blob: af8fd78d2099a0852d7867a4b2691924c1ed3982 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/inode.c
3 *
4 * (C) 1997 Linus Torvalds
5 */
6
7#include <linux/config.h>
8#include <linux/fs.h>
9#include <linux/mm.h>
10#include <linux/dcache.h>
11#include <linux/init.h>
12#include <linux/quotaops.h>
13#include <linux/slab.h>
14#include <linux/writeback.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/wait.h>
18#include <linux/hash.h>
19#include <linux/swap.h>
20#include <linux/security.h>
21#include <linux/pagemap.h>
22#include <linux/cdev.h>
23#include <linux/bootmem.h>
24
25/*
26 * This is needed for the following functions:
27 * - inode_has_buffers
28 * - invalidate_inode_buffers
29 * - fsync_bdev
30 * - invalidate_bdev
31 *
32 * FIXME: remove all knowledge of the buffer layer from this file
33 */
34#include <linux/buffer_head.h>
35
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 */
45
46/* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
47
48/* #define INODE_PARANOIA 1 */
49/* #define INODE_DEBUG 1 */
50
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
57
58static unsigned int i_hash_mask;
59static unsigned int i_hash_shift;
60
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 */
72
73LIST_HEAD(inode_in_use);
74LIST_HEAD(inode_unused);
75static struct hlist_head *inode_hashtable;
76
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 */
83DEFINE_SPINLOCK(inode_lock);
84
85/*
86 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
87 * icache shrinking path, and the umount path. Without this exclusion,
88 * by the time prune_icache calls iput for the inode whose pages it has
89 * been invalidating, or by the time it calls clear_inode & destroy_inode
90 * from its final dispose_list, the struct super_block they refer to
91 * (for inode->i_sb->s_op) may already have been freed and reused.
92 */
93DECLARE_MUTEX(iprune_sem);
94
95/*
96 * Statistics gathering..
97 */
98struct inodes_stat_t inodes_stat;
99
100static kmem_cache_t * inode_cachep;
101
102static struct inode *alloc_inode(struct super_block *sb)
103{
104 static struct address_space_operations empty_aops;
105 static struct inode_operations empty_iops;
106 static struct file_operations empty_fops;
107 struct inode *inode;
108
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);
113
114 if (inode) {
115 struct address_space * const mapping = &inode->i_data;
116
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 }
145
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;
152
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;
160
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}
171
172void destroy_inode(struct inode *inode)
173{
174 if (inode_has_buffers(inode))
175 BUG();
176 security_inode_free(inode);
177 if (inode->i_sb->s_op->destroy_inode)
178 inode->i_sb->s_op->destroy_inode(inode);
179 else
180 kmem_cache_free(inode_cachep, (inode));
181}
182
183
184/*
185 * These are initializations that only need to be done
186 * once, because the fields are idempotent across use
187 * of the inode, so let the slab aware of that.
188 */
189void inode_init_once(struct inode *inode)
190{
191 memset(inode, 0, sizeof(*inode));
192 INIT_HLIST_NODE(&inode->i_hash);
193 INIT_LIST_HEAD(&inode->i_dentry);
194 INIT_LIST_HEAD(&inode->i_devices);
195 sema_init(&inode->i_sem, 1);
196 init_rwsem(&inode->i_alloc_sem);
197 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
198 rwlock_init(&inode->i_data.tree_lock);
199 spin_lock_init(&inode->i_data.i_mmap_lock);
200 INIT_LIST_HEAD(&inode->i_data.private_list);
201 spin_lock_init(&inode->i_data.private_lock);
202 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
203 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
204 spin_lock_init(&inode->i_lock);
205 i_size_ordered_init(inode);
206}
207
208EXPORT_SYMBOL(inode_init_once);
209
210static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
211{
212 struct inode * inode = (struct inode *) foo;
213
214 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
215 SLAB_CTOR_CONSTRUCTOR)
216 inode_init_once(inode);
217}
218
219/*
220 * inode_lock must be held
221 */
222void __iget(struct inode * inode)
223{
224 if (atomic_read(&inode->i_count)) {
225 atomic_inc(&inode->i_count);
226 return;
227 }
228 atomic_inc(&inode->i_count);
229 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
230 list_move(&inode->i_list, &inode_in_use);
231 inodes_stat.nr_unused--;
232}
233
234/**
235 * clear_inode - clear an inode
236 * @inode: inode to clear
237 *
238 * This is called by the filesystem to tell us
239 * that the inode is no longer useful. We just
240 * terminate it with extreme prejudice.
241 */
242void clear_inode(struct inode *inode)
243{
244 might_sleep();
245 invalidate_inode_buffers(inode);
246
247 if (inode->i_data.nrpages)
248 BUG();
249 if (!(inode->i_state & I_FREEING))
250 BUG();
251 if (inode->i_state & I_CLEAR)
252 BUG();
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}
263
264EXPORT_SYMBOL(clear_inode);
265
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 */
273static void dispose_list(struct list_head *head)
274{
275 int nr_disposed = 0;
276
277 while (!list_empty(head)) {
278 struct inode *inode;
279
280 inode = list_entry(head->next, struct inode, i_list);
281 list_del(&inode->i_list);
282
283 if (inode->i_data.nrpages)
284 truncate_inode_pages(&inode->i_data, 0);
285 clear_inode(inode);
286 destroy_inode(inode);
287 nr_disposed++;
288 }
289 spin_lock(&inode_lock);
290 inodes_stat.nr_inodes -= nr_disposed;
291 spin_unlock(&inode_lock);
292}
293
294/*
295 * Invalidate all inodes for a device.
296 */
297static int invalidate_list(struct list_head *head, struct list_head *dispose)
298{
299 struct list_head *next;
300 int busy = 0, count = 0;
301
302 next = head->next;
303 for (;;) {
304 struct list_head * tmp = next;
305 struct inode * inode;
306
307 /*
308 * We can reschedule here without worrying about the list's
309 * consistency because the per-sb list of inodes must not
310 * change during umount anymore, and because iprune_sem keeps
311 * shrink_icache_memory() away.
312 */
313 cond_resched_lock(&inode_lock);
314
315 next = next->next;
316 if (tmp == head)
317 break;
318 inode = list_entry(tmp, struct inode, i_sb_list);
319 invalidate_inode_buffers(inode);
320 if (!atomic_read(&inode->i_count)) {
321 hlist_del_init(&inode->i_hash);
322 list_del(&inode->i_sb_list);
323 list_move(&inode->i_list, dispose);
324 inode->i_state |= I_FREEING;
325 count++;
326 continue;
327 }
328 busy = 1;
329 }
330 /* only unused inodes may be cached with i_count zero */
331 inodes_stat.nr_unused -= count;
332 return busy;
333}
334
335/*
336 * This is a two-stage process. First we collect all
337 * offending inodes onto the throw-away list, and in
338 * the second stage we actually dispose of them. This
339 * is because we don't want to sleep while messing
340 * with the global lists..
341 */
342
343/**
344 * invalidate_inodes - discard the inodes on a device
345 * @sb: superblock
346 *
347 * Discard all of the inodes for a given superblock. If the discard
348 * fails because there are busy inodes then a non zero value is returned.
349 * If the discard is successful all the inodes have been discarded.
350 */
351int invalidate_inodes(struct super_block * sb)
352{
353 int busy;
354 LIST_HEAD(throw_away);
355
356 down(&iprune_sem);
357 spin_lock(&inode_lock);
358 busy = invalidate_list(&sb->s_inodes, &throw_away);
359 spin_unlock(&inode_lock);
360
361 dispose_list(&throw_away);
362 up(&iprune_sem);
363
364 return busy;
365}
366
367EXPORT_SYMBOL(invalidate_inodes);
368
369int __invalidate_device(struct block_device *bdev, int do_sync)
370{
371 struct super_block *sb;
372 int res;
373
374 if (do_sync)
375 fsync_bdev(bdev);
376
377 res = 0;
378 sb = get_super(bdev);
379 if (sb) {
380 /*
381 * no need to lock the super, get_super holds the
382 * read semaphore so the filesystem cannot go away
383 * under us (->put_super runs with the write lock
384 * hold).
385 */
386 shrink_dcache_sb(sb);
387 res = invalidate_inodes(sb);
388 drop_super(sb);
389 }
390 invalidate_bdev(bdev, 0);
391 return res;
392}
393
394EXPORT_SYMBOL(__invalidate_device);
395
396static int can_unuse(struct inode *inode)
397{
398 if (inode->i_state)
399 return 0;
400 if (inode_has_buffers(inode))
401 return 0;
402 if (atomic_read(&inode->i_count))
403 return 0;
404 if (inode->i_data.nrpages)
405 return 0;
406 return 1;
407}
408
409/*
410 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
411 * a temporary list and then are freed outside inode_lock by dispose_list().
412 *
413 * Any inodes which are pinned purely because of attached pagecache have their
414 * pagecache removed. We expect the final iput() on that inode to add it to
415 * the front of the inode_unused list. So look for it there and if the
416 * inode is still freeable, proceed. The right inode is found 99.9% of the
417 * time in testing on a 4-way.
418 *
419 * If the inode has metadata buffers attached to mapping->private_list then
420 * try to remove them.
421 */
422static void prune_icache(int nr_to_scan)
423{
424 LIST_HEAD(freeable);
425 int nr_pruned = 0;
426 int nr_scanned;
427 unsigned long reap = 0;
428
429 down(&iprune_sem);
430 spin_lock(&inode_lock);
431 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
432 struct inode *inode;
433
434 if (list_empty(&inode_unused))
435 break;
436
437 inode = list_entry(inode_unused.prev, struct inode, i_list);
438
439 if (inode->i_state || atomic_read(&inode->i_count)) {
440 list_move(&inode->i_list, &inode_unused);
441 continue;
442 }
443 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
444 __iget(inode);
445 spin_unlock(&inode_lock);
446 if (remove_inode_buffers(inode))
447 reap += invalidate_inode_pages(&inode->i_data);
448 iput(inode);
449 spin_lock(&inode_lock);
450
451 if (inode != list_entry(inode_unused.next,
452 struct inode, i_list))
453 continue; /* wrong inode or list_empty */
454 if (!can_unuse(inode))
455 continue;
456 }
457 hlist_del_init(&inode->i_hash);
458 list_del_init(&inode->i_sb_list);
459 list_move(&inode->i_list, &freeable);
460 inode->i_state |= I_FREEING;
461 nr_pruned++;
462 }
463 inodes_stat.nr_unused -= nr_pruned;
464 spin_unlock(&inode_lock);
465
466 dispose_list(&freeable);
467 up(&iprune_sem);
468
469 if (current_is_kswapd())
470 mod_page_state(kswapd_inodesteal, reap);
471 else
472 mod_page_state(pginodesteal, reap);
473}
474
475/*
476 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
477 * "unused" means that no dentries are referring to the inodes: the files are
478 * not open and the dcache references to those inodes have already been
479 * reclaimed.
480 *
481 * This function is passed the number of inodes to scan, and it returns the
482 * total number of remaining possibly-reclaimable inodes.
483 */
484static int shrink_icache_memory(int nr, unsigned int gfp_mask)
485{
486 if (nr) {
487 /*
488 * Nasty deadlock avoidance. We may hold various FS locks,
489 * and we don't want to recurse into the FS that called us
490 * in clear_inode() and friends..
491 */
492 if (!(gfp_mask & __GFP_FS))
493 return -1;
494 prune_icache(nr);
495 }
496 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
497}
498
499static void __wait_on_freeing_inode(struct inode *inode);
500/*
501 * Called with the inode lock held.
502 * NOTE: we are not increasing the inode-refcount, you must call __iget()
503 * by hand after calling find_inode now! This simplifies iunique and won't
504 * add any additional branch in the common code.
505 */
506static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
507{
508 struct hlist_node *node;
509 struct inode * inode = NULL;
510
511repeat:
512 hlist_for_each (node, head) {
513 inode = hlist_entry(node, struct inode, i_hash);
514 if (inode->i_sb != sb)
515 continue;
516 if (!test(inode, data))
517 continue;
518 if (inode->i_state & (I_FREEING|I_CLEAR)) {
519 __wait_on_freeing_inode(inode);
520 goto repeat;
521 }
522 break;
523 }
524 return node ? inode : NULL;
525}
526
527/*
528 * find_inode_fast is the fast path version of find_inode, see the comment at
529 * iget_locked for details.
530 */
531static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
532{
533 struct hlist_node *node;
534 struct inode * inode = NULL;
535
536repeat:
537 hlist_for_each (node, head) {
538 inode = hlist_entry(node, struct inode, i_hash);
539 if (inode->i_ino != ino)
540 continue;
541 if (inode->i_sb != sb)
542 continue;
543 if (inode->i_state & (I_FREEING|I_CLEAR)) {
544 __wait_on_freeing_inode(inode);
545 goto repeat;
546 }
547 break;
548 }
549 return node ? inode : NULL;
550}
551
552/**
553 * new_inode - obtain an inode
554 * @sb: superblock
555 *
556 * Allocates a new inode for given superblock.
557 */
558struct inode *new_inode(struct super_block *sb)
559{
560 static unsigned long last_ino;
561 struct inode * inode;
562
563 spin_lock_prefetch(&inode_lock);
564
565 inode = alloc_inode(sb);
566 if (inode) {
567 spin_lock(&inode_lock);
568 inodes_stat.nr_inodes++;
569 list_add(&inode->i_list, &inode_in_use);
570 list_add(&inode->i_sb_list, &sb->s_inodes);
571 inode->i_ino = ++last_ino;
572 inode->i_state = 0;
573 spin_unlock(&inode_lock);
574 }
575 return inode;
576}
577
578EXPORT_SYMBOL(new_inode);
579
580void unlock_new_inode(struct inode *inode)
581{
582 /*
583 * This is special! We do not need the spinlock
584 * when clearing I_LOCK, because we're guaranteed
585 * that nobody else tries to do anything about the
586 * state of the inode when it is locked, as we
587 * just created it (so there can be no old holders
588 * that haven't tested I_LOCK).
589 */
590 inode->i_state &= ~(I_LOCK|I_NEW);
591 wake_up_inode(inode);
592}
593
594EXPORT_SYMBOL(unlock_new_inode);
595
596/*
597 * This is called without the inode lock held.. Be careful.
598 *
599 * We no longer cache the sb_flags in i_flags - see fs.h
600 * -- rmk@arm.uk.linux.org
601 */
602static 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)
603{
604 struct inode * inode;
605
606 inode = alloc_inode(sb);
607 if (inode) {
608 struct inode * old;
609
610 spin_lock(&inode_lock);
611 /* We released the lock, so.. */
612 old = find_inode(sb, head, test, data);
613 if (!old) {
614 if (set(inode, data))
615 goto set_failed;
616
617 inodes_stat.nr_inodes++;
618 list_add(&inode->i_list, &inode_in_use);
619 list_add(&inode->i_sb_list, &sb->s_inodes);
620 hlist_add_head(&inode->i_hash, head);
621 inode->i_state = I_LOCK|I_NEW;
622 spin_unlock(&inode_lock);
623
624 /* Return the locked inode with I_NEW set, the
625 * caller is responsible for filling in the contents
626 */
627 return inode;
628 }
629
630 /*
631 * Uhhuh, somebody else created the same inode under
632 * us. Use the old inode instead of the one we just
633 * allocated.
634 */
635 __iget(old);
636 spin_unlock(&inode_lock);
637 destroy_inode(inode);
638 inode = old;
639 wait_on_inode(inode);
640 }
641 return inode;
642
643set_failed:
644 spin_unlock(&inode_lock);
645 destroy_inode(inode);
646 return NULL;
647}
648
649/*
650 * get_new_inode_fast is the fast path version of get_new_inode, see the
651 * comment at iget_locked for details.
652 */
653static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
654{
655 struct inode * inode;
656
657 inode = alloc_inode(sb);
658 if (inode) {
659 struct inode * old;
660
661 spin_lock(&inode_lock);
662 /* We released the lock, so.. */
663 old = find_inode_fast(sb, head, ino);
664 if (!old) {
665 inode->i_ino = ino;
666 inodes_stat.nr_inodes++;
667 list_add(&inode->i_list, &inode_in_use);
668 list_add(&inode->i_sb_list, &sb->s_inodes);
669 hlist_add_head(&inode->i_hash, head);
670 inode->i_state = I_LOCK|I_NEW;
671 spin_unlock(&inode_lock);
672
673 /* Return the locked inode with I_NEW set, the
674 * caller is responsible for filling in the contents
675 */
676 return inode;
677 }
678
679 /*
680 * Uhhuh, somebody else created the same inode under
681 * us. Use the old inode instead of the one we just
682 * allocated.
683 */
684 __iget(old);
685 spin_unlock(&inode_lock);
686 destroy_inode(inode);
687 inode = old;
688 wait_on_inode(inode);
689 }
690 return inode;
691}
692
693static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
694{
695 unsigned long tmp;
696
697 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
698 L1_CACHE_BYTES;
699 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
700 return tmp & I_HASHMASK;
701}
702
703/**
704 * iunique - get a unique inode number
705 * @sb: superblock
706 * @max_reserved: highest reserved inode number
707 *
708 * Obtain an inode number that is unique on the system for a given
709 * superblock. This is used by file systems that have no natural
710 * permanent inode numbering system. An inode number is returned that
711 * is higher than the reserved limit but unique.
712 *
713 * BUGS:
714 * With a large number of inodes live on the file system this function
715 * currently becomes quite slow.
716 */
717ino_t iunique(struct super_block *sb, ino_t max_reserved)
718{
719 static ino_t counter;
720 struct inode *inode;
721 struct hlist_head * head;
722 ino_t res;
723 spin_lock(&inode_lock);
724retry:
725 if (counter > max_reserved) {
726 head = inode_hashtable + hash(sb,counter);
727 res = counter++;
728 inode = find_inode_fast(sb, head, res);
729 if (!inode) {
730 spin_unlock(&inode_lock);
731 return res;
732 }
733 } else {
734 counter = max_reserved + 1;
735 }
736 goto retry;
737
738}
739
740EXPORT_SYMBOL(iunique);
741
742struct inode *igrab(struct inode *inode)
743{
744 spin_lock(&inode_lock);
745 if (!(inode->i_state & I_FREEING))
746 __iget(inode);
747 else
748 /*
749 * Handle the case where s_op->clear_inode is not been
750 * called yet, and somebody is calling igrab
751 * while the inode is getting freed.
752 */
753 inode = NULL;
754 spin_unlock(&inode_lock);
755 return inode;
756}
757
758EXPORT_SYMBOL(igrab);
759
760/**
761 * ifind - internal function, you want ilookup5() or iget5().
762 * @sb: super block of file system to search
763 * @head: the head of the list to search
764 * @test: callback used for comparisons between inodes
765 * @data: opaque data pointer to pass to @test
766 *
767 * ifind() searches for the inode specified by @data in the inode
768 * cache. This is a generalized version of ifind_fast() for file systems where
769 * the inode number is not sufficient for unique identification of an inode.
770 *
771 * If the inode is in the cache, the inode is returned with an incremented
772 * reference count.
773 *
774 * Otherwise NULL is returned.
775 *
776 * Note, @test is called with the inode_lock held, so can't sleep.
777 */
778static inline struct inode *ifind(struct super_block *sb,
779 struct hlist_head *head, int (*test)(struct inode *, void *),
780 void *data)
781{
782 struct inode *inode;
783
784 spin_lock(&inode_lock);
785 inode = find_inode(sb, head, test, data);
786 if (inode) {
787 __iget(inode);
788 spin_unlock(&inode_lock);
789 wait_on_inode(inode);
790 return inode;
791 }
792 spin_unlock(&inode_lock);
793 return NULL;
794}
795
796/**
797 * ifind_fast - internal function, you want ilookup() or iget().
798 * @sb: super block of file system to search
799 * @head: head of the list to search
800 * @ino: inode number to search for
801 *
802 * ifind_fast() searches for the inode @ino in the inode cache. This is for
803 * file systems where the inode number is sufficient for unique identification
804 * of an inode.
805 *
806 * If the inode is in the cache, the inode is returned with an incremented
807 * reference count.
808 *
809 * Otherwise NULL is returned.
810 */
811static inline struct inode *ifind_fast(struct super_block *sb,
812 struct hlist_head *head, unsigned long ino)
813{
814 struct inode *inode;
815
816 spin_lock(&inode_lock);
817 inode = find_inode_fast(sb, head, ino);
818 if (inode) {
819 __iget(inode);
820 spin_unlock(&inode_lock);
821 wait_on_inode(inode);
822 return inode;
823 }
824 spin_unlock(&inode_lock);
825 return NULL;
826}
827
828/**
829 * ilookup5 - search for an inode in the inode cache
830 * @sb: super block of file system to search
831 * @hashval: hash value (usually inode number) to search for
832 * @test: callback used for comparisons between inodes
833 * @data: opaque data pointer to pass to @test
834 *
835 * ilookup5() uses ifind() to search for the inode specified by @hashval and
836 * @data in the inode cache. This is a generalized version of ilookup() for
837 * file systems where the inode number is not sufficient for unique
838 * identification of an inode.
839 *
840 * If the inode is in the cache, the inode is returned with an incremented
841 * reference count.
842 *
843 * Otherwise NULL is returned.
844 *
845 * Note, @test is called with the inode_lock held, so can't sleep.
846 */
847struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
848 int (*test)(struct inode *, void *), void *data)
849{
850 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
851
852 return ifind(sb, head, test, data);
853}
854
855EXPORT_SYMBOL(ilookup5);
856
857/**
858 * ilookup - search for an inode in the inode cache
859 * @sb: super block of file system to search
860 * @ino: inode number to search for
861 *
862 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
863 * This is for file systems where the inode number is sufficient for unique
864 * identification of an inode.
865 *
866 * If the inode is in the cache, the inode is returned with an incremented
867 * reference count.
868 *
869 * Otherwise NULL is returned.
870 */
871struct inode *ilookup(struct super_block *sb, unsigned long ino)
872{
873 struct hlist_head *head = inode_hashtable + hash(sb, ino);
874
875 return ifind_fast(sb, head, ino);
876}
877
878EXPORT_SYMBOL(ilookup);
879
880/**
881 * iget5_locked - obtain an inode from a mounted file system
882 * @sb: super block of file system
883 * @hashval: hash value (usually inode number) to get
884 * @test: callback used for comparisons between inodes
885 * @set: callback used to initialize a new struct inode
886 * @data: opaque data pointer to pass to @test and @set
887 *
888 * This is iget() without the read_inode() portion of get_new_inode().
889 *
890 * iget5_locked() uses ifind() to search for the inode specified by @hashval
891 * and @data in the inode cache and if present it is returned with an increased
892 * reference count. This is a generalized version of iget_locked() for file
893 * systems where the inode number is not sufficient for unique identification
894 * of an inode.
895 *
896 * If the inode is not in cache, get_new_inode() is called to allocate a new
897 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
898 * file system gets to fill it in before unlocking it via unlock_new_inode().
899 *
900 * Note both @test and @set are called with the inode_lock held, so can't sleep.
901 */
902struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
903 int (*test)(struct inode *, void *),
904 int (*set)(struct inode *, void *), void *data)
905{
906 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
907 struct inode *inode;
908
909 inode = ifind(sb, head, test, data);
910 if (inode)
911 return inode;
912 /*
913 * get_new_inode() will do the right thing, re-trying the search
914 * in case it had to block at any point.
915 */
916 return get_new_inode(sb, head, test, set, data);
917}
918
919EXPORT_SYMBOL(iget5_locked);
920
921/**
922 * iget_locked - obtain an inode from a mounted file system
923 * @sb: super block of file system
924 * @ino: inode number to get
925 *
926 * This is iget() without the read_inode() portion of get_new_inode_fast().
927 *
928 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
929 * the inode cache and if present it is returned with an increased reference
930 * count. This is for file systems where the inode number is sufficient for
931 * unique identification of an inode.
932 *
933 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
934 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
935 * The file system gets to fill it in before unlocking it via
936 * unlock_new_inode().
937 */
938struct inode *iget_locked(struct super_block *sb, unsigned long ino)
939{
940 struct hlist_head *head = inode_hashtable + hash(sb, ino);
941 struct inode *inode;
942
943 inode = ifind_fast(sb, head, ino);
944 if (inode)
945 return inode;
946 /*
947 * get_new_inode_fast() will do the right thing, re-trying the search
948 * in case it had to block at any point.
949 */
950 return get_new_inode_fast(sb, head, ino);
951}
952
953EXPORT_SYMBOL(iget_locked);
954
955/**
956 * __insert_inode_hash - hash an inode
957 * @inode: unhashed inode
958 * @hashval: unsigned long value used to locate this object in the
959 * inode_hashtable.
960 *
961 * Add an inode to the inode hash for this superblock.
962 */
963void __insert_inode_hash(struct inode *inode, unsigned long hashval)
964{
965 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
966 spin_lock(&inode_lock);
967 hlist_add_head(&inode->i_hash, head);
968 spin_unlock(&inode_lock);
969}
970
971EXPORT_SYMBOL(__insert_inode_hash);
972
973/**
974 * remove_inode_hash - remove an inode from the hash
975 * @inode: inode to unhash
976 *
977 * Remove an inode from the superblock.
978 */
979void remove_inode_hash(struct inode *inode)
980{
981 spin_lock(&inode_lock);
982 hlist_del_init(&inode->i_hash);
983 spin_unlock(&inode_lock);
984}
985
986EXPORT_SYMBOL(remove_inode_hash);
987
988/*
989 * Tell the filesystem that this inode is no longer of any interest and should
990 * be completely destroyed.
991 *
992 * We leave the inode in the inode hash table until *after* the filesystem's
993 * ->delete_inode completes. This ensures that an iget (such as nfsd might
994 * instigate) will always find up-to-date information either in the hash or on
995 * disk.
996 *
997 * I_FREEING is set so that no-one will take a new reference to the inode while
998 * it is being deleted.
999 */
1000void generic_delete_inode(struct inode *inode)
1001{
1002 struct super_operations *op = inode->i_sb->s_op;
1003
1004 list_del_init(&inode->i_list);
1005 list_del_init(&inode->i_sb_list);
1006 inode->i_state|=I_FREEING;
1007 inodes_stat.nr_inodes--;
1008 spin_unlock(&inode_lock);
1009
1010 if (inode->i_data.nrpages)
1011 truncate_inode_pages(&inode->i_data, 0);
1012
1013 security_inode_delete(inode);
1014
1015 if (op->delete_inode) {
1016 void (*delete)(struct inode *) = op->delete_inode;
1017 if (!is_bad_inode(inode))
1018 DQUOT_INIT(inode);
1019 /* s_op->delete_inode internally recalls clear_inode() */
1020 delete(inode);
1021 } else
1022 clear_inode(inode);
1023 spin_lock(&inode_lock);
1024 hlist_del_init(&inode->i_hash);
1025 spin_unlock(&inode_lock);
1026 wake_up_inode(inode);
1027 if (inode->i_state != I_CLEAR)
1028 BUG();
1029 destroy_inode(inode);
1030}
1031
1032EXPORT_SYMBOL(generic_delete_inode);
1033
1034static void generic_forget_inode(struct inode *inode)
1035{
1036 struct super_block *sb = inode->i_sb;
1037
1038 if (!hlist_unhashed(&inode->i_hash)) {
1039 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1040 list_move(&inode->i_list, &inode_unused);
1041 inodes_stat.nr_unused++;
1042 spin_unlock(&inode_lock);
1043 if (!sb || (sb->s_flags & MS_ACTIVE))
1044 return;
1045 write_inode_now(inode, 1);
1046 spin_lock(&inode_lock);
1047 inodes_stat.nr_unused--;
1048 hlist_del_init(&inode->i_hash);
1049 }
1050 list_del_init(&inode->i_list);
1051 list_del_init(&inode->i_sb_list);
1052 inode->i_state|=I_FREEING;
1053 inodes_stat.nr_inodes--;
1054 spin_unlock(&inode_lock);
1055 if (inode->i_data.nrpages)
1056 truncate_inode_pages(&inode->i_data, 0);
1057 clear_inode(inode);
1058 destroy_inode(inode);
1059}
1060
1061/*
1062 * Normal UNIX filesystem behaviour: delete the
1063 * inode when the usage count drops to zero, and
1064 * i_nlink is zero.
1065 */
1066static void generic_drop_inode(struct inode *inode)
1067{
1068 if (!inode->i_nlink)
1069 generic_delete_inode(inode);
1070 else
1071 generic_forget_inode(inode);
1072}
1073
1074/*
1075 * Called when we're dropping the last reference
1076 * to an inode.
1077 *
1078 * Call the FS "drop()" function, defaulting to
1079 * the legacy UNIX filesystem behaviour..
1080 *
1081 * NOTE! NOTE! NOTE! We're called with the inode lock
1082 * held, and the drop function is supposed to release
1083 * the lock!
1084 */
1085static inline void iput_final(struct inode *inode)
1086{
1087 struct super_operations *op = inode->i_sb->s_op;
1088 void (*drop)(struct inode *) = generic_drop_inode;
1089
1090 if (op && op->drop_inode)
1091 drop = op->drop_inode;
1092 drop(inode);
1093}
1094
1095/**
1096 * iput - put an inode
1097 * @inode: inode to put
1098 *
1099 * Puts an inode, dropping its usage count. If the inode use count hits
1100 * zero, the inode is then freed and may also be destroyed.
1101 *
1102 * Consequently, iput() can sleep.
1103 */
1104void iput(struct inode *inode)
1105{
1106 if (inode) {
1107 struct super_operations *op = inode->i_sb->s_op;
1108
1109 BUG_ON(inode->i_state == I_CLEAR);
1110
1111 if (op && op->put_inode)
1112 op->put_inode(inode);
1113
1114 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1115 iput_final(inode);
1116 }
1117}
1118
1119EXPORT_SYMBOL(iput);
1120
1121/**
1122 * bmap - find a block number in a file
1123 * @inode: inode of file
1124 * @block: block to find
1125 *
1126 * Returns the block number on the device holding the inode that
1127 * is the disk block number for the block of the file requested.
1128 * That is, asked for block 4 of inode 1 the function will return the
1129 * disk block relative to the disk start that holds that block of the
1130 * file.
1131 */
1132sector_t bmap(struct inode * inode, sector_t block)
1133{
1134 sector_t res = 0;
1135 if (inode->i_mapping->a_ops->bmap)
1136 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1137 return res;
1138}
1139
1140EXPORT_SYMBOL(bmap);
1141
1142/**
1143 * update_atime - update the access time
1144 * @inode: inode accessed
1145 *
1146 * Update the accessed time on an inode and mark it for writeback.
1147 * This function automatically handles read only file systems and media,
1148 * as well as the "noatime" flag and inode specific "noatime" markers.
1149 */
1150void update_atime(struct inode *inode)
1151{
1152 struct timespec now;
1153
1154 if (IS_NOATIME(inode))
1155 return;
1156 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1157 return;
1158 if (IS_RDONLY(inode))
1159 return;
1160
1161 now = current_fs_time(inode->i_sb);
1162 if (!timespec_equal(&inode->i_atime, &now)) {
1163 inode->i_atime = now;
1164 mark_inode_dirty_sync(inode);
1165 } else {
1166 if (!timespec_equal(&inode->i_atime, &now))
1167 inode->i_atime = now;
1168 }
1169}
1170
1171EXPORT_SYMBOL(update_atime);
1172
1173/**
1174 * inode_update_time - update mtime and ctime time
1175 * @inode: inode accessed
1176 * @ctime_too: update ctime too
1177 *
1178 * Update the mtime time on an inode and mark it for writeback.
1179 * When ctime_too is specified update the ctime too.
1180 */
1181
1182void inode_update_time(struct inode *inode, int ctime_too)
1183{
1184 struct timespec now;
1185 int sync_it = 0;
1186
1187 if (IS_NOCMTIME(inode))
1188 return;
1189 if (IS_RDONLY(inode))
1190 return;
1191
1192 now = current_fs_time(inode->i_sb);
1193 if (!timespec_equal(&inode->i_mtime, &now))
1194 sync_it = 1;
1195 inode->i_mtime = now;
1196
1197 if (ctime_too) {
1198 if (!timespec_equal(&inode->i_ctime, &now))
1199 sync_it = 1;
1200 inode->i_ctime = now;
1201 }
1202 if (sync_it)
1203 mark_inode_dirty_sync(inode);
1204}
1205
1206EXPORT_SYMBOL(inode_update_time);
1207
1208int inode_needs_sync(struct inode *inode)
1209{
1210 if (IS_SYNC(inode))
1211 return 1;
1212 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1213 return 1;
1214 return 0;
1215}
1216
1217EXPORT_SYMBOL(inode_needs_sync);
1218
1219/*
1220 * Quota functions that want to walk the inode lists..
1221 */
1222#ifdef CONFIG_QUOTA
1223
1224/* Function back in dquot.c */
1225int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1226
1227void remove_dquot_ref(struct super_block *sb, int type,
1228 struct list_head *tofree_head)
1229{
1230 struct inode *inode;
1231
1232 if (!sb->dq_op)
1233 return; /* nothing to do */
1234 spin_lock(&inode_lock); /* This lock is for inodes code */
1235
1236 /*
1237 * We don't have to lock against quota code - test IS_QUOTAINIT is
1238 * just for speedup...
1239 */
1240 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1241 if (!IS_NOQUOTA(inode))
1242 remove_inode_dquot_ref(inode, type, tofree_head);
1243
1244 spin_unlock(&inode_lock);
1245}
1246
1247#endif
1248
1249int inode_wait(void *word)
1250{
1251 schedule();
1252 return 0;
1253}
1254
1255/*
1256 * If we try to find an inode in the inode hash while it is being deleted, we
1257 * have to wait until the filesystem completes its deletion before reporting
1258 * that it isn't found. This is because iget will immediately call
1259 * ->read_inode, and we want to be sure that evidence of the deletion is found
1260 * by ->read_inode.
1261 * This is called with inode_lock held.
1262 */
1263static void __wait_on_freeing_inode(struct inode *inode)
1264{
1265 wait_queue_head_t *wq;
1266 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1267
1268 /*
1269 * I_FREEING and I_CLEAR are cleared in process context under
1270 * inode_lock, so we have to give the tasks who would clear them
1271 * a chance to run and acquire inode_lock.
1272 */
1273 if (!(inode->i_state & I_LOCK)) {
1274 spin_unlock(&inode_lock);
1275 yield();
1276 spin_lock(&inode_lock);
1277 return;
1278 }
1279 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1280 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1281 spin_unlock(&inode_lock);
1282 schedule();
1283 finish_wait(wq, &wait.wait);
1284 spin_lock(&inode_lock);
1285}
1286
1287void wake_up_inode(struct inode *inode)
1288{
1289 /*
1290 * Prevent speculative execution through spin_unlock(&inode_lock);
1291 */
1292 smp_mb();
1293 wake_up_bit(&inode->i_state, __I_LOCK);
1294}
1295
1296static __initdata unsigned long ihash_entries;
1297static int __init set_ihash_entries(char *str)
1298{
1299 if (!str)
1300 return 0;
1301 ihash_entries = simple_strtoul(str, &str, 0);
1302 return 1;
1303}
1304__setup("ihash_entries=", set_ihash_entries);
1305
1306/*
1307 * Initialize the waitqueues and inode hash table.
1308 */
1309void __init inode_init_early(void)
1310{
1311 int loop;
1312
1313 /* If hashes are distributed across NUMA nodes, defer
1314 * hash allocation until vmalloc space is available.
1315 */
1316 if (hashdist)
1317 return;
1318
1319 inode_hashtable =
1320 alloc_large_system_hash("Inode-cache",
1321 sizeof(struct hlist_head),
1322 ihash_entries,
1323 14,
1324 HASH_EARLY,
1325 &i_hash_shift,
1326 &i_hash_mask,
1327 0);
1328
1329 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1330 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1331}
1332
1333void __init inode_init(unsigned long mempages)
1334{
1335 int loop;
1336
1337 /* inode slab cache */
1338 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1339 0, SLAB_PANIC, init_once, NULL);
1340 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1341
1342 /* Hash may have been set up in inode_init_early */
1343 if (!hashdist)
1344 return;
1345
1346 inode_hashtable =
1347 alloc_large_system_hash("Inode-cache",
1348 sizeof(struct hlist_head),
1349 ihash_entries,
1350 14,
1351 0,
1352 &i_hash_shift,
1353 &i_hash_mask,
1354 0);
1355
1356 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1357 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1358}
1359
1360void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1361{
1362 inode->i_mode = mode;
1363 if (S_ISCHR(mode)) {
1364 inode->i_fop = &def_chr_fops;
1365 inode->i_rdev = rdev;
1366 } else if (S_ISBLK(mode)) {
1367 inode->i_fop = &def_blk_fops;
1368 inode->i_rdev = rdev;
1369 } else if (S_ISFIFO(mode))
1370 inode->i_fop = &def_fifo_fops;
1371 else if (S_ISSOCK(mode))
1372 inode->i_fop = &bad_sock_fops;
1373 else
1374 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1375 mode);
1376}
1377EXPORT_SYMBOL(init_special_inode);