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);
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_lru);
369 address_space_init_once(&inode->i_data);
370 i_size_ordered_init(inode);
371 #ifdef CONFIG_FSNOTIFY
372 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
375 EXPORT_SYMBOL(inode_init_once);
377 static void init_once(void *foo)
379 struct inode *inode = (struct inode *) foo;
381 inode_init_once(inode);
385 * inode->i_lock must be held
387 void __iget(struct inode *inode)
389 atomic_inc(&inode->i_count);
393 * get additional reference to inode; caller must already hold one.
395 void ihold(struct inode *inode)
397 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
399 EXPORT_SYMBOL(ihold);
401 static void inode_lru_list_add(struct inode *inode)
403 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
404 this_cpu_inc(nr_unused);
408 * Add inode to LRU if needed (inode is unused and clean).
410 * Needs inode->i_lock held.
412 void inode_add_lru(struct inode *inode)
414 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
415 I_FREEING | I_WILL_FREE)) &&
416 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
417 inode_lru_list_add(inode);
421 static void inode_lru_list_del(struct inode *inode)
424 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
425 this_cpu_dec(nr_unused);
429 * inode_sb_list_add - add inode to the superblock list of inodes
430 * @inode: inode to add
432 void inode_sb_list_add(struct inode *inode)
434 spin_lock(&inode->i_sb->s_inode_list_lock);
435 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
436 spin_unlock(&inode->i_sb->s_inode_list_lock);
438 EXPORT_SYMBOL_GPL(inode_sb_list_add);
440 static inline void inode_sb_list_del(struct inode *inode)
442 if (!list_empty(&inode->i_sb_list)) {
443 spin_lock(&inode->i_sb->s_inode_list_lock);
444 list_del_init(&inode->i_sb_list);
445 spin_unlock(&inode->i_sb->s_inode_list_lock);
449 static unsigned long hash(struct super_block *sb, unsigned long hashval)
453 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
455 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
456 return tmp & i_hash_mask;
460 * __insert_inode_hash - hash an inode
461 * @inode: unhashed inode
462 * @hashval: unsigned long value used to locate this object in the
465 * Add an inode to the inode hash for this superblock.
467 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
469 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
471 spin_lock(&inode_hash_lock);
472 spin_lock(&inode->i_lock);
473 hlist_add_head(&inode->i_hash, b);
474 spin_unlock(&inode->i_lock);
475 spin_unlock(&inode_hash_lock);
477 EXPORT_SYMBOL(__insert_inode_hash);
480 * __remove_inode_hash - remove an inode from the hash
481 * @inode: inode to unhash
483 * Remove an inode from the superblock.
485 void __remove_inode_hash(struct inode *inode)
487 spin_lock(&inode_hash_lock);
488 spin_lock(&inode->i_lock);
489 hlist_del_init(&inode->i_hash);
490 spin_unlock(&inode->i_lock);
491 spin_unlock(&inode_hash_lock);
493 EXPORT_SYMBOL(__remove_inode_hash);
495 void clear_inode(struct inode *inode)
499 * We have to cycle tree_lock here because reclaim can be still in the
500 * process of removing the last page (in __delete_from_page_cache())
501 * and we must not free mapping under it.
503 spin_lock_irq(&inode->i_data.tree_lock);
504 BUG_ON(inode->i_data.nrpages);
505 BUG_ON(inode->i_data.nrexceptional);
506 spin_unlock_irq(&inode->i_data.tree_lock);
507 BUG_ON(!list_empty(&inode->i_data.private_list));
508 BUG_ON(!(inode->i_state & I_FREEING));
509 BUG_ON(inode->i_state & I_CLEAR);
510 /* don't need i_lock here, no concurrent mods to i_state */
511 inode->i_state = I_FREEING | I_CLEAR;
513 EXPORT_SYMBOL(clear_inode);
516 * Free the inode passed in, removing it from the lists it is still connected
517 * to. We remove any pages still attached to the inode and wait for any IO that
518 * is still in progress before finally destroying the inode.
520 * An inode must already be marked I_FREEING so that we avoid the inode being
521 * moved back onto lists if we race with other code that manipulates the lists
522 * (e.g. writeback_single_inode). The caller is responsible for setting this.
524 * An inode must already be removed from the LRU list before being evicted from
525 * the cache. This should occur atomically with setting the I_FREEING state
526 * flag, so no inodes here should ever be on the LRU when being evicted.
528 static void evict(struct inode *inode)
530 const struct super_operations *op = inode->i_sb->s_op;
532 BUG_ON(!(inode->i_state & I_FREEING));
533 BUG_ON(!list_empty(&inode->i_lru));
535 if (!list_empty(&inode->i_io_list))
536 inode_io_list_del(inode);
538 inode_sb_list_del(inode);
541 * Wait for flusher thread to be done with the inode so that filesystem
542 * does not start destroying it while writeback is still running. Since
543 * the inode has I_FREEING set, flusher thread won't start new work on
544 * the inode. We just have to wait for running writeback to finish.
546 inode_wait_for_writeback(inode);
548 if (op->evict_inode) {
549 op->evict_inode(inode);
551 truncate_inode_pages_final(&inode->i_data);
554 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
556 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
559 remove_inode_hash(inode);
561 spin_lock(&inode->i_lock);
562 wake_up_bit(&inode->i_state, __I_NEW);
563 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
564 spin_unlock(&inode->i_lock);
566 destroy_inode(inode);
570 * dispose_list - dispose of the contents of a local list
571 * @head: the head of the list to free
573 * Dispose-list gets a local list with local inodes in it, so it doesn't
574 * need to worry about list corruption and SMP locks.
576 static void dispose_list(struct list_head *head)
578 while (!list_empty(head)) {
581 inode = list_first_entry(head, struct inode, i_lru);
582 list_del_init(&inode->i_lru);
590 * evict_inodes - evict all evictable inodes for a superblock
591 * @sb: superblock to operate on
593 * Make sure that no inodes with zero refcount are retained. This is
594 * called by superblock shutdown after having MS_ACTIVE flag removed,
595 * so any inode reaching zero refcount during or after that call will
596 * be immediately evicted.
598 void evict_inodes(struct super_block *sb)
600 struct inode *inode, *next;
604 spin_lock(&sb->s_inode_list_lock);
605 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
606 if (atomic_read(&inode->i_count))
609 spin_lock(&inode->i_lock);
610 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
611 spin_unlock(&inode->i_lock);
615 inode->i_state |= I_FREEING;
616 inode_lru_list_del(inode);
617 spin_unlock(&inode->i_lock);
618 list_add(&inode->i_lru, &dispose);
621 * We can have a ton of inodes to evict at unmount time given
622 * enough memory, check to see if we need to go to sleep for a
623 * bit so we don't livelock.
625 if (need_resched()) {
626 spin_unlock(&sb->s_inode_list_lock);
628 dispose_list(&dispose);
632 spin_unlock(&sb->s_inode_list_lock);
634 dispose_list(&dispose);
638 * invalidate_inodes - attempt to free all inodes on a superblock
639 * @sb: superblock to operate on
640 * @kill_dirty: flag to guide handling of dirty inodes
642 * Attempts to free all inodes for a given superblock. If there were any
643 * busy inodes return a non-zero value, else zero.
644 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
647 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
650 struct inode *inode, *next;
653 spin_lock(&sb->s_inode_list_lock);
654 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
655 spin_lock(&inode->i_lock);
656 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
657 spin_unlock(&inode->i_lock);
660 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
661 spin_unlock(&inode->i_lock);
665 if (atomic_read(&inode->i_count)) {
666 spin_unlock(&inode->i_lock);
671 inode->i_state |= I_FREEING;
672 inode_lru_list_del(inode);
673 spin_unlock(&inode->i_lock);
674 list_add(&inode->i_lru, &dispose);
676 spin_unlock(&sb->s_inode_list_lock);
678 dispose_list(&dispose);
684 * Isolate the inode from the LRU in preparation for freeing it.
686 * Any inodes which are pinned purely because of attached pagecache have their
687 * pagecache removed. If the inode has metadata buffers attached to
688 * mapping->private_list then try to remove them.
690 * If the inode has the I_REFERENCED flag set, then it means that it has been
691 * used recently - the flag is set in iput_final(). When we encounter such an
692 * inode, clear the flag and move it to the back of the LRU so it gets another
693 * pass through the LRU before it gets reclaimed. This is necessary because of
694 * the fact we are doing lazy LRU updates to minimise lock contention so the
695 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
696 * with this flag set because they are the inodes that are out of order.
698 static enum lru_status inode_lru_isolate(struct list_head *item,
699 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
701 struct list_head *freeable = arg;
702 struct inode *inode = container_of(item, struct inode, i_lru);
705 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
706 * If we fail to get the lock, just skip it.
708 if (!spin_trylock(&inode->i_lock))
712 * Referenced or dirty inodes are still in use. Give them another pass
713 * through the LRU as we canot reclaim them now.
715 if (atomic_read(&inode->i_count) ||
716 (inode->i_state & ~I_REFERENCED)) {
717 list_lru_isolate(lru, &inode->i_lru);
718 spin_unlock(&inode->i_lock);
719 this_cpu_dec(nr_unused);
723 /* recently referenced inodes get one more pass */
724 if (inode->i_state & I_REFERENCED) {
725 inode->i_state &= ~I_REFERENCED;
726 spin_unlock(&inode->i_lock);
730 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
732 spin_unlock(&inode->i_lock);
733 spin_unlock(lru_lock);
734 if (remove_inode_buffers(inode)) {
736 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
737 if (current_is_kswapd())
738 __count_vm_events(KSWAPD_INODESTEAL, reap);
740 __count_vm_events(PGINODESTEAL, reap);
741 if (current->reclaim_state)
742 current->reclaim_state->reclaimed_slab += reap;
749 WARN_ON(inode->i_state & I_NEW);
750 inode->i_state |= I_FREEING;
751 list_lru_isolate_move(lru, &inode->i_lru, freeable);
752 spin_unlock(&inode->i_lock);
754 this_cpu_dec(nr_unused);
759 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
760 * This is called from the superblock shrinker function with a number of inodes
761 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
762 * then are freed outside inode_lock by dispose_list().
764 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
769 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
770 inode_lru_isolate, &freeable);
771 dispose_list(&freeable);
775 static void __wait_on_freeing_inode(struct inode *inode);
777 * Called with the inode lock held.
779 static struct inode *find_inode(struct super_block *sb,
780 struct hlist_head *head,
781 int (*test)(struct inode *, void *),
784 struct inode *inode = NULL;
787 hlist_for_each_entry(inode, head, i_hash) {
788 if (inode->i_sb != sb)
790 if (!test(inode, data))
792 spin_lock(&inode->i_lock);
793 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
794 __wait_on_freeing_inode(inode);
798 spin_unlock(&inode->i_lock);
805 * find_inode_fast is the fast path version of find_inode, see the comment at
806 * iget_locked for details.
808 static struct inode *find_inode_fast(struct super_block *sb,
809 struct hlist_head *head, unsigned long ino)
811 struct inode *inode = NULL;
814 hlist_for_each_entry(inode, head, i_hash) {
815 if (inode->i_ino != ino)
817 if (inode->i_sb != sb)
819 spin_lock(&inode->i_lock);
820 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
821 __wait_on_freeing_inode(inode);
825 spin_unlock(&inode->i_lock);
832 * Each cpu owns a range of LAST_INO_BATCH numbers.
833 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
834 * to renew the exhausted range.
836 * This does not significantly increase overflow rate because every CPU can
837 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
838 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
839 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
840 * overflow rate by 2x, which does not seem too significant.
842 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
843 * error if st_ino won't fit in target struct field. Use 32bit counter
844 * here to attempt to avoid that.
846 #define LAST_INO_BATCH 1024
847 static DEFINE_PER_CPU(unsigned int, last_ino);
849 unsigned int get_next_ino(void)
851 unsigned int *p = &get_cpu_var(last_ino);
852 unsigned int res = *p;
855 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
856 static atomic_t shared_last_ino;
857 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
859 res = next - LAST_INO_BATCH;
864 /* get_next_ino should not provide a 0 inode number */
868 put_cpu_var(last_ino);
871 EXPORT_SYMBOL(get_next_ino);
874 * new_inode_pseudo - obtain an inode
877 * Allocates a new inode for given superblock.
878 * Inode wont be chained in superblock s_inodes list
880 * - fs can't be unmount
881 * - quotas, fsnotify, writeback can't work
883 struct inode *new_inode_pseudo(struct super_block *sb)
885 struct inode *inode = alloc_inode(sb);
888 spin_lock(&inode->i_lock);
890 spin_unlock(&inode->i_lock);
891 INIT_LIST_HEAD(&inode->i_sb_list);
897 * new_inode - obtain an inode
900 * Allocates a new inode for given superblock. The default gfp_mask
901 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
902 * If HIGHMEM pages are unsuitable or it is known that pages allocated
903 * for the page cache are not reclaimable or migratable,
904 * mapping_set_gfp_mask() must be called with suitable flags on the
905 * newly created inode's mapping
908 struct inode *new_inode(struct super_block *sb)
912 spin_lock_prefetch(&sb->s_inode_list_lock);
914 inode = new_inode_pseudo(sb);
916 inode_sb_list_add(inode);
919 EXPORT_SYMBOL(new_inode);
921 #ifdef CONFIG_DEBUG_LOCK_ALLOC
922 void lockdep_annotate_inode_mutex_key(struct inode *inode)
924 if (S_ISDIR(inode->i_mode)) {
925 struct file_system_type *type = inode->i_sb->s_type;
927 /* Set new key only if filesystem hasn't already changed it */
928 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
930 * ensure nobody is actually holding i_mutex
932 // mutex_destroy(&inode->i_mutex);
933 init_rwsem(&inode->i_rwsem);
934 lockdep_set_class(&inode->i_rwsem,
935 &type->i_mutex_dir_key);
939 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
943 * unlock_new_inode - clear the I_NEW state and wake up any waiters
944 * @inode: new inode to unlock
946 * Called when the inode is fully initialised to clear the new state of the
947 * inode and wake up anyone waiting for the inode to finish initialisation.
949 void unlock_new_inode(struct inode *inode)
951 lockdep_annotate_inode_mutex_key(inode);
952 spin_lock(&inode->i_lock);
953 WARN_ON(!(inode->i_state & I_NEW));
954 inode->i_state &= ~I_NEW;
956 wake_up_bit(&inode->i_state, __I_NEW);
957 spin_unlock(&inode->i_lock);
959 EXPORT_SYMBOL(unlock_new_inode);
962 * lock_two_nondirectories - take two i_mutexes on non-directory objects
964 * Lock any non-NULL argument that is not a directory.
965 * Zero, one or two objects may be locked by this function.
967 * @inode1: first inode to lock
968 * @inode2: second inode to lock
970 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
973 swap(inode1, inode2);
975 if (inode1 && !S_ISDIR(inode1->i_mode))
977 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
978 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
980 EXPORT_SYMBOL(lock_two_nondirectories);
983 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
984 * @inode1: first inode to unlock
985 * @inode2: second inode to unlock
987 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
989 if (inode1 && !S_ISDIR(inode1->i_mode))
990 inode_unlock(inode1);
991 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
992 inode_unlock(inode2);
994 EXPORT_SYMBOL(unlock_two_nondirectories);
997 * iget5_locked - obtain an inode from a mounted file system
998 * @sb: super block of file system
999 * @hashval: hash value (usually inode number) to get
1000 * @test: callback used for comparisons between inodes
1001 * @set: callback used to initialize a new struct inode
1002 * @data: opaque data pointer to pass to @test and @set
1004 * Search for the inode specified by @hashval and @data in the inode cache,
1005 * and if present it is return it with an increased reference count. This is
1006 * a generalized version of iget_locked() for file systems where the inode
1007 * number is not sufficient for unique identification of an inode.
1009 * If the inode is not in cache, allocate a new inode and return it locked,
1010 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1011 * before unlocking it via unlock_new_inode().
1013 * Note both @test and @set are called with the inode_hash_lock held, so can't
1016 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1017 int (*test)(struct inode *, void *),
1018 int (*set)(struct inode *, void *), void *data)
1020 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1021 struct inode *inode;
1023 spin_lock(&inode_hash_lock);
1024 inode = find_inode(sb, head, test, data);
1025 spin_unlock(&inode_hash_lock);
1028 wait_on_inode(inode);
1029 if (unlikely(inode_unhashed(inode))) {
1036 inode = alloc_inode(sb);
1040 spin_lock(&inode_hash_lock);
1041 /* We released the lock, so.. */
1042 old = find_inode(sb, head, test, data);
1044 if (set(inode, data))
1047 spin_lock(&inode->i_lock);
1048 inode->i_state = I_NEW;
1049 hlist_add_head(&inode->i_hash, head);
1050 spin_unlock(&inode->i_lock);
1051 inode_sb_list_add(inode);
1052 spin_unlock(&inode_hash_lock);
1054 /* Return the locked inode with I_NEW set, the
1055 * caller is responsible for filling in the contents
1061 * Uhhuh, somebody else created the same inode under
1062 * us. Use the old inode instead of the one we just
1065 spin_unlock(&inode_hash_lock);
1066 destroy_inode(inode);
1068 wait_on_inode(inode);
1069 if (unlikely(inode_unhashed(inode))) {
1077 spin_unlock(&inode_hash_lock);
1078 destroy_inode(inode);
1081 EXPORT_SYMBOL(iget5_locked);
1084 * iget_locked - obtain an inode from a mounted file system
1085 * @sb: super block of file system
1086 * @ino: inode number to get
1088 * Search for the inode specified by @ino in the inode cache and if present
1089 * return it with an increased reference count. This is for file systems
1090 * where the inode number is sufficient for unique identification of an inode.
1092 * If the inode is not in cache, allocate a new inode and return it locked,
1093 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1094 * before unlocking it via unlock_new_inode().
1096 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1098 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1099 struct inode *inode;
1101 spin_lock(&inode_hash_lock);
1102 inode = find_inode_fast(sb, head, ino);
1103 spin_unlock(&inode_hash_lock);
1105 wait_on_inode(inode);
1106 if (unlikely(inode_unhashed(inode))) {
1113 inode = alloc_inode(sb);
1117 spin_lock(&inode_hash_lock);
1118 /* We released the lock, so.. */
1119 old = find_inode_fast(sb, head, ino);
1122 spin_lock(&inode->i_lock);
1123 inode->i_state = I_NEW;
1124 hlist_add_head(&inode->i_hash, head);
1125 spin_unlock(&inode->i_lock);
1126 inode_sb_list_add(inode);
1127 spin_unlock(&inode_hash_lock);
1129 /* Return the locked inode with I_NEW set, the
1130 * caller is responsible for filling in the contents
1136 * Uhhuh, somebody else created the same inode under
1137 * us. Use the old inode instead of the one we just
1140 spin_unlock(&inode_hash_lock);
1141 destroy_inode(inode);
1143 wait_on_inode(inode);
1144 if (unlikely(inode_unhashed(inode))) {
1151 EXPORT_SYMBOL(iget_locked);
1154 * search the inode cache for a matching inode number.
1155 * If we find one, then the inode number we are trying to
1156 * allocate is not unique and so we should not use it.
1158 * Returns 1 if the inode number is unique, 0 if it is not.
1160 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1162 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1163 struct inode *inode;
1165 spin_lock(&inode_hash_lock);
1166 hlist_for_each_entry(inode, b, i_hash) {
1167 if (inode->i_ino == ino && inode->i_sb == sb) {
1168 spin_unlock(&inode_hash_lock);
1172 spin_unlock(&inode_hash_lock);
1178 * iunique - get a unique inode number
1180 * @max_reserved: highest reserved inode number
1182 * Obtain an inode number that is unique on the system for a given
1183 * superblock. This is used by file systems that have no natural
1184 * permanent inode numbering system. An inode number is returned that
1185 * is higher than the reserved limit but unique.
1188 * With a large number of inodes live on the file system this function
1189 * currently becomes quite slow.
1191 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1194 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1195 * error if st_ino won't fit in target struct field. Use 32bit counter
1196 * here to attempt to avoid that.
1198 static DEFINE_SPINLOCK(iunique_lock);
1199 static unsigned int counter;
1202 spin_lock(&iunique_lock);
1204 if (counter <= max_reserved)
1205 counter = max_reserved + 1;
1207 } while (!test_inode_iunique(sb, res));
1208 spin_unlock(&iunique_lock);
1212 EXPORT_SYMBOL(iunique);
1214 struct inode *igrab(struct inode *inode)
1216 spin_lock(&inode->i_lock);
1217 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1219 spin_unlock(&inode->i_lock);
1221 spin_unlock(&inode->i_lock);
1223 * Handle the case where s_op->clear_inode is not been
1224 * called yet, and somebody is calling igrab
1225 * while the inode is getting freed.
1231 EXPORT_SYMBOL(igrab);
1234 * ilookup5_nowait - search for an inode in the inode cache
1235 * @sb: super block of file system to search
1236 * @hashval: hash value (usually inode number) to search for
1237 * @test: callback used for comparisons between inodes
1238 * @data: opaque data pointer to pass to @test
1240 * Search for the inode specified by @hashval and @data in the inode cache.
1241 * If the inode is in the cache, the inode is returned with an incremented
1244 * Note: I_NEW is not waited upon so you have to be very careful what you do
1245 * with the returned inode. You probably should be using ilookup5() instead.
1247 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1249 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1250 int (*test)(struct inode *, void *), void *data)
1252 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1253 struct inode *inode;
1255 spin_lock(&inode_hash_lock);
1256 inode = find_inode(sb, head, test, data);
1257 spin_unlock(&inode_hash_lock);
1261 EXPORT_SYMBOL(ilookup5_nowait);
1264 * ilookup5 - search for an inode in the inode cache
1265 * @sb: super block of file system to search
1266 * @hashval: hash value (usually inode number) to search for
1267 * @test: callback used for comparisons between inodes
1268 * @data: opaque data pointer to pass to @test
1270 * Search for the inode specified by @hashval and @data in the inode cache,
1271 * and if the inode is in the cache, return the inode with an incremented
1272 * reference count. Waits on I_NEW before returning the inode.
1273 * returned with an incremented reference count.
1275 * This is a generalized version of ilookup() for file systems where the
1276 * inode number is not sufficient for unique identification of an inode.
1278 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1280 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1281 int (*test)(struct inode *, void *), void *data)
1283 struct inode *inode;
1285 inode = ilookup5_nowait(sb, hashval, test, data);
1287 wait_on_inode(inode);
1288 if (unlikely(inode_unhashed(inode))) {
1295 EXPORT_SYMBOL(ilookup5);
1298 * ilookup - search for an inode in the inode cache
1299 * @sb: super block of file system to search
1300 * @ino: inode number to search for
1302 * Search for the inode @ino in the inode cache, and if the inode is in the
1303 * cache, the inode is returned with an incremented reference count.
1305 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1307 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1308 struct inode *inode;
1310 spin_lock(&inode_hash_lock);
1311 inode = find_inode_fast(sb, head, ino);
1312 spin_unlock(&inode_hash_lock);
1315 wait_on_inode(inode);
1316 if (unlikely(inode_unhashed(inode))) {
1323 EXPORT_SYMBOL(ilookup);
1326 * find_inode_nowait - find an inode in the inode cache
1327 * @sb: super block of file system to search
1328 * @hashval: hash value (usually inode number) to search for
1329 * @match: callback used for comparisons between inodes
1330 * @data: opaque data pointer to pass to @match
1332 * Search for the inode specified by @hashval and @data in the inode
1333 * cache, where the helper function @match will return 0 if the inode
1334 * does not match, 1 if the inode does match, and -1 if the search
1335 * should be stopped. The @match function must be responsible for
1336 * taking the i_lock spin_lock and checking i_state for an inode being
1337 * freed or being initialized, and incrementing the reference count
1338 * before returning 1. It also must not sleep, since it is called with
1339 * the inode_hash_lock spinlock held.
1341 * This is a even more generalized version of ilookup5() when the
1342 * function must never block --- find_inode() can block in
1343 * __wait_on_freeing_inode() --- or when the caller can not increment
1344 * the reference count because the resulting iput() might cause an
1345 * inode eviction. The tradeoff is that the @match funtion must be
1346 * very carefully implemented.
1348 struct inode *find_inode_nowait(struct super_block *sb,
1349 unsigned long hashval,
1350 int (*match)(struct inode *, unsigned long,
1354 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1355 struct inode *inode, *ret_inode = NULL;
1358 spin_lock(&inode_hash_lock);
1359 hlist_for_each_entry(inode, head, i_hash) {
1360 if (inode->i_sb != sb)
1362 mval = match(inode, hashval, data);
1370 spin_unlock(&inode_hash_lock);
1373 EXPORT_SYMBOL(find_inode_nowait);
1375 int insert_inode_locked(struct inode *inode)
1377 struct super_block *sb = inode->i_sb;
1378 ino_t ino = inode->i_ino;
1379 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1382 struct inode *old = NULL;
1383 spin_lock(&inode_hash_lock);
1384 hlist_for_each_entry(old, head, i_hash) {
1385 if (old->i_ino != ino)
1387 if (old->i_sb != sb)
1389 spin_lock(&old->i_lock);
1390 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1391 spin_unlock(&old->i_lock);
1397 spin_lock(&inode->i_lock);
1398 inode->i_state |= I_NEW;
1399 hlist_add_head(&inode->i_hash, head);
1400 spin_unlock(&inode->i_lock);
1401 spin_unlock(&inode_hash_lock);
1405 spin_unlock(&old->i_lock);
1406 spin_unlock(&inode_hash_lock);
1408 if (unlikely(!inode_unhashed(old))) {
1415 EXPORT_SYMBOL(insert_inode_locked);
1417 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1418 int (*test)(struct inode *, void *), void *data)
1420 struct super_block *sb = inode->i_sb;
1421 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1424 struct inode *old = NULL;
1426 spin_lock(&inode_hash_lock);
1427 hlist_for_each_entry(old, head, i_hash) {
1428 if (old->i_sb != sb)
1430 if (!test(old, data))
1432 spin_lock(&old->i_lock);
1433 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1434 spin_unlock(&old->i_lock);
1440 spin_lock(&inode->i_lock);
1441 inode->i_state |= I_NEW;
1442 hlist_add_head(&inode->i_hash, head);
1443 spin_unlock(&inode->i_lock);
1444 spin_unlock(&inode_hash_lock);
1448 spin_unlock(&old->i_lock);
1449 spin_unlock(&inode_hash_lock);
1451 if (unlikely(!inode_unhashed(old))) {
1458 EXPORT_SYMBOL(insert_inode_locked4);
1461 int generic_delete_inode(struct inode *inode)
1465 EXPORT_SYMBOL(generic_delete_inode);
1468 * Called when we're dropping the last reference
1471 * Call the FS "drop_inode()" function, defaulting to
1472 * the legacy UNIX filesystem behaviour. If it tells
1473 * us to evict inode, do so. Otherwise, retain inode
1474 * in cache if fs is alive, sync and evict if fs is
1477 static void iput_final(struct inode *inode)
1479 struct super_block *sb = inode->i_sb;
1480 const struct super_operations *op = inode->i_sb->s_op;
1483 WARN_ON(inode->i_state & I_NEW);
1486 drop = op->drop_inode(inode);
1488 drop = generic_drop_inode(inode);
1490 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1491 inode->i_state |= I_REFERENCED;
1492 inode_add_lru(inode);
1493 spin_unlock(&inode->i_lock);
1498 inode->i_state |= I_WILL_FREE;
1499 spin_unlock(&inode->i_lock);
1500 write_inode_now(inode, 1);
1501 spin_lock(&inode->i_lock);
1502 WARN_ON(inode->i_state & I_NEW);
1503 inode->i_state &= ~I_WILL_FREE;
1506 inode->i_state |= I_FREEING;
1507 if (!list_empty(&inode->i_lru))
1508 inode_lru_list_del(inode);
1509 spin_unlock(&inode->i_lock);
1515 * iput - put an inode
1516 * @inode: inode to put
1518 * Puts an inode, dropping its usage count. If the inode use count hits
1519 * zero, the inode is then freed and may also be destroyed.
1521 * Consequently, iput() can sleep.
1523 void iput(struct inode *inode)
1527 BUG_ON(inode->i_state & I_CLEAR);
1529 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1530 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1531 atomic_inc(&inode->i_count);
1532 inode->i_state &= ~I_DIRTY_TIME;
1533 spin_unlock(&inode->i_lock);
1534 trace_writeback_lazytime_iput(inode);
1535 mark_inode_dirty_sync(inode);
1541 EXPORT_SYMBOL(iput);
1544 * bmap - find a block number in a file
1545 * @inode: inode of file
1546 * @block: block to find
1548 * Returns the block number on the device holding the inode that
1549 * is the disk block number for the block of the file requested.
1550 * That is, asked for block 4 of inode 1 the function will return the
1551 * disk block relative to the disk start that holds that block of the
1554 sector_t bmap(struct inode *inode, sector_t block)
1557 if (inode->i_mapping->a_ops->bmap)
1558 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1561 EXPORT_SYMBOL(bmap);
1564 * With relative atime, only update atime if the previous atime is
1565 * earlier than either the ctime or mtime or if at least a day has
1566 * passed since the last atime update.
1568 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1569 struct timespec now)
1572 if (!(mnt->mnt_flags & MNT_RELATIME))
1575 * Is mtime younger than atime? If yes, update atime:
1577 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1580 * Is ctime younger than atime? If yes, update atime:
1582 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1586 * Is the previous atime value older than a day? If yes,
1589 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1592 * Good, we can skip the atime update:
1597 int generic_update_time(struct inode *inode, struct timespec *time, int flags)
1599 int iflags = I_DIRTY_TIME;
1601 if (flags & S_ATIME)
1602 inode->i_atime = *time;
1603 if (flags & S_VERSION)
1604 inode_inc_iversion(inode);
1605 if (flags & S_CTIME)
1606 inode->i_ctime = *time;
1607 if (flags & S_MTIME)
1608 inode->i_mtime = *time;
1610 if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION))
1611 iflags |= I_DIRTY_SYNC;
1612 __mark_inode_dirty(inode, iflags);
1615 EXPORT_SYMBOL(generic_update_time);
1618 * This does the actual work of updating an inodes time or version. Must have
1619 * had called mnt_want_write() before calling this.
1621 static int update_time(struct inode *inode, struct timespec *time, int flags)
1623 int (*update_time)(struct inode *, struct timespec *, int);
1625 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1626 generic_update_time;
1628 return update_time(inode, time, flags);
1632 * touch_atime - update the access time
1633 * @path: the &struct path to update
1634 * @inode: inode to update
1636 * Update the accessed time on an inode and mark it for writeback.
1637 * This function automatically handles read only file systems and media,
1638 * as well as the "noatime" flag and inode specific "noatime" markers.
1640 bool atime_needs_update(const struct path *path, struct inode *inode)
1642 struct vfsmount *mnt = path->mnt;
1643 struct timespec now;
1645 if (inode->i_flags & S_NOATIME)
1647 if (IS_NOATIME(inode))
1649 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1652 if (mnt->mnt_flags & MNT_NOATIME)
1654 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1657 now = current_fs_time(inode->i_sb);
1659 if (!relatime_need_update(mnt, inode, now))
1662 if (timespec_equal(&inode->i_atime, &now))
1668 void touch_atime(const struct path *path)
1670 struct vfsmount *mnt = path->mnt;
1671 struct inode *inode = d_inode(path->dentry);
1672 struct timespec now;
1674 if (!atime_needs_update(path, inode))
1677 if (!sb_start_write_trylock(inode->i_sb))
1680 if (__mnt_want_write(mnt) != 0)
1683 * File systems can error out when updating inodes if they need to
1684 * allocate new space to modify an inode (such is the case for
1685 * Btrfs), but since we touch atime while walking down the path we
1686 * really don't care if we failed to update the atime of the file,
1687 * so just ignore the return value.
1688 * We may also fail on filesystems that have the ability to make parts
1689 * of the fs read only, e.g. subvolumes in Btrfs.
1691 now = current_fs_time(inode->i_sb);
1692 update_time(inode, &now, S_ATIME);
1693 __mnt_drop_write(mnt);
1695 sb_end_write(inode->i_sb);
1697 EXPORT_SYMBOL(touch_atime);
1700 * The logic we want is
1702 * if suid or (sgid and xgrp)
1705 int should_remove_suid(struct dentry *dentry)
1707 umode_t mode = d_inode(dentry)->i_mode;
1710 /* suid always must be killed */
1711 if (unlikely(mode & S_ISUID))
1712 kill = ATTR_KILL_SUID;
1715 * sgid without any exec bits is just a mandatory locking mark; leave
1716 * it alone. If some exec bits are set, it's a real sgid; kill it.
1718 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1719 kill |= ATTR_KILL_SGID;
1721 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1726 EXPORT_SYMBOL(should_remove_suid);
1729 * Return mask of changes for notify_change() that need to be done as a
1730 * response to write or truncate. Return 0 if nothing has to be changed.
1731 * Negative value on error (change should be denied).
1733 int dentry_needs_remove_privs(struct dentry *dentry)
1735 struct inode *inode = d_inode(dentry);
1739 if (IS_NOSEC(inode))
1742 mask = should_remove_suid(dentry);
1743 ret = security_inode_need_killpriv(dentry);
1747 mask |= ATTR_KILL_PRIV;
1750 EXPORT_SYMBOL(dentry_needs_remove_privs);
1752 static int __remove_privs(struct dentry *dentry, int kill)
1754 struct iattr newattrs;
1756 newattrs.ia_valid = ATTR_FORCE | kill;
1758 * Note we call this on write, so notify_change will not
1759 * encounter any conflicting delegations:
1761 return notify_change(dentry, &newattrs, NULL);
1765 * Remove special file priviledges (suid, capabilities) when file is written
1768 int file_remove_privs(struct file *file)
1770 struct dentry *dentry = file->f_path.dentry;
1771 struct inode *inode = d_inode(dentry);
1775 /* Fast path for nothing security related */
1776 if (IS_NOSEC(inode))
1779 kill = file_needs_remove_privs(file);
1783 error = __remove_privs(dentry, kill);
1785 inode_has_no_xattr(inode);
1789 EXPORT_SYMBOL(file_remove_privs);
1792 * file_update_time - update mtime and ctime time
1793 * @file: file accessed
1795 * Update the mtime and ctime members of an inode and mark the inode
1796 * for writeback. Note that this function is meant exclusively for
1797 * usage in the file write path of filesystems, and filesystems may
1798 * choose to explicitly ignore update via this function with the
1799 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1800 * timestamps are handled by the server. This can return an error for
1801 * file systems who need to allocate space in order to update an inode.
1804 int file_update_time(struct file *file)
1806 struct inode *inode = file_inode(file);
1807 struct timespec now;
1811 /* First try to exhaust all avenues to not sync */
1812 if (IS_NOCMTIME(inode))
1815 now = current_fs_time(inode->i_sb);
1816 if (!timespec_equal(&inode->i_mtime, &now))
1819 if (!timespec_equal(&inode->i_ctime, &now))
1822 if (IS_I_VERSION(inode))
1823 sync_it |= S_VERSION;
1828 /* Finally allowed to write? Takes lock. */
1829 if (__mnt_want_write_file(file))
1832 ret = update_time(inode, &now, sync_it);
1833 __mnt_drop_write_file(file);
1837 EXPORT_SYMBOL(file_update_time);
1839 int inode_needs_sync(struct inode *inode)
1843 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1847 EXPORT_SYMBOL(inode_needs_sync);
1850 * If we try to find an inode in the inode hash while it is being
1851 * deleted, we have to wait until the filesystem completes its
1852 * deletion before reporting that it isn't found. This function waits
1853 * until the deletion _might_ have completed. Callers are responsible
1854 * to recheck inode state.
1856 * It doesn't matter if I_NEW is not set initially, a call to
1857 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1860 static void __wait_on_freeing_inode(struct inode *inode)
1862 wait_queue_head_t *wq;
1863 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1864 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1865 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1866 spin_unlock(&inode->i_lock);
1867 spin_unlock(&inode_hash_lock);
1869 finish_wait(wq, &wait.wait);
1870 spin_lock(&inode_hash_lock);
1873 static __initdata unsigned long ihash_entries;
1874 static int __init set_ihash_entries(char *str)
1878 ihash_entries = simple_strtoul(str, &str, 0);
1881 __setup("ihash_entries=", set_ihash_entries);
1884 * Initialize the waitqueues and inode hash table.
1886 void __init inode_init_early(void)
1890 /* If hashes are distributed across NUMA nodes, defer
1891 * hash allocation until vmalloc space is available.
1897 alloc_large_system_hash("Inode-cache",
1898 sizeof(struct hlist_head),
1907 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1908 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1911 void __init inode_init(void)
1915 /* inode slab cache */
1916 inode_cachep = kmem_cache_create("inode_cache",
1917 sizeof(struct inode),
1919 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1920 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1923 /* Hash may have been set up in inode_init_early */
1928 alloc_large_system_hash("Inode-cache",
1929 sizeof(struct hlist_head),
1938 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1939 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1942 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1944 inode->i_mode = mode;
1945 if (S_ISCHR(mode)) {
1946 inode->i_fop = &def_chr_fops;
1947 inode->i_rdev = rdev;
1948 } else if (S_ISBLK(mode)) {
1949 inode->i_fop = &def_blk_fops;
1950 inode->i_rdev = rdev;
1951 } else if (S_ISFIFO(mode))
1952 inode->i_fop = &pipefifo_fops;
1953 else if (S_ISSOCK(mode))
1954 ; /* leave it no_open_fops */
1956 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1957 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1960 EXPORT_SYMBOL(init_special_inode);
1963 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1965 * @dir: Directory inode
1966 * @mode: mode of the new inode
1968 void inode_init_owner(struct inode *inode, const struct inode *dir,
1971 inode->i_uid = current_fsuid();
1972 if (dir && dir->i_mode & S_ISGID) {
1973 inode->i_gid = dir->i_gid;
1977 inode->i_gid = current_fsgid();
1978 inode->i_mode = mode;
1980 EXPORT_SYMBOL(inode_init_owner);
1983 * inode_owner_or_capable - check current task permissions to inode
1984 * @inode: inode being checked
1986 * Return true if current either has CAP_FOWNER in a namespace with the
1987 * inode owner uid mapped, or owns the file.
1989 bool inode_owner_or_capable(const struct inode *inode)
1991 struct user_namespace *ns;
1993 if (uid_eq(current_fsuid(), inode->i_uid))
1996 ns = current_user_ns();
1997 if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid))
2001 EXPORT_SYMBOL(inode_owner_or_capable);
2004 * Direct i/o helper functions
2006 static void __inode_dio_wait(struct inode *inode)
2008 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2009 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2012 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
2013 if (atomic_read(&inode->i_dio_count))
2015 } while (atomic_read(&inode->i_dio_count));
2016 finish_wait(wq, &q.wait);
2020 * inode_dio_wait - wait for outstanding DIO requests to finish
2021 * @inode: inode to wait for
2023 * Waits for all pending direct I/O requests to finish so that we can
2024 * proceed with a truncate or equivalent operation.
2026 * Must be called under a lock that serializes taking new references
2027 * to i_dio_count, usually by inode->i_mutex.
2029 void inode_dio_wait(struct inode *inode)
2031 if (atomic_read(&inode->i_dio_count))
2032 __inode_dio_wait(inode);
2034 EXPORT_SYMBOL(inode_dio_wait);
2037 * inode_set_flags - atomically set some inode flags
2039 * Note: the caller should be holding i_mutex, or else be sure that
2040 * they have exclusive access to the inode structure (i.e., while the
2041 * inode is being instantiated). The reason for the cmpxchg() loop
2042 * --- which wouldn't be necessary if all code paths which modify
2043 * i_flags actually followed this rule, is that there is at least one
2044 * code path which doesn't today so we use cmpxchg() out of an abundance
2047 * In the long run, i_mutex is overkill, and we should probably look
2048 * at using the i_lock spinlock to protect i_flags, and then make sure
2049 * it is so documented in include/linux/fs.h and that all code follows
2050 * the locking convention!!
2052 void inode_set_flags(struct inode *inode, unsigned int flags,
2055 unsigned int old_flags, new_flags;
2057 WARN_ON_ONCE(flags & ~mask);
2059 old_flags = ACCESS_ONCE(inode->i_flags);
2060 new_flags = (old_flags & ~mask) | flags;
2061 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2062 new_flags) != old_flags));
2064 EXPORT_SYMBOL(inode_set_flags);
2066 void inode_nohighmem(struct inode *inode)
2068 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2070 EXPORT_SYMBOL(inode_nohighmem);