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;
155 inode->dirtied_when = 0;
157 if (security_inode_alloc(inode))
159 spin_lock_init(&inode->i_lock);
160 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
162 mutex_init(&inode->i_mutex);
163 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
165 atomic_set(&inode->i_dio_count, 0);
167 mapping->a_ops = &empty_aops;
168 mapping->host = inode;
170 atomic_set(&mapping->i_mmap_writable, 0);
171 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
172 mapping->private_data = NULL;
173 mapping->writeback_index = 0;
174 inode->i_private = NULL;
175 inode->i_mapping = mapping;
176 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
177 #ifdef CONFIG_FS_POSIX_ACL
178 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
181 #ifdef CONFIG_FSNOTIFY
182 inode->i_fsnotify_mask = 0;
184 inode->i_flctx = NULL;
185 this_cpu_inc(nr_inodes);
191 EXPORT_SYMBOL(inode_init_always);
193 static struct inode *alloc_inode(struct super_block *sb)
197 if (sb->s_op->alloc_inode)
198 inode = sb->s_op->alloc_inode(sb);
200 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
205 if (unlikely(inode_init_always(sb, inode))) {
206 if (inode->i_sb->s_op->destroy_inode)
207 inode->i_sb->s_op->destroy_inode(inode);
209 kmem_cache_free(inode_cachep, inode);
216 void free_inode_nonrcu(struct inode *inode)
218 kmem_cache_free(inode_cachep, inode);
220 EXPORT_SYMBOL(free_inode_nonrcu);
222 void __destroy_inode(struct inode *inode)
224 BUG_ON(inode_has_buffers(inode));
225 inode_detach_wb(inode);
226 security_inode_free(inode);
227 fsnotify_inode_delete(inode);
228 locks_free_lock_context(inode->i_flctx);
229 if (!inode->i_nlink) {
230 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
231 atomic_long_dec(&inode->i_sb->s_remove_count);
234 #ifdef CONFIG_FS_POSIX_ACL
235 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
236 posix_acl_release(inode->i_acl);
237 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
238 posix_acl_release(inode->i_default_acl);
240 this_cpu_dec(nr_inodes);
242 EXPORT_SYMBOL(__destroy_inode);
244 static void i_callback(struct rcu_head *head)
246 struct inode *inode = container_of(head, struct inode, i_rcu);
247 kmem_cache_free(inode_cachep, inode);
250 static void destroy_inode(struct inode *inode)
252 BUG_ON(!list_empty(&inode->i_lru));
253 __destroy_inode(inode);
254 if (inode->i_sb->s_op->destroy_inode)
255 inode->i_sb->s_op->destroy_inode(inode);
257 call_rcu(&inode->i_rcu, i_callback);
261 * drop_nlink - directly drop an inode's link count
264 * This is a low-level filesystem helper to replace any
265 * direct filesystem manipulation of i_nlink. In cases
266 * where we are attempting to track writes to the
267 * filesystem, a decrement to zero means an imminent
268 * write when the file is truncated and actually unlinked
271 void drop_nlink(struct inode *inode)
273 WARN_ON(inode->i_nlink == 0);
276 atomic_long_inc(&inode->i_sb->s_remove_count);
278 EXPORT_SYMBOL(drop_nlink);
281 * clear_nlink - directly zero an inode's link count
284 * This is a low-level filesystem helper to replace any
285 * direct filesystem manipulation of i_nlink. See
286 * drop_nlink() for why we care about i_nlink hitting zero.
288 void clear_nlink(struct inode *inode)
290 if (inode->i_nlink) {
291 inode->__i_nlink = 0;
292 atomic_long_inc(&inode->i_sb->s_remove_count);
295 EXPORT_SYMBOL(clear_nlink);
298 * set_nlink - directly set an inode's link count
300 * @nlink: new nlink (should be non-zero)
302 * This is a low-level filesystem helper to replace any
303 * direct filesystem manipulation of i_nlink.
305 void set_nlink(struct inode *inode, unsigned int nlink)
310 /* Yes, some filesystems do change nlink from zero to one */
311 if (inode->i_nlink == 0)
312 atomic_long_dec(&inode->i_sb->s_remove_count);
314 inode->__i_nlink = nlink;
317 EXPORT_SYMBOL(set_nlink);
320 * inc_nlink - directly increment an inode's link count
323 * This is a low-level filesystem helper to replace any
324 * direct filesystem manipulation of i_nlink. Currently,
325 * it is only here for parity with dec_nlink().
327 void inc_nlink(struct inode *inode)
329 if (unlikely(inode->i_nlink == 0)) {
330 WARN_ON(!(inode->i_state & I_LINKABLE));
331 atomic_long_dec(&inode->i_sb->s_remove_count);
336 EXPORT_SYMBOL(inc_nlink);
338 void address_space_init_once(struct address_space *mapping)
340 memset(mapping, 0, sizeof(*mapping));
341 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
342 spin_lock_init(&mapping->tree_lock);
343 init_rwsem(&mapping->i_mmap_rwsem);
344 INIT_LIST_HEAD(&mapping->private_list);
345 spin_lock_init(&mapping->private_lock);
346 mapping->i_mmap = RB_ROOT;
348 EXPORT_SYMBOL(address_space_init_once);
351 * These are initializations that only need to be done
352 * once, because the fields are idempotent across use
353 * of the inode, so let the slab aware of that.
355 void inode_init_once(struct inode *inode)
357 memset(inode, 0, sizeof(*inode));
358 INIT_HLIST_NODE(&inode->i_hash);
359 INIT_LIST_HEAD(&inode->i_devices);
360 INIT_LIST_HEAD(&inode->i_io_list);
361 INIT_LIST_HEAD(&inode->i_lru);
362 address_space_init_once(&inode->i_data);
363 i_size_ordered_init(inode);
364 #ifdef CONFIG_FSNOTIFY
365 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
368 EXPORT_SYMBOL(inode_init_once);
370 static void init_once(void *foo)
372 struct inode *inode = (struct inode *) foo;
374 inode_init_once(inode);
378 * inode->i_lock must be held
380 void __iget(struct inode *inode)
382 atomic_inc(&inode->i_count);
386 * get additional reference to inode; caller must already hold one.
388 void ihold(struct inode *inode)
390 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
392 EXPORT_SYMBOL(ihold);
394 static void inode_lru_list_add(struct inode *inode)
396 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
397 this_cpu_inc(nr_unused);
401 * Add inode to LRU if needed (inode is unused and clean).
403 * Needs inode->i_lock held.
405 void inode_add_lru(struct inode *inode)
407 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
408 I_FREEING | I_WILL_FREE)) &&
409 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
410 inode_lru_list_add(inode);
414 static void inode_lru_list_del(struct inode *inode)
417 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
418 this_cpu_dec(nr_unused);
422 * inode_sb_list_add - add inode to the superblock list of inodes
423 * @inode: inode to add
425 void inode_sb_list_add(struct inode *inode)
427 spin_lock(&inode->i_sb->s_inode_list_lock);
428 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
429 spin_unlock(&inode->i_sb->s_inode_list_lock);
431 EXPORT_SYMBOL_GPL(inode_sb_list_add);
433 static inline void inode_sb_list_del(struct inode *inode)
435 if (!list_empty(&inode->i_sb_list)) {
436 spin_lock(&inode->i_sb->s_inode_list_lock);
437 list_del_init(&inode->i_sb_list);
438 spin_unlock(&inode->i_sb->s_inode_list_lock);
442 static unsigned long hash(struct super_block *sb, unsigned long hashval)
446 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
448 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
449 return tmp & i_hash_mask;
453 * __insert_inode_hash - hash an inode
454 * @inode: unhashed inode
455 * @hashval: unsigned long value used to locate this object in the
458 * Add an inode to the inode hash for this superblock.
460 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
462 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
464 spin_lock(&inode_hash_lock);
465 spin_lock(&inode->i_lock);
466 hlist_add_head(&inode->i_hash, b);
467 spin_unlock(&inode->i_lock);
468 spin_unlock(&inode_hash_lock);
470 EXPORT_SYMBOL(__insert_inode_hash);
473 * __remove_inode_hash - remove an inode from the hash
474 * @inode: inode to unhash
476 * Remove an inode from the superblock.
478 void __remove_inode_hash(struct inode *inode)
480 spin_lock(&inode_hash_lock);
481 spin_lock(&inode->i_lock);
482 hlist_del_init(&inode->i_hash);
483 spin_unlock(&inode->i_lock);
484 spin_unlock(&inode_hash_lock);
486 EXPORT_SYMBOL(__remove_inode_hash);
488 void clear_inode(struct inode *inode)
492 * We have to cycle tree_lock here because reclaim can be still in the
493 * process of removing the last page (in __delete_from_page_cache())
494 * and we must not free mapping under it.
496 spin_lock_irq(&inode->i_data.tree_lock);
497 BUG_ON(inode->i_data.nrpages);
498 BUG_ON(inode->i_data.nrshadows);
499 spin_unlock_irq(&inode->i_data.tree_lock);
500 BUG_ON(!list_empty(&inode->i_data.private_list));
501 BUG_ON(!(inode->i_state & I_FREEING));
502 BUG_ON(inode->i_state & I_CLEAR);
503 /* don't need i_lock here, no concurrent mods to i_state */
504 inode->i_state = I_FREEING | I_CLEAR;
506 EXPORT_SYMBOL(clear_inode);
509 * Free the inode passed in, removing it from the lists it is still connected
510 * to. We remove any pages still attached to the inode and wait for any IO that
511 * is still in progress before finally destroying the inode.
513 * An inode must already be marked I_FREEING so that we avoid the inode being
514 * moved back onto lists if we race with other code that manipulates the lists
515 * (e.g. writeback_single_inode). The caller is responsible for setting this.
517 * An inode must already be removed from the LRU list before being evicted from
518 * the cache. This should occur atomically with setting the I_FREEING state
519 * flag, so no inodes here should ever be on the LRU when being evicted.
521 static void evict(struct inode *inode)
523 const struct super_operations *op = inode->i_sb->s_op;
525 BUG_ON(!(inode->i_state & I_FREEING));
526 BUG_ON(!list_empty(&inode->i_lru));
528 if (!list_empty(&inode->i_io_list))
529 inode_io_list_del(inode);
531 inode_sb_list_del(inode);
534 * Wait for flusher thread to be done with the inode so that filesystem
535 * does not start destroying it while writeback is still running. Since
536 * the inode has I_FREEING set, flusher thread won't start new work on
537 * the inode. We just have to wait for running writeback to finish.
539 inode_wait_for_writeback(inode);
541 if (op->evict_inode) {
542 op->evict_inode(inode);
544 truncate_inode_pages_final(&inode->i_data);
547 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
549 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
552 remove_inode_hash(inode);
554 spin_lock(&inode->i_lock);
555 wake_up_bit(&inode->i_state, __I_NEW);
556 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
557 spin_unlock(&inode->i_lock);
559 destroy_inode(inode);
563 * dispose_list - dispose of the contents of a local list
564 * @head: the head of the list to free
566 * Dispose-list gets a local list with local inodes in it, so it doesn't
567 * need to worry about list corruption and SMP locks.
569 static void dispose_list(struct list_head *head)
571 while (!list_empty(head)) {
574 inode = list_first_entry(head, struct inode, i_lru);
575 list_del_init(&inode->i_lru);
582 * evict_inodes - evict all evictable inodes for a superblock
583 * @sb: superblock to operate on
585 * Make sure that no inodes with zero refcount are retained. This is
586 * called by superblock shutdown after having MS_ACTIVE flag removed,
587 * so any inode reaching zero refcount during or after that call will
588 * be immediately evicted.
590 void evict_inodes(struct super_block *sb)
592 struct inode *inode, *next;
595 spin_lock(&sb->s_inode_list_lock);
596 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
597 if (atomic_read(&inode->i_count))
600 spin_lock(&inode->i_lock);
601 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
602 spin_unlock(&inode->i_lock);
606 inode->i_state |= I_FREEING;
607 inode_lru_list_del(inode);
608 spin_unlock(&inode->i_lock);
609 list_add(&inode->i_lru, &dispose);
611 spin_unlock(&sb->s_inode_list_lock);
613 dispose_list(&dispose);
617 * invalidate_inodes - attempt to free all inodes on a superblock
618 * @sb: superblock to operate on
619 * @kill_dirty: flag to guide handling of dirty inodes
621 * Attempts to free all inodes for a given superblock. If there were any
622 * busy inodes return a non-zero value, else zero.
623 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
626 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
629 struct inode *inode, *next;
632 spin_lock(&sb->s_inode_list_lock);
633 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
634 spin_lock(&inode->i_lock);
635 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
636 spin_unlock(&inode->i_lock);
639 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
640 spin_unlock(&inode->i_lock);
644 if (atomic_read(&inode->i_count)) {
645 spin_unlock(&inode->i_lock);
650 inode->i_state |= I_FREEING;
651 inode_lru_list_del(inode);
652 spin_unlock(&inode->i_lock);
653 list_add(&inode->i_lru, &dispose);
655 spin_unlock(&sb->s_inode_list_lock);
657 dispose_list(&dispose);
663 * Isolate the inode from the LRU in preparation for freeing it.
665 * Any inodes which are pinned purely because of attached pagecache have their
666 * pagecache removed. If the inode has metadata buffers attached to
667 * mapping->private_list then try to remove them.
669 * If the inode has the I_REFERENCED flag set, then it means that it has been
670 * used recently - the flag is set in iput_final(). When we encounter such an
671 * inode, clear the flag and move it to the back of the LRU so it gets another
672 * pass through the LRU before it gets reclaimed. This is necessary because of
673 * the fact we are doing lazy LRU updates to minimise lock contention so the
674 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
675 * with this flag set because they are the inodes that are out of order.
677 static enum lru_status inode_lru_isolate(struct list_head *item,
678 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
680 struct list_head *freeable = arg;
681 struct inode *inode = container_of(item, struct inode, i_lru);
684 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
685 * If we fail to get the lock, just skip it.
687 if (!spin_trylock(&inode->i_lock))
691 * Referenced or dirty inodes are still in use. Give them another pass
692 * through the LRU as we canot reclaim them now.
694 if (atomic_read(&inode->i_count) ||
695 (inode->i_state & ~I_REFERENCED)) {
696 list_lru_isolate(lru, &inode->i_lru);
697 spin_unlock(&inode->i_lock);
698 this_cpu_dec(nr_unused);
702 /* recently referenced inodes get one more pass */
703 if (inode->i_state & I_REFERENCED) {
704 inode->i_state &= ~I_REFERENCED;
705 spin_unlock(&inode->i_lock);
709 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
711 spin_unlock(&inode->i_lock);
712 spin_unlock(lru_lock);
713 if (remove_inode_buffers(inode)) {
715 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
716 if (current_is_kswapd())
717 __count_vm_events(KSWAPD_INODESTEAL, reap);
719 __count_vm_events(PGINODESTEAL, reap);
720 if (current->reclaim_state)
721 current->reclaim_state->reclaimed_slab += reap;
728 WARN_ON(inode->i_state & I_NEW);
729 inode->i_state |= I_FREEING;
730 list_lru_isolate_move(lru, &inode->i_lru, freeable);
731 spin_unlock(&inode->i_lock);
733 this_cpu_dec(nr_unused);
738 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
739 * This is called from the superblock shrinker function with a number of inodes
740 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
741 * then are freed outside inode_lock by dispose_list().
743 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
748 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
749 inode_lru_isolate, &freeable);
750 dispose_list(&freeable);
754 static void __wait_on_freeing_inode(struct inode *inode);
756 * Called with the inode lock held.
758 static struct inode *find_inode(struct super_block *sb,
759 struct hlist_head *head,
760 int (*test)(struct inode *, void *),
763 struct inode *inode = NULL;
766 hlist_for_each_entry(inode, head, i_hash) {
767 if (inode->i_sb != sb)
769 if (!test(inode, data))
771 spin_lock(&inode->i_lock);
772 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
773 __wait_on_freeing_inode(inode);
777 spin_unlock(&inode->i_lock);
784 * find_inode_fast is the fast path version of find_inode, see the comment at
785 * iget_locked for details.
787 static struct inode *find_inode_fast(struct super_block *sb,
788 struct hlist_head *head, unsigned long ino)
790 struct inode *inode = NULL;
793 hlist_for_each_entry(inode, head, i_hash) {
794 if (inode->i_ino != ino)
796 if (inode->i_sb != sb)
798 spin_lock(&inode->i_lock);
799 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
800 __wait_on_freeing_inode(inode);
804 spin_unlock(&inode->i_lock);
811 * Each cpu owns a range of LAST_INO_BATCH numbers.
812 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
813 * to renew the exhausted range.
815 * This does not significantly increase overflow rate because every CPU can
816 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
817 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
818 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
819 * overflow rate by 2x, which does not seem too significant.
821 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
822 * error if st_ino won't fit in target struct field. Use 32bit counter
823 * here to attempt to avoid that.
825 #define LAST_INO_BATCH 1024
826 static DEFINE_PER_CPU(unsigned int, last_ino);
828 unsigned int get_next_ino(void)
830 unsigned int *p = &get_cpu_var(last_ino);
831 unsigned int res = *p;
834 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
835 static atomic_t shared_last_ino;
836 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
838 res = next - LAST_INO_BATCH;
843 /* get_next_ino should not provide a 0 inode number */
847 put_cpu_var(last_ino);
850 EXPORT_SYMBOL(get_next_ino);
853 * new_inode_pseudo - obtain an inode
856 * Allocates a new inode for given superblock.
857 * Inode wont be chained in superblock s_inodes list
859 * - fs can't be unmount
860 * - quotas, fsnotify, writeback can't work
862 struct inode *new_inode_pseudo(struct super_block *sb)
864 struct inode *inode = alloc_inode(sb);
867 spin_lock(&inode->i_lock);
869 spin_unlock(&inode->i_lock);
870 INIT_LIST_HEAD(&inode->i_sb_list);
876 * new_inode - obtain an inode
879 * Allocates a new inode for given superblock. The default gfp_mask
880 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
881 * If HIGHMEM pages are unsuitable or it is known that pages allocated
882 * for the page cache are not reclaimable or migratable,
883 * mapping_set_gfp_mask() must be called with suitable flags on the
884 * newly created inode's mapping
887 struct inode *new_inode(struct super_block *sb)
891 spin_lock_prefetch(&sb->s_inode_list_lock);
893 inode = new_inode_pseudo(sb);
895 inode_sb_list_add(inode);
898 EXPORT_SYMBOL(new_inode);
900 #ifdef CONFIG_DEBUG_LOCK_ALLOC
901 void lockdep_annotate_inode_mutex_key(struct inode *inode)
903 if (S_ISDIR(inode->i_mode)) {
904 struct file_system_type *type = inode->i_sb->s_type;
906 /* Set new key only if filesystem hasn't already changed it */
907 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
909 * ensure nobody is actually holding i_mutex
911 mutex_destroy(&inode->i_mutex);
912 mutex_init(&inode->i_mutex);
913 lockdep_set_class(&inode->i_mutex,
914 &type->i_mutex_dir_key);
918 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
922 * unlock_new_inode - clear the I_NEW state and wake up any waiters
923 * @inode: new inode to unlock
925 * Called when the inode is fully initialised to clear the new state of the
926 * inode and wake up anyone waiting for the inode to finish initialisation.
928 void unlock_new_inode(struct inode *inode)
930 lockdep_annotate_inode_mutex_key(inode);
931 spin_lock(&inode->i_lock);
932 WARN_ON(!(inode->i_state & I_NEW));
933 inode->i_state &= ~I_NEW;
935 wake_up_bit(&inode->i_state, __I_NEW);
936 spin_unlock(&inode->i_lock);
938 EXPORT_SYMBOL(unlock_new_inode);
941 * lock_two_nondirectories - take two i_mutexes on non-directory objects
943 * Lock any non-NULL argument that is not a directory.
944 * Zero, one or two objects may be locked by this function.
946 * @inode1: first inode to lock
947 * @inode2: second inode to lock
949 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
952 swap(inode1, inode2);
954 if (inode1 && !S_ISDIR(inode1->i_mode))
955 mutex_lock(&inode1->i_mutex);
956 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
957 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_NONDIR2);
959 EXPORT_SYMBOL(lock_two_nondirectories);
962 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
963 * @inode1: first inode to unlock
964 * @inode2: second inode to unlock
966 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
968 if (inode1 && !S_ISDIR(inode1->i_mode))
969 mutex_unlock(&inode1->i_mutex);
970 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
971 mutex_unlock(&inode2->i_mutex);
973 EXPORT_SYMBOL(unlock_two_nondirectories);
976 * iget5_locked - obtain an inode from a mounted file system
977 * @sb: super block of file system
978 * @hashval: hash value (usually inode number) to get
979 * @test: callback used for comparisons between inodes
980 * @set: callback used to initialize a new struct inode
981 * @data: opaque data pointer to pass to @test and @set
983 * Search for the inode specified by @hashval and @data in the inode cache,
984 * and if present it is return it with an increased reference count. This is
985 * a generalized version of iget_locked() for file systems where the inode
986 * number is not sufficient for unique identification of an inode.
988 * If the inode is not in cache, allocate a new inode and return it locked,
989 * hashed, and with the I_NEW flag set. The file system gets to fill it in
990 * before unlocking it via unlock_new_inode().
992 * Note both @test and @set are called with the inode_hash_lock held, so can't
995 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
996 int (*test)(struct inode *, void *),
997 int (*set)(struct inode *, void *), void *data)
999 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1000 struct inode *inode;
1002 spin_lock(&inode_hash_lock);
1003 inode = find_inode(sb, head, test, data);
1004 spin_unlock(&inode_hash_lock);
1007 wait_on_inode(inode);
1011 inode = alloc_inode(sb);
1015 spin_lock(&inode_hash_lock);
1016 /* We released the lock, so.. */
1017 old = find_inode(sb, head, test, data);
1019 if (set(inode, data))
1022 spin_lock(&inode->i_lock);
1023 inode->i_state = I_NEW;
1024 hlist_add_head(&inode->i_hash, head);
1025 spin_unlock(&inode->i_lock);
1026 inode_sb_list_add(inode);
1027 spin_unlock(&inode_hash_lock);
1029 /* Return the locked inode with I_NEW set, the
1030 * caller is responsible for filling in the contents
1036 * Uhhuh, somebody else created the same inode under
1037 * us. Use the old inode instead of the one we just
1040 spin_unlock(&inode_hash_lock);
1041 destroy_inode(inode);
1043 wait_on_inode(inode);
1048 spin_unlock(&inode_hash_lock);
1049 destroy_inode(inode);
1052 EXPORT_SYMBOL(iget5_locked);
1055 * iget_locked - obtain an inode from a mounted file system
1056 * @sb: super block of file system
1057 * @ino: inode number to get
1059 * Search for the inode specified by @ino in the inode cache and if present
1060 * return it with an increased reference count. This is for file systems
1061 * where the inode number is sufficient for unique identification of an inode.
1063 * If the inode is not in cache, allocate a new inode and return it locked,
1064 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1065 * before unlocking it via unlock_new_inode().
1067 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1069 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1070 struct inode *inode;
1072 spin_lock(&inode_hash_lock);
1073 inode = find_inode_fast(sb, head, ino);
1074 spin_unlock(&inode_hash_lock);
1076 wait_on_inode(inode);
1080 inode = alloc_inode(sb);
1084 spin_lock(&inode_hash_lock);
1085 /* We released the lock, so.. */
1086 old = find_inode_fast(sb, head, ino);
1089 spin_lock(&inode->i_lock);
1090 inode->i_state = I_NEW;
1091 hlist_add_head(&inode->i_hash, head);
1092 spin_unlock(&inode->i_lock);
1093 inode_sb_list_add(inode);
1094 spin_unlock(&inode_hash_lock);
1096 /* Return the locked inode with I_NEW set, the
1097 * caller is responsible for filling in the contents
1103 * Uhhuh, somebody else created the same inode under
1104 * us. Use the old inode instead of the one we just
1107 spin_unlock(&inode_hash_lock);
1108 destroy_inode(inode);
1110 wait_on_inode(inode);
1114 EXPORT_SYMBOL(iget_locked);
1117 * search the inode cache for a matching inode number.
1118 * If we find one, then the inode number we are trying to
1119 * allocate is not unique and so we should not use it.
1121 * Returns 1 if the inode number is unique, 0 if it is not.
1123 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1125 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1126 struct inode *inode;
1128 spin_lock(&inode_hash_lock);
1129 hlist_for_each_entry(inode, b, i_hash) {
1130 if (inode->i_ino == ino && inode->i_sb == sb) {
1131 spin_unlock(&inode_hash_lock);
1135 spin_unlock(&inode_hash_lock);
1141 * iunique - get a unique inode number
1143 * @max_reserved: highest reserved inode number
1145 * Obtain an inode number that is unique on the system for a given
1146 * superblock. This is used by file systems that have no natural
1147 * permanent inode numbering system. An inode number is returned that
1148 * is higher than the reserved limit but unique.
1151 * With a large number of inodes live on the file system this function
1152 * currently becomes quite slow.
1154 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1157 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1158 * error if st_ino won't fit in target struct field. Use 32bit counter
1159 * here to attempt to avoid that.
1161 static DEFINE_SPINLOCK(iunique_lock);
1162 static unsigned int counter;
1165 spin_lock(&iunique_lock);
1167 if (counter <= max_reserved)
1168 counter = max_reserved + 1;
1170 } while (!test_inode_iunique(sb, res));
1171 spin_unlock(&iunique_lock);
1175 EXPORT_SYMBOL(iunique);
1177 struct inode *igrab(struct inode *inode)
1179 spin_lock(&inode->i_lock);
1180 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1182 spin_unlock(&inode->i_lock);
1184 spin_unlock(&inode->i_lock);
1186 * Handle the case where s_op->clear_inode is not been
1187 * called yet, and somebody is calling igrab
1188 * while the inode is getting freed.
1194 EXPORT_SYMBOL(igrab);
1197 * ilookup5_nowait - search for an inode in the inode cache
1198 * @sb: super block of file system to search
1199 * @hashval: hash value (usually inode number) to search for
1200 * @test: callback used for comparisons between inodes
1201 * @data: opaque data pointer to pass to @test
1203 * Search for the inode specified by @hashval and @data in the inode cache.
1204 * If the inode is in the cache, the inode is returned with an incremented
1207 * Note: I_NEW is not waited upon so you have to be very careful what you do
1208 * with the returned inode. You probably should be using ilookup5() instead.
1210 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1212 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1213 int (*test)(struct inode *, void *), void *data)
1215 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1216 struct inode *inode;
1218 spin_lock(&inode_hash_lock);
1219 inode = find_inode(sb, head, test, data);
1220 spin_unlock(&inode_hash_lock);
1224 EXPORT_SYMBOL(ilookup5_nowait);
1227 * ilookup5 - search for an inode in the inode cache
1228 * @sb: super block of file system to search
1229 * @hashval: hash value (usually inode number) to search for
1230 * @test: callback used for comparisons between inodes
1231 * @data: opaque data pointer to pass to @test
1233 * Search for the inode specified by @hashval and @data in the inode cache,
1234 * and if the inode is in the cache, return the inode with an incremented
1235 * reference count. Waits on I_NEW before returning the inode.
1236 * returned with an incremented reference count.
1238 * This is a generalized version of ilookup() for file systems where the
1239 * inode number is not sufficient for unique identification of an inode.
1241 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1243 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1244 int (*test)(struct inode *, void *), void *data)
1246 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1249 wait_on_inode(inode);
1252 EXPORT_SYMBOL(ilookup5);
1255 * ilookup - search for an inode in the inode cache
1256 * @sb: super block of file system to search
1257 * @ino: inode number to search for
1259 * Search for the inode @ino in the inode cache, and if the inode is in the
1260 * cache, the inode is returned with an incremented reference count.
1262 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1264 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1265 struct inode *inode;
1267 spin_lock(&inode_hash_lock);
1268 inode = find_inode_fast(sb, head, ino);
1269 spin_unlock(&inode_hash_lock);
1272 wait_on_inode(inode);
1275 EXPORT_SYMBOL(ilookup);
1278 * find_inode_nowait - find an inode in the inode cache
1279 * @sb: super block of file system to search
1280 * @hashval: hash value (usually inode number) to search for
1281 * @match: callback used for comparisons between inodes
1282 * @data: opaque data pointer to pass to @match
1284 * Search for the inode specified by @hashval and @data in the inode
1285 * cache, where the helper function @match will return 0 if the inode
1286 * does not match, 1 if the inode does match, and -1 if the search
1287 * should be stopped. The @match function must be responsible for
1288 * taking the i_lock spin_lock and checking i_state for an inode being
1289 * freed or being initialized, and incrementing the reference count
1290 * before returning 1. It also must not sleep, since it is called with
1291 * the inode_hash_lock spinlock held.
1293 * This is a even more generalized version of ilookup5() when the
1294 * function must never block --- find_inode() can block in
1295 * __wait_on_freeing_inode() --- or when the caller can not increment
1296 * the reference count because the resulting iput() might cause an
1297 * inode eviction. The tradeoff is that the @match funtion must be
1298 * very carefully implemented.
1300 struct inode *find_inode_nowait(struct super_block *sb,
1301 unsigned long hashval,
1302 int (*match)(struct inode *, unsigned long,
1306 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1307 struct inode *inode, *ret_inode = NULL;
1310 spin_lock(&inode_hash_lock);
1311 hlist_for_each_entry(inode, head, i_hash) {
1312 if (inode->i_sb != sb)
1314 mval = match(inode, hashval, data);
1322 spin_unlock(&inode_hash_lock);
1325 EXPORT_SYMBOL(find_inode_nowait);
1327 int insert_inode_locked(struct inode *inode)
1329 struct super_block *sb = inode->i_sb;
1330 ino_t ino = inode->i_ino;
1331 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1334 struct inode *old = NULL;
1335 spin_lock(&inode_hash_lock);
1336 hlist_for_each_entry(old, head, i_hash) {
1337 if (old->i_ino != ino)
1339 if (old->i_sb != sb)
1341 spin_lock(&old->i_lock);
1342 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1343 spin_unlock(&old->i_lock);
1349 spin_lock(&inode->i_lock);
1350 inode->i_state |= I_NEW;
1351 hlist_add_head(&inode->i_hash, head);
1352 spin_unlock(&inode->i_lock);
1353 spin_unlock(&inode_hash_lock);
1357 spin_unlock(&old->i_lock);
1358 spin_unlock(&inode_hash_lock);
1360 if (unlikely(!inode_unhashed(old))) {
1367 EXPORT_SYMBOL(insert_inode_locked);
1369 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1370 int (*test)(struct inode *, void *), void *data)
1372 struct super_block *sb = inode->i_sb;
1373 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1376 struct inode *old = NULL;
1378 spin_lock(&inode_hash_lock);
1379 hlist_for_each_entry(old, head, i_hash) {
1380 if (old->i_sb != sb)
1382 if (!test(old, data))
1384 spin_lock(&old->i_lock);
1385 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1386 spin_unlock(&old->i_lock);
1392 spin_lock(&inode->i_lock);
1393 inode->i_state |= I_NEW;
1394 hlist_add_head(&inode->i_hash, head);
1395 spin_unlock(&inode->i_lock);
1396 spin_unlock(&inode_hash_lock);
1400 spin_unlock(&old->i_lock);
1401 spin_unlock(&inode_hash_lock);
1403 if (unlikely(!inode_unhashed(old))) {
1410 EXPORT_SYMBOL(insert_inode_locked4);
1413 int generic_delete_inode(struct inode *inode)
1417 EXPORT_SYMBOL(generic_delete_inode);
1420 * Called when we're dropping the last reference
1423 * Call the FS "drop_inode()" function, defaulting to
1424 * the legacy UNIX filesystem behaviour. If it tells
1425 * us to evict inode, do so. Otherwise, retain inode
1426 * in cache if fs is alive, sync and evict if fs is
1429 static void iput_final(struct inode *inode)
1431 struct super_block *sb = inode->i_sb;
1432 const struct super_operations *op = inode->i_sb->s_op;
1435 WARN_ON(inode->i_state & I_NEW);
1438 drop = op->drop_inode(inode);
1440 drop = generic_drop_inode(inode);
1442 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1443 inode->i_state |= I_REFERENCED;
1444 inode_add_lru(inode);
1445 spin_unlock(&inode->i_lock);
1450 inode->i_state |= I_WILL_FREE;
1451 spin_unlock(&inode->i_lock);
1452 write_inode_now(inode, 1);
1453 spin_lock(&inode->i_lock);
1454 WARN_ON(inode->i_state & I_NEW);
1455 inode->i_state &= ~I_WILL_FREE;
1458 inode->i_state |= I_FREEING;
1459 if (!list_empty(&inode->i_lru))
1460 inode_lru_list_del(inode);
1461 spin_unlock(&inode->i_lock);
1467 * iput - put an inode
1468 * @inode: inode to put
1470 * Puts an inode, dropping its usage count. If the inode use count hits
1471 * zero, the inode is then freed and may also be destroyed.
1473 * Consequently, iput() can sleep.
1475 void iput(struct inode *inode)
1479 BUG_ON(inode->i_state & I_CLEAR);
1481 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1482 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1483 atomic_inc(&inode->i_count);
1484 inode->i_state &= ~I_DIRTY_TIME;
1485 spin_unlock(&inode->i_lock);
1486 trace_writeback_lazytime_iput(inode);
1487 mark_inode_dirty_sync(inode);
1493 EXPORT_SYMBOL(iput);
1496 * bmap - find a block number in a file
1497 * @inode: inode of file
1498 * @block: block to find
1500 * Returns the block number on the device holding the inode that
1501 * is the disk block number for the block of the file requested.
1502 * That is, asked for block 4 of inode 1 the function will return the
1503 * disk block relative to the disk start that holds that block of the
1506 sector_t bmap(struct inode *inode, sector_t block)
1509 if (inode->i_mapping->a_ops->bmap)
1510 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1513 EXPORT_SYMBOL(bmap);
1516 * With relative atime, only update atime if the previous atime is
1517 * earlier than either the ctime or mtime or if at least a day has
1518 * passed since the last atime update.
1520 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1521 struct timespec now)
1524 if (!(mnt->mnt_flags & MNT_RELATIME))
1527 * Is mtime younger than atime? If yes, update atime:
1529 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1532 * Is ctime younger than atime? If yes, update atime:
1534 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1538 * Is the previous atime value older than a day? If yes,
1541 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1544 * Good, we can skip the atime update:
1549 int generic_update_time(struct inode *inode, struct timespec *time, int flags)
1551 int iflags = I_DIRTY_TIME;
1553 if (flags & S_ATIME)
1554 inode->i_atime = *time;
1555 if (flags & S_VERSION)
1556 inode_inc_iversion(inode);
1557 if (flags & S_CTIME)
1558 inode->i_ctime = *time;
1559 if (flags & S_MTIME)
1560 inode->i_mtime = *time;
1562 if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION))
1563 iflags |= I_DIRTY_SYNC;
1564 __mark_inode_dirty(inode, iflags);
1567 EXPORT_SYMBOL(generic_update_time);
1570 * This does the actual work of updating an inodes time or version. Must have
1571 * had called mnt_want_write() before calling this.
1573 static int update_time(struct inode *inode, struct timespec *time, int flags)
1575 int (*update_time)(struct inode *, struct timespec *, int);
1577 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1578 generic_update_time;
1580 return update_time(inode, time, flags);
1584 * touch_atime - update the access time
1585 * @path: the &struct path to update
1587 * Update the accessed time on an inode and mark it for writeback.
1588 * This function automatically handles read only file systems and media,
1589 * as well as the "noatime" flag and inode specific "noatime" markers.
1591 bool atime_needs_update(const struct path *path, struct inode *inode)
1593 struct vfsmount *mnt = path->mnt;
1594 struct timespec now;
1596 if (inode->i_flags & S_NOATIME)
1598 if (IS_NOATIME(inode))
1600 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1603 if (mnt->mnt_flags & MNT_NOATIME)
1605 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1608 now = current_fs_time(inode->i_sb);
1610 if (!relatime_need_update(mnt, inode, now))
1613 if (timespec_equal(&inode->i_atime, &now))
1619 void touch_atime(const struct path *path)
1621 struct vfsmount *mnt = path->mnt;
1622 struct inode *inode = d_inode(path->dentry);
1623 struct timespec now;
1625 if (!atime_needs_update(path, inode))
1628 if (!sb_start_write_trylock(inode->i_sb))
1631 if (__mnt_want_write(mnt) != 0)
1634 * File systems can error out when updating inodes if they need to
1635 * allocate new space to modify an inode (such is the case for
1636 * Btrfs), but since we touch atime while walking down the path we
1637 * really don't care if we failed to update the atime of the file,
1638 * so just ignore the return value.
1639 * We may also fail on filesystems that have the ability to make parts
1640 * of the fs read only, e.g. subvolumes in Btrfs.
1642 now = current_fs_time(inode->i_sb);
1643 update_time(inode, &now, S_ATIME);
1644 __mnt_drop_write(mnt);
1646 sb_end_write(inode->i_sb);
1648 EXPORT_SYMBOL(touch_atime);
1651 * The logic we want is
1653 * if suid or (sgid and xgrp)
1656 int should_remove_suid(struct dentry *dentry)
1658 umode_t mode = d_inode(dentry)->i_mode;
1661 /* suid always must be killed */
1662 if (unlikely(mode & S_ISUID))
1663 kill = ATTR_KILL_SUID;
1666 * sgid without any exec bits is just a mandatory locking mark; leave
1667 * it alone. If some exec bits are set, it's a real sgid; kill it.
1669 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1670 kill |= ATTR_KILL_SGID;
1672 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1677 EXPORT_SYMBOL(should_remove_suid);
1680 * Return mask of changes for notify_change() that need to be done as a
1681 * response to write or truncate. Return 0 if nothing has to be changed.
1682 * Negative value on error (change should be denied).
1684 int dentry_needs_remove_privs(struct dentry *dentry)
1686 struct inode *inode = d_inode(dentry);
1690 if (IS_NOSEC(inode))
1693 mask = should_remove_suid(dentry);
1694 ret = security_inode_need_killpriv(dentry);
1698 mask |= ATTR_KILL_PRIV;
1701 EXPORT_SYMBOL(dentry_needs_remove_privs);
1703 static int __remove_privs(struct dentry *dentry, int kill)
1705 struct iattr newattrs;
1707 newattrs.ia_valid = ATTR_FORCE | kill;
1709 * Note we call this on write, so notify_change will not
1710 * encounter any conflicting delegations:
1712 return notify_change(dentry, &newattrs, NULL);
1716 * Remove special file priviledges (suid, capabilities) when file is written
1719 int file_remove_privs(struct file *file)
1721 struct dentry *dentry = file->f_path.dentry;
1722 struct inode *inode = d_inode(dentry);
1726 /* Fast path for nothing security related */
1727 if (IS_NOSEC(inode))
1730 kill = file_needs_remove_privs(file);
1734 error = __remove_privs(dentry, kill);
1736 inode_has_no_xattr(inode);
1740 EXPORT_SYMBOL(file_remove_privs);
1743 * file_update_time - update mtime and ctime time
1744 * @file: file accessed
1746 * Update the mtime and ctime members of an inode and mark the inode
1747 * for writeback. Note that this function is meant exclusively for
1748 * usage in the file write path of filesystems, and filesystems may
1749 * choose to explicitly ignore update via this function with the
1750 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1751 * timestamps are handled by the server. This can return an error for
1752 * file systems who need to allocate space in order to update an inode.
1755 int file_update_time(struct file *file)
1757 struct inode *inode = file_inode(file);
1758 struct timespec now;
1762 /* First try to exhaust all avenues to not sync */
1763 if (IS_NOCMTIME(inode))
1766 now = current_fs_time(inode->i_sb);
1767 if (!timespec_equal(&inode->i_mtime, &now))
1770 if (!timespec_equal(&inode->i_ctime, &now))
1773 if (IS_I_VERSION(inode))
1774 sync_it |= S_VERSION;
1779 /* Finally allowed to write? Takes lock. */
1780 if (__mnt_want_write_file(file))
1783 ret = update_time(inode, &now, sync_it);
1784 __mnt_drop_write_file(file);
1788 EXPORT_SYMBOL(file_update_time);
1790 int inode_needs_sync(struct inode *inode)
1794 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1798 EXPORT_SYMBOL(inode_needs_sync);
1801 * If we try to find an inode in the inode hash while it is being
1802 * deleted, we have to wait until the filesystem completes its
1803 * deletion before reporting that it isn't found. This function waits
1804 * until the deletion _might_ have completed. Callers are responsible
1805 * to recheck inode state.
1807 * It doesn't matter if I_NEW is not set initially, a call to
1808 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1811 static void __wait_on_freeing_inode(struct inode *inode)
1813 wait_queue_head_t *wq;
1814 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1815 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1816 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1817 spin_unlock(&inode->i_lock);
1818 spin_unlock(&inode_hash_lock);
1820 finish_wait(wq, &wait.wait);
1821 spin_lock(&inode_hash_lock);
1824 static __initdata unsigned long ihash_entries;
1825 static int __init set_ihash_entries(char *str)
1829 ihash_entries = simple_strtoul(str, &str, 0);
1832 __setup("ihash_entries=", set_ihash_entries);
1835 * Initialize the waitqueues and inode hash table.
1837 void __init inode_init_early(void)
1841 /* If hashes are distributed across NUMA nodes, defer
1842 * hash allocation until vmalloc space is available.
1848 alloc_large_system_hash("Inode-cache",
1849 sizeof(struct hlist_head),
1858 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1859 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1862 void __init inode_init(void)
1866 /* inode slab cache */
1867 inode_cachep = kmem_cache_create("inode_cache",
1868 sizeof(struct inode),
1870 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1874 /* Hash may have been set up in inode_init_early */
1879 alloc_large_system_hash("Inode-cache",
1880 sizeof(struct hlist_head),
1889 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1890 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1893 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1895 inode->i_mode = mode;
1896 if (S_ISCHR(mode)) {
1897 inode->i_fop = &def_chr_fops;
1898 inode->i_rdev = rdev;
1899 } else if (S_ISBLK(mode)) {
1900 inode->i_fop = &def_blk_fops;
1901 inode->i_rdev = rdev;
1902 } else if (S_ISFIFO(mode))
1903 inode->i_fop = &pipefifo_fops;
1904 else if (S_ISSOCK(mode))
1905 ; /* leave it no_open_fops */
1907 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1908 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1911 EXPORT_SYMBOL(init_special_inode);
1914 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1916 * @dir: Directory inode
1917 * @mode: mode of the new inode
1919 void inode_init_owner(struct inode *inode, const struct inode *dir,
1922 inode->i_uid = current_fsuid();
1923 if (dir && dir->i_mode & S_ISGID) {
1924 inode->i_gid = dir->i_gid;
1928 inode->i_gid = current_fsgid();
1929 inode->i_mode = mode;
1931 EXPORT_SYMBOL(inode_init_owner);
1934 * inode_owner_or_capable - check current task permissions to inode
1935 * @inode: inode being checked
1937 * Return true if current either has CAP_FOWNER in a namespace with the
1938 * inode owner uid mapped, or owns the file.
1940 bool inode_owner_or_capable(const struct inode *inode)
1942 struct user_namespace *ns;
1944 if (uid_eq(current_fsuid(), inode->i_uid))
1947 ns = current_user_ns();
1948 if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid))
1952 EXPORT_SYMBOL(inode_owner_or_capable);
1955 * Direct i/o helper functions
1957 static void __inode_dio_wait(struct inode *inode)
1959 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1960 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1963 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1964 if (atomic_read(&inode->i_dio_count))
1966 } while (atomic_read(&inode->i_dio_count));
1967 finish_wait(wq, &q.wait);
1971 * inode_dio_wait - wait for outstanding DIO requests to finish
1972 * @inode: inode to wait for
1974 * Waits for all pending direct I/O requests to finish so that we can
1975 * proceed with a truncate or equivalent operation.
1977 * Must be called under a lock that serializes taking new references
1978 * to i_dio_count, usually by inode->i_mutex.
1980 void inode_dio_wait(struct inode *inode)
1982 if (atomic_read(&inode->i_dio_count))
1983 __inode_dio_wait(inode);
1985 EXPORT_SYMBOL(inode_dio_wait);
1988 * inode_set_flags - atomically set some inode flags
1990 * Note: the caller should be holding i_mutex, or else be sure that
1991 * they have exclusive access to the inode structure (i.e., while the
1992 * inode is being instantiated). The reason for the cmpxchg() loop
1993 * --- which wouldn't be necessary if all code paths which modify
1994 * i_flags actually followed this rule, is that there is at least one
1995 * code path which doesn't today so we use cmpxchg() out of an abundance
1998 * In the long run, i_mutex is overkill, and we should probably look
1999 * at using the i_lock spinlock to protect i_flags, and then make sure
2000 * it is so documented in include/linux/fs.h and that all code follows
2001 * the locking convention!!
2003 void inode_set_flags(struct inode *inode, unsigned int flags,
2006 unsigned int old_flags, new_flags;
2008 WARN_ON_ONCE(flags & ~mask);
2010 old_flags = ACCESS_ONCE(inode->i_flags);
2011 new_flags = (old_flags & ~mask) | flags;
2012 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2013 new_flags) != old_flags));
2015 EXPORT_SYMBOL(inode_set_flags);