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 <trace/events/writeback.h>
25 * Inode locking rules:
27 * inode->i_lock protects:
28 * inode->i_state, inode->i_hash, __iget()
29 * Inode LRU list locks protect:
30 * inode->i_sb->s_inode_lru, inode->i_lru
31 * inode->i_sb->s_inode_list_lock protects:
32 * inode->i_sb->s_inodes, inode->i_sb_list
33 * bdi->wb.list_lock protects:
34 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
35 * inode_hash_lock protects:
36 * inode_hashtable, inode->i_hash
40 * inode->i_sb->s_inode_list_lock
42 * Inode LRU list locks
48 * inode->i_sb->s_inode_list_lock
55 static unsigned int i_hash_mask __read_mostly;
56 static unsigned int i_hash_shift __read_mostly;
57 static struct hlist_head *inode_hashtable __read_mostly;
58 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
61 * Empty aops. Can be used for the cases where the user does not
62 * define any of the address_space operations.
64 const struct address_space_operations empty_aops = {
66 EXPORT_SYMBOL(empty_aops);
69 * Statistics gathering..
71 struct inodes_stat_t inodes_stat;
73 static DEFINE_PER_CPU(unsigned long, nr_inodes);
74 static DEFINE_PER_CPU(unsigned long, nr_unused);
76 static struct kmem_cache *inode_cachep __read_mostly;
78 static long get_nr_inodes(void)
82 for_each_possible_cpu(i)
83 sum += per_cpu(nr_inodes, i);
84 return sum < 0 ? 0 : sum;
87 static inline long get_nr_inodes_unused(void)
91 for_each_possible_cpu(i)
92 sum += per_cpu(nr_unused, i);
93 return sum < 0 ? 0 : sum;
96 long get_nr_dirty_inodes(void)
98 /* not actually dirty inodes, but a wild approximation */
99 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
100 return nr_dirty > 0 ? nr_dirty : 0;
104 * Handle nr_inode sysctl
107 int proc_nr_inodes(struct ctl_table *table, int write,
108 void __user *buffer, size_t *lenp, loff_t *ppos)
110 inodes_stat.nr_inodes = get_nr_inodes();
111 inodes_stat.nr_unused = get_nr_inodes_unused();
112 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
116 static int no_open(struct inode *inode, struct file *file)
122 * inode_init_always - perform inode structure intialisation
123 * @sb: superblock inode belongs to
124 * @inode: inode to initialise
126 * These are initializations that need to be done on every inode
127 * allocation as the fields are not initialised by slab allocation.
129 int inode_init_always(struct super_block *sb, struct inode *inode)
131 static const struct inode_operations empty_iops;
132 static const struct file_operations no_open_fops = {.open = no_open};
133 struct address_space *const mapping = &inode->i_data;
136 inode->i_blkbits = sb->s_blocksize_bits;
138 atomic_set(&inode->i_count, 1);
139 inode->i_op = &empty_iops;
140 inode->i_fop = &no_open_fops;
141 inode->__i_nlink = 1;
142 inode->i_opflags = 0;
143 i_uid_write(inode, 0);
144 i_gid_write(inode, 0);
145 atomic_set(&inode->i_writecount, 0);
149 inode->i_generation = 0;
150 inode->i_pipe = NULL;
151 inode->i_bdev = NULL;
152 inode->i_cdev = NULL;
153 inode->i_link = NULL;
154 inode->i_dir_seq = 0;
156 inode->dirtied_when = 0;
158 #ifdef CONFIG_CGROUP_WRITEBACK
159 inode->i_wb_frn_winner = 0;
160 inode->i_wb_frn_avg_time = 0;
161 inode->i_wb_frn_history = 0;
164 if (security_inode_alloc(inode))
166 spin_lock_init(&inode->i_lock);
167 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
169 init_rwsem(&inode->i_rwsem);
170 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
172 atomic_set(&inode->i_dio_count, 0);
174 mapping->a_ops = &empty_aops;
175 mapping->host = inode;
177 atomic_set(&mapping->i_mmap_writable, 0);
178 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
179 mapping->private_data = NULL;
180 mapping->writeback_index = 0;
181 inode->i_private = NULL;
182 inode->i_mapping = mapping;
183 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
184 #ifdef CONFIG_FS_POSIX_ACL
185 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
188 #ifdef CONFIG_FSNOTIFY
189 inode->i_fsnotify_mask = 0;
191 inode->i_flctx = NULL;
192 this_cpu_inc(nr_inodes);
198 EXPORT_SYMBOL(inode_init_always);
200 static struct inode *alloc_inode(struct super_block *sb)
204 if (sb->s_op->alloc_inode)
205 inode = sb->s_op->alloc_inode(sb);
207 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
212 if (unlikely(inode_init_always(sb, inode))) {
213 if (inode->i_sb->s_op->destroy_inode)
214 inode->i_sb->s_op->destroy_inode(inode);
216 kmem_cache_free(inode_cachep, inode);
223 void free_inode_nonrcu(struct inode *inode)
225 kmem_cache_free(inode_cachep, inode);
227 EXPORT_SYMBOL(free_inode_nonrcu);
229 void __destroy_inode(struct inode *inode)
231 BUG_ON(inode_has_buffers(inode));
232 inode_detach_wb(inode);
233 security_inode_free(inode);
234 fsnotify_inode_delete(inode);
235 locks_free_lock_context(inode);
236 if (!inode->i_nlink) {
237 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
238 atomic_long_dec(&inode->i_sb->s_remove_count);
241 #ifdef CONFIG_FS_POSIX_ACL
242 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
243 posix_acl_release(inode->i_acl);
244 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
245 posix_acl_release(inode->i_default_acl);
247 this_cpu_dec(nr_inodes);
249 EXPORT_SYMBOL(__destroy_inode);
251 static void i_callback(struct rcu_head *head)
253 struct inode *inode = container_of(head, struct inode, i_rcu);
254 kmem_cache_free(inode_cachep, inode);
257 static void destroy_inode(struct inode *inode)
259 BUG_ON(!list_empty(&inode->i_lru));
260 __destroy_inode(inode);
261 if (inode->i_sb->s_op->destroy_inode)
262 inode->i_sb->s_op->destroy_inode(inode);
264 call_rcu(&inode->i_rcu, i_callback);
268 * drop_nlink - directly drop an inode's link count
271 * This is a low-level filesystem helper to replace any
272 * direct filesystem manipulation of i_nlink. In cases
273 * where we are attempting to track writes to the
274 * filesystem, a decrement to zero means an imminent
275 * write when the file is truncated and actually unlinked
278 void drop_nlink(struct inode *inode)
280 WARN_ON(inode->i_nlink == 0);
283 atomic_long_inc(&inode->i_sb->s_remove_count);
285 EXPORT_SYMBOL(drop_nlink);
288 * clear_nlink - directly zero an inode's link count
291 * This is a low-level filesystem helper to replace any
292 * direct filesystem manipulation of i_nlink. See
293 * drop_nlink() for why we care about i_nlink hitting zero.
295 void clear_nlink(struct inode *inode)
297 if (inode->i_nlink) {
298 inode->__i_nlink = 0;
299 atomic_long_inc(&inode->i_sb->s_remove_count);
302 EXPORT_SYMBOL(clear_nlink);
305 * set_nlink - directly set an inode's link count
307 * @nlink: new nlink (should be non-zero)
309 * This is a low-level filesystem helper to replace any
310 * direct filesystem manipulation of i_nlink.
312 void set_nlink(struct inode *inode, unsigned int nlink)
317 /* Yes, some filesystems do change nlink from zero to one */
318 if (inode->i_nlink == 0)
319 atomic_long_dec(&inode->i_sb->s_remove_count);
321 inode->__i_nlink = nlink;
324 EXPORT_SYMBOL(set_nlink);
327 * inc_nlink - directly increment an inode's link count
330 * This is a low-level filesystem helper to replace any
331 * direct filesystem manipulation of i_nlink. Currently,
332 * it is only here for parity with dec_nlink().
334 void inc_nlink(struct inode *inode)
336 if (unlikely(inode->i_nlink == 0)) {
337 WARN_ON(!(inode->i_state & I_LINKABLE));
338 atomic_long_dec(&inode->i_sb->s_remove_count);
343 EXPORT_SYMBOL(inc_nlink);
345 void address_space_init_once(struct address_space *mapping)
347 memset(mapping, 0, sizeof(*mapping));
348 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC | __GFP_ACCOUNT);
349 spin_lock_init(&mapping->tree_lock);
350 init_rwsem(&mapping->i_mmap_rwsem);
351 INIT_LIST_HEAD(&mapping->private_list);
352 spin_lock_init(&mapping->private_lock);
353 mapping->i_mmap = RB_ROOT;
355 EXPORT_SYMBOL(address_space_init_once);
358 * These are initializations that only need to be done
359 * once, because the fields are idempotent across use
360 * of the inode, so let the slab aware of that.
362 void inode_init_once(struct inode *inode)
364 memset(inode, 0, sizeof(*inode));
365 INIT_HLIST_NODE(&inode->i_hash);
366 INIT_LIST_HEAD(&inode->i_devices);
367 INIT_LIST_HEAD(&inode->i_io_list);
368 INIT_LIST_HEAD(&inode->i_wb_list);
369 INIT_LIST_HEAD(&inode->i_lru);
370 address_space_init_once(&inode->i_data);
371 i_size_ordered_init(inode);
372 #ifdef CONFIG_FSNOTIFY
373 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
376 EXPORT_SYMBOL(inode_init_once);
378 static void init_once(void *foo)
380 struct inode *inode = (struct inode *) foo;
382 inode_init_once(inode);
386 * inode->i_lock must be held
388 void __iget(struct inode *inode)
390 atomic_inc(&inode->i_count);
394 * get additional reference to inode; caller must already hold one.
396 void ihold(struct inode *inode)
398 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
400 EXPORT_SYMBOL(ihold);
402 static void inode_lru_list_add(struct inode *inode)
404 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
405 this_cpu_inc(nr_unused);
409 * Add inode to LRU if needed (inode is unused and clean).
411 * Needs inode->i_lock held.
413 void inode_add_lru(struct inode *inode)
415 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
416 I_FREEING | I_WILL_FREE)) &&
417 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
418 inode_lru_list_add(inode);
422 static void inode_lru_list_del(struct inode *inode)
425 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
426 this_cpu_dec(nr_unused);
430 * inode_sb_list_add - add inode to the superblock list of inodes
431 * @inode: inode to add
433 void inode_sb_list_add(struct inode *inode)
435 spin_lock(&inode->i_sb->s_inode_list_lock);
436 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
437 spin_unlock(&inode->i_sb->s_inode_list_lock);
439 EXPORT_SYMBOL_GPL(inode_sb_list_add);
441 static inline void inode_sb_list_del(struct inode *inode)
443 if (!list_empty(&inode->i_sb_list)) {
444 spin_lock(&inode->i_sb->s_inode_list_lock);
445 list_del_init(&inode->i_sb_list);
446 spin_unlock(&inode->i_sb->s_inode_list_lock);
450 static unsigned long hash(struct super_block *sb, unsigned long hashval)
454 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
456 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
457 return tmp & i_hash_mask;
461 * __insert_inode_hash - hash an inode
462 * @inode: unhashed inode
463 * @hashval: unsigned long value used to locate this object in the
466 * Add an inode to the inode hash for this superblock.
468 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
470 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
472 spin_lock(&inode_hash_lock);
473 spin_lock(&inode->i_lock);
474 hlist_add_head(&inode->i_hash, b);
475 spin_unlock(&inode->i_lock);
476 spin_unlock(&inode_hash_lock);
478 EXPORT_SYMBOL(__insert_inode_hash);
481 * __remove_inode_hash - remove an inode from the hash
482 * @inode: inode to unhash
484 * Remove an inode from the superblock.
486 void __remove_inode_hash(struct inode *inode)
488 spin_lock(&inode_hash_lock);
489 spin_lock(&inode->i_lock);
490 hlist_del_init(&inode->i_hash);
491 spin_unlock(&inode->i_lock);
492 spin_unlock(&inode_hash_lock);
494 EXPORT_SYMBOL(__remove_inode_hash);
496 void clear_inode(struct inode *inode)
500 * We have to cycle tree_lock here because reclaim can be still in the
501 * process of removing the last page (in __delete_from_page_cache())
502 * and we must not free mapping under it.
504 spin_lock_irq(&inode->i_data.tree_lock);
505 BUG_ON(inode->i_data.nrpages);
506 BUG_ON(inode->i_data.nrexceptional);
507 spin_unlock_irq(&inode->i_data.tree_lock);
508 BUG_ON(!list_empty(&inode->i_data.private_list));
509 BUG_ON(!(inode->i_state & I_FREEING));
510 BUG_ON(inode->i_state & I_CLEAR);
511 BUG_ON(!list_empty(&inode->i_wb_list));
512 /* don't need i_lock here, no concurrent mods to i_state */
513 inode->i_state = I_FREEING | I_CLEAR;
515 EXPORT_SYMBOL(clear_inode);
518 * Free the inode passed in, removing it from the lists it is still connected
519 * to. We remove any pages still attached to the inode and wait for any IO that
520 * is still in progress before finally destroying the inode.
522 * An inode must already be marked I_FREEING so that we avoid the inode being
523 * moved back onto lists if we race with other code that manipulates the lists
524 * (e.g. writeback_single_inode). The caller is responsible for setting this.
526 * An inode must already be removed from the LRU list before being evicted from
527 * the cache. This should occur atomically with setting the I_FREEING state
528 * flag, so no inodes here should ever be on the LRU when being evicted.
530 static void evict(struct inode *inode)
532 const struct super_operations *op = inode->i_sb->s_op;
534 BUG_ON(!(inode->i_state & I_FREEING));
535 BUG_ON(!list_empty(&inode->i_lru));
537 if (!list_empty(&inode->i_io_list))
538 inode_io_list_del(inode);
540 inode_sb_list_del(inode);
543 * Wait for flusher thread to be done with the inode so that filesystem
544 * does not start destroying it while writeback is still running. Since
545 * the inode has I_FREEING set, flusher thread won't start new work on
546 * the inode. We just have to wait for running writeback to finish.
548 inode_wait_for_writeback(inode);
550 if (op->evict_inode) {
551 op->evict_inode(inode);
553 truncate_inode_pages_final(&inode->i_data);
556 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
558 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
561 remove_inode_hash(inode);
563 spin_lock(&inode->i_lock);
564 wake_up_bit(&inode->i_state, __I_NEW);
565 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
566 spin_unlock(&inode->i_lock);
568 destroy_inode(inode);
572 * dispose_list - dispose of the contents of a local list
573 * @head: the head of the list to free
575 * Dispose-list gets a local list with local inodes in it, so it doesn't
576 * need to worry about list corruption and SMP locks.
578 static void dispose_list(struct list_head *head)
580 while (!list_empty(head)) {
583 inode = list_first_entry(head, struct inode, i_lru);
584 list_del_init(&inode->i_lru);
592 * evict_inodes - evict all evictable inodes for a superblock
593 * @sb: superblock to operate on
595 * Make sure that no inodes with zero refcount are retained. This is
596 * called by superblock shutdown after having MS_ACTIVE flag removed,
597 * so any inode reaching zero refcount during or after that call will
598 * be immediately evicted.
600 void evict_inodes(struct super_block *sb)
602 struct inode *inode, *next;
606 spin_lock(&sb->s_inode_list_lock);
607 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
608 if (atomic_read(&inode->i_count))
611 spin_lock(&inode->i_lock);
612 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
613 spin_unlock(&inode->i_lock);
617 inode->i_state |= I_FREEING;
618 inode_lru_list_del(inode);
619 spin_unlock(&inode->i_lock);
620 list_add(&inode->i_lru, &dispose);
623 * We can have a ton of inodes to evict at unmount time given
624 * enough memory, check to see if we need to go to sleep for a
625 * bit so we don't livelock.
627 if (need_resched()) {
628 spin_unlock(&sb->s_inode_list_lock);
630 dispose_list(&dispose);
634 spin_unlock(&sb->s_inode_list_lock);
636 dispose_list(&dispose);
640 * invalidate_inodes - attempt to free all inodes on a superblock
641 * @sb: superblock to operate on
642 * @kill_dirty: flag to guide handling of dirty inodes
644 * Attempts to free all inodes for a given superblock. If there were any
645 * busy inodes return a non-zero value, else zero.
646 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
649 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
652 struct inode *inode, *next;
655 spin_lock(&sb->s_inode_list_lock);
656 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
657 spin_lock(&inode->i_lock);
658 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
659 spin_unlock(&inode->i_lock);
662 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
663 spin_unlock(&inode->i_lock);
667 if (atomic_read(&inode->i_count)) {
668 spin_unlock(&inode->i_lock);
673 inode->i_state |= I_FREEING;
674 inode_lru_list_del(inode);
675 spin_unlock(&inode->i_lock);
676 list_add(&inode->i_lru, &dispose);
678 spin_unlock(&sb->s_inode_list_lock);
680 dispose_list(&dispose);
686 * Isolate the inode from the LRU in preparation for freeing it.
688 * Any inodes which are pinned purely because of attached pagecache have their
689 * pagecache removed. If the inode has metadata buffers attached to
690 * mapping->private_list then try to remove them.
692 * If the inode has the I_REFERENCED flag set, then it means that it has been
693 * used recently - the flag is set in iput_final(). When we encounter such an
694 * inode, clear the flag and move it to the back of the LRU so it gets another
695 * pass through the LRU before it gets reclaimed. This is necessary because of
696 * the fact we are doing lazy LRU updates to minimise lock contention so the
697 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
698 * with this flag set because they are the inodes that are out of order.
700 static enum lru_status inode_lru_isolate(struct list_head *item,
701 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
703 struct list_head *freeable = arg;
704 struct inode *inode = container_of(item, struct inode, i_lru);
707 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
708 * If we fail to get the lock, just skip it.
710 if (!spin_trylock(&inode->i_lock))
714 * Referenced or dirty inodes are still in use. Give them another pass
715 * through the LRU as we canot reclaim them now.
717 if (atomic_read(&inode->i_count) ||
718 (inode->i_state & ~I_REFERENCED)) {
719 list_lru_isolate(lru, &inode->i_lru);
720 spin_unlock(&inode->i_lock);
721 this_cpu_dec(nr_unused);
725 /* recently referenced inodes get one more pass */
726 if (inode->i_state & I_REFERENCED) {
727 inode->i_state &= ~I_REFERENCED;
728 spin_unlock(&inode->i_lock);
732 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
734 spin_unlock(&inode->i_lock);
735 spin_unlock(lru_lock);
736 if (remove_inode_buffers(inode)) {
738 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
739 if (current_is_kswapd())
740 __count_vm_events(KSWAPD_INODESTEAL, reap);
742 __count_vm_events(PGINODESTEAL, reap);
743 if (current->reclaim_state)
744 current->reclaim_state->reclaimed_slab += reap;
751 WARN_ON(inode->i_state & I_NEW);
752 inode->i_state |= I_FREEING;
753 list_lru_isolate_move(lru, &inode->i_lru, freeable);
754 spin_unlock(&inode->i_lock);
756 this_cpu_dec(nr_unused);
761 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
762 * This is called from the superblock shrinker function with a number of inodes
763 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
764 * then are freed outside inode_lock by dispose_list().
766 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
771 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
772 inode_lru_isolate, &freeable);
773 dispose_list(&freeable);
777 static void __wait_on_freeing_inode(struct inode *inode);
779 * Called with the inode lock held.
781 static struct inode *find_inode(struct super_block *sb,
782 struct hlist_head *head,
783 int (*test)(struct inode *, void *),
786 struct inode *inode = NULL;
789 hlist_for_each_entry(inode, head, i_hash) {
790 if (inode->i_sb != sb)
792 if (!test(inode, data))
794 spin_lock(&inode->i_lock);
795 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
796 __wait_on_freeing_inode(inode);
800 spin_unlock(&inode->i_lock);
807 * find_inode_fast is the fast path version of find_inode, see the comment at
808 * iget_locked for details.
810 static struct inode *find_inode_fast(struct super_block *sb,
811 struct hlist_head *head, unsigned long ino)
813 struct inode *inode = NULL;
816 hlist_for_each_entry(inode, head, i_hash) {
817 if (inode->i_ino != ino)
819 if (inode->i_sb != sb)
821 spin_lock(&inode->i_lock);
822 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
823 __wait_on_freeing_inode(inode);
827 spin_unlock(&inode->i_lock);
834 * Each cpu owns a range of LAST_INO_BATCH numbers.
835 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
836 * to renew the exhausted range.
838 * This does not significantly increase overflow rate because every CPU can
839 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
840 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
841 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
842 * overflow rate by 2x, which does not seem too significant.
844 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
845 * error if st_ino won't fit in target struct field. Use 32bit counter
846 * here to attempt to avoid that.
848 #define LAST_INO_BATCH 1024
849 static DEFINE_PER_CPU(unsigned int, last_ino);
851 unsigned int get_next_ino(void)
853 unsigned int *p = &get_cpu_var(last_ino);
854 unsigned int res = *p;
857 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
858 static atomic_t shared_last_ino;
859 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
861 res = next - LAST_INO_BATCH;
866 /* get_next_ino should not provide a 0 inode number */
870 put_cpu_var(last_ino);
873 EXPORT_SYMBOL(get_next_ino);
876 * new_inode_pseudo - obtain an inode
879 * Allocates a new inode for given superblock.
880 * Inode wont be chained in superblock s_inodes list
882 * - fs can't be unmount
883 * - quotas, fsnotify, writeback can't work
885 struct inode *new_inode_pseudo(struct super_block *sb)
887 struct inode *inode = alloc_inode(sb);
890 spin_lock(&inode->i_lock);
892 spin_unlock(&inode->i_lock);
893 INIT_LIST_HEAD(&inode->i_sb_list);
899 * new_inode - obtain an inode
902 * Allocates a new inode for given superblock. The default gfp_mask
903 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
904 * If HIGHMEM pages are unsuitable or it is known that pages allocated
905 * for the page cache are not reclaimable or migratable,
906 * mapping_set_gfp_mask() must be called with suitable flags on the
907 * newly created inode's mapping
910 struct inode *new_inode(struct super_block *sb)
914 spin_lock_prefetch(&sb->s_inode_list_lock);
916 inode = new_inode_pseudo(sb);
918 inode_sb_list_add(inode);
921 EXPORT_SYMBOL(new_inode);
923 #ifdef CONFIG_DEBUG_LOCK_ALLOC
924 void lockdep_annotate_inode_mutex_key(struct inode *inode)
926 if (S_ISDIR(inode->i_mode)) {
927 struct file_system_type *type = inode->i_sb->s_type;
929 /* Set new key only if filesystem hasn't already changed it */
930 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
932 * ensure nobody is actually holding i_mutex
934 // mutex_destroy(&inode->i_mutex);
935 init_rwsem(&inode->i_rwsem);
936 lockdep_set_class(&inode->i_rwsem,
937 &type->i_mutex_dir_key);
941 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
945 * unlock_new_inode - clear the I_NEW state and wake up any waiters
946 * @inode: new inode to unlock
948 * Called when the inode is fully initialised to clear the new state of the
949 * inode and wake up anyone waiting for the inode to finish initialisation.
951 void unlock_new_inode(struct inode *inode)
953 lockdep_annotate_inode_mutex_key(inode);
954 spin_lock(&inode->i_lock);
955 WARN_ON(!(inode->i_state & I_NEW));
956 inode->i_state &= ~I_NEW;
958 wake_up_bit(&inode->i_state, __I_NEW);
959 spin_unlock(&inode->i_lock);
961 EXPORT_SYMBOL(unlock_new_inode);
964 * lock_two_nondirectories - take two i_mutexes on non-directory objects
966 * Lock any non-NULL argument that is not a directory.
967 * Zero, one or two objects may be locked by this function.
969 * @inode1: first inode to lock
970 * @inode2: second inode to lock
972 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
975 swap(inode1, inode2);
977 if (inode1 && !S_ISDIR(inode1->i_mode))
979 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
980 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
982 EXPORT_SYMBOL(lock_two_nondirectories);
985 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
986 * @inode1: first inode to unlock
987 * @inode2: second inode to unlock
989 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
991 if (inode1 && !S_ISDIR(inode1->i_mode))
992 inode_unlock(inode1);
993 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
994 inode_unlock(inode2);
996 EXPORT_SYMBOL(unlock_two_nondirectories);
999 * iget5_locked - obtain an inode from a mounted file system
1000 * @sb: super block of file system
1001 * @hashval: hash value (usually inode number) to get
1002 * @test: callback used for comparisons between inodes
1003 * @set: callback used to initialize a new struct inode
1004 * @data: opaque data pointer to pass to @test and @set
1006 * Search for the inode specified by @hashval and @data in the inode cache,
1007 * and if present it is return it with an increased reference count. This is
1008 * a generalized version of iget_locked() for file systems where the inode
1009 * number is not sufficient for unique identification of an inode.
1011 * If the inode is not in cache, allocate a new inode and return it locked,
1012 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1013 * before unlocking it via unlock_new_inode().
1015 * Note both @test and @set are called with the inode_hash_lock held, so can't
1018 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1019 int (*test)(struct inode *, void *),
1020 int (*set)(struct inode *, void *), void *data)
1022 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1023 struct inode *inode;
1025 spin_lock(&inode_hash_lock);
1026 inode = find_inode(sb, head, test, data);
1027 spin_unlock(&inode_hash_lock);
1030 wait_on_inode(inode);
1031 if (unlikely(inode_unhashed(inode))) {
1038 inode = alloc_inode(sb);
1042 spin_lock(&inode_hash_lock);
1043 /* We released the lock, so.. */
1044 old = find_inode(sb, head, test, data);
1046 if (set(inode, data))
1049 spin_lock(&inode->i_lock);
1050 inode->i_state = I_NEW;
1051 hlist_add_head(&inode->i_hash, head);
1052 spin_unlock(&inode->i_lock);
1053 inode_sb_list_add(inode);
1054 spin_unlock(&inode_hash_lock);
1056 /* Return the locked inode with I_NEW set, the
1057 * caller is responsible for filling in the contents
1063 * Uhhuh, somebody else created the same inode under
1064 * us. Use the old inode instead of the one we just
1067 spin_unlock(&inode_hash_lock);
1068 destroy_inode(inode);
1070 wait_on_inode(inode);
1071 if (unlikely(inode_unhashed(inode))) {
1079 spin_unlock(&inode_hash_lock);
1080 destroy_inode(inode);
1083 EXPORT_SYMBOL(iget5_locked);
1086 * iget_locked - obtain an inode from a mounted file system
1087 * @sb: super block of file system
1088 * @ino: inode number to get
1090 * Search for the inode specified by @ino in the inode cache and if present
1091 * return it with an increased reference count. This is for file systems
1092 * where the inode number is sufficient for unique identification of an inode.
1094 * If the inode is not in cache, allocate a new inode and return it locked,
1095 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1096 * before unlocking it via unlock_new_inode().
1098 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1100 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1101 struct inode *inode;
1103 spin_lock(&inode_hash_lock);
1104 inode = find_inode_fast(sb, head, ino);
1105 spin_unlock(&inode_hash_lock);
1107 wait_on_inode(inode);
1108 if (unlikely(inode_unhashed(inode))) {
1115 inode = alloc_inode(sb);
1119 spin_lock(&inode_hash_lock);
1120 /* We released the lock, so.. */
1121 old = find_inode_fast(sb, head, ino);
1124 spin_lock(&inode->i_lock);
1125 inode->i_state = I_NEW;
1126 hlist_add_head(&inode->i_hash, head);
1127 spin_unlock(&inode->i_lock);
1128 inode_sb_list_add(inode);
1129 spin_unlock(&inode_hash_lock);
1131 /* Return the locked inode with I_NEW set, the
1132 * caller is responsible for filling in the contents
1138 * Uhhuh, somebody else created the same inode under
1139 * us. Use the old inode instead of the one we just
1142 spin_unlock(&inode_hash_lock);
1143 destroy_inode(inode);
1145 wait_on_inode(inode);
1146 if (unlikely(inode_unhashed(inode))) {
1153 EXPORT_SYMBOL(iget_locked);
1156 * search the inode cache for a matching inode number.
1157 * If we find one, then the inode number we are trying to
1158 * allocate is not unique and so we should not use it.
1160 * Returns 1 if the inode number is unique, 0 if it is not.
1162 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1164 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1165 struct inode *inode;
1167 spin_lock(&inode_hash_lock);
1168 hlist_for_each_entry(inode, b, i_hash) {
1169 if (inode->i_ino == ino && inode->i_sb == sb) {
1170 spin_unlock(&inode_hash_lock);
1174 spin_unlock(&inode_hash_lock);
1180 * iunique - get a unique inode number
1182 * @max_reserved: highest reserved inode number
1184 * Obtain an inode number that is unique on the system for a given
1185 * superblock. This is used by file systems that have no natural
1186 * permanent inode numbering system. An inode number is returned that
1187 * is higher than the reserved limit but unique.
1190 * With a large number of inodes live on the file system this function
1191 * currently becomes quite slow.
1193 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1196 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1197 * error if st_ino won't fit in target struct field. Use 32bit counter
1198 * here to attempt to avoid that.
1200 static DEFINE_SPINLOCK(iunique_lock);
1201 static unsigned int counter;
1204 spin_lock(&iunique_lock);
1206 if (counter <= max_reserved)
1207 counter = max_reserved + 1;
1209 } while (!test_inode_iunique(sb, res));
1210 spin_unlock(&iunique_lock);
1214 EXPORT_SYMBOL(iunique);
1216 struct inode *igrab(struct inode *inode)
1218 spin_lock(&inode->i_lock);
1219 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1221 spin_unlock(&inode->i_lock);
1223 spin_unlock(&inode->i_lock);
1225 * Handle the case where s_op->clear_inode is not been
1226 * called yet, and somebody is calling igrab
1227 * while the inode is getting freed.
1233 EXPORT_SYMBOL(igrab);
1236 * ilookup5_nowait - search for an inode in the inode cache
1237 * @sb: super block of file system to search
1238 * @hashval: hash value (usually inode number) to search for
1239 * @test: callback used for comparisons between inodes
1240 * @data: opaque data pointer to pass to @test
1242 * Search for the inode specified by @hashval and @data in the inode cache.
1243 * If the inode is in the cache, the inode is returned with an incremented
1246 * Note: I_NEW is not waited upon so you have to be very careful what you do
1247 * with the returned inode. You probably should be using ilookup5() instead.
1249 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1251 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1252 int (*test)(struct inode *, void *), void *data)
1254 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1255 struct inode *inode;
1257 spin_lock(&inode_hash_lock);
1258 inode = find_inode(sb, head, test, data);
1259 spin_unlock(&inode_hash_lock);
1263 EXPORT_SYMBOL(ilookup5_nowait);
1266 * ilookup5 - search for an inode in the inode cache
1267 * @sb: super block of file system to search
1268 * @hashval: hash value (usually inode number) to search for
1269 * @test: callback used for comparisons between inodes
1270 * @data: opaque data pointer to pass to @test
1272 * Search for the inode specified by @hashval and @data in the inode cache,
1273 * and if the inode is in the cache, return the inode with an incremented
1274 * reference count. Waits on I_NEW before returning the inode.
1275 * returned with an incremented reference count.
1277 * This is a generalized version of ilookup() for file systems where the
1278 * inode number is not sufficient for unique identification of an inode.
1280 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1282 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1283 int (*test)(struct inode *, void *), void *data)
1285 struct inode *inode;
1287 inode = ilookup5_nowait(sb, hashval, test, data);
1289 wait_on_inode(inode);
1290 if (unlikely(inode_unhashed(inode))) {
1297 EXPORT_SYMBOL(ilookup5);
1300 * ilookup - search for an inode in the inode cache
1301 * @sb: super block of file system to search
1302 * @ino: inode number to search for
1304 * Search for the inode @ino in the inode cache, and if the inode is in the
1305 * cache, the inode is returned with an incremented reference count.
1307 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1309 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1310 struct inode *inode;
1312 spin_lock(&inode_hash_lock);
1313 inode = find_inode_fast(sb, head, ino);
1314 spin_unlock(&inode_hash_lock);
1317 wait_on_inode(inode);
1318 if (unlikely(inode_unhashed(inode))) {
1325 EXPORT_SYMBOL(ilookup);
1328 * find_inode_nowait - find 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 * @match: callback used for comparisons between inodes
1332 * @data: opaque data pointer to pass to @match
1334 * Search for the inode specified by @hashval and @data in the inode
1335 * cache, where the helper function @match will return 0 if the inode
1336 * does not match, 1 if the inode does match, and -1 if the search
1337 * should be stopped. The @match function must be responsible for
1338 * taking the i_lock spin_lock and checking i_state for an inode being
1339 * freed or being initialized, and incrementing the reference count
1340 * before returning 1. It also must not sleep, since it is called with
1341 * the inode_hash_lock spinlock held.
1343 * This is a even more generalized version of ilookup5() when the
1344 * function must never block --- find_inode() can block in
1345 * __wait_on_freeing_inode() --- or when the caller can not increment
1346 * the reference count because the resulting iput() might cause an
1347 * inode eviction. The tradeoff is that the @match funtion must be
1348 * very carefully implemented.
1350 struct inode *find_inode_nowait(struct super_block *sb,
1351 unsigned long hashval,
1352 int (*match)(struct inode *, unsigned long,
1356 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1357 struct inode *inode, *ret_inode = NULL;
1360 spin_lock(&inode_hash_lock);
1361 hlist_for_each_entry(inode, head, i_hash) {
1362 if (inode->i_sb != sb)
1364 mval = match(inode, hashval, data);
1372 spin_unlock(&inode_hash_lock);
1375 EXPORT_SYMBOL(find_inode_nowait);
1377 int insert_inode_locked(struct inode *inode)
1379 struct super_block *sb = inode->i_sb;
1380 ino_t ino = inode->i_ino;
1381 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1384 struct inode *old = NULL;
1385 spin_lock(&inode_hash_lock);
1386 hlist_for_each_entry(old, head, i_hash) {
1387 if (old->i_ino != ino)
1389 if (old->i_sb != sb)
1391 spin_lock(&old->i_lock);
1392 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1393 spin_unlock(&old->i_lock);
1399 spin_lock(&inode->i_lock);
1400 inode->i_state |= I_NEW;
1401 hlist_add_head(&inode->i_hash, head);
1402 spin_unlock(&inode->i_lock);
1403 spin_unlock(&inode_hash_lock);
1407 spin_unlock(&old->i_lock);
1408 spin_unlock(&inode_hash_lock);
1410 if (unlikely(!inode_unhashed(old))) {
1417 EXPORT_SYMBOL(insert_inode_locked);
1419 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1420 int (*test)(struct inode *, void *), void *data)
1422 struct super_block *sb = inode->i_sb;
1423 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1426 struct inode *old = NULL;
1428 spin_lock(&inode_hash_lock);
1429 hlist_for_each_entry(old, head, i_hash) {
1430 if (old->i_sb != sb)
1432 if (!test(old, data))
1434 spin_lock(&old->i_lock);
1435 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1436 spin_unlock(&old->i_lock);
1442 spin_lock(&inode->i_lock);
1443 inode->i_state |= I_NEW;
1444 hlist_add_head(&inode->i_hash, head);
1445 spin_unlock(&inode->i_lock);
1446 spin_unlock(&inode_hash_lock);
1450 spin_unlock(&old->i_lock);
1451 spin_unlock(&inode_hash_lock);
1453 if (unlikely(!inode_unhashed(old))) {
1460 EXPORT_SYMBOL(insert_inode_locked4);
1463 int generic_delete_inode(struct inode *inode)
1467 EXPORT_SYMBOL(generic_delete_inode);
1470 * Called when we're dropping the last reference
1473 * Call the FS "drop_inode()" function, defaulting to
1474 * the legacy UNIX filesystem behaviour. If it tells
1475 * us to evict inode, do so. Otherwise, retain inode
1476 * in cache if fs is alive, sync and evict if fs is
1479 static void iput_final(struct inode *inode)
1481 struct super_block *sb = inode->i_sb;
1482 const struct super_operations *op = inode->i_sb->s_op;
1485 WARN_ON(inode->i_state & I_NEW);
1488 drop = op->drop_inode(inode);
1490 drop = generic_drop_inode(inode);
1492 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1493 inode->i_state |= I_REFERENCED;
1494 inode_add_lru(inode);
1495 spin_unlock(&inode->i_lock);
1500 inode->i_state |= I_WILL_FREE;
1501 spin_unlock(&inode->i_lock);
1502 write_inode_now(inode, 1);
1503 spin_lock(&inode->i_lock);
1504 WARN_ON(inode->i_state & I_NEW);
1505 inode->i_state &= ~I_WILL_FREE;
1508 inode->i_state |= I_FREEING;
1509 if (!list_empty(&inode->i_lru))
1510 inode_lru_list_del(inode);
1511 spin_unlock(&inode->i_lock);
1517 * iput - put an inode
1518 * @inode: inode to put
1520 * Puts an inode, dropping its usage count. If the inode use count hits
1521 * zero, the inode is then freed and may also be destroyed.
1523 * Consequently, iput() can sleep.
1525 void iput(struct inode *inode)
1529 BUG_ON(inode->i_state & I_CLEAR);
1531 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1532 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1533 atomic_inc(&inode->i_count);
1534 inode->i_state &= ~I_DIRTY_TIME;
1535 spin_unlock(&inode->i_lock);
1536 trace_writeback_lazytime_iput(inode);
1537 mark_inode_dirty_sync(inode);
1543 EXPORT_SYMBOL(iput);
1546 * bmap - find a block number in a file
1547 * @inode: inode of file
1548 * @block: block to find
1550 * Returns the block number on the device holding the inode that
1551 * is the disk block number for the block of the file requested.
1552 * That is, asked for block 4 of inode 1 the function will return the
1553 * disk block relative to the disk start that holds that block of the
1556 sector_t bmap(struct inode *inode, sector_t block)
1559 if (inode->i_mapping->a_ops->bmap)
1560 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1563 EXPORT_SYMBOL(bmap);
1566 * Update times in overlayed inode from underlying real inode
1568 static void update_ovl_inode_times(struct dentry *dentry, struct inode *inode,
1572 struct inode *realinode = d_real_inode(dentry);
1574 if (unlikely(inode != realinode) &&
1575 (!timespec_equal(&inode->i_mtime, &realinode->i_mtime) ||
1576 !timespec_equal(&inode->i_ctime, &realinode->i_ctime))) {
1577 inode->i_mtime = realinode->i_mtime;
1578 inode->i_ctime = realinode->i_ctime;
1584 * With relative atime, only update atime if the previous atime is
1585 * earlier than either the ctime or mtime or if at least a day has
1586 * passed since the last atime update.
1588 static int relatime_need_update(const struct path *path, struct inode *inode,
1589 struct timespec now, bool rcu)
1592 if (!(path->mnt->mnt_flags & MNT_RELATIME))
1595 update_ovl_inode_times(path->dentry, inode, rcu);
1597 * Is mtime younger than atime? If yes, update atime:
1599 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1602 * Is ctime younger than atime? If yes, update atime:
1604 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1608 * Is the previous atime value older than a day? If yes,
1611 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1614 * Good, we can skip the atime update:
1619 int generic_update_time(struct inode *inode, struct timespec *time, int flags)
1621 int iflags = I_DIRTY_TIME;
1623 if (flags & S_ATIME)
1624 inode->i_atime = *time;
1625 if (flags & S_VERSION)
1626 inode_inc_iversion(inode);
1627 if (flags & S_CTIME)
1628 inode->i_ctime = *time;
1629 if (flags & S_MTIME)
1630 inode->i_mtime = *time;
1632 if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION))
1633 iflags |= I_DIRTY_SYNC;
1634 __mark_inode_dirty(inode, iflags);
1637 EXPORT_SYMBOL(generic_update_time);
1640 * This does the actual work of updating an inodes time or version. Must have
1641 * had called mnt_want_write() before calling this.
1643 static int update_time(struct inode *inode, struct timespec *time, int flags)
1645 int (*update_time)(struct inode *, struct timespec *, int);
1647 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1648 generic_update_time;
1650 return update_time(inode, time, flags);
1654 * touch_atime - update the access time
1655 * @path: the &struct path to update
1656 * @inode: inode to update
1658 * Update the accessed time on an inode and mark it for writeback.
1659 * This function automatically handles read only file systems and media,
1660 * as well as the "noatime" flag and inode specific "noatime" markers.
1662 bool __atime_needs_update(const struct path *path, struct inode *inode,
1665 struct vfsmount *mnt = path->mnt;
1666 struct timespec now;
1668 if (inode->i_flags & S_NOATIME)
1671 /* Atime updates will likely cause i_uid and i_gid to be written
1672 * back improprely if their true value is unknown to the vfs.
1674 if (HAS_UNMAPPED_ID(inode))
1677 if (IS_NOATIME(inode))
1679 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1682 if (mnt->mnt_flags & MNT_NOATIME)
1684 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1687 now = current_fs_time(inode->i_sb);
1689 if (!relatime_need_update(path, inode, now, rcu))
1692 if (timespec_equal(&inode->i_atime, &now))
1698 void touch_atime(const struct path *path)
1700 struct vfsmount *mnt = path->mnt;
1701 struct inode *inode = d_inode(path->dentry);
1702 struct timespec now;
1704 if (!__atime_needs_update(path, inode, false))
1707 if (!sb_start_write_trylock(inode->i_sb))
1710 if (__mnt_want_write(mnt) != 0)
1713 * File systems can error out when updating inodes if they need to
1714 * allocate new space to modify an inode (such is the case for
1715 * Btrfs), but since we touch atime while walking down the path we
1716 * really don't care if we failed to update the atime of the file,
1717 * so just ignore the return value.
1718 * We may also fail on filesystems that have the ability to make parts
1719 * of the fs read only, e.g. subvolumes in Btrfs.
1721 now = current_fs_time(inode->i_sb);
1722 update_time(inode, &now, S_ATIME);
1723 __mnt_drop_write(mnt);
1725 sb_end_write(inode->i_sb);
1727 EXPORT_SYMBOL(touch_atime);
1730 * The logic we want is
1732 * if suid or (sgid and xgrp)
1735 int should_remove_suid(struct dentry *dentry)
1737 umode_t mode = d_inode(dentry)->i_mode;
1740 /* suid always must be killed */
1741 if (unlikely(mode & S_ISUID))
1742 kill = ATTR_KILL_SUID;
1745 * sgid without any exec bits is just a mandatory locking mark; leave
1746 * it alone. If some exec bits are set, it's a real sgid; kill it.
1748 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1749 kill |= ATTR_KILL_SGID;
1751 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1756 EXPORT_SYMBOL(should_remove_suid);
1759 * Return mask of changes for notify_change() that need to be done as a
1760 * response to write or truncate. Return 0 if nothing has to be changed.
1761 * Negative value on error (change should be denied).
1763 int dentry_needs_remove_privs(struct dentry *dentry)
1765 struct inode *inode = d_inode(dentry);
1769 if (IS_NOSEC(inode))
1772 mask = should_remove_suid(dentry);
1773 ret = security_inode_need_killpriv(dentry);
1777 mask |= ATTR_KILL_PRIV;
1781 static int __remove_privs(struct dentry *dentry, int kill)
1783 struct iattr newattrs;
1785 newattrs.ia_valid = ATTR_FORCE | kill;
1787 * Note we call this on write, so notify_change will not
1788 * encounter any conflicting delegations:
1790 return notify_change(dentry, &newattrs, NULL);
1794 * Remove special file priviledges (suid, capabilities) when file is written
1797 int file_remove_privs(struct file *file)
1799 struct dentry *dentry = file_dentry(file);
1800 struct inode *inode = file_inode(file);
1804 /* Fast path for nothing security related */
1805 if (IS_NOSEC(inode))
1808 kill = dentry_needs_remove_privs(dentry);
1812 error = __remove_privs(dentry, kill);
1814 inode_has_no_xattr(inode);
1818 EXPORT_SYMBOL(file_remove_privs);
1821 * file_update_time - update mtime and ctime time
1822 * @file: file accessed
1824 * Update the mtime and ctime members of an inode and mark the inode
1825 * for writeback. Note that this function is meant exclusively for
1826 * usage in the file write path of filesystems, and filesystems may
1827 * choose to explicitly ignore update via this function with the
1828 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1829 * timestamps are handled by the server. This can return an error for
1830 * file systems who need to allocate space in order to update an inode.
1833 int file_update_time(struct file *file)
1835 struct inode *inode = file_inode(file);
1836 struct timespec now;
1840 /* First try to exhaust all avenues to not sync */
1841 if (IS_NOCMTIME(inode))
1844 now = current_fs_time(inode->i_sb);
1845 if (!timespec_equal(&inode->i_mtime, &now))
1848 if (!timespec_equal(&inode->i_ctime, &now))
1851 if (IS_I_VERSION(inode))
1852 sync_it |= S_VERSION;
1857 /* Finally allowed to write? Takes lock. */
1858 if (__mnt_want_write_file(file))
1861 ret = update_time(inode, &now, sync_it);
1862 __mnt_drop_write_file(file);
1866 EXPORT_SYMBOL(file_update_time);
1868 int inode_needs_sync(struct inode *inode)
1872 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1876 EXPORT_SYMBOL(inode_needs_sync);
1879 * If we try to find an inode in the inode hash while it is being
1880 * deleted, we have to wait until the filesystem completes its
1881 * deletion before reporting that it isn't found. This function waits
1882 * until the deletion _might_ have completed. Callers are responsible
1883 * to recheck inode state.
1885 * It doesn't matter if I_NEW is not set initially, a call to
1886 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1889 static void __wait_on_freeing_inode(struct inode *inode)
1891 wait_queue_head_t *wq;
1892 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1893 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1894 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1895 spin_unlock(&inode->i_lock);
1896 spin_unlock(&inode_hash_lock);
1898 finish_wait(wq, &wait.wait);
1899 spin_lock(&inode_hash_lock);
1902 static __initdata unsigned long ihash_entries;
1903 static int __init set_ihash_entries(char *str)
1907 ihash_entries = simple_strtoul(str, &str, 0);
1910 __setup("ihash_entries=", set_ihash_entries);
1913 * Initialize the waitqueues and inode hash table.
1915 void __init inode_init_early(void)
1919 /* If hashes are distributed across NUMA nodes, defer
1920 * hash allocation until vmalloc space is available.
1926 alloc_large_system_hash("Inode-cache",
1927 sizeof(struct hlist_head),
1936 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1937 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1940 void __init inode_init(void)
1944 /* inode slab cache */
1945 inode_cachep = kmem_cache_create("inode_cache",
1946 sizeof(struct inode),
1948 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1949 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1952 /* Hash may have been set up in inode_init_early */
1957 alloc_large_system_hash("Inode-cache",
1958 sizeof(struct hlist_head),
1967 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1968 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1971 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1973 inode->i_mode = mode;
1974 if (S_ISCHR(mode)) {
1975 inode->i_fop = &def_chr_fops;
1976 inode->i_rdev = rdev;
1977 } else if (S_ISBLK(mode)) {
1978 inode->i_fop = &def_blk_fops;
1979 inode->i_rdev = rdev;
1980 } else if (S_ISFIFO(mode))
1981 inode->i_fop = &pipefifo_fops;
1982 else if (S_ISSOCK(mode))
1983 ; /* leave it no_open_fops */
1985 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1986 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1989 EXPORT_SYMBOL(init_special_inode);
1992 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1994 * @dir: Directory inode
1995 * @mode: mode of the new inode
1997 void inode_init_owner(struct inode *inode, const struct inode *dir,
2000 inode->i_uid = current_fsuid();
2001 if (dir && dir->i_mode & S_ISGID) {
2002 inode->i_gid = dir->i_gid;
2006 inode->i_gid = current_fsgid();
2007 inode->i_mode = mode;
2009 EXPORT_SYMBOL(inode_init_owner);
2012 * inode_owner_or_capable - check current task permissions to inode
2013 * @inode: inode being checked
2015 * Return true if current either has CAP_FOWNER in a namespace with the
2016 * inode owner uid mapped, or owns the file.
2018 bool inode_owner_or_capable(const struct inode *inode)
2020 struct user_namespace *ns;
2022 if (uid_eq(current_fsuid(), inode->i_uid))
2025 ns = current_user_ns();
2026 if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid))
2030 EXPORT_SYMBOL(inode_owner_or_capable);
2033 * Direct i/o helper functions
2035 static void __inode_dio_wait(struct inode *inode)
2037 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2038 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2041 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
2042 if (atomic_read(&inode->i_dio_count))
2044 } while (atomic_read(&inode->i_dio_count));
2045 finish_wait(wq, &q.wait);
2049 * inode_dio_wait - wait for outstanding DIO requests to finish
2050 * @inode: inode to wait for
2052 * Waits for all pending direct I/O requests to finish so that we can
2053 * proceed with a truncate or equivalent operation.
2055 * Must be called under a lock that serializes taking new references
2056 * to i_dio_count, usually by inode->i_mutex.
2058 void inode_dio_wait(struct inode *inode)
2060 if (atomic_read(&inode->i_dio_count))
2061 __inode_dio_wait(inode);
2063 EXPORT_SYMBOL(inode_dio_wait);
2066 * inode_set_flags - atomically set some inode flags
2068 * Note: the caller should be holding i_mutex, or else be sure that
2069 * they have exclusive access to the inode structure (i.e., while the
2070 * inode is being instantiated). The reason for the cmpxchg() loop
2071 * --- which wouldn't be necessary if all code paths which modify
2072 * i_flags actually followed this rule, is that there is at least one
2073 * code path which doesn't today so we use cmpxchg() out of an abundance
2076 * In the long run, i_mutex is overkill, and we should probably look
2077 * at using the i_lock spinlock to protect i_flags, and then make sure
2078 * it is so documented in include/linux/fs.h and that all code follows
2079 * the locking convention!!
2081 void inode_set_flags(struct inode *inode, unsigned int flags,
2084 unsigned int old_flags, new_flags;
2086 WARN_ON_ONCE(flags & ~mask);
2088 old_flags = ACCESS_ONCE(inode->i_flags);
2089 new_flags = (old_flags & ~mask) | flags;
2090 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2091 new_flags) != old_flags));
2093 EXPORT_SYMBOL(inode_set_flags);
2095 void inode_nohighmem(struct inode *inode)
2097 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2099 EXPORT_SYMBOL(inode_nohighmem);