2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
5 #include <linux/export.h>
8 #include <linux/backing-dev.h>
9 #include <linux/hash.h>
10 #include <linux/swap.h>
11 #include <linux/security.h>
12 #include <linux/cdev.h>
13 #include <linux/bootmem.h>
14 #include <linux/fsnotify.h>
15 #include <linux/mount.h>
16 #include <linux/posix_acl.h>
17 #include <linux/prefetch.h>
18 #include <linux/buffer_head.h> /* for inode_has_buffers */
19 #include <linux/ratelimit.h>
23 * Inode locking rules:
25 * inode->i_lock protects:
26 * inode->i_state, inode->i_hash, __iget()
27 * inode->i_sb->s_inode_lru_lock protects:
28 * inode->i_sb->s_inode_lru, inode->i_lru
29 * inode_sb_list_lock protects:
30 * sb->s_inodes, inode->i_sb_list
31 * bdi->wb.list_lock protects:
32 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
33 * inode_hash_lock protects:
34 * inode_hashtable, inode->i_hash
40 * inode->i_sb->s_inode_lru_lock
53 static unsigned int i_hash_mask __read_mostly;
54 static unsigned int i_hash_shift __read_mostly;
55 static struct hlist_head *inode_hashtable __read_mostly;
56 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
58 __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
61 * Empty aops. Can be used for the cases where the user does not
62 * define any of the address_space operations.
64 const struct address_space_operations empty_aops = {
66 EXPORT_SYMBOL(empty_aops);
69 * Statistics gathering..
71 struct inodes_stat_t inodes_stat;
73 static DEFINE_PER_CPU(unsigned int, nr_inodes);
74 static DEFINE_PER_CPU(unsigned int, nr_unused);
76 static struct kmem_cache *inode_cachep __read_mostly;
78 static int get_nr_inodes(void)
82 for_each_possible_cpu(i)
83 sum += per_cpu(nr_inodes, i);
84 return sum < 0 ? 0 : sum;
87 static inline int get_nr_inodes_unused(void)
91 for_each_possible_cpu(i)
92 sum += per_cpu(nr_unused, i);
93 return sum < 0 ? 0 : sum;
96 int get_nr_dirty_inodes(void)
98 /* not actually dirty inodes, but a wild approximation */
99 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
100 return nr_dirty > 0 ? nr_dirty : 0;
104 * Handle nr_inode sysctl
107 int proc_nr_inodes(ctl_table *table, int write,
108 void __user *buffer, size_t *lenp, loff_t *ppos)
110 inodes_stat.nr_inodes = get_nr_inodes();
111 inodes_stat.nr_unused = get_nr_inodes_unused();
112 return proc_dointvec(table, write, buffer, lenp, ppos);
117 * inode_init_always - perform inode structure intialisation
118 * @sb: superblock inode belongs to
119 * @inode: inode to initialise
121 * These are initializations that need to be done on every inode
122 * allocation as the fields are not initialised by slab allocation.
124 int inode_init_always(struct super_block *sb, struct inode *inode)
126 static const struct inode_operations empty_iops;
127 static const struct file_operations empty_fops;
128 struct address_space *const mapping = &inode->i_data;
131 inode->i_blkbits = sb->s_blocksize_bits;
133 atomic_set(&inode->i_count, 1);
134 inode->i_op = &empty_iops;
135 inode->i_fop = &empty_fops;
136 inode->__i_nlink = 1;
137 inode->i_opflags = 0;
138 i_uid_write(inode, 0);
139 i_gid_write(inode, 0);
140 atomic_set(&inode->i_writecount, 0);
144 inode->i_generation = 0;
146 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
148 inode->i_pipe = NULL;
149 inode->i_bdev = NULL;
150 inode->i_cdev = NULL;
152 inode->dirtied_when = 0;
154 if (security_inode_alloc(inode))
156 spin_lock_init(&inode->i_lock);
157 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
159 mutex_init(&inode->i_mutex);
160 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
162 atomic_set(&inode->i_dio_count, 0);
164 mapping->a_ops = &empty_aops;
165 mapping->host = inode;
167 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
168 mapping->assoc_mapping = NULL;
169 mapping->backing_dev_info = &default_backing_dev_info;
170 mapping->writeback_index = 0;
173 * If the block_device provides a backing_dev_info for client
174 * inodes then use that. Otherwise the inode share the bdev's
178 struct backing_dev_info *bdi;
180 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
181 mapping->backing_dev_info = bdi;
183 inode->i_private = NULL;
184 inode->i_mapping = mapping;
185 INIT_LIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
186 #ifdef CONFIG_FS_POSIX_ACL
187 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
190 #ifdef CONFIG_FSNOTIFY
191 inode->i_fsnotify_mask = 0;
194 this_cpu_inc(nr_inodes);
200 EXPORT_SYMBOL(inode_init_always);
202 static struct inode *alloc_inode(struct super_block *sb)
206 if (sb->s_op->alloc_inode)
207 inode = sb->s_op->alloc_inode(sb);
209 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
214 if (unlikely(inode_init_always(sb, inode))) {
215 if (inode->i_sb->s_op->destroy_inode)
216 inode->i_sb->s_op->destroy_inode(inode);
218 kmem_cache_free(inode_cachep, inode);
225 void free_inode_nonrcu(struct inode *inode)
227 kmem_cache_free(inode_cachep, inode);
229 EXPORT_SYMBOL(free_inode_nonrcu);
231 void __destroy_inode(struct inode *inode)
233 BUG_ON(inode_has_buffers(inode));
234 security_inode_free(inode);
235 fsnotify_inode_delete(inode);
236 if (!inode->i_nlink) {
237 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
238 atomic_long_dec(&inode->i_sb->s_remove_count);
241 #ifdef CONFIG_FS_POSIX_ACL
242 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
243 posix_acl_release(inode->i_acl);
244 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
245 posix_acl_release(inode->i_default_acl);
247 this_cpu_dec(nr_inodes);
249 EXPORT_SYMBOL(__destroy_inode);
251 static void i_callback(struct rcu_head *head)
253 struct inode *inode = container_of(head, struct inode, i_rcu);
254 kmem_cache_free(inode_cachep, inode);
257 static void destroy_inode(struct inode *inode)
259 BUG_ON(!list_empty(&inode->i_lru));
260 __destroy_inode(inode);
261 if (inode->i_sb->s_op->destroy_inode)
262 inode->i_sb->s_op->destroy_inode(inode);
264 call_rcu(&inode->i_rcu, i_callback);
268 * drop_nlink - directly drop an inode's link count
271 * This is a low-level filesystem helper to replace any
272 * direct filesystem manipulation of i_nlink. In cases
273 * where we are attempting to track writes to the
274 * filesystem, a decrement to zero means an imminent
275 * write when the file is truncated and actually unlinked
278 void drop_nlink(struct inode *inode)
280 WARN_ON(inode->i_nlink == 0);
283 atomic_long_inc(&inode->i_sb->s_remove_count);
285 EXPORT_SYMBOL(drop_nlink);
288 * clear_nlink - directly zero an inode's link count
291 * This is a low-level filesystem helper to replace any
292 * direct filesystem manipulation of i_nlink. See
293 * drop_nlink() for why we care about i_nlink hitting zero.
295 void clear_nlink(struct inode *inode)
297 if (inode->i_nlink) {
298 inode->__i_nlink = 0;
299 atomic_long_inc(&inode->i_sb->s_remove_count);
302 EXPORT_SYMBOL(clear_nlink);
305 * set_nlink - directly set an inode's link count
307 * @nlink: new nlink (should be non-zero)
309 * This is a low-level filesystem helper to replace any
310 * direct filesystem manipulation of i_nlink.
312 void set_nlink(struct inode *inode, unsigned int nlink)
317 /* Yes, some filesystems do change nlink from zero to one */
318 if (inode->i_nlink == 0)
319 atomic_long_dec(&inode->i_sb->s_remove_count);
321 inode->__i_nlink = nlink;
324 EXPORT_SYMBOL(set_nlink);
327 * inc_nlink - directly increment an inode's link count
330 * This is a low-level filesystem helper to replace any
331 * direct filesystem manipulation of i_nlink. Currently,
332 * it is only here for parity with dec_nlink().
334 void inc_nlink(struct inode *inode)
336 if (WARN_ON(inode->i_nlink == 0))
337 atomic_long_dec(&inode->i_sb->s_remove_count);
341 EXPORT_SYMBOL(inc_nlink);
343 void address_space_init_once(struct address_space *mapping)
345 memset(mapping, 0, sizeof(*mapping));
346 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
347 spin_lock_init(&mapping->tree_lock);
348 mutex_init(&mapping->i_mmap_mutex);
349 INIT_LIST_HEAD(&mapping->private_list);
350 spin_lock_init(&mapping->private_lock);
351 INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
352 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
354 EXPORT_SYMBOL(address_space_init_once);
357 * These are initializations that only need to be done
358 * once, because the fields are idempotent across use
359 * of the inode, so let the slab aware of that.
361 void inode_init_once(struct inode *inode)
363 memset(inode, 0, sizeof(*inode));
364 INIT_HLIST_NODE(&inode->i_hash);
365 INIT_LIST_HEAD(&inode->i_devices);
366 INIT_LIST_HEAD(&inode->i_wb_list);
367 INIT_LIST_HEAD(&inode->i_lru);
368 address_space_init_once(&inode->i_data);
369 i_size_ordered_init(inode);
370 #ifdef CONFIG_FSNOTIFY
371 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
374 EXPORT_SYMBOL(inode_init_once);
376 static void init_once(void *foo)
378 struct inode *inode = (struct inode *) foo;
380 inode_init_once(inode);
384 * inode->i_lock must be held
386 void __iget(struct inode *inode)
388 atomic_inc(&inode->i_count);
392 * get additional reference to inode; caller must already hold one.
394 void ihold(struct inode *inode)
396 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
398 EXPORT_SYMBOL(ihold);
400 static void inode_lru_list_add(struct inode *inode)
402 spin_lock(&inode->i_sb->s_inode_lru_lock);
403 if (list_empty(&inode->i_lru)) {
404 list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
405 inode->i_sb->s_nr_inodes_unused++;
406 this_cpu_inc(nr_unused);
408 spin_unlock(&inode->i_sb->s_inode_lru_lock);
411 static void inode_lru_list_del(struct inode *inode)
413 spin_lock(&inode->i_sb->s_inode_lru_lock);
414 if (!list_empty(&inode->i_lru)) {
415 list_del_init(&inode->i_lru);
416 inode->i_sb->s_nr_inodes_unused--;
417 this_cpu_dec(nr_unused);
419 spin_unlock(&inode->i_sb->s_inode_lru_lock);
423 * inode_sb_list_add - add inode to the superblock list of inodes
424 * @inode: inode to add
426 void inode_sb_list_add(struct inode *inode)
428 spin_lock(&inode_sb_list_lock);
429 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
430 spin_unlock(&inode_sb_list_lock);
432 EXPORT_SYMBOL_GPL(inode_sb_list_add);
434 static inline void inode_sb_list_del(struct inode *inode)
436 if (!list_empty(&inode->i_sb_list)) {
437 spin_lock(&inode_sb_list_lock);
438 list_del_init(&inode->i_sb_list);
439 spin_unlock(&inode_sb_list_lock);
443 static unsigned long hash(struct super_block *sb, unsigned long hashval)
447 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
449 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
450 return tmp & i_hash_mask;
454 * __insert_inode_hash - hash an inode
455 * @inode: unhashed inode
456 * @hashval: unsigned long value used to locate this object in the
459 * Add an inode to the inode hash for this superblock.
461 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
463 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
465 spin_lock(&inode_hash_lock);
466 spin_lock(&inode->i_lock);
467 hlist_add_head(&inode->i_hash, b);
468 spin_unlock(&inode->i_lock);
469 spin_unlock(&inode_hash_lock);
471 EXPORT_SYMBOL(__insert_inode_hash);
474 * __remove_inode_hash - remove an inode from the hash
475 * @inode: inode to unhash
477 * Remove an inode from the superblock.
479 void __remove_inode_hash(struct inode *inode)
481 spin_lock(&inode_hash_lock);
482 spin_lock(&inode->i_lock);
483 hlist_del_init(&inode->i_hash);
484 spin_unlock(&inode->i_lock);
485 spin_unlock(&inode_hash_lock);
487 EXPORT_SYMBOL(__remove_inode_hash);
489 void clear_inode(struct inode *inode)
493 * We have to cycle tree_lock here because reclaim can be still in the
494 * process of removing the last page (in __delete_from_page_cache())
495 * and we must not free mapping under it.
497 spin_lock_irq(&inode->i_data.tree_lock);
498 BUG_ON(inode->i_data.nrpages);
499 spin_unlock_irq(&inode->i_data.tree_lock);
500 BUG_ON(!list_empty(&inode->i_data.private_list));
501 BUG_ON(!(inode->i_state & I_FREEING));
502 BUG_ON(inode->i_state & I_CLEAR);
503 /* don't need i_lock here, no concurrent mods to i_state */
504 inode->i_state = I_FREEING | I_CLEAR;
506 EXPORT_SYMBOL(clear_inode);
509 * Free the inode passed in, removing it from the lists it is still connected
510 * to. We remove any pages still attached to the inode and wait for any IO that
511 * is still in progress before finally destroying the inode.
513 * An inode must already be marked I_FREEING so that we avoid the inode being
514 * moved back onto lists if we race with other code that manipulates the lists
515 * (e.g. writeback_single_inode). The caller is responsible for setting this.
517 * An inode must already be removed from the LRU list before being evicted from
518 * the cache. This should occur atomically with setting the I_FREEING state
519 * flag, so no inodes here should ever be on the LRU when being evicted.
521 static void evict(struct inode *inode)
523 const struct super_operations *op = inode->i_sb->s_op;
525 BUG_ON(!(inode->i_state & I_FREEING));
526 BUG_ON(!list_empty(&inode->i_lru));
528 if (!list_empty(&inode->i_wb_list))
529 inode_wb_list_del(inode);
531 inode_sb_list_del(inode);
534 * Wait for flusher thread to be done with the inode so that filesystem
535 * does not start destroying it while writeback is still running. Since
536 * the inode has I_FREEING set, flusher thread won't start new work on
537 * the inode. We just have to wait for running writeback to finish.
539 inode_wait_for_writeback(inode);
541 if (op->evict_inode) {
542 op->evict_inode(inode);
544 if (inode->i_data.nrpages)
545 truncate_inode_pages(&inode->i_data, 0);
548 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
550 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
553 remove_inode_hash(inode);
555 spin_lock(&inode->i_lock);
556 wake_up_bit(&inode->i_state, __I_NEW);
557 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
558 spin_unlock(&inode->i_lock);
560 destroy_inode(inode);
564 * dispose_list - dispose of the contents of a local list
565 * @head: the head of the list to free
567 * Dispose-list gets a local list with local inodes in it, so it doesn't
568 * need to worry about list corruption and SMP locks.
570 static void dispose_list(struct list_head *head)
572 while (!list_empty(head)) {
575 inode = list_first_entry(head, struct inode, i_lru);
576 list_del_init(&inode->i_lru);
583 * evict_inodes - evict all evictable inodes for a superblock
584 * @sb: superblock to operate on
586 * Make sure that no inodes with zero refcount are retained. This is
587 * called by superblock shutdown after having MS_ACTIVE flag removed,
588 * so any inode reaching zero refcount during or after that call will
589 * be immediately evicted.
591 void evict_inodes(struct super_block *sb)
593 struct inode *inode, *next;
596 spin_lock(&inode_sb_list_lock);
597 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
598 if (atomic_read(&inode->i_count))
601 spin_lock(&inode->i_lock);
602 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
603 spin_unlock(&inode->i_lock);
607 inode->i_state |= I_FREEING;
608 inode_lru_list_del(inode);
609 spin_unlock(&inode->i_lock);
610 list_add(&inode->i_lru, &dispose);
612 spin_unlock(&inode_sb_list_lock);
614 dispose_list(&dispose);
618 * invalidate_inodes - attempt to free all inodes on a superblock
619 * @sb: superblock to operate on
620 * @kill_dirty: flag to guide handling of dirty inodes
622 * Attempts to free all inodes for a given superblock. If there were any
623 * busy inodes return a non-zero value, else zero.
624 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
627 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
630 struct inode *inode, *next;
633 spin_lock(&inode_sb_list_lock);
634 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
635 spin_lock(&inode->i_lock);
636 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
637 spin_unlock(&inode->i_lock);
640 if (inode->i_state & I_DIRTY && !kill_dirty) {
641 spin_unlock(&inode->i_lock);
645 if (atomic_read(&inode->i_count)) {
646 spin_unlock(&inode->i_lock);
651 inode->i_state |= I_FREEING;
652 inode_lru_list_del(inode);
653 spin_unlock(&inode->i_lock);
654 list_add(&inode->i_lru, &dispose);
656 spin_unlock(&inode_sb_list_lock);
658 dispose_list(&dispose);
663 static int can_unuse(struct inode *inode)
665 if (inode->i_state & ~I_REFERENCED)
667 if (inode_has_buffers(inode))
669 if (atomic_read(&inode->i_count))
671 if (inode->i_data.nrpages)
677 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
678 * This is called from the superblock shrinker function with a number of inodes
679 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
680 * then are freed outside inode_lock by dispose_list().
682 * Any inodes which are pinned purely because of attached pagecache have their
683 * pagecache removed. If the inode has metadata buffers attached to
684 * mapping->private_list then try to remove them.
686 * If the inode has the I_REFERENCED flag set, then it means that it has been
687 * used recently - the flag is set in iput_final(). When we encounter such an
688 * inode, clear the flag and move it to the back of the LRU so it gets another
689 * pass through the LRU before it gets reclaimed. This is necessary because of
690 * the fact we are doing lazy LRU updates to minimise lock contention so the
691 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
692 * with this flag set because they are the inodes that are out of order.
694 void prune_icache_sb(struct super_block *sb, int nr_to_scan)
698 unsigned long reap = 0;
700 spin_lock(&sb->s_inode_lru_lock);
701 for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
704 if (list_empty(&sb->s_inode_lru))
707 inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
710 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
711 * so use a trylock. If we fail to get the lock, just move the
712 * inode to the back of the list so we don't spin on it.
714 if (!spin_trylock(&inode->i_lock)) {
715 list_move_tail(&inode->i_lru, &sb->s_inode_lru);
720 * Referenced or dirty inodes are still in use. Give them
721 * another pass through the LRU as we canot reclaim them now.
723 if (atomic_read(&inode->i_count) ||
724 (inode->i_state & ~I_REFERENCED)) {
725 list_del_init(&inode->i_lru);
726 spin_unlock(&inode->i_lock);
727 sb->s_nr_inodes_unused--;
728 this_cpu_dec(nr_unused);
732 /* recently referenced inodes get one more pass */
733 if (inode->i_state & I_REFERENCED) {
734 inode->i_state &= ~I_REFERENCED;
735 list_move(&inode->i_lru, &sb->s_inode_lru);
736 spin_unlock(&inode->i_lock);
739 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
741 spin_unlock(&inode->i_lock);
742 spin_unlock(&sb->s_inode_lru_lock);
743 if (remove_inode_buffers(inode))
744 reap += invalidate_mapping_pages(&inode->i_data,
747 spin_lock(&sb->s_inode_lru_lock);
749 if (inode != list_entry(sb->s_inode_lru.next,
750 struct inode, i_lru))
751 continue; /* wrong inode or list_empty */
752 /* avoid lock inversions with trylock */
753 if (!spin_trylock(&inode->i_lock))
755 if (!can_unuse(inode)) {
756 spin_unlock(&inode->i_lock);
760 WARN_ON(inode->i_state & I_NEW);
761 inode->i_state |= I_FREEING;
762 spin_unlock(&inode->i_lock);
764 list_move(&inode->i_lru, &freeable);
765 sb->s_nr_inodes_unused--;
766 this_cpu_dec(nr_unused);
768 if (current_is_kswapd())
769 __count_vm_events(KSWAPD_INODESTEAL, reap);
771 __count_vm_events(PGINODESTEAL, reap);
772 spin_unlock(&sb->s_inode_lru_lock);
773 if (current->reclaim_state)
774 current->reclaim_state->reclaimed_slab += reap;
776 dispose_list(&freeable);
779 static void __wait_on_freeing_inode(struct inode *inode);
781 * Called with the inode lock held.
783 static struct inode *find_inode(struct super_block *sb,
784 struct hlist_head *head,
785 int (*test)(struct inode *, void *),
788 struct hlist_node *node;
789 struct inode *inode = NULL;
792 hlist_for_each_entry(inode, node, head, i_hash) {
793 spin_lock(&inode->i_lock);
794 if (inode->i_sb != sb) {
795 spin_unlock(&inode->i_lock);
798 if (!test(inode, data)) {
799 spin_unlock(&inode->i_lock);
802 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
803 __wait_on_freeing_inode(inode);
807 spin_unlock(&inode->i_lock);
814 * find_inode_fast is the fast path version of find_inode, see the comment at
815 * iget_locked for details.
817 static struct inode *find_inode_fast(struct super_block *sb,
818 struct hlist_head *head, unsigned long ino)
820 struct hlist_node *node;
821 struct inode *inode = NULL;
824 hlist_for_each_entry(inode, node, head, i_hash) {
825 spin_lock(&inode->i_lock);
826 if (inode->i_ino != ino) {
827 spin_unlock(&inode->i_lock);
830 if (inode->i_sb != sb) {
831 spin_unlock(&inode->i_lock);
834 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
835 __wait_on_freeing_inode(inode);
839 spin_unlock(&inode->i_lock);
846 * Each cpu owns a range of LAST_INO_BATCH numbers.
847 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
848 * to renew the exhausted range.
850 * This does not significantly increase overflow rate because every CPU can
851 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
852 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
853 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
854 * overflow rate by 2x, which does not seem too significant.
856 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
857 * error if st_ino won't fit in target struct field. Use 32bit counter
858 * here to attempt to avoid that.
860 #define LAST_INO_BATCH 1024
861 static DEFINE_PER_CPU(unsigned int, last_ino);
863 unsigned int get_next_ino(void)
865 unsigned int *p = &get_cpu_var(last_ino);
866 unsigned int res = *p;
869 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
870 static atomic_t shared_last_ino;
871 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
873 res = next - LAST_INO_BATCH;
878 put_cpu_var(last_ino);
881 EXPORT_SYMBOL(get_next_ino);
884 * new_inode_pseudo - obtain an inode
887 * Allocates a new inode for given superblock.
888 * Inode wont be chained in superblock s_inodes list
890 * - fs can't be unmount
891 * - quotas, fsnotify, writeback can't work
893 struct inode *new_inode_pseudo(struct super_block *sb)
895 struct inode *inode = alloc_inode(sb);
898 spin_lock(&inode->i_lock);
900 spin_unlock(&inode->i_lock);
901 INIT_LIST_HEAD(&inode->i_sb_list);
907 * new_inode - obtain an inode
910 * Allocates a new inode for given superblock. The default gfp_mask
911 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
912 * If HIGHMEM pages are unsuitable or it is known that pages allocated
913 * for the page cache are not reclaimable or migratable,
914 * mapping_set_gfp_mask() must be called with suitable flags on the
915 * newly created inode's mapping
918 struct inode *new_inode(struct super_block *sb)
922 spin_lock_prefetch(&inode_sb_list_lock);
924 inode = new_inode_pseudo(sb);
926 inode_sb_list_add(inode);
929 EXPORT_SYMBOL(new_inode);
931 #ifdef CONFIG_DEBUG_LOCK_ALLOC
932 void lockdep_annotate_inode_mutex_key(struct inode *inode)
934 if (S_ISDIR(inode->i_mode)) {
935 struct file_system_type *type = inode->i_sb->s_type;
937 /* Set new key only if filesystem hasn't already changed it */
938 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
940 * ensure nobody is actually holding i_mutex
942 mutex_destroy(&inode->i_mutex);
943 mutex_init(&inode->i_mutex);
944 lockdep_set_class(&inode->i_mutex,
945 &type->i_mutex_dir_key);
949 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
953 * unlock_new_inode - clear the I_NEW state and wake up any waiters
954 * @inode: new inode to unlock
956 * Called when the inode is fully initialised to clear the new state of the
957 * inode and wake up anyone waiting for the inode to finish initialisation.
959 void unlock_new_inode(struct inode *inode)
961 lockdep_annotate_inode_mutex_key(inode);
962 spin_lock(&inode->i_lock);
963 WARN_ON(!(inode->i_state & I_NEW));
964 inode->i_state &= ~I_NEW;
966 wake_up_bit(&inode->i_state, __I_NEW);
967 spin_unlock(&inode->i_lock);
969 EXPORT_SYMBOL(unlock_new_inode);
972 * iget5_locked - obtain an inode from a mounted file system
973 * @sb: super block of file system
974 * @hashval: hash value (usually inode number) to get
975 * @test: callback used for comparisons between inodes
976 * @set: callback used to initialize a new struct inode
977 * @data: opaque data pointer to pass to @test and @set
979 * Search for the inode specified by @hashval and @data in the inode cache,
980 * and if present it is return it with an increased reference count. This is
981 * a generalized version of iget_locked() for file systems where the inode
982 * number is not sufficient for unique identification of an inode.
984 * If the inode is not in cache, allocate a new inode and return it locked,
985 * hashed, and with the I_NEW flag set. The file system gets to fill it in
986 * before unlocking it via unlock_new_inode().
988 * Note both @test and @set are called with the inode_hash_lock held, so can't
991 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
992 int (*test)(struct inode *, void *),
993 int (*set)(struct inode *, void *), void *data)
995 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
998 spin_lock(&inode_hash_lock);
999 inode = find_inode(sb, head, test, data);
1000 spin_unlock(&inode_hash_lock);
1003 wait_on_inode(inode);
1007 inode = alloc_inode(sb);
1011 spin_lock(&inode_hash_lock);
1012 /* We released the lock, so.. */
1013 old = find_inode(sb, head, test, data);
1015 if (set(inode, data))
1018 spin_lock(&inode->i_lock);
1019 inode->i_state = I_NEW;
1020 hlist_add_head(&inode->i_hash, head);
1021 spin_unlock(&inode->i_lock);
1022 inode_sb_list_add(inode);
1023 spin_unlock(&inode_hash_lock);
1025 /* Return the locked inode with I_NEW set, the
1026 * caller is responsible for filling in the contents
1032 * Uhhuh, somebody else created the same inode under
1033 * us. Use the old inode instead of the one we just
1036 spin_unlock(&inode_hash_lock);
1037 destroy_inode(inode);
1039 wait_on_inode(inode);
1044 spin_unlock(&inode_hash_lock);
1045 destroy_inode(inode);
1048 EXPORT_SYMBOL(iget5_locked);
1051 * iget_locked - obtain an inode from a mounted file system
1052 * @sb: super block of file system
1053 * @ino: inode number to get
1055 * Search for the inode specified by @ino in the inode cache and if present
1056 * return it with an increased reference count. This is for file systems
1057 * where the inode number is sufficient for unique identification of an inode.
1059 * If the inode is not in cache, allocate a new inode and return it locked,
1060 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1061 * before unlocking it via unlock_new_inode().
1063 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1065 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1066 struct inode *inode;
1068 spin_lock(&inode_hash_lock);
1069 inode = find_inode_fast(sb, head, ino);
1070 spin_unlock(&inode_hash_lock);
1072 wait_on_inode(inode);
1076 inode = alloc_inode(sb);
1080 spin_lock(&inode_hash_lock);
1081 /* We released the lock, so.. */
1082 old = find_inode_fast(sb, head, ino);
1085 spin_lock(&inode->i_lock);
1086 inode->i_state = I_NEW;
1087 hlist_add_head(&inode->i_hash, head);
1088 spin_unlock(&inode->i_lock);
1089 inode_sb_list_add(inode);
1090 spin_unlock(&inode_hash_lock);
1092 /* Return the locked inode with I_NEW set, the
1093 * caller is responsible for filling in the contents
1099 * Uhhuh, somebody else created the same inode under
1100 * us. Use the old inode instead of the one we just
1103 spin_unlock(&inode_hash_lock);
1104 destroy_inode(inode);
1106 wait_on_inode(inode);
1110 EXPORT_SYMBOL(iget_locked);
1113 * search the inode cache for a matching inode number.
1114 * If we find one, then the inode number we are trying to
1115 * allocate is not unique and so we should not use it.
1117 * Returns 1 if the inode number is unique, 0 if it is not.
1119 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1121 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1122 struct hlist_node *node;
1123 struct inode *inode;
1125 spin_lock(&inode_hash_lock);
1126 hlist_for_each_entry(inode, node, b, i_hash) {
1127 if (inode->i_ino == ino && inode->i_sb == sb) {
1128 spin_unlock(&inode_hash_lock);
1132 spin_unlock(&inode_hash_lock);
1138 * iunique - get a unique inode number
1140 * @max_reserved: highest reserved inode number
1142 * Obtain an inode number that is unique on the system for a given
1143 * superblock. This is used by file systems that have no natural
1144 * permanent inode numbering system. An inode number is returned that
1145 * is higher than the reserved limit but unique.
1148 * With a large number of inodes live on the file system this function
1149 * currently becomes quite slow.
1151 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1154 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1155 * error if st_ino won't fit in target struct field. Use 32bit counter
1156 * here to attempt to avoid that.
1158 static DEFINE_SPINLOCK(iunique_lock);
1159 static unsigned int counter;
1162 spin_lock(&iunique_lock);
1164 if (counter <= max_reserved)
1165 counter = max_reserved + 1;
1167 } while (!test_inode_iunique(sb, res));
1168 spin_unlock(&iunique_lock);
1172 EXPORT_SYMBOL(iunique);
1174 struct inode *igrab(struct inode *inode)
1176 spin_lock(&inode->i_lock);
1177 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1179 spin_unlock(&inode->i_lock);
1181 spin_unlock(&inode->i_lock);
1183 * Handle the case where s_op->clear_inode is not been
1184 * called yet, and somebody is calling igrab
1185 * while the inode is getting freed.
1191 EXPORT_SYMBOL(igrab);
1194 * ilookup5_nowait - search for an inode in the inode cache
1195 * @sb: super block of file system to search
1196 * @hashval: hash value (usually inode number) to search for
1197 * @test: callback used for comparisons between inodes
1198 * @data: opaque data pointer to pass to @test
1200 * Search for the inode specified by @hashval and @data in the inode cache.
1201 * If the inode is in the cache, the inode is returned with an incremented
1204 * Note: I_NEW is not waited upon so you have to be very careful what you do
1205 * with the returned inode. You probably should be using ilookup5() instead.
1207 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1209 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1210 int (*test)(struct inode *, void *), void *data)
1212 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1213 struct inode *inode;
1215 spin_lock(&inode_hash_lock);
1216 inode = find_inode(sb, head, test, data);
1217 spin_unlock(&inode_hash_lock);
1221 EXPORT_SYMBOL(ilookup5_nowait);
1224 * ilookup5 - search for an inode in the inode cache
1225 * @sb: super block of file system to search
1226 * @hashval: hash value (usually inode number) to search for
1227 * @test: callback used for comparisons between inodes
1228 * @data: opaque data pointer to pass to @test
1230 * Search for the inode specified by @hashval and @data in the inode cache,
1231 * and if the inode is in the cache, return the inode with an incremented
1232 * reference count. Waits on I_NEW before returning the inode.
1233 * returned with an incremented reference count.
1235 * This is a generalized version of ilookup() for file systems where the
1236 * inode number is not sufficient for unique identification of an inode.
1238 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1240 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1241 int (*test)(struct inode *, void *), void *data)
1243 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1246 wait_on_inode(inode);
1249 EXPORT_SYMBOL(ilookup5);
1252 * ilookup - search for an inode in the inode cache
1253 * @sb: super block of file system to search
1254 * @ino: inode number to search for
1256 * Search for the inode @ino in the inode cache, and if the inode is in the
1257 * cache, the inode is returned with an incremented reference count.
1259 struct inode *ilookup(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);
1269 wait_on_inode(inode);
1272 EXPORT_SYMBOL(ilookup);
1274 int insert_inode_locked(struct inode *inode)
1276 struct super_block *sb = inode->i_sb;
1277 ino_t ino = inode->i_ino;
1278 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1281 struct hlist_node *node;
1282 struct inode *old = NULL;
1283 spin_lock(&inode_hash_lock);
1284 hlist_for_each_entry(old, node, head, i_hash) {
1285 if (old->i_ino != ino)
1287 if (old->i_sb != sb)
1289 spin_lock(&old->i_lock);
1290 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1291 spin_unlock(&old->i_lock);
1296 if (likely(!node)) {
1297 spin_lock(&inode->i_lock);
1298 inode->i_state |= I_NEW;
1299 hlist_add_head(&inode->i_hash, head);
1300 spin_unlock(&inode->i_lock);
1301 spin_unlock(&inode_hash_lock);
1305 spin_unlock(&old->i_lock);
1306 spin_unlock(&inode_hash_lock);
1308 if (unlikely(!inode_unhashed(old))) {
1315 EXPORT_SYMBOL(insert_inode_locked);
1317 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1318 int (*test)(struct inode *, void *), void *data)
1320 struct super_block *sb = inode->i_sb;
1321 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1324 struct hlist_node *node;
1325 struct inode *old = NULL;
1327 spin_lock(&inode_hash_lock);
1328 hlist_for_each_entry(old, node, head, i_hash) {
1329 if (old->i_sb != sb)
1331 if (!test(old, data))
1333 spin_lock(&old->i_lock);
1334 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1335 spin_unlock(&old->i_lock);
1340 if (likely(!node)) {
1341 spin_lock(&inode->i_lock);
1342 inode->i_state |= I_NEW;
1343 hlist_add_head(&inode->i_hash, head);
1344 spin_unlock(&inode->i_lock);
1345 spin_unlock(&inode_hash_lock);
1349 spin_unlock(&old->i_lock);
1350 spin_unlock(&inode_hash_lock);
1352 if (unlikely(!inode_unhashed(old))) {
1359 EXPORT_SYMBOL(insert_inode_locked4);
1362 int generic_delete_inode(struct inode *inode)
1366 EXPORT_SYMBOL(generic_delete_inode);
1369 * Called when we're dropping the last reference
1372 * Call the FS "drop_inode()" function, defaulting to
1373 * the legacy UNIX filesystem behaviour. If it tells
1374 * us to evict inode, do so. Otherwise, retain inode
1375 * in cache if fs is alive, sync and evict if fs is
1378 static void iput_final(struct inode *inode)
1380 struct super_block *sb = inode->i_sb;
1381 const struct super_operations *op = inode->i_sb->s_op;
1384 WARN_ON(inode->i_state & I_NEW);
1387 drop = op->drop_inode(inode);
1389 drop = generic_drop_inode(inode);
1391 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1392 inode->i_state |= I_REFERENCED;
1393 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1394 inode_lru_list_add(inode);
1395 spin_unlock(&inode->i_lock);
1400 inode->i_state |= I_WILL_FREE;
1401 spin_unlock(&inode->i_lock);
1402 write_inode_now(inode, 1);
1403 spin_lock(&inode->i_lock);
1404 WARN_ON(inode->i_state & I_NEW);
1405 inode->i_state &= ~I_WILL_FREE;
1408 inode->i_state |= I_FREEING;
1409 if (!list_empty(&inode->i_lru))
1410 inode_lru_list_del(inode);
1411 spin_unlock(&inode->i_lock);
1417 * iput - put an inode
1418 * @inode: inode to put
1420 * Puts an inode, dropping its usage count. If the inode use count hits
1421 * zero, the inode is then freed and may also be destroyed.
1423 * Consequently, iput() can sleep.
1425 void iput(struct inode *inode)
1428 BUG_ON(inode->i_state & I_CLEAR);
1430 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1434 EXPORT_SYMBOL(iput);
1437 * bmap - find a block number in a file
1438 * @inode: inode of file
1439 * @block: block to find
1441 * Returns the block number on the device holding the inode that
1442 * is the disk block number for the block of the file requested.
1443 * That is, asked for block 4 of inode 1 the function will return the
1444 * disk block relative to the disk start that holds that block of the
1447 sector_t bmap(struct inode *inode, sector_t block)
1450 if (inode->i_mapping->a_ops->bmap)
1451 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1454 EXPORT_SYMBOL(bmap);
1457 * With relative atime, only update atime if the previous atime is
1458 * earlier than either the ctime or mtime or if at least a day has
1459 * passed since the last atime update.
1461 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1462 struct timespec now)
1465 if (!(mnt->mnt_flags & MNT_RELATIME))
1468 * Is mtime younger than atime? If yes, update atime:
1470 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1473 * Is ctime younger than atime? If yes, update atime:
1475 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1479 * Is the previous atime value older than a day? If yes,
1482 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1485 * Good, we can skip the atime update:
1491 * This does the actual work of updating an inodes time or version. Must have
1492 * had called mnt_want_write() before calling this.
1494 static int update_time(struct inode *inode, struct timespec *time, int flags)
1496 if (inode->i_op->update_time)
1497 return inode->i_op->update_time(inode, time, flags);
1499 if (flags & S_ATIME)
1500 inode->i_atime = *time;
1501 if (flags & S_VERSION)
1502 inode_inc_iversion(inode);
1503 if (flags & S_CTIME)
1504 inode->i_ctime = *time;
1505 if (flags & S_MTIME)
1506 inode->i_mtime = *time;
1507 mark_inode_dirty_sync(inode);
1512 * touch_atime - update the access time
1513 * @path: the &struct path to update
1515 * Update the accessed time on an inode and mark it for writeback.
1516 * This function automatically handles read only file systems and media,
1517 * as well as the "noatime" flag and inode specific "noatime" markers.
1519 void touch_atime(struct path *path)
1521 struct vfsmount *mnt = path->mnt;
1522 struct inode *inode = path->dentry->d_inode;
1523 struct timespec now;
1525 if (inode->i_flags & S_NOATIME)
1527 if (IS_NOATIME(inode))
1529 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1532 if (mnt->mnt_flags & MNT_NOATIME)
1534 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1537 now = current_fs_time(inode->i_sb);
1539 if (!relatime_need_update(mnt, inode, now))
1542 if (timespec_equal(&inode->i_atime, &now))
1545 if (mnt_want_write(mnt))
1549 * File systems can error out when updating inodes if they need to
1550 * allocate new space to modify an inode (such is the case for
1551 * Btrfs), but since we touch atime while walking down the path we
1552 * really don't care if we failed to update the atime of the file,
1553 * so just ignore the return value.
1555 update_time(inode, &now, S_ATIME);
1556 mnt_drop_write(mnt);
1558 EXPORT_SYMBOL(touch_atime);
1561 * The logic we want is
1563 * if suid or (sgid and xgrp)
1566 int should_remove_suid(struct dentry *dentry)
1568 umode_t mode = dentry->d_inode->i_mode;
1571 /* suid always must be killed */
1572 if (unlikely(mode & S_ISUID))
1573 kill = ATTR_KILL_SUID;
1576 * sgid without any exec bits is just a mandatory locking mark; leave
1577 * it alone. If some exec bits are set, it's a real sgid; kill it.
1579 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1580 kill |= ATTR_KILL_SGID;
1582 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1587 EXPORT_SYMBOL(should_remove_suid);
1589 static int __remove_suid(struct dentry *dentry, int kill)
1591 struct iattr newattrs;
1593 newattrs.ia_valid = ATTR_FORCE | kill;
1594 return notify_change(dentry, &newattrs);
1597 int file_remove_suid(struct file *file)
1599 struct dentry *dentry = file->f_path.dentry;
1600 struct inode *inode = dentry->d_inode;
1605 /* Fast path for nothing security related */
1606 if (IS_NOSEC(inode))
1609 killsuid = should_remove_suid(dentry);
1610 killpriv = security_inode_need_killpriv(dentry);
1615 error = security_inode_killpriv(dentry);
1616 if (!error && killsuid)
1617 error = __remove_suid(dentry, killsuid);
1618 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1619 inode->i_flags |= S_NOSEC;
1623 EXPORT_SYMBOL(file_remove_suid);
1626 * file_update_time - update mtime and ctime time
1627 * @file: file accessed
1629 * Update the mtime and ctime members of an inode and mark the inode
1630 * for writeback. Note that this function is meant exclusively for
1631 * usage in the file write path of filesystems, and filesystems may
1632 * choose to explicitly ignore update via this function with the
1633 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1634 * timestamps are handled by the server. This can return an error for
1635 * file systems who need to allocate space in order to update an inode.
1638 int file_update_time(struct file *file)
1640 struct inode *inode = file->f_path.dentry->d_inode;
1641 struct timespec now;
1645 /* First try to exhaust all avenues to not sync */
1646 if (IS_NOCMTIME(inode))
1649 now = current_fs_time(inode->i_sb);
1650 if (!timespec_equal(&inode->i_mtime, &now))
1653 if (!timespec_equal(&inode->i_ctime, &now))
1656 if (IS_I_VERSION(inode))
1657 sync_it |= S_VERSION;
1662 /* Finally allowed to write? Takes lock. */
1663 if (mnt_want_write_file(file))
1666 ret = update_time(inode, &now, sync_it);
1667 mnt_drop_write_file(file);
1671 EXPORT_SYMBOL(file_update_time);
1673 int inode_needs_sync(struct inode *inode)
1677 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1681 EXPORT_SYMBOL(inode_needs_sync);
1683 int inode_wait(void *word)
1688 EXPORT_SYMBOL(inode_wait);
1691 * If we try to find an inode in the inode hash while it is being
1692 * deleted, we have to wait until the filesystem completes its
1693 * deletion before reporting that it isn't found. This function waits
1694 * until the deletion _might_ have completed. Callers are responsible
1695 * to recheck inode state.
1697 * It doesn't matter if I_NEW is not set initially, a call to
1698 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1701 static void __wait_on_freeing_inode(struct inode *inode)
1703 wait_queue_head_t *wq;
1704 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1705 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1706 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1707 spin_unlock(&inode->i_lock);
1708 spin_unlock(&inode_hash_lock);
1710 finish_wait(wq, &wait.wait);
1711 spin_lock(&inode_hash_lock);
1714 static __initdata unsigned long ihash_entries;
1715 static int __init set_ihash_entries(char *str)
1719 ihash_entries = simple_strtoul(str, &str, 0);
1722 __setup("ihash_entries=", set_ihash_entries);
1725 * Initialize the waitqueues and inode hash table.
1727 void __init inode_init_early(void)
1731 /* If hashes are distributed across NUMA nodes, defer
1732 * hash allocation until vmalloc space is available.
1738 alloc_large_system_hash("Inode-cache",
1739 sizeof(struct hlist_head),
1748 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1749 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1752 void __init inode_init(void)
1756 /* inode slab cache */
1757 inode_cachep = kmem_cache_create("inode_cache",
1758 sizeof(struct inode),
1760 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1764 /* Hash may have been set up in inode_init_early */
1769 alloc_large_system_hash("Inode-cache",
1770 sizeof(struct hlist_head),
1779 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1780 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1783 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1785 inode->i_mode = mode;
1786 if (S_ISCHR(mode)) {
1787 inode->i_fop = &def_chr_fops;
1788 inode->i_rdev = rdev;
1789 } else if (S_ISBLK(mode)) {
1790 inode->i_fop = &def_blk_fops;
1791 inode->i_rdev = rdev;
1792 } else if (S_ISFIFO(mode))
1793 inode->i_fop = &def_fifo_fops;
1794 else if (S_ISSOCK(mode))
1795 inode->i_fop = &bad_sock_fops;
1797 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1798 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1801 EXPORT_SYMBOL(init_special_inode);
1804 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1806 * @dir: Directory inode
1807 * @mode: mode of the new inode
1809 void inode_init_owner(struct inode *inode, const struct inode *dir,
1812 inode->i_uid = current_fsuid();
1813 if (dir && dir->i_mode & S_ISGID) {
1814 inode->i_gid = dir->i_gid;
1818 inode->i_gid = current_fsgid();
1819 inode->i_mode = mode;
1821 EXPORT_SYMBOL(inode_init_owner);
1824 * inode_owner_or_capable - check current task permissions to inode
1825 * @inode: inode being checked
1827 * Return true if current either has CAP_FOWNER to the inode, or
1830 bool inode_owner_or_capable(const struct inode *inode)
1832 if (uid_eq(current_fsuid(), inode->i_uid))
1834 if (inode_capable(inode, CAP_FOWNER))
1838 EXPORT_SYMBOL(inode_owner_or_capable);
1841 * Direct i/o helper functions
1843 static void __inode_dio_wait(struct inode *inode)
1845 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1846 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1849 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1850 if (atomic_read(&inode->i_dio_count))
1852 } while (atomic_read(&inode->i_dio_count));
1853 finish_wait(wq, &q.wait);
1857 * inode_dio_wait - wait for outstanding DIO requests to finish
1858 * @inode: inode to wait for
1860 * Waits for all pending direct I/O requests to finish so that we can
1861 * proceed with a truncate or equivalent operation.
1863 * Must be called under a lock that serializes taking new references
1864 * to i_dio_count, usually by inode->i_mutex.
1866 void inode_dio_wait(struct inode *inode)
1868 if (atomic_read(&inode->i_dio_count))
1869 __inode_dio_wait(inode);
1871 EXPORT_SYMBOL(inode_dio_wait);
1874 * inode_dio_done - signal finish of a direct I/O requests
1875 * @inode: inode the direct I/O happens on
1877 * This is called once we've finished processing a direct I/O request,
1878 * and is used to wake up callers waiting for direct I/O to be quiesced.
1880 void inode_dio_done(struct inode *inode)
1882 if (atomic_dec_and_test(&inode->i_dio_count))
1883 wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1885 EXPORT_SYMBOL(inode_dio_done);