2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
5 #include <linux/export.h>
8 #include <linux/backing-dev.h>
9 #include <linux/hash.h>
10 #include <linux/swap.h>
11 #include <linux/security.h>
12 #include <linux/cdev.h>
13 #include <linux/bootmem.h>
14 #include <linux/fsnotify.h>
15 #include <linux/mount.h>
16 #include <linux/posix_acl.h>
17 #include <linux/prefetch.h>
18 #include <linux/buffer_head.h> /* for inode_has_buffers */
19 #include <linux/ratelimit.h>
20 #include <linux/list_lru.h>
21 #include <linux/iversion.h>
22 #include <trace/events/writeback.h>
26 * Inode locking rules:
28 * inode->i_lock protects:
29 * inode->i_state, inode->i_hash, __iget()
30 * Inode LRU list locks protect:
31 * inode->i_sb->s_inode_lru, inode->i_lru
32 * inode->i_sb->s_inode_list_lock protects:
33 * inode->i_sb->s_inodes, inode->i_sb_list
34 * bdi->wb.list_lock protects:
35 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
36 * inode_hash_lock protects:
37 * inode_hashtable, inode->i_hash
41 * inode->i_sb->s_inode_list_lock
43 * Inode LRU list locks
49 * inode->i_sb->s_inode_list_lock
56 static unsigned int i_hash_mask __read_mostly;
57 static unsigned int i_hash_shift __read_mostly;
58 static struct hlist_head *inode_hashtable __read_mostly;
59 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
62 * Empty aops. Can be used for the cases where the user does not
63 * define any of the address_space operations.
65 const struct address_space_operations empty_aops = {
67 EXPORT_SYMBOL(empty_aops);
70 * Statistics gathering..
72 struct inodes_stat_t inodes_stat;
74 static DEFINE_PER_CPU(unsigned long, nr_inodes);
75 static DEFINE_PER_CPU(unsigned long, nr_unused);
77 static struct kmem_cache *inode_cachep __read_mostly;
79 static long get_nr_inodes(void)
83 for_each_possible_cpu(i)
84 sum += per_cpu(nr_inodes, i);
85 return sum < 0 ? 0 : sum;
88 static inline long get_nr_inodes_unused(void)
92 for_each_possible_cpu(i)
93 sum += per_cpu(nr_unused, i);
94 return sum < 0 ? 0 : sum;
97 long get_nr_dirty_inodes(void)
99 /* not actually dirty inodes, but a wild approximation */
100 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
101 return nr_dirty > 0 ? nr_dirty : 0;
105 * Handle nr_inode sysctl
108 int proc_nr_inodes(struct ctl_table *table, int write,
109 void __user *buffer, size_t *lenp, loff_t *ppos)
111 inodes_stat.nr_inodes = get_nr_inodes();
112 inodes_stat.nr_unused = get_nr_inodes_unused();
113 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
117 static int no_open(struct inode *inode, struct file *file)
123 * inode_init_always - perform inode structure initialisation
124 * @sb: superblock inode belongs to
125 * @inode: inode to initialise
127 * These are initializations that need to be done on every inode
128 * allocation as the fields are not initialised by slab allocation.
130 int inode_init_always(struct super_block *sb, struct inode *inode)
132 static const struct inode_operations empty_iops;
133 static const struct file_operations no_open_fops = {.open = no_open};
134 struct address_space *const mapping = &inode->i_data;
137 inode->i_blkbits = sb->s_blocksize_bits;
139 atomic64_set(&inode->i_sequence, 0);
140 atomic_set(&inode->i_count, 1);
141 inode->i_op = &empty_iops;
142 inode->i_fop = &no_open_fops;
143 inode->__i_nlink = 1;
144 inode->i_opflags = 0;
146 inode->i_opflags |= IOP_XATTR;
147 i_uid_write(inode, 0);
148 i_gid_write(inode, 0);
149 atomic_set(&inode->i_writecount, 0);
151 inode->i_write_hint = WRITE_LIFE_NOT_SET;
154 inode->i_generation = 0;
155 inode->i_pipe = NULL;
156 inode->i_bdev = NULL;
157 inode->i_cdev = NULL;
158 inode->i_link = NULL;
159 inode->i_dir_seq = 0;
161 inode->dirtied_when = 0;
163 #ifdef CONFIG_CGROUP_WRITEBACK
164 inode->i_wb_frn_winner = 0;
165 inode->i_wb_frn_avg_time = 0;
166 inode->i_wb_frn_history = 0;
169 spin_lock_init(&inode->i_lock);
170 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
172 init_rwsem(&inode->i_rwsem);
173 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
175 atomic_set(&inode->i_dio_count, 0);
177 mapping->a_ops = &empty_aops;
178 mapping->host = inode;
181 atomic_set(&mapping->i_mmap_writable, 0);
182 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
183 mapping->private_data = NULL;
184 mapping->writeback_index = 0;
185 inode->i_private = NULL;
186 inode->i_mapping = mapping;
187 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
188 #ifdef CONFIG_FS_POSIX_ACL
189 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
192 #ifdef CONFIG_FSNOTIFY
193 inode->i_fsnotify_mask = 0;
195 inode->i_flctx = NULL;
197 if (unlikely(security_inode_alloc(inode)))
199 this_cpu_inc(nr_inodes);
203 EXPORT_SYMBOL(inode_init_always);
205 static struct inode *alloc_inode(struct super_block *sb)
209 if (sb->s_op->alloc_inode)
210 inode = sb->s_op->alloc_inode(sb);
212 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
217 if (unlikely(inode_init_always(sb, inode))) {
218 if (inode->i_sb->s_op->destroy_inode)
219 inode->i_sb->s_op->destroy_inode(inode);
221 kmem_cache_free(inode_cachep, inode);
228 void free_inode_nonrcu(struct inode *inode)
230 kmem_cache_free(inode_cachep, inode);
232 EXPORT_SYMBOL(free_inode_nonrcu);
234 void __destroy_inode(struct inode *inode)
236 BUG_ON(inode_has_buffers(inode));
237 inode_detach_wb(inode);
238 security_inode_free(inode);
239 fsnotify_inode_delete(inode);
240 locks_free_lock_context(inode);
241 if (!inode->i_nlink) {
242 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
243 atomic_long_dec(&inode->i_sb->s_remove_count);
246 #ifdef CONFIG_FS_POSIX_ACL
247 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
248 posix_acl_release(inode->i_acl);
249 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
250 posix_acl_release(inode->i_default_acl);
252 this_cpu_dec(nr_inodes);
254 EXPORT_SYMBOL(__destroy_inode);
256 static void i_callback(struct rcu_head *head)
258 struct inode *inode = container_of(head, struct inode, i_rcu);
259 kmem_cache_free(inode_cachep, inode);
262 static void destroy_inode(struct inode *inode)
264 BUG_ON(!list_empty(&inode->i_lru));
265 __destroy_inode(inode);
266 if (inode->i_sb->s_op->destroy_inode)
267 inode->i_sb->s_op->destroy_inode(inode);
269 call_rcu(&inode->i_rcu, i_callback);
273 * drop_nlink - directly drop an inode's link count
276 * This is a low-level filesystem helper to replace any
277 * direct filesystem manipulation of i_nlink. In cases
278 * where we are attempting to track writes to the
279 * filesystem, a decrement to zero means an imminent
280 * write when the file is truncated and actually unlinked
283 void drop_nlink(struct inode *inode)
285 WARN_ON(inode->i_nlink == 0);
288 atomic_long_inc(&inode->i_sb->s_remove_count);
290 EXPORT_SYMBOL(drop_nlink);
293 * clear_nlink - directly zero an inode's link count
296 * This is a low-level filesystem helper to replace any
297 * direct filesystem manipulation of i_nlink. See
298 * drop_nlink() for why we care about i_nlink hitting zero.
300 void clear_nlink(struct inode *inode)
302 if (inode->i_nlink) {
303 inode->__i_nlink = 0;
304 atomic_long_inc(&inode->i_sb->s_remove_count);
307 EXPORT_SYMBOL(clear_nlink);
310 * set_nlink - directly set an inode's link count
312 * @nlink: new nlink (should be non-zero)
314 * This is a low-level filesystem helper to replace any
315 * direct filesystem manipulation of i_nlink.
317 void set_nlink(struct inode *inode, unsigned int nlink)
322 /* Yes, some filesystems do change nlink from zero to one */
323 if (inode->i_nlink == 0)
324 atomic_long_dec(&inode->i_sb->s_remove_count);
326 inode->__i_nlink = nlink;
329 EXPORT_SYMBOL(set_nlink);
332 * inc_nlink - directly increment an inode's link count
335 * This is a low-level filesystem helper to replace any
336 * direct filesystem manipulation of i_nlink. Currently,
337 * it is only here for parity with dec_nlink().
339 void inc_nlink(struct inode *inode)
341 if (unlikely(inode->i_nlink == 0)) {
342 WARN_ON(!(inode->i_state & I_LINKABLE));
343 atomic_long_dec(&inode->i_sb->s_remove_count);
348 EXPORT_SYMBOL(inc_nlink);
350 static void __address_space_init_once(struct address_space *mapping)
352 INIT_RADIX_TREE(&mapping->i_pages, GFP_ATOMIC | __GFP_ACCOUNT);
353 init_rwsem(&mapping->i_mmap_rwsem);
354 INIT_LIST_HEAD(&mapping->private_list);
355 spin_lock_init(&mapping->private_lock);
356 mapping->i_mmap = RB_ROOT_CACHED;
359 void address_space_init_once(struct address_space *mapping)
361 memset(mapping, 0, sizeof(*mapping));
362 __address_space_init_once(mapping);
364 EXPORT_SYMBOL(address_space_init_once);
367 * These are initializations that only need to be done
368 * once, because the fields are idempotent across use
369 * of the inode, so let the slab aware of that.
371 void inode_init_once(struct inode *inode)
373 memset(inode, 0, sizeof(*inode));
374 INIT_HLIST_NODE(&inode->i_hash);
375 INIT_LIST_HEAD(&inode->i_devices);
376 INIT_LIST_HEAD(&inode->i_io_list);
377 INIT_LIST_HEAD(&inode->i_wb_list);
378 INIT_LIST_HEAD(&inode->i_lru);
379 __address_space_init_once(&inode->i_data);
380 i_size_ordered_init(inode);
382 EXPORT_SYMBOL(inode_init_once);
384 static void init_once(void *foo)
386 struct inode *inode = (struct inode *) foo;
388 inode_init_once(inode);
392 * inode->i_lock must be held
394 void __iget(struct inode *inode)
396 atomic_inc(&inode->i_count);
400 * get additional reference to inode; caller must already hold one.
402 void ihold(struct inode *inode)
404 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
406 EXPORT_SYMBOL(ihold);
408 static void inode_lru_list_add(struct inode *inode)
410 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
411 this_cpu_inc(nr_unused);
413 inode->i_state |= I_REFERENCED;
417 * Add inode to LRU if needed (inode is unused and clean).
419 * Needs inode->i_lock held.
421 void inode_add_lru(struct inode *inode)
423 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
424 I_FREEING | I_WILL_FREE)) &&
425 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
426 inode_lru_list_add(inode);
430 static void inode_lru_list_del(struct inode *inode)
433 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
434 this_cpu_dec(nr_unused);
438 * inode_sb_list_add - add inode to the superblock list of inodes
439 * @inode: inode to add
441 void inode_sb_list_add(struct inode *inode)
443 spin_lock(&inode->i_sb->s_inode_list_lock);
444 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
445 spin_unlock(&inode->i_sb->s_inode_list_lock);
447 EXPORT_SYMBOL_GPL(inode_sb_list_add);
449 static inline void inode_sb_list_del(struct inode *inode)
451 if (!list_empty(&inode->i_sb_list)) {
452 spin_lock(&inode->i_sb->s_inode_list_lock);
453 list_del_init(&inode->i_sb_list);
454 spin_unlock(&inode->i_sb->s_inode_list_lock);
458 static unsigned long hash(struct super_block *sb, unsigned long hashval)
462 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
464 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
465 return tmp & i_hash_mask;
469 * __insert_inode_hash - hash an inode
470 * @inode: unhashed inode
471 * @hashval: unsigned long value used to locate this object in the
474 * Add an inode to the inode hash for this superblock.
476 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
478 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
480 spin_lock(&inode_hash_lock);
481 spin_lock(&inode->i_lock);
482 hlist_add_head(&inode->i_hash, b);
483 spin_unlock(&inode->i_lock);
484 spin_unlock(&inode_hash_lock);
486 EXPORT_SYMBOL(__insert_inode_hash);
489 * __remove_inode_hash - remove an inode from the hash
490 * @inode: inode to unhash
492 * Remove an inode from the superblock.
494 void __remove_inode_hash(struct inode *inode)
496 spin_lock(&inode_hash_lock);
497 spin_lock(&inode->i_lock);
498 hlist_del_init(&inode->i_hash);
499 spin_unlock(&inode->i_lock);
500 spin_unlock(&inode_hash_lock);
502 EXPORT_SYMBOL(__remove_inode_hash);
504 void clear_inode(struct inode *inode)
507 * We have to cycle the i_pages lock here because reclaim can be in the
508 * process of removing the last page (in __delete_from_page_cache())
509 * and we must not free the mapping under it.
511 xa_lock_irq(&inode->i_data.i_pages);
512 BUG_ON(inode->i_data.nrpages);
513 BUG_ON(inode->i_data.nrexceptional);
514 xa_unlock_irq(&inode->i_data.i_pages);
515 BUG_ON(!list_empty(&inode->i_data.private_list));
516 BUG_ON(!(inode->i_state & I_FREEING));
517 BUG_ON(inode->i_state & I_CLEAR);
518 BUG_ON(!list_empty(&inode->i_wb_list));
519 /* don't need i_lock here, no concurrent mods to i_state */
520 inode->i_state = I_FREEING | I_CLEAR;
522 EXPORT_SYMBOL(clear_inode);
525 * Free the inode passed in, removing it from the lists it is still connected
526 * to. We remove any pages still attached to the inode and wait for any IO that
527 * is still in progress before finally destroying the inode.
529 * An inode must already be marked I_FREEING so that we avoid the inode being
530 * moved back onto lists if we race with other code that manipulates the lists
531 * (e.g. writeback_single_inode). The caller is responsible for setting this.
533 * An inode must already be removed from the LRU list before being evicted from
534 * the cache. This should occur atomically with setting the I_FREEING state
535 * flag, so no inodes here should ever be on the LRU when being evicted.
537 static void evict(struct inode *inode)
539 const struct super_operations *op = inode->i_sb->s_op;
541 BUG_ON(!(inode->i_state & I_FREEING));
542 BUG_ON(!list_empty(&inode->i_lru));
544 if (!list_empty(&inode->i_io_list))
545 inode_io_list_del(inode);
547 inode_sb_list_del(inode);
550 * Wait for flusher thread to be done with the inode so that filesystem
551 * does not start destroying it while writeback is still running. Since
552 * the inode has I_FREEING set, flusher thread won't start new work on
553 * the inode. We just have to wait for running writeback to finish.
555 inode_wait_for_writeback(inode);
557 if (op->evict_inode) {
558 op->evict_inode(inode);
560 truncate_inode_pages_final(&inode->i_data);
563 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
565 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
568 remove_inode_hash(inode);
570 spin_lock(&inode->i_lock);
571 wake_up_bit(&inode->i_state, __I_NEW);
572 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
573 spin_unlock(&inode->i_lock);
575 destroy_inode(inode);
579 * dispose_list - dispose of the contents of a local list
580 * @head: the head of the list to free
582 * Dispose-list gets a local list with local inodes in it, so it doesn't
583 * need to worry about list corruption and SMP locks.
585 static void dispose_list(struct list_head *head)
587 while (!list_empty(head)) {
590 inode = list_first_entry(head, struct inode, i_lru);
591 list_del_init(&inode->i_lru);
599 * evict_inodes - evict all evictable inodes for a superblock
600 * @sb: superblock to operate on
602 * Make sure that no inodes with zero refcount are retained. This is
603 * called by superblock shutdown after having SB_ACTIVE flag removed,
604 * so any inode reaching zero refcount during or after that call will
605 * be immediately evicted.
607 void evict_inodes(struct super_block *sb)
609 struct inode *inode, *next;
613 spin_lock(&sb->s_inode_list_lock);
614 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
615 if (atomic_read(&inode->i_count))
618 spin_lock(&inode->i_lock);
619 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
620 spin_unlock(&inode->i_lock);
624 inode->i_state |= I_FREEING;
625 inode_lru_list_del(inode);
626 spin_unlock(&inode->i_lock);
627 list_add(&inode->i_lru, &dispose);
630 * We can have a ton of inodes to evict at unmount time given
631 * enough memory, check to see if we need to go to sleep for a
632 * bit so we don't livelock.
634 if (need_resched()) {
635 spin_unlock(&sb->s_inode_list_lock);
637 dispose_list(&dispose);
641 spin_unlock(&sb->s_inode_list_lock);
643 dispose_list(&dispose);
645 EXPORT_SYMBOL_GPL(evict_inodes);
648 * invalidate_inodes - attempt to free all inodes on a superblock
649 * @sb: superblock to operate on
650 * @kill_dirty: flag to guide handling of dirty inodes
652 * Attempts to free all inodes for a given superblock. If there were any
653 * busy inodes return a non-zero value, else zero.
654 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
657 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
660 struct inode *inode, *next;
664 spin_lock(&sb->s_inode_list_lock);
665 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
666 spin_lock(&inode->i_lock);
667 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
668 spin_unlock(&inode->i_lock);
671 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
672 spin_unlock(&inode->i_lock);
676 if (atomic_read(&inode->i_count)) {
677 spin_unlock(&inode->i_lock);
682 inode->i_state |= I_FREEING;
683 inode_lru_list_del(inode);
684 spin_unlock(&inode->i_lock);
685 list_add(&inode->i_lru, &dispose);
686 if (need_resched()) {
687 spin_unlock(&sb->s_inode_list_lock);
689 dispose_list(&dispose);
693 spin_unlock(&sb->s_inode_list_lock);
695 dispose_list(&dispose);
701 * Isolate the inode from the LRU in preparation for freeing it.
703 * Any inodes which are pinned purely because of attached pagecache have their
704 * pagecache removed. If the inode has metadata buffers attached to
705 * mapping->private_list then try to remove them.
707 * If the inode has the I_REFERENCED flag set, then it means that it has been
708 * used recently - the flag is set in iput_final(). When we encounter such an
709 * inode, clear the flag and move it to the back of the LRU so it gets another
710 * pass through the LRU before it gets reclaimed. This is necessary because of
711 * the fact we are doing lazy LRU updates to minimise lock contention so the
712 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
713 * with this flag set because they are the inodes that are out of order.
715 static enum lru_status inode_lru_isolate(struct list_head *item,
716 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
718 struct list_head *freeable = arg;
719 struct inode *inode = container_of(item, struct inode, i_lru);
722 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
723 * If we fail to get the lock, just skip it.
725 if (!spin_trylock(&inode->i_lock))
729 * Referenced or dirty inodes are still in use. Give them another pass
730 * through the LRU as we canot reclaim them now.
732 if (atomic_read(&inode->i_count) ||
733 (inode->i_state & ~I_REFERENCED)) {
734 list_lru_isolate(lru, &inode->i_lru);
735 spin_unlock(&inode->i_lock);
736 this_cpu_dec(nr_unused);
740 /* recently referenced inodes get one more pass */
741 if (inode->i_state & I_REFERENCED) {
742 inode->i_state &= ~I_REFERENCED;
743 spin_unlock(&inode->i_lock);
747 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
749 spin_unlock(&inode->i_lock);
750 spin_unlock(lru_lock);
751 if (remove_inode_buffers(inode)) {
753 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
754 if (current_is_kswapd())
755 __count_vm_events(KSWAPD_INODESTEAL, reap);
757 __count_vm_events(PGINODESTEAL, reap);
758 if (current->reclaim_state)
759 current->reclaim_state->reclaimed_slab += reap;
766 WARN_ON(inode->i_state & I_NEW);
767 inode->i_state |= I_FREEING;
768 list_lru_isolate_move(lru, &inode->i_lru, freeable);
769 spin_unlock(&inode->i_lock);
771 this_cpu_dec(nr_unused);
776 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
777 * This is called from the superblock shrinker function with a number of inodes
778 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
779 * then are freed outside inode_lock by dispose_list().
781 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
786 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
787 inode_lru_isolate, &freeable);
788 dispose_list(&freeable);
792 static void __wait_on_freeing_inode(struct inode *inode);
794 * Called with the inode lock held.
796 static struct inode *find_inode(struct super_block *sb,
797 struct hlist_head *head,
798 int (*test)(struct inode *, void *),
801 struct inode *inode = NULL;
804 hlist_for_each_entry(inode, head, i_hash) {
805 if (inode->i_sb != sb)
807 if (!test(inode, data))
809 spin_lock(&inode->i_lock);
810 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
811 __wait_on_freeing_inode(inode);
814 if (unlikely(inode->i_state & I_CREATING)) {
815 spin_unlock(&inode->i_lock);
816 return ERR_PTR(-ESTALE);
819 spin_unlock(&inode->i_lock);
826 * find_inode_fast is the fast path version of find_inode, see the comment at
827 * iget_locked for details.
829 static struct inode *find_inode_fast(struct super_block *sb,
830 struct hlist_head *head, unsigned long ino)
832 struct inode *inode = NULL;
835 hlist_for_each_entry(inode, head, i_hash) {
836 if (inode->i_ino != ino)
838 if (inode->i_sb != sb)
840 spin_lock(&inode->i_lock);
841 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
842 __wait_on_freeing_inode(inode);
845 if (unlikely(inode->i_state & I_CREATING)) {
846 spin_unlock(&inode->i_lock);
847 return ERR_PTR(-ESTALE);
850 spin_unlock(&inode->i_lock);
857 * Each cpu owns a range of LAST_INO_BATCH numbers.
858 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
859 * to renew the exhausted range.
861 * This does not significantly increase overflow rate because every CPU can
862 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
863 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
864 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
865 * overflow rate by 2x, which does not seem too significant.
867 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
868 * error if st_ino won't fit in target struct field. Use 32bit counter
869 * here to attempt to avoid that.
871 #define LAST_INO_BATCH 1024
872 static DEFINE_PER_CPU(unsigned int, last_ino);
874 unsigned int get_next_ino(void)
876 unsigned int *p = &get_cpu_var(last_ino);
877 unsigned int res = *p;
880 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
881 static atomic_t shared_last_ino;
882 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
884 res = next - LAST_INO_BATCH;
889 /* get_next_ino should not provide a 0 inode number */
893 put_cpu_var(last_ino);
896 EXPORT_SYMBOL(get_next_ino);
899 * new_inode_pseudo - obtain an inode
902 * Allocates a new inode for given superblock.
903 * Inode wont be chained in superblock s_inodes list
905 * - fs can't be unmount
906 * - quotas, fsnotify, writeback can't work
908 struct inode *new_inode_pseudo(struct super_block *sb)
910 struct inode *inode = alloc_inode(sb);
913 spin_lock(&inode->i_lock);
915 spin_unlock(&inode->i_lock);
916 INIT_LIST_HEAD(&inode->i_sb_list);
922 * new_inode - obtain an inode
925 * Allocates a new inode for given superblock. The default gfp_mask
926 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
927 * If HIGHMEM pages are unsuitable or it is known that pages allocated
928 * for the page cache are not reclaimable or migratable,
929 * mapping_set_gfp_mask() must be called with suitable flags on the
930 * newly created inode's mapping
933 struct inode *new_inode(struct super_block *sb)
937 spin_lock_prefetch(&sb->s_inode_list_lock);
939 inode = new_inode_pseudo(sb);
941 inode_sb_list_add(inode);
944 EXPORT_SYMBOL(new_inode);
946 #ifdef CONFIG_DEBUG_LOCK_ALLOC
947 void lockdep_annotate_inode_mutex_key(struct inode *inode)
949 if (S_ISDIR(inode->i_mode)) {
950 struct file_system_type *type = inode->i_sb->s_type;
952 /* Set new key only if filesystem hasn't already changed it */
953 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
955 * ensure nobody is actually holding i_mutex
957 // mutex_destroy(&inode->i_mutex);
958 init_rwsem(&inode->i_rwsem);
959 lockdep_set_class(&inode->i_rwsem,
960 &type->i_mutex_dir_key);
964 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
968 * unlock_new_inode - clear the I_NEW state and wake up any waiters
969 * @inode: new inode to unlock
971 * Called when the inode is fully initialised to clear the new state of the
972 * inode and wake up anyone waiting for the inode to finish initialisation.
974 void unlock_new_inode(struct inode *inode)
976 lockdep_annotate_inode_mutex_key(inode);
977 spin_lock(&inode->i_lock);
978 WARN_ON(!(inode->i_state & I_NEW));
979 inode->i_state &= ~I_NEW & ~I_CREATING;
981 wake_up_bit(&inode->i_state, __I_NEW);
982 spin_unlock(&inode->i_lock);
984 EXPORT_SYMBOL(unlock_new_inode);
986 void discard_new_inode(struct inode *inode)
988 lockdep_annotate_inode_mutex_key(inode);
989 spin_lock(&inode->i_lock);
990 WARN_ON(!(inode->i_state & I_NEW));
991 inode->i_state &= ~I_NEW;
993 wake_up_bit(&inode->i_state, __I_NEW);
994 spin_unlock(&inode->i_lock);
997 EXPORT_SYMBOL(discard_new_inode);
1000 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1002 * Lock any non-NULL argument that is not a directory.
1003 * Zero, one or two objects may be locked by this function.
1005 * @inode1: first inode to lock
1006 * @inode2: second inode to lock
1008 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1010 if (inode1 > inode2)
1011 swap(inode1, inode2);
1013 if (inode1 && !S_ISDIR(inode1->i_mode))
1015 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1016 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1018 EXPORT_SYMBOL(lock_two_nondirectories);
1021 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1022 * @inode1: first inode to unlock
1023 * @inode2: second inode to unlock
1025 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1027 if (inode1 && !S_ISDIR(inode1->i_mode))
1028 inode_unlock(inode1);
1029 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1030 inode_unlock(inode2);
1032 EXPORT_SYMBOL(unlock_two_nondirectories);
1035 * inode_insert5 - obtain an inode from a mounted file system
1036 * @inode: pre-allocated inode to use for insert to cache
1037 * @hashval: hash value (usually inode number) to get
1038 * @test: callback used for comparisons between inodes
1039 * @set: callback used to initialize a new struct inode
1040 * @data: opaque data pointer to pass to @test and @set
1042 * Search for the inode specified by @hashval and @data in the inode cache,
1043 * and if present it is return it with an increased reference count. This is
1044 * a variant of iget5_locked() for callers that don't want to fail on memory
1045 * allocation of inode.
1047 * If the inode is not in cache, insert the pre-allocated inode to cache and
1048 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1049 * to fill it in before unlocking it via unlock_new_inode().
1051 * Note both @test and @set are called with the inode_hash_lock held, so can't
1054 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1055 int (*test)(struct inode *, void *),
1056 int (*set)(struct inode *, void *), void *data)
1058 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1060 bool creating = inode->i_state & I_CREATING;
1063 spin_lock(&inode_hash_lock);
1064 old = find_inode(inode->i_sb, head, test, data);
1065 if (unlikely(old)) {
1067 * Uhhuh, somebody else created the same inode under us.
1068 * Use the old inode instead of the preallocated one.
1070 spin_unlock(&inode_hash_lock);
1074 if (unlikely(inode_unhashed(old))) {
1081 if (set && unlikely(set(inode, data))) {
1087 * Return the locked inode with I_NEW set, the
1088 * caller is responsible for filling in the contents
1090 spin_lock(&inode->i_lock);
1091 inode->i_state |= I_NEW;
1092 hlist_add_head(&inode->i_hash, head);
1093 spin_unlock(&inode->i_lock);
1095 inode_sb_list_add(inode);
1097 spin_unlock(&inode_hash_lock);
1101 EXPORT_SYMBOL(inode_insert5);
1104 * iget5_locked - obtain an inode from a mounted file system
1105 * @sb: super block of file system
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 generalized version of iget_locked() for file systems where the inode
1114 * number is not sufficient for unique identification of an inode.
1116 * If the inode is not in cache, allocate a new inode and return it locked,
1117 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1118 * 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 *iget5_locked(struct super_block *sb, unsigned long hashval,
1124 int (*test)(struct inode *, void *),
1125 int (*set)(struct inode *, void *), void *data)
1127 struct inode *inode = ilookup5(sb, hashval, test, data);
1130 struct inode *new = alloc_inode(sb);
1134 inode = inode_insert5(new, hashval, test, set, data);
1135 if (unlikely(inode != new))
1141 EXPORT_SYMBOL(iget5_locked);
1144 * iget_locked - obtain an inode from a mounted file system
1145 * @sb: super block of file system
1146 * @ino: inode number to get
1148 * Search for the inode specified by @ino in the inode cache and if present
1149 * return it with an increased reference count. This is for file systems
1150 * where the inode number is sufficient for unique identification of an inode.
1152 * If the inode is not in cache, allocate a new inode and return it locked,
1153 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1154 * before unlocking it via unlock_new_inode().
1156 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1158 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1159 struct inode *inode;
1161 spin_lock(&inode_hash_lock);
1162 inode = find_inode_fast(sb, head, ino);
1163 spin_unlock(&inode_hash_lock);
1167 wait_on_inode(inode);
1168 if (unlikely(inode_unhashed(inode))) {
1175 inode = alloc_inode(sb);
1179 spin_lock(&inode_hash_lock);
1180 /* We released the lock, so.. */
1181 old = find_inode_fast(sb, head, ino);
1184 spin_lock(&inode->i_lock);
1185 inode->i_state = I_NEW;
1186 hlist_add_head(&inode->i_hash, head);
1187 spin_unlock(&inode->i_lock);
1188 inode_sb_list_add(inode);
1189 spin_unlock(&inode_hash_lock);
1191 /* Return the locked inode with I_NEW set, the
1192 * caller is responsible for filling in the contents
1198 * Uhhuh, somebody else created the same inode under
1199 * us. Use the old inode instead of the one we just
1202 spin_unlock(&inode_hash_lock);
1203 destroy_inode(inode);
1207 wait_on_inode(inode);
1208 if (unlikely(inode_unhashed(inode))) {
1215 EXPORT_SYMBOL(iget_locked);
1218 * search the inode cache for a matching inode number.
1219 * If we find one, then the inode number we are trying to
1220 * allocate is not unique and so we should not use it.
1222 * Returns 1 if the inode number is unique, 0 if it is not.
1224 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1226 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1227 struct inode *inode;
1229 spin_lock(&inode_hash_lock);
1230 hlist_for_each_entry(inode, b, i_hash) {
1231 if (inode->i_ino == ino && inode->i_sb == sb) {
1232 spin_unlock(&inode_hash_lock);
1236 spin_unlock(&inode_hash_lock);
1242 * iunique - get a unique inode number
1244 * @max_reserved: highest reserved inode number
1246 * Obtain an inode number that is unique on the system for a given
1247 * superblock. This is used by file systems that have no natural
1248 * permanent inode numbering system. An inode number is returned that
1249 * is higher than the reserved limit but unique.
1252 * With a large number of inodes live on the file system this function
1253 * currently becomes quite slow.
1255 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1258 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1259 * error if st_ino won't fit in target struct field. Use 32bit counter
1260 * here to attempt to avoid that.
1262 static DEFINE_SPINLOCK(iunique_lock);
1263 static unsigned int counter;
1266 spin_lock(&iunique_lock);
1268 if (counter <= max_reserved)
1269 counter = max_reserved + 1;
1271 } while (!test_inode_iunique(sb, res));
1272 spin_unlock(&iunique_lock);
1276 EXPORT_SYMBOL(iunique);
1278 struct inode *igrab(struct inode *inode)
1280 spin_lock(&inode->i_lock);
1281 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1283 spin_unlock(&inode->i_lock);
1285 spin_unlock(&inode->i_lock);
1287 * Handle the case where s_op->clear_inode is not been
1288 * called yet, and somebody is calling igrab
1289 * while the inode is getting freed.
1295 EXPORT_SYMBOL(igrab);
1298 * ilookup5_nowait - search for an inode in the inode cache
1299 * @sb: super block of file system to search
1300 * @hashval: hash value (usually inode number) to search for
1301 * @test: callback used for comparisons between inodes
1302 * @data: opaque data pointer to pass to @test
1304 * Search for the inode specified by @hashval and @data in the inode cache.
1305 * If the inode is in the cache, the inode is returned with an incremented
1308 * Note: I_NEW is not waited upon so you have to be very careful what you do
1309 * with the returned inode. You probably should be using ilookup5() instead.
1311 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1313 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1314 int (*test)(struct inode *, void *), void *data)
1316 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1317 struct inode *inode;
1319 spin_lock(&inode_hash_lock);
1320 inode = find_inode(sb, head, test, data);
1321 spin_unlock(&inode_hash_lock);
1323 return IS_ERR(inode) ? NULL : inode;
1325 EXPORT_SYMBOL(ilookup5_nowait);
1328 * ilookup5 - search for an inode in the inode cache
1329 * @sb: super block of file system to search
1330 * @hashval: hash value (usually inode number) to search for
1331 * @test: callback used for comparisons between inodes
1332 * @data: opaque data pointer to pass to @test
1334 * Search for the inode specified by @hashval and @data in the inode cache,
1335 * and if the inode is in the cache, return the inode with an incremented
1336 * reference count. Waits on I_NEW before returning the inode.
1337 * returned with an incremented reference count.
1339 * This is a generalized version of ilookup() for file systems where the
1340 * inode number is not sufficient for unique identification of an inode.
1342 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1344 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1345 int (*test)(struct inode *, void *), void *data)
1347 struct inode *inode;
1349 inode = ilookup5_nowait(sb, hashval, test, data);
1351 wait_on_inode(inode);
1352 if (unlikely(inode_unhashed(inode))) {
1359 EXPORT_SYMBOL(ilookup5);
1362 * ilookup - search for an inode in the inode cache
1363 * @sb: super block of file system to search
1364 * @ino: inode number to search for
1366 * Search for the inode @ino in the inode cache, and if the inode is in the
1367 * cache, the inode is returned with an incremented reference count.
1369 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1371 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1372 struct inode *inode;
1374 spin_lock(&inode_hash_lock);
1375 inode = find_inode_fast(sb, head, ino);
1376 spin_unlock(&inode_hash_lock);
1381 wait_on_inode(inode);
1382 if (unlikely(inode_unhashed(inode))) {
1389 EXPORT_SYMBOL(ilookup);
1392 * find_inode_nowait - find an inode in the inode cache
1393 * @sb: super block of file system to search
1394 * @hashval: hash value (usually inode number) to search for
1395 * @match: callback used for comparisons between inodes
1396 * @data: opaque data pointer to pass to @match
1398 * Search for the inode specified by @hashval and @data in the inode
1399 * cache, where the helper function @match will return 0 if the inode
1400 * does not match, 1 if the inode does match, and -1 if the search
1401 * should be stopped. The @match function must be responsible for
1402 * taking the i_lock spin_lock and checking i_state for an inode being
1403 * freed or being initialized, and incrementing the reference count
1404 * before returning 1. It also must not sleep, since it is called with
1405 * the inode_hash_lock spinlock held.
1407 * This is a even more generalized version of ilookup5() when the
1408 * function must never block --- find_inode() can block in
1409 * __wait_on_freeing_inode() --- or when the caller can not increment
1410 * the reference count because the resulting iput() might cause an
1411 * inode eviction. The tradeoff is that the @match funtion must be
1412 * very carefully implemented.
1414 struct inode *find_inode_nowait(struct super_block *sb,
1415 unsigned long hashval,
1416 int (*match)(struct inode *, unsigned long,
1420 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1421 struct inode *inode, *ret_inode = NULL;
1424 spin_lock(&inode_hash_lock);
1425 hlist_for_each_entry(inode, head, i_hash) {
1426 if (inode->i_sb != sb)
1428 mval = match(inode, hashval, data);
1436 spin_unlock(&inode_hash_lock);
1439 EXPORT_SYMBOL(find_inode_nowait);
1441 int insert_inode_locked(struct inode *inode)
1443 struct super_block *sb = inode->i_sb;
1444 ino_t ino = inode->i_ino;
1445 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1448 struct inode *old = NULL;
1449 spin_lock(&inode_hash_lock);
1450 hlist_for_each_entry(old, head, i_hash) {
1451 if (old->i_ino != ino)
1453 if (old->i_sb != sb)
1455 spin_lock(&old->i_lock);
1456 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1457 spin_unlock(&old->i_lock);
1463 spin_lock(&inode->i_lock);
1464 inode->i_state |= I_NEW | I_CREATING;
1465 hlist_add_head(&inode->i_hash, head);
1466 spin_unlock(&inode->i_lock);
1467 spin_unlock(&inode_hash_lock);
1470 if (unlikely(old->i_state & I_CREATING)) {
1471 spin_unlock(&old->i_lock);
1472 spin_unlock(&inode_hash_lock);
1476 spin_unlock(&old->i_lock);
1477 spin_unlock(&inode_hash_lock);
1479 if (unlikely(!inode_unhashed(old))) {
1486 EXPORT_SYMBOL(insert_inode_locked);
1488 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1489 int (*test)(struct inode *, void *), void *data)
1493 inode->i_state |= I_CREATING;
1494 old = inode_insert5(inode, hashval, test, NULL, data);
1502 EXPORT_SYMBOL(insert_inode_locked4);
1505 int generic_delete_inode(struct inode *inode)
1509 EXPORT_SYMBOL(generic_delete_inode);
1512 * Called when we're dropping the last reference
1515 * Call the FS "drop_inode()" function, defaulting to
1516 * the legacy UNIX filesystem behaviour. If it tells
1517 * us to evict inode, do so. Otherwise, retain inode
1518 * in cache if fs is alive, sync and evict if fs is
1521 static void iput_final(struct inode *inode)
1523 struct super_block *sb = inode->i_sb;
1524 const struct super_operations *op = inode->i_sb->s_op;
1527 WARN_ON(inode->i_state & I_NEW);
1530 drop = op->drop_inode(inode);
1532 drop = generic_drop_inode(inode);
1534 if (!drop && (sb->s_flags & SB_ACTIVE)) {
1535 inode_add_lru(inode);
1536 spin_unlock(&inode->i_lock);
1541 inode->i_state |= I_WILL_FREE;
1542 spin_unlock(&inode->i_lock);
1543 write_inode_now(inode, 1);
1544 spin_lock(&inode->i_lock);
1545 WARN_ON(inode->i_state & I_NEW);
1546 inode->i_state &= ~I_WILL_FREE;
1549 inode->i_state |= I_FREEING;
1550 if (!list_empty(&inode->i_lru))
1551 inode_lru_list_del(inode);
1552 spin_unlock(&inode->i_lock);
1558 * iput - put an inode
1559 * @inode: inode to put
1561 * Puts an inode, dropping its usage count. If the inode use count hits
1562 * zero, the inode is then freed and may also be destroyed.
1564 * Consequently, iput() can sleep.
1566 void iput(struct inode *inode)
1570 BUG_ON(inode->i_state & I_CLEAR);
1572 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1573 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1574 atomic_inc(&inode->i_count);
1575 spin_unlock(&inode->i_lock);
1576 trace_writeback_lazytime_iput(inode);
1577 mark_inode_dirty_sync(inode);
1583 EXPORT_SYMBOL(iput);
1586 * bmap - find a block number in a file
1587 * @inode: inode of file
1588 * @block: block to find
1590 * Returns the block number on the device holding the inode that
1591 * is the disk block number for the block of the file requested.
1592 * That is, asked for block 4 of inode 1 the function will return the
1593 * disk block relative to the disk start that holds that block of the
1596 sector_t bmap(struct inode *inode, sector_t block)
1599 if (inode->i_mapping->a_ops->bmap)
1600 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1603 EXPORT_SYMBOL(bmap);
1606 * With relative atime, only update atime if the previous atime is
1607 * earlier than either the ctime or mtime or if at least a day has
1608 * passed since the last atime update.
1610 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1611 struct timespec now)
1614 if (!(mnt->mnt_flags & MNT_RELATIME))
1617 * Is mtime younger than atime? If yes, update atime:
1619 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1622 * Is ctime younger than atime? If yes, update atime:
1624 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1628 * Is the previous atime value older than a day? If yes,
1631 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1634 * Good, we can skip the atime update:
1639 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1641 int iflags = I_DIRTY_TIME;
1644 if (flags & S_ATIME)
1645 inode->i_atime = *time;
1646 if (flags & S_VERSION)
1647 dirty = inode_maybe_inc_iversion(inode, false);
1648 if (flags & S_CTIME)
1649 inode->i_ctime = *time;
1650 if (flags & S_MTIME)
1651 inode->i_mtime = *time;
1652 if ((flags & (S_ATIME | S_CTIME | S_MTIME)) &&
1653 !(inode->i_sb->s_flags & SB_LAZYTIME))
1657 iflags |= I_DIRTY_SYNC;
1658 __mark_inode_dirty(inode, iflags);
1661 EXPORT_SYMBOL(generic_update_time);
1664 * This does the actual work of updating an inodes time or version. Must have
1665 * had called mnt_want_write() before calling this.
1667 static int update_time(struct inode *inode, struct timespec64 *time, int flags)
1669 int (*update_time)(struct inode *, struct timespec64 *, int);
1671 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1672 generic_update_time;
1674 return update_time(inode, time, flags);
1678 * touch_atime - update the access time
1679 * @path: the &struct path to update
1680 * @inode: inode to update
1682 * Update the accessed time on an inode and mark it for writeback.
1683 * This function automatically handles read only file systems and media,
1684 * as well as the "noatime" flag and inode specific "noatime" markers.
1686 bool atime_needs_update(const struct path *path, struct inode *inode)
1688 struct vfsmount *mnt = path->mnt;
1689 struct timespec64 now;
1691 if (inode->i_flags & S_NOATIME)
1694 /* Atime updates will likely cause i_uid and i_gid to be written
1695 * back improprely if their true value is unknown to the vfs.
1697 if (HAS_UNMAPPED_ID(inode))
1700 if (IS_NOATIME(inode))
1702 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1705 if (mnt->mnt_flags & MNT_NOATIME)
1707 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1710 now = current_time(inode);
1712 if (!relatime_need_update(mnt, inode, timespec64_to_timespec(now)))
1715 if (timespec64_equal(&inode->i_atime, &now))
1721 void touch_atime(const struct path *path)
1723 struct vfsmount *mnt = path->mnt;
1724 struct inode *inode = d_inode(path->dentry);
1725 struct timespec64 now;
1727 if (!atime_needs_update(path, inode))
1730 if (!sb_start_write_trylock(inode->i_sb))
1733 if (__mnt_want_write(mnt) != 0)
1736 * File systems can error out when updating inodes if they need to
1737 * allocate new space to modify an inode (such is the case for
1738 * Btrfs), but since we touch atime while walking down the path we
1739 * really don't care if we failed to update the atime of the file,
1740 * so just ignore the return value.
1741 * We may also fail on filesystems that have the ability to make parts
1742 * of the fs read only, e.g. subvolumes in Btrfs.
1744 now = current_time(inode);
1745 update_time(inode, &now, S_ATIME);
1746 __mnt_drop_write(mnt);
1748 sb_end_write(inode->i_sb);
1750 EXPORT_SYMBOL(touch_atime);
1753 * The logic we want is
1755 * if suid or (sgid and xgrp)
1758 int should_remove_suid(struct dentry *dentry)
1760 umode_t mode = d_inode(dentry)->i_mode;
1763 /* suid always must be killed */
1764 if (unlikely(mode & S_ISUID))
1765 kill = ATTR_KILL_SUID;
1768 * sgid without any exec bits is just a mandatory locking mark; leave
1769 * it alone. If some exec bits are set, it's a real sgid; kill it.
1771 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1772 kill |= ATTR_KILL_SGID;
1774 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1779 EXPORT_SYMBOL(should_remove_suid);
1782 * Return mask of changes for notify_change() that need to be done as a
1783 * response to write or truncate. Return 0 if nothing has to be changed.
1784 * Negative value on error (change should be denied).
1786 int dentry_needs_remove_privs(struct dentry *dentry)
1788 struct inode *inode = d_inode(dentry);
1792 if (IS_NOSEC(inode))
1795 mask = should_remove_suid(dentry);
1796 ret = security_inode_need_killpriv(dentry);
1800 mask |= ATTR_KILL_PRIV;
1804 static int __remove_privs(struct dentry *dentry, int kill)
1806 struct iattr newattrs;
1808 newattrs.ia_valid = ATTR_FORCE | kill;
1810 * Note we call this on write, so notify_change will not
1811 * encounter any conflicting delegations:
1813 return notify_change(dentry, &newattrs, NULL);
1817 * Remove special file priviledges (suid, capabilities) when file is written
1820 int file_remove_privs(struct file *file)
1822 struct dentry *dentry = file_dentry(file);
1823 struct inode *inode = file_inode(file);
1828 * Fast path for nothing security related.
1829 * As well for non-regular files, e.g. blkdev inodes.
1830 * For example, blkdev_write_iter() might get here
1831 * trying to remove privs which it is not allowed to.
1833 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1836 kill = dentry_needs_remove_privs(dentry);
1840 error = __remove_privs(dentry, kill);
1842 inode_has_no_xattr(inode);
1846 EXPORT_SYMBOL(file_remove_privs);
1849 * file_update_time - update mtime and ctime time
1850 * @file: file accessed
1852 * Update the mtime and ctime members of an inode and mark the inode
1853 * for writeback. Note that this function is meant exclusively for
1854 * usage in the file write path of filesystems, and filesystems may
1855 * choose to explicitly ignore update via this function with the
1856 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1857 * timestamps are handled by the server. This can return an error for
1858 * file systems who need to allocate space in order to update an inode.
1861 int file_update_time(struct file *file)
1863 struct inode *inode = file_inode(file);
1864 struct timespec64 now;
1868 /* First try to exhaust all avenues to not sync */
1869 if (IS_NOCMTIME(inode))
1872 now = current_time(inode);
1873 if (!timespec64_equal(&inode->i_mtime, &now))
1876 if (!timespec64_equal(&inode->i_ctime, &now))
1879 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
1880 sync_it |= S_VERSION;
1885 /* Finally allowed to write? Takes lock. */
1886 if (__mnt_want_write_file(file))
1889 ret = update_time(inode, &now, sync_it);
1890 __mnt_drop_write_file(file);
1894 EXPORT_SYMBOL(file_update_time);
1896 int inode_needs_sync(struct inode *inode)
1900 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1904 EXPORT_SYMBOL(inode_needs_sync);
1907 * If we try to find an inode in the inode hash while it is being
1908 * deleted, we have to wait until the filesystem completes its
1909 * deletion before reporting that it isn't found. This function waits
1910 * until the deletion _might_ have completed. Callers are responsible
1911 * to recheck inode state.
1913 * It doesn't matter if I_NEW is not set initially, a call to
1914 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1917 static void __wait_on_freeing_inode(struct inode *inode)
1919 wait_queue_head_t *wq;
1920 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1921 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1922 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1923 spin_unlock(&inode->i_lock);
1924 spin_unlock(&inode_hash_lock);
1926 finish_wait(wq, &wait.wq_entry);
1927 spin_lock(&inode_hash_lock);
1930 static __initdata unsigned long ihash_entries;
1931 static int __init set_ihash_entries(char *str)
1935 ihash_entries = simple_strtoul(str, &str, 0);
1938 __setup("ihash_entries=", set_ihash_entries);
1941 * Initialize the waitqueues and inode hash table.
1943 void __init inode_init_early(void)
1945 /* If hashes are distributed across NUMA nodes, defer
1946 * hash allocation until vmalloc space is available.
1952 alloc_large_system_hash("Inode-cache",
1953 sizeof(struct hlist_head),
1956 HASH_EARLY | HASH_ZERO,
1963 void __init inode_init(void)
1965 /* inode slab cache */
1966 inode_cachep = kmem_cache_create("inode_cache",
1967 sizeof(struct inode),
1969 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1970 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1973 /* Hash may have been set up in inode_init_early */
1978 alloc_large_system_hash("Inode-cache",
1979 sizeof(struct hlist_head),
1989 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1991 inode->i_mode = mode;
1992 if (S_ISCHR(mode)) {
1993 inode->i_fop = &def_chr_fops;
1994 inode->i_rdev = rdev;
1995 } else if (S_ISBLK(mode)) {
1996 inode->i_fop = &def_blk_fops;
1997 inode->i_rdev = rdev;
1998 } else if (S_ISFIFO(mode))
1999 inode->i_fop = &pipefifo_fops;
2000 else if (S_ISSOCK(mode))
2001 ; /* leave it no_open_fops */
2003 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2004 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2007 EXPORT_SYMBOL(init_special_inode);
2010 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2012 * @dir: Directory inode
2013 * @mode: mode of the new inode
2015 void inode_init_owner(struct inode *inode, const struct inode *dir,
2018 inode->i_uid = current_fsuid();
2019 if (dir && dir->i_mode & S_ISGID) {
2020 inode->i_gid = dir->i_gid;
2022 /* Directories are special, and always inherit S_ISGID */
2025 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2026 !in_group_p(inode->i_gid) &&
2027 !capable_wrt_inode_uidgid(dir, CAP_FSETID))
2030 inode->i_gid = current_fsgid();
2031 inode->i_mode = mode;
2033 EXPORT_SYMBOL(inode_init_owner);
2036 * inode_owner_or_capable - check current task permissions to inode
2037 * @inode: inode being checked
2039 * Return true if current either has CAP_FOWNER in a namespace with the
2040 * inode owner uid mapped, or owns the file.
2042 bool inode_owner_or_capable(const struct inode *inode)
2044 struct user_namespace *ns;
2046 if (uid_eq(current_fsuid(), inode->i_uid))
2049 ns = current_user_ns();
2050 if (kuid_has_mapping(ns, inode->i_uid) && ns_capable(ns, CAP_FOWNER))
2054 EXPORT_SYMBOL(inode_owner_or_capable);
2057 * Direct i/o helper functions
2059 static void __inode_dio_wait(struct inode *inode)
2061 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2062 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2065 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2066 if (atomic_read(&inode->i_dio_count))
2068 } while (atomic_read(&inode->i_dio_count));
2069 finish_wait(wq, &q.wq_entry);
2073 * inode_dio_wait - wait for outstanding DIO requests to finish
2074 * @inode: inode to wait for
2076 * Waits for all pending direct I/O requests to finish so that we can
2077 * proceed with a truncate or equivalent operation.
2079 * Must be called under a lock that serializes taking new references
2080 * to i_dio_count, usually by inode->i_mutex.
2082 void inode_dio_wait(struct inode *inode)
2084 if (atomic_read(&inode->i_dio_count))
2085 __inode_dio_wait(inode);
2087 EXPORT_SYMBOL(inode_dio_wait);
2090 * inode_set_flags - atomically set some inode flags
2092 * Note: the caller should be holding i_mutex, or else be sure that
2093 * they have exclusive access to the inode structure (i.e., while the
2094 * inode is being instantiated). The reason for the cmpxchg() loop
2095 * --- which wouldn't be necessary if all code paths which modify
2096 * i_flags actually followed this rule, is that there is at least one
2097 * code path which doesn't today so we use cmpxchg() out of an abundance
2100 * In the long run, i_mutex is overkill, and we should probably look
2101 * at using the i_lock spinlock to protect i_flags, and then make sure
2102 * it is so documented in include/linux/fs.h and that all code follows
2103 * the locking convention!!
2105 void inode_set_flags(struct inode *inode, unsigned int flags,
2108 unsigned int old_flags, new_flags;
2110 WARN_ON_ONCE(flags & ~mask);
2112 old_flags = READ_ONCE(inode->i_flags);
2113 new_flags = (old_flags & ~mask) | flags;
2114 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2115 new_flags) != old_flags));
2117 EXPORT_SYMBOL(inode_set_flags);
2119 void inode_nohighmem(struct inode *inode)
2121 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2123 EXPORT_SYMBOL(inode_nohighmem);
2126 * timespec64_trunc - Truncate timespec64 to a granularity
2128 * @gran: Granularity in ns.
2130 * Truncate a timespec64 to a granularity. Always rounds down. gran must
2131 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2133 struct timespec64 timespec64_trunc(struct timespec64 t, unsigned gran)
2135 /* Avoid division in the common cases 1 ns and 1 s. */
2138 } else if (gran == NSEC_PER_SEC) {
2140 } else if (gran > 1 && gran < NSEC_PER_SEC) {
2141 t.tv_nsec -= t.tv_nsec % gran;
2143 WARN(1, "illegal file time granularity: %u", gran);
2147 EXPORT_SYMBOL(timespec64_trunc);
2150 * current_time - Return FS time
2153 * Return the current time truncated to the time granularity supported by
2156 * Note that inode and inode->sb cannot be NULL.
2157 * Otherwise, the function warns and returns time without truncation.
2159 struct timespec64 current_time(struct inode *inode)
2161 struct timespec64 now = current_kernel_time64();
2163 if (unlikely(!inode->i_sb)) {
2164 WARN(1, "current_time() called with uninitialized super_block in the inode");
2168 return timespec64_trunc(now, inode->i_sb->s_time_gran);
2170 EXPORT_SYMBOL(current_time);