1 // SPDX-License-Identifier: GPL-2.0-only
3 * (C) 1997 Linus Torvalds
4 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
6 #include <linux/export.h>
8 #include <linux/filelock.h>
10 #include <linux/backing-dev.h>
11 #include <linux/hash.h>
12 #include <linux/swap.h>
13 #include <linux/security.h>
14 #include <linux/cdev.h>
15 #include <linux/memblock.h>
16 #include <linux/fsnotify.h>
17 #include <linux/mount.h>
18 #include <linux/posix_acl.h>
19 #include <linux/buffer_head.h> /* for inode_has_buffers */
20 #include <linux/ratelimit.h>
21 #include <linux/list_lru.h>
22 #include <linux/iversion.h>
23 #include <trace/events/writeback.h>
27 * Inode locking rules:
29 * inode->i_lock protects:
30 * inode->i_state, inode->i_hash, __iget(), inode->i_io_list
31 * Inode LRU list locks protect:
32 * inode->i_sb->s_inode_lru, inode->i_lru
33 * inode->i_sb->s_inode_list_lock protects:
34 * inode->i_sb->s_inodes, inode->i_sb_list
35 * bdi->wb.list_lock protects:
36 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
37 * inode_hash_lock protects:
38 * inode_hashtable, inode->i_hash
42 * inode->i_sb->s_inode_list_lock
44 * Inode LRU list locks
50 * inode->i_sb->s_inode_list_lock
57 static unsigned int i_hash_mask __ro_after_init
;
58 static unsigned int i_hash_shift __ro_after_init
;
59 static struct hlist_head
*inode_hashtable __ro_after_init
;
60 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(inode_hash_lock
);
63 * Empty aops. Can be used for the cases where the user does not
64 * define any of the address_space operations.
66 const struct address_space_operations empty_aops
= {
68 EXPORT_SYMBOL(empty_aops
);
70 static DEFINE_PER_CPU(unsigned long, nr_inodes
);
71 static DEFINE_PER_CPU(unsigned long, nr_unused
);
73 static struct kmem_cache
*inode_cachep __ro_after_init
;
75 static long get_nr_inodes(void)
79 for_each_possible_cpu(i
)
80 sum
+= per_cpu(nr_inodes
, i
);
81 return sum
< 0 ? 0 : sum
;
84 static inline long get_nr_inodes_unused(void)
88 for_each_possible_cpu(i
)
89 sum
+= per_cpu(nr_unused
, i
);
90 return sum
< 0 ? 0 : sum
;
93 long get_nr_dirty_inodes(void)
95 /* not actually dirty inodes, but a wild approximation */
96 long nr_dirty
= get_nr_inodes() - get_nr_inodes_unused();
97 return nr_dirty
> 0 ? nr_dirty
: 0;
101 * Handle nr_inode sysctl
105 * Statistics gathering..
107 static struct inodes_stat_t inodes_stat
;
109 static int proc_nr_inodes(struct ctl_table
*table
, int write
, void *buffer
,
110 size_t *lenp
, loff_t
*ppos
)
112 inodes_stat
.nr_inodes
= get_nr_inodes();
113 inodes_stat
.nr_unused
= get_nr_inodes_unused();
114 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
117 static struct ctl_table inodes_sysctls
[] = {
119 .procname
= "inode-nr",
120 .data
= &inodes_stat
,
121 .maxlen
= 2*sizeof(long),
123 .proc_handler
= proc_nr_inodes
,
126 .procname
= "inode-state",
127 .data
= &inodes_stat
,
128 .maxlen
= 7*sizeof(long),
130 .proc_handler
= proc_nr_inodes
,
134 static int __init
init_fs_inode_sysctls(void)
136 register_sysctl_init("fs", inodes_sysctls
);
139 early_initcall(init_fs_inode_sysctls
);
142 static int no_open(struct inode
*inode
, struct file
*file
)
148 * inode_init_always - perform inode structure initialisation
149 * @sb: superblock inode belongs to
150 * @inode: inode to initialise
152 * These are initializations that need to be done on every inode
153 * allocation as the fields are not initialised by slab allocation.
155 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
157 static const struct inode_operations empty_iops
;
158 static const struct file_operations no_open_fops
= {.open
= no_open
};
159 struct address_space
*const mapping
= &inode
->i_data
;
162 inode
->i_blkbits
= sb
->s_blocksize_bits
;
164 atomic64_set(&inode
->i_sequence
, 0);
165 atomic_set(&inode
->i_count
, 1);
166 inode
->i_op
= &empty_iops
;
167 inode
->i_fop
= &no_open_fops
;
169 inode
->__i_nlink
= 1;
170 inode
->i_opflags
= 0;
172 inode
->i_opflags
|= IOP_XATTR
;
173 i_uid_write(inode
, 0);
174 i_gid_write(inode
, 0);
175 atomic_set(&inode
->i_writecount
, 0);
177 inode
->i_write_hint
= WRITE_LIFE_NOT_SET
;
180 inode
->i_generation
= 0;
181 inode
->i_pipe
= NULL
;
182 inode
->i_cdev
= NULL
;
183 inode
->i_link
= NULL
;
184 inode
->i_dir_seq
= 0;
186 inode
->dirtied_when
= 0;
188 #ifdef CONFIG_CGROUP_WRITEBACK
189 inode
->i_wb_frn_winner
= 0;
190 inode
->i_wb_frn_avg_time
= 0;
191 inode
->i_wb_frn_history
= 0;
194 spin_lock_init(&inode
->i_lock
);
195 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
197 init_rwsem(&inode
->i_rwsem
);
198 lockdep_set_class(&inode
->i_rwsem
, &sb
->s_type
->i_mutex_key
);
200 atomic_set(&inode
->i_dio_count
, 0);
202 mapping
->a_ops
= &empty_aops
;
203 mapping
->host
= inode
;
206 atomic_set(&mapping
->i_mmap_writable
, 0);
207 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
208 atomic_set(&mapping
->nr_thps
, 0);
210 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
211 mapping
->i_private_data
= NULL
;
212 mapping
->writeback_index
= 0;
213 init_rwsem(&mapping
->invalidate_lock
);
214 lockdep_set_class_and_name(&mapping
->invalidate_lock
,
215 &sb
->s_type
->invalidate_lock_key
,
216 "mapping.invalidate_lock");
217 if (sb
->s_iflags
& SB_I_STABLE_WRITES
)
218 mapping_set_stable_writes(mapping
);
219 inode
->i_private
= NULL
;
220 inode
->i_mapping
= mapping
;
221 INIT_HLIST_HEAD(&inode
->i_dentry
); /* buggered by rcu freeing */
222 #ifdef CONFIG_FS_POSIX_ACL
223 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
226 #ifdef CONFIG_FSNOTIFY
227 inode
->i_fsnotify_mask
= 0;
229 inode
->i_flctx
= NULL
;
231 if (unlikely(security_inode_alloc(inode
)))
233 this_cpu_inc(nr_inodes
);
237 EXPORT_SYMBOL(inode_init_always
);
239 void free_inode_nonrcu(struct inode
*inode
)
241 kmem_cache_free(inode_cachep
, inode
);
243 EXPORT_SYMBOL(free_inode_nonrcu
);
245 static void i_callback(struct rcu_head
*head
)
247 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
248 if (inode
->free_inode
)
249 inode
->free_inode(inode
);
251 free_inode_nonrcu(inode
);
254 static struct inode
*alloc_inode(struct super_block
*sb
)
256 const struct super_operations
*ops
= sb
->s_op
;
259 if (ops
->alloc_inode
)
260 inode
= ops
->alloc_inode(sb
);
262 inode
= alloc_inode_sb(sb
, inode_cachep
, GFP_KERNEL
);
267 if (unlikely(inode_init_always(sb
, inode
))) {
268 if (ops
->destroy_inode
) {
269 ops
->destroy_inode(inode
);
270 if (!ops
->free_inode
)
273 inode
->free_inode
= ops
->free_inode
;
274 i_callback(&inode
->i_rcu
);
281 void __destroy_inode(struct inode
*inode
)
283 BUG_ON(inode_has_buffers(inode
));
284 inode_detach_wb(inode
);
285 security_inode_free(inode
);
286 fsnotify_inode_delete(inode
);
287 locks_free_lock_context(inode
);
288 if (!inode
->i_nlink
) {
289 WARN_ON(atomic_long_read(&inode
->i_sb
->s_remove_count
) == 0);
290 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
293 #ifdef CONFIG_FS_POSIX_ACL
294 if (inode
->i_acl
&& !is_uncached_acl(inode
->i_acl
))
295 posix_acl_release(inode
->i_acl
);
296 if (inode
->i_default_acl
&& !is_uncached_acl(inode
->i_default_acl
))
297 posix_acl_release(inode
->i_default_acl
);
299 this_cpu_dec(nr_inodes
);
301 EXPORT_SYMBOL(__destroy_inode
);
303 static void destroy_inode(struct inode
*inode
)
305 const struct super_operations
*ops
= inode
->i_sb
->s_op
;
307 BUG_ON(!list_empty(&inode
->i_lru
));
308 __destroy_inode(inode
);
309 if (ops
->destroy_inode
) {
310 ops
->destroy_inode(inode
);
311 if (!ops
->free_inode
)
314 inode
->free_inode
= ops
->free_inode
;
315 call_rcu(&inode
->i_rcu
, i_callback
);
319 * drop_nlink - directly drop an inode's link count
322 * This is a low-level filesystem helper to replace any
323 * direct filesystem manipulation of i_nlink. In cases
324 * where we are attempting to track writes to the
325 * filesystem, a decrement to zero means an imminent
326 * write when the file is truncated and actually unlinked
329 void drop_nlink(struct inode
*inode
)
331 WARN_ON(inode
->i_nlink
== 0);
334 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
336 EXPORT_SYMBOL(drop_nlink
);
339 * clear_nlink - directly zero an inode's link count
342 * This is a low-level filesystem helper to replace any
343 * direct filesystem manipulation of i_nlink. See
344 * drop_nlink() for why we care about i_nlink hitting zero.
346 void clear_nlink(struct inode
*inode
)
348 if (inode
->i_nlink
) {
349 inode
->__i_nlink
= 0;
350 atomic_long_inc(&inode
->i_sb
->s_remove_count
);
353 EXPORT_SYMBOL(clear_nlink
);
356 * set_nlink - directly set an inode's link count
358 * @nlink: new nlink (should be non-zero)
360 * This is a low-level filesystem helper to replace any
361 * direct filesystem manipulation of i_nlink.
363 void set_nlink(struct inode
*inode
, unsigned int nlink
)
368 /* Yes, some filesystems do change nlink from zero to one */
369 if (inode
->i_nlink
== 0)
370 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
372 inode
->__i_nlink
= nlink
;
375 EXPORT_SYMBOL(set_nlink
);
378 * inc_nlink - directly increment an inode's link count
381 * This is a low-level filesystem helper to replace any
382 * direct filesystem manipulation of i_nlink. Currently,
383 * it is only here for parity with dec_nlink().
385 void inc_nlink(struct inode
*inode
)
387 if (unlikely(inode
->i_nlink
== 0)) {
388 WARN_ON(!(inode
->i_state
& I_LINKABLE
));
389 atomic_long_dec(&inode
->i_sb
->s_remove_count
);
394 EXPORT_SYMBOL(inc_nlink
);
396 static void __address_space_init_once(struct address_space
*mapping
)
398 xa_init_flags(&mapping
->i_pages
, XA_FLAGS_LOCK_IRQ
| XA_FLAGS_ACCOUNT
);
399 init_rwsem(&mapping
->i_mmap_rwsem
);
400 INIT_LIST_HEAD(&mapping
->i_private_list
);
401 spin_lock_init(&mapping
->i_private_lock
);
402 mapping
->i_mmap
= RB_ROOT_CACHED
;
405 void address_space_init_once(struct address_space
*mapping
)
407 memset(mapping
, 0, sizeof(*mapping
));
408 __address_space_init_once(mapping
);
410 EXPORT_SYMBOL(address_space_init_once
);
413 * These are initializations that only need to be done
414 * once, because the fields are idempotent across use
415 * of the inode, so let the slab aware of that.
417 void inode_init_once(struct inode
*inode
)
419 memset(inode
, 0, sizeof(*inode
));
420 INIT_HLIST_NODE(&inode
->i_hash
);
421 INIT_LIST_HEAD(&inode
->i_devices
);
422 INIT_LIST_HEAD(&inode
->i_io_list
);
423 INIT_LIST_HEAD(&inode
->i_wb_list
);
424 INIT_LIST_HEAD(&inode
->i_lru
);
425 INIT_LIST_HEAD(&inode
->i_sb_list
);
426 __address_space_init_once(&inode
->i_data
);
427 i_size_ordered_init(inode
);
429 EXPORT_SYMBOL(inode_init_once
);
431 static void init_once(void *foo
)
433 struct inode
*inode
= (struct inode
*) foo
;
435 inode_init_once(inode
);
439 * inode->i_lock must be held
441 void __iget(struct inode
*inode
)
443 atomic_inc(&inode
->i_count
);
447 * get additional reference to inode; caller must already hold one.
449 void ihold(struct inode
*inode
)
451 WARN_ON(atomic_inc_return(&inode
->i_count
) < 2);
453 EXPORT_SYMBOL(ihold
);
455 static void __inode_add_lru(struct inode
*inode
, bool rotate
)
457 if (inode
->i_state
& (I_DIRTY_ALL
| I_SYNC
| I_FREEING
| I_WILL_FREE
))
459 if (atomic_read(&inode
->i_count
))
461 if (!(inode
->i_sb
->s_flags
& SB_ACTIVE
))
463 if (!mapping_shrinkable(&inode
->i_data
))
466 if (list_lru_add_obj(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
467 this_cpu_inc(nr_unused
);
469 inode
->i_state
|= I_REFERENCED
;
473 * Add inode to LRU if needed (inode is unused and clean).
475 * Needs inode->i_lock held.
477 void inode_add_lru(struct inode
*inode
)
479 __inode_add_lru(inode
, false);
482 static void inode_lru_list_del(struct inode
*inode
)
484 if (list_lru_del_obj(&inode
->i_sb
->s_inode_lru
, &inode
->i_lru
))
485 this_cpu_dec(nr_unused
);
489 * inode_sb_list_add - add inode to the superblock list of inodes
490 * @inode: inode to add
492 void inode_sb_list_add(struct inode
*inode
)
494 spin_lock(&inode
->i_sb
->s_inode_list_lock
);
495 list_add(&inode
->i_sb_list
, &inode
->i_sb
->s_inodes
);
496 spin_unlock(&inode
->i_sb
->s_inode_list_lock
);
498 EXPORT_SYMBOL_GPL(inode_sb_list_add
);
500 static inline void inode_sb_list_del(struct inode
*inode
)
502 if (!list_empty(&inode
->i_sb_list
)) {
503 spin_lock(&inode
->i_sb
->s_inode_list_lock
);
504 list_del_init(&inode
->i_sb_list
);
505 spin_unlock(&inode
->i_sb
->s_inode_list_lock
);
509 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
513 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
515 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> i_hash_shift
);
516 return tmp
& i_hash_mask
;
520 * __insert_inode_hash - hash an inode
521 * @inode: unhashed inode
522 * @hashval: unsigned long value used to locate this object in the
525 * Add an inode to the inode hash for this superblock.
527 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
529 struct hlist_head
*b
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
531 spin_lock(&inode_hash_lock
);
532 spin_lock(&inode
->i_lock
);
533 hlist_add_head_rcu(&inode
->i_hash
, b
);
534 spin_unlock(&inode
->i_lock
);
535 spin_unlock(&inode_hash_lock
);
537 EXPORT_SYMBOL(__insert_inode_hash
);
540 * __remove_inode_hash - remove an inode from the hash
541 * @inode: inode to unhash
543 * Remove an inode from the superblock.
545 void __remove_inode_hash(struct inode
*inode
)
547 spin_lock(&inode_hash_lock
);
548 spin_lock(&inode
->i_lock
);
549 hlist_del_init_rcu(&inode
->i_hash
);
550 spin_unlock(&inode
->i_lock
);
551 spin_unlock(&inode_hash_lock
);
553 EXPORT_SYMBOL(__remove_inode_hash
);
555 void dump_mapping(const struct address_space
*mapping
)
558 const struct address_space_operations
*a_ops
;
559 struct hlist_node
*dentry_first
;
560 struct dentry
*dentry_ptr
;
561 struct dentry dentry
;
565 * If mapping is an invalid pointer, we don't want to crash
566 * accessing it, so probe everything depending on it carefully.
568 if (get_kernel_nofault(host
, &mapping
->host
) ||
569 get_kernel_nofault(a_ops
, &mapping
->a_ops
)) {
570 pr_warn("invalid mapping:%px\n", mapping
);
575 pr_warn("aops:%ps\n", a_ops
);
579 if (get_kernel_nofault(dentry_first
, &host
->i_dentry
.first
) ||
580 get_kernel_nofault(ino
, &host
->i_ino
)) {
581 pr_warn("aops:%ps invalid inode:%px\n", a_ops
, host
);
586 pr_warn("aops:%ps ino:%lx\n", a_ops
, ino
);
590 dentry_ptr
= container_of(dentry_first
, struct dentry
, d_u
.d_alias
);
591 if (get_kernel_nofault(dentry
, dentry_ptr
)) {
592 pr_warn("aops:%ps ino:%lx invalid dentry:%px\n",
593 a_ops
, ino
, dentry_ptr
);
598 * if dentry is corrupted, the %pd handler may still crash,
599 * but it's unlikely that we reach here with a corrupt mapping
601 pr_warn("aops:%ps ino:%lx dentry name:\"%pd\"\n", a_ops
, ino
, &dentry
);
604 void clear_inode(struct inode
*inode
)
607 * We have to cycle the i_pages lock here because reclaim can be in the
608 * process of removing the last page (in __filemap_remove_folio())
609 * and we must not free the mapping under it.
611 xa_lock_irq(&inode
->i_data
.i_pages
);
612 BUG_ON(inode
->i_data
.nrpages
);
614 * Almost always, mapping_empty(&inode->i_data) here; but there are
615 * two known and long-standing ways in which nodes may get left behind
616 * (when deep radix-tree node allocation failed partway; or when THP
617 * collapse_file() failed). Until those two known cases are cleaned up,
618 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
619 * nor even WARN_ON(!mapping_empty).
621 xa_unlock_irq(&inode
->i_data
.i_pages
);
622 BUG_ON(!list_empty(&inode
->i_data
.i_private_list
));
623 BUG_ON(!(inode
->i_state
& I_FREEING
));
624 BUG_ON(inode
->i_state
& I_CLEAR
);
625 BUG_ON(!list_empty(&inode
->i_wb_list
));
626 /* don't need i_lock here, no concurrent mods to i_state */
627 inode
->i_state
= I_FREEING
| I_CLEAR
;
629 EXPORT_SYMBOL(clear_inode
);
632 * Free the inode passed in, removing it from the lists it is still connected
633 * to. We remove any pages still attached to the inode and wait for any IO that
634 * is still in progress before finally destroying the inode.
636 * An inode must already be marked I_FREEING so that we avoid the inode being
637 * moved back onto lists if we race with other code that manipulates the lists
638 * (e.g. writeback_single_inode). The caller is responsible for setting this.
640 * An inode must already be removed from the LRU list before being evicted from
641 * the cache. This should occur atomically with setting the I_FREEING state
642 * flag, so no inodes here should ever be on the LRU when being evicted.
644 static void evict(struct inode
*inode
)
646 const struct super_operations
*op
= inode
->i_sb
->s_op
;
648 BUG_ON(!(inode
->i_state
& I_FREEING
));
649 BUG_ON(!list_empty(&inode
->i_lru
));
651 if (!list_empty(&inode
->i_io_list
))
652 inode_io_list_del(inode
);
654 inode_sb_list_del(inode
);
657 * Wait for flusher thread to be done with the inode so that filesystem
658 * does not start destroying it while writeback is still running. Since
659 * the inode has I_FREEING set, flusher thread won't start new work on
660 * the inode. We just have to wait for running writeback to finish.
662 inode_wait_for_writeback(inode
);
664 if (op
->evict_inode
) {
665 op
->evict_inode(inode
);
667 truncate_inode_pages_final(&inode
->i_data
);
670 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
673 remove_inode_hash(inode
);
675 spin_lock(&inode
->i_lock
);
676 wake_up_bit(&inode
->i_state
, __I_NEW
);
677 BUG_ON(inode
->i_state
!= (I_FREEING
| I_CLEAR
));
678 spin_unlock(&inode
->i_lock
);
680 destroy_inode(inode
);
684 * dispose_list - dispose of the contents of a local list
685 * @head: the head of the list to free
687 * Dispose-list gets a local list with local inodes in it, so it doesn't
688 * need to worry about list corruption and SMP locks.
690 static void dispose_list(struct list_head
*head
)
692 while (!list_empty(head
)) {
695 inode
= list_first_entry(head
, struct inode
, i_lru
);
696 list_del_init(&inode
->i_lru
);
704 * evict_inodes - evict all evictable inodes for a superblock
705 * @sb: superblock to operate on
707 * Make sure that no inodes with zero refcount are retained. This is
708 * called by superblock shutdown after having SB_ACTIVE flag removed,
709 * so any inode reaching zero refcount during or after that call will
710 * be immediately evicted.
712 void evict_inodes(struct super_block
*sb
)
714 struct inode
*inode
, *next
;
718 spin_lock(&sb
->s_inode_list_lock
);
719 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
720 if (atomic_read(&inode
->i_count
))
723 spin_lock(&inode
->i_lock
);
724 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
725 spin_unlock(&inode
->i_lock
);
729 inode
->i_state
|= I_FREEING
;
730 inode_lru_list_del(inode
);
731 spin_unlock(&inode
->i_lock
);
732 list_add(&inode
->i_lru
, &dispose
);
735 * We can have a ton of inodes to evict at unmount time given
736 * enough memory, check to see if we need to go to sleep for a
737 * bit so we don't livelock.
739 if (need_resched()) {
740 spin_unlock(&sb
->s_inode_list_lock
);
742 dispose_list(&dispose
);
746 spin_unlock(&sb
->s_inode_list_lock
);
748 dispose_list(&dispose
);
750 EXPORT_SYMBOL_GPL(evict_inodes
);
753 * invalidate_inodes - attempt to free all inodes on a superblock
754 * @sb: superblock to operate on
756 * Attempts to free all inodes (including dirty inodes) for a given superblock.
758 void invalidate_inodes(struct super_block
*sb
)
760 struct inode
*inode
, *next
;
764 spin_lock(&sb
->s_inode_list_lock
);
765 list_for_each_entry_safe(inode
, next
, &sb
->s_inodes
, i_sb_list
) {
766 spin_lock(&inode
->i_lock
);
767 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
768 spin_unlock(&inode
->i_lock
);
771 if (atomic_read(&inode
->i_count
)) {
772 spin_unlock(&inode
->i_lock
);
776 inode
->i_state
|= I_FREEING
;
777 inode_lru_list_del(inode
);
778 spin_unlock(&inode
->i_lock
);
779 list_add(&inode
->i_lru
, &dispose
);
780 if (need_resched()) {
781 spin_unlock(&sb
->s_inode_list_lock
);
783 dispose_list(&dispose
);
787 spin_unlock(&sb
->s_inode_list_lock
);
789 dispose_list(&dispose
);
793 * Isolate the inode from the LRU in preparation for freeing it.
795 * If the inode has the I_REFERENCED flag set, then it means that it has been
796 * used recently - the flag is set in iput_final(). When we encounter such an
797 * inode, clear the flag and move it to the back of the LRU so it gets another
798 * pass through the LRU before it gets reclaimed. This is necessary because of
799 * the fact we are doing lazy LRU updates to minimise lock contention so the
800 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
801 * with this flag set because they are the inodes that are out of order.
803 static enum lru_status
inode_lru_isolate(struct list_head
*item
,
804 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
806 struct list_head
*freeable
= arg
;
807 struct inode
*inode
= container_of(item
, struct inode
, i_lru
);
810 * We are inverting the lru lock/inode->i_lock here, so use a
811 * trylock. If we fail to get the lock, just skip it.
813 if (!spin_trylock(&inode
->i_lock
))
817 * Inodes can get referenced, redirtied, or repopulated while
818 * they're already on the LRU, and this can make them
819 * unreclaimable for a while. Remove them lazily here; iput,
820 * sync, or the last page cache deletion will requeue them.
822 if (atomic_read(&inode
->i_count
) ||
823 (inode
->i_state
& ~I_REFERENCED
) ||
824 !mapping_shrinkable(&inode
->i_data
)) {
825 list_lru_isolate(lru
, &inode
->i_lru
);
826 spin_unlock(&inode
->i_lock
);
827 this_cpu_dec(nr_unused
);
831 /* Recently referenced inodes get one more pass */
832 if (inode
->i_state
& I_REFERENCED
) {
833 inode
->i_state
&= ~I_REFERENCED
;
834 spin_unlock(&inode
->i_lock
);
839 * On highmem systems, mapping_shrinkable() permits dropping
840 * page cache in order to free up struct inodes: lowmem might
841 * be under pressure before the cache inside the highmem zone.
843 if (inode_has_buffers(inode
) || !mapping_empty(&inode
->i_data
)) {
845 spin_unlock(&inode
->i_lock
);
846 spin_unlock(lru_lock
);
847 if (remove_inode_buffers(inode
)) {
849 reap
= invalidate_mapping_pages(&inode
->i_data
, 0, -1);
850 if (current_is_kswapd())
851 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
853 __count_vm_events(PGINODESTEAL
, reap
);
854 mm_account_reclaimed_pages(reap
);
861 WARN_ON(inode
->i_state
& I_NEW
);
862 inode
->i_state
|= I_FREEING
;
863 list_lru_isolate_move(lru
, &inode
->i_lru
, freeable
);
864 spin_unlock(&inode
->i_lock
);
866 this_cpu_dec(nr_unused
);
871 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
872 * This is called from the superblock shrinker function with a number of inodes
873 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
874 * then are freed outside inode_lock by dispose_list().
876 long prune_icache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
881 freed
= list_lru_shrink_walk(&sb
->s_inode_lru
, sc
,
882 inode_lru_isolate
, &freeable
);
883 dispose_list(&freeable
);
887 static void __wait_on_freeing_inode(struct inode
*inode
);
889 * Called with the inode lock held.
891 static struct inode
*find_inode(struct super_block
*sb
,
892 struct hlist_head
*head
,
893 int (*test
)(struct inode
*, void *),
896 struct inode
*inode
= NULL
;
899 hlist_for_each_entry(inode
, head
, i_hash
) {
900 if (inode
->i_sb
!= sb
)
902 if (!test(inode
, data
))
904 spin_lock(&inode
->i_lock
);
905 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
906 __wait_on_freeing_inode(inode
);
909 if (unlikely(inode
->i_state
& I_CREATING
)) {
910 spin_unlock(&inode
->i_lock
);
911 return ERR_PTR(-ESTALE
);
914 spin_unlock(&inode
->i_lock
);
921 * find_inode_fast is the fast path version of find_inode, see the comment at
922 * iget_locked for details.
924 static struct inode
*find_inode_fast(struct super_block
*sb
,
925 struct hlist_head
*head
, unsigned long ino
)
927 struct inode
*inode
= NULL
;
930 hlist_for_each_entry(inode
, head
, i_hash
) {
931 if (inode
->i_ino
!= ino
)
933 if (inode
->i_sb
!= sb
)
935 spin_lock(&inode
->i_lock
);
936 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
937 __wait_on_freeing_inode(inode
);
940 if (unlikely(inode
->i_state
& I_CREATING
)) {
941 spin_unlock(&inode
->i_lock
);
942 return ERR_PTR(-ESTALE
);
945 spin_unlock(&inode
->i_lock
);
952 * Each cpu owns a range of LAST_INO_BATCH numbers.
953 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
954 * to renew the exhausted range.
956 * This does not significantly increase overflow rate because every CPU can
957 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
958 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
959 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
960 * overflow rate by 2x, which does not seem too significant.
962 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
963 * error if st_ino won't fit in target struct field. Use 32bit counter
964 * here to attempt to avoid that.
966 #define LAST_INO_BATCH 1024
967 static DEFINE_PER_CPU(unsigned int, last_ino
);
969 unsigned int get_next_ino(void)
971 unsigned int *p
= &get_cpu_var(last_ino
);
972 unsigned int res
= *p
;
975 if (unlikely((res
& (LAST_INO_BATCH
-1)) == 0)) {
976 static atomic_t shared_last_ino
;
977 int next
= atomic_add_return(LAST_INO_BATCH
, &shared_last_ino
);
979 res
= next
- LAST_INO_BATCH
;
984 /* get_next_ino should not provide a 0 inode number */
988 put_cpu_var(last_ino
);
991 EXPORT_SYMBOL(get_next_ino
);
994 * new_inode_pseudo - obtain an inode
997 * Allocates a new inode for given superblock.
998 * Inode wont be chained in superblock s_inodes list
1000 * - fs can't be unmount
1001 * - quotas, fsnotify, writeback can't work
1003 struct inode
*new_inode_pseudo(struct super_block
*sb
)
1005 struct inode
*inode
= alloc_inode(sb
);
1008 spin_lock(&inode
->i_lock
);
1010 spin_unlock(&inode
->i_lock
);
1016 * new_inode - obtain an inode
1019 * Allocates a new inode for given superblock. The default gfp_mask
1020 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
1021 * If HIGHMEM pages are unsuitable or it is known that pages allocated
1022 * for the page cache are not reclaimable or migratable,
1023 * mapping_set_gfp_mask() must be called with suitable flags on the
1024 * newly created inode's mapping
1027 struct inode
*new_inode(struct super_block
*sb
)
1029 struct inode
*inode
;
1031 inode
= new_inode_pseudo(sb
);
1033 inode_sb_list_add(inode
);
1036 EXPORT_SYMBOL(new_inode
);
1038 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1039 void lockdep_annotate_inode_mutex_key(struct inode
*inode
)
1041 if (S_ISDIR(inode
->i_mode
)) {
1042 struct file_system_type
*type
= inode
->i_sb
->s_type
;
1044 /* Set new key only if filesystem hasn't already changed it */
1045 if (lockdep_match_class(&inode
->i_rwsem
, &type
->i_mutex_key
)) {
1047 * ensure nobody is actually holding i_mutex
1049 // mutex_destroy(&inode->i_mutex);
1050 init_rwsem(&inode
->i_rwsem
);
1051 lockdep_set_class(&inode
->i_rwsem
,
1052 &type
->i_mutex_dir_key
);
1056 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key
);
1060 * unlock_new_inode - clear the I_NEW state and wake up any waiters
1061 * @inode: new inode to unlock
1063 * Called when the inode is fully initialised to clear the new state of the
1064 * inode and wake up anyone waiting for the inode to finish initialisation.
1066 void unlock_new_inode(struct inode
*inode
)
1068 lockdep_annotate_inode_mutex_key(inode
);
1069 spin_lock(&inode
->i_lock
);
1070 WARN_ON(!(inode
->i_state
& I_NEW
));
1071 inode
->i_state
&= ~I_NEW
& ~I_CREATING
;
1073 wake_up_bit(&inode
->i_state
, __I_NEW
);
1074 spin_unlock(&inode
->i_lock
);
1076 EXPORT_SYMBOL(unlock_new_inode
);
1078 void discard_new_inode(struct inode
*inode
)
1080 lockdep_annotate_inode_mutex_key(inode
);
1081 spin_lock(&inode
->i_lock
);
1082 WARN_ON(!(inode
->i_state
& I_NEW
));
1083 inode
->i_state
&= ~I_NEW
;
1085 wake_up_bit(&inode
->i_state
, __I_NEW
);
1086 spin_unlock(&inode
->i_lock
);
1089 EXPORT_SYMBOL(discard_new_inode
);
1092 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1094 * Lock any non-NULL argument. Passed objects must not be directories.
1095 * Zero, one or two objects may be locked by this function.
1097 * @inode1: first inode to lock
1098 * @inode2: second inode to lock
1100 void lock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
1103 WARN_ON_ONCE(S_ISDIR(inode1
->i_mode
));
1105 WARN_ON_ONCE(S_ISDIR(inode2
->i_mode
));
1106 if (inode1
> inode2
)
1107 swap(inode1
, inode2
);
1110 if (inode2
&& inode2
!= inode1
)
1111 inode_lock_nested(inode2
, I_MUTEX_NONDIR2
);
1113 EXPORT_SYMBOL(lock_two_nondirectories
);
1116 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1117 * @inode1: first inode to unlock
1118 * @inode2: second inode to unlock
1120 void unlock_two_nondirectories(struct inode
*inode1
, struct inode
*inode2
)
1123 WARN_ON_ONCE(S_ISDIR(inode1
->i_mode
));
1124 inode_unlock(inode1
);
1126 if (inode2
&& inode2
!= inode1
) {
1127 WARN_ON_ONCE(S_ISDIR(inode2
->i_mode
));
1128 inode_unlock(inode2
);
1131 EXPORT_SYMBOL(unlock_two_nondirectories
);
1134 * inode_insert5 - obtain an inode from a mounted file system
1135 * @inode: pre-allocated inode to use for insert to cache
1136 * @hashval: hash value (usually inode number) to get
1137 * @test: callback used for comparisons between inodes
1138 * @set: callback used to initialize a new struct inode
1139 * @data: opaque data pointer to pass to @test and @set
1141 * Search for the inode specified by @hashval and @data in the inode cache,
1142 * and if present it is return it with an increased reference count. This is
1143 * a variant of iget5_locked() for callers that don't want to fail on memory
1144 * allocation of inode.
1146 * If the inode is not in cache, insert the pre-allocated inode to cache and
1147 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1148 * to fill it in before unlocking it via unlock_new_inode().
1150 * Note both @test and @set are called with the inode_hash_lock held, so can't
1153 struct inode
*inode_insert5(struct inode
*inode
, unsigned long hashval
,
1154 int (*test
)(struct inode
*, void *),
1155 int (*set
)(struct inode
*, void *), void *data
)
1157 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1161 spin_lock(&inode_hash_lock
);
1162 old
= find_inode(inode
->i_sb
, head
, test
, data
);
1163 if (unlikely(old
)) {
1165 * Uhhuh, somebody else created the same inode under us.
1166 * Use the old inode instead of the preallocated one.
1168 spin_unlock(&inode_hash_lock
);
1172 if (unlikely(inode_unhashed(old
))) {
1179 if (set
&& unlikely(set(inode
, data
))) {
1185 * Return the locked inode with I_NEW set, the
1186 * caller is responsible for filling in the contents
1188 spin_lock(&inode
->i_lock
);
1189 inode
->i_state
|= I_NEW
;
1190 hlist_add_head_rcu(&inode
->i_hash
, head
);
1191 spin_unlock(&inode
->i_lock
);
1194 * Add inode to the sb list if it's not already. It has I_NEW at this
1195 * point, so it should be safe to test i_sb_list locklessly.
1197 if (list_empty(&inode
->i_sb_list
))
1198 inode_sb_list_add(inode
);
1200 spin_unlock(&inode_hash_lock
);
1204 EXPORT_SYMBOL(inode_insert5
);
1207 * iget5_locked - obtain an inode from a mounted file system
1208 * @sb: super block of file system
1209 * @hashval: hash value (usually inode number) to get
1210 * @test: callback used for comparisons between inodes
1211 * @set: callback used to initialize a new struct inode
1212 * @data: opaque data pointer to pass to @test and @set
1214 * Search for the inode specified by @hashval and @data in the inode cache,
1215 * and if present it is return it with an increased reference count. This is
1216 * a generalized version of iget_locked() for file systems where the inode
1217 * number is not sufficient for unique identification of an inode.
1219 * If the inode is not in cache, allocate a new inode and return it locked,
1220 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1221 * before unlocking it via unlock_new_inode().
1223 * Note both @test and @set are called with the inode_hash_lock held, so can't
1226 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1227 int (*test
)(struct inode
*, void *),
1228 int (*set
)(struct inode
*, void *), void *data
)
1230 struct inode
*inode
= ilookup5(sb
, hashval
, test
, data
);
1233 struct inode
*new = alloc_inode(sb
);
1237 inode
= inode_insert5(new, hashval
, test
, set
, data
);
1238 if (unlikely(inode
!= new))
1244 EXPORT_SYMBOL(iget5_locked
);
1247 * iget_locked - obtain an inode from a mounted file system
1248 * @sb: super block of file system
1249 * @ino: inode number to get
1251 * Search for the inode specified by @ino in the inode cache and if present
1252 * return it with an increased reference count. This is for file systems
1253 * where the inode number is sufficient for unique identification of an inode.
1255 * If the inode is not in cache, allocate a new inode and return it locked,
1256 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1257 * before unlocking it via unlock_new_inode().
1259 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1261 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1262 struct inode
*inode
;
1264 spin_lock(&inode_hash_lock
);
1265 inode
= find_inode_fast(sb
, head
, ino
);
1266 spin_unlock(&inode_hash_lock
);
1270 wait_on_inode(inode
);
1271 if (unlikely(inode_unhashed(inode
))) {
1278 inode
= alloc_inode(sb
);
1282 spin_lock(&inode_hash_lock
);
1283 /* We released the lock, so.. */
1284 old
= find_inode_fast(sb
, head
, ino
);
1287 spin_lock(&inode
->i_lock
);
1288 inode
->i_state
= I_NEW
;
1289 hlist_add_head_rcu(&inode
->i_hash
, head
);
1290 spin_unlock(&inode
->i_lock
);
1291 inode_sb_list_add(inode
);
1292 spin_unlock(&inode_hash_lock
);
1294 /* Return the locked inode with I_NEW set, the
1295 * caller is responsible for filling in the contents
1301 * Uhhuh, somebody else created the same inode under
1302 * us. Use the old inode instead of the one we just
1305 spin_unlock(&inode_hash_lock
);
1306 destroy_inode(inode
);
1310 wait_on_inode(inode
);
1311 if (unlikely(inode_unhashed(inode
))) {
1318 EXPORT_SYMBOL(iget_locked
);
1321 * search the inode cache for a matching inode number.
1322 * If we find one, then the inode number we are trying to
1323 * allocate is not unique and so we should not use it.
1325 * Returns 1 if the inode number is unique, 0 if it is not.
1327 static int test_inode_iunique(struct super_block
*sb
, unsigned long ino
)
1329 struct hlist_head
*b
= inode_hashtable
+ hash(sb
, ino
);
1330 struct inode
*inode
;
1332 hlist_for_each_entry_rcu(inode
, b
, i_hash
) {
1333 if (inode
->i_ino
== ino
&& inode
->i_sb
== sb
)
1340 * iunique - get a unique inode number
1342 * @max_reserved: highest reserved inode number
1344 * Obtain an inode number that is unique on the system for a given
1345 * superblock. This is used by file systems that have no natural
1346 * permanent inode numbering system. An inode number is returned that
1347 * is higher than the reserved limit but unique.
1350 * With a large number of inodes live on the file system this function
1351 * currently becomes quite slow.
1353 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
1356 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1357 * error if st_ino won't fit in target struct field. Use 32bit counter
1358 * here to attempt to avoid that.
1360 static DEFINE_SPINLOCK(iunique_lock
);
1361 static unsigned int counter
;
1365 spin_lock(&iunique_lock
);
1367 if (counter
<= max_reserved
)
1368 counter
= max_reserved
+ 1;
1370 } while (!test_inode_iunique(sb
, res
));
1371 spin_unlock(&iunique_lock
);
1376 EXPORT_SYMBOL(iunique
);
1378 struct inode
*igrab(struct inode
*inode
)
1380 spin_lock(&inode
->i_lock
);
1381 if (!(inode
->i_state
& (I_FREEING
|I_WILL_FREE
))) {
1383 spin_unlock(&inode
->i_lock
);
1385 spin_unlock(&inode
->i_lock
);
1387 * Handle the case where s_op->clear_inode is not been
1388 * called yet, and somebody is calling igrab
1389 * while the inode is getting freed.
1395 EXPORT_SYMBOL(igrab
);
1398 * ilookup5_nowait - search for an inode in the inode cache
1399 * @sb: super block of file system to search
1400 * @hashval: hash value (usually inode number) to search for
1401 * @test: callback used for comparisons between inodes
1402 * @data: opaque data pointer to pass to @test
1404 * Search for the inode specified by @hashval and @data in the inode cache.
1405 * If the inode is in the cache, the inode is returned with an incremented
1408 * Note: I_NEW is not waited upon so you have to be very careful what you do
1409 * with the returned inode. You probably should be using ilookup5() instead.
1411 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1413 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
1414 int (*test
)(struct inode
*, void *), void *data
)
1416 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1417 struct inode
*inode
;
1419 spin_lock(&inode_hash_lock
);
1420 inode
= find_inode(sb
, head
, test
, data
);
1421 spin_unlock(&inode_hash_lock
);
1423 return IS_ERR(inode
) ? NULL
: inode
;
1425 EXPORT_SYMBOL(ilookup5_nowait
);
1428 * ilookup5 - search for an inode in the inode cache
1429 * @sb: super block of file system to search
1430 * @hashval: hash value (usually inode number) to search for
1431 * @test: callback used for comparisons between inodes
1432 * @data: opaque data pointer to pass to @test
1434 * Search for the inode specified by @hashval and @data in the inode cache,
1435 * and if the inode is in the cache, return the inode with an incremented
1436 * reference count. Waits on I_NEW before returning the inode.
1437 * returned with an incremented reference count.
1439 * This is a generalized version of ilookup() for file systems where the
1440 * inode number is not sufficient for unique identification of an inode.
1442 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1444 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
1445 int (*test
)(struct inode
*, void *), void *data
)
1447 struct inode
*inode
;
1449 inode
= ilookup5_nowait(sb
, hashval
, test
, data
);
1451 wait_on_inode(inode
);
1452 if (unlikely(inode_unhashed(inode
))) {
1459 EXPORT_SYMBOL(ilookup5
);
1462 * ilookup - search for an inode in the inode cache
1463 * @sb: super block of file system to search
1464 * @ino: inode number to search for
1466 * Search for the inode @ino in the inode cache, and if the inode is in the
1467 * cache, the inode is returned with an incremented reference count.
1469 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1471 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1472 struct inode
*inode
;
1474 spin_lock(&inode_hash_lock
);
1475 inode
= find_inode_fast(sb
, head
, ino
);
1476 spin_unlock(&inode_hash_lock
);
1481 wait_on_inode(inode
);
1482 if (unlikely(inode_unhashed(inode
))) {
1489 EXPORT_SYMBOL(ilookup
);
1492 * find_inode_nowait - find an inode in the inode cache
1493 * @sb: super block of file system to search
1494 * @hashval: hash value (usually inode number) to search for
1495 * @match: callback used for comparisons between inodes
1496 * @data: opaque data pointer to pass to @match
1498 * Search for the inode specified by @hashval and @data in the inode
1499 * cache, where the helper function @match will return 0 if the inode
1500 * does not match, 1 if the inode does match, and -1 if the search
1501 * should be stopped. The @match function must be responsible for
1502 * taking the i_lock spin_lock and checking i_state for an inode being
1503 * freed or being initialized, and incrementing the reference count
1504 * before returning 1. It also must not sleep, since it is called with
1505 * the inode_hash_lock spinlock held.
1507 * This is a even more generalized version of ilookup5() when the
1508 * function must never block --- find_inode() can block in
1509 * __wait_on_freeing_inode() --- or when the caller can not increment
1510 * the reference count because the resulting iput() might cause an
1511 * inode eviction. The tradeoff is that the @match funtion must be
1512 * very carefully implemented.
1514 struct inode
*find_inode_nowait(struct super_block
*sb
,
1515 unsigned long hashval
,
1516 int (*match
)(struct inode
*, unsigned long,
1520 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1521 struct inode
*inode
, *ret_inode
= NULL
;
1524 spin_lock(&inode_hash_lock
);
1525 hlist_for_each_entry(inode
, head
, i_hash
) {
1526 if (inode
->i_sb
!= sb
)
1528 mval
= match(inode
, hashval
, data
);
1536 spin_unlock(&inode_hash_lock
);
1539 EXPORT_SYMBOL(find_inode_nowait
);
1542 * find_inode_rcu - find an inode in the inode cache
1543 * @sb: Super block of file system to search
1544 * @hashval: Key to hash
1545 * @test: Function to test match on an inode
1546 * @data: Data for test function
1548 * Search for the inode specified by @hashval and @data in the inode cache,
1549 * where the helper function @test will return 0 if the inode does not match
1550 * and 1 if it does. The @test function must be responsible for taking the
1551 * i_lock spin_lock and checking i_state for an inode being freed or being
1554 * If successful, this will return the inode for which the @test function
1555 * returned 1 and NULL otherwise.
1557 * The @test function is not permitted to take a ref on any inode presented.
1558 * It is also not permitted to sleep.
1560 * The caller must hold the RCU read lock.
1562 struct inode
*find_inode_rcu(struct super_block
*sb
, unsigned long hashval
,
1563 int (*test
)(struct inode
*, void *), void *data
)
1565 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1566 struct inode
*inode
;
1568 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1569 "suspicious find_inode_rcu() usage");
1571 hlist_for_each_entry_rcu(inode
, head
, i_hash
) {
1572 if (inode
->i_sb
== sb
&&
1573 !(READ_ONCE(inode
->i_state
) & (I_FREEING
| I_WILL_FREE
)) &&
1579 EXPORT_SYMBOL(find_inode_rcu
);
1582 * find_inode_by_ino_rcu - Find an inode in the inode cache
1583 * @sb: Super block of file system to search
1584 * @ino: The inode number to match
1586 * Search for the inode specified by @hashval and @data in the inode cache,
1587 * where the helper function @test will return 0 if the inode does not match
1588 * and 1 if it does. The @test function must be responsible for taking the
1589 * i_lock spin_lock and checking i_state for an inode being freed or being
1592 * If successful, this will return the inode for which the @test function
1593 * returned 1 and NULL otherwise.
1595 * The @test function is not permitted to take a ref on any inode presented.
1596 * It is also not permitted to sleep.
1598 * The caller must hold the RCU read lock.
1600 struct inode
*find_inode_by_ino_rcu(struct super_block
*sb
,
1603 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1604 struct inode
*inode
;
1606 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1607 "suspicious find_inode_by_ino_rcu() usage");
1609 hlist_for_each_entry_rcu(inode
, head
, i_hash
) {
1610 if (inode
->i_ino
== ino
&&
1611 inode
->i_sb
== sb
&&
1612 !(READ_ONCE(inode
->i_state
) & (I_FREEING
| I_WILL_FREE
)))
1617 EXPORT_SYMBOL(find_inode_by_ino_rcu
);
1619 int insert_inode_locked(struct inode
*inode
)
1621 struct super_block
*sb
= inode
->i_sb
;
1622 ino_t ino
= inode
->i_ino
;
1623 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1626 struct inode
*old
= NULL
;
1627 spin_lock(&inode_hash_lock
);
1628 hlist_for_each_entry(old
, head
, i_hash
) {
1629 if (old
->i_ino
!= ino
)
1631 if (old
->i_sb
!= sb
)
1633 spin_lock(&old
->i_lock
);
1634 if (old
->i_state
& (I_FREEING
|I_WILL_FREE
)) {
1635 spin_unlock(&old
->i_lock
);
1641 spin_lock(&inode
->i_lock
);
1642 inode
->i_state
|= I_NEW
| I_CREATING
;
1643 hlist_add_head_rcu(&inode
->i_hash
, head
);
1644 spin_unlock(&inode
->i_lock
);
1645 spin_unlock(&inode_hash_lock
);
1648 if (unlikely(old
->i_state
& I_CREATING
)) {
1649 spin_unlock(&old
->i_lock
);
1650 spin_unlock(&inode_hash_lock
);
1654 spin_unlock(&old
->i_lock
);
1655 spin_unlock(&inode_hash_lock
);
1657 if (unlikely(!inode_unhashed(old
))) {
1664 EXPORT_SYMBOL(insert_inode_locked
);
1666 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1667 int (*test
)(struct inode
*, void *), void *data
)
1671 inode
->i_state
|= I_CREATING
;
1672 old
= inode_insert5(inode
, hashval
, test
, NULL
, data
);
1680 EXPORT_SYMBOL(insert_inode_locked4
);
1683 int generic_delete_inode(struct inode
*inode
)
1687 EXPORT_SYMBOL(generic_delete_inode
);
1690 * Called when we're dropping the last reference
1693 * Call the FS "drop_inode()" function, defaulting to
1694 * the legacy UNIX filesystem behaviour. If it tells
1695 * us to evict inode, do so. Otherwise, retain inode
1696 * in cache if fs is alive, sync and evict if fs is
1699 static void iput_final(struct inode
*inode
)
1701 struct super_block
*sb
= inode
->i_sb
;
1702 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1703 unsigned long state
;
1706 WARN_ON(inode
->i_state
& I_NEW
);
1709 drop
= op
->drop_inode(inode
);
1711 drop
= generic_drop_inode(inode
);
1714 !(inode
->i_state
& I_DONTCACHE
) &&
1715 (sb
->s_flags
& SB_ACTIVE
)) {
1716 __inode_add_lru(inode
, true);
1717 spin_unlock(&inode
->i_lock
);
1721 state
= inode
->i_state
;
1723 WRITE_ONCE(inode
->i_state
, state
| I_WILL_FREE
);
1724 spin_unlock(&inode
->i_lock
);
1726 write_inode_now(inode
, 1);
1728 spin_lock(&inode
->i_lock
);
1729 state
= inode
->i_state
;
1730 WARN_ON(state
& I_NEW
);
1731 state
&= ~I_WILL_FREE
;
1734 WRITE_ONCE(inode
->i_state
, state
| I_FREEING
);
1735 if (!list_empty(&inode
->i_lru
))
1736 inode_lru_list_del(inode
);
1737 spin_unlock(&inode
->i_lock
);
1743 * iput - put an inode
1744 * @inode: inode to put
1746 * Puts an inode, dropping its usage count. If the inode use count hits
1747 * zero, the inode is then freed and may also be destroyed.
1749 * Consequently, iput() can sleep.
1751 void iput(struct inode
*inode
)
1755 BUG_ON(inode
->i_state
& I_CLEAR
);
1757 if (atomic_dec_and_lock(&inode
->i_count
, &inode
->i_lock
)) {
1758 if (inode
->i_nlink
&& (inode
->i_state
& I_DIRTY_TIME
)) {
1759 atomic_inc(&inode
->i_count
);
1760 spin_unlock(&inode
->i_lock
);
1761 trace_writeback_lazytime_iput(inode
);
1762 mark_inode_dirty_sync(inode
);
1768 EXPORT_SYMBOL(iput
);
1772 * bmap - find a block number in a file
1773 * @inode: inode owning the block number being requested
1774 * @block: pointer containing the block to find
1776 * Replaces the value in ``*block`` with the block number on the device holding
1777 * corresponding to the requested block number in the file.
1778 * That is, asked for block 4 of inode 1 the function will replace the
1779 * 4 in ``*block``, with disk block relative to the disk start that holds that
1780 * block of the file.
1782 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1783 * hole, returns 0 and ``*block`` is also set to 0.
1785 int bmap(struct inode
*inode
, sector_t
*block
)
1787 if (!inode
->i_mapping
->a_ops
->bmap
)
1790 *block
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, *block
);
1793 EXPORT_SYMBOL(bmap
);
1797 * With relative atime, only update atime if the previous atime is
1798 * earlier than or equal to either the ctime or mtime,
1799 * or if at least a day has passed since the last atime update.
1801 static bool relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1802 struct timespec64 now
)
1804 struct timespec64 atime
, mtime
, ctime
;
1806 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1809 * Is mtime younger than or equal to atime? If yes, update atime:
1811 atime
= inode_get_atime(inode
);
1812 mtime
= inode_get_mtime(inode
);
1813 if (timespec64_compare(&mtime
, &atime
) >= 0)
1816 * Is ctime younger than or equal to atime? If yes, update atime:
1818 ctime
= inode_get_ctime(inode
);
1819 if (timespec64_compare(&ctime
, &atime
) >= 0)
1823 * Is the previous atime value older than a day? If yes,
1826 if ((long)(now
.tv_sec
- atime
.tv_sec
) >= 24*60*60)
1829 * Good, we can skip the atime update:
1835 * inode_update_timestamps - update the timestamps on the inode
1836 * @inode: inode to be updated
1837 * @flags: S_* flags that needed to be updated
1839 * The update_time function is called when an inode's timestamps need to be
1840 * updated for a read or write operation. This function handles updating the
1841 * actual timestamps. It's up to the caller to ensure that the inode is marked
1842 * dirty appropriately.
1844 * In the case where any of S_MTIME, S_CTIME, or S_VERSION need to be updated,
1845 * attempt to update all three of them. S_ATIME updates can be handled
1846 * independently of the rest.
1848 * Returns a set of S_* flags indicating which values changed.
1850 int inode_update_timestamps(struct inode
*inode
, int flags
)
1853 struct timespec64 now
;
1855 if (flags
& (S_MTIME
|S_CTIME
|S_VERSION
)) {
1856 struct timespec64 ctime
= inode_get_ctime(inode
);
1857 struct timespec64 mtime
= inode_get_mtime(inode
);
1859 now
= inode_set_ctime_current(inode
);
1860 if (!timespec64_equal(&now
, &ctime
))
1862 if (!timespec64_equal(&now
, &mtime
)) {
1863 inode_set_mtime_to_ts(inode
, now
);
1866 if (IS_I_VERSION(inode
) && inode_maybe_inc_iversion(inode
, updated
))
1867 updated
|= S_VERSION
;
1869 now
= current_time(inode
);
1872 if (flags
& S_ATIME
) {
1873 struct timespec64 atime
= inode_get_atime(inode
);
1875 if (!timespec64_equal(&now
, &atime
)) {
1876 inode_set_atime_to_ts(inode
, now
);
1882 EXPORT_SYMBOL(inode_update_timestamps
);
1885 * generic_update_time - update the timestamps on the inode
1886 * @inode: inode to be updated
1887 * @flags: S_* flags that needed to be updated
1889 * The update_time function is called when an inode's timestamps need to be
1890 * updated for a read or write operation. In the case where any of S_MTIME, S_CTIME,
1891 * or S_VERSION need to be updated we attempt to update all three of them. S_ATIME
1892 * updates can be handled done independently of the rest.
1894 * Returns a S_* mask indicating which fields were updated.
1896 int generic_update_time(struct inode
*inode
, int flags
)
1898 int updated
= inode_update_timestamps(inode
, flags
);
1899 int dirty_flags
= 0;
1901 if (updated
& (S_ATIME
|S_MTIME
|S_CTIME
))
1902 dirty_flags
= inode
->i_sb
->s_flags
& SB_LAZYTIME
? I_DIRTY_TIME
: I_DIRTY_SYNC
;
1903 if (updated
& S_VERSION
)
1904 dirty_flags
|= I_DIRTY_SYNC
;
1905 __mark_inode_dirty(inode
, dirty_flags
);
1908 EXPORT_SYMBOL(generic_update_time
);
1911 * This does the actual work of updating an inodes time or version. Must have
1912 * had called mnt_want_write() before calling this.
1914 int inode_update_time(struct inode
*inode
, int flags
)
1916 if (inode
->i_op
->update_time
)
1917 return inode
->i_op
->update_time(inode
, flags
);
1918 generic_update_time(inode
, flags
);
1921 EXPORT_SYMBOL(inode_update_time
);
1924 * atime_needs_update - update the access time
1925 * @path: the &struct path to update
1926 * @inode: inode to update
1928 * Update the accessed time on an inode and mark it for writeback.
1929 * This function automatically handles read only file systems and media,
1930 * as well as the "noatime" flag and inode specific "noatime" markers.
1932 bool atime_needs_update(const struct path
*path
, struct inode
*inode
)
1934 struct vfsmount
*mnt
= path
->mnt
;
1935 struct timespec64 now
, atime
;
1937 if (inode
->i_flags
& S_NOATIME
)
1940 /* Atime updates will likely cause i_uid and i_gid to be written
1941 * back improprely if their true value is unknown to the vfs.
1943 if (HAS_UNMAPPED_ID(mnt_idmap(mnt
), inode
))
1946 if (IS_NOATIME(inode
))
1948 if ((inode
->i_sb
->s_flags
& SB_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1951 if (mnt
->mnt_flags
& MNT_NOATIME
)
1953 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1956 now
= current_time(inode
);
1958 if (!relatime_need_update(mnt
, inode
, now
))
1961 atime
= inode_get_atime(inode
);
1962 if (timespec64_equal(&atime
, &now
))
1968 void touch_atime(const struct path
*path
)
1970 struct vfsmount
*mnt
= path
->mnt
;
1971 struct inode
*inode
= d_inode(path
->dentry
);
1973 if (!atime_needs_update(path
, inode
))
1976 if (!sb_start_write_trylock(inode
->i_sb
))
1979 if (mnt_get_write_access(mnt
) != 0)
1982 * File systems can error out when updating inodes if they need to
1983 * allocate new space to modify an inode (such is the case for
1984 * Btrfs), but since we touch atime while walking down the path we
1985 * really don't care if we failed to update the atime of the file,
1986 * so just ignore the return value.
1987 * We may also fail on filesystems that have the ability to make parts
1988 * of the fs read only, e.g. subvolumes in Btrfs.
1990 inode_update_time(inode
, S_ATIME
);
1991 mnt_put_write_access(mnt
);
1993 sb_end_write(inode
->i_sb
);
1995 EXPORT_SYMBOL(touch_atime
);
1998 * Return mask of changes for notify_change() that need to be done as a
1999 * response to write or truncate. Return 0 if nothing has to be changed.
2000 * Negative value on error (change should be denied).
2002 int dentry_needs_remove_privs(struct mnt_idmap
*idmap
,
2003 struct dentry
*dentry
)
2005 struct inode
*inode
= d_inode(dentry
);
2009 if (IS_NOSEC(inode
))
2012 mask
= setattr_should_drop_suidgid(idmap
, inode
);
2013 ret
= security_inode_need_killpriv(dentry
);
2017 mask
|= ATTR_KILL_PRIV
;
2021 static int __remove_privs(struct mnt_idmap
*idmap
,
2022 struct dentry
*dentry
, int kill
)
2024 struct iattr newattrs
;
2026 newattrs
.ia_valid
= ATTR_FORCE
| kill
;
2028 * Note we call this on write, so notify_change will not
2029 * encounter any conflicting delegations:
2031 return notify_change(idmap
, dentry
, &newattrs
, NULL
);
2034 static int __file_remove_privs(struct file
*file
, unsigned int flags
)
2036 struct dentry
*dentry
= file_dentry(file
);
2037 struct inode
*inode
= file_inode(file
);
2041 if (IS_NOSEC(inode
) || !S_ISREG(inode
->i_mode
))
2044 kill
= dentry_needs_remove_privs(file_mnt_idmap(file
), dentry
);
2049 if (flags
& IOCB_NOWAIT
)
2052 error
= __remove_privs(file_mnt_idmap(file
), dentry
, kill
);
2056 inode_has_no_xattr(inode
);
2061 * file_remove_privs - remove special file privileges (suid, capabilities)
2062 * @file: file to remove privileges from
2064 * When file is modified by a write or truncation ensure that special
2065 * file privileges are removed.
2067 * Return: 0 on success, negative errno on failure.
2069 int file_remove_privs(struct file
*file
)
2071 return __file_remove_privs(file
, 0);
2073 EXPORT_SYMBOL(file_remove_privs
);
2075 static int inode_needs_update_time(struct inode
*inode
)
2078 struct timespec64 now
= current_time(inode
);
2079 struct timespec64 ts
;
2081 /* First try to exhaust all avenues to not sync */
2082 if (IS_NOCMTIME(inode
))
2085 ts
= inode_get_mtime(inode
);
2086 if (!timespec64_equal(&ts
, &now
))
2089 ts
= inode_get_ctime(inode
);
2090 if (!timespec64_equal(&ts
, &now
))
2093 if (IS_I_VERSION(inode
) && inode_iversion_need_inc(inode
))
2094 sync_it
|= S_VERSION
;
2099 static int __file_update_time(struct file
*file
, int sync_mode
)
2102 struct inode
*inode
= file_inode(file
);
2104 /* try to update time settings */
2105 if (!mnt_get_write_access_file(file
)) {
2106 ret
= inode_update_time(inode
, sync_mode
);
2107 mnt_put_write_access_file(file
);
2114 * file_update_time - update mtime and ctime time
2115 * @file: file accessed
2117 * Update the mtime and ctime members of an inode and mark the inode for
2118 * writeback. Note that this function is meant exclusively for usage in
2119 * the file write path of filesystems, and filesystems may choose to
2120 * explicitly ignore updates via this function with the _NOCMTIME inode
2121 * flag, e.g. for network filesystem where these imestamps are handled
2122 * by the server. This can return an error for file systems who need to
2123 * allocate space in order to update an inode.
2125 * Return: 0 on success, negative errno on failure.
2127 int file_update_time(struct file
*file
)
2130 struct inode
*inode
= file_inode(file
);
2132 ret
= inode_needs_update_time(inode
);
2136 return __file_update_time(file
, ret
);
2138 EXPORT_SYMBOL(file_update_time
);
2141 * file_modified_flags - handle mandated vfs changes when modifying a file
2142 * @file: file that was modified
2143 * @flags: kiocb flags
2145 * When file has been modified ensure that special
2146 * file privileges are removed and time settings are updated.
2148 * If IOCB_NOWAIT is set, special file privileges will not be removed and
2149 * time settings will not be updated. It will return -EAGAIN.
2151 * Context: Caller must hold the file's inode lock.
2153 * Return: 0 on success, negative errno on failure.
2155 static int file_modified_flags(struct file
*file
, int flags
)
2158 struct inode
*inode
= file_inode(file
);
2161 * Clear the security bits if the process is not being run by root.
2162 * This keeps people from modifying setuid and setgid binaries.
2164 ret
= __file_remove_privs(file
, flags
);
2168 if (unlikely(file
->f_mode
& FMODE_NOCMTIME
))
2171 ret
= inode_needs_update_time(inode
);
2174 if (flags
& IOCB_NOWAIT
)
2177 return __file_update_time(file
, ret
);
2181 * file_modified - handle mandated vfs changes when modifying a file
2182 * @file: file that was modified
2184 * When file has been modified ensure that special
2185 * file privileges are removed and time settings are updated.
2187 * Context: Caller must hold the file's inode lock.
2189 * Return: 0 on success, negative errno on failure.
2191 int file_modified(struct file
*file
)
2193 return file_modified_flags(file
, 0);
2195 EXPORT_SYMBOL(file_modified
);
2198 * kiocb_modified - handle mandated vfs changes when modifying a file
2199 * @iocb: iocb that was modified
2201 * When file has been modified ensure that special
2202 * file privileges are removed and time settings are updated.
2204 * Context: Caller must hold the file's inode lock.
2206 * Return: 0 on success, negative errno on failure.
2208 int kiocb_modified(struct kiocb
*iocb
)
2210 return file_modified_flags(iocb
->ki_filp
, iocb
->ki_flags
);
2212 EXPORT_SYMBOL_GPL(kiocb_modified
);
2214 int inode_needs_sync(struct inode
*inode
)
2218 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
2222 EXPORT_SYMBOL(inode_needs_sync
);
2225 * If we try to find an inode in the inode hash while it is being
2226 * deleted, we have to wait until the filesystem completes its
2227 * deletion before reporting that it isn't found. This function waits
2228 * until the deletion _might_ have completed. Callers are responsible
2229 * to recheck inode state.
2231 * It doesn't matter if I_NEW is not set initially, a call to
2232 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2235 static void __wait_on_freeing_inode(struct inode
*inode
)
2237 wait_queue_head_t
*wq
;
2238 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
2239 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
2240 prepare_to_wait(wq
, &wait
.wq_entry
, TASK_UNINTERRUPTIBLE
);
2241 spin_unlock(&inode
->i_lock
);
2242 spin_unlock(&inode_hash_lock
);
2244 finish_wait(wq
, &wait
.wq_entry
);
2245 spin_lock(&inode_hash_lock
);
2248 static __initdata
unsigned long ihash_entries
;
2249 static int __init
set_ihash_entries(char *str
)
2253 ihash_entries
= simple_strtoul(str
, &str
, 0);
2256 __setup("ihash_entries=", set_ihash_entries
);
2259 * Initialize the waitqueues and inode hash table.
2261 void __init
inode_init_early(void)
2263 /* If hashes are distributed across NUMA nodes, defer
2264 * hash allocation until vmalloc space is available.
2270 alloc_large_system_hash("Inode-cache",
2271 sizeof(struct hlist_head
),
2274 HASH_EARLY
| HASH_ZERO
,
2281 void __init
inode_init(void)
2283 /* inode slab cache */
2284 inode_cachep
= kmem_cache_create("inode_cache",
2285 sizeof(struct inode
),
2287 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
2288 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
),
2291 /* Hash may have been set up in inode_init_early */
2296 alloc_large_system_hash("Inode-cache",
2297 sizeof(struct hlist_head
),
2307 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
2309 inode
->i_mode
= mode
;
2310 if (S_ISCHR(mode
)) {
2311 inode
->i_fop
= &def_chr_fops
;
2312 inode
->i_rdev
= rdev
;
2313 } else if (S_ISBLK(mode
)) {
2314 if (IS_ENABLED(CONFIG_BLOCK
))
2315 inode
->i_fop
= &def_blk_fops
;
2316 inode
->i_rdev
= rdev
;
2317 } else if (S_ISFIFO(mode
))
2318 inode
->i_fop
= &pipefifo_fops
;
2319 else if (S_ISSOCK(mode
))
2320 ; /* leave it no_open_fops */
2322 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
2323 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
2326 EXPORT_SYMBOL(init_special_inode
);
2329 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2330 * @idmap: idmap of the mount the inode was created from
2332 * @dir: Directory inode
2333 * @mode: mode of the new inode
2335 * If the inode has been created through an idmapped mount the idmap of
2336 * the vfsmount must be passed through @idmap. This function will then take
2337 * care to map the inode according to @idmap before checking permissions
2338 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2339 * checking is to be performed on the raw inode simply pass @nop_mnt_idmap.
2341 void inode_init_owner(struct mnt_idmap
*idmap
, struct inode
*inode
,
2342 const struct inode
*dir
, umode_t mode
)
2344 inode_fsuid_set(inode
, idmap
);
2345 if (dir
&& dir
->i_mode
& S_ISGID
) {
2346 inode
->i_gid
= dir
->i_gid
;
2348 /* Directories are special, and always inherit S_ISGID */
2352 inode_fsgid_set(inode
, idmap
);
2353 inode
->i_mode
= mode
;
2355 EXPORT_SYMBOL(inode_init_owner
);
2358 * inode_owner_or_capable - check current task permissions to inode
2359 * @idmap: idmap of the mount the inode was found from
2360 * @inode: inode being checked
2362 * Return true if current either has CAP_FOWNER in a namespace with the
2363 * inode owner uid mapped, or owns the file.
2365 * If the inode has been found through an idmapped mount the idmap of
2366 * the vfsmount must be passed through @idmap. This function will then take
2367 * care to map the inode according to @idmap before checking permissions.
2368 * On non-idmapped mounts or if permission checking is to be performed on the
2369 * raw inode simply pass @nop_mnt_idmap.
2371 bool inode_owner_or_capable(struct mnt_idmap
*idmap
,
2372 const struct inode
*inode
)
2375 struct user_namespace
*ns
;
2377 vfsuid
= i_uid_into_vfsuid(idmap
, inode
);
2378 if (vfsuid_eq_kuid(vfsuid
, current_fsuid()))
2381 ns
= current_user_ns();
2382 if (vfsuid_has_mapping(ns
, vfsuid
) && ns_capable(ns
, CAP_FOWNER
))
2386 EXPORT_SYMBOL(inode_owner_or_capable
);
2389 * Direct i/o helper functions
2391 static void __inode_dio_wait(struct inode
*inode
)
2393 wait_queue_head_t
*wq
= bit_waitqueue(&inode
->i_state
, __I_DIO_WAKEUP
);
2394 DEFINE_WAIT_BIT(q
, &inode
->i_state
, __I_DIO_WAKEUP
);
2397 prepare_to_wait(wq
, &q
.wq_entry
, TASK_UNINTERRUPTIBLE
);
2398 if (atomic_read(&inode
->i_dio_count
))
2400 } while (atomic_read(&inode
->i_dio_count
));
2401 finish_wait(wq
, &q
.wq_entry
);
2405 * inode_dio_wait - wait for outstanding DIO requests to finish
2406 * @inode: inode to wait for
2408 * Waits for all pending direct I/O requests to finish so that we can
2409 * proceed with a truncate or equivalent operation.
2411 * Must be called under a lock that serializes taking new references
2412 * to i_dio_count, usually by inode->i_mutex.
2414 void inode_dio_wait(struct inode
*inode
)
2416 if (atomic_read(&inode
->i_dio_count
))
2417 __inode_dio_wait(inode
);
2419 EXPORT_SYMBOL(inode_dio_wait
);
2422 * inode_set_flags - atomically set some inode flags
2424 * Note: the caller should be holding i_mutex, or else be sure that
2425 * they have exclusive access to the inode structure (i.e., while the
2426 * inode is being instantiated). The reason for the cmpxchg() loop
2427 * --- which wouldn't be necessary if all code paths which modify
2428 * i_flags actually followed this rule, is that there is at least one
2429 * code path which doesn't today so we use cmpxchg() out of an abundance
2432 * In the long run, i_mutex is overkill, and we should probably look
2433 * at using the i_lock spinlock to protect i_flags, and then make sure
2434 * it is so documented in include/linux/fs.h and that all code follows
2435 * the locking convention!!
2437 void inode_set_flags(struct inode
*inode
, unsigned int flags
,
2440 WARN_ON_ONCE(flags
& ~mask
);
2441 set_mask_bits(&inode
->i_flags
, mask
, flags
);
2443 EXPORT_SYMBOL(inode_set_flags
);
2445 void inode_nohighmem(struct inode
*inode
)
2447 mapping_set_gfp_mask(inode
->i_mapping
, GFP_USER
);
2449 EXPORT_SYMBOL(inode_nohighmem
);
2452 * timestamp_truncate - Truncate timespec to a granularity
2454 * @inode: inode being updated
2456 * Truncate a timespec to the granularity supported by the fs
2457 * containing the inode. Always rounds down. gran must
2458 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2460 struct timespec64
timestamp_truncate(struct timespec64 t
, struct inode
*inode
)
2462 struct super_block
*sb
= inode
->i_sb
;
2463 unsigned int gran
= sb
->s_time_gran
;
2465 t
.tv_sec
= clamp(t
.tv_sec
, sb
->s_time_min
, sb
->s_time_max
);
2466 if (unlikely(t
.tv_sec
== sb
->s_time_max
|| t
.tv_sec
== sb
->s_time_min
))
2469 /* Avoid division in the common cases 1 ns and 1 s. */
2472 else if (gran
== NSEC_PER_SEC
)
2474 else if (gran
> 1 && gran
< NSEC_PER_SEC
)
2475 t
.tv_nsec
-= t
.tv_nsec
% gran
;
2477 WARN(1, "invalid file time granularity: %u", gran
);
2480 EXPORT_SYMBOL(timestamp_truncate
);
2483 * current_time - Return FS time
2486 * Return the current time truncated to the time granularity supported by
2489 * Note that inode and inode->sb cannot be NULL.
2490 * Otherwise, the function warns and returns time without truncation.
2492 struct timespec64
current_time(struct inode
*inode
)
2494 struct timespec64 now
;
2496 ktime_get_coarse_real_ts64(&now
);
2497 return timestamp_truncate(now
, inode
);
2499 EXPORT_SYMBOL(current_time
);
2502 * inode_set_ctime_current - set the ctime to current_time
2505 * Set the inode->i_ctime to the current value for the inode. Returns
2506 * the current value that was assigned to i_ctime.
2508 struct timespec64
inode_set_ctime_current(struct inode
*inode
)
2510 struct timespec64 now
= current_time(inode
);
2512 inode_set_ctime(inode
, now
.tv_sec
, now
.tv_nsec
);
2515 EXPORT_SYMBOL(inode_set_ctime_current
);
2518 * in_group_or_capable - check whether caller is CAP_FSETID privileged
2519 * @idmap: idmap of the mount @inode was found from
2520 * @inode: inode to check
2521 * @vfsgid: the new/current vfsgid of @inode
2523 * Check wether @vfsgid is in the caller's group list or if the caller is
2524 * privileged with CAP_FSETID over @inode. This can be used to determine
2525 * whether the setgid bit can be kept or must be dropped.
2527 * Return: true if the caller is sufficiently privileged, false if not.
2529 bool in_group_or_capable(struct mnt_idmap
*idmap
,
2530 const struct inode
*inode
, vfsgid_t vfsgid
)
2532 if (vfsgid_in_group_p(vfsgid
))
2534 if (capable_wrt_inode_uidgid(idmap
, inode
, CAP_FSETID
))
2540 * mode_strip_sgid - handle the sgid bit for non-directories
2541 * @idmap: idmap of the mount the inode was created from
2542 * @dir: parent directory inode
2543 * @mode: mode of the file to be created in @dir
2545 * If the @mode of the new file has both the S_ISGID and S_IXGRP bit
2546 * raised and @dir has the S_ISGID bit raised ensure that the caller is
2547 * either in the group of the parent directory or they have CAP_FSETID
2548 * in their user namespace and are privileged over the parent directory.
2549 * In all other cases, strip the S_ISGID bit from @mode.
2551 * Return: the new mode to use for the file
2553 umode_t
mode_strip_sgid(struct mnt_idmap
*idmap
,
2554 const struct inode
*dir
, umode_t mode
)
2556 if ((mode
& (S_ISGID
| S_IXGRP
)) != (S_ISGID
| S_IXGRP
))
2558 if (S_ISDIR(mode
) || !dir
|| !(dir
->i_mode
& S_ISGID
))
2560 if (in_group_or_capable(idmap
, dir
, i_gid_into_vfsgid(idmap
, dir
)))
2562 return mode
& ~S_ISGID
;
2564 EXPORT_SYMBOL(mode_strip_sgid
);