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);
85 EXPORT_SYMBOL(inode_lock);
88 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
89 * icache shrinking path, and the umount path. Without this exclusion,
90 * by the time prune_icache calls iput for the inode whose pages it has
91 * been invalidating, or by the time it calls clear_inode & destroy_inode
92 * from its final dispose_list, the struct super_block they refer to
93 * (for inode->i_sb->s_op) may already have been freed and reused.
95 * We make this an rwsem because the fastpath is icache shrinking. In
96 * some cases a filesystem may be doing a significant amount of work in
97 * its inode reclaim code, so this should improve parallelism.
99 static DECLARE_RWSEM(iprune_sem);
102 * Statistics gathering..
104 struct inodes_stat_t inodes_stat;
106 static DEFINE_PER_CPU(unsigned int, nr_inodes);
108 static struct kmem_cache *inode_cachep __read_mostly;
110 static int get_nr_inodes(void)
114 for_each_possible_cpu(i)
115 sum += per_cpu(nr_inodes, i);
116 return sum < 0 ? 0 : sum;
119 static inline int get_nr_inodes_unused(void)
121 return inodes_stat.nr_unused;
124 int get_nr_dirty_inodes(void)
126 /* not actually dirty inodes, but a wild approximation */
127 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
128 return nr_dirty > 0 ? nr_dirty : 0;
132 * Handle nr_inode sysctl
135 int proc_nr_inodes(ctl_table *table, int write,
136 void __user *buffer, size_t *lenp, loff_t *ppos)
138 inodes_stat.nr_inodes = get_nr_inodes();
139 return proc_dointvec(table, write, buffer, lenp, ppos);
143 static void wake_up_inode(struct inode *inode)
146 * Prevent speculative execution through spin_unlock(&inode_lock);
149 wake_up_bit(&inode->i_state, __I_NEW);
153 * inode_init_always - perform inode structure intialisation
154 * @sb: superblock inode belongs to
155 * @inode: inode to initialise
157 * These are initializations that need to be done on every inode
158 * allocation as the fields are not initialised by slab allocation.
160 int inode_init_always(struct super_block *sb, struct inode *inode)
162 static const struct address_space_operations empty_aops;
163 static const struct inode_operations empty_iops;
164 static const struct file_operations empty_fops;
165 struct address_space *const mapping = &inode->i_data;
168 inode->i_blkbits = sb->s_blocksize_bits;
170 atomic_set(&inode->i_count, 1);
171 inode->i_op = &empty_iops;
172 inode->i_fop = &empty_fops;
176 atomic_set(&inode->i_writecount, 0);
180 inode->i_generation = 0;
182 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
184 inode->i_pipe = NULL;
185 inode->i_bdev = NULL;
186 inode->i_cdev = NULL;
188 inode->dirtied_when = 0;
190 if (security_inode_alloc(inode))
192 spin_lock_init(&inode->i_lock);
193 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
195 mutex_init(&inode->i_mutex);
196 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
198 init_rwsem(&inode->i_alloc_sem);
199 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
201 mapping->a_ops = &empty_aops;
202 mapping->host = inode;
204 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
205 mapping->assoc_mapping = NULL;
206 mapping->backing_dev_info = &default_backing_dev_info;
207 mapping->writeback_index = 0;
210 * If the block_device provides a backing_dev_info for client
211 * inodes then use that. Otherwise the inode share the bdev's
215 struct backing_dev_info *bdi;
217 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
218 mapping->backing_dev_info = bdi;
220 inode->i_private = NULL;
221 inode->i_mapping = mapping;
222 #ifdef CONFIG_FS_POSIX_ACL
223 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
226 #ifdef CONFIG_FSNOTIFY
227 inode->i_fsnotify_mask = 0;
230 this_cpu_inc(nr_inodes);
236 EXPORT_SYMBOL(inode_init_always);
238 static struct inode *alloc_inode(struct super_block *sb)
242 if (sb->s_op->alloc_inode)
243 inode = sb->s_op->alloc_inode(sb);
245 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
250 if (unlikely(inode_init_always(sb, inode))) {
251 if (inode->i_sb->s_op->destroy_inode)
252 inode->i_sb->s_op->destroy_inode(inode);
254 kmem_cache_free(inode_cachep, inode);
261 void free_inode_nonrcu(struct inode *inode)
263 kmem_cache_free(inode_cachep, inode);
265 EXPORT_SYMBOL(free_inode_nonrcu);
267 void __destroy_inode(struct inode *inode)
269 BUG_ON(inode_has_buffers(inode));
270 security_inode_free(inode);
271 fsnotify_inode_delete(inode);
272 #ifdef CONFIG_FS_POSIX_ACL
273 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
274 posix_acl_release(inode->i_acl);
275 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
276 posix_acl_release(inode->i_default_acl);
278 this_cpu_dec(nr_inodes);
280 EXPORT_SYMBOL(__destroy_inode);
282 static void i_callback(struct rcu_head *head)
284 struct inode *inode = container_of(head, struct inode, i_rcu);
285 INIT_LIST_HEAD(&inode->i_dentry);
286 kmem_cache_free(inode_cachep, inode);
289 static void destroy_inode(struct inode *inode)
291 BUG_ON(!list_empty(&inode->i_lru));
292 __destroy_inode(inode);
293 if (inode->i_sb->s_op->destroy_inode)
294 inode->i_sb->s_op->destroy_inode(inode);
296 call_rcu(&inode->i_rcu, i_callback);
299 void address_space_init_once(struct address_space *mapping)
301 memset(mapping, 0, sizeof(*mapping));
302 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
303 spin_lock_init(&mapping->tree_lock);
304 spin_lock_init(&mapping->i_mmap_lock);
305 INIT_LIST_HEAD(&mapping->private_list);
306 spin_lock_init(&mapping->private_lock);
307 INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
308 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
309 mutex_init(&mapping->unmap_mutex);
311 EXPORT_SYMBOL(address_space_init_once);
314 * These are initializations that only need to be done
315 * once, because the fields are idempotent across use
316 * of the inode, so let the slab aware of that.
318 void inode_init_once(struct inode *inode)
320 memset(inode, 0, sizeof(*inode));
321 INIT_HLIST_NODE(&inode->i_hash);
322 INIT_LIST_HEAD(&inode->i_dentry);
323 INIT_LIST_HEAD(&inode->i_devices);
324 INIT_LIST_HEAD(&inode->i_wb_list);
325 INIT_LIST_HEAD(&inode->i_lru);
326 address_space_init_once(&inode->i_data);
327 i_size_ordered_init(inode);
328 #ifdef CONFIG_FSNOTIFY
329 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
332 EXPORT_SYMBOL(inode_init_once);
334 static void init_once(void *foo)
336 struct inode *inode = (struct inode *) foo;
338 inode_init_once(inode);
342 * inode_lock must be held
344 void __iget(struct inode *inode)
346 atomic_inc(&inode->i_count);
350 * get additional reference to inode; caller must already hold one.
352 void ihold(struct inode *inode)
354 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
356 EXPORT_SYMBOL(ihold);
358 static void inode_lru_list_add(struct inode *inode)
360 if (list_empty(&inode->i_lru)) {
361 list_add(&inode->i_lru, &inode_lru);
362 inodes_stat.nr_unused++;
366 static void inode_lru_list_del(struct inode *inode)
368 if (!list_empty(&inode->i_lru)) {
369 list_del_init(&inode->i_lru);
370 inodes_stat.nr_unused--;
374 static inline void __inode_sb_list_add(struct inode *inode)
376 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
380 * inode_sb_list_add - add inode to the superblock list of inodes
381 * @inode: inode to add
383 void inode_sb_list_add(struct inode *inode)
385 spin_lock(&inode_lock);
386 __inode_sb_list_add(inode);
387 spin_unlock(&inode_lock);
389 EXPORT_SYMBOL_GPL(inode_sb_list_add);
391 static inline void __inode_sb_list_del(struct inode *inode)
393 list_del_init(&inode->i_sb_list);
396 static unsigned long hash(struct super_block *sb, unsigned long hashval)
400 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
402 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
403 return tmp & I_HASHMASK;
407 * __insert_inode_hash - hash an inode
408 * @inode: unhashed inode
409 * @hashval: unsigned long value used to locate this object in the
412 * Add an inode to the inode hash for this superblock.
414 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
416 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
418 spin_lock(&inode_lock);
419 hlist_add_head(&inode->i_hash, b);
420 spin_unlock(&inode_lock);
422 EXPORT_SYMBOL(__insert_inode_hash);
425 * __remove_inode_hash - remove an inode from the hash
426 * @inode: inode to unhash
428 * Remove an inode from the superblock.
430 static void __remove_inode_hash(struct inode *inode)
432 hlist_del_init(&inode->i_hash);
436 * remove_inode_hash - remove an inode from the hash
437 * @inode: inode to unhash
439 * Remove an inode from the superblock.
441 void remove_inode_hash(struct inode *inode)
443 spin_lock(&inode_lock);
444 hlist_del_init(&inode->i_hash);
445 spin_unlock(&inode_lock);
447 EXPORT_SYMBOL(remove_inode_hash);
449 void end_writeback(struct inode *inode)
452 BUG_ON(inode->i_data.nrpages);
453 BUG_ON(!list_empty(&inode->i_data.private_list));
454 BUG_ON(!(inode->i_state & I_FREEING));
455 BUG_ON(inode->i_state & I_CLEAR);
456 inode_sync_wait(inode);
457 /* don't need i_lock here, no concurrent mods to i_state */
458 inode->i_state = I_FREEING | I_CLEAR;
460 EXPORT_SYMBOL(end_writeback);
462 static void evict(struct inode *inode)
464 const struct super_operations *op = inode->i_sb->s_op;
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)
480 * dispose_list - dispose of the contents of a local list
481 * @head: the head of the list to free
483 * Dispose-list gets a local list with local inodes in it, so it doesn't
484 * need to worry about list corruption and SMP locks.
486 static void dispose_list(struct list_head *head)
488 while (!list_empty(head)) {
491 inode = list_first_entry(head, struct inode, i_lru);
492 list_del_init(&inode->i_lru);
496 spin_lock(&inode_lock);
497 __remove_inode_hash(inode);
498 __inode_sb_list_del(inode);
499 spin_unlock(&inode_lock);
501 wake_up_inode(inode);
502 destroy_inode(inode);
507 * evict_inodes - evict all evictable inodes for a superblock
508 * @sb: superblock to operate on
510 * Make sure that no inodes with zero refcount are retained. This is
511 * called by superblock shutdown after having MS_ACTIVE flag removed,
512 * so any inode reaching zero refcount during or after that call will
513 * be immediately evicted.
515 void evict_inodes(struct super_block *sb)
517 struct inode *inode, *next;
520 down_write(&iprune_sem);
522 spin_lock(&inode_lock);
523 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
524 if (atomic_read(&inode->i_count))
527 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
532 inode->i_state |= I_FREEING;
535 * Move the inode off the IO lists and LRU once I_FREEING is
536 * set so that it won't get moved back on there if it is dirty.
538 list_move(&inode->i_lru, &dispose);
539 list_del_init(&inode->i_wb_list);
540 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
541 inodes_stat.nr_unused--;
543 spin_unlock(&inode_lock);
545 dispose_list(&dispose);
546 up_write(&iprune_sem);
550 * invalidate_inodes - attempt to free all inodes on a superblock
551 * @sb: superblock to operate on
552 * @kill_dirty: flag to guide handling of dirty inodes
554 * Attempts to free all inodes for a given superblock. If there were any
555 * busy inodes return a non-zero value, else zero.
556 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
559 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
562 struct inode *inode, *next;
565 down_write(&iprune_sem);
567 spin_lock(&inode_lock);
568 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
569 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE))
571 if (inode->i_state & I_DIRTY && !kill_dirty) {
575 if (atomic_read(&inode->i_count)) {
580 inode->i_state |= I_FREEING;
583 * Move the inode off the IO lists and LRU once I_FREEING is
584 * set so that it won't get moved back on there if it is dirty.
586 list_move(&inode->i_lru, &dispose);
587 list_del_init(&inode->i_wb_list);
588 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
589 inodes_stat.nr_unused--;
591 spin_unlock(&inode_lock);
593 dispose_list(&dispose);
594 up_write(&iprune_sem);
599 static int can_unuse(struct inode *inode)
601 if (inode->i_state & ~I_REFERENCED)
603 if (inode_has_buffers(inode))
605 if (atomic_read(&inode->i_count))
607 if (inode->i_data.nrpages)
613 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
614 * temporary list and then are freed outside inode_lock by dispose_list().
616 * Any inodes which are pinned purely because of attached pagecache have their
617 * pagecache removed. If the inode has metadata buffers attached to
618 * mapping->private_list then try to remove them.
620 * If the inode has the I_REFERENCED flag set, then it means that it has been
621 * used recently - the flag is set in iput_final(). When we encounter such an
622 * inode, clear the flag and move it to the back of the LRU so it gets another
623 * pass through the LRU before it gets reclaimed. This is necessary because of
624 * the fact we are doing lazy LRU updates to minimise lock contention so the
625 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
626 * with this flag set because they are the inodes that are out of order.
628 static void prune_icache(int nr_to_scan)
632 unsigned long reap = 0;
634 down_read(&iprune_sem);
635 spin_lock(&inode_lock);
636 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
639 if (list_empty(&inode_lru))
642 inode = list_entry(inode_lru.prev, struct inode, i_lru);
645 * Referenced or dirty inodes are still in use. Give them
646 * another pass through the LRU as we canot reclaim them now.
648 if (atomic_read(&inode->i_count) ||
649 (inode->i_state & ~I_REFERENCED)) {
650 list_del_init(&inode->i_lru);
651 inodes_stat.nr_unused--;
655 /* recently referenced inodes get one more pass */
656 if (inode->i_state & I_REFERENCED) {
657 list_move(&inode->i_lru, &inode_lru);
658 inode->i_state &= ~I_REFERENCED;
661 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
663 spin_unlock(&inode_lock);
664 if (remove_inode_buffers(inode))
665 reap += invalidate_mapping_pages(&inode->i_data,
668 spin_lock(&inode_lock);
670 if (inode != list_entry(inode_lru.next,
671 struct inode, i_lru))
672 continue; /* wrong inode or list_empty */
673 if (!can_unuse(inode))
676 WARN_ON(inode->i_state & I_NEW);
677 inode->i_state |= I_FREEING;
680 * Move the inode off the IO lists and LRU once I_FREEING is
681 * set so that it won't get moved back on there if it is dirty.
683 list_move(&inode->i_lru, &freeable);
684 list_del_init(&inode->i_wb_list);
685 inodes_stat.nr_unused--;
687 if (current_is_kswapd())
688 __count_vm_events(KSWAPD_INODESTEAL, reap);
690 __count_vm_events(PGINODESTEAL, reap);
691 spin_unlock(&inode_lock);
693 dispose_list(&freeable);
694 up_read(&iprune_sem);
698 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
699 * "unused" means that no dentries are referring to the inodes: the files are
700 * not open and the dcache references to those inodes have already been
703 * This function is passed the number of inodes to scan, and it returns the
704 * total number of remaining possibly-reclaimable inodes.
706 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
710 * Nasty deadlock avoidance. We may hold various FS locks,
711 * and we don't want to recurse into the FS that called us
712 * in clear_inode() and friends..
714 if (!(gfp_mask & __GFP_FS))
718 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
721 static struct shrinker icache_shrinker = {
722 .shrink = shrink_icache_memory,
723 .seeks = DEFAULT_SEEKS,
726 static void __wait_on_freeing_inode(struct inode *inode);
728 * Called with the inode lock held.
730 static struct inode *find_inode(struct super_block *sb,
731 struct hlist_head *head,
732 int (*test)(struct inode *, void *),
735 struct hlist_node *node;
736 struct inode *inode = NULL;
739 hlist_for_each_entry(inode, node, head, i_hash) {
740 if (inode->i_sb != sb)
742 if (!test(inode, data))
744 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
745 __wait_on_freeing_inode(inode);
755 * find_inode_fast is the fast path version of find_inode, see the comment at
756 * iget_locked for details.
758 static struct inode *find_inode_fast(struct super_block *sb,
759 struct hlist_head *head, unsigned long ino)
761 struct hlist_node *node;
762 struct inode *inode = NULL;
765 hlist_for_each_entry(inode, node, head, i_hash) {
766 if (inode->i_ino != ino)
768 if (inode->i_sb != sb)
770 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
771 __wait_on_freeing_inode(inode);
781 * Each cpu owns a range of LAST_INO_BATCH numbers.
782 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
783 * to renew the exhausted range.
785 * This does not significantly increase overflow rate because every CPU can
786 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
787 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
788 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
789 * overflow rate by 2x, which does not seem too significant.
791 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
792 * error if st_ino won't fit in target struct field. Use 32bit counter
793 * here to attempt to avoid that.
795 #define LAST_INO_BATCH 1024
796 static DEFINE_PER_CPU(unsigned int, last_ino);
798 unsigned int get_next_ino(void)
800 unsigned int *p = &get_cpu_var(last_ino);
801 unsigned int res = *p;
804 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
805 static atomic_t shared_last_ino;
806 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
808 res = next - LAST_INO_BATCH;
813 put_cpu_var(last_ino);
816 EXPORT_SYMBOL(get_next_ino);
819 * new_inode - obtain an inode
822 * Allocates a new inode for given superblock. The default gfp_mask
823 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
824 * If HIGHMEM pages are unsuitable or it is known that pages allocated
825 * for the page cache are not reclaimable or migratable,
826 * mapping_set_gfp_mask() must be called with suitable flags on the
827 * newly created inode's mapping
830 struct inode *new_inode(struct super_block *sb)
834 spin_lock_prefetch(&inode_lock);
836 inode = alloc_inode(sb);
838 spin_lock(&inode_lock);
839 __inode_sb_list_add(inode);
841 spin_unlock(&inode_lock);
845 EXPORT_SYMBOL(new_inode);
847 void unlock_new_inode(struct inode *inode)
849 #ifdef CONFIG_DEBUG_LOCK_ALLOC
850 if (S_ISDIR(inode->i_mode)) {
851 struct file_system_type *type = inode->i_sb->s_type;
853 /* Set new key only if filesystem hasn't already changed it */
854 if (!lockdep_match_class(&inode->i_mutex,
855 &type->i_mutex_key)) {
857 * ensure nobody is actually holding i_mutex
859 mutex_destroy(&inode->i_mutex);
860 mutex_init(&inode->i_mutex);
861 lockdep_set_class(&inode->i_mutex,
862 &type->i_mutex_dir_key);
867 * This is special! We do not need the spinlock when clearing I_NEW,
868 * because we're guaranteed that nobody else tries to do anything about
869 * the state of the inode when it is locked, as we just created it (so
870 * there can be no old holders that haven't tested I_NEW).
871 * However we must emit the memory barrier so that other CPUs reliably
872 * see the clearing of I_NEW after the other inode initialisation has
876 WARN_ON(!(inode->i_state & I_NEW));
877 inode->i_state &= ~I_NEW;
878 wake_up_inode(inode);
880 EXPORT_SYMBOL(unlock_new_inode);
883 * This is called without the inode lock held.. Be careful.
885 * We no longer cache the sb_flags in i_flags - see fs.h
886 * -- rmk@arm.uk.linux.org
888 static struct inode *get_new_inode(struct super_block *sb,
889 struct hlist_head *head,
890 int (*test)(struct inode *, void *),
891 int (*set)(struct inode *, void *),
896 inode = alloc_inode(sb);
900 spin_lock(&inode_lock);
901 /* We released the lock, so.. */
902 old = find_inode(sb, head, test, data);
904 if (set(inode, data))
907 hlist_add_head(&inode->i_hash, head);
908 __inode_sb_list_add(inode);
909 inode->i_state = I_NEW;
910 spin_unlock(&inode_lock);
912 /* Return the locked inode with I_NEW set, the
913 * caller is responsible for filling in the contents
919 * Uhhuh, somebody else created the same inode under
920 * us. Use the old inode instead of the one we just
923 spin_unlock(&inode_lock);
924 destroy_inode(inode);
926 wait_on_inode(inode);
931 spin_unlock(&inode_lock);
932 destroy_inode(inode);
937 * get_new_inode_fast is the fast path version of get_new_inode, see the
938 * comment at iget_locked for details.
940 static struct inode *get_new_inode_fast(struct super_block *sb,
941 struct hlist_head *head, unsigned long ino)
945 inode = alloc_inode(sb);
949 spin_lock(&inode_lock);
950 /* We released the lock, so.. */
951 old = find_inode_fast(sb, head, ino);
954 hlist_add_head(&inode->i_hash, head);
955 __inode_sb_list_add(inode);
956 inode->i_state = I_NEW;
957 spin_unlock(&inode_lock);
959 /* Return the locked inode with I_NEW set, the
960 * caller is responsible for filling in the contents
966 * Uhhuh, somebody else created the same inode under
967 * us. Use the old inode instead of the one we just
970 spin_unlock(&inode_lock);
971 destroy_inode(inode);
973 wait_on_inode(inode);
979 * search the inode cache for a matching inode number.
980 * If we find one, then the inode number we are trying to
981 * allocate is not unique and so we should not use it.
983 * Returns 1 if the inode number is unique, 0 if it is not.
985 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
987 struct hlist_head *b = inode_hashtable + hash(sb, ino);
988 struct hlist_node *node;
991 hlist_for_each_entry(inode, node, b, i_hash) {
992 if (inode->i_ino == ino && inode->i_sb == sb)
1000 * iunique - get a unique inode number
1002 * @max_reserved: highest reserved inode number
1004 * Obtain an inode number that is unique on the system for a given
1005 * superblock. This is used by file systems that have no natural
1006 * permanent inode numbering system. An inode number is returned that
1007 * is higher than the reserved limit but unique.
1010 * With a large number of inodes live on the file system this function
1011 * currently becomes quite slow.
1013 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1016 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1017 * error if st_ino won't fit in target struct field. Use 32bit counter
1018 * here to attempt to avoid that.
1020 static DEFINE_SPINLOCK(iunique_lock);
1021 static unsigned int counter;
1024 spin_lock(&inode_lock);
1025 spin_lock(&iunique_lock);
1027 if (counter <= max_reserved)
1028 counter = max_reserved + 1;
1030 } while (!test_inode_iunique(sb, res));
1031 spin_unlock(&iunique_lock);
1032 spin_unlock(&inode_lock);
1036 EXPORT_SYMBOL(iunique);
1038 struct inode *igrab(struct inode *inode)
1040 spin_lock(&inode_lock);
1041 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
1045 * Handle the case where s_op->clear_inode is not been
1046 * called yet, and somebody is calling igrab
1047 * while the inode is getting freed.
1050 spin_unlock(&inode_lock);
1053 EXPORT_SYMBOL(igrab);
1056 * ifind - internal function, you want ilookup5() or iget5().
1057 * @sb: super block of file system to search
1058 * @head: the head of the list to search
1059 * @test: callback used for comparisons between inodes
1060 * @data: opaque data pointer to pass to @test
1061 * @wait: if true wait for the inode to be unlocked, if false do not
1063 * ifind() searches for the inode specified by @data in the inode
1064 * cache. This is a generalized version of ifind_fast() for file systems where
1065 * the inode number is not sufficient for unique identification of an inode.
1067 * If the inode is in the cache, the inode is returned with an incremented
1070 * Otherwise NULL is returned.
1072 * Note, @test is called with the inode_lock held, so can't sleep.
1074 static struct inode *ifind(struct super_block *sb,
1075 struct hlist_head *head, int (*test)(struct inode *, void *),
1076 void *data, const int wait)
1078 struct inode *inode;
1080 spin_lock(&inode_lock);
1081 inode = find_inode(sb, head, test, data);
1083 spin_unlock(&inode_lock);
1085 wait_on_inode(inode);
1088 spin_unlock(&inode_lock);
1093 * ifind_fast - internal function, you want ilookup() or iget().
1094 * @sb: super block of file system to search
1095 * @head: head of the list to search
1096 * @ino: inode number to search for
1098 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1099 * file systems where the inode number is sufficient for unique identification
1102 * If the inode is in the cache, the inode is returned with an incremented
1105 * Otherwise NULL is returned.
1107 static struct inode *ifind_fast(struct super_block *sb,
1108 struct hlist_head *head, unsigned long ino)
1110 struct inode *inode;
1112 spin_lock(&inode_lock);
1113 inode = find_inode_fast(sb, head, ino);
1115 spin_unlock(&inode_lock);
1116 wait_on_inode(inode);
1119 spin_unlock(&inode_lock);
1124 * ilookup5_nowait - 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 is returned with an incremented
1136 * reference count. Note, the inode lock is not waited upon so you have to be
1137 * very careful what you do with the returned inode. You probably should be
1138 * using ilookup5() instead.
1140 * Otherwise NULL is returned.
1142 * Note, @test is called with the inode_lock held, so can't sleep.
1144 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1145 int (*test)(struct inode *, void *), void *data)
1147 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1149 return ifind(sb, head, test, data, 0);
1151 EXPORT_SYMBOL(ilookup5_nowait);
1154 * ilookup5 - search for an inode in the inode cache
1155 * @sb: super block of file system to search
1156 * @hashval: hash value (usually inode number) to search for
1157 * @test: callback used for comparisons between inodes
1158 * @data: opaque data pointer to pass to @test
1160 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1161 * @data in the inode cache. This is a generalized version of ilookup() for
1162 * file systems where the inode number is not sufficient for unique
1163 * identification of an inode.
1165 * If the inode is in the cache, the inode lock is waited upon and the inode is
1166 * returned with an incremented reference count.
1168 * Otherwise NULL is returned.
1170 * Note, @test is called with the inode_lock held, so can't sleep.
1172 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1173 int (*test)(struct inode *, void *), void *data)
1175 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1177 return ifind(sb, head, test, data, 1);
1179 EXPORT_SYMBOL(ilookup5);
1182 * ilookup - search for an inode in the inode cache
1183 * @sb: super block of file system to search
1184 * @ino: inode number to search for
1186 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1187 * This is for file systems where the inode number is sufficient for unique
1188 * identification of an inode.
1190 * If the inode is in the cache, the inode is returned with an incremented
1193 * Otherwise NULL is returned.
1195 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1197 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1199 return ifind_fast(sb, head, ino);
1201 EXPORT_SYMBOL(ilookup);
1204 * iget5_locked - obtain an inode from a mounted file system
1205 * @sb: super block of file system
1206 * @hashval: hash value (usually inode number) to get
1207 * @test: callback used for comparisons between inodes
1208 * @set: callback used to initialize a new struct inode
1209 * @data: opaque data pointer to pass to @test and @set
1211 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1212 * and @data in the inode cache and if present it is returned with an increased
1213 * reference count. This is a generalized version of iget_locked() for file
1214 * systems where the inode number is not sufficient for unique identification
1217 * If the inode is not in cache, get_new_inode() is called to allocate a new
1218 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1219 * file system gets to fill it in before unlocking it via unlock_new_inode().
1221 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1223 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1224 int (*test)(struct inode *, void *),
1225 int (*set)(struct inode *, void *), void *data)
1227 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1228 struct inode *inode;
1230 inode = ifind(sb, head, test, data, 1);
1234 * get_new_inode() will do the right thing, re-trying the search
1235 * in case it had to block at any point.
1237 return get_new_inode(sb, head, test, set, data);
1239 EXPORT_SYMBOL(iget5_locked);
1242 * iget_locked - obtain an inode from a mounted file system
1243 * @sb: super block of file system
1244 * @ino: inode number to get
1246 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1247 * the inode cache and if present it is returned with an increased reference
1248 * count. This is for file systems where the inode number is sufficient for
1249 * unique identification of an inode.
1251 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1252 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1253 * The file system gets to fill it in before unlocking it via
1254 * unlock_new_inode().
1256 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1258 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1259 struct inode *inode;
1261 inode = ifind_fast(sb, head, ino);
1265 * get_new_inode_fast() will do the right thing, re-trying the search
1266 * in case it had to block at any point.
1268 return get_new_inode_fast(sb, head, ino);
1270 EXPORT_SYMBOL(iget_locked);
1272 int insert_inode_locked(struct inode *inode)
1274 struct super_block *sb = inode->i_sb;
1275 ino_t ino = inode->i_ino;
1276 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1278 inode->i_state |= I_NEW;
1280 struct hlist_node *node;
1281 struct inode *old = NULL;
1282 spin_lock(&inode_lock);
1283 hlist_for_each_entry(old, node, head, i_hash) {
1284 if (old->i_ino != ino)
1286 if (old->i_sb != sb)
1288 if (old->i_state & (I_FREEING|I_WILL_FREE))
1292 if (likely(!node)) {
1293 hlist_add_head(&inode->i_hash, head);
1294 spin_unlock(&inode_lock);
1298 spin_unlock(&inode_lock);
1300 if (unlikely(!inode_unhashed(old))) {
1307 EXPORT_SYMBOL(insert_inode_locked);
1309 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1310 int (*test)(struct inode *, void *), void *data)
1312 struct super_block *sb = inode->i_sb;
1313 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1315 inode->i_state |= I_NEW;
1318 struct hlist_node *node;
1319 struct inode *old = NULL;
1321 spin_lock(&inode_lock);
1322 hlist_for_each_entry(old, node, head, i_hash) {
1323 if (old->i_sb != sb)
1325 if (!test(old, data))
1327 if (old->i_state & (I_FREEING|I_WILL_FREE))
1331 if (likely(!node)) {
1332 hlist_add_head(&inode->i_hash, head);
1333 spin_unlock(&inode_lock);
1337 spin_unlock(&inode_lock);
1339 if (unlikely(!inode_unhashed(old))) {
1346 EXPORT_SYMBOL(insert_inode_locked4);
1349 int generic_delete_inode(struct inode *inode)
1353 EXPORT_SYMBOL(generic_delete_inode);
1356 * Normal UNIX filesystem behaviour: delete the
1357 * inode when the usage count drops to zero, and
1360 int generic_drop_inode(struct inode *inode)
1362 return !inode->i_nlink || inode_unhashed(inode);
1364 EXPORT_SYMBOL_GPL(generic_drop_inode);
1367 * Called when we're dropping the last reference
1370 * Call the FS "drop_inode()" function, defaulting to
1371 * the legacy UNIX filesystem behaviour. If it tells
1372 * us to evict inode, do so. Otherwise, retain inode
1373 * in cache if fs is alive, sync and evict if fs is
1376 static void iput_final(struct inode *inode)
1378 struct super_block *sb = inode->i_sb;
1379 const struct super_operations *op = inode->i_sb->s_op;
1382 if (op && op->drop_inode)
1383 drop = op->drop_inode(inode);
1385 drop = generic_drop_inode(inode);
1388 if (sb->s_flags & MS_ACTIVE) {
1389 inode->i_state |= I_REFERENCED;
1390 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1391 inode_lru_list_add(inode);
1393 spin_unlock(&inode_lock);
1396 WARN_ON(inode->i_state & I_NEW);
1397 inode->i_state |= I_WILL_FREE;
1398 spin_unlock(&inode_lock);
1399 write_inode_now(inode, 1);
1400 spin_lock(&inode_lock);
1401 WARN_ON(inode->i_state & I_NEW);
1402 inode->i_state &= ~I_WILL_FREE;
1403 __remove_inode_hash(inode);
1406 WARN_ON(inode->i_state & I_NEW);
1407 inode->i_state |= I_FREEING;
1410 * Move the inode off the IO lists and LRU once I_FREEING is
1411 * set so that it won't get moved back on there if it is dirty.
1413 inode_lru_list_del(inode);
1414 list_del_init(&inode->i_wb_list);
1416 __inode_sb_list_del(inode);
1417 spin_unlock(&inode_lock);
1419 remove_inode_hash(inode);
1420 wake_up_inode(inode);
1421 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1422 destroy_inode(inode);
1426 * iput - put an inode
1427 * @inode: inode to put
1429 * Puts an inode, dropping its usage count. If the inode use count hits
1430 * zero, the inode is then freed and may also be destroyed.
1432 * Consequently, iput() can sleep.
1434 void iput(struct inode *inode)
1437 BUG_ON(inode->i_state & I_CLEAR);
1439 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1443 EXPORT_SYMBOL(iput);
1446 * bmap - find a block number in a file
1447 * @inode: inode of file
1448 * @block: block to find
1450 * Returns the block number on the device holding the inode that
1451 * is the disk block number for the block of the file requested.
1452 * That is, asked for block 4 of inode 1 the function will return the
1453 * disk block relative to the disk start that holds that block of the
1456 sector_t bmap(struct inode *inode, sector_t block)
1459 if (inode->i_mapping->a_ops->bmap)
1460 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1463 EXPORT_SYMBOL(bmap);
1466 * With relative atime, only update atime if the previous atime is
1467 * earlier than either the ctime or mtime or if at least a day has
1468 * passed since the last atime update.
1470 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1471 struct timespec now)
1474 if (!(mnt->mnt_flags & MNT_RELATIME))
1477 * Is mtime younger than atime? If yes, update atime:
1479 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1482 * Is ctime younger than atime? If yes, update atime:
1484 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1488 * Is the previous atime value older than a day? If yes,
1491 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1494 * Good, we can skip the atime update:
1500 * touch_atime - update the access time
1501 * @mnt: mount the inode is accessed on
1502 * @dentry: dentry accessed
1504 * Update the accessed time on an inode and mark it for writeback.
1505 * This function automatically handles read only file systems and media,
1506 * as well as the "noatime" flag and inode specific "noatime" markers.
1508 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1510 struct inode *inode = dentry->d_inode;
1511 struct timespec now;
1513 if (inode->i_flags & S_NOATIME)
1515 if (IS_NOATIME(inode))
1517 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1520 if (mnt->mnt_flags & MNT_NOATIME)
1522 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1525 now = current_fs_time(inode->i_sb);
1527 if (!relatime_need_update(mnt, inode, now))
1530 if (timespec_equal(&inode->i_atime, &now))
1533 if (mnt_want_write(mnt))
1536 inode->i_atime = now;
1537 mark_inode_dirty_sync(inode);
1538 mnt_drop_write(mnt);
1540 EXPORT_SYMBOL(touch_atime);
1543 * file_update_time - update mtime and ctime time
1544 * @file: file accessed
1546 * Update the mtime and ctime members of an inode and mark the inode
1547 * for writeback. Note that this function is meant exclusively for
1548 * usage in the file write path of filesystems, and filesystems may
1549 * choose to explicitly ignore update via this function with the
1550 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1551 * timestamps are handled by the server.
1554 void file_update_time(struct file *file)
1556 struct inode *inode = file->f_path.dentry->d_inode;
1557 struct timespec now;
1558 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1560 /* First try to exhaust all avenues to not sync */
1561 if (IS_NOCMTIME(inode))
1564 now = current_fs_time(inode->i_sb);
1565 if (!timespec_equal(&inode->i_mtime, &now))
1568 if (!timespec_equal(&inode->i_ctime, &now))
1571 if (IS_I_VERSION(inode))
1572 sync_it |= S_VERSION;
1577 /* Finally allowed to write? Takes lock. */
1578 if (mnt_want_write_file(file))
1581 /* Only change inode inside the lock region */
1582 if (sync_it & S_VERSION)
1583 inode_inc_iversion(inode);
1584 if (sync_it & S_CTIME)
1585 inode->i_ctime = now;
1586 if (sync_it & S_MTIME)
1587 inode->i_mtime = now;
1588 mark_inode_dirty_sync(inode);
1589 mnt_drop_write(file->f_path.mnt);
1591 EXPORT_SYMBOL(file_update_time);
1593 int inode_needs_sync(struct inode *inode)
1597 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1601 EXPORT_SYMBOL(inode_needs_sync);
1603 int inode_wait(void *word)
1608 EXPORT_SYMBOL(inode_wait);
1611 * If we try to find an inode in the inode hash while it is being
1612 * deleted, we have to wait until the filesystem completes its
1613 * deletion before reporting that it isn't found. This function waits
1614 * until the deletion _might_ have completed. Callers are responsible
1615 * to recheck inode state.
1617 * It doesn't matter if I_NEW is not set initially, a call to
1618 * wake_up_inode() after removing from the hash list will DTRT.
1620 * This is called with inode_lock held.
1622 static void __wait_on_freeing_inode(struct inode *inode)
1624 wait_queue_head_t *wq;
1625 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1626 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1627 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1628 spin_unlock(&inode_lock);
1630 finish_wait(wq, &wait.wait);
1631 spin_lock(&inode_lock);
1634 static __initdata unsigned long ihash_entries;
1635 static int __init set_ihash_entries(char *str)
1639 ihash_entries = simple_strtoul(str, &str, 0);
1642 __setup("ihash_entries=", set_ihash_entries);
1645 * Initialize the waitqueues and inode hash table.
1647 void __init inode_init_early(void)
1651 /* If hashes are distributed across NUMA nodes, defer
1652 * hash allocation until vmalloc space is available.
1658 alloc_large_system_hash("Inode-cache",
1659 sizeof(struct hlist_head),
1667 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1668 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1671 void __init inode_init(void)
1675 /* inode slab cache */
1676 inode_cachep = kmem_cache_create("inode_cache",
1677 sizeof(struct inode),
1679 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1682 register_shrinker(&icache_shrinker);
1684 /* Hash may have been set up in inode_init_early */
1689 alloc_large_system_hash("Inode-cache",
1690 sizeof(struct hlist_head),
1698 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1699 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1702 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1704 inode->i_mode = mode;
1705 if (S_ISCHR(mode)) {
1706 inode->i_fop = &def_chr_fops;
1707 inode->i_rdev = rdev;
1708 } else if (S_ISBLK(mode)) {
1709 inode->i_fop = &def_blk_fops;
1710 inode->i_rdev = rdev;
1711 } else if (S_ISFIFO(mode))
1712 inode->i_fop = &def_fifo_fops;
1713 else if (S_ISSOCK(mode))
1714 inode->i_fop = &bad_sock_fops;
1716 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1717 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1720 EXPORT_SYMBOL(init_special_inode);
1723 * Init uid,gid,mode for new inode according to posix standards
1725 * @dir: Directory inode
1726 * @mode: mode of the new inode
1728 void inode_init_owner(struct inode *inode, const struct inode *dir,
1731 inode->i_uid = current_fsuid();
1732 if (dir && dir->i_mode & S_ISGID) {
1733 inode->i_gid = dir->i_gid;
1737 inode->i_gid = current_fsgid();
1738 inode->i_mode = mode;
1740 EXPORT_SYMBOL(inode_init_owner);
1742 #define CREATE_TRACE_POINTS
1743 #include <trace/events/vfs.h>