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 <linux/rw_hint.h>
24 #include <trace/events/writeback.h>
28 * Inode locking rules:
30 * inode->i_lock protects:
31 * inode->i_state, inode->i_hash, __iget(), inode->i_io_list
32 * Inode LRU list locks protect:
33 * inode->i_sb->s_inode_lru, inode->i_lru
34 * inode->i_sb->s_inode_list_lock protects:
35 * inode->i_sb->s_inodes, inode->i_sb_list
36 * bdi->wb.list_lock protects:
37 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
38 * inode_hash_lock protects:
39 * inode_hashtable, inode->i_hash
43 * inode->i_sb->s_inode_list_lock
45 * Inode LRU list locks
51 * inode->i_sb->s_inode_list_lock
58 static unsigned int i_hash_mask __ro_after_init;
59 static unsigned int i_hash_shift __ro_after_init;
60 static struct hlist_head *inode_hashtable __ro_after_init;
61 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
64 * Empty aops. Can be used for the cases where the user does not
65 * define any of the address_space operations.
67 const struct address_space_operations empty_aops = {
69 EXPORT_SYMBOL(empty_aops);
71 static DEFINE_PER_CPU(unsigned long, nr_inodes);
72 static DEFINE_PER_CPU(unsigned long, nr_unused);
74 static struct kmem_cache *inode_cachep __ro_after_init;
76 static long get_nr_inodes(void)
80 for_each_possible_cpu(i)
81 sum += per_cpu(nr_inodes, i);
82 return sum < 0 ? 0 : sum;
85 static inline long get_nr_inodes_unused(void)
89 for_each_possible_cpu(i)
90 sum += per_cpu(nr_unused, i);
91 return sum < 0 ? 0 : sum;
94 long get_nr_dirty_inodes(void)
96 /* not actually dirty inodes, but a wild approximation */
97 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
98 return nr_dirty > 0 ? nr_dirty : 0;
102 * Handle nr_inode sysctl
106 * Statistics gathering..
108 static struct inodes_stat_t inodes_stat;
110 static int proc_nr_inodes(const struct ctl_table *table, int write, void *buffer,
111 size_t *lenp, loff_t *ppos)
113 inodes_stat.nr_inodes = get_nr_inodes();
114 inodes_stat.nr_unused = get_nr_inodes_unused();
115 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
118 static struct ctl_table inodes_sysctls[] = {
120 .procname = "inode-nr",
121 .data = &inodes_stat,
122 .maxlen = 2*sizeof(long),
124 .proc_handler = proc_nr_inodes,
127 .procname = "inode-state",
128 .data = &inodes_stat,
129 .maxlen = 7*sizeof(long),
131 .proc_handler = proc_nr_inodes,
135 static int __init init_fs_inode_sysctls(void)
137 register_sysctl_init("fs", inodes_sysctls);
140 early_initcall(init_fs_inode_sysctls);
143 static int no_open(struct inode *inode, struct file *file)
149 * inode_init_always - perform inode structure initialisation
150 * @sb: superblock inode belongs to
151 * @inode: inode to initialise
153 * These are initializations that need to be done on every inode
154 * allocation as the fields are not initialised by slab allocation.
156 int inode_init_always(struct super_block *sb, struct inode *inode)
158 static const struct inode_operations empty_iops;
159 static const struct file_operations no_open_fops = {.open = no_open};
160 struct address_space *const mapping = &inode->i_data;
163 inode->i_blkbits = sb->s_blocksize_bits;
166 atomic64_set(&inode->i_sequence, 0);
167 atomic_set(&inode->i_count, 1);
168 inode->i_op = &empty_iops;
169 inode->i_fop = &no_open_fops;
171 inode->__i_nlink = 1;
172 inode->i_opflags = 0;
174 inode->i_opflags |= IOP_XATTR;
175 i_uid_write(inode, 0);
176 i_gid_write(inode, 0);
177 atomic_set(&inode->i_writecount, 0);
179 inode->i_write_hint = WRITE_LIFE_NOT_SET;
182 inode->i_generation = 0;
183 inode->i_pipe = NULL;
184 inode->i_cdev = NULL;
185 inode->i_link = NULL;
186 inode->i_dir_seq = 0;
188 inode->dirtied_when = 0;
190 #ifdef CONFIG_CGROUP_WRITEBACK
191 inode->i_wb_frn_winner = 0;
192 inode->i_wb_frn_avg_time = 0;
193 inode->i_wb_frn_history = 0;
196 spin_lock_init(&inode->i_lock);
197 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
199 init_rwsem(&inode->i_rwsem);
200 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
202 atomic_set(&inode->i_dio_count, 0);
204 mapping->a_ops = &empty_aops;
205 mapping->host = inode;
208 atomic_set(&mapping->i_mmap_writable, 0);
209 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
210 atomic_set(&mapping->nr_thps, 0);
212 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
213 mapping->i_private_data = NULL;
214 mapping->writeback_index = 0;
215 init_rwsem(&mapping->invalidate_lock);
216 lockdep_set_class_and_name(&mapping->invalidate_lock,
217 &sb->s_type->invalidate_lock_key,
218 "mapping.invalidate_lock");
219 if (sb->s_iflags & SB_I_STABLE_WRITES)
220 mapping_set_stable_writes(mapping);
221 inode->i_private = NULL;
222 inode->i_mapping = mapping;
223 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
224 #ifdef CONFIG_FS_POSIX_ACL
225 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
228 #ifdef CONFIG_FSNOTIFY
229 inode->i_fsnotify_mask = 0;
231 inode->i_flctx = NULL;
233 if (unlikely(security_inode_alloc(inode)))
236 this_cpu_inc(nr_inodes);
240 EXPORT_SYMBOL(inode_init_always);
242 void free_inode_nonrcu(struct inode *inode)
244 kmem_cache_free(inode_cachep, inode);
246 EXPORT_SYMBOL(free_inode_nonrcu);
248 static void i_callback(struct rcu_head *head)
250 struct inode *inode = container_of(head, struct inode, i_rcu);
251 if (inode->free_inode)
252 inode->free_inode(inode);
254 free_inode_nonrcu(inode);
257 static struct inode *alloc_inode(struct super_block *sb)
259 const struct super_operations *ops = sb->s_op;
262 if (ops->alloc_inode)
263 inode = ops->alloc_inode(sb);
265 inode = alloc_inode_sb(sb, inode_cachep, GFP_KERNEL);
270 if (unlikely(inode_init_always(sb, inode))) {
271 if (ops->destroy_inode) {
272 ops->destroy_inode(inode);
273 if (!ops->free_inode)
276 inode->free_inode = ops->free_inode;
277 i_callback(&inode->i_rcu);
284 void __destroy_inode(struct inode *inode)
286 BUG_ON(inode_has_buffers(inode));
287 inode_detach_wb(inode);
288 security_inode_free(inode);
289 fsnotify_inode_delete(inode);
290 locks_free_lock_context(inode);
291 if (!inode->i_nlink) {
292 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
293 atomic_long_dec(&inode->i_sb->s_remove_count);
296 #ifdef CONFIG_FS_POSIX_ACL
297 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
298 posix_acl_release(inode->i_acl);
299 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
300 posix_acl_release(inode->i_default_acl);
302 this_cpu_dec(nr_inodes);
304 EXPORT_SYMBOL(__destroy_inode);
306 static void destroy_inode(struct inode *inode)
308 const struct super_operations *ops = inode->i_sb->s_op;
310 BUG_ON(!list_empty(&inode->i_lru));
311 __destroy_inode(inode);
312 if (ops->destroy_inode) {
313 ops->destroy_inode(inode);
314 if (!ops->free_inode)
317 inode->free_inode = ops->free_inode;
318 call_rcu(&inode->i_rcu, i_callback);
322 * drop_nlink - directly drop an inode's link count
325 * This is a low-level filesystem helper to replace any
326 * direct filesystem manipulation of i_nlink. In cases
327 * where we are attempting to track writes to the
328 * filesystem, a decrement to zero means an imminent
329 * write when the file is truncated and actually unlinked
332 void drop_nlink(struct inode *inode)
334 WARN_ON(inode->i_nlink == 0);
337 atomic_long_inc(&inode->i_sb->s_remove_count);
339 EXPORT_SYMBOL(drop_nlink);
342 * clear_nlink - directly zero an inode's link count
345 * This is a low-level filesystem helper to replace any
346 * direct filesystem manipulation of i_nlink. See
347 * drop_nlink() for why we care about i_nlink hitting zero.
349 void clear_nlink(struct inode *inode)
351 if (inode->i_nlink) {
352 inode->__i_nlink = 0;
353 atomic_long_inc(&inode->i_sb->s_remove_count);
356 EXPORT_SYMBOL(clear_nlink);
359 * set_nlink - directly set an inode's link count
361 * @nlink: new nlink (should be non-zero)
363 * This is a low-level filesystem helper to replace any
364 * direct filesystem manipulation of i_nlink.
366 void set_nlink(struct inode *inode, unsigned int nlink)
371 /* Yes, some filesystems do change nlink from zero to one */
372 if (inode->i_nlink == 0)
373 atomic_long_dec(&inode->i_sb->s_remove_count);
375 inode->__i_nlink = nlink;
378 EXPORT_SYMBOL(set_nlink);
381 * inc_nlink - directly increment an inode's link count
384 * This is a low-level filesystem helper to replace any
385 * direct filesystem manipulation of i_nlink. Currently,
386 * it is only here for parity with dec_nlink().
388 void inc_nlink(struct inode *inode)
390 if (unlikely(inode->i_nlink == 0)) {
391 WARN_ON(!(inode->i_state & I_LINKABLE));
392 atomic_long_dec(&inode->i_sb->s_remove_count);
397 EXPORT_SYMBOL(inc_nlink);
399 static void __address_space_init_once(struct address_space *mapping)
401 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
402 init_rwsem(&mapping->i_mmap_rwsem);
403 INIT_LIST_HEAD(&mapping->i_private_list);
404 spin_lock_init(&mapping->i_private_lock);
405 mapping->i_mmap = RB_ROOT_CACHED;
408 void address_space_init_once(struct address_space *mapping)
410 memset(mapping, 0, sizeof(*mapping));
411 __address_space_init_once(mapping);
413 EXPORT_SYMBOL(address_space_init_once);
416 * These are initializations that only need to be done
417 * once, because the fields are idempotent across use
418 * of the inode, so let the slab aware of that.
420 void inode_init_once(struct inode *inode)
422 memset(inode, 0, sizeof(*inode));
423 INIT_HLIST_NODE(&inode->i_hash);
424 INIT_LIST_HEAD(&inode->i_devices);
425 INIT_LIST_HEAD(&inode->i_io_list);
426 INIT_LIST_HEAD(&inode->i_wb_list);
427 INIT_LIST_HEAD(&inode->i_lru);
428 INIT_LIST_HEAD(&inode->i_sb_list);
429 __address_space_init_once(&inode->i_data);
430 i_size_ordered_init(inode);
432 EXPORT_SYMBOL(inode_init_once);
434 static void init_once(void *foo)
436 struct inode *inode = (struct inode *) foo;
438 inode_init_once(inode);
442 * inode->i_lock must be held
444 void __iget(struct inode *inode)
446 atomic_inc(&inode->i_count);
450 * get additional reference to inode; caller must already hold one.
452 void ihold(struct inode *inode)
454 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
456 EXPORT_SYMBOL(ihold);
458 static void __inode_add_lru(struct inode *inode, bool rotate)
460 if (inode->i_state & (I_DIRTY_ALL | I_SYNC | I_FREEING | I_WILL_FREE))
462 if (atomic_read(&inode->i_count))
464 if (!(inode->i_sb->s_flags & SB_ACTIVE))
466 if (!mapping_shrinkable(&inode->i_data))
469 if (list_lru_add_obj(&inode->i_sb->s_inode_lru, &inode->i_lru))
470 this_cpu_inc(nr_unused);
472 inode->i_state |= I_REFERENCED;
475 struct wait_queue_head *inode_bit_waitqueue(struct wait_bit_queue_entry *wqe,
476 struct inode *inode, u32 bit)
480 bit_address = inode_state_wait_address(inode, bit);
481 init_wait_var_entry(wqe, bit_address, 0);
482 return __var_waitqueue(bit_address);
484 EXPORT_SYMBOL(inode_bit_waitqueue);
487 * Add inode to LRU if needed (inode is unused and clean).
489 * Needs inode->i_lock held.
491 void inode_add_lru(struct inode *inode)
493 __inode_add_lru(inode, false);
496 static void inode_lru_list_del(struct inode *inode)
498 if (list_lru_del_obj(&inode->i_sb->s_inode_lru, &inode->i_lru))
499 this_cpu_dec(nr_unused);
502 static void inode_pin_lru_isolating(struct inode *inode)
504 lockdep_assert_held(&inode->i_lock);
505 WARN_ON(inode->i_state & (I_LRU_ISOLATING | I_FREEING | I_WILL_FREE));
506 inode->i_state |= I_LRU_ISOLATING;
509 static void inode_unpin_lru_isolating(struct inode *inode)
511 spin_lock(&inode->i_lock);
512 WARN_ON(!(inode->i_state & I_LRU_ISOLATING));
513 inode->i_state &= ~I_LRU_ISOLATING;
514 /* Called with inode->i_lock which ensures memory ordering. */
515 inode_wake_up_bit(inode, __I_LRU_ISOLATING);
516 spin_unlock(&inode->i_lock);
519 static void inode_wait_for_lru_isolating(struct inode *inode)
521 struct wait_bit_queue_entry wqe;
522 struct wait_queue_head *wq_head;
524 lockdep_assert_held(&inode->i_lock);
525 if (!(inode->i_state & I_LRU_ISOLATING))
528 wq_head = inode_bit_waitqueue(&wqe, inode, __I_LRU_ISOLATING);
530 prepare_to_wait_event(wq_head, &wqe.wq_entry, TASK_UNINTERRUPTIBLE);
532 * Checking I_LRU_ISOLATING with inode->i_lock guarantees
535 if (!(inode->i_state & I_LRU_ISOLATING))
537 spin_unlock(&inode->i_lock);
539 spin_lock(&inode->i_lock);
541 finish_wait(wq_head, &wqe.wq_entry);
542 WARN_ON(inode->i_state & I_LRU_ISOLATING);
546 * inode_sb_list_add - add inode to the superblock list of inodes
547 * @inode: inode to add
549 void inode_sb_list_add(struct inode *inode)
551 spin_lock(&inode->i_sb->s_inode_list_lock);
552 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
553 spin_unlock(&inode->i_sb->s_inode_list_lock);
555 EXPORT_SYMBOL_GPL(inode_sb_list_add);
557 static inline void inode_sb_list_del(struct inode *inode)
559 if (!list_empty(&inode->i_sb_list)) {
560 spin_lock(&inode->i_sb->s_inode_list_lock);
561 list_del_init(&inode->i_sb_list);
562 spin_unlock(&inode->i_sb->s_inode_list_lock);
566 static unsigned long hash(struct super_block *sb, unsigned long hashval)
570 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
572 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
573 return tmp & i_hash_mask;
577 * __insert_inode_hash - hash an inode
578 * @inode: unhashed inode
579 * @hashval: unsigned long value used to locate this object in the
582 * Add an inode to the inode hash for this superblock.
584 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
586 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
588 spin_lock(&inode_hash_lock);
589 spin_lock(&inode->i_lock);
590 hlist_add_head_rcu(&inode->i_hash, b);
591 spin_unlock(&inode->i_lock);
592 spin_unlock(&inode_hash_lock);
594 EXPORT_SYMBOL(__insert_inode_hash);
597 * __remove_inode_hash - remove an inode from the hash
598 * @inode: inode to unhash
600 * Remove an inode from the superblock.
602 void __remove_inode_hash(struct inode *inode)
604 spin_lock(&inode_hash_lock);
605 spin_lock(&inode->i_lock);
606 hlist_del_init_rcu(&inode->i_hash);
607 spin_unlock(&inode->i_lock);
608 spin_unlock(&inode_hash_lock);
610 EXPORT_SYMBOL(__remove_inode_hash);
612 void dump_mapping(const struct address_space *mapping)
615 const struct address_space_operations *a_ops;
616 struct hlist_node *dentry_first;
617 struct dentry *dentry_ptr;
618 struct dentry dentry;
623 * If mapping is an invalid pointer, we don't want to crash
624 * accessing it, so probe everything depending on it carefully.
626 if (get_kernel_nofault(host, &mapping->host) ||
627 get_kernel_nofault(a_ops, &mapping->a_ops)) {
628 pr_warn("invalid mapping:%px\n", mapping);
633 pr_warn("aops:%ps\n", a_ops);
637 if (get_kernel_nofault(dentry_first, &host->i_dentry.first) ||
638 get_kernel_nofault(ino, &host->i_ino)) {
639 pr_warn("aops:%ps invalid inode:%px\n", a_ops, host);
644 pr_warn("aops:%ps ino:%lx\n", a_ops, ino);
648 dentry_ptr = container_of(dentry_first, struct dentry, d_u.d_alias);
649 if (get_kernel_nofault(dentry, dentry_ptr) ||
650 !dentry.d_parent || !dentry.d_name.name) {
651 pr_warn("aops:%ps ino:%lx invalid dentry:%px\n",
652 a_ops, ino, dentry_ptr);
656 if (strncpy_from_kernel_nofault(fname, dentry.d_name.name, 63) < 0)
657 strscpy(fname, "<invalid>");
659 * Even if strncpy_from_kernel_nofault() succeeded,
660 * the fname could be unreliable
662 pr_warn("aops:%ps ino:%lx dentry name(?):\"%s\"\n",
666 void clear_inode(struct inode *inode)
669 * We have to cycle the i_pages lock here because reclaim can be in the
670 * process of removing the last page (in __filemap_remove_folio())
671 * and we must not free the mapping under it.
673 xa_lock_irq(&inode->i_data.i_pages);
674 BUG_ON(inode->i_data.nrpages);
676 * Almost always, mapping_empty(&inode->i_data) here; but there are
677 * two known and long-standing ways in which nodes may get left behind
678 * (when deep radix-tree node allocation failed partway; or when THP
679 * collapse_file() failed). Until those two known cases are cleaned up,
680 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
681 * nor even WARN_ON(!mapping_empty).
683 xa_unlock_irq(&inode->i_data.i_pages);
684 BUG_ON(!list_empty(&inode->i_data.i_private_list));
685 BUG_ON(!(inode->i_state & I_FREEING));
686 BUG_ON(inode->i_state & I_CLEAR);
687 BUG_ON(!list_empty(&inode->i_wb_list));
688 /* don't need i_lock here, no concurrent mods to i_state */
689 inode->i_state = I_FREEING | I_CLEAR;
691 EXPORT_SYMBOL(clear_inode);
694 * Free the inode passed in, removing it from the lists it is still connected
695 * to. We remove any pages still attached to the inode and wait for any IO that
696 * is still in progress before finally destroying the inode.
698 * An inode must already be marked I_FREEING so that we avoid the inode being
699 * moved back onto lists if we race with other code that manipulates the lists
700 * (e.g. writeback_single_inode). The caller is responsible for setting this.
702 * An inode must already be removed from the LRU list before being evicted from
703 * the cache. This should occur atomically with setting the I_FREEING state
704 * flag, so no inodes here should ever be on the LRU when being evicted.
706 static void evict(struct inode *inode)
708 const struct super_operations *op = inode->i_sb->s_op;
710 BUG_ON(!(inode->i_state & I_FREEING));
711 BUG_ON(!list_empty(&inode->i_lru));
713 if (!list_empty(&inode->i_io_list))
714 inode_io_list_del(inode);
716 inode_sb_list_del(inode);
718 spin_lock(&inode->i_lock);
719 inode_wait_for_lru_isolating(inode);
722 * Wait for flusher thread to be done with the inode so that filesystem
723 * does not start destroying it while writeback is still running. Since
724 * the inode has I_FREEING set, flusher thread won't start new work on
725 * the inode. We just have to wait for running writeback to finish.
727 inode_wait_for_writeback(inode);
728 spin_unlock(&inode->i_lock);
730 if (op->evict_inode) {
731 op->evict_inode(inode);
733 truncate_inode_pages_final(&inode->i_data);
736 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
739 remove_inode_hash(inode);
742 * Wake up waiters in __wait_on_freeing_inode().
744 * Lockless hash lookup may end up finding the inode before we removed
745 * it above, but only lock it *after* we are done with the wakeup below.
746 * In this case the potential waiter cannot safely block.
748 * The inode being unhashed after the call to remove_inode_hash() is
749 * used as an indicator whether blocking on it is safe.
751 spin_lock(&inode->i_lock);
753 * Pairs with the barrier in prepare_to_wait_event() to make sure
754 * ___wait_var_event() either sees the bit cleared or
755 * waitqueue_active() check in wake_up_var() sees the waiter.
758 inode_wake_up_bit(inode, __I_NEW);
759 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
760 spin_unlock(&inode->i_lock);
762 destroy_inode(inode);
766 * dispose_list - dispose of the contents of a local list
767 * @head: the head of the list to free
769 * Dispose-list gets a local list with local inodes in it, so it doesn't
770 * need to worry about list corruption and SMP locks.
772 static void dispose_list(struct list_head *head)
774 while (!list_empty(head)) {
777 inode = list_first_entry(head, struct inode, i_lru);
778 list_del_init(&inode->i_lru);
786 * evict_inodes - evict all evictable inodes for a superblock
787 * @sb: superblock to operate on
789 * Make sure that no inodes with zero refcount are retained. This is
790 * called by superblock shutdown after having SB_ACTIVE flag removed,
791 * so any inode reaching zero refcount during or after that call will
792 * be immediately evicted.
794 void evict_inodes(struct super_block *sb)
796 struct inode *inode, *next;
800 spin_lock(&sb->s_inode_list_lock);
801 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
802 if (atomic_read(&inode->i_count))
805 spin_lock(&inode->i_lock);
806 if (atomic_read(&inode->i_count)) {
807 spin_unlock(&inode->i_lock);
810 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
811 spin_unlock(&inode->i_lock);
815 inode->i_state |= I_FREEING;
816 inode_lru_list_del(inode);
817 spin_unlock(&inode->i_lock);
818 list_add(&inode->i_lru, &dispose);
821 * We can have a ton of inodes to evict at unmount time given
822 * enough memory, check to see if we need to go to sleep for a
823 * bit so we don't livelock.
825 if (need_resched()) {
826 spin_unlock(&sb->s_inode_list_lock);
828 dispose_list(&dispose);
832 spin_unlock(&sb->s_inode_list_lock);
834 dispose_list(&dispose);
836 EXPORT_SYMBOL_GPL(evict_inodes);
839 * invalidate_inodes - attempt to free all inodes on a superblock
840 * @sb: superblock to operate on
842 * Attempts to free all inodes (including dirty inodes) for a given superblock.
844 void invalidate_inodes(struct super_block *sb)
846 struct inode *inode, *next;
850 spin_lock(&sb->s_inode_list_lock);
851 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
852 spin_lock(&inode->i_lock);
853 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
854 spin_unlock(&inode->i_lock);
857 if (atomic_read(&inode->i_count)) {
858 spin_unlock(&inode->i_lock);
862 inode->i_state |= I_FREEING;
863 inode_lru_list_del(inode);
864 spin_unlock(&inode->i_lock);
865 list_add(&inode->i_lru, &dispose);
866 if (need_resched()) {
867 spin_unlock(&sb->s_inode_list_lock);
869 dispose_list(&dispose);
873 spin_unlock(&sb->s_inode_list_lock);
875 dispose_list(&dispose);
879 * Isolate the inode from the LRU in preparation for freeing it.
881 * If the inode has the I_REFERENCED flag set, then it means that it has been
882 * used recently - the flag is set in iput_final(). When we encounter such an
883 * inode, clear the flag and move it to the back of the LRU so it gets another
884 * pass through the LRU before it gets reclaimed. This is necessary because of
885 * the fact we are doing lazy LRU updates to minimise lock contention so the
886 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
887 * with this flag set because they are the inodes that are out of order.
889 static enum lru_status inode_lru_isolate(struct list_head *item,
890 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
892 struct list_head *freeable = arg;
893 struct inode *inode = container_of(item, struct inode, i_lru);
896 * We are inverting the lru lock/inode->i_lock here, so use a
897 * trylock. If we fail to get the lock, just skip it.
899 if (!spin_trylock(&inode->i_lock))
903 * Inodes can get referenced, redirtied, or repopulated while
904 * they're already on the LRU, and this can make them
905 * unreclaimable for a while. Remove them lazily here; iput,
906 * sync, or the last page cache deletion will requeue them.
908 if (atomic_read(&inode->i_count) ||
909 (inode->i_state & ~I_REFERENCED) ||
910 !mapping_shrinkable(&inode->i_data)) {
911 list_lru_isolate(lru, &inode->i_lru);
912 spin_unlock(&inode->i_lock);
913 this_cpu_dec(nr_unused);
917 /* Recently referenced inodes get one more pass */
918 if (inode->i_state & I_REFERENCED) {
919 inode->i_state &= ~I_REFERENCED;
920 spin_unlock(&inode->i_lock);
925 * On highmem systems, mapping_shrinkable() permits dropping
926 * page cache in order to free up struct inodes: lowmem might
927 * be under pressure before the cache inside the highmem zone.
929 if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
930 inode_pin_lru_isolating(inode);
931 spin_unlock(&inode->i_lock);
932 spin_unlock(lru_lock);
933 if (remove_inode_buffers(inode)) {
935 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
936 if (current_is_kswapd())
937 __count_vm_events(KSWAPD_INODESTEAL, reap);
939 __count_vm_events(PGINODESTEAL, reap);
940 mm_account_reclaimed_pages(reap);
942 inode_unpin_lru_isolating(inode);
947 WARN_ON(inode->i_state & I_NEW);
948 inode->i_state |= I_FREEING;
949 list_lru_isolate_move(lru, &inode->i_lru, freeable);
950 spin_unlock(&inode->i_lock);
952 this_cpu_dec(nr_unused);
957 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
958 * This is called from the superblock shrinker function with a number of inodes
959 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
960 * then are freed outside inode_lock by dispose_list().
962 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
967 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
968 inode_lru_isolate, &freeable);
969 dispose_list(&freeable);
973 static void __wait_on_freeing_inode(struct inode *inode, bool is_inode_hash_locked);
975 * Called with the inode lock held.
977 static struct inode *find_inode(struct super_block *sb,
978 struct hlist_head *head,
979 int (*test)(struct inode *, void *),
980 void *data, bool is_inode_hash_locked)
982 struct inode *inode = NULL;
984 if (is_inode_hash_locked)
985 lockdep_assert_held(&inode_hash_lock);
987 lockdep_assert_not_held(&inode_hash_lock);
991 hlist_for_each_entry_rcu(inode, head, i_hash) {
992 if (inode->i_sb != sb)
994 if (!test(inode, data))
996 spin_lock(&inode->i_lock);
997 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
998 __wait_on_freeing_inode(inode, is_inode_hash_locked);
1001 if (unlikely(inode->i_state & I_CREATING)) {
1002 spin_unlock(&inode->i_lock);
1004 return ERR_PTR(-ESTALE);
1007 spin_unlock(&inode->i_lock);
1016 * find_inode_fast is the fast path version of find_inode, see the comment at
1017 * iget_locked for details.
1019 static struct inode *find_inode_fast(struct super_block *sb,
1020 struct hlist_head *head, unsigned long ino,
1021 bool is_inode_hash_locked)
1023 struct inode *inode = NULL;
1025 if (is_inode_hash_locked)
1026 lockdep_assert_held(&inode_hash_lock);
1028 lockdep_assert_not_held(&inode_hash_lock);
1032 hlist_for_each_entry_rcu(inode, head, i_hash) {
1033 if (inode->i_ino != ino)
1035 if (inode->i_sb != sb)
1037 spin_lock(&inode->i_lock);
1038 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1039 __wait_on_freeing_inode(inode, is_inode_hash_locked);
1042 if (unlikely(inode->i_state & I_CREATING)) {
1043 spin_unlock(&inode->i_lock);
1045 return ERR_PTR(-ESTALE);
1048 spin_unlock(&inode->i_lock);
1057 * Each cpu owns a range of LAST_INO_BATCH numbers.
1058 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
1059 * to renew the exhausted range.
1061 * This does not significantly increase overflow rate because every CPU can
1062 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
1063 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
1064 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
1065 * overflow rate by 2x, which does not seem too significant.
1067 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1068 * error if st_ino won't fit in target struct field. Use 32bit counter
1069 * here to attempt to avoid that.
1071 #define LAST_INO_BATCH 1024
1072 static DEFINE_PER_CPU(unsigned int, last_ino);
1074 unsigned int get_next_ino(void)
1076 unsigned int *p = &get_cpu_var(last_ino);
1077 unsigned int res = *p;
1080 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
1081 static atomic_t shared_last_ino;
1082 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
1084 res = next - LAST_INO_BATCH;
1089 /* get_next_ino should not provide a 0 inode number */
1093 put_cpu_var(last_ino);
1096 EXPORT_SYMBOL(get_next_ino);
1099 * new_inode_pseudo - obtain an inode
1102 * Allocates a new inode for given superblock.
1103 * Inode wont be chained in superblock s_inodes list
1105 * - fs can't be unmount
1106 * - quotas, fsnotify, writeback can't work
1108 struct inode *new_inode_pseudo(struct super_block *sb)
1110 return alloc_inode(sb);
1114 * new_inode - obtain an inode
1117 * Allocates a new inode for given superblock. The default gfp_mask
1118 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
1119 * If HIGHMEM pages are unsuitable or it is known that pages allocated
1120 * for the page cache are not reclaimable or migratable,
1121 * mapping_set_gfp_mask() must be called with suitable flags on the
1122 * newly created inode's mapping
1125 struct inode *new_inode(struct super_block *sb)
1127 struct inode *inode;
1129 inode = new_inode_pseudo(sb);
1131 inode_sb_list_add(inode);
1134 EXPORT_SYMBOL(new_inode);
1136 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1137 void lockdep_annotate_inode_mutex_key(struct inode *inode)
1139 if (S_ISDIR(inode->i_mode)) {
1140 struct file_system_type *type = inode->i_sb->s_type;
1142 /* Set new key only if filesystem hasn't already changed it */
1143 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
1145 * ensure nobody is actually holding i_mutex
1147 // mutex_destroy(&inode->i_mutex);
1148 init_rwsem(&inode->i_rwsem);
1149 lockdep_set_class(&inode->i_rwsem,
1150 &type->i_mutex_dir_key);
1154 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1158 * unlock_new_inode - clear the I_NEW state and wake up any waiters
1159 * @inode: new inode to unlock
1161 * Called when the inode is fully initialised to clear the new state of the
1162 * inode and wake up anyone waiting for the inode to finish initialisation.
1164 void unlock_new_inode(struct inode *inode)
1166 lockdep_annotate_inode_mutex_key(inode);
1167 spin_lock(&inode->i_lock);
1168 WARN_ON(!(inode->i_state & I_NEW));
1169 inode->i_state &= ~I_NEW & ~I_CREATING;
1171 * Pairs with the barrier in prepare_to_wait_event() to make sure
1172 * ___wait_var_event() either sees the bit cleared or
1173 * waitqueue_active() check in wake_up_var() sees the waiter.
1176 inode_wake_up_bit(inode, __I_NEW);
1177 spin_unlock(&inode->i_lock);
1179 EXPORT_SYMBOL(unlock_new_inode);
1181 void discard_new_inode(struct inode *inode)
1183 lockdep_annotate_inode_mutex_key(inode);
1184 spin_lock(&inode->i_lock);
1185 WARN_ON(!(inode->i_state & I_NEW));
1186 inode->i_state &= ~I_NEW;
1188 * Pairs with the barrier in prepare_to_wait_event() to make sure
1189 * ___wait_var_event() either sees the bit cleared or
1190 * waitqueue_active() check in wake_up_var() sees the waiter.
1193 inode_wake_up_bit(inode, __I_NEW);
1194 spin_unlock(&inode->i_lock);
1197 EXPORT_SYMBOL(discard_new_inode);
1200 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1202 * Lock any non-NULL argument. Passed objects must not be directories.
1203 * Zero, one or two objects may be locked by this function.
1205 * @inode1: first inode to lock
1206 * @inode2: second inode to lock
1208 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1211 WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1213 WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1214 if (inode1 > inode2)
1215 swap(inode1, inode2);
1218 if (inode2 && inode2 != inode1)
1219 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1221 EXPORT_SYMBOL(lock_two_nondirectories);
1224 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1225 * @inode1: first inode to unlock
1226 * @inode2: second inode to unlock
1228 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1231 WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1232 inode_unlock(inode1);
1234 if (inode2 && inode2 != inode1) {
1235 WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1236 inode_unlock(inode2);
1239 EXPORT_SYMBOL(unlock_two_nondirectories);
1242 * inode_insert5 - obtain an inode from a mounted file system
1243 * @inode: pre-allocated inode to use for insert to cache
1244 * @hashval: hash value (usually inode number) to get
1245 * @test: callback used for comparisons between inodes
1246 * @set: callback used to initialize a new struct inode
1247 * @data: opaque data pointer to pass to @test and @set
1249 * Search for the inode specified by @hashval and @data in the inode cache,
1250 * and if present it is return it with an increased reference count. This is
1251 * a variant of iget5_locked() for callers that don't want to fail on memory
1252 * allocation of inode.
1254 * If the inode is not in cache, insert the pre-allocated inode to cache and
1255 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1256 * to fill it in before unlocking it via unlock_new_inode().
1258 * Note both @test and @set are called with the inode_hash_lock held, so can't
1261 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1262 int (*test)(struct inode *, void *),
1263 int (*set)(struct inode *, void *), void *data)
1265 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1269 spin_lock(&inode_hash_lock);
1270 old = find_inode(inode->i_sb, head, test, data, true);
1271 if (unlikely(old)) {
1273 * Uhhuh, somebody else created the same inode under us.
1274 * Use the old inode instead of the preallocated one.
1276 spin_unlock(&inode_hash_lock);
1280 if (unlikely(inode_unhashed(old))) {
1287 if (set && unlikely(set(inode, data))) {
1293 * Return the locked inode with I_NEW set, the
1294 * caller is responsible for filling in the contents
1296 spin_lock(&inode->i_lock);
1297 inode->i_state |= I_NEW;
1298 hlist_add_head_rcu(&inode->i_hash, head);
1299 spin_unlock(&inode->i_lock);
1302 * Add inode to the sb list if it's not already. It has I_NEW at this
1303 * point, so it should be safe to test i_sb_list locklessly.
1305 if (list_empty(&inode->i_sb_list))
1306 inode_sb_list_add(inode);
1308 spin_unlock(&inode_hash_lock);
1312 EXPORT_SYMBOL(inode_insert5);
1315 * iget5_locked - obtain an inode from a mounted file system
1316 * @sb: super block of file system
1317 * @hashval: hash value (usually inode number) to get
1318 * @test: callback used for comparisons between inodes
1319 * @set: callback used to initialize a new struct inode
1320 * @data: opaque data pointer to pass to @test and @set
1322 * Search for the inode specified by @hashval and @data in the inode cache,
1323 * and if present it is return it with an increased reference count. This is
1324 * a generalized version of iget_locked() for file systems where the inode
1325 * number is not sufficient for unique identification of an inode.
1327 * If the inode is not in cache, allocate a new inode and return it locked,
1328 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1329 * before unlocking it via unlock_new_inode().
1331 * Note both @test and @set are called with the inode_hash_lock held, so can't
1334 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1335 int (*test)(struct inode *, void *),
1336 int (*set)(struct inode *, void *), void *data)
1338 struct inode *inode = ilookup5(sb, hashval, test, data);
1341 struct inode *new = alloc_inode(sb);
1344 inode = inode_insert5(new, hashval, test, set, data);
1345 if (unlikely(inode != new))
1351 EXPORT_SYMBOL(iget5_locked);
1354 * iget5_locked_rcu - obtain an inode from a mounted file system
1355 * @sb: super block of file system
1356 * @hashval: hash value (usually inode number) to get
1357 * @test: callback used for comparisons between inodes
1358 * @set: callback used to initialize a new struct inode
1359 * @data: opaque data pointer to pass to @test and @set
1361 * This is equivalent to iget5_locked, except the @test callback must
1362 * tolerate the inode not being stable, including being mid-teardown.
1364 struct inode *iget5_locked_rcu(struct super_block *sb, unsigned long hashval,
1365 int (*test)(struct inode *, void *),
1366 int (*set)(struct inode *, void *), void *data)
1368 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1369 struct inode *inode, *new;
1372 inode = find_inode(sb, head, test, data, false);
1376 wait_on_inode(inode);
1377 if (unlikely(inode_unhashed(inode))) {
1384 new = alloc_inode(sb);
1386 inode = inode_insert5(new, hashval, test, set, data);
1387 if (unlikely(inode != new))
1392 EXPORT_SYMBOL_GPL(iget5_locked_rcu);
1395 * iget_locked - obtain an inode from a mounted file system
1396 * @sb: super block of file system
1397 * @ino: inode number to get
1399 * Search for the inode specified by @ino in the inode cache and if present
1400 * return it with an increased reference count. This is for file systems
1401 * where the inode number is sufficient for unique identification of an inode.
1403 * If the inode is not in cache, allocate a new inode and return it locked,
1404 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1405 * before unlocking it via unlock_new_inode().
1407 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1409 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1410 struct inode *inode;
1412 inode = find_inode_fast(sb, head, ino, false);
1416 wait_on_inode(inode);
1417 if (unlikely(inode_unhashed(inode))) {
1424 inode = alloc_inode(sb);
1428 spin_lock(&inode_hash_lock);
1429 /* We released the lock, so.. */
1430 old = find_inode_fast(sb, head, ino, true);
1433 spin_lock(&inode->i_lock);
1434 inode->i_state = I_NEW;
1435 hlist_add_head_rcu(&inode->i_hash, head);
1436 spin_unlock(&inode->i_lock);
1437 inode_sb_list_add(inode);
1438 spin_unlock(&inode_hash_lock);
1440 /* Return the locked inode with I_NEW set, the
1441 * caller is responsible for filling in the contents
1447 * Uhhuh, somebody else created the same inode under
1448 * us. Use the old inode instead of the one we just
1451 spin_unlock(&inode_hash_lock);
1452 destroy_inode(inode);
1456 wait_on_inode(inode);
1457 if (unlikely(inode_unhashed(inode))) {
1464 EXPORT_SYMBOL(iget_locked);
1467 * search the inode cache for a matching inode number.
1468 * If we find one, then the inode number we are trying to
1469 * allocate is not unique and so we should not use it.
1471 * Returns 1 if the inode number is unique, 0 if it is not.
1473 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1475 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1476 struct inode *inode;
1478 hlist_for_each_entry_rcu(inode, b, i_hash) {
1479 if (inode->i_ino == ino && inode->i_sb == sb)
1486 * iunique - get a unique inode number
1488 * @max_reserved: highest reserved inode number
1490 * Obtain an inode number that is unique on the system for a given
1491 * superblock. This is used by file systems that have no natural
1492 * permanent inode numbering system. An inode number is returned that
1493 * is higher than the reserved limit but unique.
1496 * With a large number of inodes live on the file system this function
1497 * currently becomes quite slow.
1499 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1502 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1503 * error if st_ino won't fit in target struct field. Use 32bit counter
1504 * here to attempt to avoid that.
1506 static DEFINE_SPINLOCK(iunique_lock);
1507 static unsigned int counter;
1511 spin_lock(&iunique_lock);
1513 if (counter <= max_reserved)
1514 counter = max_reserved + 1;
1516 } while (!test_inode_iunique(sb, res));
1517 spin_unlock(&iunique_lock);
1522 EXPORT_SYMBOL(iunique);
1524 struct inode *igrab(struct inode *inode)
1526 spin_lock(&inode->i_lock);
1527 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1529 spin_unlock(&inode->i_lock);
1531 spin_unlock(&inode->i_lock);
1533 * Handle the case where s_op->clear_inode is not been
1534 * called yet, and somebody is calling igrab
1535 * while the inode is getting freed.
1541 EXPORT_SYMBOL(igrab);
1544 * ilookup5_nowait - search for an inode in the inode cache
1545 * @sb: super block of file system to search
1546 * @hashval: hash value (usually inode number) to search for
1547 * @test: callback used for comparisons between inodes
1548 * @data: opaque data pointer to pass to @test
1550 * Search for the inode specified by @hashval and @data in the inode cache.
1551 * If the inode is in the cache, the inode is returned with an incremented
1554 * Note: I_NEW is not waited upon so you have to be very careful what you do
1555 * with the returned inode. You probably should be using ilookup5() instead.
1557 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1559 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1560 int (*test)(struct inode *, void *), void *data)
1562 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1563 struct inode *inode;
1565 spin_lock(&inode_hash_lock);
1566 inode = find_inode(sb, head, test, data, true);
1567 spin_unlock(&inode_hash_lock);
1569 return IS_ERR(inode) ? NULL : inode;
1571 EXPORT_SYMBOL(ilookup5_nowait);
1574 * ilookup5 - search for an inode in the inode cache
1575 * @sb: super block of file system to search
1576 * @hashval: hash value (usually inode number) to search for
1577 * @test: callback used for comparisons between inodes
1578 * @data: opaque data pointer to pass to @test
1580 * Search for the inode specified by @hashval and @data in the inode cache,
1581 * and if the inode is in the cache, return the inode with an incremented
1582 * reference count. Waits on I_NEW before returning the inode.
1583 * returned with an incremented reference count.
1585 * This is a generalized version of ilookup() for file systems where the
1586 * inode number is not sufficient for unique identification of an inode.
1588 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1590 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1591 int (*test)(struct inode *, void *), void *data)
1593 struct inode *inode;
1595 inode = ilookup5_nowait(sb, hashval, test, data);
1597 wait_on_inode(inode);
1598 if (unlikely(inode_unhashed(inode))) {
1605 EXPORT_SYMBOL(ilookup5);
1608 * ilookup - search for an inode in the inode cache
1609 * @sb: super block of file system to search
1610 * @ino: inode number to search for
1612 * Search for the inode @ino in the inode cache, and if the inode is in the
1613 * cache, the inode is returned with an incremented reference count.
1615 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1617 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1618 struct inode *inode;
1620 inode = find_inode_fast(sb, head, ino, false);
1625 wait_on_inode(inode);
1626 if (unlikely(inode_unhashed(inode))) {
1633 EXPORT_SYMBOL(ilookup);
1636 * find_inode_nowait - find an inode in the inode cache
1637 * @sb: super block of file system to search
1638 * @hashval: hash value (usually inode number) to search for
1639 * @match: callback used for comparisons between inodes
1640 * @data: opaque data pointer to pass to @match
1642 * Search for the inode specified by @hashval and @data in the inode
1643 * cache, where the helper function @match will return 0 if the inode
1644 * does not match, 1 if the inode does match, and -1 if the search
1645 * should be stopped. The @match function must be responsible for
1646 * taking the i_lock spin_lock and checking i_state for an inode being
1647 * freed or being initialized, and incrementing the reference count
1648 * before returning 1. It also must not sleep, since it is called with
1649 * the inode_hash_lock spinlock held.
1651 * This is a even more generalized version of ilookup5() when the
1652 * function must never block --- find_inode() can block in
1653 * __wait_on_freeing_inode() --- or when the caller can not increment
1654 * the reference count because the resulting iput() might cause an
1655 * inode eviction. The tradeoff is that the @match funtion must be
1656 * very carefully implemented.
1658 struct inode *find_inode_nowait(struct super_block *sb,
1659 unsigned long hashval,
1660 int (*match)(struct inode *, unsigned long,
1664 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1665 struct inode *inode, *ret_inode = NULL;
1668 spin_lock(&inode_hash_lock);
1669 hlist_for_each_entry(inode, head, i_hash) {
1670 if (inode->i_sb != sb)
1672 mval = match(inode, hashval, data);
1680 spin_unlock(&inode_hash_lock);
1683 EXPORT_SYMBOL(find_inode_nowait);
1686 * find_inode_rcu - find an inode in the inode cache
1687 * @sb: Super block of file system to search
1688 * @hashval: Key to hash
1689 * @test: Function to test match on an inode
1690 * @data: Data for test function
1692 * Search for the inode specified by @hashval and @data in the inode cache,
1693 * where the helper function @test will return 0 if the inode does not match
1694 * and 1 if it does. The @test function must be responsible for taking the
1695 * i_lock spin_lock and checking i_state for an inode being freed or being
1698 * If successful, this will return the inode for which the @test function
1699 * returned 1 and NULL otherwise.
1701 * The @test function is not permitted to take a ref on any inode presented.
1702 * It is also not permitted to sleep.
1704 * The caller must hold the RCU read lock.
1706 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1707 int (*test)(struct inode *, void *), void *data)
1709 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1710 struct inode *inode;
1712 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1713 "suspicious find_inode_rcu() usage");
1715 hlist_for_each_entry_rcu(inode, head, i_hash) {
1716 if (inode->i_sb == sb &&
1717 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1723 EXPORT_SYMBOL(find_inode_rcu);
1726 * find_inode_by_ino_rcu - Find an inode in the inode cache
1727 * @sb: Super block of file system to search
1728 * @ino: The inode number to match
1730 * Search for the inode specified by @hashval and @data in the inode cache,
1731 * where the helper function @test will return 0 if the inode does not match
1732 * and 1 if it does. The @test function must be responsible for taking the
1733 * i_lock spin_lock and checking i_state for an inode being freed or being
1736 * If successful, this will return the inode for which the @test function
1737 * returned 1 and NULL otherwise.
1739 * The @test function is not permitted to take a ref on any inode presented.
1740 * It is also not permitted to sleep.
1742 * The caller must hold the RCU read lock.
1744 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1747 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1748 struct inode *inode;
1750 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1751 "suspicious find_inode_by_ino_rcu() usage");
1753 hlist_for_each_entry_rcu(inode, head, i_hash) {
1754 if (inode->i_ino == ino &&
1755 inode->i_sb == sb &&
1756 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1761 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1763 int insert_inode_locked(struct inode *inode)
1765 struct super_block *sb = inode->i_sb;
1766 ino_t ino = inode->i_ino;
1767 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1770 struct inode *old = NULL;
1771 spin_lock(&inode_hash_lock);
1772 hlist_for_each_entry(old, head, i_hash) {
1773 if (old->i_ino != ino)
1775 if (old->i_sb != sb)
1777 spin_lock(&old->i_lock);
1778 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1779 spin_unlock(&old->i_lock);
1785 spin_lock(&inode->i_lock);
1786 inode->i_state |= I_NEW | I_CREATING;
1787 hlist_add_head_rcu(&inode->i_hash, head);
1788 spin_unlock(&inode->i_lock);
1789 spin_unlock(&inode_hash_lock);
1792 if (unlikely(old->i_state & I_CREATING)) {
1793 spin_unlock(&old->i_lock);
1794 spin_unlock(&inode_hash_lock);
1798 spin_unlock(&old->i_lock);
1799 spin_unlock(&inode_hash_lock);
1801 if (unlikely(!inode_unhashed(old))) {
1808 EXPORT_SYMBOL(insert_inode_locked);
1810 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1811 int (*test)(struct inode *, void *), void *data)
1815 inode->i_state |= I_CREATING;
1816 old = inode_insert5(inode, hashval, test, NULL, data);
1824 EXPORT_SYMBOL(insert_inode_locked4);
1827 int generic_delete_inode(struct inode *inode)
1831 EXPORT_SYMBOL(generic_delete_inode);
1834 * Called when we're dropping the last reference
1837 * Call the FS "drop_inode()" function, defaulting to
1838 * the legacy UNIX filesystem behaviour. If it tells
1839 * us to evict inode, do so. Otherwise, retain inode
1840 * in cache if fs is alive, sync and evict if fs is
1843 static void iput_final(struct inode *inode)
1845 struct super_block *sb = inode->i_sb;
1846 const struct super_operations *op = inode->i_sb->s_op;
1847 unsigned long state;
1850 WARN_ON(inode->i_state & I_NEW);
1853 drop = op->drop_inode(inode);
1855 drop = generic_drop_inode(inode);
1858 !(inode->i_state & I_DONTCACHE) &&
1859 (sb->s_flags & SB_ACTIVE)) {
1860 __inode_add_lru(inode, true);
1861 spin_unlock(&inode->i_lock);
1865 state = inode->i_state;
1867 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1868 spin_unlock(&inode->i_lock);
1870 write_inode_now(inode, 1);
1872 spin_lock(&inode->i_lock);
1873 state = inode->i_state;
1874 WARN_ON(state & I_NEW);
1875 state &= ~I_WILL_FREE;
1878 WRITE_ONCE(inode->i_state, state | I_FREEING);
1879 if (!list_empty(&inode->i_lru))
1880 inode_lru_list_del(inode);
1881 spin_unlock(&inode->i_lock);
1887 * iput - put an inode
1888 * @inode: inode to put
1890 * Puts an inode, dropping its usage count. If the inode use count hits
1891 * zero, the inode is then freed and may also be destroyed.
1893 * Consequently, iput() can sleep.
1895 void iput(struct inode *inode)
1899 BUG_ON(inode->i_state & I_CLEAR);
1901 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1902 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1903 atomic_inc(&inode->i_count);
1904 spin_unlock(&inode->i_lock);
1905 trace_writeback_lazytime_iput(inode);
1906 mark_inode_dirty_sync(inode);
1912 EXPORT_SYMBOL(iput);
1916 * bmap - find a block number in a file
1917 * @inode: inode owning the block number being requested
1918 * @block: pointer containing the block to find
1920 * Replaces the value in ``*block`` with the block number on the device holding
1921 * corresponding to the requested block number in the file.
1922 * That is, asked for block 4 of inode 1 the function will replace the
1923 * 4 in ``*block``, with disk block relative to the disk start that holds that
1924 * block of the file.
1926 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1927 * hole, returns 0 and ``*block`` is also set to 0.
1929 int bmap(struct inode *inode, sector_t *block)
1931 if (!inode->i_mapping->a_ops->bmap)
1934 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1937 EXPORT_SYMBOL(bmap);
1941 * With relative atime, only update atime if the previous atime is
1942 * earlier than or equal to either the ctime or mtime,
1943 * or if at least a day has passed since the last atime update.
1945 static bool relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1946 struct timespec64 now)
1948 struct timespec64 atime, mtime, ctime;
1950 if (!(mnt->mnt_flags & MNT_RELATIME))
1953 * Is mtime younger than or equal to atime? If yes, update atime:
1955 atime = inode_get_atime(inode);
1956 mtime = inode_get_mtime(inode);
1957 if (timespec64_compare(&mtime, &atime) >= 0)
1960 * Is ctime younger than or equal to atime? If yes, update atime:
1962 ctime = inode_get_ctime(inode);
1963 if (timespec64_compare(&ctime, &atime) >= 0)
1967 * Is the previous atime value older than a day? If yes,
1970 if ((long)(now.tv_sec - atime.tv_sec) >= 24*60*60)
1973 * Good, we can skip the atime update:
1979 * inode_update_timestamps - update the timestamps on the inode
1980 * @inode: inode to be updated
1981 * @flags: S_* flags that needed to be updated
1983 * The update_time function is called when an inode's timestamps need to be
1984 * updated for a read or write operation. This function handles updating the
1985 * actual timestamps. It's up to the caller to ensure that the inode is marked
1986 * dirty appropriately.
1988 * In the case where any of S_MTIME, S_CTIME, or S_VERSION need to be updated,
1989 * attempt to update all three of them. S_ATIME updates can be handled
1990 * independently of the rest.
1992 * Returns a set of S_* flags indicating which values changed.
1994 int inode_update_timestamps(struct inode *inode, int flags)
1997 struct timespec64 now;
1999 if (flags & (S_MTIME|S_CTIME|S_VERSION)) {
2000 struct timespec64 ctime = inode_get_ctime(inode);
2001 struct timespec64 mtime = inode_get_mtime(inode);
2003 now = inode_set_ctime_current(inode);
2004 if (!timespec64_equal(&now, &ctime))
2006 if (!timespec64_equal(&now, &mtime)) {
2007 inode_set_mtime_to_ts(inode, now);
2010 if (IS_I_VERSION(inode) && inode_maybe_inc_iversion(inode, updated))
2011 updated |= S_VERSION;
2013 now = current_time(inode);
2016 if (flags & S_ATIME) {
2017 struct timespec64 atime = inode_get_atime(inode);
2019 if (!timespec64_equal(&now, &atime)) {
2020 inode_set_atime_to_ts(inode, now);
2026 EXPORT_SYMBOL(inode_update_timestamps);
2029 * generic_update_time - update the timestamps on the inode
2030 * @inode: inode to be updated
2031 * @flags: S_* flags that needed to be updated
2033 * The update_time function is called when an inode's timestamps need to be
2034 * updated for a read or write operation. In the case where any of S_MTIME, S_CTIME,
2035 * or S_VERSION need to be updated we attempt to update all three of them. S_ATIME
2036 * updates can be handled done independently of the rest.
2038 * Returns a S_* mask indicating which fields were updated.
2040 int generic_update_time(struct inode *inode, int flags)
2042 int updated = inode_update_timestamps(inode, flags);
2043 int dirty_flags = 0;
2045 if (updated & (S_ATIME|S_MTIME|S_CTIME))
2046 dirty_flags = inode->i_sb->s_flags & SB_LAZYTIME ? I_DIRTY_TIME : I_DIRTY_SYNC;
2047 if (updated & S_VERSION)
2048 dirty_flags |= I_DIRTY_SYNC;
2049 __mark_inode_dirty(inode, dirty_flags);
2052 EXPORT_SYMBOL(generic_update_time);
2055 * This does the actual work of updating an inodes time or version. Must have
2056 * had called mnt_want_write() before calling this.
2058 int inode_update_time(struct inode *inode, int flags)
2060 if (inode->i_op->update_time)
2061 return inode->i_op->update_time(inode, flags);
2062 generic_update_time(inode, flags);
2065 EXPORT_SYMBOL(inode_update_time);
2068 * atime_needs_update - update the access time
2069 * @path: the &struct path to update
2070 * @inode: inode to update
2072 * Update the accessed time on an inode and mark it for writeback.
2073 * This function automatically handles read only file systems and media,
2074 * as well as the "noatime" flag and inode specific "noatime" markers.
2076 bool atime_needs_update(const struct path *path, struct inode *inode)
2078 struct vfsmount *mnt = path->mnt;
2079 struct timespec64 now, atime;
2081 if (inode->i_flags & S_NOATIME)
2084 /* Atime updates will likely cause i_uid and i_gid to be written
2085 * back improprely if their true value is unknown to the vfs.
2087 if (HAS_UNMAPPED_ID(mnt_idmap(mnt), inode))
2090 if (IS_NOATIME(inode))
2092 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
2095 if (mnt->mnt_flags & MNT_NOATIME)
2097 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
2100 now = current_time(inode);
2102 if (!relatime_need_update(mnt, inode, now))
2105 atime = inode_get_atime(inode);
2106 if (timespec64_equal(&atime, &now))
2112 void touch_atime(const struct path *path)
2114 struct vfsmount *mnt = path->mnt;
2115 struct inode *inode = d_inode(path->dentry);
2117 if (!atime_needs_update(path, inode))
2120 if (!sb_start_write_trylock(inode->i_sb))
2123 if (mnt_get_write_access(mnt) != 0)
2126 * File systems can error out when updating inodes if they need to
2127 * allocate new space to modify an inode (such is the case for
2128 * Btrfs), but since we touch atime while walking down the path we
2129 * really don't care if we failed to update the atime of the file,
2130 * so just ignore the return value.
2131 * We may also fail on filesystems that have the ability to make parts
2132 * of the fs read only, e.g. subvolumes in Btrfs.
2134 inode_update_time(inode, S_ATIME);
2135 mnt_put_write_access(mnt);
2137 sb_end_write(inode->i_sb);
2139 EXPORT_SYMBOL(touch_atime);
2142 * Return mask of changes for notify_change() that need to be done as a
2143 * response to write or truncate. Return 0 if nothing has to be changed.
2144 * Negative value on error (change should be denied).
2146 int dentry_needs_remove_privs(struct mnt_idmap *idmap,
2147 struct dentry *dentry)
2149 struct inode *inode = d_inode(dentry);
2153 if (IS_NOSEC(inode))
2156 mask = setattr_should_drop_suidgid(idmap, inode);
2157 ret = security_inode_need_killpriv(dentry);
2161 mask |= ATTR_KILL_PRIV;
2165 static int __remove_privs(struct mnt_idmap *idmap,
2166 struct dentry *dentry, int kill)
2168 struct iattr newattrs;
2170 newattrs.ia_valid = ATTR_FORCE | kill;
2172 * Note we call this on write, so notify_change will not
2173 * encounter any conflicting delegations:
2175 return notify_change(idmap, dentry, &newattrs, NULL);
2178 int file_remove_privs_flags(struct file *file, unsigned int flags)
2180 struct dentry *dentry = file_dentry(file);
2181 struct inode *inode = file_inode(file);
2185 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
2188 kill = dentry_needs_remove_privs(file_mnt_idmap(file), dentry);
2193 if (flags & IOCB_NOWAIT)
2196 error = __remove_privs(file_mnt_idmap(file), dentry, kill);
2200 inode_has_no_xattr(inode);
2203 EXPORT_SYMBOL_GPL(file_remove_privs_flags);
2206 * file_remove_privs - remove special file privileges (suid, capabilities)
2207 * @file: file to remove privileges from
2209 * When file is modified by a write or truncation ensure that special
2210 * file privileges are removed.
2212 * Return: 0 on success, negative errno on failure.
2214 int file_remove_privs(struct file *file)
2216 return file_remove_privs_flags(file, 0);
2218 EXPORT_SYMBOL(file_remove_privs);
2220 static int inode_needs_update_time(struct inode *inode)
2223 struct timespec64 now = current_time(inode);
2224 struct timespec64 ts;
2226 /* First try to exhaust all avenues to not sync */
2227 if (IS_NOCMTIME(inode))
2230 ts = inode_get_mtime(inode);
2231 if (!timespec64_equal(&ts, &now))
2234 ts = inode_get_ctime(inode);
2235 if (!timespec64_equal(&ts, &now))
2238 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2239 sync_it |= S_VERSION;
2244 static int __file_update_time(struct file *file, int sync_mode)
2247 struct inode *inode = file_inode(file);
2249 /* try to update time settings */
2250 if (!mnt_get_write_access_file(file)) {
2251 ret = inode_update_time(inode, sync_mode);
2252 mnt_put_write_access_file(file);
2259 * file_update_time - update mtime and ctime time
2260 * @file: file accessed
2262 * Update the mtime and ctime members of an inode and mark the inode for
2263 * writeback. Note that this function is meant exclusively for usage in
2264 * the file write path of filesystems, and filesystems may choose to
2265 * explicitly ignore updates via this function with the _NOCMTIME inode
2266 * flag, e.g. for network filesystem where these imestamps are handled
2267 * by the server. This can return an error for file systems who need to
2268 * allocate space in order to update an inode.
2270 * Return: 0 on success, negative errno on failure.
2272 int file_update_time(struct file *file)
2275 struct inode *inode = file_inode(file);
2277 ret = inode_needs_update_time(inode);
2281 return __file_update_time(file, ret);
2283 EXPORT_SYMBOL(file_update_time);
2286 * file_modified_flags - handle mandated vfs changes when modifying a file
2287 * @file: file that was modified
2288 * @flags: kiocb flags
2290 * When file has been modified ensure that special
2291 * file privileges are removed and time settings are updated.
2293 * If IOCB_NOWAIT is set, special file privileges will not be removed and
2294 * time settings will not be updated. It will return -EAGAIN.
2296 * Context: Caller must hold the file's inode lock.
2298 * Return: 0 on success, negative errno on failure.
2300 static int file_modified_flags(struct file *file, int flags)
2303 struct inode *inode = file_inode(file);
2306 * Clear the security bits if the process is not being run by root.
2307 * This keeps people from modifying setuid and setgid binaries.
2309 ret = file_remove_privs_flags(file, flags);
2313 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2316 ret = inode_needs_update_time(inode);
2319 if (flags & IOCB_NOWAIT)
2322 return __file_update_time(file, ret);
2326 * file_modified - handle mandated vfs changes when modifying a file
2327 * @file: file that was modified
2329 * When file has been modified ensure that special
2330 * file privileges are removed and time settings are updated.
2332 * Context: Caller must hold the file's inode lock.
2334 * Return: 0 on success, negative errno on failure.
2336 int file_modified(struct file *file)
2338 return file_modified_flags(file, 0);
2340 EXPORT_SYMBOL(file_modified);
2343 * kiocb_modified - handle mandated vfs changes when modifying a file
2344 * @iocb: iocb that was modified
2346 * When file has been modified ensure that special
2347 * file privileges are removed and time settings are updated.
2349 * Context: Caller must hold the file's inode lock.
2351 * Return: 0 on success, negative errno on failure.
2353 int kiocb_modified(struct kiocb *iocb)
2355 return file_modified_flags(iocb->ki_filp, iocb->ki_flags);
2357 EXPORT_SYMBOL_GPL(kiocb_modified);
2359 int inode_needs_sync(struct inode *inode)
2363 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2367 EXPORT_SYMBOL(inode_needs_sync);
2370 * If we try to find an inode in the inode hash while it is being
2371 * deleted, we have to wait until the filesystem completes its
2372 * deletion before reporting that it isn't found. This function waits
2373 * until the deletion _might_ have completed. Callers are responsible
2374 * to recheck inode state.
2376 * It doesn't matter if I_NEW is not set initially, a call to
2377 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2380 static void __wait_on_freeing_inode(struct inode *inode, bool is_inode_hash_locked)
2382 struct wait_bit_queue_entry wqe;
2383 struct wait_queue_head *wq_head;
2386 * Handle racing against evict(), see that routine for more details.
2388 if (unlikely(inode_unhashed(inode))) {
2389 WARN_ON(is_inode_hash_locked);
2390 spin_unlock(&inode->i_lock);
2394 wq_head = inode_bit_waitqueue(&wqe, inode, __I_NEW);
2395 prepare_to_wait_event(wq_head, &wqe.wq_entry, TASK_UNINTERRUPTIBLE);
2396 spin_unlock(&inode->i_lock);
2398 if (is_inode_hash_locked)
2399 spin_unlock(&inode_hash_lock);
2401 finish_wait(wq_head, &wqe.wq_entry);
2402 if (is_inode_hash_locked)
2403 spin_lock(&inode_hash_lock);
2407 static __initdata unsigned long ihash_entries;
2408 static int __init set_ihash_entries(char *str)
2412 ihash_entries = simple_strtoul(str, &str, 0);
2415 __setup("ihash_entries=", set_ihash_entries);
2418 * Initialize the waitqueues and inode hash table.
2420 void __init inode_init_early(void)
2422 /* If hashes are distributed across NUMA nodes, defer
2423 * hash allocation until vmalloc space is available.
2429 alloc_large_system_hash("Inode-cache",
2430 sizeof(struct hlist_head),
2433 HASH_EARLY | HASH_ZERO,
2440 void __init inode_init(void)
2442 /* inode slab cache */
2443 inode_cachep = kmem_cache_create("inode_cache",
2444 sizeof(struct inode),
2446 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2450 /* Hash may have been set up in inode_init_early */
2455 alloc_large_system_hash("Inode-cache",
2456 sizeof(struct hlist_head),
2466 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2468 inode->i_mode = mode;
2469 if (S_ISCHR(mode)) {
2470 inode->i_fop = &def_chr_fops;
2471 inode->i_rdev = rdev;
2472 } else if (S_ISBLK(mode)) {
2473 if (IS_ENABLED(CONFIG_BLOCK))
2474 inode->i_fop = &def_blk_fops;
2475 inode->i_rdev = rdev;
2476 } else if (S_ISFIFO(mode))
2477 inode->i_fop = &pipefifo_fops;
2478 else if (S_ISSOCK(mode))
2479 ; /* leave it no_open_fops */
2481 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2482 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2485 EXPORT_SYMBOL(init_special_inode);
2488 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2489 * @idmap: idmap of the mount the inode was created from
2491 * @dir: Directory inode
2492 * @mode: mode of the new inode
2494 * If the inode has been created through an idmapped mount the idmap of
2495 * the vfsmount must be passed through @idmap. This function will then take
2496 * care to map the inode according to @idmap before checking permissions
2497 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2498 * checking is to be performed on the raw inode simply pass @nop_mnt_idmap.
2500 void inode_init_owner(struct mnt_idmap *idmap, struct inode *inode,
2501 const struct inode *dir, umode_t mode)
2503 inode_fsuid_set(inode, idmap);
2504 if (dir && dir->i_mode & S_ISGID) {
2505 inode->i_gid = dir->i_gid;
2507 /* Directories are special, and always inherit S_ISGID */
2511 inode_fsgid_set(inode, idmap);
2512 inode->i_mode = mode;
2514 EXPORT_SYMBOL(inode_init_owner);
2517 * inode_owner_or_capable - check current task permissions to inode
2518 * @idmap: idmap of the mount the inode was found from
2519 * @inode: inode being checked
2521 * Return true if current either has CAP_FOWNER in a namespace with the
2522 * inode owner uid mapped, or owns the file.
2524 * If the inode has been found through an idmapped mount the idmap of
2525 * the vfsmount must be passed through @idmap. This function will then take
2526 * care to map the inode according to @idmap before checking permissions.
2527 * On non-idmapped mounts or if permission checking is to be performed on the
2528 * raw inode simply pass @nop_mnt_idmap.
2530 bool inode_owner_or_capable(struct mnt_idmap *idmap,
2531 const struct inode *inode)
2534 struct user_namespace *ns;
2536 vfsuid = i_uid_into_vfsuid(idmap, inode);
2537 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
2540 ns = current_user_ns();
2541 if (vfsuid_has_mapping(ns, vfsuid) && ns_capable(ns, CAP_FOWNER))
2545 EXPORT_SYMBOL(inode_owner_or_capable);
2548 * Direct i/o helper functions
2550 bool inode_dio_finished(const struct inode *inode)
2552 return atomic_read(&inode->i_dio_count) == 0;
2554 EXPORT_SYMBOL(inode_dio_finished);
2557 * inode_dio_wait - wait for outstanding DIO requests to finish
2558 * @inode: inode to wait for
2560 * Waits for all pending direct I/O requests to finish so that we can
2561 * proceed with a truncate or equivalent operation.
2563 * Must be called under a lock that serializes taking new references
2564 * to i_dio_count, usually by inode->i_mutex.
2566 void inode_dio_wait(struct inode *inode)
2568 wait_var_event(&inode->i_dio_count, inode_dio_finished(inode));
2570 EXPORT_SYMBOL(inode_dio_wait);
2572 void inode_dio_wait_interruptible(struct inode *inode)
2574 wait_var_event_interruptible(&inode->i_dio_count,
2575 inode_dio_finished(inode));
2577 EXPORT_SYMBOL(inode_dio_wait_interruptible);
2580 * inode_set_flags - atomically set some inode flags
2582 * Note: the caller should be holding i_mutex, or else be sure that
2583 * they have exclusive access to the inode structure (i.e., while the
2584 * inode is being instantiated). The reason for the cmpxchg() loop
2585 * --- which wouldn't be necessary if all code paths which modify
2586 * i_flags actually followed this rule, is that there is at least one
2587 * code path which doesn't today so we use cmpxchg() out of an abundance
2590 * In the long run, i_mutex is overkill, and we should probably look
2591 * at using the i_lock spinlock to protect i_flags, and then make sure
2592 * it is so documented in include/linux/fs.h and that all code follows
2593 * the locking convention!!
2595 void inode_set_flags(struct inode *inode, unsigned int flags,
2598 WARN_ON_ONCE(flags & ~mask);
2599 set_mask_bits(&inode->i_flags, mask, flags);
2601 EXPORT_SYMBOL(inode_set_flags);
2603 void inode_nohighmem(struct inode *inode)
2605 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2607 EXPORT_SYMBOL(inode_nohighmem);
2610 * timestamp_truncate - Truncate timespec to a granularity
2612 * @inode: inode being updated
2614 * Truncate a timespec to the granularity supported by the fs
2615 * containing the inode. Always rounds down. gran must
2616 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2618 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2620 struct super_block *sb = inode->i_sb;
2621 unsigned int gran = sb->s_time_gran;
2623 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2624 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2627 /* Avoid division in the common cases 1 ns and 1 s. */
2630 else if (gran == NSEC_PER_SEC)
2632 else if (gran > 1 && gran < NSEC_PER_SEC)
2633 t.tv_nsec -= t.tv_nsec % gran;
2635 WARN(1, "invalid file time granularity: %u", gran);
2638 EXPORT_SYMBOL(timestamp_truncate);
2641 * current_time - Return FS time
2644 * Return the current time truncated to the time granularity supported by
2647 * Note that inode and inode->sb cannot be NULL.
2648 * Otherwise, the function warns and returns time without truncation.
2650 struct timespec64 current_time(struct inode *inode)
2652 struct timespec64 now;
2654 ktime_get_coarse_real_ts64(&now);
2655 return timestamp_truncate(now, inode);
2657 EXPORT_SYMBOL(current_time);
2660 * inode_set_ctime_current - set the ctime to current_time
2663 * Set the inode->i_ctime to the current value for the inode. Returns
2664 * the current value that was assigned to i_ctime.
2666 struct timespec64 inode_set_ctime_current(struct inode *inode)
2668 struct timespec64 now = current_time(inode);
2670 inode_set_ctime_to_ts(inode, now);
2673 EXPORT_SYMBOL(inode_set_ctime_current);
2676 * in_group_or_capable - check whether caller is CAP_FSETID privileged
2677 * @idmap: idmap of the mount @inode was found from
2678 * @inode: inode to check
2679 * @vfsgid: the new/current vfsgid of @inode
2681 * Check wether @vfsgid is in the caller's group list or if the caller is
2682 * privileged with CAP_FSETID over @inode. This can be used to determine
2683 * whether the setgid bit can be kept or must be dropped.
2685 * Return: true if the caller is sufficiently privileged, false if not.
2687 bool in_group_or_capable(struct mnt_idmap *idmap,
2688 const struct inode *inode, vfsgid_t vfsgid)
2690 if (vfsgid_in_group_p(vfsgid))
2692 if (capable_wrt_inode_uidgid(idmap, inode, CAP_FSETID))
2696 EXPORT_SYMBOL(in_group_or_capable);
2699 * mode_strip_sgid - handle the sgid bit for non-directories
2700 * @idmap: idmap of the mount the inode was created from
2701 * @dir: parent directory inode
2702 * @mode: mode of the file to be created in @dir
2704 * If the @mode of the new file has both the S_ISGID and S_IXGRP bit
2705 * raised and @dir has the S_ISGID bit raised ensure that the caller is
2706 * either in the group of the parent directory or they have CAP_FSETID
2707 * in their user namespace and are privileged over the parent directory.
2708 * In all other cases, strip the S_ISGID bit from @mode.
2710 * Return: the new mode to use for the file
2712 umode_t mode_strip_sgid(struct mnt_idmap *idmap,
2713 const struct inode *dir, umode_t mode)
2715 if ((mode & (S_ISGID | S_IXGRP)) != (S_ISGID | S_IXGRP))
2717 if (S_ISDIR(mode) || !dir || !(dir->i_mode & S_ISGID))
2719 if (in_group_or_capable(idmap, dir, i_gid_into_vfsgid(idmap, dir)))
2721 return mode & ~S_ISGID;
2723 EXPORT_SYMBOL(mode_strip_sgid);