staging/usbip: convert to kthread
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / inode.c
blobda85e56378f3e1d98c241901171e24e83882f189
1 /*
2 * linux/fs/inode.c
4 * (C) 1997 Linus Torvalds
5 */
7 #include <linux/fs.h>
8 #include <linux/mm.h>
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:
31 * - inode_has_buffers
32 * - invalidate_bdev
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".
45 * Famous last words.
48 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
50 /* #define INODE_PARANOIA 1 */
51 /* #define INODE_DEBUG 1 */
54 * Inode lookup is no longer as critical as it used to be:
55 * most of the lookups are going to be through the dcache.
57 #define I_HASHBITS i_hash_shift
58 #define I_HASHMASK i_hash_mask
60 static unsigned int i_hash_mask __read_mostly;
61 static unsigned int i_hash_shift __read_mostly;
64 * Each inode can be on two separate lists. One is
65 * the hash list of the inode, used for lookups. The
66 * other linked list is the "type" list:
67 * "in_use" - valid inode, i_count > 0, i_nlink > 0
68 * "dirty" - as "in_use" but also dirty
69 * "unused" - valid inode, i_count = 0
71 * A "dirty" list is maintained for each super block,
72 * allowing for low-overhead inode sync() operations.
75 static LIST_HEAD(inode_lru);
76 static struct hlist_head *inode_hashtable __read_mostly;
79 * A simple spinlock to protect the list manipulations.
81 * NOTE! You also have to own the lock if you change
82 * the i_state of an inode while it is in use..
84 DEFINE_SPINLOCK(inode_lock);
87 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
88 * icache shrinking path, and the umount path. Without this exclusion,
89 * by the time prune_icache calls iput for the inode whose pages it has
90 * been invalidating, or by the time it calls clear_inode & destroy_inode
91 * from its final dispose_list, the struct super_block they refer to
92 * (for inode->i_sb->s_op) may already have been freed and reused.
94 * We make this an rwsem because the fastpath is icache shrinking. In
95 * some cases a filesystem may be doing a significant amount of work in
96 * its inode reclaim code, so this should improve parallelism.
98 static DECLARE_RWSEM(iprune_sem);
101 * Statistics gathering..
103 struct inodes_stat_t inodes_stat;
105 static DEFINE_PER_CPU(unsigned int, nr_inodes);
107 static struct kmem_cache *inode_cachep __read_mostly;
109 static int get_nr_inodes(void)
111 int i;
112 int sum = 0;
113 for_each_possible_cpu(i)
114 sum += per_cpu(nr_inodes, i);
115 return sum < 0 ? 0 : sum;
118 static inline int get_nr_inodes_unused(void)
120 return inodes_stat.nr_unused;
123 int get_nr_dirty_inodes(void)
125 /* not actually dirty inodes, but a wild approximation */
126 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
127 return nr_dirty > 0 ? nr_dirty : 0;
131 * Handle nr_inode sysctl
133 #ifdef CONFIG_SYSCTL
134 int proc_nr_inodes(ctl_table *table, int write,
135 void __user *buffer, size_t *lenp, loff_t *ppos)
137 inodes_stat.nr_inodes = get_nr_inodes();
138 return proc_dointvec(table, write, buffer, lenp, ppos);
140 #endif
142 static void wake_up_inode(struct inode *inode)
145 * Prevent speculative execution through spin_unlock(&inode_lock);
147 smp_mb();
148 wake_up_bit(&inode->i_state, __I_NEW);
152 * inode_init_always - perform inode structure intialisation
153 * @sb: superblock inode belongs to
154 * @inode: inode to initialise
156 * These are initializations that need to be done on every inode
157 * allocation as the fields are not initialised by slab allocation.
159 int inode_init_always(struct super_block *sb, struct inode *inode)
161 static const struct address_space_operations empty_aops;
162 static const struct inode_operations empty_iops;
163 static const struct file_operations empty_fops;
164 struct address_space *const mapping = &inode->i_data;
166 inode->i_sb = sb;
167 inode->i_blkbits = sb->s_blocksize_bits;
168 inode->i_flags = 0;
169 atomic_set(&inode->i_count, 1);
170 inode->i_op = &empty_iops;
171 inode->i_fop = &empty_fops;
172 inode->i_nlink = 1;
173 inode->i_uid = 0;
174 inode->i_gid = 0;
175 atomic_set(&inode->i_writecount, 0);
176 inode->i_size = 0;
177 inode->i_blocks = 0;
178 inode->i_bytes = 0;
179 inode->i_generation = 0;
180 #ifdef CONFIG_QUOTA
181 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
182 #endif
183 inode->i_pipe = NULL;
184 inode->i_bdev = NULL;
185 inode->i_cdev = NULL;
186 inode->i_rdev = 0;
187 inode->dirtied_when = 0;
189 if (security_inode_alloc(inode))
190 goto out;
191 spin_lock_init(&inode->i_lock);
192 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
194 mutex_init(&inode->i_mutex);
195 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
197 init_rwsem(&inode->i_alloc_sem);
198 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
200 mapping->a_ops = &empty_aops;
201 mapping->host = inode;
202 mapping->flags = 0;
203 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
204 mapping->assoc_mapping = NULL;
205 mapping->backing_dev_info = &default_backing_dev_info;
206 mapping->writeback_index = 0;
209 * If the block_device provides a backing_dev_info for client
210 * inodes then use that. Otherwise the inode share the bdev's
211 * backing_dev_info.
213 if (sb->s_bdev) {
214 struct backing_dev_info *bdi;
216 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
217 mapping->backing_dev_info = bdi;
219 inode->i_private = NULL;
220 inode->i_mapping = mapping;
221 #ifdef CONFIG_FS_POSIX_ACL
222 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
223 #endif
225 #ifdef CONFIG_FSNOTIFY
226 inode->i_fsnotify_mask = 0;
227 #endif
229 this_cpu_inc(nr_inodes);
231 return 0;
232 out:
233 return -ENOMEM;
235 EXPORT_SYMBOL(inode_init_always);
237 static struct inode *alloc_inode(struct super_block *sb)
239 struct inode *inode;
241 if (sb->s_op->alloc_inode)
242 inode = sb->s_op->alloc_inode(sb);
243 else
244 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
246 if (!inode)
247 return NULL;
249 if (unlikely(inode_init_always(sb, inode))) {
250 if (inode->i_sb->s_op->destroy_inode)
251 inode->i_sb->s_op->destroy_inode(inode);
252 else
253 kmem_cache_free(inode_cachep, inode);
254 return NULL;
257 return inode;
260 void free_inode_nonrcu(struct inode *inode)
262 kmem_cache_free(inode_cachep, inode);
264 EXPORT_SYMBOL(free_inode_nonrcu);
266 void __destroy_inode(struct inode *inode)
268 BUG_ON(inode_has_buffers(inode));
269 security_inode_free(inode);
270 fsnotify_inode_delete(inode);
271 #ifdef CONFIG_FS_POSIX_ACL
272 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
273 posix_acl_release(inode->i_acl);
274 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
275 posix_acl_release(inode->i_default_acl);
276 #endif
277 this_cpu_dec(nr_inodes);
279 EXPORT_SYMBOL(__destroy_inode);
281 static void i_callback(struct rcu_head *head)
283 struct inode *inode = container_of(head, struct inode, i_rcu);
284 INIT_LIST_HEAD(&inode->i_dentry);
285 kmem_cache_free(inode_cachep, inode);
288 static void destroy_inode(struct inode *inode)
290 BUG_ON(!list_empty(&inode->i_lru));
291 __destroy_inode(inode);
292 if (inode->i_sb->s_op->destroy_inode)
293 inode->i_sb->s_op->destroy_inode(inode);
294 else
295 call_rcu(&inode->i_rcu, i_callback);
299 * These are initializations that only need to be done
300 * once, because the fields are idempotent across use
301 * of the inode, so let the slab aware of that.
303 void inode_init_once(struct inode *inode)
305 memset(inode, 0, sizeof(*inode));
306 INIT_HLIST_NODE(&inode->i_hash);
307 INIT_LIST_HEAD(&inode->i_dentry);
308 INIT_LIST_HEAD(&inode->i_devices);
309 INIT_LIST_HEAD(&inode->i_wb_list);
310 INIT_LIST_HEAD(&inode->i_lru);
311 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
312 spin_lock_init(&inode->i_data.tree_lock);
313 spin_lock_init(&inode->i_data.i_mmap_lock);
314 INIT_LIST_HEAD(&inode->i_data.private_list);
315 spin_lock_init(&inode->i_data.private_lock);
316 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
317 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
318 i_size_ordered_init(inode);
319 #ifdef CONFIG_FSNOTIFY
320 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
321 #endif
323 EXPORT_SYMBOL(inode_init_once);
325 static void init_once(void *foo)
327 struct inode *inode = (struct inode *) foo;
329 inode_init_once(inode);
333 * inode_lock must be held
335 void __iget(struct inode *inode)
337 atomic_inc(&inode->i_count);
341 * get additional reference to inode; caller must already hold one.
343 void ihold(struct inode *inode)
345 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
347 EXPORT_SYMBOL(ihold);
349 static void inode_lru_list_add(struct inode *inode)
351 if (list_empty(&inode->i_lru)) {
352 list_add(&inode->i_lru, &inode_lru);
353 inodes_stat.nr_unused++;
357 static void inode_lru_list_del(struct inode *inode)
359 if (!list_empty(&inode->i_lru)) {
360 list_del_init(&inode->i_lru);
361 inodes_stat.nr_unused--;
365 static inline void __inode_sb_list_add(struct inode *inode)
367 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
371 * inode_sb_list_add - add inode to the superblock list of inodes
372 * @inode: inode to add
374 void inode_sb_list_add(struct inode *inode)
376 spin_lock(&inode_lock);
377 __inode_sb_list_add(inode);
378 spin_unlock(&inode_lock);
380 EXPORT_SYMBOL_GPL(inode_sb_list_add);
382 static inline void __inode_sb_list_del(struct inode *inode)
384 list_del_init(&inode->i_sb_list);
387 static unsigned long hash(struct super_block *sb, unsigned long hashval)
389 unsigned long tmp;
391 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
392 L1_CACHE_BYTES;
393 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
394 return tmp & I_HASHMASK;
398 * __insert_inode_hash - hash an inode
399 * @inode: unhashed inode
400 * @hashval: unsigned long value used to locate this object in the
401 * inode_hashtable.
403 * Add an inode to the inode hash for this superblock.
405 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
407 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
409 spin_lock(&inode_lock);
410 hlist_add_head(&inode->i_hash, b);
411 spin_unlock(&inode_lock);
413 EXPORT_SYMBOL(__insert_inode_hash);
416 * __remove_inode_hash - remove an inode from the hash
417 * @inode: inode to unhash
419 * Remove an inode from the superblock.
421 static void __remove_inode_hash(struct inode *inode)
423 hlist_del_init(&inode->i_hash);
427 * remove_inode_hash - remove an inode from the hash
428 * @inode: inode to unhash
430 * Remove an inode from the superblock.
432 void remove_inode_hash(struct inode *inode)
434 spin_lock(&inode_lock);
435 hlist_del_init(&inode->i_hash);
436 spin_unlock(&inode_lock);
438 EXPORT_SYMBOL(remove_inode_hash);
440 void end_writeback(struct inode *inode)
442 might_sleep();
443 BUG_ON(inode->i_data.nrpages);
444 BUG_ON(!list_empty(&inode->i_data.private_list));
445 BUG_ON(!(inode->i_state & I_FREEING));
446 BUG_ON(inode->i_state & I_CLEAR);
447 inode_sync_wait(inode);
448 /* don't need i_lock here, no concurrent mods to i_state */
449 inode->i_state = I_FREEING | I_CLEAR;
451 EXPORT_SYMBOL(end_writeback);
453 static void evict(struct inode *inode)
455 const struct super_operations *op = inode->i_sb->s_op;
457 if (op->evict_inode) {
458 op->evict_inode(inode);
459 } else {
460 if (inode->i_data.nrpages)
461 truncate_inode_pages(&inode->i_data, 0);
462 end_writeback(inode);
464 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
465 bd_forget(inode);
466 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
467 cd_forget(inode);
471 * dispose_list - dispose of the contents of a local list
472 * @head: the head of the list to free
474 * Dispose-list gets a local list with local inodes in it, so it doesn't
475 * need to worry about list corruption and SMP locks.
477 static void dispose_list(struct list_head *head)
479 while (!list_empty(head)) {
480 struct inode *inode;
482 inode = list_first_entry(head, struct inode, i_lru);
483 list_del_init(&inode->i_lru);
485 evict(inode);
487 spin_lock(&inode_lock);
488 __remove_inode_hash(inode);
489 __inode_sb_list_del(inode);
490 spin_unlock(&inode_lock);
492 wake_up_inode(inode);
493 destroy_inode(inode);
498 * evict_inodes - evict all evictable inodes for a superblock
499 * @sb: superblock to operate on
501 * Make sure that no inodes with zero refcount are retained. This is
502 * called by superblock shutdown after having MS_ACTIVE flag removed,
503 * so any inode reaching zero refcount during or after that call will
504 * be immediately evicted.
506 void evict_inodes(struct super_block *sb)
508 struct inode *inode, *next;
509 LIST_HEAD(dispose);
511 down_write(&iprune_sem);
513 spin_lock(&inode_lock);
514 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
515 if (atomic_read(&inode->i_count))
516 continue;
518 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
519 WARN_ON(1);
520 continue;
523 inode->i_state |= I_FREEING;
526 * Move the inode off the IO lists and LRU once I_FREEING is
527 * set so that it won't get moved back on there if it is dirty.
529 list_move(&inode->i_lru, &dispose);
530 list_del_init(&inode->i_wb_list);
531 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
532 inodes_stat.nr_unused--;
534 spin_unlock(&inode_lock);
536 dispose_list(&dispose);
537 up_write(&iprune_sem);
541 * invalidate_inodes - attempt to free all inodes on a superblock
542 * @sb: superblock to operate on
544 * Attempts to free all inodes for a given superblock. If there were any
545 * busy inodes return a non-zero value, else zero.
547 int invalidate_inodes(struct super_block *sb)
549 int busy = 0;
550 struct inode *inode, *next;
551 LIST_HEAD(dispose);
553 down_write(&iprune_sem);
555 spin_lock(&inode_lock);
556 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
557 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE))
558 continue;
559 if (atomic_read(&inode->i_count)) {
560 busy = 1;
561 continue;
564 inode->i_state |= I_FREEING;
567 * Move the inode off the IO lists and LRU once I_FREEING is
568 * set so that it won't get moved back on there if it is dirty.
570 list_move(&inode->i_lru, &dispose);
571 list_del_init(&inode->i_wb_list);
572 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
573 inodes_stat.nr_unused--;
575 spin_unlock(&inode_lock);
577 dispose_list(&dispose);
578 up_write(&iprune_sem);
580 return busy;
583 static int can_unuse(struct inode *inode)
585 if (inode->i_state & ~I_REFERENCED)
586 return 0;
587 if (inode_has_buffers(inode))
588 return 0;
589 if (atomic_read(&inode->i_count))
590 return 0;
591 if (inode->i_data.nrpages)
592 return 0;
593 return 1;
597 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
598 * temporary list and then are freed outside inode_lock by dispose_list().
600 * Any inodes which are pinned purely because of attached pagecache have their
601 * pagecache removed. If the inode has metadata buffers attached to
602 * mapping->private_list then try to remove them.
604 * If the inode has the I_REFERENCED flag set, then it means that it has been
605 * used recently - the flag is set in iput_final(). When we encounter such an
606 * inode, clear the flag and move it to the back of the LRU so it gets another
607 * pass through the LRU before it gets reclaimed. This is necessary because of
608 * the fact we are doing lazy LRU updates to minimise lock contention so the
609 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
610 * with this flag set because they are the inodes that are out of order.
612 static void prune_icache(int nr_to_scan)
614 LIST_HEAD(freeable);
615 int nr_scanned;
616 unsigned long reap = 0;
618 down_read(&iprune_sem);
619 spin_lock(&inode_lock);
620 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
621 struct inode *inode;
623 if (list_empty(&inode_lru))
624 break;
626 inode = list_entry(inode_lru.prev, struct inode, i_lru);
629 * Referenced or dirty inodes are still in use. Give them
630 * another pass through the LRU as we canot reclaim them now.
632 if (atomic_read(&inode->i_count) ||
633 (inode->i_state & ~I_REFERENCED)) {
634 list_del_init(&inode->i_lru);
635 inodes_stat.nr_unused--;
636 continue;
639 /* recently referenced inodes get one more pass */
640 if (inode->i_state & I_REFERENCED) {
641 list_move(&inode->i_lru, &inode_lru);
642 inode->i_state &= ~I_REFERENCED;
643 continue;
645 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
646 __iget(inode);
647 spin_unlock(&inode_lock);
648 if (remove_inode_buffers(inode))
649 reap += invalidate_mapping_pages(&inode->i_data,
650 0, -1);
651 iput(inode);
652 spin_lock(&inode_lock);
654 if (inode != list_entry(inode_lru.next,
655 struct inode, i_lru))
656 continue; /* wrong inode or list_empty */
657 if (!can_unuse(inode))
658 continue;
660 WARN_ON(inode->i_state & I_NEW);
661 inode->i_state |= I_FREEING;
664 * Move the inode off the IO lists and LRU once I_FREEING is
665 * set so that it won't get moved back on there if it is dirty.
667 list_move(&inode->i_lru, &freeable);
668 list_del_init(&inode->i_wb_list);
669 inodes_stat.nr_unused--;
671 if (current_is_kswapd())
672 __count_vm_events(KSWAPD_INODESTEAL, reap);
673 else
674 __count_vm_events(PGINODESTEAL, reap);
675 spin_unlock(&inode_lock);
677 dispose_list(&freeable);
678 up_read(&iprune_sem);
682 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
683 * "unused" means that no dentries are referring to the inodes: the files are
684 * not open and the dcache references to those inodes have already been
685 * reclaimed.
687 * This function is passed the number of inodes to scan, and it returns the
688 * total number of remaining possibly-reclaimable inodes.
690 static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
692 if (nr) {
694 * Nasty deadlock avoidance. We may hold various FS locks,
695 * and we don't want to recurse into the FS that called us
696 * in clear_inode() and friends..
698 if (!(gfp_mask & __GFP_FS))
699 return -1;
700 prune_icache(nr);
702 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
705 static struct shrinker icache_shrinker = {
706 .shrink = shrink_icache_memory,
707 .seeks = DEFAULT_SEEKS,
710 static void __wait_on_freeing_inode(struct inode *inode);
712 * Called with the inode lock held.
714 static struct inode *find_inode(struct super_block *sb,
715 struct hlist_head *head,
716 int (*test)(struct inode *, void *),
717 void *data)
719 struct hlist_node *node;
720 struct inode *inode = NULL;
722 repeat:
723 hlist_for_each_entry(inode, node, head, i_hash) {
724 if (inode->i_sb != sb)
725 continue;
726 if (!test(inode, data))
727 continue;
728 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
729 __wait_on_freeing_inode(inode);
730 goto repeat;
732 __iget(inode);
733 return inode;
735 return NULL;
739 * find_inode_fast is the fast path version of find_inode, see the comment at
740 * iget_locked for details.
742 static struct inode *find_inode_fast(struct super_block *sb,
743 struct hlist_head *head, unsigned long ino)
745 struct hlist_node *node;
746 struct inode *inode = NULL;
748 repeat:
749 hlist_for_each_entry(inode, node, head, i_hash) {
750 if (inode->i_ino != ino)
751 continue;
752 if (inode->i_sb != sb)
753 continue;
754 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
755 __wait_on_freeing_inode(inode);
756 goto repeat;
758 __iget(inode);
759 return inode;
761 return NULL;
765 * Each cpu owns a range of LAST_INO_BATCH numbers.
766 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
767 * to renew the exhausted range.
769 * This does not significantly increase overflow rate because every CPU can
770 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
771 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
772 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
773 * overflow rate by 2x, which does not seem too significant.
775 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
776 * error if st_ino won't fit in target struct field. Use 32bit counter
777 * here to attempt to avoid that.
779 #define LAST_INO_BATCH 1024
780 static DEFINE_PER_CPU(unsigned int, last_ino);
782 unsigned int get_next_ino(void)
784 unsigned int *p = &get_cpu_var(last_ino);
785 unsigned int res = *p;
787 #ifdef CONFIG_SMP
788 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
789 static atomic_t shared_last_ino;
790 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
792 res = next - LAST_INO_BATCH;
794 #endif
796 *p = ++res;
797 put_cpu_var(last_ino);
798 return res;
800 EXPORT_SYMBOL(get_next_ino);
803 * new_inode - obtain an inode
804 * @sb: superblock
806 * Allocates a new inode for given superblock. The default gfp_mask
807 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
808 * If HIGHMEM pages are unsuitable or it is known that pages allocated
809 * for the page cache are not reclaimable or migratable,
810 * mapping_set_gfp_mask() must be called with suitable flags on the
811 * newly created inode's mapping
814 struct inode *new_inode(struct super_block *sb)
816 struct inode *inode;
818 spin_lock_prefetch(&inode_lock);
820 inode = alloc_inode(sb);
821 if (inode) {
822 spin_lock(&inode_lock);
823 __inode_sb_list_add(inode);
824 inode->i_state = 0;
825 spin_unlock(&inode_lock);
827 return inode;
829 EXPORT_SYMBOL(new_inode);
831 void unlock_new_inode(struct inode *inode)
833 #ifdef CONFIG_DEBUG_LOCK_ALLOC
834 if (S_ISDIR(inode->i_mode)) {
835 struct file_system_type *type = inode->i_sb->s_type;
837 /* Set new key only if filesystem hasn't already changed it */
838 if (!lockdep_match_class(&inode->i_mutex,
839 &type->i_mutex_key)) {
841 * ensure nobody is actually holding i_mutex
843 mutex_destroy(&inode->i_mutex);
844 mutex_init(&inode->i_mutex);
845 lockdep_set_class(&inode->i_mutex,
846 &type->i_mutex_dir_key);
849 #endif
851 * This is special! We do not need the spinlock when clearing I_NEW,
852 * because we're guaranteed that nobody else tries to do anything about
853 * the state of the inode when it is locked, as we just created it (so
854 * there can be no old holders that haven't tested I_NEW).
855 * However we must emit the memory barrier so that other CPUs reliably
856 * see the clearing of I_NEW after the other inode initialisation has
857 * completed.
859 smp_mb();
860 WARN_ON(!(inode->i_state & I_NEW));
861 inode->i_state &= ~I_NEW;
862 wake_up_inode(inode);
864 EXPORT_SYMBOL(unlock_new_inode);
867 * This is called without the inode lock held.. Be careful.
869 * We no longer cache the sb_flags in i_flags - see fs.h
870 * -- rmk@arm.uk.linux.org
872 static struct inode *get_new_inode(struct super_block *sb,
873 struct hlist_head *head,
874 int (*test)(struct inode *, void *),
875 int (*set)(struct inode *, void *),
876 void *data)
878 struct inode *inode;
880 inode = alloc_inode(sb);
881 if (inode) {
882 struct inode *old;
884 spin_lock(&inode_lock);
885 /* We released the lock, so.. */
886 old = find_inode(sb, head, test, data);
887 if (!old) {
888 if (set(inode, data))
889 goto set_failed;
891 hlist_add_head(&inode->i_hash, head);
892 __inode_sb_list_add(inode);
893 inode->i_state = I_NEW;
894 spin_unlock(&inode_lock);
896 /* Return the locked inode with I_NEW set, the
897 * caller is responsible for filling in the contents
899 return inode;
903 * Uhhuh, somebody else created the same inode under
904 * us. Use the old inode instead of the one we just
905 * allocated.
907 spin_unlock(&inode_lock);
908 destroy_inode(inode);
909 inode = old;
910 wait_on_inode(inode);
912 return inode;
914 set_failed:
915 spin_unlock(&inode_lock);
916 destroy_inode(inode);
917 return NULL;
921 * get_new_inode_fast is the fast path version of get_new_inode, see the
922 * comment at iget_locked for details.
924 static struct inode *get_new_inode_fast(struct super_block *sb,
925 struct hlist_head *head, unsigned long ino)
927 struct inode *inode;
929 inode = alloc_inode(sb);
930 if (inode) {
931 struct inode *old;
933 spin_lock(&inode_lock);
934 /* We released the lock, so.. */
935 old = find_inode_fast(sb, head, ino);
936 if (!old) {
937 inode->i_ino = ino;
938 hlist_add_head(&inode->i_hash, head);
939 __inode_sb_list_add(inode);
940 inode->i_state = I_NEW;
941 spin_unlock(&inode_lock);
943 /* Return the locked inode with I_NEW set, the
944 * caller is responsible for filling in the contents
946 return inode;
950 * Uhhuh, somebody else created the same inode under
951 * us. Use the old inode instead of the one we just
952 * allocated.
954 spin_unlock(&inode_lock);
955 destroy_inode(inode);
956 inode = old;
957 wait_on_inode(inode);
959 return inode;
963 * search the inode cache for a matching inode number.
964 * If we find one, then the inode number we are trying to
965 * allocate is not unique and so we should not use it.
967 * Returns 1 if the inode number is unique, 0 if it is not.
969 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
971 struct hlist_head *b = inode_hashtable + hash(sb, ino);
972 struct hlist_node *node;
973 struct inode *inode;
975 hlist_for_each_entry(inode, node, b, i_hash) {
976 if (inode->i_ino == ino && inode->i_sb == sb)
977 return 0;
980 return 1;
984 * iunique - get a unique inode number
985 * @sb: superblock
986 * @max_reserved: highest reserved inode number
988 * Obtain an inode number that is unique on the system for a given
989 * superblock. This is used by file systems that have no natural
990 * permanent inode numbering system. An inode number is returned that
991 * is higher than the reserved limit but unique.
993 * BUGS:
994 * With a large number of inodes live on the file system this function
995 * currently becomes quite slow.
997 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1000 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1001 * error if st_ino won't fit in target struct field. Use 32bit counter
1002 * here to attempt to avoid that.
1004 static DEFINE_SPINLOCK(iunique_lock);
1005 static unsigned int counter;
1006 ino_t res;
1008 spin_lock(&inode_lock);
1009 spin_lock(&iunique_lock);
1010 do {
1011 if (counter <= max_reserved)
1012 counter = max_reserved + 1;
1013 res = counter++;
1014 } while (!test_inode_iunique(sb, res));
1015 spin_unlock(&iunique_lock);
1016 spin_unlock(&inode_lock);
1018 return res;
1020 EXPORT_SYMBOL(iunique);
1022 struct inode *igrab(struct inode *inode)
1024 spin_lock(&inode_lock);
1025 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
1026 __iget(inode);
1027 else
1029 * Handle the case where s_op->clear_inode is not been
1030 * called yet, and somebody is calling igrab
1031 * while the inode is getting freed.
1033 inode = NULL;
1034 spin_unlock(&inode_lock);
1035 return inode;
1037 EXPORT_SYMBOL(igrab);
1040 * ifind - internal function, you want ilookup5() or iget5().
1041 * @sb: super block of file system to search
1042 * @head: the head of the list to search
1043 * @test: callback used for comparisons between inodes
1044 * @data: opaque data pointer to pass to @test
1045 * @wait: if true wait for the inode to be unlocked, if false do not
1047 * ifind() searches for the inode specified by @data in the inode
1048 * cache. This is a generalized version of ifind_fast() for file systems where
1049 * the inode number is not sufficient for unique identification of an inode.
1051 * If the inode is in the cache, the inode is returned with an incremented
1052 * reference count.
1054 * Otherwise NULL is returned.
1056 * Note, @test is called with the inode_lock held, so can't sleep.
1058 static struct inode *ifind(struct super_block *sb,
1059 struct hlist_head *head, int (*test)(struct inode *, void *),
1060 void *data, const int wait)
1062 struct inode *inode;
1064 spin_lock(&inode_lock);
1065 inode = find_inode(sb, head, test, data);
1066 if (inode) {
1067 spin_unlock(&inode_lock);
1068 if (likely(wait))
1069 wait_on_inode(inode);
1070 return inode;
1072 spin_unlock(&inode_lock);
1073 return NULL;
1077 * ifind_fast - internal function, you want ilookup() or iget().
1078 * @sb: super block of file system to search
1079 * @head: head of the list to search
1080 * @ino: inode number to search for
1082 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1083 * file systems where the inode number is sufficient for unique identification
1084 * of an inode.
1086 * If the inode is in the cache, the inode is returned with an incremented
1087 * reference count.
1089 * Otherwise NULL is returned.
1091 static struct inode *ifind_fast(struct super_block *sb,
1092 struct hlist_head *head, unsigned long ino)
1094 struct inode *inode;
1096 spin_lock(&inode_lock);
1097 inode = find_inode_fast(sb, head, ino);
1098 if (inode) {
1099 spin_unlock(&inode_lock);
1100 wait_on_inode(inode);
1101 return inode;
1103 spin_unlock(&inode_lock);
1104 return NULL;
1108 * ilookup5_nowait - search for an inode in the inode cache
1109 * @sb: super block of file system to search
1110 * @hashval: hash value (usually inode number) to search for
1111 * @test: callback used for comparisons between inodes
1112 * @data: opaque data pointer to pass to @test
1114 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1115 * @data in the inode cache. This is a generalized version of ilookup() for
1116 * file systems where the inode number is not sufficient for unique
1117 * identification of an inode.
1119 * If the inode is in the cache, the inode is returned with an incremented
1120 * reference count. Note, the inode lock is not waited upon so you have to be
1121 * very careful what you do with the returned inode. You probably should be
1122 * using ilookup5() instead.
1124 * Otherwise NULL is returned.
1126 * Note, @test is called with the inode_lock held, so can't sleep.
1128 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1129 int (*test)(struct inode *, void *), void *data)
1131 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1133 return ifind(sb, head, test, data, 0);
1135 EXPORT_SYMBOL(ilookup5_nowait);
1138 * ilookup5 - search for an inode in the inode cache
1139 * @sb: super block of file system to search
1140 * @hashval: hash value (usually inode number) to search for
1141 * @test: callback used for comparisons between inodes
1142 * @data: opaque data pointer to pass to @test
1144 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1145 * @data in the inode cache. This is a generalized version of ilookup() for
1146 * file systems where the inode number is not sufficient for unique
1147 * identification of an inode.
1149 * If the inode is in the cache, the inode lock is waited upon and the inode is
1150 * returned with an incremented reference count.
1152 * Otherwise NULL is returned.
1154 * Note, @test is called with the inode_lock held, so can't sleep.
1156 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1157 int (*test)(struct inode *, void *), void *data)
1159 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1161 return ifind(sb, head, test, data, 1);
1163 EXPORT_SYMBOL(ilookup5);
1166 * ilookup - search for an inode in the inode cache
1167 * @sb: super block of file system to search
1168 * @ino: inode number to search for
1170 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1171 * This is for file systems where the inode number is sufficient for unique
1172 * identification of an inode.
1174 * If the inode is in the cache, the inode is returned with an incremented
1175 * reference count.
1177 * Otherwise NULL is returned.
1179 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1181 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1183 return ifind_fast(sb, head, ino);
1185 EXPORT_SYMBOL(ilookup);
1188 * iget5_locked - obtain an inode from a mounted file system
1189 * @sb: super block of file system
1190 * @hashval: hash value (usually inode number) to get
1191 * @test: callback used for comparisons between inodes
1192 * @set: callback used to initialize a new struct inode
1193 * @data: opaque data pointer to pass to @test and @set
1195 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1196 * and @data in the inode cache and if present it is returned with an increased
1197 * reference count. This is a generalized version of iget_locked() for file
1198 * systems where the inode number is not sufficient for unique identification
1199 * of an inode.
1201 * If the inode is not in cache, get_new_inode() is called to allocate a new
1202 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1203 * file system gets to fill it in before unlocking it via unlock_new_inode().
1205 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1207 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1208 int (*test)(struct inode *, void *),
1209 int (*set)(struct inode *, void *), void *data)
1211 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1212 struct inode *inode;
1214 inode = ifind(sb, head, test, data, 1);
1215 if (inode)
1216 return inode;
1218 * get_new_inode() will do the right thing, re-trying the search
1219 * in case it had to block at any point.
1221 return get_new_inode(sb, head, test, set, data);
1223 EXPORT_SYMBOL(iget5_locked);
1226 * iget_locked - obtain an inode from a mounted file system
1227 * @sb: super block of file system
1228 * @ino: inode number to get
1230 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1231 * the inode cache and if present it is returned with an increased reference
1232 * count. This is for file systems where the inode number is sufficient for
1233 * unique identification of an inode.
1235 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1236 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1237 * The file system gets to fill it in before unlocking it via
1238 * unlock_new_inode().
1240 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1242 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1243 struct inode *inode;
1245 inode = ifind_fast(sb, head, ino);
1246 if (inode)
1247 return inode;
1249 * get_new_inode_fast() will do the right thing, re-trying the search
1250 * in case it had to block at any point.
1252 return get_new_inode_fast(sb, head, ino);
1254 EXPORT_SYMBOL(iget_locked);
1256 int insert_inode_locked(struct inode *inode)
1258 struct super_block *sb = inode->i_sb;
1259 ino_t ino = inode->i_ino;
1260 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1262 inode->i_state |= I_NEW;
1263 while (1) {
1264 struct hlist_node *node;
1265 struct inode *old = NULL;
1266 spin_lock(&inode_lock);
1267 hlist_for_each_entry(old, node, head, i_hash) {
1268 if (old->i_ino != ino)
1269 continue;
1270 if (old->i_sb != sb)
1271 continue;
1272 if (old->i_state & (I_FREEING|I_WILL_FREE))
1273 continue;
1274 break;
1276 if (likely(!node)) {
1277 hlist_add_head(&inode->i_hash, head);
1278 spin_unlock(&inode_lock);
1279 return 0;
1281 __iget(old);
1282 spin_unlock(&inode_lock);
1283 wait_on_inode(old);
1284 if (unlikely(!inode_unhashed(old))) {
1285 iput(old);
1286 return -EBUSY;
1288 iput(old);
1291 EXPORT_SYMBOL(insert_inode_locked);
1293 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1294 int (*test)(struct inode *, void *), void *data)
1296 struct super_block *sb = inode->i_sb;
1297 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1299 inode->i_state |= I_NEW;
1301 while (1) {
1302 struct hlist_node *node;
1303 struct inode *old = NULL;
1305 spin_lock(&inode_lock);
1306 hlist_for_each_entry(old, node, head, i_hash) {
1307 if (old->i_sb != sb)
1308 continue;
1309 if (!test(old, data))
1310 continue;
1311 if (old->i_state & (I_FREEING|I_WILL_FREE))
1312 continue;
1313 break;
1315 if (likely(!node)) {
1316 hlist_add_head(&inode->i_hash, head);
1317 spin_unlock(&inode_lock);
1318 return 0;
1320 __iget(old);
1321 spin_unlock(&inode_lock);
1322 wait_on_inode(old);
1323 if (unlikely(!inode_unhashed(old))) {
1324 iput(old);
1325 return -EBUSY;
1327 iput(old);
1330 EXPORT_SYMBOL(insert_inode_locked4);
1333 int generic_delete_inode(struct inode *inode)
1335 return 1;
1337 EXPORT_SYMBOL(generic_delete_inode);
1340 * Normal UNIX filesystem behaviour: delete the
1341 * inode when the usage count drops to zero, and
1342 * i_nlink is zero.
1344 int generic_drop_inode(struct inode *inode)
1346 return !inode->i_nlink || inode_unhashed(inode);
1348 EXPORT_SYMBOL_GPL(generic_drop_inode);
1351 * Called when we're dropping the last reference
1352 * to an inode.
1354 * Call the FS "drop_inode()" function, defaulting to
1355 * the legacy UNIX filesystem behaviour. If it tells
1356 * us to evict inode, do so. Otherwise, retain inode
1357 * in cache if fs is alive, sync and evict if fs is
1358 * shutting down.
1360 static void iput_final(struct inode *inode)
1362 struct super_block *sb = inode->i_sb;
1363 const struct super_operations *op = inode->i_sb->s_op;
1364 int drop;
1366 if (op && op->drop_inode)
1367 drop = op->drop_inode(inode);
1368 else
1369 drop = generic_drop_inode(inode);
1371 if (!drop) {
1372 if (sb->s_flags & MS_ACTIVE) {
1373 inode->i_state |= I_REFERENCED;
1374 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1375 inode_lru_list_add(inode);
1377 spin_unlock(&inode_lock);
1378 return;
1380 WARN_ON(inode->i_state & I_NEW);
1381 inode->i_state |= I_WILL_FREE;
1382 spin_unlock(&inode_lock);
1383 write_inode_now(inode, 1);
1384 spin_lock(&inode_lock);
1385 WARN_ON(inode->i_state & I_NEW);
1386 inode->i_state &= ~I_WILL_FREE;
1387 __remove_inode_hash(inode);
1390 WARN_ON(inode->i_state & I_NEW);
1391 inode->i_state |= I_FREEING;
1394 * Move the inode off the IO lists and LRU once I_FREEING is
1395 * set so that it won't get moved back on there if it is dirty.
1397 inode_lru_list_del(inode);
1398 list_del_init(&inode->i_wb_list);
1400 __inode_sb_list_del(inode);
1401 spin_unlock(&inode_lock);
1402 evict(inode);
1403 remove_inode_hash(inode);
1404 wake_up_inode(inode);
1405 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1406 destroy_inode(inode);
1410 * iput - put an inode
1411 * @inode: inode to put
1413 * Puts an inode, dropping its usage count. If the inode use count hits
1414 * zero, the inode is then freed and may also be destroyed.
1416 * Consequently, iput() can sleep.
1418 void iput(struct inode *inode)
1420 if (inode) {
1421 BUG_ON(inode->i_state & I_CLEAR);
1423 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1424 iput_final(inode);
1427 EXPORT_SYMBOL(iput);
1430 * bmap - find a block number in a file
1431 * @inode: inode of file
1432 * @block: block to find
1434 * Returns the block number on the device holding the inode that
1435 * is the disk block number for the block of the file requested.
1436 * That is, asked for block 4 of inode 1 the function will return the
1437 * disk block relative to the disk start that holds that block of the
1438 * file.
1440 sector_t bmap(struct inode *inode, sector_t block)
1442 sector_t res = 0;
1443 if (inode->i_mapping->a_ops->bmap)
1444 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1445 return res;
1447 EXPORT_SYMBOL(bmap);
1450 * With relative atime, only update atime if the previous atime is
1451 * earlier than either the ctime or mtime or if at least a day has
1452 * passed since the last atime update.
1454 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1455 struct timespec now)
1458 if (!(mnt->mnt_flags & MNT_RELATIME))
1459 return 1;
1461 * Is mtime younger than atime? If yes, update atime:
1463 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1464 return 1;
1466 * Is ctime younger than atime? If yes, update atime:
1468 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1469 return 1;
1472 * Is the previous atime value older than a day? If yes,
1473 * update atime:
1475 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1476 return 1;
1478 * Good, we can skip the atime update:
1480 return 0;
1484 * touch_atime - update the access time
1485 * @mnt: mount the inode is accessed on
1486 * @dentry: dentry accessed
1488 * Update the accessed time on an inode and mark it for writeback.
1489 * This function automatically handles read only file systems and media,
1490 * as well as the "noatime" flag and inode specific "noatime" markers.
1492 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1494 struct inode *inode = dentry->d_inode;
1495 struct timespec now;
1497 if (inode->i_flags & S_NOATIME)
1498 return;
1499 if (IS_NOATIME(inode))
1500 return;
1501 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1502 return;
1504 if (mnt->mnt_flags & MNT_NOATIME)
1505 return;
1506 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1507 return;
1509 now = current_fs_time(inode->i_sb);
1511 if (!relatime_need_update(mnt, inode, now))
1512 return;
1514 if (timespec_equal(&inode->i_atime, &now))
1515 return;
1517 if (mnt_want_write(mnt))
1518 return;
1520 inode->i_atime = now;
1521 mark_inode_dirty_sync(inode);
1522 mnt_drop_write(mnt);
1524 EXPORT_SYMBOL(touch_atime);
1527 * file_update_time - update mtime and ctime time
1528 * @file: file accessed
1530 * Update the mtime and ctime members of an inode and mark the inode
1531 * for writeback. Note that this function is meant exclusively for
1532 * usage in the file write path of filesystems, and filesystems may
1533 * choose to explicitly ignore update via this function with the
1534 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1535 * timestamps are handled by the server.
1538 void file_update_time(struct file *file)
1540 struct inode *inode = file->f_path.dentry->d_inode;
1541 struct timespec now;
1542 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1544 /* First try to exhaust all avenues to not sync */
1545 if (IS_NOCMTIME(inode))
1546 return;
1548 now = current_fs_time(inode->i_sb);
1549 if (!timespec_equal(&inode->i_mtime, &now))
1550 sync_it = S_MTIME;
1552 if (!timespec_equal(&inode->i_ctime, &now))
1553 sync_it |= S_CTIME;
1555 if (IS_I_VERSION(inode))
1556 sync_it |= S_VERSION;
1558 if (!sync_it)
1559 return;
1561 /* Finally allowed to write? Takes lock. */
1562 if (mnt_want_write_file(file))
1563 return;
1565 /* Only change inode inside the lock region */
1566 if (sync_it & S_VERSION)
1567 inode_inc_iversion(inode);
1568 if (sync_it & S_CTIME)
1569 inode->i_ctime = now;
1570 if (sync_it & S_MTIME)
1571 inode->i_mtime = now;
1572 mark_inode_dirty_sync(inode);
1573 mnt_drop_write(file->f_path.mnt);
1575 EXPORT_SYMBOL(file_update_time);
1577 int inode_needs_sync(struct inode *inode)
1579 if (IS_SYNC(inode))
1580 return 1;
1581 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1582 return 1;
1583 return 0;
1585 EXPORT_SYMBOL(inode_needs_sync);
1587 int inode_wait(void *word)
1589 schedule();
1590 return 0;
1592 EXPORT_SYMBOL(inode_wait);
1595 * If we try to find an inode in the inode hash while it is being
1596 * deleted, we have to wait until the filesystem completes its
1597 * deletion before reporting that it isn't found. This function waits
1598 * until the deletion _might_ have completed. Callers are responsible
1599 * to recheck inode state.
1601 * It doesn't matter if I_NEW is not set initially, a call to
1602 * wake_up_inode() after removing from the hash list will DTRT.
1604 * This is called with inode_lock held.
1606 static void __wait_on_freeing_inode(struct inode *inode)
1608 wait_queue_head_t *wq;
1609 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1610 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1611 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1612 spin_unlock(&inode_lock);
1613 schedule();
1614 finish_wait(wq, &wait.wait);
1615 spin_lock(&inode_lock);
1618 static __initdata unsigned long ihash_entries;
1619 static int __init set_ihash_entries(char *str)
1621 if (!str)
1622 return 0;
1623 ihash_entries = simple_strtoul(str, &str, 0);
1624 return 1;
1626 __setup("ihash_entries=", set_ihash_entries);
1629 * Initialize the waitqueues and inode hash table.
1631 void __init inode_init_early(void)
1633 int loop;
1635 /* If hashes are distributed across NUMA nodes, defer
1636 * hash allocation until vmalloc space is available.
1638 if (hashdist)
1639 return;
1641 inode_hashtable =
1642 alloc_large_system_hash("Inode-cache",
1643 sizeof(struct hlist_head),
1644 ihash_entries,
1646 HASH_EARLY,
1647 &i_hash_shift,
1648 &i_hash_mask,
1651 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1652 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1655 void __init inode_init(void)
1657 int loop;
1659 /* inode slab cache */
1660 inode_cachep = kmem_cache_create("inode_cache",
1661 sizeof(struct inode),
1663 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1664 SLAB_MEM_SPREAD),
1665 init_once);
1666 register_shrinker(&icache_shrinker);
1668 /* Hash may have been set up in inode_init_early */
1669 if (!hashdist)
1670 return;
1672 inode_hashtable =
1673 alloc_large_system_hash("Inode-cache",
1674 sizeof(struct hlist_head),
1675 ihash_entries,
1678 &i_hash_shift,
1679 &i_hash_mask,
1682 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1683 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1686 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1688 inode->i_mode = mode;
1689 if (S_ISCHR(mode)) {
1690 inode->i_fop = &def_chr_fops;
1691 inode->i_rdev = rdev;
1692 } else if (S_ISBLK(mode)) {
1693 inode->i_fop = &def_blk_fops;
1694 inode->i_rdev = rdev;
1695 } else if (S_ISFIFO(mode))
1696 inode->i_fop = &def_fifo_fops;
1697 else if (S_ISSOCK(mode))
1698 inode->i_fop = &bad_sock_fops;
1699 else
1700 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1701 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1702 inode->i_ino);
1704 EXPORT_SYMBOL(init_special_inode);
1707 * Init uid,gid,mode for new inode according to posix standards
1708 * @inode: New inode
1709 * @dir: Directory inode
1710 * @mode: mode of the new inode
1712 void inode_init_owner(struct inode *inode, const struct inode *dir,
1713 mode_t mode)
1715 inode->i_uid = current_fsuid();
1716 if (dir && dir->i_mode & S_ISGID) {
1717 inode->i_gid = dir->i_gid;
1718 if (S_ISDIR(mode))
1719 mode |= S_ISGID;
1720 } else
1721 inode->i_gid = current_fsgid();
1722 inode->i_mode = mode;
1724 EXPORT_SYMBOL(inode_init_owner);