2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
7 #include <linux/dcache.h>
8 #include <linux/init.h>
9 #include <linux/slab.h>
10 #include <linux/writeback.h>
11 #include <linux/module.h>
12 #include <linux/backing-dev.h>
13 #include <linux/wait.h>
14 #include <linux/rwsem.h>
15 #include <linux/hash.h>
16 #include <linux/swap.h>
17 #include <linux/security.h>
18 #include <linux/pagemap.h>
19 #include <linux/cdev.h>
20 #include <linux/bootmem.h>
21 #include <linux/fsnotify.h>
22 #include <linux/mount.h>
23 #include <linux/async.h>
24 #include <linux/posix_acl.h>
25 #include <linux/prefetch.h>
26 #include <linux/ima.h>
27 #include <linux/cred.h>
28 #include <linux/buffer_head.h> /* for inode_has_buffers */
32 * Inode locking rules:
34 * inode->i_lock protects:
35 * inode->i_state, inode->i_hash, __iget()
36 * inode_lru_lock protects:
37 * inode_lru, inode->i_lru
38 * inode_sb_list_lock protects:
39 * sb->s_inodes, inode->i_sb_list
40 * inode_wb_list_lock protects:
41 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
42 * inode_hash_lock protects:
43 * inode_hashtable, inode->i_hash
62 static unsigned int i_hash_mask __read_mostly;
63 static unsigned int i_hash_shift __read_mostly;
64 static struct hlist_head *inode_hashtable __read_mostly;
65 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
67 static LIST_HEAD(inode_lru);
68 static DEFINE_SPINLOCK(inode_lru_lock);
70 __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
71 __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_wb_list_lock);
74 * iprune_sem provides exclusion between the icache shrinking and the
77 * We don't actually need it to protect anything in the umount path,
78 * but only need to cycle through it to make sure any inode that
79 * prune_icache took off the LRU list has been fully torn down by the
80 * time we are past evict_inodes.
82 static DECLARE_RWSEM(iprune_sem);
85 * Empty aops. Can be used for the cases where the user does not
86 * define any of the address_space operations.
88 const struct address_space_operations empty_aops = {
90 EXPORT_SYMBOL(empty_aops);
93 * Statistics gathering..
95 struct inodes_stat_t inodes_stat;
97 static DEFINE_PER_CPU(unsigned int, nr_inodes);
99 static struct kmem_cache *inode_cachep __read_mostly;
101 static int get_nr_inodes(void)
105 for_each_possible_cpu(i)
106 sum += per_cpu(nr_inodes, i);
107 return sum < 0 ? 0 : sum;
110 static inline int get_nr_inodes_unused(void)
112 return inodes_stat.nr_unused;
115 int get_nr_dirty_inodes(void)
117 /* not actually dirty inodes, but a wild approximation */
118 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
119 return nr_dirty > 0 ? nr_dirty : 0;
123 * Handle nr_inode sysctl
126 int proc_nr_inodes(ctl_table *table, int write,
127 void __user *buffer, size_t *lenp, loff_t *ppos)
129 inodes_stat.nr_inodes = get_nr_inodes();
130 return proc_dointvec(table, write, buffer, lenp, ppos);
135 * inode_init_always - perform inode structure intialisation
136 * @sb: superblock inode belongs to
137 * @inode: inode to initialise
139 * These are initializations that need to be done on every inode
140 * allocation as the fields are not initialised by slab allocation.
142 int inode_init_always(struct super_block *sb, struct inode *inode)
144 static const struct inode_operations empty_iops;
145 static const struct file_operations empty_fops;
146 struct address_space *const mapping = &inode->i_data;
149 inode->i_blkbits = sb->s_blocksize_bits;
151 atomic_set(&inode->i_count, 1);
152 inode->i_op = &empty_iops;
153 inode->i_fop = &empty_fops;
157 atomic_set(&inode->i_writecount, 0);
161 inode->i_generation = 0;
163 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
165 inode->i_pipe = NULL;
166 inode->i_bdev = NULL;
167 inode->i_cdev = NULL;
169 inode->dirtied_when = 0;
171 if (security_inode_alloc(inode))
173 spin_lock_init(&inode->i_lock);
174 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
176 mutex_init(&inode->i_mutex);
177 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
179 init_rwsem(&inode->i_alloc_sem);
180 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
182 mapping->a_ops = &empty_aops;
183 mapping->host = inode;
185 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
186 mapping->assoc_mapping = NULL;
187 mapping->backing_dev_info = &default_backing_dev_info;
188 mapping->writeback_index = 0;
191 * If the block_device provides a backing_dev_info for client
192 * inodes then use that. Otherwise the inode share the bdev's
196 struct backing_dev_info *bdi;
198 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
199 mapping->backing_dev_info = bdi;
201 inode->i_private = NULL;
202 inode->i_mapping = mapping;
203 #ifdef CONFIG_FS_POSIX_ACL
204 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
207 #ifdef CONFIG_FSNOTIFY
208 inode->i_fsnotify_mask = 0;
211 this_cpu_inc(nr_inodes);
217 EXPORT_SYMBOL(inode_init_always);
219 static struct inode *alloc_inode(struct super_block *sb)
223 if (sb->s_op->alloc_inode)
224 inode = sb->s_op->alloc_inode(sb);
226 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
231 if (unlikely(inode_init_always(sb, inode))) {
232 if (inode->i_sb->s_op->destroy_inode)
233 inode->i_sb->s_op->destroy_inode(inode);
235 kmem_cache_free(inode_cachep, inode);
242 void free_inode_nonrcu(struct inode *inode)
244 kmem_cache_free(inode_cachep, inode);
246 EXPORT_SYMBOL(free_inode_nonrcu);
248 void __destroy_inode(struct inode *inode)
250 BUG_ON(inode_has_buffers(inode));
251 security_inode_free(inode);
252 fsnotify_inode_delete(inode);
253 #ifdef CONFIG_FS_POSIX_ACL
254 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
255 posix_acl_release(inode->i_acl);
256 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
257 posix_acl_release(inode->i_default_acl);
259 this_cpu_dec(nr_inodes);
261 EXPORT_SYMBOL(__destroy_inode);
263 static void i_callback(struct rcu_head *head)
265 struct inode *inode = container_of(head, struct inode, i_rcu);
266 INIT_LIST_HEAD(&inode->i_dentry);
267 kmem_cache_free(inode_cachep, inode);
270 static void destroy_inode(struct inode *inode)
272 BUG_ON(!list_empty(&inode->i_lru));
273 __destroy_inode(inode);
274 if (inode->i_sb->s_op->destroy_inode)
275 inode->i_sb->s_op->destroy_inode(inode);
277 call_rcu(&inode->i_rcu, i_callback);
280 void address_space_init_once(struct address_space *mapping)
282 memset(mapping, 0, sizeof(*mapping));
283 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
284 spin_lock_init(&mapping->tree_lock);
285 mutex_init(&mapping->i_mmap_mutex);
286 INIT_LIST_HEAD(&mapping->private_list);
287 spin_lock_init(&mapping->private_lock);
288 INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
289 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
291 EXPORT_SYMBOL(address_space_init_once);
294 * These are initializations that only need to be done
295 * once, because the fields are idempotent across use
296 * of the inode, so let the slab aware of that.
298 void inode_init_once(struct inode *inode)
300 memset(inode, 0, sizeof(*inode));
301 INIT_HLIST_NODE(&inode->i_hash);
302 INIT_LIST_HEAD(&inode->i_dentry);
303 INIT_LIST_HEAD(&inode->i_devices);
304 INIT_LIST_HEAD(&inode->i_wb_list);
305 INIT_LIST_HEAD(&inode->i_lru);
306 address_space_init_once(&inode->i_data);
307 i_size_ordered_init(inode);
308 #ifdef CONFIG_FSNOTIFY
309 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
312 EXPORT_SYMBOL(inode_init_once);
314 static void init_once(void *foo)
316 struct inode *inode = (struct inode *) foo;
318 inode_init_once(inode);
322 * inode->i_lock must be held
324 void __iget(struct inode *inode)
326 atomic_inc(&inode->i_count);
330 * get additional reference to inode; caller must already hold one.
332 void ihold(struct inode *inode)
334 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
336 EXPORT_SYMBOL(ihold);
338 static void inode_lru_list_add(struct inode *inode)
340 spin_lock(&inode_lru_lock);
341 if (list_empty(&inode->i_lru)) {
342 list_add(&inode->i_lru, &inode_lru);
343 inodes_stat.nr_unused++;
345 spin_unlock(&inode_lru_lock);
348 static void inode_lru_list_del(struct inode *inode)
350 spin_lock(&inode_lru_lock);
351 if (!list_empty(&inode->i_lru)) {
352 list_del_init(&inode->i_lru);
353 inodes_stat.nr_unused--;
355 spin_unlock(&inode_lru_lock);
359 * inode_sb_list_add - add inode to the superblock list of inodes
360 * @inode: inode to add
362 void inode_sb_list_add(struct inode *inode)
364 spin_lock(&inode_sb_list_lock);
365 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
366 spin_unlock(&inode_sb_list_lock);
368 EXPORT_SYMBOL_GPL(inode_sb_list_add);
370 static inline void inode_sb_list_del(struct inode *inode)
372 spin_lock(&inode_sb_list_lock);
373 list_del_init(&inode->i_sb_list);
374 spin_unlock(&inode_sb_list_lock);
377 static unsigned long hash(struct super_block *sb, unsigned long hashval)
381 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
383 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
384 return tmp & i_hash_mask;
388 * __insert_inode_hash - hash an inode
389 * @inode: unhashed inode
390 * @hashval: unsigned long value used to locate this object in the
393 * Add an inode to the inode hash for this superblock.
395 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
397 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
399 spin_lock(&inode_hash_lock);
400 spin_lock(&inode->i_lock);
401 hlist_add_head(&inode->i_hash, b);
402 spin_unlock(&inode->i_lock);
403 spin_unlock(&inode_hash_lock);
405 EXPORT_SYMBOL(__insert_inode_hash);
408 * remove_inode_hash - remove an inode from the hash
409 * @inode: inode to unhash
411 * Remove an inode from the superblock.
413 void remove_inode_hash(struct inode *inode)
415 spin_lock(&inode_hash_lock);
416 spin_lock(&inode->i_lock);
417 hlist_del_init(&inode->i_hash);
418 spin_unlock(&inode->i_lock);
419 spin_unlock(&inode_hash_lock);
421 EXPORT_SYMBOL(remove_inode_hash);
423 void end_writeback(struct inode *inode)
427 * We have to cycle tree_lock here because reclaim can be still in the
428 * process of removing the last page (in __delete_from_page_cache())
429 * and we must not free mapping under it.
431 spin_lock_irq(&inode->i_data.tree_lock);
432 BUG_ON(inode->i_data.nrpages);
433 spin_unlock_irq(&inode->i_data.tree_lock);
434 BUG_ON(!list_empty(&inode->i_data.private_list));
435 BUG_ON(!(inode->i_state & I_FREEING));
436 BUG_ON(inode->i_state & I_CLEAR);
437 inode_sync_wait(inode);
438 /* don't need i_lock here, no concurrent mods to i_state */
439 inode->i_state = I_FREEING | I_CLEAR;
441 EXPORT_SYMBOL(end_writeback);
444 * Free the inode passed in, removing it from the lists it is still connected
445 * to. We remove any pages still attached to the inode and wait for any IO that
446 * is still in progress before finally destroying the inode.
448 * An inode must already be marked I_FREEING so that we avoid the inode being
449 * moved back onto lists if we race with other code that manipulates the lists
450 * (e.g. writeback_single_inode). The caller is responsible for setting this.
452 * An inode must already be removed from the LRU list before being evicted from
453 * the cache. This should occur atomically with setting the I_FREEING state
454 * flag, so no inodes here should ever be on the LRU when being evicted.
456 static void evict(struct inode *inode)
458 const struct super_operations *op = inode->i_sb->s_op;
460 BUG_ON(!(inode->i_state & I_FREEING));
461 BUG_ON(!list_empty(&inode->i_lru));
463 inode_wb_list_del(inode);
464 inode_sb_list_del(inode);
466 if (op->evict_inode) {
467 op->evict_inode(inode);
469 if (inode->i_data.nrpages)
470 truncate_inode_pages(&inode->i_data, 0);
471 end_writeback(inode);
473 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
475 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
478 remove_inode_hash(inode);
480 spin_lock(&inode->i_lock);
481 wake_up_bit(&inode->i_state, __I_NEW);
482 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
483 spin_unlock(&inode->i_lock);
485 destroy_inode(inode);
489 * dispose_list - dispose of the contents of a local list
490 * @head: the head of the list to free
492 * Dispose-list gets a local list with local inodes in it, so it doesn't
493 * need to worry about list corruption and SMP locks.
495 static void dispose_list(struct list_head *head)
497 while (!list_empty(head)) {
500 inode = list_first_entry(head, struct inode, i_lru);
501 list_del_init(&inode->i_lru);
508 * evict_inodes - evict all evictable inodes for a superblock
509 * @sb: superblock to operate on
511 * Make sure that no inodes with zero refcount are retained. This is
512 * called by superblock shutdown after having MS_ACTIVE flag removed,
513 * so any inode reaching zero refcount during or after that call will
514 * be immediately evicted.
516 void evict_inodes(struct super_block *sb)
518 struct inode *inode, *next;
521 spin_lock(&inode_sb_list_lock);
522 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
523 if (atomic_read(&inode->i_count))
526 spin_lock(&inode->i_lock);
527 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
528 spin_unlock(&inode->i_lock);
532 inode->i_state |= I_FREEING;
533 inode_lru_list_del(inode);
534 spin_unlock(&inode->i_lock);
535 list_add(&inode->i_lru, &dispose);
537 spin_unlock(&inode_sb_list_lock);
539 dispose_list(&dispose);
542 * Cycle through iprune_sem to make sure any inode that prune_icache
543 * moved off the list before we took the lock has been fully torn
546 down_write(&iprune_sem);
547 up_write(&iprune_sem);
551 * invalidate_inodes - attempt to free all inodes on a superblock
552 * @sb: superblock to operate on
553 * @kill_dirty: flag to guide handling of dirty inodes
555 * Attempts to free all inodes for a given superblock. If there were any
556 * busy inodes return a non-zero value, else zero.
557 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
560 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
563 struct inode *inode, *next;
566 spin_lock(&inode_sb_list_lock);
567 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
568 spin_lock(&inode->i_lock);
569 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
570 spin_unlock(&inode->i_lock);
573 if (inode->i_state & I_DIRTY && !kill_dirty) {
574 spin_unlock(&inode->i_lock);
578 if (atomic_read(&inode->i_count)) {
579 spin_unlock(&inode->i_lock);
584 inode->i_state |= I_FREEING;
585 inode_lru_list_del(inode);
586 spin_unlock(&inode->i_lock);
587 list_add(&inode->i_lru, &dispose);
589 spin_unlock(&inode_sb_list_lock);
591 dispose_list(&dispose);
596 static int can_unuse(struct inode *inode)
598 if (inode->i_state & ~I_REFERENCED)
600 if (inode_has_buffers(inode))
602 if (atomic_read(&inode->i_count))
604 if (inode->i_data.nrpages)
610 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
611 * temporary list and then are freed outside inode_lru_lock by dispose_list().
613 * Any inodes which are pinned purely because of attached pagecache have their
614 * pagecache removed. If the inode has metadata buffers attached to
615 * mapping->private_list then try to remove them.
617 * If the inode has the I_REFERENCED flag set, then it means that it has been
618 * used recently - the flag is set in iput_final(). When we encounter such an
619 * inode, clear the flag and move it to the back of the LRU so it gets another
620 * pass through the LRU before it gets reclaimed. This is necessary because of
621 * the fact we are doing lazy LRU updates to minimise lock contention so the
622 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
623 * with this flag set because they are the inodes that are out of order.
625 static void prune_icache(int nr_to_scan)
629 unsigned long reap = 0;
631 down_read(&iprune_sem);
632 spin_lock(&inode_lru_lock);
633 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
636 if (list_empty(&inode_lru))
639 inode = list_entry(inode_lru.prev, struct inode, i_lru);
642 * we are inverting the inode_lru_lock/inode->i_lock here,
643 * so use a trylock. If we fail to get the lock, just move the
644 * inode to the back of the list so we don't spin on it.
646 if (!spin_trylock(&inode->i_lock)) {
647 list_move(&inode->i_lru, &inode_lru);
652 * Referenced or dirty inodes are still in use. Give them
653 * another pass through the LRU as we canot reclaim them now.
655 if (atomic_read(&inode->i_count) ||
656 (inode->i_state & ~I_REFERENCED)) {
657 list_del_init(&inode->i_lru);
658 spin_unlock(&inode->i_lock);
659 inodes_stat.nr_unused--;
663 /* recently referenced inodes get one more pass */
664 if (inode->i_state & I_REFERENCED) {
665 inode->i_state &= ~I_REFERENCED;
666 list_move(&inode->i_lru, &inode_lru);
667 spin_unlock(&inode->i_lock);
670 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
672 spin_unlock(&inode->i_lock);
673 spin_unlock(&inode_lru_lock);
674 if (remove_inode_buffers(inode))
675 reap += invalidate_mapping_pages(&inode->i_data,
678 spin_lock(&inode_lru_lock);
680 if (inode != list_entry(inode_lru.next,
681 struct inode, i_lru))
682 continue; /* wrong inode or list_empty */
683 /* avoid lock inversions with trylock */
684 if (!spin_trylock(&inode->i_lock))
686 if (!can_unuse(inode)) {
687 spin_unlock(&inode->i_lock);
691 WARN_ON(inode->i_state & I_NEW);
692 inode->i_state |= I_FREEING;
693 spin_unlock(&inode->i_lock);
695 list_move(&inode->i_lru, &freeable);
696 inodes_stat.nr_unused--;
698 if (current_is_kswapd())
699 __count_vm_events(KSWAPD_INODESTEAL, reap);
701 __count_vm_events(PGINODESTEAL, reap);
702 spin_unlock(&inode_lru_lock);
704 dispose_list(&freeable);
705 up_read(&iprune_sem);
709 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
710 * "unused" means that no dentries are referring to the inodes: the files are
711 * not open and the dcache references to those inodes have already been
714 * This function is passed the number of inodes to scan, and it returns the
715 * total number of remaining possibly-reclaimable inodes.
717 static int shrink_icache_memory(struct shrinker *shrink,
718 struct shrink_control *sc)
720 int nr = sc->nr_to_scan;
721 gfp_t gfp_mask = sc->gfp_mask;
725 * Nasty deadlock avoidance. We may hold various FS locks,
726 * and we don't want to recurse into the FS that called us
727 * in clear_inode() and friends..
729 if (!(gfp_mask & __GFP_FS))
733 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
736 static struct shrinker icache_shrinker = {
737 .shrink = shrink_icache_memory,
738 .seeks = DEFAULT_SEEKS,
741 static void __wait_on_freeing_inode(struct inode *inode);
743 * Called with the inode lock held.
745 static struct inode *find_inode(struct super_block *sb,
746 struct hlist_head *head,
747 int (*test)(struct inode *, void *),
750 struct hlist_node *node;
751 struct inode *inode = NULL;
754 hlist_for_each_entry(inode, node, head, i_hash) {
755 spin_lock(&inode->i_lock);
756 if (inode->i_sb != sb) {
757 spin_unlock(&inode->i_lock);
760 if (!test(inode, data)) {
761 spin_unlock(&inode->i_lock);
764 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
765 __wait_on_freeing_inode(inode);
769 spin_unlock(&inode->i_lock);
776 * find_inode_fast is the fast path version of find_inode, see the comment at
777 * iget_locked for details.
779 static struct inode *find_inode_fast(struct super_block *sb,
780 struct hlist_head *head, unsigned long ino)
782 struct hlist_node *node;
783 struct inode *inode = NULL;
786 hlist_for_each_entry(inode, node, head, i_hash) {
787 spin_lock(&inode->i_lock);
788 if (inode->i_ino != ino) {
789 spin_unlock(&inode->i_lock);
792 if (inode->i_sb != sb) {
793 spin_unlock(&inode->i_lock);
796 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
797 __wait_on_freeing_inode(inode);
801 spin_unlock(&inode->i_lock);
808 * Each cpu owns a range of LAST_INO_BATCH numbers.
809 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
810 * to renew the exhausted range.
812 * This does not significantly increase overflow rate because every CPU can
813 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
814 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
815 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
816 * overflow rate by 2x, which does not seem too significant.
818 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
819 * error if st_ino won't fit in target struct field. Use 32bit counter
820 * here to attempt to avoid that.
822 #define LAST_INO_BATCH 1024
823 static DEFINE_PER_CPU(unsigned int, last_ino);
825 unsigned int get_next_ino(void)
827 unsigned int *p = &get_cpu_var(last_ino);
828 unsigned int res = *p;
831 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
832 static atomic_t shared_last_ino;
833 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
835 res = next - LAST_INO_BATCH;
840 put_cpu_var(last_ino);
843 EXPORT_SYMBOL(get_next_ino);
846 * new_inode - obtain an inode
849 * Allocates a new inode for given superblock. The default gfp_mask
850 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
851 * If HIGHMEM pages are unsuitable or it is known that pages allocated
852 * for the page cache are not reclaimable or migratable,
853 * mapping_set_gfp_mask() must be called with suitable flags on the
854 * newly created inode's mapping
857 struct inode *new_inode(struct super_block *sb)
861 spin_lock_prefetch(&inode_sb_list_lock);
863 inode = alloc_inode(sb);
865 spin_lock(&inode->i_lock);
867 spin_unlock(&inode->i_lock);
868 inode_sb_list_add(inode);
872 EXPORT_SYMBOL(new_inode);
875 * unlock_new_inode - clear the I_NEW state and wake up any waiters
876 * @inode: new inode to unlock
878 * Called when the inode is fully initialised to clear the new state of the
879 * inode and wake up anyone waiting for the inode to finish initialisation.
881 void unlock_new_inode(struct inode *inode)
883 #ifdef CONFIG_DEBUG_LOCK_ALLOC
884 if (S_ISDIR(inode->i_mode)) {
885 struct file_system_type *type = inode->i_sb->s_type;
887 /* Set new key only if filesystem hasn't already changed it */
888 if (!lockdep_match_class(&inode->i_mutex,
889 &type->i_mutex_key)) {
891 * ensure nobody is actually holding i_mutex
893 mutex_destroy(&inode->i_mutex);
894 mutex_init(&inode->i_mutex);
895 lockdep_set_class(&inode->i_mutex,
896 &type->i_mutex_dir_key);
900 spin_lock(&inode->i_lock);
901 WARN_ON(!(inode->i_state & I_NEW));
902 inode->i_state &= ~I_NEW;
903 wake_up_bit(&inode->i_state, __I_NEW);
904 spin_unlock(&inode->i_lock);
906 EXPORT_SYMBOL(unlock_new_inode);
909 * iget5_locked - obtain an inode from a mounted file system
910 * @sb: super block of file system
911 * @hashval: hash value (usually inode number) to get
912 * @test: callback used for comparisons between inodes
913 * @set: callback used to initialize a new struct inode
914 * @data: opaque data pointer to pass to @test and @set
916 * Search for the inode specified by @hashval and @data in the inode cache,
917 * and if present it is return it with an increased reference count. This is
918 * a generalized version of iget_locked() for file systems where the inode
919 * number is not sufficient for unique identification of an inode.
921 * If the inode is not in cache, allocate a new inode and return it locked,
922 * hashed, and with the I_NEW flag set. The file system gets to fill it in
923 * before unlocking it via unlock_new_inode().
925 * Note both @test and @set are called with the inode_hash_lock held, so can't
928 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
929 int (*test)(struct inode *, void *),
930 int (*set)(struct inode *, void *), void *data)
932 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
935 spin_lock(&inode_hash_lock);
936 inode = find_inode(sb, head, test, data);
937 spin_unlock(&inode_hash_lock);
940 wait_on_inode(inode);
944 inode = alloc_inode(sb);
948 spin_lock(&inode_hash_lock);
949 /* We released the lock, so.. */
950 old = find_inode(sb, head, test, data);
952 if (set(inode, data))
955 spin_lock(&inode->i_lock);
956 inode->i_state = I_NEW;
957 hlist_add_head(&inode->i_hash, head);
958 spin_unlock(&inode->i_lock);
959 inode_sb_list_add(inode);
960 spin_unlock(&inode_hash_lock);
962 /* Return the locked inode with I_NEW set, the
963 * caller is responsible for filling in the contents
969 * Uhhuh, somebody else created the same inode under
970 * us. Use the old inode instead of the one we just
973 spin_unlock(&inode_hash_lock);
974 destroy_inode(inode);
976 wait_on_inode(inode);
981 spin_unlock(&inode_hash_lock);
982 destroy_inode(inode);
985 EXPORT_SYMBOL(iget5_locked);
988 * iget_locked - obtain an inode from a mounted file system
989 * @sb: super block of file system
990 * @ino: inode number to get
992 * Search for the inode specified by @ino in the inode cache and if present
993 * return it with an increased reference count. This is for file systems
994 * where the inode number is sufficient for unique identification of an inode.
996 * If the inode is not in cache, allocate a new inode and return it locked,
997 * hashed, and with the I_NEW flag set. The file system gets to fill it in
998 * before unlocking it via unlock_new_inode().
1000 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1002 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1003 struct inode *inode;
1005 spin_lock(&inode_hash_lock);
1006 inode = find_inode_fast(sb, head, ino);
1007 spin_unlock(&inode_hash_lock);
1009 wait_on_inode(inode);
1013 inode = alloc_inode(sb);
1017 spin_lock(&inode_hash_lock);
1018 /* We released the lock, so.. */
1019 old = find_inode_fast(sb, head, ino);
1022 spin_lock(&inode->i_lock);
1023 inode->i_state = I_NEW;
1024 hlist_add_head(&inode->i_hash, head);
1025 spin_unlock(&inode->i_lock);
1026 inode_sb_list_add(inode);
1027 spin_unlock(&inode_hash_lock);
1029 /* Return the locked inode with I_NEW set, the
1030 * caller is responsible for filling in the contents
1036 * Uhhuh, somebody else created the same inode under
1037 * us. Use the old inode instead of the one we just
1040 spin_unlock(&inode_hash_lock);
1041 destroy_inode(inode);
1043 wait_on_inode(inode);
1047 EXPORT_SYMBOL(iget_locked);
1050 * search the inode cache for a matching inode number.
1051 * If we find one, then the inode number we are trying to
1052 * allocate is not unique and so we should not use it.
1054 * Returns 1 if the inode number is unique, 0 if it is not.
1056 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1058 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1059 struct hlist_node *node;
1060 struct inode *inode;
1062 spin_lock(&inode_hash_lock);
1063 hlist_for_each_entry(inode, node, b, i_hash) {
1064 if (inode->i_ino == ino && inode->i_sb == sb) {
1065 spin_unlock(&inode_hash_lock);
1069 spin_unlock(&inode_hash_lock);
1075 * iunique - get a unique inode number
1077 * @max_reserved: highest reserved inode number
1079 * Obtain an inode number that is unique on the system for a given
1080 * superblock. This is used by file systems that have no natural
1081 * permanent inode numbering system. An inode number is returned that
1082 * is higher than the reserved limit but unique.
1085 * With a large number of inodes live on the file system this function
1086 * currently becomes quite slow.
1088 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1091 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1092 * error if st_ino won't fit in target struct field. Use 32bit counter
1093 * here to attempt to avoid that.
1095 static DEFINE_SPINLOCK(iunique_lock);
1096 static unsigned int counter;
1099 spin_lock(&iunique_lock);
1101 if (counter <= max_reserved)
1102 counter = max_reserved + 1;
1104 } while (!test_inode_iunique(sb, res));
1105 spin_unlock(&iunique_lock);
1109 EXPORT_SYMBOL(iunique);
1111 struct inode *igrab(struct inode *inode)
1113 spin_lock(&inode->i_lock);
1114 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1116 spin_unlock(&inode->i_lock);
1118 spin_unlock(&inode->i_lock);
1120 * Handle the case where s_op->clear_inode is not been
1121 * called yet, and somebody is calling igrab
1122 * while the inode is getting freed.
1128 EXPORT_SYMBOL(igrab);
1131 * ilookup5_nowait - search for an inode in the inode cache
1132 * @sb: super block of file system to search
1133 * @hashval: hash value (usually inode number) to search for
1134 * @test: callback used for comparisons between inodes
1135 * @data: opaque data pointer to pass to @test
1137 * Search for the inode specified by @hashval and @data in the inode cache.
1138 * If the inode is in the cache, the inode is returned with an incremented
1141 * Note: I_NEW is not waited upon so you have to be very careful what you do
1142 * with the returned inode. You probably should be using ilookup5() instead.
1144 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1146 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1147 int (*test)(struct inode *, void *), void *data)
1149 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1150 struct inode *inode;
1152 spin_lock(&inode_hash_lock);
1153 inode = find_inode(sb, head, test, data);
1154 spin_unlock(&inode_hash_lock);
1158 EXPORT_SYMBOL(ilookup5_nowait);
1161 * ilookup5 - search for an inode in the inode cache
1162 * @sb: super block of file system to search
1163 * @hashval: hash value (usually inode number) to search for
1164 * @test: callback used for comparisons between inodes
1165 * @data: opaque data pointer to pass to @test
1167 * Search for the inode specified by @hashval and @data in the inode cache,
1168 * and if the inode is in the cache, return the inode with an incremented
1169 * reference count. Waits on I_NEW before returning the inode.
1170 * returned with an incremented reference count.
1172 * This is a generalized version of ilookup() for file systems where the
1173 * inode number is not sufficient for unique identification of an inode.
1175 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1177 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1178 int (*test)(struct inode *, void *), void *data)
1180 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1183 wait_on_inode(inode);
1186 EXPORT_SYMBOL(ilookup5);
1189 * ilookup - search for an inode in the inode cache
1190 * @sb: super block of file system to search
1191 * @ino: inode number to search for
1193 * Search for the inode @ino in the inode cache, and if the inode is in the
1194 * cache, the inode is returned with an incremented reference count.
1196 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1198 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1199 struct inode *inode;
1201 spin_lock(&inode_hash_lock);
1202 inode = find_inode_fast(sb, head, ino);
1203 spin_unlock(&inode_hash_lock);
1206 wait_on_inode(inode);
1209 EXPORT_SYMBOL(ilookup);
1211 int insert_inode_locked(struct inode *inode)
1213 struct super_block *sb = inode->i_sb;
1214 ino_t ino = inode->i_ino;
1215 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1218 struct hlist_node *node;
1219 struct inode *old = NULL;
1220 spin_lock(&inode_hash_lock);
1221 hlist_for_each_entry(old, node, head, i_hash) {
1222 if (old->i_ino != ino)
1224 if (old->i_sb != sb)
1226 spin_lock(&old->i_lock);
1227 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1228 spin_unlock(&old->i_lock);
1233 if (likely(!node)) {
1234 spin_lock(&inode->i_lock);
1235 inode->i_state |= I_NEW;
1236 hlist_add_head(&inode->i_hash, head);
1237 spin_unlock(&inode->i_lock);
1238 spin_unlock(&inode_hash_lock);
1242 spin_unlock(&old->i_lock);
1243 spin_unlock(&inode_hash_lock);
1245 if (unlikely(!inode_unhashed(old))) {
1252 EXPORT_SYMBOL(insert_inode_locked);
1254 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1255 int (*test)(struct inode *, void *), void *data)
1257 struct super_block *sb = inode->i_sb;
1258 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1261 struct hlist_node *node;
1262 struct inode *old = NULL;
1264 spin_lock(&inode_hash_lock);
1265 hlist_for_each_entry(old, node, head, i_hash) {
1266 if (old->i_sb != sb)
1268 if (!test(old, data))
1270 spin_lock(&old->i_lock);
1271 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1272 spin_unlock(&old->i_lock);
1277 if (likely(!node)) {
1278 spin_lock(&inode->i_lock);
1279 inode->i_state |= I_NEW;
1280 hlist_add_head(&inode->i_hash, head);
1281 spin_unlock(&inode->i_lock);
1282 spin_unlock(&inode_hash_lock);
1286 spin_unlock(&old->i_lock);
1287 spin_unlock(&inode_hash_lock);
1289 if (unlikely(!inode_unhashed(old))) {
1296 EXPORT_SYMBOL(insert_inode_locked4);
1299 int generic_delete_inode(struct inode *inode)
1303 EXPORT_SYMBOL(generic_delete_inode);
1306 * Normal UNIX filesystem behaviour: delete the
1307 * inode when the usage count drops to zero, and
1310 int generic_drop_inode(struct inode *inode)
1312 return !inode->i_nlink || inode_unhashed(inode);
1314 EXPORT_SYMBOL_GPL(generic_drop_inode);
1317 * Called when we're dropping the last reference
1320 * Call the FS "drop_inode()" function, defaulting to
1321 * the legacy UNIX filesystem behaviour. If it tells
1322 * us to evict inode, do so. Otherwise, retain inode
1323 * in cache if fs is alive, sync and evict if fs is
1326 static void iput_final(struct inode *inode)
1328 struct super_block *sb = inode->i_sb;
1329 const struct super_operations *op = inode->i_sb->s_op;
1332 WARN_ON(inode->i_state & I_NEW);
1334 if (op && op->drop_inode)
1335 drop = op->drop_inode(inode);
1337 drop = generic_drop_inode(inode);
1339 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1340 inode->i_state |= I_REFERENCED;
1341 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1342 inode_lru_list_add(inode);
1343 spin_unlock(&inode->i_lock);
1348 inode->i_state |= I_WILL_FREE;
1349 spin_unlock(&inode->i_lock);
1350 write_inode_now(inode, 1);
1351 spin_lock(&inode->i_lock);
1352 WARN_ON(inode->i_state & I_NEW);
1353 inode->i_state &= ~I_WILL_FREE;
1356 inode->i_state |= I_FREEING;
1357 inode_lru_list_del(inode);
1358 spin_unlock(&inode->i_lock);
1364 * iput - put an inode
1365 * @inode: inode to put
1367 * Puts an inode, dropping its usage count. If the inode use count hits
1368 * zero, the inode is then freed and may also be destroyed.
1370 * Consequently, iput() can sleep.
1372 void iput(struct inode *inode)
1375 BUG_ON(inode->i_state & I_CLEAR);
1377 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1381 EXPORT_SYMBOL(iput);
1384 * bmap - find a block number in a file
1385 * @inode: inode of file
1386 * @block: block to find
1388 * Returns the block number on the device holding the inode that
1389 * is the disk block number for the block of the file requested.
1390 * That is, asked for block 4 of inode 1 the function will return the
1391 * disk block relative to the disk start that holds that block of the
1394 sector_t bmap(struct inode *inode, sector_t block)
1397 if (inode->i_mapping->a_ops->bmap)
1398 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1401 EXPORT_SYMBOL(bmap);
1404 * With relative atime, only update atime if the previous atime is
1405 * earlier than either the ctime or mtime or if at least a day has
1406 * passed since the last atime update.
1408 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1409 struct timespec now)
1412 if (!(mnt->mnt_flags & MNT_RELATIME))
1415 * Is mtime younger than atime? If yes, update atime:
1417 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1420 * Is ctime younger than atime? If yes, update atime:
1422 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1426 * Is the previous atime value older than a day? If yes,
1429 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1432 * Good, we can skip the atime update:
1438 * touch_atime - update the access time
1439 * @mnt: mount the inode is accessed on
1440 * @dentry: dentry accessed
1442 * Update the accessed time on an inode and mark it for writeback.
1443 * This function automatically handles read only file systems and media,
1444 * as well as the "noatime" flag and inode specific "noatime" markers.
1446 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1448 struct inode *inode = dentry->d_inode;
1449 struct timespec now;
1451 if (inode->i_flags & S_NOATIME)
1453 if (IS_NOATIME(inode))
1455 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1458 if (mnt->mnt_flags & MNT_NOATIME)
1460 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1463 now = current_fs_time(inode->i_sb);
1465 if (!relatime_need_update(mnt, inode, now))
1468 if (timespec_equal(&inode->i_atime, &now))
1471 if (mnt_want_write(mnt))
1474 inode->i_atime = now;
1475 mark_inode_dirty_sync(inode);
1476 mnt_drop_write(mnt);
1478 EXPORT_SYMBOL(touch_atime);
1481 * file_update_time - update mtime and ctime time
1482 * @file: file accessed
1484 * Update the mtime and ctime members of an inode and mark the inode
1485 * for writeback. Note that this function is meant exclusively for
1486 * usage in the file write path of filesystems, and filesystems may
1487 * choose to explicitly ignore update via this function with the
1488 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1489 * timestamps are handled by the server.
1492 void file_update_time(struct file *file)
1494 struct inode *inode = file->f_path.dentry->d_inode;
1495 struct timespec now;
1496 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1498 /* First try to exhaust all avenues to not sync */
1499 if (IS_NOCMTIME(inode))
1502 now = current_fs_time(inode->i_sb);
1503 if (!timespec_equal(&inode->i_mtime, &now))
1506 if (!timespec_equal(&inode->i_ctime, &now))
1509 if (IS_I_VERSION(inode))
1510 sync_it |= S_VERSION;
1515 /* Finally allowed to write? Takes lock. */
1516 if (mnt_want_write_file(file))
1519 /* Only change inode inside the lock region */
1520 if (sync_it & S_VERSION)
1521 inode_inc_iversion(inode);
1522 if (sync_it & S_CTIME)
1523 inode->i_ctime = now;
1524 if (sync_it & S_MTIME)
1525 inode->i_mtime = now;
1526 mark_inode_dirty_sync(inode);
1527 mnt_drop_write(file->f_path.mnt);
1529 EXPORT_SYMBOL(file_update_time);
1531 int inode_needs_sync(struct inode *inode)
1535 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1539 EXPORT_SYMBOL(inode_needs_sync);
1541 int inode_wait(void *word)
1546 EXPORT_SYMBOL(inode_wait);
1549 * If we try to find an inode in the inode hash while it is being
1550 * deleted, we have to wait until the filesystem completes its
1551 * deletion before reporting that it isn't found. This function waits
1552 * until the deletion _might_ have completed. Callers are responsible
1553 * to recheck inode state.
1555 * It doesn't matter if I_NEW is not set initially, a call to
1556 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1559 static void __wait_on_freeing_inode(struct inode *inode)
1561 wait_queue_head_t *wq;
1562 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1563 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1564 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1565 spin_unlock(&inode->i_lock);
1566 spin_unlock(&inode_hash_lock);
1568 finish_wait(wq, &wait.wait);
1569 spin_lock(&inode_hash_lock);
1572 static __initdata unsigned long ihash_entries;
1573 static int __init set_ihash_entries(char *str)
1577 ihash_entries = simple_strtoul(str, &str, 0);
1580 __setup("ihash_entries=", set_ihash_entries);
1583 * Initialize the waitqueues and inode hash table.
1585 void __init inode_init_early(void)
1589 /* If hashes are distributed across NUMA nodes, defer
1590 * hash allocation until vmalloc space is available.
1596 alloc_large_system_hash("Inode-cache",
1597 sizeof(struct hlist_head),
1605 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1606 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1609 void __init inode_init(void)
1613 /* inode slab cache */
1614 inode_cachep = kmem_cache_create("inode_cache",
1615 sizeof(struct inode),
1617 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1620 register_shrinker(&icache_shrinker);
1622 /* Hash may have been set up in inode_init_early */
1627 alloc_large_system_hash("Inode-cache",
1628 sizeof(struct hlist_head),
1636 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1637 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1640 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1642 inode->i_mode = mode;
1643 if (S_ISCHR(mode)) {
1644 inode->i_fop = &def_chr_fops;
1645 inode->i_rdev = rdev;
1646 } else if (S_ISBLK(mode)) {
1647 inode->i_fop = &def_blk_fops;
1648 inode->i_rdev = rdev;
1649 } else if (S_ISFIFO(mode))
1650 inode->i_fop = &def_fifo_fops;
1651 else if (S_ISSOCK(mode))
1652 inode->i_fop = &bad_sock_fops;
1654 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1655 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1658 EXPORT_SYMBOL(init_special_inode);
1661 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1663 * @dir: Directory inode
1664 * @mode: mode of the new inode
1666 void inode_init_owner(struct inode *inode, const struct inode *dir,
1669 inode->i_uid = current_fsuid();
1670 if (dir && dir->i_mode & S_ISGID) {
1671 inode->i_gid = dir->i_gid;
1675 inode->i_gid = current_fsgid();
1676 inode->i_mode = mode;
1678 EXPORT_SYMBOL(inode_init_owner);
1681 * inode_owner_or_capable - check current task permissions to inode
1682 * @inode: inode being checked
1684 * Return true if current either has CAP_FOWNER to the inode, or
1687 bool inode_owner_or_capable(const struct inode *inode)
1689 struct user_namespace *ns = inode_userns(inode);
1691 if (current_user_ns() == ns && current_fsuid() == inode->i_uid)
1693 if (ns_capable(ns, CAP_FOWNER))
1697 EXPORT_SYMBOL(inode_owner_or_capable);