2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
7 #include <linux/dcache.h>
8 #include <linux/init.h>
9 #include <linux/slab.h>
10 #include <linux/writeback.h>
11 #include <linux/module.h>
12 #include <linux/backing-dev.h>
13 #include <linux/wait.h>
14 #include <linux/rwsem.h>
15 #include <linux/hash.h>
16 #include <linux/swap.h>
17 #include <linux/security.h>
18 #include <linux/pagemap.h>
19 #include <linux/cdev.h>
20 #include <linux/bootmem.h>
21 #include <linux/fsnotify.h>
22 #include <linux/mount.h>
23 #include <linux/async.h>
24 #include <linux/posix_acl.h>
25 #include <linux/prefetch.h>
26 #include <linux/ima.h>
27 #include <linux/cred.h>
28 #include <linux/buffer_head.h> /* for inode_has_buffers */
32 * Inode locking rules:
34 * inode->i_lock protects:
35 * inode->i_state, inode->i_hash, __iget()
36 * inode->i_sb->s_inode_lru_lock protects:
37 * inode->i_sb->s_inode_lru, inode->i_lru
38 * inode_sb_list_lock protects:
39 * sb->s_inodes, inode->i_sb_list
40 * inode_wb_list_lock protects:
41 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
42 * inode_hash_lock protects:
43 * inode_hashtable, inode->i_hash
49 * inode->i_sb->s_inode_lru_lock
62 static unsigned int i_hash_mask __read_mostly;
63 static unsigned int i_hash_shift __read_mostly;
64 static struct hlist_head *inode_hashtable __read_mostly;
65 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
67 __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
68 __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_wb_list_lock);
71 * Empty aops. Can be used for the cases where the user does not
72 * define any of the address_space operations.
74 const struct address_space_operations empty_aops = {
76 EXPORT_SYMBOL(empty_aops);
79 * Statistics gathering..
81 struct inodes_stat_t inodes_stat;
83 static DEFINE_PER_CPU(unsigned int, nr_inodes);
84 static DEFINE_PER_CPU(unsigned int, nr_unused);
86 static struct kmem_cache *inode_cachep __read_mostly;
88 static int get_nr_inodes(void)
92 for_each_possible_cpu(i)
93 sum += per_cpu(nr_inodes, i);
94 return sum < 0 ? 0 : sum;
97 static inline int get_nr_inodes_unused(void)
101 for_each_possible_cpu(i)
102 sum += per_cpu(nr_unused, i);
103 return sum < 0 ? 0 : sum;
106 int get_nr_dirty_inodes(void)
108 /* not actually dirty inodes, but a wild approximation */
109 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
110 return nr_dirty > 0 ? nr_dirty : 0;
114 * Handle nr_inode sysctl
117 int proc_nr_inodes(ctl_table *table, int write,
118 void __user *buffer, size_t *lenp, loff_t *ppos)
120 inodes_stat.nr_inodes = get_nr_inodes();
121 inodes_stat.nr_unused = get_nr_inodes_unused();
122 return proc_dointvec(table, write, buffer, lenp, ppos);
127 * inode_init_always - perform inode structure intialisation
128 * @sb: superblock inode belongs to
129 * @inode: inode to initialise
131 * These are initializations that need to be done on every inode
132 * allocation as the fields are not initialised by slab allocation.
134 int inode_init_always(struct super_block *sb, struct inode *inode)
136 static const struct inode_operations empty_iops;
137 static const struct file_operations empty_fops;
138 struct address_space *const mapping = &inode->i_data;
141 inode->i_blkbits = sb->s_blocksize_bits;
143 atomic_set(&inode->i_count, 1);
144 inode->i_op = &empty_iops;
145 inode->i_fop = &empty_fops;
149 atomic_set(&inode->i_writecount, 0);
153 inode->i_generation = 0;
155 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
157 inode->i_pipe = NULL;
158 inode->i_bdev = NULL;
159 inode->i_cdev = NULL;
161 inode->dirtied_when = 0;
163 if (security_inode_alloc(inode))
165 spin_lock_init(&inode->i_lock);
166 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
168 mutex_init(&inode->i_mutex);
169 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
171 atomic_set(&inode->i_dio_count, 0);
173 mapping->a_ops = &empty_aops;
174 mapping->host = inode;
176 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
177 mapping->assoc_mapping = NULL;
178 mapping->backing_dev_info = &default_backing_dev_info;
179 mapping->writeback_index = 0;
182 * If the block_device provides a backing_dev_info for client
183 * inodes then use that. Otherwise the inode share the bdev's
187 struct backing_dev_info *bdi;
189 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
190 mapping->backing_dev_info = bdi;
192 inode->i_private = NULL;
193 inode->i_mapping = mapping;
194 #ifdef CONFIG_FS_POSIX_ACL
195 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
198 #ifdef CONFIG_FSNOTIFY
199 inode->i_fsnotify_mask = 0;
202 this_cpu_inc(nr_inodes);
208 EXPORT_SYMBOL(inode_init_always);
210 static struct inode *alloc_inode(struct super_block *sb)
214 if (sb->s_op->alloc_inode)
215 inode = sb->s_op->alloc_inode(sb);
217 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
222 if (unlikely(inode_init_always(sb, inode))) {
223 if (inode->i_sb->s_op->destroy_inode)
224 inode->i_sb->s_op->destroy_inode(inode);
226 kmem_cache_free(inode_cachep, inode);
233 void free_inode_nonrcu(struct inode *inode)
235 kmem_cache_free(inode_cachep, inode);
237 EXPORT_SYMBOL(free_inode_nonrcu);
239 void __destroy_inode(struct inode *inode)
241 BUG_ON(inode_has_buffers(inode));
242 security_inode_free(inode);
243 fsnotify_inode_delete(inode);
244 #ifdef CONFIG_FS_POSIX_ACL
245 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
246 posix_acl_release(inode->i_acl);
247 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
248 posix_acl_release(inode->i_default_acl);
250 this_cpu_dec(nr_inodes);
252 EXPORT_SYMBOL(__destroy_inode);
254 static void i_callback(struct rcu_head *head)
256 struct inode *inode = container_of(head, struct inode, i_rcu);
257 INIT_LIST_HEAD(&inode->i_dentry);
258 kmem_cache_free(inode_cachep, inode);
261 static void destroy_inode(struct inode *inode)
263 BUG_ON(!list_empty(&inode->i_lru));
264 __destroy_inode(inode);
265 if (inode->i_sb->s_op->destroy_inode)
266 inode->i_sb->s_op->destroy_inode(inode);
268 call_rcu(&inode->i_rcu, i_callback);
271 void address_space_init_once(struct address_space *mapping)
273 memset(mapping, 0, sizeof(*mapping));
274 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
275 spin_lock_init(&mapping->tree_lock);
276 mutex_init(&mapping->i_mmap_mutex);
277 INIT_LIST_HEAD(&mapping->private_list);
278 spin_lock_init(&mapping->private_lock);
279 INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
280 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
282 EXPORT_SYMBOL(address_space_init_once);
285 * These are initializations that only need to be done
286 * once, because the fields are idempotent across use
287 * of the inode, so let the slab aware of that.
289 void inode_init_once(struct inode *inode)
291 memset(inode, 0, sizeof(*inode));
292 INIT_HLIST_NODE(&inode->i_hash);
293 INIT_LIST_HEAD(&inode->i_dentry);
294 INIT_LIST_HEAD(&inode->i_devices);
295 INIT_LIST_HEAD(&inode->i_wb_list);
296 INIT_LIST_HEAD(&inode->i_lru);
297 address_space_init_once(&inode->i_data);
298 i_size_ordered_init(inode);
299 #ifdef CONFIG_FSNOTIFY
300 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
303 EXPORT_SYMBOL(inode_init_once);
305 static void init_once(void *foo)
307 struct inode *inode = (struct inode *) foo;
309 inode_init_once(inode);
313 * inode->i_lock must be held
315 void __iget(struct inode *inode)
317 atomic_inc(&inode->i_count);
321 * get additional reference to inode; caller must already hold one.
323 void ihold(struct inode *inode)
325 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
327 EXPORT_SYMBOL(ihold);
329 static void inode_lru_list_add(struct inode *inode)
331 spin_lock(&inode->i_sb->s_inode_lru_lock);
332 if (list_empty(&inode->i_lru)) {
333 list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
334 inode->i_sb->s_nr_inodes_unused++;
335 this_cpu_inc(nr_unused);
337 spin_unlock(&inode->i_sb->s_inode_lru_lock);
340 static void inode_lru_list_del(struct inode *inode)
342 spin_lock(&inode->i_sb->s_inode_lru_lock);
343 if (!list_empty(&inode->i_lru)) {
344 list_del_init(&inode->i_lru);
345 inode->i_sb->s_nr_inodes_unused--;
346 this_cpu_dec(nr_unused);
348 spin_unlock(&inode->i_sb->s_inode_lru_lock);
352 * inode_sb_list_add - add inode to the superblock list of inodes
353 * @inode: inode to add
355 void inode_sb_list_add(struct inode *inode)
357 spin_lock(&inode_sb_list_lock);
358 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
359 spin_unlock(&inode_sb_list_lock);
361 EXPORT_SYMBOL_GPL(inode_sb_list_add);
363 static inline void inode_sb_list_del(struct inode *inode)
365 if (!list_empty(&inode->i_sb_list)) {
366 spin_lock(&inode_sb_list_lock);
367 list_del_init(&inode->i_sb_list);
368 spin_unlock(&inode_sb_list_lock);
372 static unsigned long hash(struct super_block *sb, unsigned long hashval)
376 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
378 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
379 return tmp & i_hash_mask;
383 * __insert_inode_hash - hash an inode
384 * @inode: unhashed inode
385 * @hashval: unsigned long value used to locate this object in the
388 * Add an inode to the inode hash for this superblock.
390 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
392 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
394 spin_lock(&inode_hash_lock);
395 spin_lock(&inode->i_lock);
396 hlist_add_head(&inode->i_hash, b);
397 spin_unlock(&inode->i_lock);
398 spin_unlock(&inode_hash_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 void remove_inode_hash(struct inode *inode)
410 spin_lock(&inode_hash_lock);
411 spin_lock(&inode->i_lock);
412 hlist_del_init(&inode->i_hash);
413 spin_unlock(&inode->i_lock);
414 spin_unlock(&inode_hash_lock);
416 EXPORT_SYMBOL(remove_inode_hash);
418 void end_writeback(struct inode *inode)
422 * We have to cycle tree_lock here because reclaim can be still in the
423 * process of removing the last page (in __delete_from_page_cache())
424 * and we must not free mapping under it.
426 spin_lock_irq(&inode->i_data.tree_lock);
427 BUG_ON(inode->i_data.nrpages);
428 spin_unlock_irq(&inode->i_data.tree_lock);
429 BUG_ON(!list_empty(&inode->i_data.private_list));
430 BUG_ON(!(inode->i_state & I_FREEING));
431 BUG_ON(inode->i_state & I_CLEAR);
432 inode_sync_wait(inode);
433 /* don't need i_lock here, no concurrent mods to i_state */
434 inode->i_state = I_FREEING | I_CLEAR;
436 EXPORT_SYMBOL(end_writeback);
439 * Free the inode passed in, removing it from the lists it is still connected
440 * to. We remove any pages still attached to the inode and wait for any IO that
441 * is still in progress before finally destroying the inode.
443 * An inode must already be marked I_FREEING so that we avoid the inode being
444 * moved back onto lists if we race with other code that manipulates the lists
445 * (e.g. writeback_single_inode). The caller is responsible for setting this.
447 * An inode must already be removed from the LRU list before being evicted from
448 * the cache. This should occur atomically with setting the I_FREEING state
449 * flag, so no inodes here should ever be on the LRU when being evicted.
451 static void evict(struct inode *inode)
453 const struct super_operations *op = inode->i_sb->s_op;
455 BUG_ON(!(inode->i_state & I_FREEING));
456 BUG_ON(!list_empty(&inode->i_lru));
458 inode_wb_list_del(inode);
459 inode_sb_list_del(inode);
461 if (op->evict_inode) {
462 op->evict_inode(inode);
464 if (inode->i_data.nrpages)
465 truncate_inode_pages(&inode->i_data, 0);
466 end_writeback(inode);
468 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
470 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
473 remove_inode_hash(inode);
475 spin_lock(&inode->i_lock);
476 wake_up_bit(&inode->i_state, __I_NEW);
477 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
478 spin_unlock(&inode->i_lock);
480 destroy_inode(inode);
484 * dispose_list - dispose of the contents of a local list
485 * @head: the head of the list to free
487 * Dispose-list gets a local list with local inodes in it, so it doesn't
488 * need to worry about list corruption and SMP locks.
490 static void dispose_list(struct list_head *head)
492 while (!list_empty(head)) {
495 inode = list_first_entry(head, struct inode, i_lru);
496 list_del_init(&inode->i_lru);
503 * evict_inodes - evict all evictable inodes for a superblock
504 * @sb: superblock to operate on
506 * Make sure that no inodes with zero refcount are retained. This is
507 * called by superblock shutdown after having MS_ACTIVE flag removed,
508 * so any inode reaching zero refcount during or after that call will
509 * be immediately evicted.
511 void evict_inodes(struct super_block *sb)
513 struct inode *inode, *next;
516 spin_lock(&inode_sb_list_lock);
517 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
518 if (atomic_read(&inode->i_count))
521 spin_lock(&inode->i_lock);
522 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
523 spin_unlock(&inode->i_lock);
527 inode->i_state |= I_FREEING;
528 inode_lru_list_del(inode);
529 spin_unlock(&inode->i_lock);
530 list_add(&inode->i_lru, &dispose);
532 spin_unlock(&inode_sb_list_lock);
534 dispose_list(&dispose);
538 * invalidate_inodes - attempt to free all inodes on a superblock
539 * @sb: superblock to operate on
540 * @kill_dirty: flag to guide handling of dirty inodes
542 * Attempts to free all inodes for a given superblock. If there were any
543 * busy inodes return a non-zero value, else zero.
544 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
547 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
550 struct inode *inode, *next;
553 spin_lock(&inode_sb_list_lock);
554 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
555 spin_lock(&inode->i_lock);
556 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
557 spin_unlock(&inode->i_lock);
560 if (inode->i_state & I_DIRTY && !kill_dirty) {
561 spin_unlock(&inode->i_lock);
565 if (atomic_read(&inode->i_count)) {
566 spin_unlock(&inode->i_lock);
571 inode->i_state |= I_FREEING;
572 inode_lru_list_del(inode);
573 spin_unlock(&inode->i_lock);
574 list_add(&inode->i_lru, &dispose);
576 spin_unlock(&inode_sb_list_lock);
578 dispose_list(&dispose);
583 static int can_unuse(struct inode *inode)
585 if (inode->i_state & ~I_REFERENCED)
587 if (inode_has_buffers(inode))
589 if (atomic_read(&inode->i_count))
591 if (inode->i_data.nrpages)
597 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
598 * This is called from the superblock shrinker function with a number of inodes
599 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
600 * then are freed outside inode_lock by dispose_list().
602 * Any inodes which are pinned purely because of attached pagecache have their
603 * pagecache removed. If the inode has metadata buffers attached to
604 * mapping->private_list then try to remove them.
606 * If the inode has the I_REFERENCED flag set, then it means that it has been
607 * used recently - the flag is set in iput_final(). When we encounter such an
608 * inode, clear the flag and move it to the back of the LRU so it gets another
609 * pass through the LRU before it gets reclaimed. This is necessary because of
610 * the fact we are doing lazy LRU updates to minimise lock contention so the
611 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
612 * with this flag set because they are the inodes that are out of order.
614 void prune_icache_sb(struct super_block *sb, int nr_to_scan)
618 unsigned long reap = 0;
620 spin_lock(&sb->s_inode_lru_lock);
621 for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
624 if (list_empty(&sb->s_inode_lru))
627 inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
630 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
631 * so use a trylock. If we fail to get the lock, just move the
632 * inode to the back of the list so we don't spin on it.
634 if (!spin_trylock(&inode->i_lock)) {
635 list_move(&inode->i_lru, &sb->s_inode_lru);
640 * Referenced or dirty inodes are still in use. Give them
641 * another pass through the LRU as we canot reclaim them now.
643 if (atomic_read(&inode->i_count) ||
644 (inode->i_state & ~I_REFERENCED)) {
645 list_del_init(&inode->i_lru);
646 spin_unlock(&inode->i_lock);
647 sb->s_nr_inodes_unused--;
648 this_cpu_dec(nr_unused);
652 /* recently referenced inodes get one more pass */
653 if (inode->i_state & I_REFERENCED) {
654 inode->i_state &= ~I_REFERENCED;
655 list_move(&inode->i_lru, &sb->s_inode_lru);
656 spin_unlock(&inode->i_lock);
659 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
661 spin_unlock(&inode->i_lock);
662 spin_unlock(&sb->s_inode_lru_lock);
663 if (remove_inode_buffers(inode))
664 reap += invalidate_mapping_pages(&inode->i_data,
667 spin_lock(&sb->s_inode_lru_lock);
669 if (inode != list_entry(sb->s_inode_lru.next,
670 struct inode, i_lru))
671 continue; /* wrong inode or list_empty */
672 /* avoid lock inversions with trylock */
673 if (!spin_trylock(&inode->i_lock))
675 if (!can_unuse(inode)) {
676 spin_unlock(&inode->i_lock);
680 WARN_ON(inode->i_state & I_NEW);
681 inode->i_state |= I_FREEING;
682 spin_unlock(&inode->i_lock);
684 list_move(&inode->i_lru, &freeable);
685 sb->s_nr_inodes_unused--;
686 this_cpu_dec(nr_unused);
688 if (current_is_kswapd())
689 __count_vm_events(KSWAPD_INODESTEAL, reap);
691 __count_vm_events(PGINODESTEAL, reap);
692 spin_unlock(&sb->s_inode_lru_lock);
694 dispose_list(&freeable);
697 static void __wait_on_freeing_inode(struct inode *inode);
699 * Called with the inode lock held.
701 static struct inode *find_inode(struct super_block *sb,
702 struct hlist_head *head,
703 int (*test)(struct inode *, void *),
706 struct hlist_node *node;
707 struct inode *inode = NULL;
710 hlist_for_each_entry(inode, node, head, i_hash) {
711 spin_lock(&inode->i_lock);
712 if (inode->i_sb != sb) {
713 spin_unlock(&inode->i_lock);
716 if (!test(inode, data)) {
717 spin_unlock(&inode->i_lock);
720 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
721 __wait_on_freeing_inode(inode);
725 spin_unlock(&inode->i_lock);
732 * find_inode_fast is the fast path version of find_inode, see the comment at
733 * iget_locked for details.
735 static struct inode *find_inode_fast(struct super_block *sb,
736 struct hlist_head *head, unsigned long ino)
738 struct hlist_node *node;
739 struct inode *inode = NULL;
742 hlist_for_each_entry(inode, node, head, i_hash) {
743 spin_lock(&inode->i_lock);
744 if (inode->i_ino != ino) {
745 spin_unlock(&inode->i_lock);
748 if (inode->i_sb != sb) {
749 spin_unlock(&inode->i_lock);
752 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
753 __wait_on_freeing_inode(inode);
757 spin_unlock(&inode->i_lock);
764 * Each cpu owns a range of LAST_INO_BATCH numbers.
765 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
766 * to renew the exhausted range.
768 * This does not significantly increase overflow rate because every CPU can
769 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
770 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
771 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
772 * overflow rate by 2x, which does not seem too significant.
774 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
775 * error if st_ino won't fit in target struct field. Use 32bit counter
776 * here to attempt to avoid that.
778 #define LAST_INO_BATCH 1024
779 static DEFINE_PER_CPU(unsigned int, last_ino);
781 unsigned int get_next_ino(void)
783 unsigned int *p = &get_cpu_var(last_ino);
784 unsigned int res = *p;
787 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
788 static atomic_t shared_last_ino;
789 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
791 res = next - LAST_INO_BATCH;
796 put_cpu_var(last_ino);
799 EXPORT_SYMBOL(get_next_ino);
802 * new_inode_pseudo - obtain an inode
805 * Allocates a new inode for given superblock.
806 * Inode wont be chained in superblock s_inodes list
808 * - fs can't be unmount
809 * - quotas, fsnotify, writeback can't work
811 struct inode *new_inode_pseudo(struct super_block *sb)
813 struct inode *inode = alloc_inode(sb);
816 spin_lock(&inode->i_lock);
818 spin_unlock(&inode->i_lock);
819 INIT_LIST_HEAD(&inode->i_sb_list);
825 * new_inode - obtain an inode
828 * Allocates a new inode for given superblock. The default gfp_mask
829 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
830 * If HIGHMEM pages are unsuitable or it is known that pages allocated
831 * for the page cache are not reclaimable or migratable,
832 * mapping_set_gfp_mask() must be called with suitable flags on the
833 * newly created inode's mapping
836 struct inode *new_inode(struct super_block *sb)
840 spin_lock_prefetch(&inode_sb_list_lock);
842 inode = new_inode_pseudo(sb);
844 inode_sb_list_add(inode);
847 EXPORT_SYMBOL(new_inode);
850 * unlock_new_inode - clear the I_NEW state and wake up any waiters
851 * @inode: new inode to unlock
853 * Called when the inode is fully initialised to clear the new state of the
854 * inode and wake up anyone waiting for the inode to finish initialisation.
856 void unlock_new_inode(struct inode *inode)
858 #ifdef CONFIG_DEBUG_LOCK_ALLOC
859 if (S_ISDIR(inode->i_mode)) {
860 struct file_system_type *type = inode->i_sb->s_type;
862 /* Set new key only if filesystem hasn't already changed it */
863 if (!lockdep_match_class(&inode->i_mutex,
864 &type->i_mutex_key)) {
866 * ensure nobody is actually holding i_mutex
868 mutex_destroy(&inode->i_mutex);
869 mutex_init(&inode->i_mutex);
870 lockdep_set_class(&inode->i_mutex,
871 &type->i_mutex_dir_key);
875 spin_lock(&inode->i_lock);
876 WARN_ON(!(inode->i_state & I_NEW));
877 inode->i_state &= ~I_NEW;
878 wake_up_bit(&inode->i_state, __I_NEW);
879 spin_unlock(&inode->i_lock);
881 EXPORT_SYMBOL(unlock_new_inode);
884 * iget5_locked - obtain an inode from a mounted file system
885 * @sb: super block of file system
886 * @hashval: hash value (usually inode number) to get
887 * @test: callback used for comparisons between inodes
888 * @set: callback used to initialize a new struct inode
889 * @data: opaque data pointer to pass to @test and @set
891 * Search for the inode specified by @hashval and @data in the inode cache,
892 * and if present it is return it with an increased reference count. This is
893 * a generalized version of iget_locked() for file systems where the inode
894 * number is not sufficient for unique identification of an inode.
896 * If the inode is not in cache, allocate a new inode and return it locked,
897 * hashed, and with the I_NEW flag set. The file system gets to fill it in
898 * before unlocking it via unlock_new_inode().
900 * Note both @test and @set are called with the inode_hash_lock held, so can't
903 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
904 int (*test)(struct inode *, void *),
905 int (*set)(struct inode *, void *), void *data)
907 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
910 spin_lock(&inode_hash_lock);
911 inode = find_inode(sb, head, test, data);
912 spin_unlock(&inode_hash_lock);
915 wait_on_inode(inode);
919 inode = alloc_inode(sb);
923 spin_lock(&inode_hash_lock);
924 /* We released the lock, so.. */
925 old = find_inode(sb, head, test, data);
927 if (set(inode, data))
930 spin_lock(&inode->i_lock);
931 inode->i_state = I_NEW;
932 hlist_add_head(&inode->i_hash, head);
933 spin_unlock(&inode->i_lock);
934 inode_sb_list_add(inode);
935 spin_unlock(&inode_hash_lock);
937 /* Return the locked inode with I_NEW set, the
938 * caller is responsible for filling in the contents
944 * Uhhuh, somebody else created the same inode under
945 * us. Use the old inode instead of the one we just
948 spin_unlock(&inode_hash_lock);
949 destroy_inode(inode);
951 wait_on_inode(inode);
956 spin_unlock(&inode_hash_lock);
957 destroy_inode(inode);
960 EXPORT_SYMBOL(iget5_locked);
963 * iget_locked - obtain an inode from a mounted file system
964 * @sb: super block of file system
965 * @ino: inode number to get
967 * Search for the inode specified by @ino in the inode cache and if present
968 * return it with an increased reference count. This is for file systems
969 * where the inode number is sufficient for unique identification of an inode.
971 * If the inode is not in cache, allocate a new inode and return it locked,
972 * hashed, and with the I_NEW flag set. The file system gets to fill it in
973 * before unlocking it via unlock_new_inode().
975 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
977 struct hlist_head *head = inode_hashtable + hash(sb, ino);
980 spin_lock(&inode_hash_lock);
981 inode = find_inode_fast(sb, head, ino);
982 spin_unlock(&inode_hash_lock);
984 wait_on_inode(inode);
988 inode = alloc_inode(sb);
992 spin_lock(&inode_hash_lock);
993 /* We released the lock, so.. */
994 old = find_inode_fast(sb, head, ino);
997 spin_lock(&inode->i_lock);
998 inode->i_state = I_NEW;
999 hlist_add_head(&inode->i_hash, head);
1000 spin_unlock(&inode->i_lock);
1001 inode_sb_list_add(inode);
1002 spin_unlock(&inode_hash_lock);
1004 /* Return the locked inode with I_NEW set, the
1005 * caller is responsible for filling in the contents
1011 * Uhhuh, somebody else created the same inode under
1012 * us. Use the old inode instead of the one we just
1015 spin_unlock(&inode_hash_lock);
1016 destroy_inode(inode);
1018 wait_on_inode(inode);
1022 EXPORT_SYMBOL(iget_locked);
1025 * search the inode cache for a matching inode number.
1026 * If we find one, then the inode number we are trying to
1027 * allocate is not unique and so we should not use it.
1029 * Returns 1 if the inode number is unique, 0 if it is not.
1031 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1033 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1034 struct hlist_node *node;
1035 struct inode *inode;
1037 spin_lock(&inode_hash_lock);
1038 hlist_for_each_entry(inode, node, b, i_hash) {
1039 if (inode->i_ino == ino && inode->i_sb == sb) {
1040 spin_unlock(&inode_hash_lock);
1044 spin_unlock(&inode_hash_lock);
1050 * iunique - get a unique inode number
1052 * @max_reserved: highest reserved inode number
1054 * Obtain an inode number that is unique on the system for a given
1055 * superblock. This is used by file systems that have no natural
1056 * permanent inode numbering system. An inode number is returned that
1057 * is higher than the reserved limit but unique.
1060 * With a large number of inodes live on the file system this function
1061 * currently becomes quite slow.
1063 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1066 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1067 * error if st_ino won't fit in target struct field. Use 32bit counter
1068 * here to attempt to avoid that.
1070 static DEFINE_SPINLOCK(iunique_lock);
1071 static unsigned int counter;
1074 spin_lock(&iunique_lock);
1076 if (counter <= max_reserved)
1077 counter = max_reserved + 1;
1079 } while (!test_inode_iunique(sb, res));
1080 spin_unlock(&iunique_lock);
1084 EXPORT_SYMBOL(iunique);
1086 struct inode *igrab(struct inode *inode)
1088 spin_lock(&inode->i_lock);
1089 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1091 spin_unlock(&inode->i_lock);
1093 spin_unlock(&inode->i_lock);
1095 * Handle the case where s_op->clear_inode is not been
1096 * called yet, and somebody is calling igrab
1097 * while the inode is getting freed.
1103 EXPORT_SYMBOL(igrab);
1106 * ilookup5_nowait - 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 * Search for the inode specified by @hashval and @data in the inode cache.
1113 * If the inode is in the cache, the inode is returned with an incremented
1116 * Note: I_NEW is not waited upon so you have to be very careful what you do
1117 * with the returned inode. You probably should be using ilookup5() instead.
1119 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1121 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1122 int (*test)(struct inode *, void *), void *data)
1124 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1125 struct inode *inode;
1127 spin_lock(&inode_hash_lock);
1128 inode = find_inode(sb, head, test, data);
1129 spin_unlock(&inode_hash_lock);
1133 EXPORT_SYMBOL(ilookup5_nowait);
1136 * ilookup5 - search for an inode in the inode cache
1137 * @sb: super block of file system to search
1138 * @hashval: hash value (usually inode number) to search for
1139 * @test: callback used for comparisons between inodes
1140 * @data: opaque data pointer to pass to @test
1142 * Search for the inode specified by @hashval and @data in the inode cache,
1143 * and if the inode is in the cache, return the inode with an incremented
1144 * reference count. Waits on I_NEW before returning the inode.
1145 * returned with an incremented reference count.
1147 * This is a generalized version of ilookup() for file systems where the
1148 * inode number is not sufficient for unique identification of an inode.
1150 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1152 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1153 int (*test)(struct inode *, void *), void *data)
1155 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1158 wait_on_inode(inode);
1161 EXPORT_SYMBOL(ilookup5);
1164 * ilookup - search for an inode in the inode cache
1165 * @sb: super block of file system to search
1166 * @ino: inode number to search for
1168 * Search for the inode @ino in the inode cache, and if the inode is in the
1169 * cache, the inode is returned with an incremented reference count.
1171 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1173 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1174 struct inode *inode;
1176 spin_lock(&inode_hash_lock);
1177 inode = find_inode_fast(sb, head, ino);
1178 spin_unlock(&inode_hash_lock);
1181 wait_on_inode(inode);
1184 EXPORT_SYMBOL(ilookup);
1186 int insert_inode_locked(struct inode *inode)
1188 struct super_block *sb = inode->i_sb;
1189 ino_t ino = inode->i_ino;
1190 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1193 struct hlist_node *node;
1194 struct inode *old = NULL;
1195 spin_lock(&inode_hash_lock);
1196 hlist_for_each_entry(old, node, head, i_hash) {
1197 if (old->i_ino != ino)
1199 if (old->i_sb != sb)
1201 spin_lock(&old->i_lock);
1202 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1203 spin_unlock(&old->i_lock);
1208 if (likely(!node)) {
1209 spin_lock(&inode->i_lock);
1210 inode->i_state |= I_NEW;
1211 hlist_add_head(&inode->i_hash, head);
1212 spin_unlock(&inode->i_lock);
1213 spin_unlock(&inode_hash_lock);
1217 spin_unlock(&old->i_lock);
1218 spin_unlock(&inode_hash_lock);
1220 if (unlikely(!inode_unhashed(old))) {
1227 EXPORT_SYMBOL(insert_inode_locked);
1229 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1230 int (*test)(struct inode *, void *), void *data)
1232 struct super_block *sb = inode->i_sb;
1233 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1236 struct hlist_node *node;
1237 struct inode *old = NULL;
1239 spin_lock(&inode_hash_lock);
1240 hlist_for_each_entry(old, node, head, i_hash) {
1241 if (old->i_sb != sb)
1243 if (!test(old, data))
1245 spin_lock(&old->i_lock);
1246 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1247 spin_unlock(&old->i_lock);
1252 if (likely(!node)) {
1253 spin_lock(&inode->i_lock);
1254 inode->i_state |= I_NEW;
1255 hlist_add_head(&inode->i_hash, head);
1256 spin_unlock(&inode->i_lock);
1257 spin_unlock(&inode_hash_lock);
1261 spin_unlock(&old->i_lock);
1262 spin_unlock(&inode_hash_lock);
1264 if (unlikely(!inode_unhashed(old))) {
1271 EXPORT_SYMBOL(insert_inode_locked4);
1274 int generic_delete_inode(struct inode *inode)
1278 EXPORT_SYMBOL(generic_delete_inode);
1281 * Normal UNIX filesystem behaviour: delete the
1282 * inode when the usage count drops to zero, and
1285 int generic_drop_inode(struct inode *inode)
1287 return !inode->i_nlink || inode_unhashed(inode);
1289 EXPORT_SYMBOL_GPL(generic_drop_inode);
1292 * Called when we're dropping the last reference
1295 * Call the FS "drop_inode()" function, defaulting to
1296 * the legacy UNIX filesystem behaviour. If it tells
1297 * us to evict inode, do so. Otherwise, retain inode
1298 * in cache if fs is alive, sync and evict if fs is
1301 static void iput_final(struct inode *inode)
1303 struct super_block *sb = inode->i_sb;
1304 const struct super_operations *op = inode->i_sb->s_op;
1307 WARN_ON(inode->i_state & I_NEW);
1310 drop = op->drop_inode(inode);
1312 drop = generic_drop_inode(inode);
1314 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1315 inode->i_state |= I_REFERENCED;
1316 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1317 inode_lru_list_add(inode);
1318 spin_unlock(&inode->i_lock);
1323 inode->i_state |= I_WILL_FREE;
1324 spin_unlock(&inode->i_lock);
1325 write_inode_now(inode, 1);
1326 spin_lock(&inode->i_lock);
1327 WARN_ON(inode->i_state & I_NEW);
1328 inode->i_state &= ~I_WILL_FREE;
1331 inode->i_state |= I_FREEING;
1332 inode_lru_list_del(inode);
1333 spin_unlock(&inode->i_lock);
1339 * iput - put an inode
1340 * @inode: inode to put
1342 * Puts an inode, dropping its usage count. If the inode use count hits
1343 * zero, the inode is then freed and may also be destroyed.
1345 * Consequently, iput() can sleep.
1347 void iput(struct inode *inode)
1350 BUG_ON(inode->i_state & I_CLEAR);
1352 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1356 EXPORT_SYMBOL(iput);
1359 * bmap - find a block number in a file
1360 * @inode: inode of file
1361 * @block: block to find
1363 * Returns the block number on the device holding the inode that
1364 * is the disk block number for the block of the file requested.
1365 * That is, asked for block 4 of inode 1 the function will return the
1366 * disk block relative to the disk start that holds that block of the
1369 sector_t bmap(struct inode *inode, sector_t block)
1372 if (inode->i_mapping->a_ops->bmap)
1373 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1376 EXPORT_SYMBOL(bmap);
1379 * With relative atime, only update atime if the previous atime is
1380 * earlier than either the ctime or mtime or if at least a day has
1381 * passed since the last atime update.
1383 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1384 struct timespec now)
1387 if (!(mnt->mnt_flags & MNT_RELATIME))
1390 * Is mtime younger than atime? If yes, update atime:
1392 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1395 * Is ctime younger than atime? If yes, update atime:
1397 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1401 * Is the previous atime value older than a day? If yes,
1404 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1407 * Good, we can skip the atime update:
1413 * touch_atime - update the access time
1414 * @mnt: mount the inode is accessed on
1415 * @dentry: dentry accessed
1417 * Update the accessed time on an inode and mark it for writeback.
1418 * This function automatically handles read only file systems and media,
1419 * as well as the "noatime" flag and inode specific "noatime" markers.
1421 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1423 struct inode *inode = dentry->d_inode;
1424 struct timespec now;
1426 if (inode->i_flags & S_NOATIME)
1428 if (IS_NOATIME(inode))
1430 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1433 if (mnt->mnt_flags & MNT_NOATIME)
1435 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1438 now = current_fs_time(inode->i_sb);
1440 if (!relatime_need_update(mnt, inode, now))
1443 if (timespec_equal(&inode->i_atime, &now))
1446 if (mnt_want_write(mnt))
1449 inode->i_atime = now;
1450 mark_inode_dirty_sync(inode);
1451 mnt_drop_write(mnt);
1453 EXPORT_SYMBOL(touch_atime);
1456 * file_update_time - update mtime and ctime time
1457 * @file: file accessed
1459 * Update the mtime and ctime members of an inode and mark the inode
1460 * for writeback. Note that this function is meant exclusively for
1461 * usage in the file write path of filesystems, and filesystems may
1462 * choose to explicitly ignore update via this function with the
1463 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1464 * timestamps are handled by the server.
1467 void file_update_time(struct file *file)
1469 struct inode *inode = file->f_path.dentry->d_inode;
1470 struct timespec now;
1471 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1473 /* First try to exhaust all avenues to not sync */
1474 if (IS_NOCMTIME(inode))
1477 now = current_fs_time(inode->i_sb);
1478 if (!timespec_equal(&inode->i_mtime, &now))
1481 if (!timespec_equal(&inode->i_ctime, &now))
1484 if (IS_I_VERSION(inode))
1485 sync_it |= S_VERSION;
1490 /* Finally allowed to write? Takes lock. */
1491 if (mnt_want_write_file(file))
1494 /* Only change inode inside the lock region */
1495 if (sync_it & S_VERSION)
1496 inode_inc_iversion(inode);
1497 if (sync_it & S_CTIME)
1498 inode->i_ctime = now;
1499 if (sync_it & S_MTIME)
1500 inode->i_mtime = now;
1501 mark_inode_dirty_sync(inode);
1502 mnt_drop_write(file->f_path.mnt);
1504 EXPORT_SYMBOL(file_update_time);
1506 int inode_needs_sync(struct inode *inode)
1510 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1514 EXPORT_SYMBOL(inode_needs_sync);
1516 int inode_wait(void *word)
1521 EXPORT_SYMBOL(inode_wait);
1524 * If we try to find an inode in the inode hash while it is being
1525 * deleted, we have to wait until the filesystem completes its
1526 * deletion before reporting that it isn't found. This function waits
1527 * until the deletion _might_ have completed. Callers are responsible
1528 * to recheck inode state.
1530 * It doesn't matter if I_NEW is not set initially, a call to
1531 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1534 static void __wait_on_freeing_inode(struct inode *inode)
1536 wait_queue_head_t *wq;
1537 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1538 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1539 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1540 spin_unlock(&inode->i_lock);
1541 spin_unlock(&inode_hash_lock);
1543 finish_wait(wq, &wait.wait);
1544 spin_lock(&inode_hash_lock);
1547 static __initdata unsigned long ihash_entries;
1548 static int __init set_ihash_entries(char *str)
1552 ihash_entries = simple_strtoul(str, &str, 0);
1555 __setup("ihash_entries=", set_ihash_entries);
1558 * Initialize the waitqueues and inode hash table.
1560 void __init inode_init_early(void)
1564 /* If hashes are distributed across NUMA nodes, defer
1565 * hash allocation until vmalloc space is available.
1571 alloc_large_system_hash("Inode-cache",
1572 sizeof(struct hlist_head),
1580 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1581 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1584 void __init inode_init(void)
1588 /* inode slab cache */
1589 inode_cachep = kmem_cache_create("inode_cache",
1590 sizeof(struct inode),
1592 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1596 /* Hash may have been set up in inode_init_early */
1601 alloc_large_system_hash("Inode-cache",
1602 sizeof(struct hlist_head),
1610 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1611 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1614 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1616 inode->i_mode = mode;
1617 if (S_ISCHR(mode)) {
1618 inode->i_fop = &def_chr_fops;
1619 inode->i_rdev = rdev;
1620 } else if (S_ISBLK(mode)) {
1621 inode->i_fop = &def_blk_fops;
1622 inode->i_rdev = rdev;
1623 } else if (S_ISFIFO(mode))
1624 inode->i_fop = &def_fifo_fops;
1625 else if (S_ISSOCK(mode))
1626 inode->i_fop = &bad_sock_fops;
1628 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1629 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1632 EXPORT_SYMBOL(init_special_inode);
1635 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1637 * @dir: Directory inode
1638 * @mode: mode of the new inode
1640 void inode_init_owner(struct inode *inode, const struct inode *dir,
1643 inode->i_uid = current_fsuid();
1644 if (dir && dir->i_mode & S_ISGID) {
1645 inode->i_gid = dir->i_gid;
1649 inode->i_gid = current_fsgid();
1650 inode->i_mode = mode;
1652 EXPORT_SYMBOL(inode_init_owner);
1655 * inode_owner_or_capable - check current task permissions to inode
1656 * @inode: inode being checked
1658 * Return true if current either has CAP_FOWNER to the inode, or
1661 bool inode_owner_or_capable(const struct inode *inode)
1663 struct user_namespace *ns = inode_userns(inode);
1665 if (current_user_ns() == ns && current_fsuid() == inode->i_uid)
1667 if (ns_capable(ns, CAP_FOWNER))
1671 EXPORT_SYMBOL(inode_owner_or_capable);