 0;
326 atomic_long_inc(&inode->i_sb->s_remove_count);
329 EXPORT_SYMBOL(clear_nlink);
332 * set_nlink - directly set an inode's link count
334 * @nlink: new nlink (should be non-zero)
336 * This is a low-level filesystem helper to replace any
337 * direct filesystem manipulation of i_nlink.
339 void set_nlink(struct inode *inode, unsigned int nlink)
344 /* Yes, some filesystems do change nlink from zero to one */
345 if (inode->i_nlink == 0)
346 atomic_long_dec(&inode->i_sb->s_remove_count);
348 inode->__i_nlink = nlink;
351 EXPORT_SYMBOL(set_nlink);
354 * inc_nlink - directly increment an inode's link count
357 * This is a low-level filesystem helper to replace any
358 * direct filesystem manipulation of i_nlink. Currently,
359 * it is only here for parity with dec_nlink().
361 void inc_nlink(struct inode *inode)
363 if (unlikely(inode->i_nlink == 0)) {
364 WARN_ON(!(inode->i_state & I_LINKABLE));
365 atomic_long_dec(&inode->i_sb->s_remove_count);
370 EXPORT_SYMBOL(inc_nlink);
372 static void __address_space_init_once(struct address_space *mapping)
374 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
375 init_rwsem(&mapping->i_mmap_rwsem);
376 INIT_LIST_HEAD(&mapping->private_list);
377 spin_lock_init(&mapping->private_lock);
378 mapping->i_mmap = RB_ROOT_CACHED;
381 void address_space_init_once(struct address_space *mapping)
383 memset(mapping, 0, sizeof(*mapping));
384 __address_space_init_once(mapping);
386 EXPORT_SYMBOL(address_space_init_once);
389 * These are initializations that only need to be done
390 * once, because the fields are idempotent across use
391 * of the inode, so let the slab aware of that.
393 void inode_init_once(struct inode *inode)
395 memset(inode, 0, sizeof(*inode));
396 INIT_HLIST_NODE(&inode->i_hash);
397 INIT_LIST_HEAD(&inode->i_devices);
398 INIT_LIST_HEAD(&inode->i_io_list);
399 INIT_LIST_HEAD(&inode->i_wb_list);
400 INIT_LIST_HEAD(&inode->i_lru);
401 __address_space_init_once(&inode->i_data);
402 i_size_ordered_init(inode);
404 EXPORT_SYMBOL(inode_init_once);
406 static void init_once(void *foo)
408 struct inode *inode = (struct inode *) foo;
410 inode_init_once(inode);
414 * inode->i_lock must be held
416 void __iget(struct inode *inode)
418 atomic_inc(&inode->i_count);
422 * get additional reference to inode; caller must already hold one.
424 void ihold(struct inode *inode)
426 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
428 EXPORT_SYMBOL(ihold);
430 static void inode_lru_list_add(struct inode *inode)
432 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
433 this_cpu_inc(nr_unused);
435 inode->i_state |= I_REFERENCED;
439 * Add inode to LRU if needed (inode is unused and clean).
441 * Needs inode->i_lock held.
443 void inode_add_lru(struct inode *inode)
445 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
446 I_FREEING | I_WILL_FREE)) &&
447 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
448 inode_lru_list_add(inode);
452 static void inode_lru_list_del(struct inode *inode)
455 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
456 this_cpu_dec(nr_unused);
460 * inode_sb_list_add - add inode to the superblock list of inodes
461 * @inode: inode to add
463 void inode_sb_list_add(struct inode *inode)
465 spin_lock(&inode->i_sb->s_inode_list_lock);
466 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
467 spin_unlock(&inode->i_sb->s_inode_list_lock);
469 EXPORT_SYMBOL_GPL(inode_sb_list_add);
471 static inline void inode_sb_list_del(struct inode *inode)
473 if (!list_empty(&inode->i_sb_list)) {
474 spin_lock(&inode->i_sb->s_inode_list_lock);
475 list_del_init(&inode->i_sb_list);
476 spin_unlock(&inode->i_sb->s_inode_list_lock);
480 static unsigned long hash(struct super_block *sb, unsigned long hashval)
484 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
486 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
487 return tmp & i_hash_mask;
491 * __insert_inode_hash - hash an inode
492 * @inode: unhashed inode
493 * @hashval: unsigned long value used to locate this object in the
496 * Add an inode to the inode hash for this superblock.
498 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
500 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
502 spin_lock(&inode_hash_lock);
503 spin_lock(&inode->i_lock);
504 hlist_add_head_rcu(&inode->i_hash, b);
505 spin_unlock(&inode->i_lock);
506 spin_unlock(&inode_hash_lock);
508 EXPORT_SYMBOL(__insert_inode_hash);
511 * __remove_inode_hash - remove an inode from the hash
512 * @inode: inode to unhash
514 * Remove an inode from the superblock.
516 void __remove_inode_hash(struct inode *inode)
518 spin_lock(&inode_hash_lock);
519 spin_lock(&inode->i_lock);
520 hlist_del_init_rcu(&inode->i_hash);
521 spin_unlock(&inode->i_lock);
522 spin_unlock(&inode_hash_lock);
524 EXPORT_SYMBOL(__remove_inode_hash);
526 void clear_inode(struct inode *inode)
529 * We have to cycle the i_pages lock here because reclaim can be in the
530 * process of removing the last page (in __delete_from_page_cache())
531 * and we must not free the mapping under it.
533 xa_lock_irq(&inode->i_data.i_pages);
534 BUG_ON(inode->i_data.nrpages);
536 * Almost always, mapping_empty(&inode->i_data) here; but there are
537 * two known and long-standing ways in which nodes may get left behind
538 * (when deep radix-tree node allocation failed partway; or when THP
539 * collapse_file() failed). Until those two known cases are cleaned up,
540 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
541 * nor even WARN_ON(!mapping_empty).
543 xa_unlock_irq(&inode->i_data.i_pages);
544 BUG_ON(!list_empty(&inode->i_data.private_list));
545 BUG_ON(!(inode->i_state & I_FREEING));
546 BUG_ON(inode->i_state & I_CLEAR);
547 BUG_ON(!list_empty(&inode->i_wb_list));
548 /* don't need i_lock here, no concurrent mods to i_state */
549 inode->i_state = I_FREEING | I_CLEAR;
551 EXPORT_SYMBOL(clear_inode);
554 * Free the inode passed in, removing it from the lists it is still connected
555 * to. We remove any pages still attached to the inode and wait for any IO that
556 * is still in progress before finally destroying the inode.
558 * An inode must already be marked I_FREEING so that we avoid the inode being
559 * moved back onto lists if we race with other code that manipulates the lists
560 * (e.g. writeback_single_inode). The caller is responsible for setting this.
562 * An inode must already be removed from the LRU list before being evicted from
563 * the cache. This should occur atomically with setting the I_FREEING state
564 * flag, so no inodes here should ever be on the LRU when being evicted.
566 static void evict(struct inode *inode)
568 const struct super_operations *op = inode->i_sb->s_op;
570 BUG_ON(!(inode->i_state & I_FREEING));
571 BUG_ON(!list_empty(&inode->i_lru));
573 if (!list_empty(&inode->i_io_list))
574 inode_io_list_del(inode);
576 inode_sb_list_del(inode);
579 * Wait for flusher thread to be done with the inode so that filesystem
580 * does not start destroying it while writeback is still running. Since
581 * the inode has I_FREEING set, flusher thread won't start new work on
582 * the inode. We just have to wait for running writeback to finish.
584 inode_wait_for_writeback(inode);
586 if (op->evict_inode) {
587 op->evict_inode(inode);
589 truncate_inode_pages_final(&inode->i_data);
592 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
595 remove_inode_hash(inode);
597 spin_lock(&inode->i_lock);
598 wake_up_bit(&inode->i_state, __I_NEW);
599 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
600 spin_unlock(&inode->i_lock);
602 destroy_inode(inode);
606 * dispose_list - dispose of the contents of a local list
607 * @head: the head of the list to free
609 * Dispose-list gets a local list with local inodes in it, so it doesn't
610 * need to worry about list corruption and SMP locks.
612 static void dispose_list(struct list_head *head)
614 while (!list_empty(head)) {
617 inode = list_first_entry(head, struct inode, i_lru);
618 list_del_init(&inode->i_lru);
626 * evict_inodes - evict all evictable inodes for a superblock
627 * @sb: superblock to operate on
629 * Make sure that no inodes with zero refcount are retained. This is
630 * called by superblock shutdown after having SB_ACTIVE flag removed,
631 * so any inode reaching zero refcount during or after that call will
632 * be immediately evicted.
634 void evict_inodes(struct super_block *sb)
636 struct inode *inode, *next;
640 spin_lock(&sb->s_inode_list_lock);
641 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
642 if (atomic_read(&inode->i_count))
645 spin_lock(&inode->i_lock);
646 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
647 spin_unlock(&inode->i_lock);
651 inode->i_state |= I_FREEING;
652 inode_lru_list_del(inode);
653 spin_unlock(&inode->i_lock);
654 list_add(&inode->i_lru, &dispose);
657 * We can have a ton of inodes to evict at unmount time given
658 * enough memory, check to see if we need to go to sleep for a
659 * bit so we don't livelock.
661 if (need_resched()) {
662 spin_unlock(&sb->s_inode_list_lock);
664 dispose_list(&dispose);
668 spin_unlock(&sb->s_inode_list_lock);
670 dispose_list(&dispose);
672 EXPORT_SYMBOL_GPL(evict_inodes);
675 * invalidate_inodes - attempt to free all inodes on a superblock
676 * @sb: superblock to operate on
677 * @kill_dirty: flag to guide handling of dirty inodes
679 * Attempts to free all inodes for a given superblock. If there were any
680 * busy inodes return a non-zero value, else zero.
681 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
684 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
687 struct inode *inode, *next;
691 spin_lock(&sb->s_inode_list_lock);
692 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
693 spin_lock(&inode->i_lock);
694 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
695 spin_unlock(&inode->i_lock);
698 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
699 spin_unlock(&inode->i_lock);
703 if (atomic_read(&inode->i_count)) {
704 spin_unlock(&inode->i_lock);
709 inode->i_state |= I_FREEING;
710 inode_lru_list_del(inode);
711 spin_unlock(&inode->i_lock);
712 list_add(&inode->i_lru, &dispose);
713 if (need_resched()) {
714 spin_unlock(&sb->s_inode_list_lock);
716 dispose_list(&dispose);
720 spin_unlock(&sb->s_inode_list_lock);
722 dispose_list(&dispose);
728 * Isolate the inode from the LRU in preparation for freeing it.
730 * Any inodes which are pinned purely because of attached pagecache have their
731 * pagecache removed. If the inode has metadata buffers attached to
732 * mapping->private_list then try to remove them.
734 * If the inode has the I_REFERENCED flag set, then it means that it has been
735 * used recently - the flag is set in iput_final(). When we encounter such an
736 * inode, clear the flag and move it to the back of the LRU so it gets another
737 * pass through the LRU before it gets reclaimed. This is necessary because of
738 * the fact we are doing lazy LRU updates to minimise lock contention so the
739 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
740 * with this flag set because they are the inodes that are out of order.
742 static enum lru_status inode_lru_isolate(struct list_head *item,
743 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
745 struct list_head *freeable = arg;
746 struct inode *inode = container_of(item, struct inode, i_lru);
749 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
750 * If we fail to get the lock, just skip it.
752 if (!spin_trylock(&inode->i_lock))
756 * Referenced or dirty inodes are still in use. Give them another pass
757 * through the LRU as we canot reclaim them now.
759 if (atomic_read(&inode->i_count) ||
760 (inode->i_state & ~I_REFERENCED)) {
761 list_lru_isolate(lru, &inode->i_lru);
762 spin_unlock(&inode->i_lock);
763 this_cpu_dec(nr_unused);
767 /* recently referenced inodes get one more pass */
768 if (inode->i_state & I_REFERENCED) {
769 inode->i_state &= ~I_REFERENCED;
770 spin_unlock(&inode->i_lock);
774 if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
776 spin_unlock(&inode->i_lock);
777 spin_unlock(lru_lock);
778 if (remove_inode_buffers(inode)) {
780 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
781 if (current_is_kswapd())
782 __count_vm_events(KSWAPD_INODESTEAL, reap);
784 __count_vm_events(PGINODESTEAL, reap);
785 if (current->reclaim_state)
786 current->reclaim_state->reclaimed_slab += reap;
793 WARN_ON(inode->i_state & I_NEW);
794 inode->i_state |= I_FREEING;
795 list_lru_isolate_move(lru, &inode->i_lru, freeable);
796 spin_unlock(&inode->i_lock);
798 this_cpu_dec(nr_unused);
803 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
804 * This is called from the superblock shrinker function with a number of inodes
805 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
806 * then are freed outside inode_lock by dispose_list().
808 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
813 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
814 inode_lru_isolate, &freeable);
815 dispose_list(&freeable);
819 static void __wait_on_freeing_inode(struct inode *inode);
821 * Called with the inode lock held.
823 static struct inode *find_inode(struct super_block *sb,
824 struct hlist_head *head,
825 int (*test)(struct inode *, void *),
828 struct inode *inode = NULL;
831 hlist_for_each_entry(inode, head, i_hash) {
832 if (inode->i_sb != sb)
834 if (!test(inode, data))
836 spin_lock(&inode->i_lock);
837 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
838 __wait_on_freeing_inode(inode);
841 if (unlikely(inode->i_state & I_CREATING)) {
842 spin_unlock(&inode->i_lock);
843 return ERR_PTR(-ESTALE);
846 spin_unlock(&inode->i_lock);
853 * find_inode_fast is the fast path version of find_inode, see the comment at
854 * iget_locked for details.
856 static struct inode *find_inode_fast(struct super_block *sb,
857 struct hlist_head *head, unsigned long ino)
859 struct inode *inode = NULL;
862 hlist_for_each_entry(inode, head, i_hash) {
863 if (inode->i_ino != ino)
865 if (inode->i_sb != sb)
867 spin_lock(&inode->i_lock);
868 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
869 __wait_on_freeing_inode(inode);
872 if (unlikely(inode->i_state & I_CREATING)) {
873 spin_unlock(&inode->i_lock);
874 return ERR_PTR(-ESTALE);
877 spin_unlock(&inode->i_lock);
884 * Each cpu owns a range of LAST_INO_BATCH numbers.
885 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
886 * to renew the exhausted range.
888 * This does not significantly increase overflow rate because every CPU can
889 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
890 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
891 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
892 * overflow rate by 2x, which does not seem too significant.
894 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
895 * error if st_ino won't fit in target struct field. Use 32bit counter
896 * here to attempt to avoid that.
898 #define LAST_INO_BATCH 1024
899 static DEFINE_PER_CPU(unsigned int, last_ino);
901 unsigned int get_next_ino(void)
903 unsigned int *p = &get_cpu_var(last_ino);
904 unsigned int res = *p;
907 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
908 static atomic_t shared_last_ino;
909 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
911 res = next - LAST_INO_BATCH;
916 /* get_next_ino should not provide a 0 inode number */
920 put_cpu_var(last_ino);
923 EXPORT_SYMBOL(get_next_ino);
926 * new_inode_pseudo - obtain an inode
929 * Allocates a new inode for given superblock.
930 * Inode wont be chained in superblock s_inodes list
932 * - fs can't be unmount
933 * - quotas, fsnotify, writeback can't work
935 struct inode *new_inode_pseudo(struct super_block *sb)
937 struct inode *inode = alloc_inode(sb);
940 spin_lock(&inode->i_lock);
942 spin_unlock(&inode->i_lock);
943 INIT_LIST_HEAD(&inode->i_sb_list);
949 * new_inode - obtain an inode
952 * Allocates a new inode for given superblock. The default gfp_mask
953 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
954 * If HIGHMEM pages are unsuitable or it is known that pages allocated
955 * for the page cache are not reclaimable or migratable,
956 * mapping_set_gfp_mask() must be called with suitable flags on the
957 * newly created inode's mapping
960 struct inode *new_inode(struct super_block *sb)
964 spin_lock_prefetch(&sb->s_inode_list_lock);
966 inode = new_inode_pseudo(sb);
968 inode_sb_list_add(inode);
971 EXPORT_SYMBOL(new_inode);
973 #ifdef CONFIG_DEBUG_LOCK_ALLOC
974 void lockdep_annotate_inode_mutex_key(struct inode *inode)
976 if (S_ISDIR(inode->i_mode)) {
977 struct file_system_type *type = inode->i_sb->s_type;
979 /* Set new key only if filesystem hasn't already changed it */
980 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
982 * ensure nobody is actually holding i_mutex
984 // mutex_destroy(&inode->i_mutex);
985 init_rwsem(&inode->i_rwsem);
986 lockdep_set_class(&inode->i_rwsem,
987 &type->i_mutex_dir_key);
991 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
995 * unlock_new_inode - clear the I_NEW state and wake up any waiters
996 * @inode: new inode to unlock
998 * Called when the inode is fully initialised to clear the new state of the
999 * inode and wake up anyone waiting for the inode to finish initialisation.
1001 void unlock_new_inode(struct inode *inode)
1003 lockdep_annotate_inode_mutex_key(inode);
1004 spin_lock(&inode->i_lock);
1005 WARN_ON(!(inode->i_state & I_NEW));
1006 inode->i_state &= ~I_NEW & ~I_CREATING;
1008 wake_up_bit(&inode->i_state, __I_NEW);
1009 spin_unlock(&inode->i_lock);
1011 EXPORT_SYMBOL(unlock_new_inode);
1013 void discard_new_inode(struct inode *inode)
1015 lockdep_annotate_inode_mutex_key(inode);
1016 spin_lock(&inode->i_lock);
1017 WARN_ON(!(inode->i_state & I_NEW));
1018 inode->i_state &= ~I_NEW;
1020 wake_up_bit(&inode->i_state, __I_NEW);
1021 spin_unlock(&inode->i_lock);
1024 EXPORT_SYMBOL(discard_new_inode);
1027 * lock_two_inodes - lock two inodes (may be regular files but also dirs)
1029 * Lock any non-NULL argument. The caller must make sure that if he is passing
1030 * in two directories, one is not ancestor of the other. Zero, one or two
1031 * objects may be locked by this function.
1033 * @inode1: first inode to lock
1034 * @inode2: second inode to lock
1035 * @subclass1: inode lock subclass for the first lock obtained
1036 * @subclass2: inode lock subclass for the second lock obtained
1038 void lock_two_inodes(struct inode *inode1, struct inode *inode2,
1039 unsigned subclass1, unsigned subclass2)
1041 if (!inode1 || !inode2) {
1043 * Make sure @subclass1 will be used for the acquired lock.
1044 * This is not strictly necessary (no current caller cares) but
1045 * let's keep things consistent.
1048 swap(inode1, inode2);
1053 * If one object is directory and the other is not, we must make sure
1054 * to lock directory first as the other object may be its child.
1056 if (S_ISDIR(inode2->i_mode) == S_ISDIR(inode1->i_mode)) {
1057 if (inode1 > inode2)
1058 swap(inode1, inode2);
1059 } else if (!S_ISDIR(inode1->i_mode))
1060 swap(inode1, inode2);
1063 inode_lock_nested(inode1, subclass1);
1064 if (inode2 && inode2 != inode1)
1065 inode_lock_nested(inode2, subclass2);
1069 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1071 * Lock any non-NULL argument that is not a directory.
1072 * Zero, one or two objects may be locked by this function.
1074 * @inode1: first inode to lock
1075 * @inode2: second inode to lock
1077 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1079 if (inode1 > inode2)
1080 swap(inode1, inode2);
1082 if (inode1 && !S_ISDIR(inode1->i_mode))
1084 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1085 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1087 EXPORT_SYMBOL(lock_two_nondirectories);
1090 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1091 * @inode1: first inode to unlock
1092 * @inode2: second inode to unlock
1094 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1096 if (inode1 && !S_ISDIR(inode1->i_mode))
1097 inode_unlock(inode1);
1098 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1099 inode_unlock(inode2);
1101 EXPORT_SYMBOL(unlock_two_nondirectories);
1104 * inode_insert5 - obtain an inode from a mounted file system
1105 * @inode: pre-allocated inode to use for insert to cache
1106 * @hashval: hash value (usually inode number) to get
1107 * @test: callback used for comparisons between inodes
1108 * @set: callback used to initialize a new struct inode
1109 * @data: opaque data pointer to pass to @test and @set
1111 * Search for the inode specified by @hashval and @data in the inode cache,
1112 * and if present it is return it with an increased reference count. This is
1113 * a variant of iget5_locked() for callers that don't want to fail on memory
1114 * allocation of inode.
1116 * If the inode is not in cache, insert the pre-allocated inode to cache and
1117 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1118 * to fill it in before unlocking it via unlock_new_inode().
1120 * Note both @test and @set are called with the inode_hash_lock held, so can't
1123 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1124 int (*test)(struct inode *, void *),
1125 int (*set)(struct inode *, void *), void *data)
1127 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1129 bool creating = inode->i_state & I_CREATING;
1132 spin_lock(&inode_hash_lock);
1133 old = find_inode(inode->i_sb, head, test, data);
1134 if (unlikely(old)) {
1136 * Uhhuh, somebody else created the same inode under us.
1137 * Use the old inode instead of the preallocated one.
1139 spin_unlock(&inode_hash_lock);
1143 if (unlikely(inode_unhashed(old))) {
1150 if (set && unlikely(set(inode, data))) {
1156 * Return the locked inode with I_NEW set, the
1157 * caller is responsible for filling in the contents
1159 spin_lock(&inode->i_lock);
1160 inode->i_state |= I_NEW;
1161 hlist_add_head_rcu(&inode->i_hash, head);
1162 spin_unlock(&inode->i_lock);
1164 inode_sb_list_add(inode);
1166 spin_unlock(&inode_hash_lock);
1170 EXPORT_SYMBOL(inode_insert5);
1173 * iget5_locked - obtain an inode from a mounted file system
1174 * @sb: super block of file system
1175 * @hashval: hash value (usually inode number) to get
1176 * @test: callback used for comparisons between inodes
1177 * @set: callback used to initialize a new struct inode
1178 * @data: opaque data pointer to pass to @test and @set
1180 * Search for the inode specified by @hashval and @data in the inode cache,
1181 * and if present it is return it with an increased reference count. This is
1182 * a generalized version of iget_locked() for file systems where the inode
1183 * number is not sufficient for unique identification of an inode.
1185 * If the inode is not in cache, allocate a new inode and return it locked,
1186 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1187 * before unlocking it via unlock_new_inode().
1189 * Note both @test and @set are called with the inode_hash_lock held, so can't
1192 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1193 int (*test)(struct inode *, void *),
1194 int (*set)(struct inode *, void *), void *data)
1196 struct inode *inode = ilookup5(sb, hashval, test, data);
1199 struct inode *new = alloc_inode(sb);
1203 inode = inode_insert5(new, hashval, test, set, data);
1204 if (unlikely(inode != new))
1210 EXPORT_SYMBOL(iget5_locked);
1213 * iget_locked - obtain an inode from a mounted file system
1214 * @sb: super block of file system
1215 * @ino: inode number to get
1217 * Search for the inode specified by @ino in the inode cache and if present
1218 * return it with an increased reference count. This is for file systems
1219 * where the inode number is sufficient for unique identification of an inode.
1221 * If the inode is not in cache, allocate a new inode and return it locked,
1222 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1223 * before unlocking it via unlock_new_inode().
1225 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1227 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1228 struct inode *inode;
1230 spin_lock(&inode_hash_lock);
1231 inode = find_inode_fast(sb, head, ino);
1232 spin_unlock(&inode_hash_lock);
1236 wait_on_inode(inode);
1237 if (unlikely(inode_unhashed(inode))) {
1244 inode = alloc_inode(sb);
1248 spin_lock(&inode_hash_lock);
1249 /* We released the lock, so.. */
1250 old = find_inode_fast(sb, head, ino);
1253 spin_lock(&inode->i_lock);
1254 inode->i_state = I_NEW;
1255 hlist_add_head_rcu(&inode->i_hash, head);
1256 spin_unlock(&inode->i_lock);
1257 inode_sb_list_add(inode);
1258 spin_unlock(&inode_hash_lock);
1260 /* Return the locked inode with I_NEW set, the
1261 * caller is responsible for filling in the contents
1267 * Uhhuh, somebody else created the same inode under
1268 * us. Use the old inode instead of the one we just
1271 spin_unlock(&inode_hash_lock);
1272 destroy_inode(inode);
1276 wait_on_inode(inode);
1277 if (unlikely(inode_unhashed(inode))) {
1284 EXPORT_SYMBOL(iget_locked);
1287 * search the inode cache for a matching inode number.
1288 * If we find one, then the inode number we are trying to
1289 * allocate is not unique and so we should not use it.
1291 * Returns 1 if the inode number is unique, 0 if it is not.
1293 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1295 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1296 struct inode *inode;
1298 hlist_for_each_entry_rcu(inode, b, i_hash) {
1299 if (inode->i_ino == ino && inode->i_sb == sb)
1306 * iunique - get a unique inode number
1308 * @max_reserved: highest reserved inode number
1310 * Obtain an inode number that is unique on the system for a given
1311 * superblock. This is used by file systems that have no natural
1312 * permanent inode numbering system. An inode number is returned that
1313 * is higher than the reserved limit but unique.
1316 * With a large number of inodes live on the file system this function
1317 * currently becomes quite slow.
1319 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1322 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1323 * error if st_ino won't fit in target struct field. Use 32bit counter
1324 * here to attempt to avoid that.
1326 static DEFINE_SPINLOCK(iunique_lock);
1327 static unsigned int counter;
1331 spin_lock(&iunique_lock);
1333 if (counter <= max_reserved)
1334 counter = max_reserved + 1;
1336 } while (!test_inode_iunique(sb, res));
1337 spin_unlock(&iunique_lock);
1342 EXPORT_SYMBOL(iunique);
1344 struct inode *igrab(struct inode *inode)
1346 spin_lock(&inode->i_lock);
1347 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1349 spin_unlock(&inode->i_lock);
1351 spin_unlock(&inode->i_lock);
1353 * Handle the case where s_op->clear_inode is not been
1354 * called yet, and somebody is calling igrab
1355 * while the inode is getting freed.
1361 EXPORT_SYMBOL(igrab);
1364 * ilookup5_nowait - search for an inode in the inode cache
1365 * @sb: super block of file system to search
1366 * @hashval: hash value (usually inode number) to search for
1367 * @test: callback used for comparisons between inodes
1368 * @data: opaque data pointer to pass to @test
1370 * Search for the inode specified by @hashval and @data in the inode cache.
1371 * If the inode is in the cache, the inode is returned with an incremented
1374 * Note: I_NEW is not waited upon so you have to be very careful what you do
1375 * with the returned inode. You probably should be using ilookup5() instead.
1377 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1379 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1380 int (*test)(struct inode *, void *), void *data)
1382 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1383 struct inode *inode;
1385 spin_lock(&inode_hash_lock);
1386 inode = find_inode(sb, head, test, data);
1387 spin_unlock(&inode_hash_lock);
1389 return IS_ERR(inode) ? NULL : inode;
1391 EXPORT_SYMBOL(ilookup5_nowait);
1394 * ilookup5 - search for an inode in the inode cache
1395 * @sb: super block of file system to search
1396 * @hashval: hash value (usually inode number) to search for
1397 * @test: callback used for comparisons between inodes
1398 * @data: opaque data pointer to pass to @test
1400 * Search for the inode specified by @hashval and @data in the inode cache,
1401 * and if the inode is in the cache, return the inode with an incremented
1402 * reference count. Waits on I_NEW before returning the inode.
1403 * returned with an incremented reference count.
1405 * This is a generalized version of ilookup() for file systems where the
1406 * inode number is not sufficient for unique identification of an inode.
1408 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1410 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1411 int (*test)(struct inode *, void *), void *data)
1413 struct inode *inode;
1415 inode = ilookup5_nowait(sb, hashval, test, data);
1417 wait_on_inode(inode);
1418 if (unlikely(inode_unhashed(inode))) {
1425 EXPORT_SYMBOL(ilookup5);
1428 * ilookup - search for an inode in the inode cache
1429 * @sb: super block of file system to search
1430 * @ino: inode number to search for
1432 * Search for the inode @ino in the inode cache, and if the inode is in the
1433 * cache, the inode is returned with an incremented reference count.
1435 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1437 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1438 struct inode *inode;
1440 spin_lock(&inode_hash_lock);
1441 inode = find_inode_fast(sb, head, ino);
1442 spin_unlock(&inode_hash_lock);
1447 wait_on_inode(inode);
1448 if (unlikely(inode_unhashed(inode))) {
1455 EXPORT_SYMBOL(ilookup);
1458 * find_inode_nowait - find an inode in the inode cache
1459 * @sb: super block of file system to search
1460 * @hashval: hash value (usually inode number) to search for
1461 * @match: callback used for comparisons between inodes
1462 * @data: opaque data pointer to pass to @match
1464 * Search for the inode specified by @hashval and @data in the inode
1465 * cache, where the helper function @match will return 0 if the inode
1466 * does not match, 1 if the inode does match, and -1 if the search
1467 * should be stopped. The @match function must be responsible for
1468 * taking the i_lock spin_lock and checking i_state for an inode being
1469 * freed or being initialized, and incrementing the reference count
1470 * before returning 1. It also must not sleep, since it is called with
1471 * the inode_hash_lock spinlock held.
1473 * This is a even more generalized version of ilookup5() when the
1474 * function must never block --- find_inode() can block in
1475 * __wait_on_freeing_inode() --- or when the caller can not increment
1476 * the reference count because the resulting iput() might cause an
1477 * inode eviction. The tradeoff is that the @match funtion must be
1478 * very carefully implemented.
1480 struct inode *find_inode_nowait(struct super_block *sb,
1481 unsigned long hashval,
1482 int (*match)(struct inode *, unsigned long,
1486 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1487 struct inode *inode, *ret_inode = NULL;
1490 spin_lock(&inode_hash_lock);
1491 hlist_for_each_entry(inode, head, i_hash) {
1492 if (inode->i_sb != sb)
1494 mval = match(inode, hashval, data);
1502 spin_unlock(&inode_hash_lock);
1505 EXPORT_SYMBOL(find_inode_nowait);
1508 * find_inode_rcu - find an inode in the inode cache
1509 * @sb: Super block of file system to search
1510 * @hashval: Key to hash
1511 * @test: Function to test match on an inode
1512 * @data: Data for test function
1514 * Search for the inode specified by @hashval and @data in the inode cache,
1515 * where the helper function @test will return 0 if the inode does not match
1516 * and 1 if it does. The @test function must be responsible for taking the
1517 * i_lock spin_lock and checking i_state for an inode being freed or being
1520 * If successful, this will return the inode for which the @test function
1521 * returned 1 and NULL otherwise.
1523 * The @test function is not permitted to take a ref on any inode presented.
1524 * It is also not permitted to sleep.
1526 * The caller must hold the RCU read lock.
1528 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1529 int (*test)(struct inode *, void *), void *data)
1531 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1532 struct inode *inode;
1534 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1535 "suspicious find_inode_rcu() usage");
1537 hlist_for_each_entry_rcu(inode, head, i_hash) {
1538 if (inode->i_sb == sb &&
1539 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1545 EXPORT_SYMBOL(find_inode_rcu);
1548 * find_inode_by_ino_rcu - Find an inode in the inode cache
1549 * @sb: Super block of file system to search
1550 * @ino: The inode number to match
1552 * Search for the inode specified by @hashval and @data in the inode cache,
1553 * where the helper function @test will return 0 if the inode does not match
1554 * and 1 if it does. The @test function must be responsible for taking the
1555 * i_lock spin_lock and checking i_state for an inode being freed or being
1558 * If successful, this will return the inode for which the @test function
1559 * returned 1 and NULL otherwise.
1561 * The @test function is not permitted to take a ref on any inode presented.
1562 * It is also not permitted to sleep.
1564 * The caller must hold the RCU read lock.
1566 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1569 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1570 struct inode *inode;
1572 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1573 "suspicious find_inode_by_ino_rcu() usage");
1575 hlist_for_each_entry_rcu(inode, head, i_hash) {
1576 if (inode->i_ino == ino &&
1577 inode->i_sb == sb &&
1578 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1583 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1585 int insert_inode_locked(struct inode *inode)
1587 struct super_block *sb = inode->i_sb;
1588 ino_t ino = inode->i_ino;
1589 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1592 struct inode *old = NULL;
1593 spin_lock(&inode_hash_lock);
1594 hlist_for_each_entry(old, head, i_hash) {
1595 if (old->i_ino != ino)
1597 if (old->i_sb != sb)
1599 spin_lock(&old->i_lock);
1600 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1601 spin_unlock(&old->i_lock);
1607 spin_lock(&inode->i_lock);
1608 inode->i_state |= I_NEW | I_CREATING;
1609 hlist_add_head_rcu(&inode->i_hash, head);
1610 spin_unlock(&inode->i_lock);
1611 spin_unlock(&inode_hash_lock);
1614 if (unlikely(old->i_state & I_CREATING)) {
1615 spin_unlock(&old->i_lock);
1616 spin_unlock(&inode_hash_lock);
1620 spin_unlock(&old->i_lock);
1621 spin_unlock(&inode_hash_lock);
1623 if (unlikely(!inode_unhashed(old))) {
1630 EXPORT_SYMBOL(insert_inode_locked);
1632 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1633 int (*test)(struct inode *, void *), void *data)
1637 inode->i_state |= I_CREATING;
1638 old = inode_insert5(inode, hashval, test, NULL, data);
1646 EXPORT_SYMBOL(insert_inode_locked4);
1649 int generic_delete_inode(struct inode *inode)
1653 EXPORT_SYMBOL(generic_delete_inode);
1656 * Called when we're dropping the last reference
1659 * Call the FS "drop_inode()" function, defaulting to
1660 * the legacy UNIX filesystem behaviour. If it tells
1661 * us to evict inode, do so. Otherwise, retain inode
1662 * in cache if fs is alive, sync and evict if fs is
1665 static void iput_final(struct inode *inode)
1667 struct super_block *sb = inode->i_sb;
1668 const struct super_operations *op = inode->i_sb->s_op;
1669 unsigned long state;
1672 WARN_ON(inode->i_state & I_NEW);
1675 drop = op->drop_inode(inode);
1677 drop = generic_drop_inode(inode);
1680 !(inode->i_state & I_DONTCACHE) &&
1681 (sb->s_flags & SB_ACTIVE)) {
1682 inode_add_lru(inode);
1683 spin_unlock(&inode->i_lock);
1687 state = inode->i_state;
1689 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1690 spin_unlock(&inode->i_lock);
1692 write_inode_now(inode, 1);
1694 spin_lock(&inode->i_lock);
1695 state = inode->i_state;
1696 WARN_ON(state & I_NEW);
1697 state &= ~I_WILL_FREE;
1700 WRITE_ONCE(inode->i_state, state | I_FREEING);
1701 if (!list_empty(&inode->i_lru))
1702 inode_lru_list_del(inode);
1703 spin_unlock(&inode->i_lock);
1709 * iput - put an inode
1710 * @inode: inode to put
1712 * Puts an inode, dropping its usage count. If the inode use count hits
1713 * zero, the inode is then freed and may also be destroyed.
1715 * Consequently, iput() can sleep.
1717 void iput(struct inode *inode)
1721 BUG_ON(inode->i_state & I_CLEAR);
1723 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1724 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1725 atomic_inc(&inode->i_count);
1726 spin_unlock(&inode->i_lock);
1727 trace_writeback_lazytime_iput(inode);
1728 mark_inode_dirty_sync(inode);
1734 EXPORT_SYMBOL(iput);
1738 * bmap - find a block number in a file
1739 * @inode: inode owning the block number being requested
1740 * @block: pointer containing the block to find
1742 * Replaces the value in ``*block`` with the block number on the device holding
1743 * corresponding to the requested block number in the file.
1744 * That is, asked for block 4 of inode 1 the function will replace the
1745 * 4 in ``*block``, with disk block relative to the disk start that holds that
1746 * block of the file.
1748 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1749 * hole, returns 0 and ``*block`` is also set to 0.
1751 int bmap(struct inode *inode, sector_t *block)
1753 if (!inode->i_mapping->a_ops->bmap)
1756 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1759 EXPORT_SYMBOL(bmap);
1763 * With relative atime, only update atime if the previous atime is
1764 * earlier than either the ctime or mtime or if at least a day has
1765 * passed since the last atime update.
1767 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1768 struct timespec64 now)
1771 if (!(mnt->mnt_flags & MNT_RELATIME))
1774 * Is mtime younger than atime? If yes, update atime:
1776 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1779 * Is ctime younger than atime? If yes, update atime:
1781 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1785 * Is the previous atime value older than a day? If yes,
1788 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1791 * Good, we can skip the atime update:
1796 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1798 int dirty_flags = 0;
1800 if (flags & (S_ATIME | S_CTIME | S_MTIME)) {
1801 if (flags & S_ATIME)
1802 inode->i_atime = *time;
1803 if (flags & S_CTIME)
1804 inode->i_ctime = *time;
1805 if (flags & S_MTIME)
1806 inode->i_mtime = *time;
1808 if (inode->i_sb->s_flags & SB_LAZYTIME)
1809 dirty_flags |= I_DIRTY_TIME;
1811 dirty_flags |= I_DIRTY_SYNC;
1814 if ((flags & S_VERSION) && inode_maybe_inc_iversion(inode, false))
1815 dirty_flags |= I_DIRTY_SYNC;
1817 __mark_inode_dirty(inode, dirty_flags);
1820 EXPORT_SYMBOL(generic_update_time);
1823 * This does the actual work of updating an inodes time or version. Must have
1824 * had called mnt_want_write() before calling this.
1826 int inode_update_time(struct inode *inode, struct timespec64 *time, int flags)
1828 if (inode->i_op->update_time)
1829 return inode->i_op->update_time(inode, time, flags);
1830 return generic_update_time(inode, time, flags);
1832 EXPORT_SYMBOL(inode_update_time);
1835 * atime_needs_update - update the access time
1836 * @path: the &struct path to update
1837 * @inode: inode to update
1839 * Update the accessed time on an inode and mark it for writeback.
1840 * This function automatically handles read only file systems and media,
1841 * as well as the "noatime" flag and inode specific "noatime" markers.
1843 bool atime_needs_update(const struct path *path, struct inode *inode)
1845 struct vfsmount *mnt = path->mnt;
1846 struct timespec64 now;
1848 if (inode->i_flags & S_NOATIME)
1851 /* Atime updates will likely cause i_uid and i_gid to be written
1852 * back improprely if their true value is unknown to the vfs.
1854 if (HAS_UNMAPPED_ID(mnt_user_ns(mnt), inode))
1857 if (IS_NOATIME(inode))
1859 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1862 if (mnt->mnt_flags & MNT_NOATIME)
1864 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1867 now = current_time(inode);
1869 if (!relatime_need_update(mnt, inode, now))
1872 if (timespec64_equal(&inode->i_atime, &now))
1878 void touch_atime(const struct path *path)
1880 struct vfsmount *mnt = path->mnt;
1881 struct inode *inode = d_inode(path->dentry);
1882 struct timespec64 now;
1884 if (!atime_needs_update(path, inode))
1887 if (!sb_start_write_trylock(inode->i_sb))
1890 if (__mnt_want_write(mnt) != 0)
1893 * File systems can error out when updating inodes if they need to
1894 * allocate new space to modify an inode (such is the case for
1895 * Btrfs), but since we touch atime while walking down the path we
1896 * really don't care if we failed to update the atime of the file,
1897 * so just ignore the return value.
1898 * We may also fail on filesystems that have the ability to make parts
1899 * of the fs read only, e.g. subvolumes in Btrfs.
1901 now = current_time(inode);
1902 inode_update_time(inode, &now, S_ATIME);
1903 __mnt_drop_write(mnt);
1905 sb_end_write(inode->i_sb);
1907 EXPORT_SYMBOL(touch_atime);
1910 * Return mask of changes for notify_change() that need to be done as a
1911 * response to write or truncate. Return 0 if nothing has to be changed.
1912 * Negative value on error (change should be denied).
1914 int dentry_needs_remove_privs(struct user_namespace *mnt_userns,
1915 struct dentry *dentry)
1917 struct inode *inode = d_inode(dentry);
1921 if (IS_NOSEC(inode))
1924 mask = setattr_should_drop_suidgid(mnt_userns, inode);
1925 ret = security_inode_need_killpriv(dentry);
1929 mask |= ATTR_KILL_PRIV;
1933 static int __remove_privs(struct user_namespace *mnt_userns,
1934 struct dentry *dentry, int kill)
1936 struct iattr newattrs;
1938 newattrs.ia_valid = ATTR_FORCE | kill;
1940 * Note we call this on write, so notify_change will not
1941 * encounter any conflicting delegations:
1943 return notify_change(mnt_userns, dentry, &newattrs, NULL);
1947 * Remove special file priviledges (suid, capabilities) when file is written
1950 int file_remove_privs(struct file *file)
1952 struct dentry *dentry = file_dentry(file);
1953 struct inode *inode = file_inode(file);
1958 * Fast path for nothing security related.
1959 * As well for non-regular files, e.g. blkdev inodes.
1960 * For example, blkdev_write_iter() might get here
1961 * trying to remove privs which it is not allowed to.
1963 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1966 kill = dentry_needs_remove_privs(file_mnt_user_ns(file), dentry);
1970 error = __remove_privs(file_mnt_user_ns(file), dentry, kill);
1972 inode_has_no_xattr(inode);
1976 EXPORT_SYMBOL(file_remove_privs);
1979 * file_update_time - update mtime and ctime time
1980 * @file: file accessed
1982 * Update the mtime and ctime members of an inode and mark the inode
1983 * for writeback. Note that this function is meant exclusively for
1984 * usage in the file write path of filesystems, and filesystems may
1985 * choose to explicitly ignore update via this function with the
1986 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1987 * timestamps are handled by the server. This can return an error for
1988 * file systems who need to allocate space in order to update an inode.
1991 int file_update_time(struct file *file)
1993 struct inode *inode = file_inode(file);
1994 struct timespec64 now;
1998 /* First try to exhaust all avenues to not sync */
1999 if (IS_NOCMTIME(inode))
2002 now = current_time(inode);
2003 if (!timespec64_equal(&inode->i_mtime, &now))
2006 if (!timespec64_equal(&inode->i_ctime, &now))
2009 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2010 sync_it |= S_VERSION;
2015 /* Finally allowed to write? Takes lock. */
2016 if (__mnt_want_write_file(file))
2019 ret = inode_update_time(inode, &now, sync_it);
2020 __mnt_drop_write_file(file);
2024 EXPORT_SYMBOL(file_update_time);
2026 /* Caller must hold the file's inode lock */
2027 int file_modified(struct file *file)
2032 * Clear the security bits if the process is not being run by root.
2033 * This keeps people from modifying setuid and setgid binaries.
2035 err = file_remove_privs(file);
2039 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2042 return file_update_time(file);
2044 EXPORT_SYMBOL(file_modified);
2046 int inode_needs_sync(struct inode *inode)
2050 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2054 EXPORT_SYMBOL(inode_needs_sync);
2057 * If we try to find an inode in the inode hash while it is being
2058 * deleted, we have to wait until the filesystem completes its
2059 * deletion before reporting that it isn't found. This function waits
2060 * until the deletion _might_ have completed. Callers are responsible
2061 * to recheck inode state.
2063 * It doesn't matter if I_NEW is not set initially, a call to
2064 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2067 static void __wait_on_freeing_inode(struct inode *inode)
2069 wait_queue_head_t *wq;
2070 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2071 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2072 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2073 spin_unlock(&inode->i_lock);
2074 spin_unlock(&inode_hash_lock);
2076 finish_wait(wq, &wait.wq_entry);
2077 spin_lock(&inode_hash_lock);
2080 static __initdata unsigned long ihash_entries;
2081 static int __init set_ihash_entries(char *str)
2085 ihash_entries = simple_strtoul(str, &str, 0);
2088 __setup("ihash_entries=", set_ihash_entries);
2091 * Initialize the waitqueues and inode hash table.
2093 void __init inode_init_early(void)
2095 /* If hashes are distributed across NUMA nodes, defer
2096 * hash allocation until vmalloc space is available.
2102 alloc_large_system_hash("Inode-cache",
2103 sizeof(struct hlist_head),
2106 HASH_EARLY | HASH_ZERO,
2113 void __init inode_init(void)
2115 /* inode slab cache */
2116 inode_cachep = kmem_cache_create("inode_cache",
2117 sizeof(struct inode),
2119 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2120 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2123 /* Hash may have been set up in inode_init_early */
2128 alloc_large_system_hash("Inode-cache",
2129 sizeof(struct hlist_head),
2139 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2141 inode->i_mode = mode;
2142 if (S_ISCHR(mode)) {
2143 inode->i_fop = &def_chr_fops;
2144 inode->i_rdev = rdev;
2145 } else if (S_ISBLK(mode)) {
2146 inode->i_fop = &def_blk_fops;
2147 inode->i_rdev = rdev;
2148 } else if (S_ISFIFO(mode))
2149 inode->i_fop = &pipefifo_fops;
2150 else if (S_ISSOCK(mode))
2151 ; /* leave it no_open_fops */
2153 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2154 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2157 EXPORT_SYMBOL(init_special_inode);
2160 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2161 * @mnt_userns: User namespace of the mount the inode was created from
2163 * @dir: Directory inode
2164 * @mode: mode of the new inode
2166 * If the inode has been created through an idmapped mount the user namespace of
2167 * the vfsmount must be passed through @mnt_userns. This function will then take
2168 * care to map the inode according to @mnt_userns before checking permissions
2169 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2170 * checking is to be performed on the raw inode simply passs init_user_ns.
2172 void inode_init_owner(struct user_namespace *mnt_userns, struct inode *inode,
2173 const struct inode *dir, umode_t mode)
2175 inode_fsuid_set(inode, mnt_userns);
2176 if (dir && dir->i_mode & S_ISGID) {
2177 inode->i_gid = dir->i_gid;
2179 /* Directories are special, and always inherit S_ISGID */
2183 inode_fsgid_set(inode, mnt_userns);
2184 inode->i_mode = mode;
2186 EXPORT_SYMBOL(inode_init_owner);
2189 * inode_owner_or_capable - check current task permissions to inode
2190 * @mnt_userns: user namespace of the mount the inode was found from
2191 * @inode: inode being checked
2193 * Return true if current either has CAP_FOWNER in a namespace with the
2194 * inode owner uid mapped, or owns the file.
2196 * If the inode has been found through an idmapped mount the user namespace of
2197 * the vfsmount must be passed through @mnt_userns. This function will then take
2198 * care to map the inode according to @mnt_userns before checking permissions.
2199 * On non-idmapped mounts or if permission checking is to be performed on the
2200 * raw inode simply passs init_user_ns.
2202 bool inode_owner_or_capable(struct user_namespace *mnt_userns,
2203 const struct inode *inode)
2206 struct user_namespace *ns;
2208 i_uid = i_uid_into_mnt(mnt_userns, inode);
2209 if (uid_eq(current_fsuid(), i_uid))
2212 ns = current_user_ns();
2213 if (kuid_has_mapping(ns, i_uid) && ns_capable(ns, CAP_FOWNER))
2217 EXPORT_SYMBOL(inode_owner_or_capable);
2220 * Direct i/o helper functions
2222 static void __inode_dio_wait(struct inode *inode)
2224 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2225 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2228 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2229 if (atomic_read(&inode->i_dio_count))
2231 } while (atomic_read(&inode->i_dio_count));
2232 finish_wait(wq, &q.wq_entry);
2236 * inode_dio_wait - wait for outstanding DIO requests to finish
2237 * @inode: inode to wait for
2239 * Waits for all pending direct I/O requests to finish so that we can
2240 * proceed with a truncate or equivalent operation.
2242 * Must be called under a lock that serializes taking new references
2243 * to i_dio_count, usually by inode->i_mutex.
2245 void inode_dio_wait(struct inode *inode)
2247 if (atomic_read(&inode->i_dio_count))
2248 __inode_dio_wait(inode);
2250 EXPORT_SYMBOL(inode_dio_wait);
2253 * inode_set_flags - atomically set some inode flags
2255 * Note: the caller should be holding i_mutex, or else be sure that
2256 * they have exclusive access to the inode structure (i.e., while the
2257 * inode is being instantiated). The reason for the cmpxchg() loop
2258 * --- which wouldn't be necessary if all code paths which modify
2259 * i_flags actually followed this rule, is that there is at least one
2260 * code path which doesn't today so we use cmpxchg() out of an abundance
2263 * In the long run, i_mutex is overkill, and we should probably look
2264 * at using the i_lock spinlock to protect i_flags, and then make sure
2265 * it is so documented in include/linux/fs.h and that all code follows
2266 * the locking convention!!
2268 void inode_set_flags(struct inode *inode, unsigned int flags,
2271 WARN_ON_ONCE(flags & ~mask);
2272 set_mask_bits(&inode->i_flags, mask, flags);
2274 EXPORT_SYMBOL(inode_set_flags);
2276 void inode_nohighmem(struct inode *inode)
2278 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2280 EXPORT_SYMBOL(inode_nohighmem);
2283 * timestamp_truncate - Truncate timespec to a granularity
2285 * @inode: inode being updated
2287 * Truncate a timespec to the granularity supported by the fs
2288 * containing the inode. Always rounds down. gran must
2289 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2291 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2293 struct super_block *sb = inode->i_sb;
2294 unsigned int gran = sb->s_time_gran;
2296 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2297 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2300 /* Avoid division in the common cases 1 ns and 1 s. */
2303 else if (gran == NSEC_PER_SEC)
2305 else if (gran > 1 && gran < NSEC_PER_SEC)
2306 t.tv_nsec -= t.tv_nsec % gran;
2308 WARN(1, "invalid file time granularity: %u", gran);
2311 EXPORT_SYMBOL(timestamp_truncate);
2314 * current_time - Return FS time
2317 * Return the current time truncated to the time granularity supported by
2320 * Note that inode and inode->sb cannot be NULL.
2321 * Otherwise, the function warns and returns time without truncation.
2323 struct timespec64 current_time(struct inode *inode)
2325 struct timespec64 now;
2327 ktime_get_coarse_real_ts64(&now);
2329 if (unlikely(!inode->i_sb)) {
2330 WARN(1, "current_time() called with uninitialized super_block in the inode");
2334 return timestamp_truncate(now, inode);
2336 EXPORT_SYMBOL(current_time);
2339 * in_group_or_capable - check whether caller is CAP_FSETID privileged
2340 * @mnt_userns: user namespace of the mount @inode was found from
2341 * @inode: inode to check
2342 * @gid: the new/current gid of @inode
2344 * Check wether @gid is in the caller's group list or if the caller is
2345 * privileged with CAP_FSETID over @inode. This can be used to determine
2346 * whether the setgid bit can be kept or must be dropped.
2348 * Return: true if the caller is sufficiently privileged, false if not.
2350 bool in_group_or_capable(struct user_namespace *mnt_userns,
2351 const struct inode *inode, kgid_t gid)
2353 if (in_group_p(gid))
2355 if (capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FSETID))
2361 * mode_strip_sgid - handle the sgid bit for non-directories
2362 * @mnt_userns: User namespace of the mount the inode was created from
2363 * @dir: parent directory inode
2364 * @mode: mode of the file to be created in @dir
2366 * If the @mode of the new file has both the S_ISGID and S_IXGRP bit
2367 * raised and @dir has the S_ISGID bit raised ensure that the caller is
2368 * either in the group of the parent directory or they have CAP_FSETID
2369 * in their user namespace and are privileged over the parent directory.
2370 * In all other cases, strip the S_ISGID bit from @mode.
2372 * Return: the new mode to use for the file
2374 umode_t mode_strip_sgid(struct user_namespace *mnt_userns,
2375 const struct inode *dir, umode_t mode)
2377 if ((mode & (S_ISGID | S_IXGRP)) != (S_ISGID | S_IXGRP))
2379 if (S_ISDIR(mode) || !dir || !(dir->i_mode & S_ISGID))
2381 if (in_group_or_capable(mnt_userns, dir,
2382 i_gid_into_mnt(mnt_userns, dir)))
2384 return mode & ~S_ISGID;
2386 EXPORT_SYMBOL(mode_strip_sgid);