4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/slab.h>
12 #include <linux/writeback.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/wait.h>
16 #include <linux/rwsem.h>
17 #include <linux/hash.h>
18 #include <linux/swap.h>
19 #include <linux/security.h>
20 #include <linux/pagemap.h>
21 #include <linux/cdev.h>
22 #include <linux/bootmem.h>
23 #include <linux/fsnotify.h>
24 #include <linux/mount.h>
25 #include <linux/async.h>
26 #include <linux/posix_acl.h>
27 #include <linux/ima.h>
30 * This is needed for the following functions:
34 * FIXME: remove all knowledge of the buffer layer from this file
36 #include <linux/buffer_head.h>
39 * New inode.c implementation.
41 * This implementation has the basic premise of trying
42 * to be extremely low-overhead and SMP-safe, yet be
43 * simple enough to be "obviously correct".
48 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
50 /* #define INODE_PARANOIA 1 */
51 /* #define INODE_DEBUG 1 */
54 * Inode lookup is no longer as critical as it used to be:
55 * most of the lookups are going to be through the dcache.
57 #define I_HASHBITS i_hash_shift
58 #define I_HASHMASK i_hash_mask
60 static unsigned int i_hash_mask __read_mostly;
61 static unsigned int i_hash_shift __read_mostly;
64 * Each inode can be on two separate lists. One is
65 * the hash list of the inode, used for lookups. The
66 * other linked list is the "type" list:
67 * "in_use" - valid inode, i_count > 0, i_nlink > 0
68 * "dirty" - as "in_use" but also dirty
69 * "unused" - valid inode, i_count = 0
71 * A "dirty" list is maintained for each super block,
72 * allowing for low-overhead inode sync() operations.
75 static LIST_HEAD(inode_lru);
76 static struct hlist_head *inode_hashtable __read_mostly;
79 * A simple spinlock to protect the list manipulations.
81 * NOTE! You also have to own the lock if you change
82 * the i_state of an inode while it is in use..
84 DEFINE_SPINLOCK(inode_lock);
87 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
88 * icache shrinking path, and the umount path. Without this exclusion,
89 * by the time prune_icache calls iput for the inode whose pages it has
90 * been invalidating, or by the time it calls clear_inode & destroy_inode
91 * from its final dispose_list, the struct super_block they refer to
92 * (for inode->i_sb->s_op) may already have been freed and reused.
94 * We make this an rwsem because the fastpath is icache shrinking. In
95 * some cases a filesystem may be doing a significant amount of work in
96 * its inode reclaim code, so this should improve parallelism.
98 static DECLARE_RWSEM(iprune_sem);
101 * Statistics gathering..
103 struct inodes_stat_t inodes_stat;
105 static DEFINE_PER_CPU(unsigned int, nr_inodes);
107 static struct kmem_cache *inode_cachep __read_mostly;
109 static int get_nr_inodes(void)
113 for_each_possible_cpu(i)
114 sum += per_cpu(nr_inodes, i);
115 return sum < 0 ? 0 : sum;
118 static inline int get_nr_inodes_unused(void)
120 return inodes_stat.nr_unused;
123 int get_nr_dirty_inodes(void)
125 /* not actually dirty inodes, but a wild approximation */
126 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
127 return nr_dirty > 0 ? nr_dirty : 0;
131 * Handle nr_inode sysctl
134 int proc_nr_inodes(ctl_table *table, int write,
135 void __user *buffer, size_t *lenp, loff_t *ppos)
137 inodes_stat.nr_inodes = get_nr_inodes();
138 return proc_dointvec(table, write, buffer, lenp, ppos);
142 static void wake_up_inode(struct inode *inode)
145 * Prevent speculative execution through spin_unlock(&inode_lock);
148 wake_up_bit(&inode->i_state, __I_NEW);
152 * inode_init_always - perform inode structure intialisation
153 * @sb: superblock inode belongs to
154 * @inode: inode to initialise
156 * These are initializations that need to be done on every inode
157 * allocation as the fields are not initialised by slab allocation.
159 int inode_init_always(struct super_block *sb, struct inode *inode)
161 static const struct address_space_operations empty_aops;
162 static const struct inode_operations empty_iops;
163 static const struct file_operations empty_fops;
164 struct address_space *const mapping = &inode->i_data;
167 inode->i_blkbits = sb->s_blocksize_bits;
169 atomic_set(&inode->i_count, 1);
170 inode->i_op = &empty_iops;
171 inode->i_fop = &empty_fops;
175 atomic_set(&inode->i_writecount, 0);
179 inode->i_generation = 0;
181 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
183 inode->i_pipe = NULL;
184 inode->i_bdev = NULL;
185 inode->i_cdev = NULL;
187 inode->dirtied_when = 0;
189 if (security_inode_alloc(inode))
191 spin_lock_init(&inode->i_lock);
192 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
194 mutex_init(&inode->i_mutex);
195 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
197 init_rwsem(&inode->i_alloc_sem);
198 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
200 mapping->a_ops = &empty_aops;
201 mapping->host = inode;
203 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
204 mapping->assoc_mapping = NULL;
205 mapping->backing_dev_info = &default_backing_dev_info;
206 mapping->writeback_index = 0;
209 * If the block_device provides a backing_dev_info for client
210 * inodes then use that. Otherwise the inode share the bdev's
214 struct backing_dev_info *bdi;
216 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
217 mapping->backing_dev_info = bdi;
219 inode->i_private = NULL;
220 inode->i_mapping = mapping;
221 #ifdef CONFIG_FS_POSIX_ACL
222 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
225 #ifdef CONFIG_FSNOTIFY
226 inode->i_fsnotify_mask = 0;
229 this_cpu_inc(nr_inodes);
235 EXPORT_SYMBOL(inode_init_always);
237 static struct inode *alloc_inode(struct super_block *sb)
241 if (sb->s_op->alloc_inode)
242 inode = sb->s_op->alloc_inode(sb);
244 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
249 if (unlikely(inode_init_always(sb, inode))) {
250 if (inode->i_sb->s_op->destroy_inode)
251 inode->i_sb->s_op->destroy_inode(inode);
253 kmem_cache_free(inode_cachep, inode);
260 void __destroy_inode(struct inode *inode)
262 BUG_ON(inode_has_buffers(inode));
263 security_inode_free(inode);
264 fsnotify_inode_delete(inode);
265 #ifdef CONFIG_FS_POSIX_ACL
266 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
267 posix_acl_release(inode->i_acl);
268 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
269 posix_acl_release(inode->i_default_acl);
271 this_cpu_dec(nr_inodes);
273 EXPORT_SYMBOL(__destroy_inode);
275 static void destroy_inode(struct inode *inode)
277 BUG_ON(!list_empty(&inode->i_lru));
278 __destroy_inode(inode);
279 if (inode->i_sb->s_op->destroy_inode)
280 inode->i_sb->s_op->destroy_inode(inode);
282 kmem_cache_free(inode_cachep, (inode));
286 * These are initializations that only need to be done
287 * once, because the fields are idempotent across use
288 * of the inode, so let the slab aware of that.
290 void inode_init_once(struct inode *inode)
292 memset(inode, 0, sizeof(*inode));
293 INIT_HLIST_NODE(&inode->i_hash);
294 INIT_LIST_HEAD(&inode->i_dentry);
295 INIT_LIST_HEAD(&inode->i_devices);
296 INIT_LIST_HEAD(&inode->i_wb_list);
297 INIT_LIST_HEAD(&inode->i_lru);
298 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
299 spin_lock_init(&inode->i_data.tree_lock);
300 spin_lock_init(&inode->i_data.i_mmap_lock);
301 INIT_LIST_HEAD(&inode->i_data.private_list);
302 spin_lock_init(&inode->i_data.private_lock);
303 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
304 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
305 i_size_ordered_init(inode);
306 #ifdef CONFIG_FSNOTIFY
307 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
310 EXPORT_SYMBOL(inode_init_once);
312 static void init_once(void *foo)
314 struct inode *inode = (struct inode *) foo;
316 inode_init_once(inode);
320 * inode_lock must be held
322 void __iget(struct inode *inode)
324 atomic_inc(&inode->i_count);
328 * get additional reference to inode; caller must already hold one.
330 void ihold(struct inode *inode)
332 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
334 EXPORT_SYMBOL(ihold);
336 static void inode_lru_list_add(struct inode *inode)
338 if (list_empty(&inode->i_lru)) {
339 list_add(&inode->i_lru, &inode_lru);
340 inodes_stat.nr_unused++;
344 static void inode_lru_list_del(struct inode *inode)
346 if (!list_empty(&inode->i_lru)) {
347 list_del_init(&inode->i_lru);
348 inodes_stat.nr_unused--;
352 static inline void __inode_sb_list_add(struct inode *inode)
354 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
358 * inode_sb_list_add - add inode to the superblock list of inodes
359 * @inode: inode to add
361 void inode_sb_list_add(struct inode *inode)
363 spin_lock(&inode_lock);
364 __inode_sb_list_add(inode);
365 spin_unlock(&inode_lock);
367 EXPORT_SYMBOL_GPL(inode_sb_list_add);
369 static inline void __inode_sb_list_del(struct inode *inode)
371 list_del_init(&inode->i_sb_list);
374 static unsigned long hash(struct super_block *sb, unsigned long hashval)
378 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
380 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
381 return tmp & I_HASHMASK;
385 * __insert_inode_hash - hash an inode
386 * @inode: unhashed inode
387 * @hashval: unsigned long value used to locate this object in the
390 * Add an inode to the inode hash for this superblock.
392 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
394 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
396 spin_lock(&inode_lock);
397 hlist_add_head(&inode->i_hash, b);
398 spin_unlock(&inode_lock);
400 EXPORT_SYMBOL(__insert_inode_hash);
403 * __remove_inode_hash - remove an inode from the hash
404 * @inode: inode to unhash
406 * Remove an inode from the superblock.
408 static void __remove_inode_hash(struct inode *inode)
410 hlist_del_init(&inode->i_hash);
414 * remove_inode_hash - remove an inode from the hash
415 * @inode: inode to unhash
417 * Remove an inode from the superblock.
419 void remove_inode_hash(struct inode *inode)
421 spin_lock(&inode_lock);
422 hlist_del_init(&inode->i_hash);
423 spin_unlock(&inode_lock);
425 EXPORT_SYMBOL(remove_inode_hash);
427 void end_writeback(struct inode *inode)
430 BUG_ON(inode->i_data.nrpages);
431 BUG_ON(!list_empty(&inode->i_data.private_list));
432 BUG_ON(!(inode->i_state & I_FREEING));
433 BUG_ON(inode->i_state & I_CLEAR);
434 inode_sync_wait(inode);
435 inode->i_state = I_FREEING | I_CLEAR;
437 EXPORT_SYMBOL(end_writeback);
439 static void evict(struct inode *inode)
441 const struct super_operations *op = inode->i_sb->s_op;
443 if (op->evict_inode) {
444 op->evict_inode(inode);
446 if (inode->i_data.nrpages)
447 truncate_inode_pages(&inode->i_data, 0);
448 end_writeback(inode);
450 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
452 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
457 * dispose_list - dispose of the contents of a local list
458 * @head: the head of the list to free
460 * Dispose-list gets a local list with local inodes in it, so it doesn't
461 * need to worry about list corruption and SMP locks.
463 static void dispose_list(struct list_head *head)
465 while (!list_empty(head)) {
468 inode = list_first_entry(head, struct inode, i_lru);
469 list_del_init(&inode->i_lru);
473 spin_lock(&inode_lock);
474 __remove_inode_hash(inode);
475 __inode_sb_list_del(inode);
476 spin_unlock(&inode_lock);
478 wake_up_inode(inode);
479 destroy_inode(inode);
484 * evict_inodes - evict all evictable inodes for a superblock
485 * @sb: superblock to operate on
487 * Make sure that no inodes with zero refcount are retained. This is
488 * called by superblock shutdown after having MS_ACTIVE flag removed,
489 * so any inode reaching zero refcount during or after that call will
490 * be immediately evicted.
492 void evict_inodes(struct super_block *sb)
494 struct inode *inode, *next;
497 down_write(&iprune_sem);
499 spin_lock(&inode_lock);
500 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
501 if (atomic_read(&inode->i_count))
504 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
509 inode->i_state |= I_FREEING;
512 * Move the inode off the IO lists and LRU once I_FREEING is
513 * set so that it won't get moved back on there if it is dirty.
515 list_move(&inode->i_lru, &dispose);
516 list_del_init(&inode->i_wb_list);
517 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
518 inodes_stat.nr_unused--;
520 spin_unlock(&inode_lock);
522 dispose_list(&dispose);
523 up_write(&iprune_sem);
527 * invalidate_inodes - attempt to free all inodes on a superblock
528 * @sb: superblock to operate on
530 * Attempts to free all inodes for a given superblock. If there were any
531 * busy inodes return a non-zero value, else zero.
533 int invalidate_inodes(struct super_block *sb)
536 struct inode *inode, *next;
539 down_write(&iprune_sem);
541 spin_lock(&inode_lock);
542 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
543 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE))
545 if (atomic_read(&inode->i_count)) {
550 inode->i_state |= I_FREEING;
553 * Move the inode off the IO lists and LRU once I_FREEING is
554 * set so that it won't get moved back on there if it is dirty.
556 list_move(&inode->i_lru, &dispose);
557 list_del_init(&inode->i_wb_list);
558 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
559 inodes_stat.nr_unused--;
561 spin_unlock(&inode_lock);
563 dispose_list(&dispose);
564 up_write(&iprune_sem);
569 static int can_unuse(struct inode *inode)
571 if (inode->i_state & ~I_REFERENCED)
573 if (inode_has_buffers(inode))
575 if (atomic_read(&inode->i_count))
577 if (inode->i_data.nrpages)
583 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
584 * temporary list and then are freed outside inode_lock by dispose_list().
586 * Any inodes which are pinned purely because of attached pagecache have their
587 * pagecache removed. If the inode has metadata buffers attached to
588 * mapping->private_list then try to remove them.
590 * If the inode has the I_REFERENCED flag set, then it means that it has been
591 * used recently - the flag is set in iput_final(). When we encounter such an
592 * inode, clear the flag and move it to the back of the LRU so it gets another
593 * pass through the LRU before it gets reclaimed. This is necessary because of
594 * the fact we are doing lazy LRU updates to minimise lock contention so the
595 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
596 * with this flag set because they are the inodes that are out of order.
598 static void prune_icache(int nr_to_scan)
602 unsigned long reap = 0;
604 down_read(&iprune_sem);
605 spin_lock(&inode_lock);
606 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
609 if (list_empty(&inode_lru))
612 inode = list_entry(inode_lru.prev, struct inode, i_lru);
615 * Referenced or dirty inodes are still in use. Give them
616 * another pass through the LRU as we canot reclaim them now.
618 if (atomic_read(&inode->i_count) ||
619 (inode->i_state & ~I_REFERENCED)) {
620 list_del_init(&inode->i_lru);
621 inodes_stat.nr_unused--;
625 /* recently referenced inodes get one more pass */
626 if (inode->i_state & I_REFERENCED) {
627 list_move(&inode->i_lru, &inode_lru);
628 inode->i_state &= ~I_REFERENCED;
631 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
633 spin_unlock(&inode_lock);
634 if (remove_inode_buffers(inode))
635 reap += invalidate_mapping_pages(&inode->i_data,
638 spin_lock(&inode_lock);
640 if (inode != list_entry(inode_lru.next,
641 struct inode, i_lru))
642 continue; /* wrong inode or list_empty */
643 if (!can_unuse(inode))
646 WARN_ON(inode->i_state & I_NEW);
647 inode->i_state |= I_FREEING;
650 * Move the inode off the IO lists and LRU once I_FREEING is
651 * set so that it won't get moved back on there if it is dirty.
653 list_move(&inode->i_lru, &freeable);
654 list_del_init(&inode->i_wb_list);
655 inodes_stat.nr_unused--;
657 if (current_is_kswapd())
658 __count_vm_events(KSWAPD_INODESTEAL, reap);
660 __count_vm_events(PGINODESTEAL, reap);
661 spin_unlock(&inode_lock);
663 dispose_list(&freeable);
664 up_read(&iprune_sem);
668 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
669 * "unused" means that no dentries are referring to the inodes: the files are
670 * not open and the dcache references to those inodes have already been
673 * This function is passed the number of inodes to scan, and it returns the
674 * total number of remaining possibly-reclaimable inodes.
676 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
680 * Nasty deadlock avoidance. We may hold various FS locks,
681 * and we don't want to recurse into the FS that called us
682 * in clear_inode() and friends..
684 if (!(gfp_mask & __GFP_FS))
688 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
691 static struct shrinker icache_shrinker = {
692 .shrink = shrink_icache_memory,
693 .seeks = DEFAULT_SEEKS,
696 static void __wait_on_freeing_inode(struct inode *inode);
698 * Called with the inode lock held.
700 static struct inode *find_inode(struct super_block *sb,
701 struct hlist_head *head,
702 int (*test)(struct inode *, void *),
705 struct hlist_node *node;
706 struct inode *inode = NULL;
709 hlist_for_each_entry(inode, node, head, i_hash) {
710 if (inode->i_sb != sb)
712 if (!test(inode, data))
714 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
715 __wait_on_freeing_inode(inode);
725 * find_inode_fast is the fast path version of find_inode, see the comment at
726 * iget_locked for details.
728 static struct inode *find_inode_fast(struct super_block *sb,
729 struct hlist_head *head, unsigned long ino)
731 struct hlist_node *node;
732 struct inode *inode = NULL;
735 hlist_for_each_entry(inode, node, head, i_hash) {
736 if (inode->i_ino != ino)
738 if (inode->i_sb != sb)
740 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
741 __wait_on_freeing_inode(inode);
751 * Each cpu owns a range of LAST_INO_BATCH numbers.
752 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
753 * to renew the exhausted range.
755 * This does not significantly increase overflow rate because every CPU can
756 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
757 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
758 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
759 * overflow rate by 2x, which does not seem too significant.
761 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
762 * error if st_ino won't fit in target struct field. Use 32bit counter
763 * here to attempt to avoid that.
765 #define LAST_INO_BATCH 1024
766 static DEFINE_PER_CPU(unsigned int, last_ino);
768 unsigned int get_next_ino(void)
770 unsigned int *p = &get_cpu_var(last_ino);
771 unsigned int res = *p;
774 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
775 static atomic_t shared_last_ino;
776 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
778 res = next - LAST_INO_BATCH;
783 put_cpu_var(last_ino);
786 EXPORT_SYMBOL(get_next_ino);
789 * new_inode - obtain an inode
792 * Allocates a new inode for given superblock. The default gfp_mask
793 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
794 * If HIGHMEM pages are unsuitable or it is known that pages allocated
795 * for the page cache are not reclaimable or migratable,
796 * mapping_set_gfp_mask() must be called with suitable flags on the
797 * newly created inode's mapping
800 struct inode *new_inode(struct super_block *sb)
804 spin_lock_prefetch(&inode_lock);
806 inode = alloc_inode(sb);
808 spin_lock(&inode_lock);
809 __inode_sb_list_add(inode);
811 spin_unlock(&inode_lock);
815 EXPORT_SYMBOL(new_inode);
817 void unlock_new_inode(struct inode *inode)
819 #ifdef CONFIG_DEBUG_LOCK_ALLOC
820 if (S_ISDIR(inode->i_mode)) {
821 struct file_system_type *type = inode->i_sb->s_type;
823 /* Set new key only if filesystem hasn't already changed it */
824 if (!lockdep_match_class(&inode->i_mutex,
825 &type->i_mutex_key)) {
827 * ensure nobody is actually holding i_mutex
829 mutex_destroy(&inode->i_mutex);
830 mutex_init(&inode->i_mutex);
831 lockdep_set_class(&inode->i_mutex,
832 &type->i_mutex_dir_key);
837 * This is special! We do not need the spinlock when clearing I_NEW,
838 * because we're guaranteed that nobody else tries to do anything about
839 * the state of the inode when it is locked, as we just created it (so
840 * there can be no old holders that haven't tested I_NEW).
841 * However we must emit the memory barrier so that other CPUs reliably
842 * see the clearing of I_NEW after the other inode initialisation has
846 WARN_ON(!(inode->i_state & I_NEW));
847 inode->i_state &= ~I_NEW;
848 wake_up_inode(inode);
850 EXPORT_SYMBOL(unlock_new_inode);
853 * This is called without the inode lock held.. Be careful.
855 * We no longer cache the sb_flags in i_flags - see fs.h
856 * -- rmk@arm.uk.linux.org
858 static struct inode *get_new_inode(struct super_block *sb,
859 struct hlist_head *head,
860 int (*test)(struct inode *, void *),
861 int (*set)(struct inode *, void *),
866 inode = alloc_inode(sb);
870 spin_lock(&inode_lock);
871 /* We released the lock, so.. */
872 old = find_inode(sb, head, test, data);
874 if (set(inode, data))
877 hlist_add_head(&inode->i_hash, head);
878 __inode_sb_list_add(inode);
879 inode->i_state = I_NEW;
880 spin_unlock(&inode_lock);
882 /* Return the locked inode with I_NEW set, the
883 * caller is responsible for filling in the contents
889 * Uhhuh, somebody else created the same inode under
890 * us. Use the old inode instead of the one we just
893 spin_unlock(&inode_lock);
894 destroy_inode(inode);
896 wait_on_inode(inode);
901 spin_unlock(&inode_lock);
902 destroy_inode(inode);
907 * get_new_inode_fast is the fast path version of get_new_inode, see the
908 * comment at iget_locked for details.
910 static struct inode *get_new_inode_fast(struct super_block *sb,
911 struct hlist_head *head, unsigned long ino)
915 inode = alloc_inode(sb);
919 spin_lock(&inode_lock);
920 /* We released the lock, so.. */
921 old = find_inode_fast(sb, head, ino);
924 hlist_add_head(&inode->i_hash, head);
925 __inode_sb_list_add(inode);
926 inode->i_state = I_NEW;
927 spin_unlock(&inode_lock);
929 /* Return the locked inode with I_NEW set, the
930 * caller is responsible for filling in the contents
936 * Uhhuh, somebody else created the same inode under
937 * us. Use the old inode instead of the one we just
940 spin_unlock(&inode_lock);
941 destroy_inode(inode);
943 wait_on_inode(inode);
949 * search the inode cache for a matching inode number.
950 * If we find one, then the inode number we are trying to
951 * allocate is not unique and so we should not use it.
953 * Returns 1 if the inode number is unique, 0 if it is not.
955 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
957 struct hlist_head *b = inode_hashtable + hash(sb, ino);
958 struct hlist_node *node;
961 hlist_for_each_entry(inode, node, b, i_hash) {
962 if (inode->i_ino == ino && inode->i_sb == sb)
970 * iunique - get a unique inode number
972 * @max_reserved: highest reserved inode number
974 * Obtain an inode number that is unique on the system for a given
975 * superblock. This is used by file systems that have no natural
976 * permanent inode numbering system. An inode number is returned that
977 * is higher than the reserved limit but unique.
980 * With a large number of inodes live on the file system this function
981 * currently becomes quite slow.
983 ino_t iunique(struct super_block *sb, ino_t max_reserved)
986 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
987 * error if st_ino won't fit in target struct field. Use 32bit counter
988 * here to attempt to avoid that.
990 static DEFINE_SPINLOCK(iunique_lock);
991 static unsigned int counter;
994 spin_lock(&inode_lock);
995 spin_lock(&iunique_lock);
997 if (counter <= max_reserved)
998 counter = max_reserved + 1;
1000 } while (!test_inode_iunique(sb, res));
1001 spin_unlock(&iunique_lock);
1002 spin_unlock(&inode_lock);
1006 EXPORT_SYMBOL(iunique);
1008 struct inode *igrab(struct inode *inode)
1010 spin_lock(&inode_lock);
1011 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
1015 * Handle the case where s_op->clear_inode is not been
1016 * called yet, and somebody is calling igrab
1017 * while the inode is getting freed.
1020 spin_unlock(&inode_lock);
1023 EXPORT_SYMBOL(igrab);
1026 * ifind - internal function, you want ilookup5() or iget5().
1027 * @sb: super block of file system to search
1028 * @head: the head of the list to search
1029 * @test: callback used for comparisons between inodes
1030 * @data: opaque data pointer to pass to @test
1031 * @wait: if true wait for the inode to be unlocked, if false do not
1033 * ifind() searches for the inode specified by @data in the inode
1034 * cache. This is a generalized version of ifind_fast() for file systems where
1035 * the inode number is not sufficient for unique identification of an inode.
1037 * If the inode is in the cache, the inode is returned with an incremented
1040 * Otherwise NULL is returned.
1042 * Note, @test is called with the inode_lock held, so can't sleep.
1044 static struct inode *ifind(struct super_block *sb,
1045 struct hlist_head *head, int (*test)(struct inode *, void *),
1046 void *data, const int wait)
1048 struct inode *inode;
1050 spin_lock(&inode_lock);
1051 inode = find_inode(sb, head, test, data);
1053 spin_unlock(&inode_lock);
1055 wait_on_inode(inode);
1058 spin_unlock(&inode_lock);
1063 * ifind_fast - internal function, you want ilookup() or iget().
1064 * @sb: super block of file system to search
1065 * @head: head of the list to search
1066 * @ino: inode number to search for
1068 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1069 * file systems where the inode number is sufficient for unique identification
1072 * If the inode is in the cache, the inode is returned with an incremented
1075 * Otherwise NULL is returned.
1077 static struct inode *ifind_fast(struct super_block *sb,
1078 struct hlist_head *head, unsigned long ino)
1080 struct inode *inode;
1082 spin_lock(&inode_lock);
1083 inode = find_inode_fast(sb, head, ino);
1085 spin_unlock(&inode_lock);
1086 wait_on_inode(inode);
1089 spin_unlock(&inode_lock);
1094 * ilookup5_nowait - search for an inode in the inode cache
1095 * @sb: super block of file system to search
1096 * @hashval: hash value (usually inode number) to search for
1097 * @test: callback used for comparisons between inodes
1098 * @data: opaque data pointer to pass to @test
1100 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1101 * @data in the inode cache. This is a generalized version of ilookup() for
1102 * file systems where the inode number is not sufficient for unique
1103 * identification of an inode.
1105 * If the inode is in the cache, the inode is returned with an incremented
1106 * reference count. Note, the inode lock is not waited upon so you have to be
1107 * very careful what you do with the returned inode. You probably should be
1108 * using ilookup5() instead.
1110 * Otherwise NULL is returned.
1112 * Note, @test is called with the inode_lock held, so can't sleep.
1114 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1115 int (*test)(struct inode *, void *), void *data)
1117 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1119 return ifind(sb, head, test, data, 0);
1121 EXPORT_SYMBOL(ilookup5_nowait);
1124 * ilookup5 - search for an inode in the inode cache
1125 * @sb: super block of file system to search
1126 * @hashval: hash value (usually inode number) to search for
1127 * @test: callback used for comparisons between inodes
1128 * @data: opaque data pointer to pass to @test
1130 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1131 * @data in the inode cache. This is a generalized version of ilookup() for
1132 * file systems where the inode number is not sufficient for unique
1133 * identification of an inode.
1135 * If the inode is in the cache, the inode lock is waited upon and the inode is
1136 * returned with an incremented reference count.
1138 * Otherwise NULL is returned.
1140 * Note, @test is called with the inode_lock held, so can't sleep.
1142 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1143 int (*test)(struct inode *, void *), void *data)
1145 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1147 return ifind(sb, head, test, data, 1);
1149 EXPORT_SYMBOL(ilookup5);
1152 * ilookup - search for an inode in the inode cache
1153 * @sb: super block of file system to search
1154 * @ino: inode number to search for
1156 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1157 * This is for file systems where the inode number is sufficient for unique
1158 * identification of an inode.
1160 * If the inode is in the cache, the inode is returned with an incremented
1163 * Otherwise NULL is returned.
1165 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1167 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1169 return ifind_fast(sb, head, ino);
1171 EXPORT_SYMBOL(ilookup);
1174 * iget5_locked - obtain an inode from a mounted file system
1175 * @sb: super block of file system
1176 * @hashval: hash value (usually inode number) to get
1177 * @test: callback used for comparisons between inodes
1178 * @set: callback used to initialize a new struct inode
1179 * @data: opaque data pointer to pass to @test and @set
1181 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1182 * and @data in the inode cache and if present it is returned with an increased
1183 * reference count. This is a generalized version of iget_locked() for file
1184 * systems where the inode number is not sufficient for unique identification
1187 * If the inode is not in cache, get_new_inode() is called to allocate a new
1188 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1189 * file system gets to fill it in before unlocking it via unlock_new_inode().
1191 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1193 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1194 int (*test)(struct inode *, void *),
1195 int (*set)(struct inode *, void *), void *data)
1197 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1198 struct inode *inode;
1200 inode = ifind(sb, head, test, data, 1);
1204 * get_new_inode() will do the right thing, re-trying the search
1205 * in case it had to block at any point.
1207 return get_new_inode(sb, head, test, set, data);
1209 EXPORT_SYMBOL(iget5_locked);
1212 * iget_locked - obtain an inode from a mounted file system
1213 * @sb: super block of file system
1214 * @ino: inode number to get
1216 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1217 * the inode cache and if present it is returned with an increased reference
1218 * count. This is for file systems where the inode number is sufficient for
1219 * unique identification of an inode.
1221 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1222 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1223 * The file system gets to fill it in before unlocking it via
1224 * unlock_new_inode().
1226 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1228 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1229 struct inode *inode;
1231 inode = ifind_fast(sb, head, ino);
1235 * get_new_inode_fast() will do the right thing, re-trying the search
1236 * in case it had to block at any point.
1238 return get_new_inode_fast(sb, head, ino);
1240 EXPORT_SYMBOL(iget_locked);
1242 int insert_inode_locked(struct inode *inode)
1244 struct super_block *sb = inode->i_sb;
1245 ino_t ino = inode->i_ino;
1246 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1248 inode->i_state |= I_NEW;
1250 struct hlist_node *node;
1251 struct inode *old = NULL;
1252 spin_lock(&inode_lock);
1253 hlist_for_each_entry(old, node, head, i_hash) {
1254 if (old->i_ino != ino)
1256 if (old->i_sb != sb)
1258 if (old->i_state & (I_FREEING|I_WILL_FREE))
1262 if (likely(!node)) {
1263 hlist_add_head(&inode->i_hash, head);
1264 spin_unlock(&inode_lock);
1268 spin_unlock(&inode_lock);
1270 if (unlikely(!inode_unhashed(old))) {
1277 EXPORT_SYMBOL(insert_inode_locked);
1279 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1280 int (*test)(struct inode *, void *), void *data)
1282 struct super_block *sb = inode->i_sb;
1283 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1285 inode->i_state |= I_NEW;
1288 struct hlist_node *node;
1289 struct inode *old = NULL;
1291 spin_lock(&inode_lock);
1292 hlist_for_each_entry(old, node, head, i_hash) {
1293 if (old->i_sb != sb)
1295 if (!test(old, data))
1297 if (old->i_state & (I_FREEING|I_WILL_FREE))
1301 if (likely(!node)) {
1302 hlist_add_head(&inode->i_hash, head);
1303 spin_unlock(&inode_lock);
1307 spin_unlock(&inode_lock);
1309 if (unlikely(!inode_unhashed(old))) {
1316 EXPORT_SYMBOL(insert_inode_locked4);
1319 int generic_delete_inode(struct inode *inode)
1323 EXPORT_SYMBOL(generic_delete_inode);
1326 * Normal UNIX filesystem behaviour: delete the
1327 * inode when the usage count drops to zero, and
1330 int generic_drop_inode(struct inode *inode)
1332 return !inode->i_nlink || inode_unhashed(inode);
1334 EXPORT_SYMBOL_GPL(generic_drop_inode);
1337 * Called when we're dropping the last reference
1340 * Call the FS "drop_inode()" function, defaulting to
1341 * the legacy UNIX filesystem behaviour. If it tells
1342 * us to evict inode, do so. Otherwise, retain inode
1343 * in cache if fs is alive, sync and evict if fs is
1346 static void iput_final(struct inode *inode)
1348 struct super_block *sb = inode->i_sb;
1349 const struct super_operations *op = inode->i_sb->s_op;
1352 if (op && op->drop_inode)
1353 drop = op->drop_inode(inode);
1355 drop = generic_drop_inode(inode);
1358 if (sb->s_flags & MS_ACTIVE) {
1359 inode->i_state |= I_REFERENCED;
1360 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1361 inode_lru_list_add(inode);
1363 spin_unlock(&inode_lock);
1366 WARN_ON(inode->i_state & I_NEW);
1367 inode->i_state |= I_WILL_FREE;
1368 spin_unlock(&inode_lock);
1369 write_inode_now(inode, 1);
1370 spin_lock(&inode_lock);
1371 WARN_ON(inode->i_state & I_NEW);
1372 inode->i_state &= ~I_WILL_FREE;
1373 __remove_inode_hash(inode);
1376 WARN_ON(inode->i_state & I_NEW);
1377 inode->i_state |= I_FREEING;
1380 * Move the inode off the IO lists and LRU once I_FREEING is
1381 * set so that it won't get moved back on there if it is dirty.
1383 inode_lru_list_del(inode);
1384 list_del_init(&inode->i_wb_list);
1386 __inode_sb_list_del(inode);
1387 spin_unlock(&inode_lock);
1389 remove_inode_hash(inode);
1390 wake_up_inode(inode);
1391 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1392 destroy_inode(inode);
1396 * iput - put an inode
1397 * @inode: inode to put
1399 * Puts an inode, dropping its usage count. If the inode use count hits
1400 * zero, the inode is then freed and may also be destroyed.
1402 * Consequently, iput() can sleep.
1404 void iput(struct inode *inode)
1407 BUG_ON(inode->i_state & I_CLEAR);
1409 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1413 EXPORT_SYMBOL(iput);
1416 * bmap - find a block number in a file
1417 * @inode: inode of file
1418 * @block: block to find
1420 * Returns the block number on the device holding the inode that
1421 * is the disk block number for the block of the file requested.
1422 * That is, asked for block 4 of inode 1 the function will return the
1423 * disk block relative to the disk start that holds that block of the
1426 sector_t bmap(struct inode *inode, sector_t block)
1429 if (inode->i_mapping->a_ops->bmap)
1430 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1433 EXPORT_SYMBOL(bmap);
1436 * With relative atime, only update atime if the previous atime is
1437 * earlier than either the ctime or mtime or if at least a day has
1438 * passed since the last atime update.
1440 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1441 struct timespec now)
1444 if (!(mnt->mnt_flags & MNT_RELATIME))
1447 * Is mtime younger than atime? If yes, update atime:
1449 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1452 * Is ctime younger than atime? If yes, update atime:
1454 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1458 * Is the previous atime value older than a day? If yes,
1461 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1464 * Good, we can skip the atime update:
1470 * touch_atime - update the access time
1471 * @mnt: mount the inode is accessed on
1472 * @dentry: dentry accessed
1474 * Update the accessed time on an inode and mark it for writeback.
1475 * This function automatically handles read only file systems and media,
1476 * as well as the "noatime" flag and inode specific "noatime" markers.
1478 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1480 struct inode *inode = dentry->d_inode;
1481 struct timespec now;
1483 if (inode->i_flags & S_NOATIME)
1485 if (IS_NOATIME(inode))
1487 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1490 if (mnt->mnt_flags & MNT_NOATIME)
1492 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1495 now = current_fs_time(inode->i_sb);
1497 if (!relatime_need_update(mnt, inode, now))
1500 if (timespec_equal(&inode->i_atime, &now))
1503 if (mnt_want_write(mnt))
1506 inode->i_atime = now;
1507 mark_inode_dirty_sync(inode);
1508 mnt_drop_write(mnt);
1510 EXPORT_SYMBOL(touch_atime);
1513 * file_update_time - update mtime and ctime time
1514 * @file: file accessed
1516 * Update the mtime and ctime members of an inode and mark the inode
1517 * for writeback. Note that this function is meant exclusively for
1518 * usage in the file write path of filesystems, and filesystems may
1519 * choose to explicitly ignore update via this function with the
1520 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1521 * timestamps are handled by the server.
1524 void file_update_time(struct file *file)
1526 struct inode *inode = file->f_path.dentry->d_inode;
1527 struct timespec now;
1528 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1530 /* First try to exhaust all avenues to not sync */
1531 if (IS_NOCMTIME(inode))
1534 now = current_fs_time(inode->i_sb);
1535 if (!timespec_equal(&inode->i_mtime, &now))
1538 if (!timespec_equal(&inode->i_ctime, &now))
1541 if (IS_I_VERSION(inode))
1542 sync_it |= S_VERSION;
1547 /* Finally allowed to write? Takes lock. */
1548 if (mnt_want_write_file(file))
1551 /* Only change inode inside the lock region */
1552 if (sync_it & S_VERSION)
1553 inode_inc_iversion(inode);
1554 if (sync_it & S_CTIME)
1555 inode->i_ctime = now;
1556 if (sync_it & S_MTIME)
1557 inode->i_mtime = now;
1558 mark_inode_dirty_sync(inode);
1559 mnt_drop_write(file->f_path.mnt);
1561 EXPORT_SYMBOL(file_update_time);
1563 int inode_needs_sync(struct inode *inode)
1567 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1571 EXPORT_SYMBOL(inode_needs_sync);
1573 int inode_wait(void *word)
1578 EXPORT_SYMBOL(inode_wait);
1581 * If we try to find an inode in the inode hash while it is being
1582 * deleted, we have to wait until the filesystem completes its
1583 * deletion before reporting that it isn't found. This function waits
1584 * until the deletion _might_ have completed. Callers are responsible
1585 * to recheck inode state.
1587 * It doesn't matter if I_NEW is not set initially, a call to
1588 * wake_up_inode() after removing from the hash list will DTRT.
1590 * This is called with inode_lock held.
1592 static void __wait_on_freeing_inode(struct inode *inode)
1594 wait_queue_head_t *wq;
1595 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1596 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1597 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1598 spin_unlock(&inode_lock);
1600 finish_wait(wq, &wait.wait);
1601 spin_lock(&inode_lock);
1604 static __initdata unsigned long ihash_entries;
1605 static int __init set_ihash_entries(char *str)
1609 ihash_entries = simple_strtoul(str, &str, 0);
1612 __setup("ihash_entries=", set_ihash_entries);
1615 * Initialize the waitqueues and inode hash table.
1617 void __init inode_init_early(void)
1621 /* If hashes are distributed across NUMA nodes, defer
1622 * hash allocation until vmalloc space is available.
1628 alloc_large_system_hash("Inode-cache",
1629 sizeof(struct hlist_head),
1637 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1638 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1641 void __init inode_init(void)
1645 /* inode slab cache */
1646 inode_cachep = kmem_cache_create("inode_cache",
1647 sizeof(struct inode),
1649 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1652 register_shrinker(&icache_shrinker);
1654 /* Hash may have been set up in inode_init_early */
1659 alloc_large_system_hash("Inode-cache",
1660 sizeof(struct hlist_head),
1668 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1669 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1672 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1674 inode->i_mode = mode;
1675 if (S_ISCHR(mode)) {
1676 inode->i_fop = &def_chr_fops;
1677 inode->i_rdev = rdev;
1678 } else if (S_ISBLK(mode)) {
1679 inode->i_fop = &def_blk_fops;
1680 inode->i_rdev = rdev;
1681 } else if (S_ISFIFO(mode))
1682 inode->i_fop = &def_fifo_fops;
1683 else if (S_ISSOCK(mode))
1684 inode->i_fop = &bad_sock_fops;
1686 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1687 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1690 EXPORT_SYMBOL(init_special_inode);
1693 * Init uid,gid,mode for new inode according to posix standards
1695 * @dir: Directory inode
1696 * @mode: mode of the new inode
1698 void inode_init_owner(struct inode *inode, const struct inode *dir,
1701 inode->i_uid = current_fsuid();
1702 if (dir && dir->i_mode & S_ISGID) {
1703 inode->i_gid = dir->i_gid;
1707 inode->i_gid = current_fsgid();
1708 inode->i_mode = mode;
1710 EXPORT_SYMBOL(inode_init_owner);