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>
24 * Inode locking rules:
26 * inode->i_lock protects:
27 * inode->i_state, inode->i_hash, __iget()
28 * Inode LRU list locks protect:
29 * inode->i_sb->s_inode_lru, inode->i_lru
30 * inode_sb_list_lock protects:
31 * sb->s_inodes, inode->i_sb_list
32 * bdi->wb.list_lock protects:
33 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
34 * inode_hash_lock protects:
35 * inode_hashtable, inode->i_hash
41 * Inode LRU list locks
54 static unsigned int i_hash_mask __read_mostly
;
55 static unsigned int i_hash_shift __read_mostly
;
56 static struct hlist_head
*inode_hashtable __read_mostly
;
57 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_hash_lock
);
59 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_sb_list_lock
);
62 * Empty aops. Can be used for the cases where the user does not
63 * define any of the address_space operations.
65 const struct address_space_operations empty_aops
= {
67 EXPORT_SYMBOL(empty_aops
);
70 * Statistics gathering..
72 struct inodes_stat_t inodes_stat
;
74 static DEFINE_PER_CPU(unsigned long, nr_inodes
);
75 static DEFINE_PER_CPU(unsigned long, nr_unused
);
77 static struct kmem_cache
*inode_cachep __read_mostly
;
79 static long get_nr_inodes(void)
83 for_each_possible_cpu(i
)
84 sum
+= per_cpu(nr_inodes
, i
);
85 return sum
< 0 ? 0 : sum
;
88 static inline long get_nr_inodes_unused(void)
92 for_each_possible_cpu(i
)
93 sum
+= per_cpu(nr_unused
, i
);
94 return sum
< 0 ? 0 : sum
;
97 long get_nr_dirty_inodes(void)
99 /* not actually dirty inodes, but a wild approximation */
100 long nr_dirty
= get_nr_inodes() - get_nr_inodes_unused();
101 return nr_dirty
> 0 ? nr_dirty
: 0;
105 * Handle nr_inode sysctl
108 int proc_nr_inodes(struct ctl_table
*table
, int write
,
109 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
111 inodes_stat
.nr_inodes
= get_nr_inodes();
112 inodes_stat
.nr_unused
= get_nr_inodes_unused();
113 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
117 static int no_open(struct inode
*inode
, struct file
*file
)
123 * inode_init_always - perform inode structure intialisation
124 * @sb: superblock inode belongs to
125 * @inode: inode to initialise
127 * These are initializations that need to be done on every inode
128 * allocation as the fields are not initialised by slab allocation.
130 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
132 static const struct inode_operations empty_iops
;
133 static const struct file_operations no_open_fops
= {.open
= no_open
};
134 struct address_space
*const mapping
= &inode
->i_data
;
137 inode
->i_blkbits
= sb
->s_blocksize_bits
;
139 atomic_set(&inode
->i_count
, 1);
140 inode
->i_op
= &empty_iops
;
141 inode
->i_fop
= &no_open_fops
;
142 inode
->__i_nlink
= 1;
143 inode
->i_opflags
= 0;
144 i_uid_write(inode
, 0);
145 i_gid_write(inode
, 0);
146 atomic_set(&inode
->i_writecount
, 0);
150 inode
->i_generation
= 0;
151 inode
->i_pipe
= NULL
;
152 inode
->i_bdev
= NULL
;
153 inode
->i_cdev
= 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
->backing_dev_info
= &default_backing_dev_info
;
174 mapping
->writeback_index
= 0;
177 * If the block_device provides a backing_dev_info for client
178 * inodes then use that. Otherwise the inode share the bdev's
182 struct backing_dev_info
*bdi
;
184 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
185 mapping
->backing_dev_info
= bdi
;
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
;
194 #ifdef CONFIG_FSNOTIFY
195 inode
->i_fsnotify_mask
= 0;
197 inode
->i_flctx
= NULL
;
198 this_cpu_inc(nr_inodes
);
204 EXPORT_SYMBOL(inode_init_always
);
206 static struct inode
*alloc_inode(struct super_block
*sb
)
210 if (sb
->s_op
->alloc_inode
)
211 inode
= sb
->s_op
->alloc_inode(sb
);
213 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
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
);
222 kmem_cache_free(inode_cachep
, 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 security_inode_free(inode
);
239 fsnotify_inode_delete(inode
);
240 locks_free_lock_context(inode
->i_flctx
);
241 if (!inode
->i_nlink
) {
242 WARN_ON(atomic_long_read(&inode
->i_sb
->s_remove_count
) == 0);
243 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
246 #ifdef CONFIG_FS_POSIX_ACL
247 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
248 posix_acl_release(inode
->i_acl
);
249 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
250 posix_acl_release(inode
->i_default_acl
);
252 this_cpu_dec(nr_inodes
);
254 EXPORT_SYMBOL(__destroy_inode
);
256 static void i_callback(struct rcu_head
*head
)
258 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
259 kmem_cache_free(inode_cachep
, inode
);
262 static void destroy_inode(struct inode
*inode
)
264 BUG_ON(!list_empty(&inode
->i_lru
));
265 __destroy_inode(inode
);
266 if (inode
->i_sb
->s_op
->destroy_inode
)
267 inode
->i_sb
->s_op
->destroy_inode(inode
);
269 call_rcu(&inode
->i_rcu
, i_callback
);
273 * drop_nlink - directly drop an inode's link count
276 * This is a low-level filesystem helper to replace any
277 * direct filesystem manipulation of i_nlink. In cases
278 * where we are attempting to track writes to the
279 * filesystem, a decrement to zero means an imminent
280 * write when the file is truncated and actually unlinked
283 void drop_nlink(struct inode
*inode
)
285 WARN_ON(inode
->i_nlink
== 0);
288 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
290 EXPORT_SYMBOL(drop_nlink
);
293 * clear_nlink - directly zero an inode's link count
296 * This is a low-level filesystem helper to replace any
297 * direct filesystem manipulation of i_nlink. See
298 * drop_nlink() for why we care about i_nlink hitting zero.
300 void clear_nlink(struct inode
*inode
)
302 if (inode
->i_nlink
) {
303 inode
->__i_nlink
= 0;
304 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
307 EXPORT_SYMBOL(clear_nlink
);
310 * set_nlink - directly set an inode's link count
312 * @nlink: new nlink (should be non-zero)
314 * This is a low-level filesystem helper to replace any
315 * direct filesystem manipulation of i_nlink.
317 void set_nlink(struct inode
*inode
, unsigned int nlink
)
322 /* Yes, some filesystems do change nlink from zero to one */
323 if (inode
->i_nlink
== 0)
324 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
326 inode
->__i_nlink
= nlink
;
329 EXPORT_SYMBOL(set_nlink
);
332 * inc_nlink - directly increment an inode's link count
335 * This is a low-level filesystem helper to replace any
336 * direct filesystem manipulation of i_nlink. Currently,
337 * it is only here for parity with dec_nlink().
339 void inc_nlink(struct inode
*inode
)
341 if (unlikely(inode
->i_nlink
== 0)) {
342 WARN_ON(!(inode
->i_state
& I_LINKABLE
));
343 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
348 EXPORT_SYMBOL(inc_nlink
);
350 void address_space_init_once(struct address_space
*mapping
)
352 memset(mapping
, 0, sizeof(*mapping
));
353 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
354 spin_lock_init(&mapping
->tree_lock
);
355 init_rwsem(&mapping
->i_mmap_rwsem
);
356 INIT_LIST_HEAD(&mapping
->private_list
);
357 spin_lock_init(&mapping
->private_lock
);
358 mapping
->i_mmap
= RB_ROOT
;
359 INIT_LIST_HEAD(&mapping
->i_mmap_nonlinear
);
361 EXPORT_SYMBOL(address_space_init_once
);
364 * These are initializations that only need to be done
365 * once, because the fields are idempotent across use
366 * of the inode, so let the slab aware of that.
368 void inode_init_once(struct inode
*inode
)
370 memset(inode
, 0, sizeof(*inode
));
371 INIT_HLIST_NODE(&inode
->i_hash
);
372 INIT_LIST_HEAD(&inode
->i_devices
);
373 INIT_LIST_HEAD(&inode
->i_wb_list
);
374 INIT_LIST_HEAD(&inode
->i_lru
);
375 address_space_init_once(&inode
->i_data
);
376 i_size_ordered_init(inode
);
377 #ifdef CONFIG_FSNOTIFY
378 INIT_HLIST_HEAD(&inode
->i_fsnotify_marks
);
381 EXPORT_SYMBOL(inode_init_once
);
383 static void init_once(void *foo
)
385 struct inode
*inode
= (struct inode
*) foo
;
387 inode_init_once(inode
);
391 * inode->i_lock must be held
393 void __iget(struct inode
*inode
)
395 atomic_inc(&inode
->i_count
);
399 * get additional reference to inode; caller must already hold one.
401 void ihold(struct inode
*inode
)
403 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
405 EXPORT_SYMBOL(ihold
);
407 static void inode_lru_list_add(struct inode
*inode
)
409 if (list_lru_add(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
410 this_cpu_inc(nr_unused
);
414 * Add inode to LRU if needed (inode is unused and clean).
416 * Needs inode->i_lock held.
418 void inode_add_lru(struct inode
*inode
)
420 if (!(inode
->i_state
& (I_DIRTY
| I_SYNC
| I_FREEING
| I_WILL_FREE
)) &&
421 !atomic_read(&inode
->i_count
) && inode
->i_sb
->s_flags
& MS_ACTIVE
)
422 inode_lru_list_add(inode
);
426 static void inode_lru_list_del(struct inode
*inode
)
429 if (list_lru_del(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
430 this_cpu_dec(nr_unused
);
434 * inode_sb_list_add - add inode to the superblock list of inodes
435 * @inode: inode to add
437 void inode_sb_list_add(struct inode
*inode
)
439 spin_lock(&inode_sb_list_lock
);
440 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
441 spin_unlock(&inode_sb_list_lock
);
443 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
445 static inline void inode_sb_list_del(struct inode
*inode
)
447 if (!list_empty(&inode
->i_sb_list
)) {
448 spin_lock(&inode_sb_list_lock
);
449 list_del_init(&inode
->i_sb_list
);
450 spin_unlock(&inode_sb_list_lock
);
454 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
458 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
460 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> i_hash_shift
);
461 return tmp
& i_hash_mask
;
465 * __insert_inode_hash - hash an inode
466 * @inode: unhashed inode
467 * @hashval: unsigned long value used to locate this object in the
470 * Add an inode to the inode hash for this superblock.
472 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
474 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
476 spin_lock(&inode_hash_lock
);
477 spin_lock(&inode
->i_lock
);
478 hlist_add_head(&inode
->i_hash
, b
);
479 spin_unlock(&inode
->i_lock
);
480 spin_unlock(&inode_hash_lock
);
482 EXPORT_SYMBOL(__insert_inode_hash
);
485 * __remove_inode_hash - remove an inode from the hash
486 * @inode: inode to unhash
488 * Remove an inode from the superblock.
490 void __remove_inode_hash(struct inode
*inode
)
492 spin_lock(&inode_hash_lock
);
493 spin_lock(&inode
->i_lock
);
494 hlist_del_init(&inode
->i_hash
);
495 spin_unlock(&inode
->i_lock
);
496 spin_unlock(&inode_hash_lock
);
498 EXPORT_SYMBOL(__remove_inode_hash
);
500 void clear_inode(struct inode
*inode
)
504 * We have to cycle tree_lock here because reclaim can be still in the
505 * process of removing the last page (in __delete_from_page_cache())
506 * and we must not free mapping under it.
508 spin_lock_irq(&inode
->i_data
.tree_lock
);
509 BUG_ON(inode
->i_data
.nrpages
);
510 BUG_ON(inode
->i_data
.nrshadows
);
511 spin_unlock_irq(&inode
->i_data
.tree_lock
);
512 BUG_ON(!list_empty(&inode
->i_data
.private_list
));
513 BUG_ON(!(inode
->i_state
& I_FREEING
));
514 BUG_ON(inode
->i_state
& I_CLEAR
);
515 /* don't need i_lock here, no concurrent mods to i_state */
516 inode
->i_state
= I_FREEING
| I_CLEAR
;
518 EXPORT_SYMBOL(clear_inode
);
521 * Free the inode passed in, removing it from the lists it is still connected
522 * to. We remove any pages still attached to the inode and wait for any IO that
523 * is still in progress before finally destroying the inode.
525 * An inode must already be marked I_FREEING so that we avoid the inode being
526 * moved back onto lists if we race with other code that manipulates the lists
527 * (e.g. writeback_single_inode). The caller is responsible for setting this.
529 * An inode must already be removed from the LRU list before being evicted from
530 * the cache. This should occur atomically with setting the I_FREEING state
531 * flag, so no inodes here should ever be on the LRU when being evicted.
533 static void evict(struct inode
*inode
)
535 const struct super_operations
*op
= inode
->i_sb
->s_op
;
537 BUG_ON(!(inode
->i_state
& I_FREEING
));
538 BUG_ON(!list_empty(&inode
->i_lru
));
540 if (!list_empty(&inode
->i_wb_list
))
541 inode_wb_list_del(inode
);
543 inode_sb_list_del(inode
);
546 * Wait for flusher thread to be done with the inode so that filesystem
547 * does not start destroying it while writeback is still running. Since
548 * the inode has I_FREEING set, flusher thread won't start new work on
549 * the inode. We just have to wait for running writeback to finish.
551 inode_wait_for_writeback(inode
);
553 if (op
->evict_inode
) {
554 op
->evict_inode(inode
);
556 truncate_inode_pages_final(&inode
->i_data
);
559 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
561 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
564 remove_inode_hash(inode
);
566 spin_lock(&inode
->i_lock
);
567 wake_up_bit(&inode
->i_state
, __I_NEW
);
568 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
569 spin_unlock(&inode
->i_lock
);
571 destroy_inode(inode
);
575 * dispose_list - dispose of the contents of a local list
576 * @head: the head of the list to free
578 * Dispose-list gets a local list with local inodes in it, so it doesn't
579 * need to worry about list corruption and SMP locks.
581 static void dispose_list(struct list_head
*head
)
583 while (!list_empty(head
)) {
586 inode
= list_first_entry(head
, struct inode
, i_lru
);
587 list_del_init(&inode
->i_lru
);
594 * evict_inodes - evict all evictable inodes for a superblock
595 * @sb: superblock to operate on
597 * Make sure that no inodes with zero refcount are retained. This is
598 * called by superblock shutdown after having MS_ACTIVE flag removed,
599 * so any inode reaching zero refcount during or after that call will
600 * be immediately evicted.
602 void evict_inodes(struct super_block
*sb
)
604 struct inode
*inode
, *next
;
607 spin_lock(&inode_sb_list_lock
);
608 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
609 if (atomic_read(&inode
->i_count
))
612 spin_lock(&inode
->i_lock
);
613 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
614 spin_unlock(&inode
->i_lock
);
618 inode
->i_state
|= I_FREEING
;
619 inode_lru_list_del(inode
);
620 spin_unlock(&inode
->i_lock
);
621 list_add(&inode
->i_lru
, &dispose
);
623 spin_unlock(&inode_sb_list_lock
);
625 dispose_list(&dispose
);
629 * invalidate_inodes - attempt to free all inodes on a superblock
630 * @sb: superblock to operate on
631 * @kill_dirty: flag to guide handling of dirty inodes
633 * Attempts to free all inodes for a given superblock. If there were any
634 * busy inodes return a non-zero value, else zero.
635 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
638 int invalidate_inodes(struct super_block
*sb
, bool kill_dirty
)
641 struct inode
*inode
, *next
;
644 spin_lock(&inode_sb_list_lock
);
645 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
646 spin_lock(&inode
->i_lock
);
647 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
648 spin_unlock(&inode
->i_lock
);
651 if (inode
->i_state
& I_DIRTY
&& !kill_dirty
) {
652 spin_unlock(&inode
->i_lock
);
656 if (atomic_read(&inode
->i_count
)) {
657 spin_unlock(&inode
->i_lock
);
662 inode
->i_state
|= I_FREEING
;
663 inode_lru_list_del(inode
);
664 spin_unlock(&inode
->i_lock
);
665 list_add(&inode
->i_lru
, &dispose
);
667 spin_unlock(&inode_sb_list_lock
);
669 dispose_list(&dispose
);
675 * Isolate the inode from the LRU in preparation for freeing it.
677 * Any inodes which are pinned purely because of attached pagecache have their
678 * pagecache removed. If the inode has metadata buffers attached to
679 * mapping->private_list then try to remove them.
681 * If the inode has the I_REFERENCED flag set, then it means that it has been
682 * used recently - the flag is set in iput_final(). When we encounter such an
683 * inode, clear the flag and move it to the back of the LRU so it gets another
684 * pass through the LRU before it gets reclaimed. This is necessary because of
685 * the fact we are doing lazy LRU updates to minimise lock contention so the
686 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
687 * with this flag set because they are the inodes that are out of order.
689 static enum lru_status
690 inode_lru_isolate(struct list_head
*item
, spinlock_t
*lru_lock
, void *arg
)
692 struct list_head
*freeable
= arg
;
693 struct inode
*inode
= container_of(item
, struct inode
, i_lru
);
696 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
697 * If we fail to get the lock, just skip it.
699 if (!spin_trylock(&inode
->i_lock
))
703 * Referenced or dirty inodes are still in use. Give them another pass
704 * through the LRU as we canot reclaim them now.
706 if (atomic_read(&inode
->i_count
) ||
707 (inode
->i_state
& ~I_REFERENCED
)) {
708 list_del_init(&inode
->i_lru
);
709 spin_unlock(&inode
->i_lock
);
710 this_cpu_dec(nr_unused
);
714 /* recently referenced inodes get one more pass */
715 if (inode
->i_state
& I_REFERENCED
) {
716 inode
->i_state
&= ~I_REFERENCED
;
717 spin_unlock(&inode
->i_lock
);
721 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
723 spin_unlock(&inode
->i_lock
);
724 spin_unlock(lru_lock
);
725 if (remove_inode_buffers(inode
)) {
727 reap
= invalidate_mapping_pages(&inode
->i_data
, 0, -1);
728 if (current_is_kswapd())
729 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
731 __count_vm_events(PGINODESTEAL
, reap
);
732 if (current
->reclaim_state
)
733 current
->reclaim_state
->reclaimed_slab
+= reap
;
740 WARN_ON(inode
->i_state
& I_NEW
);
741 inode
->i_state
|= I_FREEING
;
742 list_move(&inode
->i_lru
, freeable
);
743 spin_unlock(&inode
->i_lock
);
745 this_cpu_dec(nr_unused
);
750 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
751 * This is called from the superblock shrinker function with a number of inodes
752 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
753 * then are freed outside inode_lock by dispose_list().
755 long prune_icache_sb(struct super_block
*sb
, unsigned long nr_to_scan
,
761 freed
= list_lru_walk_node(&sb
->s_inode_lru
, nid
, inode_lru_isolate
,
762 &freeable
, &nr_to_scan
);
763 dispose_list(&freeable
);
767 static void __wait_on_freeing_inode(struct inode
*inode
);
769 * Called with the inode lock held.
771 static struct inode
*find_inode(struct super_block
*sb
,
772 struct hlist_head
*head
,
773 int (*test
)(struct inode
*, void *),
776 struct inode
*inode
= NULL
;
779 hlist_for_each_entry(inode
, head
, i_hash
) {
780 if (inode
->i_sb
!= sb
)
782 if (!test(inode
, data
))
784 spin_lock(&inode
->i_lock
);
785 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
786 __wait_on_freeing_inode(inode
);
790 spin_unlock(&inode
->i_lock
);
797 * find_inode_fast is the fast path version of find_inode, see the comment at
798 * iget_locked for details.
800 static struct inode
*find_inode_fast(struct super_block
*sb
,
801 struct hlist_head
*head
, unsigned long ino
)
803 struct inode
*inode
= NULL
;
806 hlist_for_each_entry(inode
, head
, i_hash
) {
807 if (inode
->i_ino
!= ino
)
809 if (inode
->i_sb
!= sb
)
811 spin_lock(&inode
->i_lock
);
812 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
813 __wait_on_freeing_inode(inode
);
817 spin_unlock(&inode
->i_lock
);
824 * Each cpu owns a range of LAST_INO_BATCH numbers.
825 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
826 * to renew the exhausted range.
828 * This does not significantly increase overflow rate because every CPU can
829 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
830 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
831 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
832 * overflow rate by 2x, which does not seem too significant.
834 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
835 * error if st_ino won't fit in target struct field. Use 32bit counter
836 * here to attempt to avoid that.
838 #define LAST_INO_BATCH 1024
839 static DEFINE_PER_CPU(unsigned int, last_ino
);
841 unsigned int get_next_ino(void)
843 unsigned int *p
= &get_cpu_var(last_ino
);
844 unsigned int res
= *p
;
847 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
848 static atomic_t shared_last_ino
;
849 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
851 res
= next
- LAST_INO_BATCH
;
856 put_cpu_var(last_ino
);
859 EXPORT_SYMBOL(get_next_ino
);
862 * new_inode_pseudo - obtain an inode
865 * Allocates a new inode for given superblock.
866 * Inode wont be chained in superblock s_inodes list
868 * - fs can't be unmount
869 * - quotas, fsnotify, writeback can't work
871 struct inode
*new_inode_pseudo(struct super_block
*sb
)
873 struct inode
*inode
= alloc_inode(sb
);
876 spin_lock(&inode
->i_lock
);
878 spin_unlock(&inode
->i_lock
);
879 INIT_LIST_HEAD(&inode
->i_sb_list
);
885 * new_inode - obtain an inode
888 * Allocates a new inode for given superblock. The default gfp_mask
889 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
890 * If HIGHMEM pages are unsuitable or it is known that pages allocated
891 * for the page cache are not reclaimable or migratable,
892 * mapping_set_gfp_mask() must be called with suitable flags on the
893 * newly created inode's mapping
896 struct inode
*new_inode(struct super_block
*sb
)
900 spin_lock_prefetch(&inode_sb_list_lock
);
902 inode
= new_inode_pseudo(sb
);
904 inode_sb_list_add(inode
);
907 EXPORT_SYMBOL(new_inode
);
909 #ifdef CONFIG_DEBUG_LOCK_ALLOC
910 void lockdep_annotate_inode_mutex_key(struct inode
*inode
)
912 if (S_ISDIR(inode
->i_mode
)) {
913 struct file_system_type
*type
= inode
->i_sb
->s_type
;
915 /* Set new key only if filesystem hasn't already changed it */
916 if (lockdep_match_class(&inode
->i_mutex
, &type
->i_mutex_key
)) {
918 * ensure nobody is actually holding i_mutex
920 mutex_destroy(&inode
->i_mutex
);
921 mutex_init(&inode
->i_mutex
);
922 lockdep_set_class(&inode
->i_mutex
,
923 &type
->i_mutex_dir_key
);
927 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key
);
931 * unlock_new_inode - clear the I_NEW state and wake up any waiters
932 * @inode: new inode to unlock
934 * Called when the inode is fully initialised to clear the new state of the
935 * inode and wake up anyone waiting for the inode to finish initialisation.
937 void unlock_new_inode(struct inode
*inode
)
939 lockdep_annotate_inode_mutex_key(inode
);
940 spin_lock(&inode
->i_lock
);
941 WARN_ON(!(inode
->i_state
& I_NEW
));
942 inode
->i_state
&= ~I_NEW
;
944 wake_up_bit(&inode
->i_state
, __I_NEW
);
945 spin_unlock(&inode
->i_lock
);
947 EXPORT_SYMBOL(unlock_new_inode
);
950 * lock_two_nondirectories - take two i_mutexes on non-directory objects
952 * Lock any non-NULL argument that is not a directory.
953 * Zero, one or two objects may be locked by this function.
955 * @inode1: first inode to lock
956 * @inode2: second inode to lock
958 void lock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
961 swap(inode1
, inode2
);
963 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
964 mutex_lock(&inode1
->i_mutex
);
965 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
966 mutex_lock_nested(&inode2
->i_mutex
, I_MUTEX_NONDIR2
);
968 EXPORT_SYMBOL(lock_two_nondirectories
);
971 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
972 * @inode1: first inode to unlock
973 * @inode2: second inode to unlock
975 void unlock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
977 if (inode1
&& !S_ISDIR(inode1
->i_mode
))
978 mutex_unlock(&inode1
->i_mutex
);
979 if (inode2
&& !S_ISDIR(inode2
->i_mode
) && inode2
!= inode1
)
980 mutex_unlock(&inode2
->i_mutex
);
982 EXPORT_SYMBOL(unlock_two_nondirectories
);
985 * iget5_locked - obtain an inode from a mounted file system
986 * @sb: super block of file system
987 * @hashval: hash value (usually inode number) to get
988 * @test: callback used for comparisons between inodes
989 * @set: callback used to initialize a new struct inode
990 * @data: opaque data pointer to pass to @test and @set
992 * Search for the inode specified by @hashval and @data in the inode cache,
993 * and if present it is return it with an increased reference count. This is
994 * a generalized version of iget_locked() for file systems where the inode
995 * number is not sufficient for unique identification of an inode.
997 * If the inode is not in cache, allocate a new inode and return it locked,
998 * hashed, and with the I_NEW flag set. The file system gets to fill it in
999 * before unlocking it via unlock_new_inode().
1001 * Note both @test and @set are called with the inode_hash_lock held, so can't
1004 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1005 int (*test
)(struct inode
*, void *),
1006 int (*set
)(struct inode
*, void *), void *data
)
1008 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1009 struct inode
*inode
;
1011 spin_lock(&inode_hash_lock
);
1012 inode
= find_inode(sb
, head
, test
, data
);
1013 spin_unlock(&inode_hash_lock
);
1016 wait_on_inode(inode
);
1020 inode
= alloc_inode(sb
);
1024 spin_lock(&inode_hash_lock
);
1025 /* We released the lock, so.. */
1026 old
= find_inode(sb
, head
, test
, data
);
1028 if (set(inode
, data
))
1031 spin_lock(&inode
->i_lock
);
1032 inode
->i_state
= I_NEW
;
1033 hlist_add_head(&inode
->i_hash
, head
);
1034 spin_unlock(&inode
->i_lock
);
1035 inode_sb_list_add(inode
);
1036 spin_unlock(&inode_hash_lock
);
1038 /* Return the locked inode with I_NEW set, the
1039 * caller is responsible for filling in the contents
1045 * Uhhuh, somebody else created the same inode under
1046 * us. Use the old inode instead of the one we just
1049 spin_unlock(&inode_hash_lock
);
1050 destroy_inode(inode
);
1052 wait_on_inode(inode
);
1057 spin_unlock(&inode_hash_lock
);
1058 destroy_inode(inode
);
1061 EXPORT_SYMBOL(iget5_locked
);
1064 * iget_locked - obtain an inode from a mounted file system
1065 * @sb: super block of file system
1066 * @ino: inode number to get
1068 * Search for the inode specified by @ino in the inode cache and if present
1069 * return it with an increased reference count. This is for file systems
1070 * where the inode number is sufficient for unique identification of an inode.
1072 * If the inode is not in cache, allocate a new inode and return it locked,
1073 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1074 * before unlocking it via unlock_new_inode().
1076 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1078 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1079 struct inode
*inode
;
1081 spin_lock(&inode_hash_lock
);
1082 inode
= find_inode_fast(sb
, head
, ino
);
1083 spin_unlock(&inode_hash_lock
);
1085 wait_on_inode(inode
);
1089 inode
= alloc_inode(sb
);
1093 spin_lock(&inode_hash_lock
);
1094 /* We released the lock, so.. */
1095 old
= find_inode_fast(sb
, head
, ino
);
1098 spin_lock(&inode
->i_lock
);
1099 inode
->i_state
= I_NEW
;
1100 hlist_add_head(&inode
->i_hash
, head
);
1101 spin_unlock(&inode
->i_lock
);
1102 inode_sb_list_add(inode
);
1103 spin_unlock(&inode_hash_lock
);
1105 /* Return the locked inode with I_NEW set, the
1106 * caller is responsible for filling in the contents
1112 * Uhhuh, somebody else created the same inode under
1113 * us. Use the old inode instead of the one we just
1116 spin_unlock(&inode_hash_lock
);
1117 destroy_inode(inode
);
1119 wait_on_inode(inode
);
1123 EXPORT_SYMBOL(iget_locked
);
1126 * search the inode cache for a matching inode number.
1127 * If we find one, then the inode number we are trying to
1128 * allocate is not unique and so we should not use it.
1130 * Returns 1 if the inode number is unique, 0 if it is not.
1132 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1134 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1135 struct inode
*inode
;
1137 spin_lock(&inode_hash_lock
);
1138 hlist_for_each_entry(inode
, b
, i_hash
) {
1139 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
) {
1140 spin_unlock(&inode_hash_lock
);
1144 spin_unlock(&inode_hash_lock
);
1150 * iunique - get a unique inode number
1152 * @max_reserved: highest reserved inode number
1154 * Obtain an inode number that is unique on the system for a given
1155 * superblock. This is used by file systems that have no natural
1156 * permanent inode numbering system. An inode number is returned that
1157 * is higher than the reserved limit but unique.
1160 * With a large number of inodes live on the file system this function
1161 * currently becomes quite slow.
1163 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1166 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1167 * error if st_ino won't fit in target struct field. Use 32bit counter
1168 * here to attempt to avoid that.
1170 static DEFINE_SPINLOCK(iunique_lock
);
1171 static unsigned int counter
;
1174 spin_lock(&iunique_lock
);
1176 if (counter
<= max_reserved
)
1177 counter
= max_reserved
+ 1;
1179 } while (!test_inode_iunique(sb
, res
));
1180 spin_unlock(&iunique_lock
);
1184 EXPORT_SYMBOL(iunique
);
1186 struct inode
*igrab(struct inode
*inode
)
1188 spin_lock(&inode
->i_lock
);
1189 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1191 spin_unlock(&inode
->i_lock
);
1193 spin_unlock(&inode
->i_lock
);
1195 * Handle the case where s_op->clear_inode is not been
1196 * called yet, and somebody is calling igrab
1197 * while the inode is getting freed.
1203 EXPORT_SYMBOL(igrab
);
1206 * ilookup5_nowait - search for an inode in the inode cache
1207 * @sb: super block of file system to search
1208 * @hashval: hash value (usually inode number) to search for
1209 * @test: callback used for comparisons between inodes
1210 * @data: opaque data pointer to pass to @test
1212 * Search for the inode specified by @hashval and @data in the inode cache.
1213 * If the inode is in the cache, the inode is returned with an incremented
1216 * Note: I_NEW is not waited upon so you have to be very careful what you do
1217 * with the returned inode. You probably should be using ilookup5() instead.
1219 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1221 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1222 int (*test
)(struct inode
*, void *), void *data
)
1224 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1225 struct inode
*inode
;
1227 spin_lock(&inode_hash_lock
);
1228 inode
= find_inode(sb
, head
, test
, data
);
1229 spin_unlock(&inode_hash_lock
);
1233 EXPORT_SYMBOL(ilookup5_nowait
);
1236 * ilookup5 - search for an inode in the inode cache
1237 * @sb: super block of file system to search
1238 * @hashval: hash value (usually inode number) to search for
1239 * @test: callback used for comparisons between inodes
1240 * @data: opaque data pointer to pass to @test
1242 * Search for the inode specified by @hashval and @data in the inode cache,
1243 * and if the inode is in the cache, return the inode with an incremented
1244 * reference count. Waits on I_NEW before returning the inode.
1245 * returned with an incremented reference count.
1247 * This is a generalized version of ilookup() for file systems where the
1248 * inode number is not sufficient for unique identification of an inode.
1250 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1252 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1253 int (*test
)(struct inode
*, void *), void *data
)
1255 struct inode
*inode
= ilookup5_nowait(sb
, hashval
, test
, data
);
1258 wait_on_inode(inode
);
1261 EXPORT_SYMBOL(ilookup5
);
1264 * ilookup - search for an inode in the inode cache
1265 * @sb: super block of file system to search
1266 * @ino: inode number to search for
1268 * Search for the inode @ino in the inode cache, and if the inode is in the
1269 * cache, the inode is returned with an incremented reference count.
1271 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1273 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1274 struct inode
*inode
;
1276 spin_lock(&inode_hash_lock
);
1277 inode
= find_inode_fast(sb
, head
, ino
);
1278 spin_unlock(&inode_hash_lock
);
1281 wait_on_inode(inode
);
1284 EXPORT_SYMBOL(ilookup
);
1286 int insert_inode_locked(struct inode
*inode
)
1288 struct super_block
*sb
= inode
->i_sb
;
1289 ino_t ino
= inode
->i_ino
;
1290 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1293 struct inode
*old
= NULL
;
1294 spin_lock(&inode_hash_lock
);
1295 hlist_for_each_entry(old
, head
, i_hash
) {
1296 if (old
->i_ino
!= ino
)
1298 if (old
->i_sb
!= sb
)
1300 spin_lock(&old
->i_lock
);
1301 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1302 spin_unlock(&old
->i_lock
);
1308 spin_lock(&inode
->i_lock
);
1309 inode
->i_state
|= I_NEW
;
1310 hlist_add_head(&inode
->i_hash
, head
);
1311 spin_unlock(&inode
->i_lock
);
1312 spin_unlock(&inode_hash_lock
);
1316 spin_unlock(&old
->i_lock
);
1317 spin_unlock(&inode_hash_lock
);
1319 if (unlikely(!inode_unhashed(old
))) {
1326 EXPORT_SYMBOL(insert_inode_locked
);
1328 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1329 int (*test
)(struct inode
*, void *), void *data
)
1331 struct super_block
*sb
= inode
->i_sb
;
1332 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1335 struct inode
*old
= NULL
;
1337 spin_lock(&inode_hash_lock
);
1338 hlist_for_each_entry(old
, head
, i_hash
) {
1339 if (old
->i_sb
!= sb
)
1341 if (!test(old
, data
))
1343 spin_lock(&old
->i_lock
);
1344 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1345 spin_unlock(&old
->i_lock
);
1351 spin_lock(&inode
->i_lock
);
1352 inode
->i_state
|= I_NEW
;
1353 hlist_add_head(&inode
->i_hash
, head
);
1354 spin_unlock(&inode
->i_lock
);
1355 spin_unlock(&inode_hash_lock
);
1359 spin_unlock(&old
->i_lock
);
1360 spin_unlock(&inode_hash_lock
);
1362 if (unlikely(!inode_unhashed(old
))) {
1369 EXPORT_SYMBOL(insert_inode_locked4
);
1372 int generic_delete_inode(struct inode
*inode
)
1376 EXPORT_SYMBOL(generic_delete_inode
);
1379 * Called when we're dropping the last reference
1382 * Call the FS "drop_inode()" function, defaulting to
1383 * the legacy UNIX filesystem behaviour. If it tells
1384 * us to evict inode, do so. Otherwise, retain inode
1385 * in cache if fs is alive, sync and evict if fs is
1388 static void iput_final(struct inode
*inode
)
1390 struct super_block
*sb
= inode
->i_sb
;
1391 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1394 WARN_ON(inode
->i_state
& I_NEW
);
1397 drop
= op
->drop_inode(inode
);
1399 drop
= generic_drop_inode(inode
);
1401 if (!drop
&& (sb
->s_flags
& MS_ACTIVE
)) {
1402 inode
->i_state
|= I_REFERENCED
;
1403 inode_add_lru(inode
);
1404 spin_unlock(&inode
->i_lock
);
1409 inode
->i_state
|= I_WILL_FREE
;
1410 spin_unlock(&inode
->i_lock
);
1411 write_inode_now(inode
, 1);
1412 spin_lock(&inode
->i_lock
);
1413 WARN_ON(inode
->i_state
& I_NEW
);
1414 inode
->i_state
&= ~I_WILL_FREE
;
1417 inode
->i_state
|= I_FREEING
;
1418 if (!list_empty(&inode
->i_lru
))
1419 inode_lru_list_del(inode
);
1420 spin_unlock(&inode
->i_lock
);
1426 * iput - put an inode
1427 * @inode: inode to put
1429 * Puts an inode, dropping its usage count. If the inode use count hits
1430 * zero, the inode is then freed and may also be destroyed.
1432 * Consequently, iput() can sleep.
1434 void iput(struct inode
*inode
)
1437 BUG_ON(inode
->i_state
& I_CLEAR
);
1439 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
))
1443 EXPORT_SYMBOL(iput
);
1446 * bmap - find a block number in a file
1447 * @inode: inode of file
1448 * @block: block to find
1450 * Returns the block number on the device holding the inode that
1451 * is the disk block number for the block of the file requested.
1452 * That is, asked for block 4 of inode 1 the function will return the
1453 * disk block relative to the disk start that holds that block of the
1456 sector_t
bmap(struct inode
*inode
, sector_t block
)
1459 if (inode
->i_mapping
->a_ops
->bmap
)
1460 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1463 EXPORT_SYMBOL(bmap
);
1466 * With relative atime, only update atime if the previous atime is
1467 * earlier than either the ctime or mtime or if at least a day has
1468 * passed since the last atime update.
1470 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1471 struct timespec now
)
1474 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1477 * Is mtime younger than atime? If yes, update atime:
1479 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1482 * Is ctime younger than atime? If yes, update atime:
1484 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1488 * Is the previous atime value older than a day? If yes,
1491 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1494 * Good, we can skip the atime update:
1500 * This does the actual work of updating an inodes time or version. Must have
1501 * had called mnt_want_write() before calling this.
1503 static int update_time(struct inode
*inode
, struct timespec
*time
, int flags
)
1505 if (inode
->i_op
->update_time
)
1506 return inode
->i_op
->update_time(inode
, time
, flags
);
1508 if (flags
& S_ATIME
)
1509 inode
->i_atime
= *time
;
1510 if (flags
& S_VERSION
)
1511 inode_inc_iversion(inode
);
1512 if (flags
& S_CTIME
)
1513 inode
->i_ctime
= *time
;
1514 if (flags
& S_MTIME
)
1515 inode
->i_mtime
= *time
;
1516 mark_inode_dirty_sync(inode
);
1521 * touch_atime - update the access time
1522 * @path: the &struct path to update
1524 * Update the accessed time on an inode and mark it for writeback.
1525 * This function automatically handles read only file systems and media,
1526 * as well as the "noatime" flag and inode specific "noatime" markers.
1528 void touch_atime(const struct path
*path
)
1530 struct vfsmount
*mnt
= path
->mnt
;
1531 struct inode
*inode
= path
->dentry
->d_inode
;
1532 struct timespec now
;
1534 if (inode
->i_flags
& S_NOATIME
)
1536 if (IS_NOATIME(inode
))
1538 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1541 if (mnt
->mnt_flags
& MNT_NOATIME
)
1543 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1546 now
= current_fs_time(inode
->i_sb
);
1548 if (!relatime_need_update(mnt
, inode
, now
))
1551 if (timespec_equal(&inode
->i_atime
, &now
))
1554 if (!sb_start_write_trylock(inode
->i_sb
))
1557 if (__mnt_want_write(mnt
))
1560 * File systems can error out when updating inodes if they need to
1561 * allocate new space to modify an inode (such is the case for
1562 * Btrfs), but since we touch atime while walking down the path we
1563 * really don't care if we failed to update the atime of the file,
1564 * so just ignore the return value.
1565 * We may also fail on filesystems that have the ability to make parts
1566 * of the fs read only, e.g. subvolumes in Btrfs.
1568 update_time(inode
, &now
, S_ATIME
);
1569 __mnt_drop_write(mnt
);
1571 sb_end_write(inode
->i_sb
);
1573 EXPORT_SYMBOL(touch_atime
);
1576 * The logic we want is
1578 * if suid or (sgid and xgrp)
1581 int should_remove_suid(struct dentry
*dentry
)
1583 umode_t mode
= dentry
->d_inode
->i_mode
;
1586 /* suid always must be killed */
1587 if (unlikely(mode
& S_ISUID
))
1588 kill
= ATTR_KILL_SUID
;
1591 * sgid without any exec bits is just a mandatory locking mark; leave
1592 * it alone. If some exec bits are set, it's a real sgid; kill it.
1594 if (unlikely((mode
& S_ISGID
) && (mode
& S_IXGRP
)))
1595 kill
|= ATTR_KILL_SGID
;
1597 if (unlikely(kill
&& !capable(CAP_FSETID
) && S_ISREG(mode
)))
1602 EXPORT_SYMBOL(should_remove_suid
);
1604 static int __remove_suid(struct dentry
*dentry
, int kill
)
1606 struct iattr newattrs
;
1608 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
1610 * Note we call this on write, so notify_change will not
1611 * encounter any conflicting delegations:
1613 return notify_change(dentry
, &newattrs
, NULL
);
1616 int file_remove_suid(struct file
*file
)
1618 struct dentry
*dentry
= file
->f_path
.dentry
;
1619 struct inode
*inode
= dentry
->d_inode
;
1624 /* Fast path for nothing security related */
1625 if (IS_NOSEC(inode
))
1628 killsuid
= should_remove_suid(dentry
);
1629 killpriv
= security_inode_need_killpriv(dentry
);
1634 error
= security_inode_killpriv(dentry
);
1635 if (!error
&& killsuid
)
1636 error
= __remove_suid(dentry
, killsuid
);
1637 if (!error
&& (inode
->i_sb
->s_flags
& MS_NOSEC
))
1638 inode
->i_flags
|= S_NOSEC
;
1642 EXPORT_SYMBOL(file_remove_suid
);
1645 * file_update_time - update mtime and ctime time
1646 * @file: file accessed
1648 * Update the mtime and ctime members of an inode and mark the inode
1649 * for writeback. Note that this function is meant exclusively for
1650 * usage in the file write path of filesystems, and filesystems may
1651 * choose to explicitly ignore update via this function with the
1652 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1653 * timestamps are handled by the server. This can return an error for
1654 * file systems who need to allocate space in order to update an inode.
1657 int file_update_time(struct file
*file
)
1659 struct inode
*inode
= file_inode(file
);
1660 struct timespec now
;
1664 /* First try to exhaust all avenues to not sync */
1665 if (IS_NOCMTIME(inode
))
1668 now
= current_fs_time(inode
->i_sb
);
1669 if (!timespec_equal(&inode
->i_mtime
, &now
))
1672 if (!timespec_equal(&inode
->i_ctime
, &now
))
1675 if (IS_I_VERSION(inode
))
1676 sync_it
|= S_VERSION
;
1681 /* Finally allowed to write? Takes lock. */
1682 if (__mnt_want_write_file(file
))
1685 ret
= update_time(inode
, &now
, sync_it
);
1686 __mnt_drop_write_file(file
);
1690 EXPORT_SYMBOL(file_update_time
);
1692 int inode_needs_sync(struct inode
*inode
)
1696 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1700 EXPORT_SYMBOL(inode_needs_sync
);
1703 * If we try to find an inode in the inode hash while it is being
1704 * deleted, we have to wait until the filesystem completes its
1705 * deletion before reporting that it isn't found. This function waits
1706 * until the deletion _might_ have completed. Callers are responsible
1707 * to recheck inode state.
1709 * It doesn't matter if I_NEW is not set initially, a call to
1710 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1713 static void __wait_on_freeing_inode(struct inode
*inode
)
1715 wait_queue_head_t
*wq
;
1716 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1717 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1718 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1719 spin_unlock(&inode
->i_lock
);
1720 spin_unlock(&inode_hash_lock
);
1722 finish_wait(wq
, &wait
.wait
);
1723 spin_lock(&inode_hash_lock
);
1726 static __initdata
unsigned long ihash_entries
;
1727 static int __init
set_ihash_entries(char *str
)
1731 ihash_entries
= simple_strtoul(str
, &str
, 0);
1734 __setup("ihash_entries=", set_ihash_entries
);
1737 * Initialize the waitqueues and inode hash table.
1739 void __init
inode_init_early(void)
1743 /* If hashes are distributed across NUMA nodes, defer
1744 * hash allocation until vmalloc space is available.
1750 alloc_large_system_hash("Inode-cache",
1751 sizeof(struct hlist_head
),
1760 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1761 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1764 void __init
inode_init(void)
1768 /* inode slab cache */
1769 inode_cachep
= kmem_cache_create("inode_cache",
1770 sizeof(struct inode
),
1772 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1776 /* Hash may have been set up in inode_init_early */
1781 alloc_large_system_hash("Inode-cache",
1782 sizeof(struct hlist_head
),
1791 for (loop
= 0; loop
< (1U << i_hash_shift
); loop
++)
1792 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1795 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1797 inode
->i_mode
= mode
;
1798 if (S_ISCHR(mode
)) {
1799 inode
->i_fop
= &def_chr_fops
;
1800 inode
->i_rdev
= rdev
;
1801 } else if (S_ISBLK(mode
)) {
1802 inode
->i_fop
= &def_blk_fops
;
1803 inode
->i_rdev
= rdev
;
1804 } else if (S_ISFIFO(mode
))
1805 inode
->i_fop
= &pipefifo_fops
;
1806 else if (S_ISSOCK(mode
))
1807 ; /* leave it no_open_fops */
1809 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1810 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1813 EXPORT_SYMBOL(init_special_inode
);
1816 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1818 * @dir: Directory inode
1819 * @mode: mode of the new inode
1821 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1824 inode
->i_uid
= current_fsuid();
1825 if (dir
&& dir
->i_mode
& S_ISGID
) {
1826 inode
->i_gid
= dir
->i_gid
;
1830 inode
->i_gid
= current_fsgid();
1831 inode
->i_mode
= mode
;
1833 EXPORT_SYMBOL(inode_init_owner
);
1836 * inode_owner_or_capable - check current task permissions to inode
1837 * @inode: inode being checked
1839 * Return true if current either has CAP_FOWNER in a namespace with the
1840 * inode owner uid mapped, or owns the file.
1842 bool inode_owner_or_capable(const struct inode
*inode
)
1844 struct user_namespace
*ns
;
1846 if (uid_eq(current_fsuid(), inode
->i_uid
))
1849 ns
= current_user_ns();
1850 if (ns_capable(ns
, CAP_FOWNER
) && kuid_has_mapping(ns
, inode
->i_uid
))
1854 EXPORT_SYMBOL(inode_owner_or_capable
);
1857 * Direct i/o helper functions
1859 static void __inode_dio_wait(struct inode
*inode
)
1861 wait_queue_head_t
*wq
= bit_waitqueue(&inode
->i_state
, __I_DIO_WAKEUP
);
1862 DEFINE_WAIT_BIT(q
, &inode
->i_state
, __I_DIO_WAKEUP
);
1865 prepare_to_wait(wq
, &q
.wait
, TASK_UNINTERRUPTIBLE
);
1866 if (atomic_read(&inode
->i_dio_count
))
1868 } while (atomic_read(&inode
->i_dio_count
));
1869 finish_wait(wq
, &q
.wait
);
1873 * inode_dio_wait - wait for outstanding DIO requests to finish
1874 * @inode: inode to wait for
1876 * Waits for all pending direct I/O requests to finish so that we can
1877 * proceed with a truncate or equivalent operation.
1879 * Must be called under a lock that serializes taking new references
1880 * to i_dio_count, usually by inode->i_mutex.
1882 void inode_dio_wait(struct inode
*inode
)
1884 if (atomic_read(&inode
->i_dio_count
))
1885 __inode_dio_wait(inode
);
1887 EXPORT_SYMBOL(inode_dio_wait
);
1890 * inode_dio_done - signal finish of a direct I/O requests
1891 * @inode: inode the direct I/O happens on
1893 * This is called once we've finished processing a direct I/O request,
1894 * and is used to wake up callers waiting for direct I/O to be quiesced.
1896 void inode_dio_done(struct inode
*inode
)
1898 if (atomic_dec_and_test(&inode
->i_dio_count
))
1899 wake_up_bit(&inode
->i_state
, __I_DIO_WAKEUP
);
1901 EXPORT_SYMBOL(inode_dio_done
);
1904 * inode_set_flags - atomically set some inode flags
1906 * Note: the caller should be holding i_mutex, or else be sure that
1907 * they have exclusive access to the inode structure (i.e., while the
1908 * inode is being instantiated). The reason for the cmpxchg() loop
1909 * --- which wouldn't be necessary if all code paths which modify
1910 * i_flags actually followed this rule, is that there is at least one
1911 * code path which doesn't today --- for example,
1912 * __generic_file_aio_write() calls file_remove_suid() without holding
1913 * i_mutex --- so we use cmpxchg() out of an abundance of caution.
1915 * In the long run, i_mutex is overkill, and we should probably look
1916 * at using the i_lock spinlock to protect i_flags, and then make sure
1917 * it is so documented in include/linux/fs.h and that all code follows
1918 * the locking convention!!
1920 void inode_set_flags(struct inode
*inode
, unsigned int flags
,
1923 unsigned int old_flags
, new_flags
;
1925 WARN_ON_ONCE(flags
& ~mask
);
1927 old_flags
= ACCESS_ONCE(inode
->i_flags
);
1928 new_flags
= (old_flags
& ~mask
) | flags
;
1929 } while (unlikely(cmpxchg(&inode
->i_flags
, old_flags
,
1930 new_flags
) != old_flags
));
1932 EXPORT_SYMBOL(inode_set_flags
);