Linux 4.19.133
[linux/fpc-iii.git] / fs / inode.c
blobc9eb5041ffae199f46c397280296258bd329efb1
1 /*
2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
4 */
5 #include <linux/export.h>
6 #include <linux/fs.h>
7 #include <linux/mm.h>
8 #include <linux/backing-dev.h>
9 #include <linux/hash.h>
10 #include <linux/swap.h>
11 #include <linux/security.h>
12 #include <linux/cdev.h>
13 #include <linux/bootmem.h>
14 #include <linux/fsnotify.h>
15 #include <linux/mount.h>
16 #include <linux/posix_acl.h>
17 #include <linux/prefetch.h>
18 #include <linux/buffer_head.h> /* for inode_has_buffers */
19 #include <linux/ratelimit.h>
20 #include <linux/list_lru.h>
21 #include <linux/iversion.h>
22 #include <trace/events/writeback.h>
23 #include "internal.h"
26 * Inode locking rules:
28 * inode->i_lock protects:
29 * inode->i_state, inode->i_hash, __iget()
30 * Inode LRU list locks protect:
31 * inode->i_sb->s_inode_lru, inode->i_lru
32 * inode->i_sb->s_inode_list_lock protects:
33 * inode->i_sb->s_inodes, inode->i_sb_list
34 * bdi->wb.list_lock protects:
35 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
36 * inode_hash_lock protects:
37 * inode_hashtable, inode->i_hash
39 * Lock ordering:
41 * inode->i_sb->s_inode_list_lock
42 * inode->i_lock
43 * Inode LRU list locks
45 * bdi->wb.list_lock
46 * inode->i_lock
48 * inode_hash_lock
49 * inode->i_sb->s_inode_list_lock
50 * inode->i_lock
52 * iunique_lock
53 * inode_hash_lock
56 static unsigned int i_hash_mask __read_mostly;
57 static unsigned int i_hash_shift __read_mostly;
58 static struct hlist_head *inode_hashtable __read_mostly;
59 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
62 * Empty aops. Can be used for the cases where the user does not
63 * define any of the address_space operations.
65 const struct address_space_operations empty_aops = {
67 EXPORT_SYMBOL(empty_aops);
70 * Statistics gathering..
72 struct inodes_stat_t inodes_stat;
74 static DEFINE_PER_CPU(unsigned long, nr_inodes);
75 static DEFINE_PER_CPU(unsigned long, nr_unused);
77 static struct kmem_cache *inode_cachep __read_mostly;
79 static long get_nr_inodes(void)
81 int i;
82 long sum = 0;
83 for_each_possible_cpu(i)
84 sum += per_cpu(nr_inodes, i);
85 return sum < 0 ? 0 : sum;
88 static inline long get_nr_inodes_unused(void)
90 int i;
91 long sum = 0;
92 for_each_possible_cpu(i)
93 sum += per_cpu(nr_unused, i);
94 return sum < 0 ? 0 : sum;
97 long get_nr_dirty_inodes(void)
99 /* not actually dirty inodes, but a wild approximation */
100 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
101 return nr_dirty > 0 ? nr_dirty : 0;
105 * Handle nr_inode sysctl
107 #ifdef CONFIG_SYSCTL
108 int proc_nr_inodes(struct ctl_table *table, int write,
109 void __user *buffer, size_t *lenp, loff_t *ppos)
111 inodes_stat.nr_inodes = get_nr_inodes();
112 inodes_stat.nr_unused = get_nr_inodes_unused();
113 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
115 #endif
117 static int no_open(struct inode *inode, struct file *file)
119 return -ENXIO;
123 * inode_init_always - perform inode structure initialisation
124 * @sb: superblock inode belongs to
125 * @inode: inode to initialise
127 * These are initializations that need to be done on every inode
128 * allocation as the fields are not initialised by slab allocation.
130 int inode_init_always(struct super_block *sb, struct inode *inode)
132 static const struct inode_operations empty_iops;
133 static const struct file_operations no_open_fops = {.open = no_open};
134 struct address_space *const mapping = &inode->i_data;
136 inode->i_sb = sb;
137 inode->i_blkbits = sb->s_blocksize_bits;
138 inode->i_flags = 0;
139 atomic64_set(&inode->i_sequence, 0);
140 atomic_set(&inode->i_count, 1);
141 inode->i_op = &empty_iops;
142 inode->i_fop = &no_open_fops;
143 inode->__i_nlink = 1;
144 inode->i_opflags = 0;
145 if (sb->s_xattr)
146 inode->i_opflags |= IOP_XATTR;
147 i_uid_write(inode, 0);
148 i_gid_write(inode, 0);
149 atomic_set(&inode->i_writecount, 0);
150 inode->i_size = 0;
151 inode->i_write_hint = WRITE_LIFE_NOT_SET;
152 inode->i_blocks = 0;
153 inode->i_bytes = 0;
154 inode->i_generation = 0;
155 inode->i_pipe = NULL;
156 inode->i_bdev = NULL;
157 inode->i_cdev = NULL;
158 inode->i_link = NULL;
159 inode->i_dir_seq = 0;
160 inode->i_rdev = 0;
161 inode->dirtied_when = 0;
163 #ifdef CONFIG_CGROUP_WRITEBACK
164 inode->i_wb_frn_winner = 0;
165 inode->i_wb_frn_avg_time = 0;
166 inode->i_wb_frn_history = 0;
167 #endif
169 if (security_inode_alloc(inode))
170 goto out;
171 spin_lock_init(&inode->i_lock);
172 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
174 init_rwsem(&inode->i_rwsem);
175 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
177 atomic_set(&inode->i_dio_count, 0);
179 mapping->a_ops = &empty_aops;
180 mapping->host = inode;
181 mapping->flags = 0;
182 mapping->wb_err = 0;
183 atomic_set(&mapping->i_mmap_writable, 0);
184 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
185 mapping->private_data = NULL;
186 mapping->writeback_index = 0;
187 inode->i_private = NULL;
188 inode->i_mapping = mapping;
189 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
190 #ifdef CONFIG_FS_POSIX_ACL
191 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
192 #endif
194 #ifdef CONFIG_FSNOTIFY
195 inode->i_fsnotify_mask = 0;
196 #endif
197 inode->i_flctx = NULL;
198 this_cpu_inc(nr_inodes);
200 return 0;
201 out:
202 return -ENOMEM;
204 EXPORT_SYMBOL(inode_init_always);
206 static struct inode *alloc_inode(struct super_block *sb)
208 struct inode *inode;
210 if (sb->s_op->alloc_inode)
211 inode = sb->s_op->alloc_inode(sb);
212 else
213 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
215 if (!inode)
216 return NULL;
218 if (unlikely(inode_init_always(sb, inode))) {
219 if (inode->i_sb->s_op->destroy_inode)
220 inode->i_sb->s_op->destroy_inode(inode);
221 else
222 kmem_cache_free(inode_cachep, inode);
223 return NULL;
226 return inode;
229 void free_inode_nonrcu(struct inode *inode)
231 kmem_cache_free(inode_cachep, inode);
233 EXPORT_SYMBOL(free_inode_nonrcu);
235 void __destroy_inode(struct inode *inode)
237 BUG_ON(inode_has_buffers(inode));
238 inode_detach_wb(inode);
239 security_inode_free(inode);
240 fsnotify_inode_delete(inode);
241 locks_free_lock_context(inode);
242 if (!inode->i_nlink) {
243 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
244 atomic_long_dec(&inode->i_sb->s_remove_count);
247 #ifdef CONFIG_FS_POSIX_ACL
248 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
249 posix_acl_release(inode->i_acl);
250 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
251 posix_acl_release(inode->i_default_acl);
252 #endif
253 this_cpu_dec(nr_inodes);
255 EXPORT_SYMBOL(__destroy_inode);
257 static void i_callback(struct rcu_head *head)
259 struct inode *inode = container_of(head, struct inode, i_rcu);
260 kmem_cache_free(inode_cachep, inode);
263 static void destroy_inode(struct inode *inode)
265 BUG_ON(!list_empty(&inode->i_lru));
266 __destroy_inode(inode);
267 if (inode->i_sb->s_op->destroy_inode)
268 inode->i_sb->s_op->destroy_inode(inode);
269 else
270 call_rcu(&inode->i_rcu, i_callback);
274 * drop_nlink - directly drop an inode's link count
275 * @inode: inode
277 * This is a low-level filesystem helper to replace any
278 * direct filesystem manipulation of i_nlink. In cases
279 * where we are attempting to track writes to the
280 * filesystem, a decrement to zero means an imminent
281 * write when the file is truncated and actually unlinked
282 * on the filesystem.
284 void drop_nlink(struct inode *inode)
286 WARN_ON(inode->i_nlink == 0);
287 inode->__i_nlink--;
288 if (!inode->i_nlink)
289 atomic_long_inc(&inode->i_sb->s_remove_count);
291 EXPORT_SYMBOL(drop_nlink);
294 * clear_nlink - directly zero an inode's link count
295 * @inode: inode
297 * This is a low-level filesystem helper to replace any
298 * direct filesystem manipulation of i_nlink. See
299 * drop_nlink() for why we care about i_nlink hitting zero.
301 void clear_nlink(struct inode *inode)
303 if (inode->i_nlink) {
304 inode->__i_nlink = 0;
305 atomic_long_inc(&inode->i_sb->s_remove_count);
308 EXPORT_SYMBOL(clear_nlink);
311 * set_nlink - directly set an inode's link count
312 * @inode: inode
313 * @nlink: new nlink (should be non-zero)
315 * This is a low-level filesystem helper to replace any
316 * direct filesystem manipulation of i_nlink.
318 void set_nlink(struct inode *inode, unsigned int nlink)
320 if (!nlink) {
321 clear_nlink(inode);
322 } else {
323 /* Yes, some filesystems do change nlink from zero to one */
324 if (inode->i_nlink == 0)
325 atomic_long_dec(&inode->i_sb->s_remove_count);
327 inode->__i_nlink = nlink;
330 EXPORT_SYMBOL(set_nlink);
333 * inc_nlink - directly increment an inode's link count
334 * @inode: inode
336 * This is a low-level filesystem helper to replace any
337 * direct filesystem manipulation of i_nlink. Currently,
338 * it is only here for parity with dec_nlink().
340 void inc_nlink(struct inode *inode)
342 if (unlikely(inode->i_nlink == 0)) {
343 WARN_ON(!(inode->i_state & I_LINKABLE));
344 atomic_long_dec(&inode->i_sb->s_remove_count);
347 inode->__i_nlink++;
349 EXPORT_SYMBOL(inc_nlink);
351 static void __address_space_init_once(struct address_space *mapping)
353 INIT_RADIX_TREE(&mapping->i_pages, GFP_ATOMIC | __GFP_ACCOUNT);
354 init_rwsem(&mapping->i_mmap_rwsem);
355 INIT_LIST_HEAD(&mapping->private_list);
356 spin_lock_init(&mapping->private_lock);
357 mapping->i_mmap = RB_ROOT_CACHED;
360 void address_space_init_once(struct address_space *mapping)
362 memset(mapping, 0, sizeof(*mapping));
363 __address_space_init_once(mapping);
365 EXPORT_SYMBOL(address_space_init_once);
368 * These are initializations that only need to be done
369 * once, because the fields are idempotent across use
370 * of the inode, so let the slab aware of that.
372 void inode_init_once(struct inode *inode)
374 memset(inode, 0, sizeof(*inode));
375 INIT_HLIST_NODE(&inode->i_hash);
376 INIT_LIST_HEAD(&inode->i_devices);
377 INIT_LIST_HEAD(&inode->i_io_list);
378 INIT_LIST_HEAD(&inode->i_wb_list);
379 INIT_LIST_HEAD(&inode->i_lru);
380 __address_space_init_once(&inode->i_data);
381 i_size_ordered_init(inode);
383 EXPORT_SYMBOL(inode_init_once);
385 static void init_once(void *foo)
387 struct inode *inode = (struct inode *) foo;
389 inode_init_once(inode);
393 * inode->i_lock must be held
395 void __iget(struct inode *inode)
397 atomic_inc(&inode->i_count);
401 * get additional reference to inode; caller must already hold one.
403 void ihold(struct inode *inode)
405 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
407 EXPORT_SYMBOL(ihold);
409 static void inode_lru_list_add(struct inode *inode)
411 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
412 this_cpu_inc(nr_unused);
413 else
414 inode->i_state |= I_REFERENCED;
418 * Add inode to LRU if needed (inode is unused and clean).
420 * Needs inode->i_lock held.
422 void inode_add_lru(struct inode *inode)
424 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
425 I_FREEING | I_WILL_FREE)) &&
426 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
427 inode_lru_list_add(inode);
431 static void inode_lru_list_del(struct inode *inode)
434 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
435 this_cpu_dec(nr_unused);
439 * inode_sb_list_add - add inode to the superblock list of inodes
440 * @inode: inode to add
442 void inode_sb_list_add(struct inode *inode)
444 spin_lock(&inode->i_sb->s_inode_list_lock);
445 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
446 spin_unlock(&inode->i_sb->s_inode_list_lock);
448 EXPORT_SYMBOL_GPL(inode_sb_list_add);
450 static inline void inode_sb_list_del(struct inode *inode)
452 if (!list_empty(&inode->i_sb_list)) {
453 spin_lock(&inode->i_sb->s_inode_list_lock);
454 list_del_init(&inode->i_sb_list);
455 spin_unlock(&inode->i_sb->s_inode_list_lock);
459 static unsigned long hash(struct super_block *sb, unsigned long hashval)
461 unsigned long tmp;
463 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
464 L1_CACHE_BYTES;
465 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
466 return tmp & i_hash_mask;
470 * __insert_inode_hash - hash an inode
471 * @inode: unhashed inode
472 * @hashval: unsigned long value used to locate this object in the
473 * inode_hashtable.
475 * Add an inode to the inode hash for this superblock.
477 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
479 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
481 spin_lock(&inode_hash_lock);
482 spin_lock(&inode->i_lock);
483 hlist_add_head(&inode->i_hash, b);
484 spin_unlock(&inode->i_lock);
485 spin_unlock(&inode_hash_lock);
487 EXPORT_SYMBOL(__insert_inode_hash);
490 * __remove_inode_hash - remove an inode from the hash
491 * @inode: inode to unhash
493 * Remove an inode from the superblock.
495 void __remove_inode_hash(struct inode *inode)
497 spin_lock(&inode_hash_lock);
498 spin_lock(&inode->i_lock);
499 hlist_del_init(&inode->i_hash);
500 spin_unlock(&inode->i_lock);
501 spin_unlock(&inode_hash_lock);
503 EXPORT_SYMBOL(__remove_inode_hash);
505 void clear_inode(struct inode *inode)
508 * We have to cycle the i_pages lock here because reclaim can be in the
509 * process of removing the last page (in __delete_from_page_cache())
510 * and we must not free the mapping under it.
512 xa_lock_irq(&inode->i_data.i_pages);
513 BUG_ON(inode->i_data.nrpages);
514 BUG_ON(inode->i_data.nrexceptional);
515 xa_unlock_irq(&inode->i_data.i_pages);
516 BUG_ON(!list_empty(&inode->i_data.private_list));
517 BUG_ON(!(inode->i_state & I_FREEING));
518 BUG_ON(inode->i_state & I_CLEAR);
519 BUG_ON(!list_empty(&inode->i_wb_list));
520 /* don't need i_lock here, no concurrent mods to i_state */
521 inode->i_state = I_FREEING | I_CLEAR;
523 EXPORT_SYMBOL(clear_inode);
526 * Free the inode passed in, removing it from the lists it is still connected
527 * to. We remove any pages still attached to the inode and wait for any IO that
528 * is still in progress before finally destroying the inode.
530 * An inode must already be marked I_FREEING so that we avoid the inode being
531 * moved back onto lists if we race with other code that manipulates the lists
532 * (e.g. writeback_single_inode). The caller is responsible for setting this.
534 * An inode must already be removed from the LRU list before being evicted from
535 * the cache. This should occur atomically with setting the I_FREEING state
536 * flag, so no inodes here should ever be on the LRU when being evicted.
538 static void evict(struct inode *inode)
540 const struct super_operations *op = inode->i_sb->s_op;
542 BUG_ON(!(inode->i_state & I_FREEING));
543 BUG_ON(!list_empty(&inode->i_lru));
545 if (!list_empty(&inode->i_io_list))
546 inode_io_list_del(inode);
548 inode_sb_list_del(inode);
551 * Wait for flusher thread to be done with the inode so that filesystem
552 * does not start destroying it while writeback is still running. Since
553 * the inode has I_FREEING set, flusher thread won't start new work on
554 * the inode. We just have to wait for running writeback to finish.
556 inode_wait_for_writeback(inode);
558 if (op->evict_inode) {
559 op->evict_inode(inode);
560 } else {
561 truncate_inode_pages_final(&inode->i_data);
562 clear_inode(inode);
564 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
565 bd_forget(inode);
566 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
567 cd_forget(inode);
569 remove_inode_hash(inode);
571 spin_lock(&inode->i_lock);
572 wake_up_bit(&inode->i_state, __I_NEW);
573 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
574 spin_unlock(&inode->i_lock);
576 destroy_inode(inode);
580 * dispose_list - dispose of the contents of a local list
581 * @head: the head of the list to free
583 * Dispose-list gets a local list with local inodes in it, so it doesn't
584 * need to worry about list corruption and SMP locks.
586 static void dispose_list(struct list_head *head)
588 while (!list_empty(head)) {
589 struct inode *inode;
591 inode = list_first_entry(head, struct inode, i_lru);
592 list_del_init(&inode->i_lru);
594 evict(inode);
595 cond_resched();
600 * evict_inodes - evict all evictable inodes for a superblock
601 * @sb: superblock to operate on
603 * Make sure that no inodes with zero refcount are retained. This is
604 * called by superblock shutdown after having SB_ACTIVE flag removed,
605 * so any inode reaching zero refcount during or after that call will
606 * be immediately evicted.
608 void evict_inodes(struct super_block *sb)
610 struct inode *inode, *next;
611 LIST_HEAD(dispose);
613 again:
614 spin_lock(&sb->s_inode_list_lock);
615 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
616 if (atomic_read(&inode->i_count))
617 continue;
619 spin_lock(&inode->i_lock);
620 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
621 spin_unlock(&inode->i_lock);
622 continue;
625 inode->i_state |= I_FREEING;
626 inode_lru_list_del(inode);
627 spin_unlock(&inode->i_lock);
628 list_add(&inode->i_lru, &dispose);
631 * We can have a ton of inodes to evict at unmount time given
632 * enough memory, check to see if we need to go to sleep for a
633 * bit so we don't livelock.
635 if (need_resched()) {
636 spin_unlock(&sb->s_inode_list_lock);
637 cond_resched();
638 dispose_list(&dispose);
639 goto again;
642 spin_unlock(&sb->s_inode_list_lock);
644 dispose_list(&dispose);
646 EXPORT_SYMBOL_GPL(evict_inodes);
649 * invalidate_inodes - attempt to free all inodes on a superblock
650 * @sb: superblock to operate on
651 * @kill_dirty: flag to guide handling of dirty inodes
653 * Attempts to free all inodes for a given superblock. If there were any
654 * busy inodes return a non-zero value, else zero.
655 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
656 * them as busy.
658 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
660 int busy = 0;
661 struct inode *inode, *next;
662 LIST_HEAD(dispose);
664 again:
665 spin_lock(&sb->s_inode_list_lock);
666 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
667 spin_lock(&inode->i_lock);
668 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
669 spin_unlock(&inode->i_lock);
670 continue;
672 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
673 spin_unlock(&inode->i_lock);
674 busy = 1;
675 continue;
677 if (atomic_read(&inode->i_count)) {
678 spin_unlock(&inode->i_lock);
679 busy = 1;
680 continue;
683 inode->i_state |= I_FREEING;
684 inode_lru_list_del(inode);
685 spin_unlock(&inode->i_lock);
686 list_add(&inode->i_lru, &dispose);
687 if (need_resched()) {
688 spin_unlock(&sb->s_inode_list_lock);
689 cond_resched();
690 dispose_list(&dispose);
691 goto again;
694 spin_unlock(&sb->s_inode_list_lock);
696 dispose_list(&dispose);
698 return busy;
702 * Isolate the inode from the LRU in preparation for freeing it.
704 * Any inodes which are pinned purely because of attached pagecache have their
705 * pagecache removed. If the inode has metadata buffers attached to
706 * mapping->private_list then try to remove them.
708 * If the inode has the I_REFERENCED flag set, then it means that it has been
709 * used recently - the flag is set in iput_final(). When we encounter such an
710 * inode, clear the flag and move it to the back of the LRU so it gets another
711 * pass through the LRU before it gets reclaimed. This is necessary because of
712 * the fact we are doing lazy LRU updates to minimise lock contention so the
713 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
714 * with this flag set because they are the inodes that are out of order.
716 static enum lru_status inode_lru_isolate(struct list_head *item,
717 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
719 struct list_head *freeable = arg;
720 struct inode *inode = container_of(item, struct inode, i_lru);
723 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
724 * If we fail to get the lock, just skip it.
726 if (!spin_trylock(&inode->i_lock))
727 return LRU_SKIP;
730 * Referenced or dirty inodes are still in use. Give them another pass
731 * through the LRU as we canot reclaim them now.
733 if (atomic_read(&inode->i_count) ||
734 (inode->i_state & ~I_REFERENCED)) {
735 list_lru_isolate(lru, &inode->i_lru);
736 spin_unlock(&inode->i_lock);
737 this_cpu_dec(nr_unused);
738 return LRU_REMOVED;
741 /* recently referenced inodes get one more pass */
742 if (inode->i_state & I_REFERENCED) {
743 inode->i_state &= ~I_REFERENCED;
744 spin_unlock(&inode->i_lock);
745 return LRU_ROTATE;
748 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
749 __iget(inode);
750 spin_unlock(&inode->i_lock);
751 spin_unlock(lru_lock);
752 if (remove_inode_buffers(inode)) {
753 unsigned long reap;
754 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
755 if (current_is_kswapd())
756 __count_vm_events(KSWAPD_INODESTEAL, reap);
757 else
758 __count_vm_events(PGINODESTEAL, reap);
759 if (current->reclaim_state)
760 current->reclaim_state->reclaimed_slab += reap;
762 iput(inode);
763 spin_lock(lru_lock);
764 return LRU_RETRY;
767 WARN_ON(inode->i_state & I_NEW);
768 inode->i_state |= I_FREEING;
769 list_lru_isolate_move(lru, &inode->i_lru, freeable);
770 spin_unlock(&inode->i_lock);
772 this_cpu_dec(nr_unused);
773 return LRU_REMOVED;
777 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
778 * This is called from the superblock shrinker function with a number of inodes
779 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
780 * then are freed outside inode_lock by dispose_list().
782 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
784 LIST_HEAD(freeable);
785 long freed;
787 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
788 inode_lru_isolate, &freeable);
789 dispose_list(&freeable);
790 return freed;
793 static void __wait_on_freeing_inode(struct inode *inode);
795 * Called with the inode lock held.
797 static struct inode *find_inode(struct super_block *sb,
798 struct hlist_head *head,
799 int (*test)(struct inode *, void *),
800 void *data)
802 struct inode *inode = NULL;
804 repeat:
805 hlist_for_each_entry(inode, head, i_hash) {
806 if (inode->i_sb != sb)
807 continue;
808 if (!test(inode, data))
809 continue;
810 spin_lock(&inode->i_lock);
811 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
812 __wait_on_freeing_inode(inode);
813 goto repeat;
815 if (unlikely(inode->i_state & I_CREATING)) {
816 spin_unlock(&inode->i_lock);
817 return ERR_PTR(-ESTALE);
819 __iget(inode);
820 spin_unlock(&inode->i_lock);
821 return inode;
823 return NULL;
827 * find_inode_fast is the fast path version of find_inode, see the comment at
828 * iget_locked for details.
830 static struct inode *find_inode_fast(struct super_block *sb,
831 struct hlist_head *head, unsigned long ino)
833 struct inode *inode = NULL;
835 repeat:
836 hlist_for_each_entry(inode, head, i_hash) {
837 if (inode->i_ino != ino)
838 continue;
839 if (inode->i_sb != sb)
840 continue;
841 spin_lock(&inode->i_lock);
842 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
843 __wait_on_freeing_inode(inode);
844 goto repeat;
846 if (unlikely(inode->i_state & I_CREATING)) {
847 spin_unlock(&inode->i_lock);
848 return ERR_PTR(-ESTALE);
850 __iget(inode);
851 spin_unlock(&inode->i_lock);
852 return inode;
854 return NULL;
858 * Each cpu owns a range of LAST_INO_BATCH numbers.
859 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
860 * to renew the exhausted range.
862 * This does not significantly increase overflow rate because every CPU can
863 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
864 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
865 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
866 * overflow rate by 2x, which does not seem too significant.
868 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
869 * error if st_ino won't fit in target struct field. Use 32bit counter
870 * here to attempt to avoid that.
872 #define LAST_INO_BATCH 1024
873 static DEFINE_PER_CPU(unsigned int, last_ino);
875 unsigned int get_next_ino(void)
877 unsigned int *p = &get_cpu_var(last_ino);
878 unsigned int res = *p;
880 #ifdef CONFIG_SMP
881 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
882 static atomic_t shared_last_ino;
883 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
885 res = next - LAST_INO_BATCH;
887 #endif
889 res++;
890 /* get_next_ino should not provide a 0 inode number */
891 if (unlikely(!res))
892 res++;
893 *p = res;
894 put_cpu_var(last_ino);
895 return res;
897 EXPORT_SYMBOL(get_next_ino);
900 * new_inode_pseudo - obtain an inode
901 * @sb: superblock
903 * Allocates a new inode for given superblock.
904 * Inode wont be chained in superblock s_inodes list
905 * This means :
906 * - fs can't be unmount
907 * - quotas, fsnotify, writeback can't work
909 struct inode *new_inode_pseudo(struct super_block *sb)
911 struct inode *inode = alloc_inode(sb);
913 if (inode) {
914 spin_lock(&inode->i_lock);
915 inode->i_state = 0;
916 spin_unlock(&inode->i_lock);
917 INIT_LIST_HEAD(&inode->i_sb_list);
919 return inode;
923 * new_inode - obtain an inode
924 * @sb: superblock
926 * Allocates a new inode for given superblock. The default gfp_mask
927 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
928 * If HIGHMEM pages are unsuitable or it is known that pages allocated
929 * for the page cache are not reclaimable or migratable,
930 * mapping_set_gfp_mask() must be called with suitable flags on the
931 * newly created inode's mapping
934 struct inode *new_inode(struct super_block *sb)
936 struct inode *inode;
938 spin_lock_prefetch(&sb->s_inode_list_lock);
940 inode = new_inode_pseudo(sb);
941 if (inode)
942 inode_sb_list_add(inode);
943 return inode;
945 EXPORT_SYMBOL(new_inode);
947 #ifdef CONFIG_DEBUG_LOCK_ALLOC
948 void lockdep_annotate_inode_mutex_key(struct inode *inode)
950 if (S_ISDIR(inode->i_mode)) {
951 struct file_system_type *type = inode->i_sb->s_type;
953 /* Set new key only if filesystem hasn't already changed it */
954 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
956 * ensure nobody is actually holding i_mutex
958 // mutex_destroy(&inode->i_mutex);
959 init_rwsem(&inode->i_rwsem);
960 lockdep_set_class(&inode->i_rwsem,
961 &type->i_mutex_dir_key);
965 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
966 #endif
969 * unlock_new_inode - clear the I_NEW state and wake up any waiters
970 * @inode: new inode to unlock
972 * Called when the inode is fully initialised to clear the new state of the
973 * inode and wake up anyone waiting for the inode to finish initialisation.
975 void unlock_new_inode(struct inode *inode)
977 lockdep_annotate_inode_mutex_key(inode);
978 spin_lock(&inode->i_lock);
979 WARN_ON(!(inode->i_state & I_NEW));
980 inode->i_state &= ~I_NEW & ~I_CREATING;
981 smp_mb();
982 wake_up_bit(&inode->i_state, __I_NEW);
983 spin_unlock(&inode->i_lock);
985 EXPORT_SYMBOL(unlock_new_inode);
987 void discard_new_inode(struct inode *inode)
989 lockdep_annotate_inode_mutex_key(inode);
990 spin_lock(&inode->i_lock);
991 WARN_ON(!(inode->i_state & I_NEW));
992 inode->i_state &= ~I_NEW;
993 smp_mb();
994 wake_up_bit(&inode->i_state, __I_NEW);
995 spin_unlock(&inode->i_lock);
996 iput(inode);
998 EXPORT_SYMBOL(discard_new_inode);
1001 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1003 * Lock any non-NULL argument that is not a directory.
1004 * Zero, one or two objects may be locked by this function.
1006 * @inode1: first inode to lock
1007 * @inode2: second inode to lock
1009 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1011 if (inode1 > inode2)
1012 swap(inode1, inode2);
1014 if (inode1 && !S_ISDIR(inode1->i_mode))
1015 inode_lock(inode1);
1016 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1017 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1019 EXPORT_SYMBOL(lock_two_nondirectories);
1022 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1023 * @inode1: first inode to unlock
1024 * @inode2: second inode to unlock
1026 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1028 if (inode1 && !S_ISDIR(inode1->i_mode))
1029 inode_unlock(inode1);
1030 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1031 inode_unlock(inode2);
1033 EXPORT_SYMBOL(unlock_two_nondirectories);
1036 * inode_insert5 - obtain an inode from a mounted file system
1037 * @inode: pre-allocated inode to use for insert to cache
1038 * @hashval: hash value (usually inode number) to get
1039 * @test: callback used for comparisons between inodes
1040 * @set: callback used to initialize a new struct inode
1041 * @data: opaque data pointer to pass to @test and @set
1043 * Search for the inode specified by @hashval and @data in the inode cache,
1044 * and if present it is return it with an increased reference count. This is
1045 * a variant of iget5_locked() for callers that don't want to fail on memory
1046 * allocation of inode.
1048 * If the inode is not in cache, insert the pre-allocated inode to cache and
1049 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1050 * to fill it in before unlocking it via unlock_new_inode().
1052 * Note both @test and @set are called with the inode_hash_lock held, so can't
1053 * sleep.
1055 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1056 int (*test)(struct inode *, void *),
1057 int (*set)(struct inode *, void *), void *data)
1059 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1060 struct inode *old;
1061 bool creating = inode->i_state & I_CREATING;
1063 again:
1064 spin_lock(&inode_hash_lock);
1065 old = find_inode(inode->i_sb, head, test, data);
1066 if (unlikely(old)) {
1068 * Uhhuh, somebody else created the same inode under us.
1069 * Use the old inode instead of the preallocated one.
1071 spin_unlock(&inode_hash_lock);
1072 if (IS_ERR(old))
1073 return NULL;
1074 wait_on_inode(old);
1075 if (unlikely(inode_unhashed(old))) {
1076 iput(old);
1077 goto again;
1079 return old;
1082 if (set && unlikely(set(inode, data))) {
1083 inode = NULL;
1084 goto unlock;
1088 * Return the locked inode with I_NEW set, the
1089 * caller is responsible for filling in the contents
1091 spin_lock(&inode->i_lock);
1092 inode->i_state |= I_NEW;
1093 hlist_add_head(&inode->i_hash, head);
1094 spin_unlock(&inode->i_lock);
1095 if (!creating)
1096 inode_sb_list_add(inode);
1097 unlock:
1098 spin_unlock(&inode_hash_lock);
1100 return inode;
1102 EXPORT_SYMBOL(inode_insert5);
1105 * iget5_locked - obtain an inode from a mounted file system
1106 * @sb: super block of file system
1107 * @hashval: hash value (usually inode number) to get
1108 * @test: callback used for comparisons between inodes
1109 * @set: callback used to initialize a new struct inode
1110 * @data: opaque data pointer to pass to @test and @set
1112 * Search for the inode specified by @hashval and @data in the inode cache,
1113 * and if present it is return it with an increased reference count. This is
1114 * a generalized version of iget_locked() for file systems where the inode
1115 * number is not sufficient for unique identification of an inode.
1117 * If the inode is not in cache, allocate a new inode and return it locked,
1118 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1119 * before unlocking it via unlock_new_inode().
1121 * Note both @test and @set are called with the inode_hash_lock held, so can't
1122 * sleep.
1124 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1125 int (*test)(struct inode *, void *),
1126 int (*set)(struct inode *, void *), void *data)
1128 struct inode *inode = ilookup5(sb, hashval, test, data);
1130 if (!inode) {
1131 struct inode *new = alloc_inode(sb);
1133 if (new) {
1134 new->i_state = 0;
1135 inode = inode_insert5(new, hashval, test, set, data);
1136 if (unlikely(inode != new))
1137 destroy_inode(new);
1140 return inode;
1142 EXPORT_SYMBOL(iget5_locked);
1145 * iget_locked - obtain an inode from a mounted file system
1146 * @sb: super block of file system
1147 * @ino: inode number to get
1149 * Search for the inode specified by @ino in the inode cache and if present
1150 * return it with an increased reference count. This is for file systems
1151 * where the inode number is sufficient for unique identification of an inode.
1153 * If the inode is not in cache, allocate a new inode and return it locked,
1154 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1155 * before unlocking it via unlock_new_inode().
1157 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1159 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1160 struct inode *inode;
1161 again:
1162 spin_lock(&inode_hash_lock);
1163 inode = find_inode_fast(sb, head, ino);
1164 spin_unlock(&inode_hash_lock);
1165 if (inode) {
1166 if (IS_ERR(inode))
1167 return NULL;
1168 wait_on_inode(inode);
1169 if (unlikely(inode_unhashed(inode))) {
1170 iput(inode);
1171 goto again;
1173 return inode;
1176 inode = alloc_inode(sb);
1177 if (inode) {
1178 struct inode *old;
1180 spin_lock(&inode_hash_lock);
1181 /* We released the lock, so.. */
1182 old = find_inode_fast(sb, head, ino);
1183 if (!old) {
1184 inode->i_ino = ino;
1185 spin_lock(&inode->i_lock);
1186 inode->i_state = I_NEW;
1187 hlist_add_head(&inode->i_hash, head);
1188 spin_unlock(&inode->i_lock);
1189 inode_sb_list_add(inode);
1190 spin_unlock(&inode_hash_lock);
1192 /* Return the locked inode with I_NEW set, the
1193 * caller is responsible for filling in the contents
1195 return inode;
1199 * Uhhuh, somebody else created the same inode under
1200 * us. Use the old inode instead of the one we just
1201 * allocated.
1203 spin_unlock(&inode_hash_lock);
1204 destroy_inode(inode);
1205 if (IS_ERR(old))
1206 return NULL;
1207 inode = old;
1208 wait_on_inode(inode);
1209 if (unlikely(inode_unhashed(inode))) {
1210 iput(inode);
1211 goto again;
1214 return inode;
1216 EXPORT_SYMBOL(iget_locked);
1219 * search the inode cache for a matching inode number.
1220 * If we find one, then the inode number we are trying to
1221 * allocate is not unique and so we should not use it.
1223 * Returns 1 if the inode number is unique, 0 if it is not.
1225 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1227 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1228 struct inode *inode;
1230 spin_lock(&inode_hash_lock);
1231 hlist_for_each_entry(inode, b, i_hash) {
1232 if (inode->i_ino == ino && inode->i_sb == sb) {
1233 spin_unlock(&inode_hash_lock);
1234 return 0;
1237 spin_unlock(&inode_hash_lock);
1239 return 1;
1243 * iunique - get a unique inode number
1244 * @sb: superblock
1245 * @max_reserved: highest reserved inode number
1247 * Obtain an inode number that is unique on the system for a given
1248 * superblock. This is used by file systems that have no natural
1249 * permanent inode numbering system. An inode number is returned that
1250 * is higher than the reserved limit but unique.
1252 * BUGS:
1253 * With a large number of inodes live on the file system this function
1254 * currently becomes quite slow.
1256 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1259 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1260 * error if st_ino won't fit in target struct field. Use 32bit counter
1261 * here to attempt to avoid that.
1263 static DEFINE_SPINLOCK(iunique_lock);
1264 static unsigned int counter;
1265 ino_t res;
1267 spin_lock(&iunique_lock);
1268 do {
1269 if (counter <= max_reserved)
1270 counter = max_reserved + 1;
1271 res = counter++;
1272 } while (!test_inode_iunique(sb, res));
1273 spin_unlock(&iunique_lock);
1275 return res;
1277 EXPORT_SYMBOL(iunique);
1279 struct inode *igrab(struct inode *inode)
1281 spin_lock(&inode->i_lock);
1282 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1283 __iget(inode);
1284 spin_unlock(&inode->i_lock);
1285 } else {
1286 spin_unlock(&inode->i_lock);
1288 * Handle the case where s_op->clear_inode is not been
1289 * called yet, and somebody is calling igrab
1290 * while the inode is getting freed.
1292 inode = NULL;
1294 return inode;
1296 EXPORT_SYMBOL(igrab);
1299 * ilookup5_nowait - search for an inode in the inode cache
1300 * @sb: super block of file system to search
1301 * @hashval: hash value (usually inode number) to search for
1302 * @test: callback used for comparisons between inodes
1303 * @data: opaque data pointer to pass to @test
1305 * Search for the inode specified by @hashval and @data in the inode cache.
1306 * If the inode is in the cache, the inode is returned with an incremented
1307 * reference count.
1309 * Note: I_NEW is not waited upon so you have to be very careful what you do
1310 * with the returned inode. You probably should be using ilookup5() instead.
1312 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1314 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1315 int (*test)(struct inode *, void *), void *data)
1317 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1318 struct inode *inode;
1320 spin_lock(&inode_hash_lock);
1321 inode = find_inode(sb, head, test, data);
1322 spin_unlock(&inode_hash_lock);
1324 return IS_ERR(inode) ? NULL : inode;
1326 EXPORT_SYMBOL(ilookup5_nowait);
1329 * ilookup5 - search for an inode in the inode cache
1330 * @sb: super block of file system to search
1331 * @hashval: hash value (usually inode number) to search for
1332 * @test: callback used for comparisons between inodes
1333 * @data: opaque data pointer to pass to @test
1335 * Search for the inode specified by @hashval and @data in the inode cache,
1336 * and if the inode is in the cache, return the inode with an incremented
1337 * reference count. Waits on I_NEW before returning the inode.
1338 * returned with an incremented reference count.
1340 * This is a generalized version of ilookup() for file systems where the
1341 * inode number is not sufficient for unique identification of an inode.
1343 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1345 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1346 int (*test)(struct inode *, void *), void *data)
1348 struct inode *inode;
1349 again:
1350 inode = ilookup5_nowait(sb, hashval, test, data);
1351 if (inode) {
1352 wait_on_inode(inode);
1353 if (unlikely(inode_unhashed(inode))) {
1354 iput(inode);
1355 goto again;
1358 return inode;
1360 EXPORT_SYMBOL(ilookup5);
1363 * ilookup - search for an inode in the inode cache
1364 * @sb: super block of file system to search
1365 * @ino: inode number to search for
1367 * Search for the inode @ino in the inode cache, and if the inode is in the
1368 * cache, the inode is returned with an incremented reference count.
1370 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1372 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1373 struct inode *inode;
1374 again:
1375 spin_lock(&inode_hash_lock);
1376 inode = find_inode_fast(sb, head, ino);
1377 spin_unlock(&inode_hash_lock);
1379 if (inode) {
1380 if (IS_ERR(inode))
1381 return NULL;
1382 wait_on_inode(inode);
1383 if (unlikely(inode_unhashed(inode))) {
1384 iput(inode);
1385 goto again;
1388 return inode;
1390 EXPORT_SYMBOL(ilookup);
1393 * find_inode_nowait - find an inode in the inode cache
1394 * @sb: super block of file system to search
1395 * @hashval: hash value (usually inode number) to search for
1396 * @match: callback used for comparisons between inodes
1397 * @data: opaque data pointer to pass to @match
1399 * Search for the inode specified by @hashval and @data in the inode
1400 * cache, where the helper function @match will return 0 if the inode
1401 * does not match, 1 if the inode does match, and -1 if the search
1402 * should be stopped. The @match function must be responsible for
1403 * taking the i_lock spin_lock and checking i_state for an inode being
1404 * freed or being initialized, and incrementing the reference count
1405 * before returning 1. It also must not sleep, since it is called with
1406 * the inode_hash_lock spinlock held.
1408 * This is a even more generalized version of ilookup5() when the
1409 * function must never block --- find_inode() can block in
1410 * __wait_on_freeing_inode() --- or when the caller can not increment
1411 * the reference count because the resulting iput() might cause an
1412 * inode eviction. The tradeoff is that the @match funtion must be
1413 * very carefully implemented.
1415 struct inode *find_inode_nowait(struct super_block *sb,
1416 unsigned long hashval,
1417 int (*match)(struct inode *, unsigned long,
1418 void *),
1419 void *data)
1421 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1422 struct inode *inode, *ret_inode = NULL;
1423 int mval;
1425 spin_lock(&inode_hash_lock);
1426 hlist_for_each_entry(inode, head, i_hash) {
1427 if (inode->i_sb != sb)
1428 continue;
1429 mval = match(inode, hashval, data);
1430 if (mval == 0)
1431 continue;
1432 if (mval == 1)
1433 ret_inode = inode;
1434 goto out;
1436 out:
1437 spin_unlock(&inode_hash_lock);
1438 return ret_inode;
1440 EXPORT_SYMBOL(find_inode_nowait);
1442 int insert_inode_locked(struct inode *inode)
1444 struct super_block *sb = inode->i_sb;
1445 ino_t ino = inode->i_ino;
1446 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1448 while (1) {
1449 struct inode *old = NULL;
1450 spin_lock(&inode_hash_lock);
1451 hlist_for_each_entry(old, head, i_hash) {
1452 if (old->i_ino != ino)
1453 continue;
1454 if (old->i_sb != sb)
1455 continue;
1456 spin_lock(&old->i_lock);
1457 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1458 spin_unlock(&old->i_lock);
1459 continue;
1461 break;
1463 if (likely(!old)) {
1464 spin_lock(&inode->i_lock);
1465 inode->i_state |= I_NEW | I_CREATING;
1466 hlist_add_head(&inode->i_hash, head);
1467 spin_unlock(&inode->i_lock);
1468 spin_unlock(&inode_hash_lock);
1469 return 0;
1471 if (unlikely(old->i_state & I_CREATING)) {
1472 spin_unlock(&old->i_lock);
1473 spin_unlock(&inode_hash_lock);
1474 return -EBUSY;
1476 __iget(old);
1477 spin_unlock(&old->i_lock);
1478 spin_unlock(&inode_hash_lock);
1479 wait_on_inode(old);
1480 if (unlikely(!inode_unhashed(old))) {
1481 iput(old);
1482 return -EBUSY;
1484 iput(old);
1487 EXPORT_SYMBOL(insert_inode_locked);
1489 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1490 int (*test)(struct inode *, void *), void *data)
1492 struct inode *old;
1494 inode->i_state |= I_CREATING;
1495 old = inode_insert5(inode, hashval, test, NULL, data);
1497 if (old != inode) {
1498 iput(old);
1499 return -EBUSY;
1501 return 0;
1503 EXPORT_SYMBOL(insert_inode_locked4);
1506 int generic_delete_inode(struct inode *inode)
1508 return 1;
1510 EXPORT_SYMBOL(generic_delete_inode);
1513 * Called when we're dropping the last reference
1514 * to an inode.
1516 * Call the FS "drop_inode()" function, defaulting to
1517 * the legacy UNIX filesystem behaviour. If it tells
1518 * us to evict inode, do so. Otherwise, retain inode
1519 * in cache if fs is alive, sync and evict if fs is
1520 * shutting down.
1522 static void iput_final(struct inode *inode)
1524 struct super_block *sb = inode->i_sb;
1525 const struct super_operations *op = inode->i_sb->s_op;
1526 int drop;
1528 WARN_ON(inode->i_state & I_NEW);
1530 if (op->drop_inode)
1531 drop = op->drop_inode(inode);
1532 else
1533 drop = generic_drop_inode(inode);
1535 if (!drop && (sb->s_flags & SB_ACTIVE)) {
1536 inode_add_lru(inode);
1537 spin_unlock(&inode->i_lock);
1538 return;
1541 if (!drop) {
1542 inode->i_state |= I_WILL_FREE;
1543 spin_unlock(&inode->i_lock);
1544 write_inode_now(inode, 1);
1545 spin_lock(&inode->i_lock);
1546 WARN_ON(inode->i_state & I_NEW);
1547 inode->i_state &= ~I_WILL_FREE;
1550 inode->i_state |= I_FREEING;
1551 if (!list_empty(&inode->i_lru))
1552 inode_lru_list_del(inode);
1553 spin_unlock(&inode->i_lock);
1555 evict(inode);
1559 * iput - put an inode
1560 * @inode: inode to put
1562 * Puts an inode, dropping its usage count. If the inode use count hits
1563 * zero, the inode is then freed and may also be destroyed.
1565 * Consequently, iput() can sleep.
1567 void iput(struct inode *inode)
1569 if (!inode)
1570 return;
1571 BUG_ON(inode->i_state & I_CLEAR);
1572 retry:
1573 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1574 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1575 atomic_inc(&inode->i_count);
1576 spin_unlock(&inode->i_lock);
1577 trace_writeback_lazytime_iput(inode);
1578 mark_inode_dirty_sync(inode);
1579 goto retry;
1581 iput_final(inode);
1584 EXPORT_SYMBOL(iput);
1587 * bmap - find a block number in a file
1588 * @inode: inode of file
1589 * @block: block to find
1591 * Returns the block number on the device holding the inode that
1592 * is the disk block number for the block of the file requested.
1593 * That is, asked for block 4 of inode 1 the function will return the
1594 * disk block relative to the disk start that holds that block of the
1595 * file.
1597 sector_t bmap(struct inode *inode, sector_t block)
1599 sector_t res = 0;
1600 if (inode->i_mapping->a_ops->bmap)
1601 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1602 return res;
1604 EXPORT_SYMBOL(bmap);
1607 * With relative atime, only update atime if the previous atime is
1608 * earlier than either the ctime or mtime or if at least a day has
1609 * passed since the last atime update.
1611 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1612 struct timespec now)
1615 if (!(mnt->mnt_flags & MNT_RELATIME))
1616 return 1;
1618 * Is mtime younger than atime? If yes, update atime:
1620 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1621 return 1;
1623 * Is ctime younger than atime? If yes, update atime:
1625 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1626 return 1;
1629 * Is the previous atime value older than a day? If yes,
1630 * update atime:
1632 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1633 return 1;
1635 * Good, we can skip the atime update:
1637 return 0;
1640 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1642 int iflags = I_DIRTY_TIME;
1643 bool dirty = false;
1645 if (flags & S_ATIME)
1646 inode->i_atime = *time;
1647 if (flags & S_VERSION)
1648 dirty = inode_maybe_inc_iversion(inode, false);
1649 if (flags & S_CTIME)
1650 inode->i_ctime = *time;
1651 if (flags & S_MTIME)
1652 inode->i_mtime = *time;
1653 if ((flags & (S_ATIME | S_CTIME | S_MTIME)) &&
1654 !(inode->i_sb->s_flags & SB_LAZYTIME))
1655 dirty = true;
1657 if (dirty)
1658 iflags |= I_DIRTY_SYNC;
1659 __mark_inode_dirty(inode, iflags);
1660 return 0;
1662 EXPORT_SYMBOL(generic_update_time);
1665 * This does the actual work of updating an inodes time or version. Must have
1666 * had called mnt_want_write() before calling this.
1668 static int update_time(struct inode *inode, struct timespec64 *time, int flags)
1670 int (*update_time)(struct inode *, struct timespec64 *, int);
1672 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1673 generic_update_time;
1675 return update_time(inode, time, flags);
1679 * touch_atime - update the access time
1680 * @path: the &struct path to update
1681 * @inode: inode to update
1683 * Update the accessed time on an inode and mark it for writeback.
1684 * This function automatically handles read only file systems and media,
1685 * as well as the "noatime" flag and inode specific "noatime" markers.
1687 bool atime_needs_update(const struct path *path, struct inode *inode)
1689 struct vfsmount *mnt = path->mnt;
1690 struct timespec64 now;
1692 if (inode->i_flags & S_NOATIME)
1693 return false;
1695 /* Atime updates will likely cause i_uid and i_gid to be written
1696 * back improprely if their true value is unknown to the vfs.
1698 if (HAS_UNMAPPED_ID(inode))
1699 return false;
1701 if (IS_NOATIME(inode))
1702 return false;
1703 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1704 return false;
1706 if (mnt->mnt_flags & MNT_NOATIME)
1707 return false;
1708 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1709 return false;
1711 now = current_time(inode);
1713 if (!relatime_need_update(mnt, inode, timespec64_to_timespec(now)))
1714 return false;
1716 if (timespec64_equal(&inode->i_atime, &now))
1717 return false;
1719 return true;
1722 void touch_atime(const struct path *path)
1724 struct vfsmount *mnt = path->mnt;
1725 struct inode *inode = d_inode(path->dentry);
1726 struct timespec64 now;
1728 if (!atime_needs_update(path, inode))
1729 return;
1731 if (!sb_start_write_trylock(inode->i_sb))
1732 return;
1734 if (__mnt_want_write(mnt) != 0)
1735 goto skip_update;
1737 * File systems can error out when updating inodes if they need to
1738 * allocate new space to modify an inode (such is the case for
1739 * Btrfs), but since we touch atime while walking down the path we
1740 * really don't care if we failed to update the atime of the file,
1741 * so just ignore the return value.
1742 * We may also fail on filesystems that have the ability to make parts
1743 * of the fs read only, e.g. subvolumes in Btrfs.
1745 now = current_time(inode);
1746 update_time(inode, &now, S_ATIME);
1747 __mnt_drop_write(mnt);
1748 skip_update:
1749 sb_end_write(inode->i_sb);
1751 EXPORT_SYMBOL(touch_atime);
1754 * The logic we want is
1756 * if suid or (sgid and xgrp)
1757 * remove privs
1759 int should_remove_suid(struct dentry *dentry)
1761 umode_t mode = d_inode(dentry)->i_mode;
1762 int kill = 0;
1764 /* suid always must be killed */
1765 if (unlikely(mode & S_ISUID))
1766 kill = ATTR_KILL_SUID;
1769 * sgid without any exec bits is just a mandatory locking mark; leave
1770 * it alone. If some exec bits are set, it's a real sgid; kill it.
1772 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1773 kill |= ATTR_KILL_SGID;
1775 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1776 return kill;
1778 return 0;
1780 EXPORT_SYMBOL(should_remove_suid);
1783 * Return mask of changes for notify_change() that need to be done as a
1784 * response to write or truncate. Return 0 if nothing has to be changed.
1785 * Negative value on error (change should be denied).
1787 int dentry_needs_remove_privs(struct dentry *dentry)
1789 struct inode *inode = d_inode(dentry);
1790 int mask = 0;
1791 int ret;
1793 if (IS_NOSEC(inode))
1794 return 0;
1796 mask = should_remove_suid(dentry);
1797 ret = security_inode_need_killpriv(dentry);
1798 if (ret < 0)
1799 return ret;
1800 if (ret)
1801 mask |= ATTR_KILL_PRIV;
1802 return mask;
1805 static int __remove_privs(struct dentry *dentry, int kill)
1807 struct iattr newattrs;
1809 newattrs.ia_valid = ATTR_FORCE | kill;
1811 * Note we call this on write, so notify_change will not
1812 * encounter any conflicting delegations:
1814 return notify_change(dentry, &newattrs, NULL);
1818 * Remove special file priviledges (suid, capabilities) when file is written
1819 * to or truncated.
1821 int file_remove_privs(struct file *file)
1823 struct dentry *dentry = file_dentry(file);
1824 struct inode *inode = file_inode(file);
1825 int kill;
1826 int error = 0;
1829 * Fast path for nothing security related.
1830 * As well for non-regular files, e.g. blkdev inodes.
1831 * For example, blkdev_write_iter() might get here
1832 * trying to remove privs which it is not allowed to.
1834 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1835 return 0;
1837 kill = dentry_needs_remove_privs(dentry);
1838 if (kill < 0)
1839 return kill;
1840 if (kill)
1841 error = __remove_privs(dentry, kill);
1842 if (!error)
1843 inode_has_no_xattr(inode);
1845 return error;
1847 EXPORT_SYMBOL(file_remove_privs);
1850 * file_update_time - update mtime and ctime time
1851 * @file: file accessed
1853 * Update the mtime and ctime members of an inode and mark the inode
1854 * for writeback. Note that this function is meant exclusively for
1855 * usage in the file write path of filesystems, and filesystems may
1856 * choose to explicitly ignore update via this function with the
1857 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1858 * timestamps are handled by the server. This can return an error for
1859 * file systems who need to allocate space in order to update an inode.
1862 int file_update_time(struct file *file)
1864 struct inode *inode = file_inode(file);
1865 struct timespec64 now;
1866 int sync_it = 0;
1867 int ret;
1869 /* First try to exhaust all avenues to not sync */
1870 if (IS_NOCMTIME(inode))
1871 return 0;
1873 now = current_time(inode);
1874 if (!timespec64_equal(&inode->i_mtime, &now))
1875 sync_it = S_MTIME;
1877 if (!timespec64_equal(&inode->i_ctime, &now))
1878 sync_it |= S_CTIME;
1880 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
1881 sync_it |= S_VERSION;
1883 if (!sync_it)
1884 return 0;
1886 /* Finally allowed to write? Takes lock. */
1887 if (__mnt_want_write_file(file))
1888 return 0;
1890 ret = update_time(inode, &now, sync_it);
1891 __mnt_drop_write_file(file);
1893 return ret;
1895 EXPORT_SYMBOL(file_update_time);
1897 int inode_needs_sync(struct inode *inode)
1899 if (IS_SYNC(inode))
1900 return 1;
1901 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1902 return 1;
1903 return 0;
1905 EXPORT_SYMBOL(inode_needs_sync);
1908 * If we try to find an inode in the inode hash while it is being
1909 * deleted, we have to wait until the filesystem completes its
1910 * deletion before reporting that it isn't found. This function waits
1911 * until the deletion _might_ have completed. Callers are responsible
1912 * to recheck inode state.
1914 * It doesn't matter if I_NEW is not set initially, a call to
1915 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1916 * will DTRT.
1918 static void __wait_on_freeing_inode(struct inode *inode)
1920 wait_queue_head_t *wq;
1921 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1922 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1923 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1924 spin_unlock(&inode->i_lock);
1925 spin_unlock(&inode_hash_lock);
1926 schedule();
1927 finish_wait(wq, &wait.wq_entry);
1928 spin_lock(&inode_hash_lock);
1931 static __initdata unsigned long ihash_entries;
1932 static int __init set_ihash_entries(char *str)
1934 if (!str)
1935 return 0;
1936 ihash_entries = simple_strtoul(str, &str, 0);
1937 return 1;
1939 __setup("ihash_entries=", set_ihash_entries);
1942 * Initialize the waitqueues and inode hash table.
1944 void __init inode_init_early(void)
1946 /* If hashes are distributed across NUMA nodes, defer
1947 * hash allocation until vmalloc space is available.
1949 if (hashdist)
1950 return;
1952 inode_hashtable =
1953 alloc_large_system_hash("Inode-cache",
1954 sizeof(struct hlist_head),
1955 ihash_entries,
1957 HASH_EARLY | HASH_ZERO,
1958 &i_hash_shift,
1959 &i_hash_mask,
1964 void __init inode_init(void)
1966 /* inode slab cache */
1967 inode_cachep = kmem_cache_create("inode_cache",
1968 sizeof(struct inode),
1970 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1971 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1972 init_once);
1974 /* Hash may have been set up in inode_init_early */
1975 if (!hashdist)
1976 return;
1978 inode_hashtable =
1979 alloc_large_system_hash("Inode-cache",
1980 sizeof(struct hlist_head),
1981 ihash_entries,
1983 HASH_ZERO,
1984 &i_hash_shift,
1985 &i_hash_mask,
1990 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1992 inode->i_mode = mode;
1993 if (S_ISCHR(mode)) {
1994 inode->i_fop = &def_chr_fops;
1995 inode->i_rdev = rdev;
1996 } else if (S_ISBLK(mode)) {
1997 inode->i_fop = &def_blk_fops;
1998 inode->i_rdev = rdev;
1999 } else if (S_ISFIFO(mode))
2000 inode->i_fop = &pipefifo_fops;
2001 else if (S_ISSOCK(mode))
2002 ; /* leave it no_open_fops */
2003 else
2004 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2005 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2006 inode->i_ino);
2008 EXPORT_SYMBOL(init_special_inode);
2011 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2012 * @inode: New inode
2013 * @dir: Directory inode
2014 * @mode: mode of the new inode
2016 void inode_init_owner(struct inode *inode, const struct inode *dir,
2017 umode_t mode)
2019 inode->i_uid = current_fsuid();
2020 if (dir && dir->i_mode & S_ISGID) {
2021 inode->i_gid = dir->i_gid;
2023 /* Directories are special, and always inherit S_ISGID */
2024 if (S_ISDIR(mode))
2025 mode |= S_ISGID;
2026 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2027 !in_group_p(inode->i_gid) &&
2028 !capable_wrt_inode_uidgid(dir, CAP_FSETID))
2029 mode &= ~S_ISGID;
2030 } else
2031 inode->i_gid = current_fsgid();
2032 inode->i_mode = mode;
2034 EXPORT_SYMBOL(inode_init_owner);
2037 * inode_owner_or_capable - check current task permissions to inode
2038 * @inode: inode being checked
2040 * Return true if current either has CAP_FOWNER in a namespace with the
2041 * inode owner uid mapped, or owns the file.
2043 bool inode_owner_or_capable(const struct inode *inode)
2045 struct user_namespace *ns;
2047 if (uid_eq(current_fsuid(), inode->i_uid))
2048 return true;
2050 ns = current_user_ns();
2051 if (kuid_has_mapping(ns, inode->i_uid) && ns_capable(ns, CAP_FOWNER))
2052 return true;
2053 return false;
2055 EXPORT_SYMBOL(inode_owner_or_capable);
2058 * Direct i/o helper functions
2060 static void __inode_dio_wait(struct inode *inode)
2062 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2063 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2065 do {
2066 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2067 if (atomic_read(&inode->i_dio_count))
2068 schedule();
2069 } while (atomic_read(&inode->i_dio_count));
2070 finish_wait(wq, &q.wq_entry);
2074 * inode_dio_wait - wait for outstanding DIO requests to finish
2075 * @inode: inode to wait for
2077 * Waits for all pending direct I/O requests to finish so that we can
2078 * proceed with a truncate or equivalent operation.
2080 * Must be called under a lock that serializes taking new references
2081 * to i_dio_count, usually by inode->i_mutex.
2083 void inode_dio_wait(struct inode *inode)
2085 if (atomic_read(&inode->i_dio_count))
2086 __inode_dio_wait(inode);
2088 EXPORT_SYMBOL(inode_dio_wait);
2091 * inode_set_flags - atomically set some inode flags
2093 * Note: the caller should be holding i_mutex, or else be sure that
2094 * they have exclusive access to the inode structure (i.e., while the
2095 * inode is being instantiated). The reason for the cmpxchg() loop
2096 * --- which wouldn't be necessary if all code paths which modify
2097 * i_flags actually followed this rule, is that there is at least one
2098 * code path which doesn't today so we use cmpxchg() out of an abundance
2099 * of caution.
2101 * In the long run, i_mutex is overkill, and we should probably look
2102 * at using the i_lock spinlock to protect i_flags, and then make sure
2103 * it is so documented in include/linux/fs.h and that all code follows
2104 * the locking convention!!
2106 void inode_set_flags(struct inode *inode, unsigned int flags,
2107 unsigned int mask)
2109 unsigned int old_flags, new_flags;
2111 WARN_ON_ONCE(flags & ~mask);
2112 do {
2113 old_flags = READ_ONCE(inode->i_flags);
2114 new_flags = (old_flags & ~mask) | flags;
2115 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2116 new_flags) != old_flags));
2118 EXPORT_SYMBOL(inode_set_flags);
2120 void inode_nohighmem(struct inode *inode)
2122 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2124 EXPORT_SYMBOL(inode_nohighmem);
2127 * timespec64_trunc - Truncate timespec64 to a granularity
2128 * @t: Timespec64
2129 * @gran: Granularity in ns.
2131 * Truncate a timespec64 to a granularity. Always rounds down. gran must
2132 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2134 struct timespec64 timespec64_trunc(struct timespec64 t, unsigned gran)
2136 /* Avoid division in the common cases 1 ns and 1 s. */
2137 if (gran == 1) {
2138 /* nothing */
2139 } else if (gran == NSEC_PER_SEC) {
2140 t.tv_nsec = 0;
2141 } else if (gran > 1 && gran < NSEC_PER_SEC) {
2142 t.tv_nsec -= t.tv_nsec % gran;
2143 } else {
2144 WARN(1, "illegal file time granularity: %u", gran);
2146 return t;
2148 EXPORT_SYMBOL(timespec64_trunc);
2151 * current_time - Return FS time
2152 * @inode: inode.
2154 * Return the current time truncated to the time granularity supported by
2155 * the fs.
2157 * Note that inode and inode->sb cannot be NULL.
2158 * Otherwise, the function warns and returns time without truncation.
2160 struct timespec64 current_time(struct inode *inode)
2162 struct timespec64 now = current_kernel_time64();
2164 if (unlikely(!inode->i_sb)) {
2165 WARN(1, "current_time() called with uninitialized super_block in the inode");
2166 return now;
2169 return timespec64_trunc(now, inode->i_sb->s_time_gran);
2171 EXPORT_SYMBOL(current_time);