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>
29 * This is needed for the following functions:
33 * FIXME: remove all knowledge of the buffer layer from this file
35 #include <linux/buffer_head.h>
38 * New inode.c implementation.
40 * This implementation has the basic premise of trying
41 * to be extremely low-overhead and SMP-safe, yet be
42 * simple enough to be "obviously correct".
47 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
49 /* #define INODE_PARANOIA 1 */
50 /* #define INODE_DEBUG 1 */
53 * Inode lookup is no longer as critical as it used to be:
54 * most of the lookups are going to be through the dcache.
56 #define I_HASHBITS i_hash_shift
57 #define I_HASHMASK i_hash_mask
59 static unsigned int i_hash_mask __read_mostly;
60 static unsigned int i_hash_shift __read_mostly;
63 * Each inode can be on two separate lists. One is
64 * the hash list of the inode, used for lookups. The
65 * other linked list is the "type" list:
66 * "in_use" - valid inode, i_count > 0, i_nlink > 0
67 * "dirty" - as "in_use" but also dirty
68 * "unused" - valid inode, i_count = 0
70 * A "dirty" list is maintained for each super block,
71 * allowing for low-overhead inode sync() operations.
74 static LIST_HEAD(inode_lru);
75 static struct hlist_head *inode_hashtable __read_mostly;
78 * A simple spinlock to protect the list manipulations.
80 * NOTE! You also have to own the lock if you change
81 * the i_state of an inode while it is in use..
83 DEFINE_SPINLOCK(inode_lock);
86 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
87 * icache shrinking path, and the umount path. Without this exclusion,
88 * by the time prune_icache calls iput for the inode whose pages it has
89 * been invalidating, or by the time it calls clear_inode & destroy_inode
90 * from its final dispose_list, the struct super_block they refer to
91 * (for inode->i_sb->s_op) may already have been freed and reused.
93 * We make this an rwsem because the fastpath is icache shrinking. In
94 * some cases a filesystem may be doing a significant amount of work in
95 * its inode reclaim code, so this should improve parallelism.
97 static DECLARE_RWSEM(iprune_sem);
100 * Statistics gathering..
102 struct inodes_stat_t inodes_stat;
104 static struct percpu_counter nr_inodes __cacheline_aligned_in_smp;
105 static struct percpu_counter nr_inodes_unused __cacheline_aligned_in_smp;
107 static struct kmem_cache *inode_cachep __read_mostly;
109 static inline int get_nr_inodes(void)
111 return percpu_counter_sum_positive(&nr_inodes);
114 static inline int get_nr_inodes_unused(void)
116 return percpu_counter_sum_positive(&nr_inodes_unused);
119 int get_nr_dirty_inodes(void)
121 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
122 return nr_dirty > 0 ? nr_dirty : 0;
127 * Handle nr_inode sysctl
130 int proc_nr_inodes(ctl_table *table, int write,
131 void __user *buffer, size_t *lenp, loff_t *ppos)
133 inodes_stat.nr_inodes = get_nr_inodes();
134 inodes_stat.nr_unused = get_nr_inodes_unused();
135 return proc_dointvec(table, write, buffer, lenp, ppos);
139 static void wake_up_inode(struct inode *inode)
142 * Prevent speculative execution through spin_unlock(&inode_lock);
145 wake_up_bit(&inode->i_state, __I_NEW);
149 * inode_init_always - perform inode structure intialisation
150 * @sb: superblock inode belongs to
151 * @inode: inode to initialise
153 * These are initializations that need to be done on every inode
154 * allocation as the fields are not initialised by slab allocation.
156 int inode_init_always(struct super_block *sb, struct inode *inode)
158 static const struct address_space_operations empty_aops;
159 static const struct inode_operations empty_iops;
160 static const struct file_operations empty_fops;
161 struct address_space *const mapping = &inode->i_data;
164 inode->i_blkbits = sb->s_blocksize_bits;
166 atomic_set(&inode->i_count, 1);
167 inode->i_op = &empty_iops;
168 inode->i_fop = &empty_fops;
172 atomic_set(&inode->i_writecount, 0);
176 inode->i_generation = 0;
178 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
180 inode->i_pipe = NULL;
181 inode->i_bdev = NULL;
182 inode->i_cdev = NULL;
184 inode->dirtied_when = 0;
186 if (security_inode_alloc(inode))
188 spin_lock_init(&inode->i_lock);
189 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
191 mutex_init(&inode->i_mutex);
192 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
194 init_rwsem(&inode->i_alloc_sem);
195 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
197 mapping->a_ops = &empty_aops;
198 mapping->host = inode;
200 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
201 mapping->assoc_mapping = NULL;
202 mapping->backing_dev_info = &default_backing_dev_info;
203 mapping->writeback_index = 0;
206 * If the block_device provides a backing_dev_info for client
207 * inodes then use that. Otherwise the inode share the bdev's
211 struct backing_dev_info *bdi;
213 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
214 mapping->backing_dev_info = bdi;
216 inode->i_private = NULL;
217 inode->i_mapping = mapping;
218 #ifdef CONFIG_FS_POSIX_ACL
219 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
222 #ifdef CONFIG_FSNOTIFY
223 inode->i_fsnotify_mask = 0;
226 percpu_counter_inc(&nr_inodes);
232 EXPORT_SYMBOL(inode_init_always);
234 static struct inode *alloc_inode(struct super_block *sb)
238 if (sb->s_op->alloc_inode)
239 inode = sb->s_op->alloc_inode(sb);
241 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
246 if (unlikely(inode_init_always(sb, inode))) {
247 if (inode->i_sb->s_op->destroy_inode)
248 inode->i_sb->s_op->destroy_inode(inode);
250 kmem_cache_free(inode_cachep, inode);
257 void __destroy_inode(struct inode *inode)
259 BUG_ON(inode_has_buffers(inode));
260 security_inode_free(inode);
261 fsnotify_inode_delete(inode);
262 #ifdef CONFIG_FS_POSIX_ACL
263 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
264 posix_acl_release(inode->i_acl);
265 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
266 posix_acl_release(inode->i_default_acl);
268 percpu_counter_dec(&nr_inodes);
270 EXPORT_SYMBOL(__destroy_inode);
272 static void destroy_inode(struct inode *inode)
274 BUG_ON(!list_empty(&inode->i_lru));
275 __destroy_inode(inode);
276 if (inode->i_sb->s_op->destroy_inode)
277 inode->i_sb->s_op->destroy_inode(inode);
279 kmem_cache_free(inode_cachep, (inode));
283 * These are initializations that only need to be done
284 * once, because the fields are idempotent across use
285 * of the inode, so let the slab aware of that.
287 void inode_init_once(struct inode *inode)
289 memset(inode, 0, sizeof(*inode));
290 INIT_HLIST_NODE(&inode->i_hash);
291 INIT_LIST_HEAD(&inode->i_dentry);
292 INIT_LIST_HEAD(&inode->i_devices);
293 INIT_LIST_HEAD(&inode->i_wb_list);
294 INIT_LIST_HEAD(&inode->i_lru);
295 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
296 spin_lock_init(&inode->i_data.tree_lock);
297 spin_lock_init(&inode->i_data.i_mmap_lock);
298 INIT_LIST_HEAD(&inode->i_data.private_list);
299 spin_lock_init(&inode->i_data.private_lock);
300 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
301 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
302 i_size_ordered_init(inode);
303 #ifdef CONFIG_FSNOTIFY
304 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
307 EXPORT_SYMBOL(inode_init_once);
309 static void init_once(void *foo)
311 struct inode *inode = (struct inode *) foo;
313 inode_init_once(inode);
317 * inode_lock must be held
319 void __iget(struct inode *inode)
321 atomic_inc(&inode->i_count);
325 * get additional reference to inode; caller must already hold one.
327 void ihold(struct inode *inode)
329 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
331 EXPORT_SYMBOL(ihold);
333 static void inode_lru_list_add(struct inode *inode)
335 if (list_empty(&inode->i_lru)) {
336 list_add(&inode->i_lru, &inode_lru);
337 percpu_counter_inc(&nr_inodes_unused);
341 static void inode_lru_list_del(struct inode *inode)
343 if (!list_empty(&inode->i_lru)) {
344 list_del_init(&inode->i_lru);
345 percpu_counter_dec(&nr_inodes_unused);
349 static inline void __inode_sb_list_add(struct inode *inode)
351 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
355 * inode_sb_list_add - add inode to the superblock list of inodes
356 * @inode: inode to add
358 void inode_sb_list_add(struct inode *inode)
360 spin_lock(&inode_lock);
361 __inode_sb_list_add(inode);
362 spin_unlock(&inode_lock);
364 EXPORT_SYMBOL_GPL(inode_sb_list_add);
366 static inline void __inode_sb_list_del(struct inode *inode)
368 list_del_init(&inode->i_sb_list);
371 static unsigned long hash(struct super_block *sb, unsigned long hashval)
375 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
377 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
378 return tmp & I_HASHMASK;
382 * __insert_inode_hash - hash an inode
383 * @inode: unhashed inode
384 * @hashval: unsigned long value used to locate this object in the
387 * Add an inode to the inode hash for this superblock.
389 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
391 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
393 spin_lock(&inode_lock);
394 hlist_add_head(&inode->i_hash, b);
395 spin_unlock(&inode_lock);
397 EXPORT_SYMBOL(__insert_inode_hash);
400 * __remove_inode_hash - remove an inode from the hash
401 * @inode: inode to unhash
403 * Remove an inode from the superblock.
405 static void __remove_inode_hash(struct inode *inode)
407 hlist_del_init(&inode->i_hash);
411 * remove_inode_hash - remove an inode from the hash
412 * @inode: inode to unhash
414 * Remove an inode from the superblock.
416 void remove_inode_hash(struct inode *inode)
418 spin_lock(&inode_lock);
419 hlist_del_init(&inode->i_hash);
420 spin_unlock(&inode_lock);
422 EXPORT_SYMBOL(remove_inode_hash);
424 void end_writeback(struct inode *inode)
427 BUG_ON(inode->i_data.nrpages);
428 BUG_ON(!list_empty(&inode->i_data.private_list));
429 BUG_ON(!(inode->i_state & I_FREEING));
430 BUG_ON(inode->i_state & I_CLEAR);
431 inode_sync_wait(inode);
432 inode->i_state = I_FREEING | I_CLEAR;
434 EXPORT_SYMBOL(end_writeback);
436 static void evict(struct inode *inode)
438 const struct super_operations *op = inode->i_sb->s_op;
440 if (op->evict_inode) {
441 op->evict_inode(inode);
443 if (inode->i_data.nrpages)
444 truncate_inode_pages(&inode->i_data, 0);
445 end_writeback(inode);
447 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
449 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
454 * dispose_list - dispose of the contents of a local list
455 * @head: the head of the list to free
457 * Dispose-list gets a local list with local inodes in it, so it doesn't
458 * need to worry about list corruption and SMP locks.
460 static void dispose_list(struct list_head *head)
462 while (!list_empty(head)) {
465 inode = list_first_entry(head, struct inode, i_lru);
466 list_del_init(&inode->i_lru);
470 spin_lock(&inode_lock);
471 __remove_inode_hash(inode);
472 __inode_sb_list_del(inode);
473 spin_unlock(&inode_lock);
475 wake_up_inode(inode);
476 destroy_inode(inode);
481 * Invalidate all inodes for a device.
483 static int invalidate_list(struct list_head *head, struct list_head *dispose)
485 struct list_head *next;
490 struct list_head *tmp = next;
494 * We can reschedule here without worrying about the list's
495 * consistency because the per-sb list of inodes must not
496 * change during umount anymore, and because iprune_sem keeps
497 * shrink_icache_memory() away.
499 cond_resched_lock(&inode_lock);
504 inode = list_entry(tmp, struct inode, i_sb_list);
505 if (inode->i_state & I_NEW)
507 if (atomic_read(&inode->i_count)) {
512 inode->i_state |= I_FREEING;
515 * Move the inode off the IO lists and LRU once I_FREEING is
516 * set so that it won't get moved back on there if it is dirty.
518 list_move(&inode->i_lru, dispose);
519 list_del_init(&inode->i_wb_list);
520 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
521 percpu_counter_dec(&nr_inodes_unused);
527 * invalidate_inodes - discard the inodes on a device
530 * Discard all of the inodes for a given superblock. If the discard
531 * fails because there are busy inodes then a non zero value is returned.
532 * If the discard is successful all the inodes have been discarded.
534 int invalidate_inodes(struct super_block *sb)
537 LIST_HEAD(throw_away);
539 down_write(&iprune_sem);
540 spin_lock(&inode_lock);
541 fsnotify_unmount_inodes(&sb->s_inodes);
542 busy = invalidate_list(&sb->s_inodes, &throw_away);
543 spin_unlock(&inode_lock);
545 dispose_list(&throw_away);
546 up_write(&iprune_sem);
551 static int can_unuse(struct inode *inode)
553 if (inode->i_state & ~I_REFERENCED)
555 if (inode_has_buffers(inode))
557 if (atomic_read(&inode->i_count))
559 if (inode->i_data.nrpages)
565 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
566 * temporary list and then are freed outside inode_lock by dispose_list().
568 * Any inodes which are pinned purely because of attached pagecache have their
569 * pagecache removed. If the inode has metadata buffers attached to
570 * mapping->private_list then try to remove them.
572 * If the inode has the I_REFERENCED flag set, then it means that it has been
573 * used recently - the flag is set in iput_final(). When we encounter such an
574 * inode, clear the flag and move it to the back of the LRU so it gets another
575 * pass through the LRU before it gets reclaimed. This is necessary because of
576 * the fact we are doing lazy LRU updates to minimise lock contention so the
577 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
578 * with this flag set because they are the inodes that are out of order.
580 static void prune_icache(int nr_to_scan)
584 unsigned long reap = 0;
586 down_read(&iprune_sem);
587 spin_lock(&inode_lock);
588 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
591 if (list_empty(&inode_lru))
594 inode = list_entry(inode_lru.prev, struct inode, i_lru);
597 * Referenced or dirty inodes are still in use. Give them
598 * another pass through the LRU as we canot reclaim them now.
600 if (atomic_read(&inode->i_count) ||
601 (inode->i_state & ~I_REFERENCED)) {
602 list_del_init(&inode->i_lru);
603 percpu_counter_dec(&nr_inodes_unused);
607 /* recently referenced inodes get one more pass */
608 if (inode->i_state & I_REFERENCED) {
609 list_move(&inode->i_lru, &inode_lru);
610 inode->i_state &= ~I_REFERENCED;
613 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
615 spin_unlock(&inode_lock);
616 if (remove_inode_buffers(inode))
617 reap += invalidate_mapping_pages(&inode->i_data,
620 spin_lock(&inode_lock);
622 if (inode != list_entry(inode_lru.next,
623 struct inode, i_lru))
624 continue; /* wrong inode or list_empty */
625 if (!can_unuse(inode))
628 WARN_ON(inode->i_state & I_NEW);
629 inode->i_state |= I_FREEING;
632 * Move the inode off the IO lists and LRU once I_FREEING is
633 * set so that it won't get moved back on there if it is dirty.
635 list_move(&inode->i_lru, &freeable);
636 list_del_init(&inode->i_wb_list);
637 percpu_counter_dec(&nr_inodes_unused);
639 if (current_is_kswapd())
640 __count_vm_events(KSWAPD_INODESTEAL, reap);
642 __count_vm_events(PGINODESTEAL, reap);
643 spin_unlock(&inode_lock);
645 dispose_list(&freeable);
646 up_read(&iprune_sem);
650 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
651 * "unused" means that no dentries are referring to the inodes: the files are
652 * not open and the dcache references to those inodes have already been
655 * This function is passed the number of inodes to scan, and it returns the
656 * total number of remaining possibly-reclaimable inodes.
658 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
662 * Nasty deadlock avoidance. We may hold various FS locks,
663 * and we don't want to recurse into the FS that called us
664 * in clear_inode() and friends..
666 if (!(gfp_mask & __GFP_FS))
670 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
673 static struct shrinker icache_shrinker = {
674 .shrink = shrink_icache_memory,
675 .seeks = DEFAULT_SEEKS,
678 static void __wait_on_freeing_inode(struct inode *inode);
680 * Called with the inode lock held.
682 static struct inode *find_inode(struct super_block *sb,
683 struct hlist_head *head,
684 int (*test)(struct inode *, void *),
687 struct hlist_node *node;
688 struct inode *inode = NULL;
691 hlist_for_each_entry(inode, node, head, i_hash) {
692 if (inode->i_sb != sb)
694 if (!test(inode, data))
696 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
697 __wait_on_freeing_inode(inode);
707 * find_inode_fast is the fast path version of find_inode, see the comment at
708 * iget_locked for details.
710 static struct inode *find_inode_fast(struct super_block *sb,
711 struct hlist_head *head, unsigned long ino)
713 struct hlist_node *node;
714 struct inode *inode = NULL;
717 hlist_for_each_entry(inode, node, head, i_hash) {
718 if (inode->i_ino != ino)
720 if (inode->i_sb != sb)
722 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
723 __wait_on_freeing_inode(inode);
733 * Each cpu owns a range of LAST_INO_BATCH numbers.
734 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
735 * to renew the exhausted range.
737 * This does not significantly increase overflow rate because every CPU can
738 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
739 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
740 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
741 * overflow rate by 2x, which does not seem too significant.
743 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
744 * error if st_ino won't fit in target struct field. Use 32bit counter
745 * here to attempt to avoid that.
747 #define LAST_INO_BATCH 1024
748 static DEFINE_PER_CPU(unsigned int, last_ino);
750 unsigned int get_next_ino(void)
752 unsigned int *p = &get_cpu_var(last_ino);
753 unsigned int res = *p;
756 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
757 static atomic_t shared_last_ino;
758 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
760 res = next - LAST_INO_BATCH;
765 put_cpu_var(last_ino);
768 EXPORT_SYMBOL(get_next_ino);
771 * new_inode - obtain an inode
774 * Allocates a new inode for given superblock. The default gfp_mask
775 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
776 * If HIGHMEM pages are unsuitable or it is known that pages allocated
777 * for the page cache are not reclaimable or migratable,
778 * mapping_set_gfp_mask() must be called with suitable flags on the
779 * newly created inode's mapping
782 struct inode *new_inode(struct super_block *sb)
786 spin_lock_prefetch(&inode_lock);
788 inode = alloc_inode(sb);
790 spin_lock(&inode_lock);
791 __inode_sb_list_add(inode);
793 spin_unlock(&inode_lock);
797 EXPORT_SYMBOL(new_inode);
799 void unlock_new_inode(struct inode *inode)
801 #ifdef CONFIG_DEBUG_LOCK_ALLOC
802 if (S_ISDIR(inode->i_mode)) {
803 struct file_system_type *type = inode->i_sb->s_type;
805 /* Set new key only if filesystem hasn't already changed it */
806 if (!lockdep_match_class(&inode->i_mutex,
807 &type->i_mutex_key)) {
809 * ensure nobody is actually holding i_mutex
811 mutex_destroy(&inode->i_mutex);
812 mutex_init(&inode->i_mutex);
813 lockdep_set_class(&inode->i_mutex,
814 &type->i_mutex_dir_key);
819 * This is special! We do not need the spinlock when clearing I_NEW,
820 * because we're guaranteed that nobody else tries to do anything about
821 * the state of the inode when it is locked, as we just created it (so
822 * there can be no old holders that haven't tested I_NEW).
823 * However we must emit the memory barrier so that other CPUs reliably
824 * see the clearing of I_NEW after the other inode initialisation has
828 WARN_ON(!(inode->i_state & I_NEW));
829 inode->i_state &= ~I_NEW;
830 wake_up_inode(inode);
832 EXPORT_SYMBOL(unlock_new_inode);
835 * This is called without the inode lock held.. Be careful.
837 * We no longer cache the sb_flags in i_flags - see fs.h
838 * -- rmk@arm.uk.linux.org
840 static struct inode *get_new_inode(struct super_block *sb,
841 struct hlist_head *head,
842 int (*test)(struct inode *, void *),
843 int (*set)(struct inode *, void *),
848 inode = alloc_inode(sb);
852 spin_lock(&inode_lock);
853 /* We released the lock, so.. */
854 old = find_inode(sb, head, test, data);
856 if (set(inode, data))
859 hlist_add_head(&inode->i_hash, head);
860 __inode_sb_list_add(inode);
861 inode->i_state = I_NEW;
862 spin_unlock(&inode_lock);
864 /* Return the locked inode with I_NEW set, the
865 * caller is responsible for filling in the contents
871 * Uhhuh, somebody else created the same inode under
872 * us. Use the old inode instead of the one we just
875 spin_unlock(&inode_lock);
876 destroy_inode(inode);
878 wait_on_inode(inode);
883 spin_unlock(&inode_lock);
884 destroy_inode(inode);
889 * get_new_inode_fast is the fast path version of get_new_inode, see the
890 * comment at iget_locked for details.
892 static struct inode *get_new_inode_fast(struct super_block *sb,
893 struct hlist_head *head, unsigned long ino)
897 inode = alloc_inode(sb);
901 spin_lock(&inode_lock);
902 /* We released the lock, so.. */
903 old = find_inode_fast(sb, head, ino);
906 hlist_add_head(&inode->i_hash, head);
907 __inode_sb_list_add(inode);
908 inode->i_state = I_NEW;
909 spin_unlock(&inode_lock);
911 /* Return the locked inode with I_NEW set, the
912 * caller is responsible for filling in the contents
918 * Uhhuh, somebody else created the same inode under
919 * us. Use the old inode instead of the one we just
922 spin_unlock(&inode_lock);
923 destroy_inode(inode);
925 wait_on_inode(inode);
931 * search the inode cache for a matching inode number.
932 * If we find one, then the inode number we are trying to
933 * allocate is not unique and so we should not use it.
935 * Returns 1 if the inode number is unique, 0 if it is not.
937 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
939 struct hlist_head *b = inode_hashtable + hash(sb, ino);
940 struct hlist_node *node;
943 hlist_for_each_entry(inode, node, b, i_hash) {
944 if (inode->i_ino == ino && inode->i_sb == sb)
952 * iunique - get a unique inode number
954 * @max_reserved: highest reserved inode number
956 * Obtain an inode number that is unique on the system for a given
957 * superblock. This is used by file systems that have no natural
958 * permanent inode numbering system. An inode number is returned that
959 * is higher than the reserved limit but unique.
962 * With a large number of inodes live on the file system this function
963 * currently becomes quite slow.
965 ino_t iunique(struct super_block *sb, ino_t max_reserved)
968 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
969 * error if st_ino won't fit in target struct field. Use 32bit counter
970 * here to attempt to avoid that.
972 static DEFINE_SPINLOCK(iunique_lock);
973 static unsigned int counter;
976 spin_lock(&inode_lock);
977 spin_lock(&iunique_lock);
979 if (counter <= max_reserved)
980 counter = max_reserved + 1;
982 } while (!test_inode_iunique(sb, res));
983 spin_unlock(&iunique_lock);
984 spin_unlock(&inode_lock);
988 EXPORT_SYMBOL(iunique);
990 struct inode *igrab(struct inode *inode)
992 spin_lock(&inode_lock);
993 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
997 * Handle the case where s_op->clear_inode is not been
998 * called yet, and somebody is calling igrab
999 * while the inode is getting freed.
1002 spin_unlock(&inode_lock);
1005 EXPORT_SYMBOL(igrab);
1008 * ifind - internal function, you want ilookup5() or iget5().
1009 * @sb: super block of file system to search
1010 * @head: the head of the list to search
1011 * @test: callback used for comparisons between inodes
1012 * @data: opaque data pointer to pass to @test
1013 * @wait: if true wait for the inode to be unlocked, if false do not
1015 * ifind() searches for the inode specified by @data in the inode
1016 * cache. This is a generalized version of ifind_fast() for file systems where
1017 * the inode number is not sufficient for unique identification of an inode.
1019 * If the inode is in the cache, the inode is returned with an incremented
1022 * Otherwise NULL is returned.
1024 * Note, @test is called with the inode_lock held, so can't sleep.
1026 static struct inode *ifind(struct super_block *sb,
1027 struct hlist_head *head, int (*test)(struct inode *, void *),
1028 void *data, const int wait)
1030 struct inode *inode;
1032 spin_lock(&inode_lock);
1033 inode = find_inode(sb, head, test, data);
1035 spin_unlock(&inode_lock);
1037 wait_on_inode(inode);
1040 spin_unlock(&inode_lock);
1045 * ifind_fast - internal function, you want ilookup() or iget().
1046 * @sb: super block of file system to search
1047 * @head: head of the list to search
1048 * @ino: inode number to search for
1050 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1051 * file systems where the inode number is sufficient for unique identification
1054 * If the inode is in the cache, the inode is returned with an incremented
1057 * Otherwise NULL is returned.
1059 static struct inode *ifind_fast(struct super_block *sb,
1060 struct hlist_head *head, unsigned long ino)
1062 struct inode *inode;
1064 spin_lock(&inode_lock);
1065 inode = find_inode_fast(sb, head, ino);
1067 spin_unlock(&inode_lock);
1068 wait_on_inode(inode);
1071 spin_unlock(&inode_lock);
1076 * ilookup5_nowait - search for an inode in the inode cache
1077 * @sb: super block of file system to search
1078 * @hashval: hash value (usually inode number) to search for
1079 * @test: callback used for comparisons between inodes
1080 * @data: opaque data pointer to pass to @test
1082 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1083 * @data in the inode cache. This is a generalized version of ilookup() for
1084 * file systems where the inode number is not sufficient for unique
1085 * identification of an inode.
1087 * If the inode is in the cache, the inode is returned with an incremented
1088 * reference count. Note, the inode lock is not waited upon so you have to be
1089 * very careful what you do with the returned inode. You probably should be
1090 * using ilookup5() instead.
1092 * Otherwise NULL is returned.
1094 * Note, @test is called with the inode_lock held, so can't sleep.
1096 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1097 int (*test)(struct inode *, void *), void *data)
1099 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1101 return ifind(sb, head, test, data, 0);
1103 EXPORT_SYMBOL(ilookup5_nowait);
1106 * ilookup5 - search for an inode in the inode cache
1107 * @sb: super block of file system to search
1108 * @hashval: hash value (usually inode number) to search for
1109 * @test: callback used for comparisons between inodes
1110 * @data: opaque data pointer to pass to @test
1112 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1113 * @data in the inode cache. This is a generalized version of ilookup() for
1114 * file systems where the inode number is not sufficient for unique
1115 * identification of an inode.
1117 * If the inode is in the cache, the inode lock is waited upon and the inode is
1118 * returned with an incremented reference count.
1120 * Otherwise NULL is returned.
1122 * Note, @test is called with the inode_lock held, so can't sleep.
1124 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1125 int (*test)(struct inode *, void *), void *data)
1127 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1129 return ifind(sb, head, test, data, 1);
1131 EXPORT_SYMBOL(ilookup5);
1134 * ilookup - search for an inode in the inode cache
1135 * @sb: super block of file system to search
1136 * @ino: inode number to search for
1138 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1139 * This is for file systems where the inode number is sufficient for unique
1140 * identification of an inode.
1142 * If the inode is in the cache, the inode is returned with an incremented
1145 * Otherwise NULL is returned.
1147 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1149 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1151 return ifind_fast(sb, head, ino);
1153 EXPORT_SYMBOL(ilookup);
1156 * iget5_locked - obtain an inode from a mounted file system
1157 * @sb: super block of file system
1158 * @hashval: hash value (usually inode number) to get
1159 * @test: callback used for comparisons between inodes
1160 * @set: callback used to initialize a new struct inode
1161 * @data: opaque data pointer to pass to @test and @set
1163 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1164 * and @data in the inode cache and if present it is returned with an increased
1165 * reference count. This is a generalized version of iget_locked() for file
1166 * systems where the inode number is not sufficient for unique identification
1169 * If the inode is not in cache, get_new_inode() is called to allocate a new
1170 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1171 * file system gets to fill it in before unlocking it via unlock_new_inode().
1173 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1175 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1176 int (*test)(struct inode *, void *),
1177 int (*set)(struct inode *, void *), void *data)
1179 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1180 struct inode *inode;
1182 inode = ifind(sb, head, test, data, 1);
1186 * get_new_inode() will do the right thing, re-trying the search
1187 * in case it had to block at any point.
1189 return get_new_inode(sb, head, test, set, data);
1191 EXPORT_SYMBOL(iget5_locked);
1194 * iget_locked - obtain an inode from a mounted file system
1195 * @sb: super block of file system
1196 * @ino: inode number to get
1198 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1199 * the inode cache and if present it is returned with an increased reference
1200 * count. This is for file systems where the inode number is sufficient for
1201 * unique identification of an inode.
1203 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1204 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1205 * The file system gets to fill it in before unlocking it via
1206 * unlock_new_inode().
1208 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1210 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1211 struct inode *inode;
1213 inode = ifind_fast(sb, head, ino);
1217 * get_new_inode_fast() will do the right thing, re-trying the search
1218 * in case it had to block at any point.
1220 return get_new_inode_fast(sb, head, ino);
1222 EXPORT_SYMBOL(iget_locked);
1224 int insert_inode_locked(struct inode *inode)
1226 struct super_block *sb = inode->i_sb;
1227 ino_t ino = inode->i_ino;
1228 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1230 inode->i_state |= I_NEW;
1232 struct hlist_node *node;
1233 struct inode *old = NULL;
1234 spin_lock(&inode_lock);
1235 hlist_for_each_entry(old, node, head, i_hash) {
1236 if (old->i_ino != ino)
1238 if (old->i_sb != sb)
1240 if (old->i_state & (I_FREEING|I_WILL_FREE))
1244 if (likely(!node)) {
1245 hlist_add_head(&inode->i_hash, head);
1246 spin_unlock(&inode_lock);
1250 spin_unlock(&inode_lock);
1252 if (unlikely(!inode_unhashed(old))) {
1259 EXPORT_SYMBOL(insert_inode_locked);
1261 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1262 int (*test)(struct inode *, void *), void *data)
1264 struct super_block *sb = inode->i_sb;
1265 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1267 inode->i_state |= I_NEW;
1270 struct hlist_node *node;
1271 struct inode *old = NULL;
1273 spin_lock(&inode_lock);
1274 hlist_for_each_entry(old, node, head, i_hash) {
1275 if (old->i_sb != sb)
1277 if (!test(old, data))
1279 if (old->i_state & (I_FREEING|I_WILL_FREE))
1283 if (likely(!node)) {
1284 hlist_add_head(&inode->i_hash, head);
1285 spin_unlock(&inode_lock);
1289 spin_unlock(&inode_lock);
1291 if (unlikely(!inode_unhashed(old))) {
1298 EXPORT_SYMBOL(insert_inode_locked4);
1301 int generic_delete_inode(struct inode *inode)
1305 EXPORT_SYMBOL(generic_delete_inode);
1308 * Normal UNIX filesystem behaviour: delete the
1309 * inode when the usage count drops to zero, and
1312 int generic_drop_inode(struct inode *inode)
1314 return !inode->i_nlink || inode_unhashed(inode);
1316 EXPORT_SYMBOL_GPL(generic_drop_inode);
1319 * Called when we're dropping the last reference
1322 * Call the FS "drop_inode()" function, defaulting to
1323 * the legacy UNIX filesystem behaviour. If it tells
1324 * us to evict inode, do so. Otherwise, retain inode
1325 * in cache if fs is alive, sync and evict if fs is
1328 static void iput_final(struct inode *inode)
1330 struct super_block *sb = inode->i_sb;
1331 const struct super_operations *op = inode->i_sb->s_op;
1334 if (op && op->drop_inode)
1335 drop = op->drop_inode(inode);
1337 drop = generic_drop_inode(inode);
1340 if (sb->s_flags & MS_ACTIVE) {
1341 inode->i_state |= I_REFERENCED;
1342 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1343 inode_lru_list_add(inode);
1345 spin_unlock(&inode_lock);
1348 WARN_ON(inode->i_state & I_NEW);
1349 inode->i_state |= I_WILL_FREE;
1350 spin_unlock(&inode_lock);
1351 write_inode_now(inode, 1);
1352 spin_lock(&inode_lock);
1353 WARN_ON(inode->i_state & I_NEW);
1354 inode->i_state &= ~I_WILL_FREE;
1355 __remove_inode_hash(inode);
1358 WARN_ON(inode->i_state & I_NEW);
1359 inode->i_state |= I_FREEING;
1362 * Move the inode off the IO lists and LRU once I_FREEING is
1363 * set so that it won't get moved back on there if it is dirty.
1365 inode_lru_list_del(inode);
1366 list_del_init(&inode->i_wb_list);
1368 __inode_sb_list_del(inode);
1369 spin_unlock(&inode_lock);
1371 remove_inode_hash(inode);
1372 wake_up_inode(inode);
1373 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1374 destroy_inode(inode);
1378 * iput - put an inode
1379 * @inode: inode to put
1381 * Puts an inode, dropping its usage count. If the inode use count hits
1382 * zero, the inode is then freed and may also be destroyed.
1384 * Consequently, iput() can sleep.
1386 void iput(struct inode *inode)
1389 BUG_ON(inode->i_state & I_CLEAR);
1391 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1395 EXPORT_SYMBOL(iput);
1398 * bmap - find a block number in a file
1399 * @inode: inode of file
1400 * @block: block to find
1402 * Returns the block number on the device holding the inode that
1403 * is the disk block number for the block of the file requested.
1404 * That is, asked for block 4 of inode 1 the function will return the
1405 * disk block relative to the disk start that holds that block of the
1408 sector_t bmap(struct inode *inode, sector_t block)
1411 if (inode->i_mapping->a_ops->bmap)
1412 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1415 EXPORT_SYMBOL(bmap);
1418 * With relative atime, only update atime if the previous atime is
1419 * earlier than either the ctime or mtime or if at least a day has
1420 * passed since the last atime update.
1422 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1423 struct timespec now)
1426 if (!(mnt->mnt_flags & MNT_RELATIME))
1429 * Is mtime younger than atime? If yes, update atime:
1431 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1434 * Is ctime younger than atime? If yes, update atime:
1436 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1440 * Is the previous atime value older than a day? If yes,
1443 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1446 * Good, we can skip the atime update:
1452 * touch_atime - update the access time
1453 * @mnt: mount the inode is accessed on
1454 * @dentry: dentry accessed
1456 * Update the accessed time on an inode and mark it for writeback.
1457 * This function automatically handles read only file systems and media,
1458 * as well as the "noatime" flag and inode specific "noatime" markers.
1460 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1462 struct inode *inode = dentry->d_inode;
1463 struct timespec now;
1465 if (inode->i_flags & S_NOATIME)
1467 if (IS_NOATIME(inode))
1469 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1472 if (mnt->mnt_flags & MNT_NOATIME)
1474 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1477 now = current_fs_time(inode->i_sb);
1479 if (!relatime_need_update(mnt, inode, now))
1482 if (timespec_equal(&inode->i_atime, &now))
1485 if (mnt_want_write(mnt))
1488 inode->i_atime = now;
1489 mark_inode_dirty_sync(inode);
1490 mnt_drop_write(mnt);
1492 EXPORT_SYMBOL(touch_atime);
1495 * file_update_time - update mtime and ctime time
1496 * @file: file accessed
1498 * Update the mtime and ctime members of an inode and mark the inode
1499 * for writeback. Note that this function is meant exclusively for
1500 * usage in the file write path of filesystems, and filesystems may
1501 * choose to explicitly ignore update via this function with the
1502 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1503 * timestamps are handled by the server.
1506 void file_update_time(struct file *file)
1508 struct inode *inode = file->f_path.dentry->d_inode;
1509 struct timespec now;
1510 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1512 /* First try to exhaust all avenues to not sync */
1513 if (IS_NOCMTIME(inode))
1516 now = current_fs_time(inode->i_sb);
1517 if (!timespec_equal(&inode->i_mtime, &now))
1520 if (!timespec_equal(&inode->i_ctime, &now))
1523 if (IS_I_VERSION(inode))
1524 sync_it |= S_VERSION;
1529 /* Finally allowed to write? Takes lock. */
1530 if (mnt_want_write_file(file))
1533 /* Only change inode inside the lock region */
1534 if (sync_it & S_VERSION)
1535 inode_inc_iversion(inode);
1536 if (sync_it & S_CTIME)
1537 inode->i_ctime = now;
1538 if (sync_it & S_MTIME)
1539 inode->i_mtime = now;
1540 mark_inode_dirty_sync(inode);
1541 mnt_drop_write(file->f_path.mnt);
1543 EXPORT_SYMBOL(file_update_time);
1545 int inode_needs_sync(struct inode *inode)
1549 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1553 EXPORT_SYMBOL(inode_needs_sync);
1555 int inode_wait(void *word)
1560 EXPORT_SYMBOL(inode_wait);
1563 * If we try to find an inode in the inode hash while it is being
1564 * deleted, we have to wait until the filesystem completes its
1565 * deletion before reporting that it isn't found. This function waits
1566 * until the deletion _might_ have completed. Callers are responsible
1567 * to recheck inode state.
1569 * It doesn't matter if I_NEW is not set initially, a call to
1570 * wake_up_inode() after removing from the hash list will DTRT.
1572 * This is called with inode_lock held.
1574 static void __wait_on_freeing_inode(struct inode *inode)
1576 wait_queue_head_t *wq;
1577 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1578 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1579 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1580 spin_unlock(&inode_lock);
1582 finish_wait(wq, &wait.wait);
1583 spin_lock(&inode_lock);
1586 static __initdata unsigned long ihash_entries;
1587 static int __init set_ihash_entries(char *str)
1591 ihash_entries = simple_strtoul(str, &str, 0);
1594 __setup("ihash_entries=", set_ihash_entries);
1597 * Initialize the waitqueues and inode hash table.
1599 void __init inode_init_early(void)
1603 /* If hashes are distributed across NUMA nodes, defer
1604 * hash allocation until vmalloc space is available.
1610 alloc_large_system_hash("Inode-cache",
1611 sizeof(struct hlist_head),
1619 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1620 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1623 void __init inode_init(void)
1627 /* inode slab cache */
1628 inode_cachep = kmem_cache_create("inode_cache",
1629 sizeof(struct inode),
1631 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1634 register_shrinker(&icache_shrinker);
1635 percpu_counter_init(&nr_inodes, 0);
1636 percpu_counter_init(&nr_inodes_unused, 0);
1638 /* Hash may have been set up in inode_init_early */
1643 alloc_large_system_hash("Inode-cache",
1644 sizeof(struct hlist_head),
1652 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1653 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1656 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1658 inode->i_mode = mode;
1659 if (S_ISCHR(mode)) {
1660 inode->i_fop = &def_chr_fops;
1661 inode->i_rdev = rdev;
1662 } else if (S_ISBLK(mode)) {
1663 inode->i_fop = &def_blk_fops;
1664 inode->i_rdev = rdev;
1665 } else if (S_ISFIFO(mode))
1666 inode->i_fop = &def_fifo_fops;
1667 else if (S_ISSOCK(mode))
1668 inode->i_fop = &bad_sock_fops;
1670 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1671 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1674 EXPORT_SYMBOL(init_special_inode);
1677 * Init uid,gid,mode for new inode according to posix standards
1679 * @dir: Directory inode
1680 * @mode: mode of the new inode
1682 void inode_init_owner(struct inode *inode, const struct inode *dir,
1685 inode->i_uid = current_fsuid();
1686 if (dir && dir->i_mode & S_ISGID) {
1687 inode->i_gid = dir->i_gid;
1691 inode->i_gid = current_fsgid();
1692 inode->i_mode = mode;
1694 EXPORT_SYMBOL(inode_init_owner);