ia64/linux-2.6.18-xen.hg

view fs/inode.c @ 524:7f8b544237bf

netfront: Allow netfront in domain 0.

This is useful if your physical network device is in a utility domain.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
author Keir Fraser <keir.fraser@citrix.com>
date Tue Apr 15 15:18:58 2008 +0100 (2008-04-15)
parents 831230e53067
children
line source
1 /*
2 * linux/fs/inode.c
3 *
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/quotaops.h>
12 #include <linux/slab.h>
13 #include <linux/writeback.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/wait.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/inotify.h>
24 #include <linux/mount.h>
26 /*
27 * This is needed for the following functions:
28 * - inode_has_buffers
29 * - invalidate_inode_buffers
30 * - invalidate_bdev
31 *
32 * FIXME: remove all knowledge of the buffer layer from this file
33 */
34 #include <linux/buffer_head.h>
36 /*
37 * New inode.c implementation.
38 *
39 * This implementation has the basic premise of trying
40 * to be extremely low-overhead and SMP-safe, yet be
41 * simple enough to be "obviously correct".
42 *
43 * Famous last words.
44 */
46 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
48 /* #define INODE_PARANOIA 1 */
49 /* #define INODE_DEBUG 1 */
51 /*
52 * Inode lookup is no longer as critical as it used to be:
53 * most of the lookups are going to be through the dcache.
54 */
55 #define I_HASHBITS i_hash_shift
56 #define I_HASHMASK i_hash_mask
58 static unsigned int i_hash_mask __read_mostly;
59 static unsigned int i_hash_shift __read_mostly;
61 /*
62 * Each inode can be on two separate lists. One is
63 * the hash list of the inode, used for lookups. The
64 * other linked list is the "type" list:
65 * "in_use" - valid inode, i_count > 0, i_nlink > 0
66 * "dirty" - as "in_use" but also dirty
67 * "unused" - valid inode, i_count = 0
68 *
69 * A "dirty" list is maintained for each super block,
70 * allowing for low-overhead inode sync() operations.
71 */
73 LIST_HEAD(inode_in_use);
74 LIST_HEAD(inode_unused);
75 static struct hlist_head *inode_hashtable __read_mostly;
77 /*
78 * A simple spinlock to protect the list manipulations.
79 *
80 * NOTE! You also have to own the lock if you change
81 * the i_state of an inode while it is in use..
82 */
83 DEFINE_SPINLOCK(inode_lock);
85 /*
86 * iprune_mutex 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.
92 */
93 static DEFINE_MUTEX(iprune_mutex);
95 /*
96 * Statistics gathering..
97 */
98 struct inodes_stat_t inodes_stat;
100 static kmem_cache_t * inode_cachep __read_mostly;
102 static struct inode *alloc_inode(struct super_block *sb)
103 {
104 static const struct address_space_operations empty_aops;
105 static struct inode_operations empty_iops;
106 static const struct file_operations empty_fops;
107 struct inode *inode;
109 if (sb->s_op->alloc_inode)
110 inode = sb->s_op->alloc_inode(sb);
111 else
112 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
114 if (inode) {
115 struct address_space * const mapping = &inode->i_data;
117 inode->i_sb = sb;
118 inode->i_blkbits = sb->s_blocksize_bits;
119 inode->i_flags = 0;
120 atomic_set(&inode->i_count, 1);
121 inode->i_op = &empty_iops;
122 inode->i_fop = &empty_fops;
123 inode->i_nlink = 1;
124 atomic_set(&inode->i_writecount, 0);
125 inode->i_size = 0;
126 inode->i_blocks = 0;
127 inode->i_bytes = 0;
128 inode->i_generation = 0;
129 #ifdef CONFIG_QUOTA
130 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
131 #endif
132 inode->i_pipe = NULL;
133 inode->i_bdev = NULL;
134 inode->i_cdev = NULL;
135 inode->i_rdev = 0;
136 inode->i_security = NULL;
137 inode->dirtied_when = 0;
138 if (security_inode_alloc(inode)) {
139 if (inode->i_sb->s_op->destroy_inode)
140 inode->i_sb->s_op->destroy_inode(inode);
141 else
142 kmem_cache_free(inode_cachep, (inode));
143 return NULL;
144 }
146 mapping->a_ops = &empty_aops;
147 mapping->host = inode;
148 mapping->flags = 0;
149 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
150 mapping->assoc_mapping = NULL;
151 mapping->backing_dev_info = &default_backing_dev_info;
153 /*
154 * If the block_device provides a backing_dev_info for client
155 * inodes then use that. Otherwise the inode share the bdev's
156 * backing_dev_info.
157 */
158 if (sb->s_bdev) {
159 struct backing_dev_info *bdi;
161 bdi = sb->s_bdev->bd_inode_backing_dev_info;
162 if (!bdi)
163 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
164 mapping->backing_dev_info = bdi;
165 }
166 memset(&inode->u, 0, sizeof(inode->u));
167 inode->i_mapping = mapping;
168 }
169 return inode;
170 }
172 void destroy_inode(struct inode *inode)
173 {
174 BUG_ON(inode_has_buffers(inode));
175 security_inode_free(inode);
176 if (inode->i_sb->s_op->destroy_inode)
177 inode->i_sb->s_op->destroy_inode(inode);
178 else
179 kmem_cache_free(inode_cachep, (inode));
180 }
183 /*
184 * These are initializations that only need to be done
185 * once, because the fields are idempotent across use
186 * of the inode, so let the slab aware of that.
187 */
188 void inode_init_once(struct inode *inode)
189 {
190 memset(inode, 0, sizeof(*inode));
191 INIT_HLIST_NODE(&inode->i_hash);
192 INIT_LIST_HEAD(&inode->i_dentry);
193 INIT_LIST_HEAD(&inode->i_devices);
194 mutex_init(&inode->i_mutex);
195 init_rwsem(&inode->i_alloc_sem);
196 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
197 rwlock_init(&inode->i_data.tree_lock);
198 spin_lock_init(&inode->i_data.i_mmap_lock);
199 INIT_LIST_HEAD(&inode->i_data.private_list);
200 spin_lock_init(&inode->i_data.private_lock);
201 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
202 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
203 spin_lock_init(&inode->i_lock);
204 i_size_ordered_init(inode);
205 #ifdef CONFIG_INOTIFY
206 INIT_LIST_HEAD(&inode->inotify_watches);
207 mutex_init(&inode->inotify_mutex);
208 #endif
209 }
211 EXPORT_SYMBOL(inode_init_once);
213 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
214 {
215 struct inode * inode = (struct inode *) foo;
217 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
218 SLAB_CTOR_CONSTRUCTOR)
219 inode_init_once(inode);
220 }
222 /*
223 * inode_lock must be held
224 */
225 void __iget(struct inode * inode)
226 {
227 if (atomic_read(&inode->i_count)) {
228 atomic_inc(&inode->i_count);
229 return;
230 }
231 atomic_inc(&inode->i_count);
232 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
233 list_move(&inode->i_list, &inode_in_use);
234 inodes_stat.nr_unused--;
235 }
237 /**
238 * clear_inode - clear an inode
239 * @inode: inode to clear
240 *
241 * This is called by the filesystem to tell us
242 * that the inode is no longer useful. We just
243 * terminate it with extreme prejudice.
244 */
245 void clear_inode(struct inode *inode)
246 {
247 might_sleep();
248 invalidate_inode_buffers(inode);
250 BUG_ON(inode->i_data.nrpages);
251 BUG_ON(!(inode->i_state & I_FREEING));
252 BUG_ON(inode->i_state & I_CLEAR);
253 wait_on_inode(inode);
254 DQUOT_DROP(inode);
255 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
256 inode->i_sb->s_op->clear_inode(inode);
257 if (inode->i_bdev)
258 bd_forget(inode);
259 if (inode->i_cdev)
260 cd_forget(inode);
261 inode->i_state = I_CLEAR;
262 }
264 EXPORT_SYMBOL(clear_inode);
266 /*
267 * dispose_list - dispose of the contents of a local list
268 * @head: the head of the list to free
269 *
270 * Dispose-list gets a local list with local inodes in it, so it doesn't
271 * need to worry about list corruption and SMP locks.
272 */
273 static void dispose_list(struct list_head *head)
274 {
275 int nr_disposed = 0;
277 while (!list_empty(head)) {
278 struct inode *inode;
280 inode = list_entry(head->next, struct inode, i_list);
281 list_del(&inode->i_list);
283 if (inode->i_data.nrpages)
284 truncate_inode_pages(&inode->i_data, 0);
285 clear_inode(inode);
287 spin_lock(&inode_lock);
288 hlist_del_init(&inode->i_hash);
289 list_del_init(&inode->i_sb_list);
290 spin_unlock(&inode_lock);
292 wake_up_inode(inode);
293 destroy_inode(inode);
294 nr_disposed++;
295 }
296 spin_lock(&inode_lock);
297 inodes_stat.nr_inodes -= nr_disposed;
298 spin_unlock(&inode_lock);
299 }
301 /*
302 * Invalidate all inodes for a device.
303 */
304 static int invalidate_list(struct list_head *head, struct list_head *dispose)
305 {
306 struct list_head *next;
307 int busy = 0, count = 0;
309 next = head->next;
310 for (;;) {
311 struct list_head * tmp = next;
312 struct inode * inode;
314 /*
315 * We can reschedule here without worrying about the list's
316 * consistency because the per-sb list of inodes must not
317 * change during umount anymore, and because iprune_mutex keeps
318 * shrink_icache_memory() away.
319 */
320 cond_resched_lock(&inode_lock);
322 next = next->next;
323 if (tmp == head)
324 break;
325 inode = list_entry(tmp, struct inode, i_sb_list);
326 invalidate_inode_buffers(inode);
327 if (!atomic_read(&inode->i_count)) {
328 list_move(&inode->i_list, dispose);
329 inode->i_state |= I_FREEING;
330 count++;
331 continue;
332 }
333 busy = 1;
334 }
335 /* only unused inodes may be cached with i_count zero */
336 inodes_stat.nr_unused -= count;
337 return busy;
338 }
340 /**
341 * invalidate_inodes - discard the inodes on a device
342 * @sb: superblock
343 *
344 * Discard all of the inodes for a given superblock. If the discard
345 * fails because there are busy inodes then a non zero value is returned.
346 * If the discard is successful all the inodes have been discarded.
347 */
348 int invalidate_inodes(struct super_block * sb)
349 {
350 int busy;
351 LIST_HEAD(throw_away);
353 mutex_lock(&iprune_mutex);
354 spin_lock(&inode_lock);
355 inotify_unmount_inodes(&sb->s_inodes);
356 busy = invalidate_list(&sb->s_inodes, &throw_away);
357 spin_unlock(&inode_lock);
359 dispose_list(&throw_away);
360 mutex_unlock(&iprune_mutex);
362 return busy;
363 }
365 EXPORT_SYMBOL(invalidate_inodes);
367 int __invalidate_device(struct block_device *bdev)
368 {
369 struct super_block *sb = get_super(bdev);
370 int res = 0;
372 if (sb) {
373 /*
374 * no need to lock the super, get_super holds the
375 * read mutex so the filesystem cannot go away
376 * under us (->put_super runs with the write lock
377 * hold).
378 */
379 shrink_dcache_sb(sb);
380 res = invalidate_inodes(sb);
381 drop_super(sb);
382 }
383 invalidate_bdev(bdev, 0);
384 return res;
385 }
386 EXPORT_SYMBOL(__invalidate_device);
388 static int can_unuse(struct inode *inode)
389 {
390 if (inode->i_state)
391 return 0;
392 if (inode_has_buffers(inode))
393 return 0;
394 if (atomic_read(&inode->i_count))
395 return 0;
396 if (inode->i_data.nrpages)
397 return 0;
398 return 1;
399 }
401 /*
402 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
403 * a temporary list and then are freed outside inode_lock by dispose_list().
404 *
405 * Any inodes which are pinned purely because of attached pagecache have their
406 * pagecache removed. We expect the final iput() on that inode to add it to
407 * the front of the inode_unused list. So look for it there and if the
408 * inode is still freeable, proceed. The right inode is found 99.9% of the
409 * time in testing on a 4-way.
410 *
411 * If the inode has metadata buffers attached to mapping->private_list then
412 * try to remove them.
413 */
414 static void prune_icache(int nr_to_scan)
415 {
416 LIST_HEAD(freeable);
417 int nr_pruned = 0;
418 int nr_scanned;
419 unsigned long reap = 0;
421 mutex_lock(&iprune_mutex);
422 spin_lock(&inode_lock);
423 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
424 struct inode *inode;
426 if (list_empty(&inode_unused))
427 break;
429 inode = list_entry(inode_unused.prev, struct inode, i_list);
431 if (inode->i_state || atomic_read(&inode->i_count)) {
432 list_move(&inode->i_list, &inode_unused);
433 continue;
434 }
435 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
436 __iget(inode);
437 spin_unlock(&inode_lock);
438 if (remove_inode_buffers(inode))
439 reap += invalidate_inode_pages(&inode->i_data);
440 iput(inode);
441 spin_lock(&inode_lock);
443 if (inode != list_entry(inode_unused.next,
444 struct inode, i_list))
445 continue; /* wrong inode or list_empty */
446 if (!can_unuse(inode))
447 continue;
448 }
449 list_move(&inode->i_list, &freeable);
450 inode->i_state |= I_FREEING;
451 nr_pruned++;
452 }
453 inodes_stat.nr_unused -= nr_pruned;
454 if (current_is_kswapd())
455 __count_vm_events(KSWAPD_INODESTEAL, reap);
456 else
457 __count_vm_events(PGINODESTEAL, reap);
458 spin_unlock(&inode_lock);
460 dispose_list(&freeable);
461 mutex_unlock(&iprune_mutex);
462 }
464 /*
465 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
466 * "unused" means that no dentries are referring to the inodes: the files are
467 * not open and the dcache references to those inodes have already been
468 * reclaimed.
469 *
470 * This function is passed the number of inodes to scan, and it returns the
471 * total number of remaining possibly-reclaimable inodes.
472 */
473 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
474 {
475 if (nr) {
476 /*
477 * Nasty deadlock avoidance. We may hold various FS locks,
478 * and we don't want to recurse into the FS that called us
479 * in clear_inode() and friends..
480 */
481 if (!(gfp_mask & __GFP_FS))
482 return -1;
483 prune_icache(nr);
484 }
485 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
486 }
488 static void __wait_on_freeing_inode(struct inode *inode);
489 /*
490 * Called with the inode lock held.
491 * NOTE: we are not increasing the inode-refcount, you must call __iget()
492 * by hand after calling find_inode now! This simplifies iunique and won't
493 * add any additional branch in the common code.
494 */
495 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
496 {
497 struct hlist_node *node;
498 struct inode * inode = NULL;
500 repeat:
501 hlist_for_each (node, head) {
502 inode = hlist_entry(node, struct inode, i_hash);
503 if (inode->i_sb != sb)
504 continue;
505 if (!test(inode, data))
506 continue;
507 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
508 __wait_on_freeing_inode(inode);
509 goto repeat;
510 }
511 break;
512 }
513 return node ? inode : NULL;
514 }
516 /*
517 * find_inode_fast is the fast path version of find_inode, see the comment at
518 * iget_locked for details.
519 */
520 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
521 {
522 struct hlist_node *node;
523 struct inode * inode = NULL;
525 repeat:
526 hlist_for_each (node, head) {
527 inode = hlist_entry(node, struct inode, i_hash);
528 if (inode->i_ino != ino)
529 continue;
530 if (inode->i_sb != sb)
531 continue;
532 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
533 __wait_on_freeing_inode(inode);
534 goto repeat;
535 }
536 break;
537 }
538 return node ? inode : NULL;
539 }
541 /**
542 * new_inode - obtain an inode
543 * @sb: superblock
544 *
545 * Allocates a new inode for given superblock.
546 */
547 struct inode *new_inode(struct super_block *sb)
548 {
549 static unsigned long last_ino;
550 struct inode * inode;
552 spin_lock_prefetch(&inode_lock);
554 inode = alloc_inode(sb);
555 if (inode) {
556 spin_lock(&inode_lock);
557 inodes_stat.nr_inodes++;
558 list_add(&inode->i_list, &inode_in_use);
559 list_add(&inode->i_sb_list, &sb->s_inodes);
560 inode->i_ino = ++last_ino;
561 inode->i_state = 0;
562 spin_unlock(&inode_lock);
563 }
564 return inode;
565 }
567 EXPORT_SYMBOL(new_inode);
569 void unlock_new_inode(struct inode *inode)
570 {
571 /*
572 * This is special! We do not need the spinlock
573 * when clearing I_LOCK, because we're guaranteed
574 * that nobody else tries to do anything about the
575 * state of the inode when it is locked, as we
576 * just created it (so there can be no old holders
577 * that haven't tested I_LOCK).
578 */
579 inode->i_state &= ~(I_LOCK|I_NEW);
580 wake_up_inode(inode);
581 }
583 EXPORT_SYMBOL(unlock_new_inode);
585 /*
586 * This is called without the inode lock held.. Be careful.
587 *
588 * We no longer cache the sb_flags in i_flags - see fs.h
589 * -- rmk@arm.uk.linux.org
590 */
591 static struct inode * get_new_inode(struct super_block *sb, struct hlist_head *head, int (*test)(struct inode *, void *), int (*set)(struct inode *, void *), void *data)
592 {
593 struct inode * inode;
595 inode = alloc_inode(sb);
596 if (inode) {
597 struct inode * old;
599 spin_lock(&inode_lock);
600 /* We released the lock, so.. */
601 old = find_inode(sb, head, test, data);
602 if (!old) {
603 if (set(inode, data))
604 goto set_failed;
606 inodes_stat.nr_inodes++;
607 list_add(&inode->i_list, &inode_in_use);
608 list_add(&inode->i_sb_list, &sb->s_inodes);
609 hlist_add_head(&inode->i_hash, head);
610 inode->i_state = I_LOCK|I_NEW;
611 spin_unlock(&inode_lock);
613 /* Return the locked inode with I_NEW set, the
614 * caller is responsible for filling in the contents
615 */
616 return inode;
617 }
619 /*
620 * Uhhuh, somebody else created the same inode under
621 * us. Use the old inode instead of the one we just
622 * allocated.
623 */
624 __iget(old);
625 spin_unlock(&inode_lock);
626 destroy_inode(inode);
627 inode = old;
628 wait_on_inode(inode);
629 }
630 return inode;
632 set_failed:
633 spin_unlock(&inode_lock);
634 destroy_inode(inode);
635 return NULL;
636 }
638 /*
639 * get_new_inode_fast is the fast path version of get_new_inode, see the
640 * comment at iget_locked for details.
641 */
642 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
643 {
644 struct inode * inode;
646 inode = alloc_inode(sb);
647 if (inode) {
648 struct inode * old;
650 spin_lock(&inode_lock);
651 /* We released the lock, so.. */
652 old = find_inode_fast(sb, head, ino);
653 if (!old) {
654 inode->i_ino = ino;
655 inodes_stat.nr_inodes++;
656 list_add(&inode->i_list, &inode_in_use);
657 list_add(&inode->i_sb_list, &sb->s_inodes);
658 hlist_add_head(&inode->i_hash, head);
659 inode->i_state = I_LOCK|I_NEW;
660 spin_unlock(&inode_lock);
662 /* Return the locked inode with I_NEW set, the
663 * caller is responsible for filling in the contents
664 */
665 return inode;
666 }
668 /*
669 * Uhhuh, somebody else created the same inode under
670 * us. Use the old inode instead of the one we just
671 * allocated.
672 */
673 __iget(old);
674 spin_unlock(&inode_lock);
675 destroy_inode(inode);
676 inode = old;
677 wait_on_inode(inode);
678 }
679 return inode;
680 }
682 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
683 {
684 unsigned long tmp;
686 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
687 L1_CACHE_BYTES;
688 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
689 return tmp & I_HASHMASK;
690 }
692 /**
693 * iunique - get a unique inode number
694 * @sb: superblock
695 * @max_reserved: highest reserved inode number
696 *
697 * Obtain an inode number that is unique on the system for a given
698 * superblock. This is used by file systems that have no natural
699 * permanent inode numbering system. An inode number is returned that
700 * is higher than the reserved limit but unique.
701 *
702 * BUGS:
703 * With a large number of inodes live on the file system this function
704 * currently becomes quite slow.
705 */
706 ino_t iunique(struct super_block *sb, ino_t max_reserved)
707 {
708 static ino_t counter;
709 struct inode *inode;
710 struct hlist_head * head;
711 ino_t res;
712 spin_lock(&inode_lock);
713 retry:
714 if (counter > max_reserved) {
715 head = inode_hashtable + hash(sb,counter);
716 res = counter++;
717 inode = find_inode_fast(sb, head, res);
718 if (!inode) {
719 spin_unlock(&inode_lock);
720 return res;
721 }
722 } else {
723 counter = max_reserved + 1;
724 }
725 goto retry;
727 }
729 EXPORT_SYMBOL(iunique);
731 struct inode *igrab(struct inode *inode)
732 {
733 spin_lock(&inode_lock);
734 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
735 __iget(inode);
736 else
737 /*
738 * Handle the case where s_op->clear_inode is not been
739 * called yet, and somebody is calling igrab
740 * while the inode is getting freed.
741 */
742 inode = NULL;
743 spin_unlock(&inode_lock);
744 return inode;
745 }
747 EXPORT_SYMBOL(igrab);
749 /**
750 * ifind - internal function, you want ilookup5() or iget5().
751 * @sb: super block of file system to search
752 * @head: the head of the list to search
753 * @test: callback used for comparisons between inodes
754 * @data: opaque data pointer to pass to @test
755 * @wait: if true wait for the inode to be unlocked, if false do not
756 *
757 * ifind() searches for the inode specified by @data in the inode
758 * cache. This is a generalized version of ifind_fast() for file systems where
759 * the inode number is not sufficient for unique identification of an inode.
760 *
761 * If the inode is in the cache, the inode is returned with an incremented
762 * reference count.
763 *
764 * Otherwise NULL is returned.
765 *
766 * Note, @test is called with the inode_lock held, so can't sleep.
767 */
768 static struct inode *ifind(struct super_block *sb,
769 struct hlist_head *head, int (*test)(struct inode *, void *),
770 void *data, const int wait)
771 {
772 struct inode *inode;
774 spin_lock(&inode_lock);
775 inode = find_inode(sb, head, test, data);
776 if (inode) {
777 __iget(inode);
778 spin_unlock(&inode_lock);
779 if (likely(wait))
780 wait_on_inode(inode);
781 return inode;
782 }
783 spin_unlock(&inode_lock);
784 return NULL;
785 }
787 /**
788 * ifind_fast - internal function, you want ilookup() or iget().
789 * @sb: super block of file system to search
790 * @head: head of the list to search
791 * @ino: inode number to search for
792 *
793 * ifind_fast() searches for the inode @ino in the inode cache. This is for
794 * file systems where the inode number is sufficient for unique identification
795 * of an inode.
796 *
797 * If the inode is in the cache, the inode is returned with an incremented
798 * reference count.
799 *
800 * Otherwise NULL is returned.
801 */
802 static struct inode *ifind_fast(struct super_block *sb,
803 struct hlist_head *head, unsigned long ino)
804 {
805 struct inode *inode;
807 spin_lock(&inode_lock);
808 inode = find_inode_fast(sb, head, ino);
809 if (inode) {
810 __iget(inode);
811 spin_unlock(&inode_lock);
812 wait_on_inode(inode);
813 return inode;
814 }
815 spin_unlock(&inode_lock);
816 return NULL;
817 }
819 /**
820 * ilookup5_nowait - search for an inode in the inode cache
821 * @sb: super block of file system to search
822 * @hashval: hash value (usually inode number) to search for
823 * @test: callback used for comparisons between inodes
824 * @data: opaque data pointer to pass to @test
825 *
826 * ilookup5() uses ifind() to search for the inode specified by @hashval and
827 * @data in the inode cache. This is a generalized version of ilookup() for
828 * file systems where the inode number is not sufficient for unique
829 * identification of an inode.
830 *
831 * If the inode is in the cache, the inode is returned with an incremented
832 * reference count. Note, the inode lock is not waited upon so you have to be
833 * very careful what you do with the returned inode. You probably should be
834 * using ilookup5() instead.
835 *
836 * Otherwise NULL is returned.
837 *
838 * Note, @test is called with the inode_lock held, so can't sleep.
839 */
840 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
841 int (*test)(struct inode *, void *), void *data)
842 {
843 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
845 return ifind(sb, head, test, data, 0);
846 }
848 EXPORT_SYMBOL(ilookup5_nowait);
850 /**
851 * ilookup5 - search for an inode in the inode cache
852 * @sb: super block of file system to search
853 * @hashval: hash value (usually inode number) to search for
854 * @test: callback used for comparisons between inodes
855 * @data: opaque data pointer to pass to @test
856 *
857 * ilookup5() uses ifind() to search for the inode specified by @hashval and
858 * @data in the inode cache. This is a generalized version of ilookup() for
859 * file systems where the inode number is not sufficient for unique
860 * identification of an inode.
861 *
862 * If the inode is in the cache, the inode lock is waited upon and the inode is
863 * returned with an incremented reference count.
864 *
865 * Otherwise NULL is returned.
866 *
867 * Note, @test is called with the inode_lock held, so can't sleep.
868 */
869 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
870 int (*test)(struct inode *, void *), void *data)
871 {
872 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
874 return ifind(sb, head, test, data, 1);
875 }
877 EXPORT_SYMBOL(ilookup5);
879 /**
880 * ilookup - search for an inode in the inode cache
881 * @sb: super block of file system to search
882 * @ino: inode number to search for
883 *
884 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
885 * This is for file systems where the inode number is sufficient for unique
886 * identification of an inode.
887 *
888 * If the inode is in the cache, the inode is returned with an incremented
889 * reference count.
890 *
891 * Otherwise NULL is returned.
892 */
893 struct inode *ilookup(struct super_block *sb, unsigned long ino)
894 {
895 struct hlist_head *head = inode_hashtable + hash(sb, ino);
897 return ifind_fast(sb, head, ino);
898 }
900 EXPORT_SYMBOL(ilookup);
902 /**
903 * iget5_locked - obtain an inode from a mounted file system
904 * @sb: super block of file system
905 * @hashval: hash value (usually inode number) to get
906 * @test: callback used for comparisons between inodes
907 * @set: callback used to initialize a new struct inode
908 * @data: opaque data pointer to pass to @test and @set
909 *
910 * This is iget() without the read_inode() portion of get_new_inode().
911 *
912 * iget5_locked() uses ifind() to search for the inode specified by @hashval
913 * and @data in the inode cache and if present it is returned with an increased
914 * reference count. This is a generalized version of iget_locked() for file
915 * systems where the inode number is not sufficient for unique identification
916 * of an inode.
917 *
918 * If the inode is not in cache, get_new_inode() is called to allocate a new
919 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
920 * file system gets to fill it in before unlocking it via unlock_new_inode().
921 *
922 * Note both @test and @set are called with the inode_lock held, so can't sleep.
923 */
924 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
925 int (*test)(struct inode *, void *),
926 int (*set)(struct inode *, void *), void *data)
927 {
928 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
929 struct inode *inode;
931 inode = ifind(sb, head, test, data, 1);
932 if (inode)
933 return inode;
934 /*
935 * get_new_inode() will do the right thing, re-trying the search
936 * in case it had to block at any point.
937 */
938 return get_new_inode(sb, head, test, set, data);
939 }
941 EXPORT_SYMBOL(iget5_locked);
943 /**
944 * iget_locked - obtain an inode from a mounted file system
945 * @sb: super block of file system
946 * @ino: inode number to get
947 *
948 * This is iget() without the read_inode() portion of get_new_inode_fast().
949 *
950 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
951 * the inode cache and if present it is returned with an increased reference
952 * count. This is for file systems where the inode number is sufficient for
953 * unique identification of an inode.
954 *
955 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
956 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
957 * The file system gets to fill it in before unlocking it via
958 * unlock_new_inode().
959 */
960 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
961 {
962 struct hlist_head *head = inode_hashtable + hash(sb, ino);
963 struct inode *inode;
965 inode = ifind_fast(sb, head, ino);
966 if (inode)
967 return inode;
968 /*
969 * get_new_inode_fast() will do the right thing, re-trying the search
970 * in case it had to block at any point.
971 */
972 return get_new_inode_fast(sb, head, ino);
973 }
975 EXPORT_SYMBOL(iget_locked);
977 /**
978 * __insert_inode_hash - hash an inode
979 * @inode: unhashed inode
980 * @hashval: unsigned long value used to locate this object in the
981 * inode_hashtable.
982 *
983 * Add an inode to the inode hash for this superblock.
984 */
985 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
986 {
987 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
988 spin_lock(&inode_lock);
989 hlist_add_head(&inode->i_hash, head);
990 spin_unlock(&inode_lock);
991 }
993 EXPORT_SYMBOL(__insert_inode_hash);
995 /**
996 * remove_inode_hash - remove an inode from the hash
997 * @inode: inode to unhash
998 *
999 * Remove an inode from the superblock.
1000 */
1001 void remove_inode_hash(struct inode *inode)
1003 spin_lock(&inode_lock);
1004 hlist_del_init(&inode->i_hash);
1005 spin_unlock(&inode_lock);
1008 EXPORT_SYMBOL(remove_inode_hash);
1010 /*
1011 * Tell the filesystem that this inode is no longer of any interest and should
1012 * be completely destroyed.
1014 * We leave the inode in the inode hash table until *after* the filesystem's
1015 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1016 * instigate) will always find up-to-date information either in the hash or on
1017 * disk.
1019 * I_FREEING is set so that no-one will take a new reference to the inode while
1020 * it is being deleted.
1021 */
1022 void generic_delete_inode(struct inode *inode)
1024 struct super_operations *op = inode->i_sb->s_op;
1026 list_del_init(&inode->i_list);
1027 list_del_init(&inode->i_sb_list);
1028 inode->i_state|=I_FREEING;
1029 inodes_stat.nr_inodes--;
1030 spin_unlock(&inode_lock);
1032 security_inode_delete(inode);
1034 if (op->delete_inode) {
1035 void (*delete)(struct inode *) = op->delete_inode;
1036 if (!is_bad_inode(inode))
1037 DQUOT_INIT(inode);
1038 /* Filesystems implementing their own
1039 * s_op->delete_inode are required to call
1040 * truncate_inode_pages and clear_inode()
1041 * internally */
1042 delete(inode);
1043 } else {
1044 truncate_inode_pages(&inode->i_data, 0);
1045 clear_inode(inode);
1047 spin_lock(&inode_lock);
1048 hlist_del_init(&inode->i_hash);
1049 spin_unlock(&inode_lock);
1050 wake_up_inode(inode);
1051 BUG_ON(inode->i_state != I_CLEAR);
1052 destroy_inode(inode);
1055 EXPORT_SYMBOL(generic_delete_inode);
1057 static void generic_forget_inode(struct inode *inode)
1059 struct super_block *sb = inode->i_sb;
1061 if (!hlist_unhashed(&inode->i_hash)) {
1062 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1063 list_move(&inode->i_list, &inode_unused);
1064 inodes_stat.nr_unused++;
1065 if (!sb || (sb->s_flags & MS_ACTIVE)) {
1066 spin_unlock(&inode_lock);
1067 return;
1069 inode->i_state |= I_WILL_FREE;
1070 spin_unlock(&inode_lock);
1071 write_inode_now(inode, 1);
1072 spin_lock(&inode_lock);
1073 inode->i_state &= ~I_WILL_FREE;
1074 inodes_stat.nr_unused--;
1075 hlist_del_init(&inode->i_hash);
1077 list_del_init(&inode->i_list);
1078 list_del_init(&inode->i_sb_list);
1079 inode->i_state |= I_FREEING;
1080 inodes_stat.nr_inodes--;
1081 spin_unlock(&inode_lock);
1082 if (inode->i_data.nrpages)
1083 truncate_inode_pages(&inode->i_data, 0);
1084 clear_inode(inode);
1085 wake_up_inode(inode);
1086 destroy_inode(inode);
1089 /*
1090 * Normal UNIX filesystem behaviour: delete the
1091 * inode when the usage count drops to zero, and
1092 * i_nlink is zero.
1093 */
1094 void generic_drop_inode(struct inode *inode)
1096 if (!inode->i_nlink)
1097 generic_delete_inode(inode);
1098 else
1099 generic_forget_inode(inode);
1102 EXPORT_SYMBOL_GPL(generic_drop_inode);
1104 /*
1105 * Called when we're dropping the last reference
1106 * to an inode.
1108 * Call the FS "drop()" function, defaulting to
1109 * the legacy UNIX filesystem behaviour..
1111 * NOTE! NOTE! NOTE! We're called with the inode lock
1112 * held, and the drop function is supposed to release
1113 * the lock!
1114 */
1115 static inline void iput_final(struct inode *inode)
1117 struct super_operations *op = inode->i_sb->s_op;
1118 void (*drop)(struct inode *) = generic_drop_inode;
1120 if (op && op->drop_inode)
1121 drop = op->drop_inode;
1122 drop(inode);
1125 /**
1126 * iput - put an inode
1127 * @inode: inode to put
1129 * Puts an inode, dropping its usage count. If the inode use count hits
1130 * zero, the inode is then freed and may also be destroyed.
1132 * Consequently, iput() can sleep.
1133 */
1134 void iput(struct inode *inode)
1136 if (inode) {
1137 struct super_operations *op = inode->i_sb->s_op;
1139 BUG_ON(inode->i_state == I_CLEAR);
1141 if (op && op->put_inode)
1142 op->put_inode(inode);
1144 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1145 iput_final(inode);
1149 EXPORT_SYMBOL(iput);
1151 /**
1152 * bmap - find a block number in a file
1153 * @inode: inode of file
1154 * @block: block to find
1156 * Returns the block number on the device holding the inode that
1157 * is the disk block number for the block of the file requested.
1158 * That is, asked for block 4 of inode 1 the function will return the
1159 * disk block relative to the disk start that holds that block of the
1160 * file.
1161 */
1162 sector_t bmap(struct inode * inode, sector_t block)
1164 sector_t res = 0;
1165 if (inode->i_mapping->a_ops->bmap)
1166 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1167 return res;
1170 EXPORT_SYMBOL(bmap);
1172 /**
1173 * touch_atime - update the access time
1174 * @mnt: mount the inode is accessed on
1175 * @dentry: dentry accessed
1177 * Update the accessed time on an inode and mark it for writeback.
1178 * This function automatically handles read only file systems and media,
1179 * as well as the "noatime" flag and inode specific "noatime" markers.
1180 */
1181 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1183 struct inode *inode = dentry->d_inode;
1184 struct timespec now;
1186 if (IS_RDONLY(inode))
1187 return;
1189 if ((inode->i_flags & S_NOATIME) ||
1190 (inode->i_sb->s_flags & MS_NOATIME) ||
1191 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
1192 return;
1194 /*
1195 * We may have a NULL vfsmount when coming from NFSD
1196 */
1197 if (mnt &&
1198 ((mnt->mnt_flags & MNT_NOATIME) ||
1199 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))))
1200 return;
1202 now = current_fs_time(inode->i_sb);
1203 if (!timespec_equal(&inode->i_atime, &now)) {
1204 inode->i_atime = now;
1205 mark_inode_dirty_sync(inode);
1209 EXPORT_SYMBOL(touch_atime);
1211 /**
1212 * file_update_time - update mtime and ctime time
1213 * @file: file accessed
1215 * Update the mtime and ctime members of an inode and mark the inode
1216 * for writeback. Note that this function is meant exclusively for
1217 * usage in the file write path of filesystems, and filesystems may
1218 * choose to explicitly ignore update via this function with the
1219 * S_NOCTIME inode flag, e.g. for network filesystem where these
1220 * timestamps are handled by the server.
1221 */
1223 void file_update_time(struct file *file)
1225 struct inode *inode = file->f_dentry->d_inode;
1226 struct timespec now;
1227 int sync_it = 0;
1229 if (IS_NOCMTIME(inode))
1230 return;
1231 if (IS_RDONLY(inode))
1232 return;
1234 now = current_fs_time(inode->i_sb);
1235 if (!timespec_equal(&inode->i_mtime, &now))
1236 sync_it = 1;
1237 inode->i_mtime = now;
1239 if (!timespec_equal(&inode->i_ctime, &now))
1240 sync_it = 1;
1241 inode->i_ctime = now;
1243 if (sync_it)
1244 mark_inode_dirty_sync(inode);
1247 EXPORT_SYMBOL(file_update_time);
1249 int inode_needs_sync(struct inode *inode)
1251 if (IS_SYNC(inode))
1252 return 1;
1253 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1254 return 1;
1255 return 0;
1258 EXPORT_SYMBOL(inode_needs_sync);
1260 /*
1261 * Quota functions that want to walk the inode lists..
1262 */
1263 #ifdef CONFIG_QUOTA
1265 /* Function back in dquot.c */
1266 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1268 void remove_dquot_ref(struct super_block *sb, int type,
1269 struct list_head *tofree_head)
1271 struct inode *inode;
1273 if (!sb->dq_op)
1274 return; /* nothing to do */
1275 spin_lock(&inode_lock); /* This lock is for inodes code */
1277 /*
1278 * We don't have to lock against quota code - test IS_QUOTAINIT is
1279 * just for speedup...
1280 */
1281 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1282 if (!IS_NOQUOTA(inode))
1283 remove_inode_dquot_ref(inode, type, tofree_head);
1285 spin_unlock(&inode_lock);
1288 #endif
1290 int inode_wait(void *word)
1292 schedule();
1293 return 0;
1296 /*
1297 * If we try to find an inode in the inode hash while it is being
1298 * deleted, we have to wait until the filesystem completes its
1299 * deletion before reporting that it isn't found. This function waits
1300 * until the deletion _might_ have completed. Callers are responsible
1301 * to recheck inode state.
1303 * It doesn't matter if I_LOCK is not set initially, a call to
1304 * wake_up_inode() after removing from the hash list will DTRT.
1306 * This is called with inode_lock held.
1307 */
1308 static void __wait_on_freeing_inode(struct inode *inode)
1310 wait_queue_head_t *wq;
1311 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1312 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1313 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1314 spin_unlock(&inode_lock);
1315 schedule();
1316 finish_wait(wq, &wait.wait);
1317 spin_lock(&inode_lock);
1320 void wake_up_inode(struct inode *inode)
1322 /*
1323 * Prevent speculative execution through spin_unlock(&inode_lock);
1324 */
1325 smp_mb();
1326 wake_up_bit(&inode->i_state, __I_LOCK);
1329 static __initdata unsigned long ihash_entries;
1330 static int __init set_ihash_entries(char *str)
1332 if (!str)
1333 return 0;
1334 ihash_entries = simple_strtoul(str, &str, 0);
1335 return 1;
1337 __setup("ihash_entries=", set_ihash_entries);
1339 /*
1340 * Initialize the waitqueues and inode hash table.
1341 */
1342 void __init inode_init_early(void)
1344 int loop;
1346 /* If hashes are distributed across NUMA nodes, defer
1347 * hash allocation until vmalloc space is available.
1348 */
1349 if (hashdist)
1350 return;
1352 inode_hashtable =
1353 alloc_large_system_hash("Inode-cache",
1354 sizeof(struct hlist_head),
1355 ihash_entries,
1356 14,
1357 HASH_EARLY,
1358 &i_hash_shift,
1359 &i_hash_mask,
1360 0);
1362 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1363 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1366 void __init inode_init(unsigned long mempages)
1368 int loop;
1370 /* inode slab cache */
1371 inode_cachep = kmem_cache_create("inode_cache",
1372 sizeof(struct inode),
1373 0,
1374 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1375 SLAB_MEM_SPREAD),
1376 init_once,
1377 NULL);
1378 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1380 /* Hash may have been set up in inode_init_early */
1381 if (!hashdist)
1382 return;
1384 inode_hashtable =
1385 alloc_large_system_hash("Inode-cache",
1386 sizeof(struct hlist_head),
1387 ihash_entries,
1388 14,
1389 0,
1390 &i_hash_shift,
1391 &i_hash_mask,
1392 0);
1394 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1395 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1398 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1400 inode->i_mode = mode;
1401 if (S_ISCHR(mode)) {
1402 inode->i_fop = &def_chr_fops;
1403 inode->i_rdev = rdev;
1404 } else if (S_ISBLK(mode)) {
1405 inode->i_fop = &def_blk_fops;
1406 inode->i_rdev = rdev;
1407 } else if (S_ISFIFO(mode))
1408 inode->i_fop = &def_fifo_fops;
1409 else if (S_ISSOCK(mode))
1410 inode->i_fop = &bad_sock_fops;
1411 else
1412 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1413 mode);
1415 EXPORT_SYMBOL(init_special_inode);