blob: 1f9a3a2b89bc67add68e90faa78612ebf6e884cf [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070029 * - invalidate_bdev
30 *
31 * FIXME: remove all knowledge of the buffer layer from this file
32 */
33#include <linux/buffer_head.h>
34
35/*
36 * New inode.c implementation.
37 *
38 * This implementation has the basic premise of trying
39 * to be extremely low-overhead and SMP-safe, yet be
40 * simple enough to be "obviously correct".
41 *
42 * Famous last words.
43 */
44
45/* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
46
47/* #define INODE_PARANOIA 1 */
48/* #define INODE_DEBUG 1 */
49
50/*
51 * Inode lookup is no longer as critical as it used to be:
52 * most of the lookups are going to be through the dcache.
53 */
54#define I_HASHBITS i_hash_shift
55#define I_HASHMASK i_hash_mask
56
57static unsigned int i_hash_mask;
58static unsigned int i_hash_shift;
59
60/*
61 * Each inode can be on two separate lists. One is
62 * the hash list of the inode, used for lookups. The
63 * other linked list is the "type" list:
64 * "in_use" - valid inode, i_count > 0, i_nlink > 0
65 * "dirty" - as "in_use" but also dirty
66 * "unused" - valid inode, i_count = 0
67 *
68 * A "dirty" list is maintained for each super block,
69 * allowing for low-overhead inode sync() operations.
70 */
71
72LIST_HEAD(inode_in_use);
73LIST_HEAD(inode_unused);
74static struct hlist_head *inode_hashtable;
75
76/*
77 * A simple spinlock to protect the list manipulations.
78 *
79 * NOTE! You also have to own the lock if you change
80 * the i_state of an inode while it is in use..
81 */
82DEFINE_SPINLOCK(inode_lock);
83
84/*
85 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
86 * icache shrinking path, and the umount path. Without this exclusion,
87 * by the time prune_icache calls iput for the inode whose pages it has
88 * been invalidating, or by the time it calls clear_inode & destroy_inode
89 * from its final dispose_list, the struct super_block they refer to
90 * (for inode->i_sb->s_op) may already have been freed and reused.
91 */
92DECLARE_MUTEX(iprune_sem);
93
94/*
95 * Statistics gathering..
96 */
97struct inodes_stat_t inodes_stat;
98
99static kmem_cache_t * inode_cachep;
100
101static struct inode *alloc_inode(struct super_block *sb)
102{
103 static struct address_space_operations empty_aops;
104 static struct inode_operations empty_iops;
105 static struct file_operations empty_fops;
106 struct inode *inode;
107
108 if (sb->s_op->alloc_inode)
109 inode = sb->s_op->alloc_inode(sb);
110 else
111 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
112
113 if (inode) {
114 struct address_space * const mapping = &inode->i_data;
115
116 inode->i_sb = sb;
117 inode->i_blkbits = sb->s_blocksize_bits;
118 inode->i_flags = 0;
119 atomic_set(&inode->i_count, 1);
120 inode->i_op = &empty_iops;
121 inode->i_fop = &empty_fops;
122 inode->i_nlink = 1;
123 atomic_set(&inode->i_writecount, 0);
124 inode->i_size = 0;
125 inode->i_blocks = 0;
126 inode->i_bytes = 0;
127 inode->i_generation = 0;
128#ifdef CONFIG_QUOTA
129 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
130#endif
131 inode->i_pipe = NULL;
132 inode->i_bdev = NULL;
133 inode->i_cdev = NULL;
134 inode->i_rdev = 0;
135 inode->i_security = NULL;
136 inode->dirtied_when = 0;
137 if (security_inode_alloc(inode)) {
138 if (inode->i_sb->s_op->destroy_inode)
139 inode->i_sb->s_op->destroy_inode(inode);
140 else
141 kmem_cache_free(inode_cachep, (inode));
142 return NULL;
143 }
144
145 mapping->a_ops = &empty_aops;
146 mapping->host = inode;
147 mapping->flags = 0;
148 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
149 mapping->assoc_mapping = NULL;
150 mapping->backing_dev_info = &default_backing_dev_info;
151
152 /*
153 * If the block_device provides a backing_dev_info for client
154 * inodes then use that. Otherwise the inode share the bdev's
155 * backing_dev_info.
156 */
157 if (sb->s_bdev) {
158 struct backing_dev_info *bdi;
159
160 bdi = sb->s_bdev->bd_inode_backing_dev_info;
161 if (!bdi)
162 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
163 mapping->backing_dev_info = bdi;
164 }
165 memset(&inode->u, 0, sizeof(inode->u));
166 inode->i_mapping = mapping;
167 }
168 return inode;
169}
170
171void destroy_inode(struct inode *inode)
172{
173 if (inode_has_buffers(inode))
174 BUG();
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}
181
182
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 */
188void 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 sema_init(&inode->i_sem, 1);
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}
206
207EXPORT_SYMBOL(inode_init_once);
208
209static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
210{
211 struct inode * inode = (struct inode *) foo;
212
213 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
214 SLAB_CTOR_CONSTRUCTOR)
215 inode_init_once(inode);
216}
217
218/*
219 * inode_lock must be held
220 */
221void __iget(struct inode * inode)
222{
223 if (atomic_read(&inode->i_count)) {
224 atomic_inc(&inode->i_count);
225 return;
226 }
227 atomic_inc(&inode->i_count);
228 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
229 list_move(&inode->i_list, &inode_in_use);
230 inodes_stat.nr_unused--;
231}
232
233/**
234 * clear_inode - clear an inode
235 * @inode: inode to clear
236 *
237 * This is called by the filesystem to tell us
238 * that the inode is no longer useful. We just
239 * terminate it with extreme prejudice.
240 */
241void clear_inode(struct inode *inode)
242{
243 might_sleep();
244 invalidate_inode_buffers(inode);
245
246 if (inode->i_data.nrpages)
247 BUG();
248 if (!(inode->i_state & I_FREEING))
249 BUG();
250 if (inode->i_state & I_CLEAR)
251 BUG();
252 wait_on_inode(inode);
253 DQUOT_DROP(inode);
254 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
255 inode->i_sb->s_op->clear_inode(inode);
256 if (inode->i_bdev)
257 bd_forget(inode);
258 if (inode->i_cdev)
259 cd_forget(inode);
260 inode->i_state = I_CLEAR;
261}
262
263EXPORT_SYMBOL(clear_inode);
264
265/*
266 * dispose_list - dispose of the contents of a local list
267 * @head: the head of the list to free
268 *
269 * Dispose-list gets a local list with local inodes in it, so it doesn't
270 * need to worry about list corruption and SMP locks.
271 */
272static void dispose_list(struct list_head *head)
273{
274 int nr_disposed = 0;
275
276 while (!list_empty(head)) {
277 struct inode *inode;
278
279 inode = list_entry(head->next, struct inode, i_list);
280 list_del(&inode->i_list);
281
282 if (inode->i_data.nrpages)
283 truncate_inode_pages(&inode->i_data, 0);
284 clear_inode(inode);
285 destroy_inode(inode);
286 nr_disposed++;
287 }
288 spin_lock(&inode_lock);
289 inodes_stat.nr_inodes -= nr_disposed;
290 spin_unlock(&inode_lock);
291}
292
293/*
294 * Invalidate all inodes for a device.
295 */
296static int invalidate_list(struct list_head *head, struct list_head *dispose)
297{
298 struct list_head *next;
299 int busy = 0, count = 0;
300
301 next = head->next;
302 for (;;) {
303 struct list_head * tmp = next;
304 struct inode * inode;
305
306 /*
307 * We can reschedule here without worrying about the list's
308 * consistency because the per-sb list of inodes must not
309 * change during umount anymore, and because iprune_sem keeps
310 * shrink_icache_memory() away.
311 */
312 cond_resched_lock(&inode_lock);
313
314 next = next->next;
315 if (tmp == head)
316 break;
317 inode = list_entry(tmp, struct inode, i_sb_list);
318 invalidate_inode_buffers(inode);
319 if (!atomic_read(&inode->i_count)) {
320 hlist_del_init(&inode->i_hash);
321 list_del(&inode->i_sb_list);
322 list_move(&inode->i_list, dispose);
323 inode->i_state |= I_FREEING;
324 count++;
325 continue;
326 }
327 busy = 1;
328 }
329 /* only unused inodes may be cached with i_count zero */
330 inodes_stat.nr_unused -= count;
331 return busy;
332}
333
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334/**
335 * invalidate_inodes - discard the inodes on a device
336 * @sb: superblock
337 *
338 * Discard all of the inodes for a given superblock. If the discard
339 * fails because there are busy inodes then a non zero value is returned.
340 * If the discard is successful all the inodes have been discarded.
341 */
342int invalidate_inodes(struct super_block * sb)
343{
344 int busy;
345 LIST_HEAD(throw_away);
346
347 down(&iprune_sem);
348 spin_lock(&inode_lock);
349 busy = invalidate_list(&sb->s_inodes, &throw_away);
350 spin_unlock(&inode_lock);
351
352 dispose_list(&throw_away);
353 up(&iprune_sem);
354
355 return busy;
356}
357
358EXPORT_SYMBOL(invalidate_inodes);
359
Christoph Hellwig2ef41632005-05-05 16:15:59 -0700360int __invalidate_device(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361{
Christoph Hellwig2ef41632005-05-05 16:15:59 -0700362 struct super_block *sb = get_super(bdev);
363 int res = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 if (sb) {
366 /*
367 * no need to lock the super, get_super holds the
368 * read semaphore so the filesystem cannot go away
369 * under us (->put_super runs with the write lock
370 * hold).
371 */
372 shrink_dcache_sb(sb);
373 res = invalidate_inodes(sb);
374 drop_super(sb);
375 }
376 invalidate_bdev(bdev, 0);
377 return res;
378}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379EXPORT_SYMBOL(__invalidate_device);
380
381static int can_unuse(struct inode *inode)
382{
383 if (inode->i_state)
384 return 0;
385 if (inode_has_buffers(inode))
386 return 0;
387 if (atomic_read(&inode->i_count))
388 return 0;
389 if (inode->i_data.nrpages)
390 return 0;
391 return 1;
392}
393
394/*
395 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
396 * a temporary list and then are freed outside inode_lock by dispose_list().
397 *
398 * Any inodes which are pinned purely because of attached pagecache have their
399 * pagecache removed. We expect the final iput() on that inode to add it to
400 * the front of the inode_unused list. So look for it there and if the
401 * inode is still freeable, proceed. The right inode is found 99.9% of the
402 * time in testing on a 4-way.
403 *
404 * If the inode has metadata buffers attached to mapping->private_list then
405 * try to remove them.
406 */
407static void prune_icache(int nr_to_scan)
408{
409 LIST_HEAD(freeable);
410 int nr_pruned = 0;
411 int nr_scanned;
412 unsigned long reap = 0;
413
414 down(&iprune_sem);
415 spin_lock(&inode_lock);
416 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
417 struct inode *inode;
418
419 if (list_empty(&inode_unused))
420 break;
421
422 inode = list_entry(inode_unused.prev, struct inode, i_list);
423
424 if (inode->i_state || atomic_read(&inode->i_count)) {
425 list_move(&inode->i_list, &inode_unused);
426 continue;
427 }
428 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
429 __iget(inode);
430 spin_unlock(&inode_lock);
431 if (remove_inode_buffers(inode))
432 reap += invalidate_inode_pages(&inode->i_data);
433 iput(inode);
434 spin_lock(&inode_lock);
435
436 if (inode != list_entry(inode_unused.next,
437 struct inode, i_list))
438 continue; /* wrong inode or list_empty */
439 if (!can_unuse(inode))
440 continue;
441 }
442 hlist_del_init(&inode->i_hash);
443 list_del_init(&inode->i_sb_list);
444 list_move(&inode->i_list, &freeable);
445 inode->i_state |= I_FREEING;
446 nr_pruned++;
447 }
448 inodes_stat.nr_unused -= nr_pruned;
449 spin_unlock(&inode_lock);
450
451 dispose_list(&freeable);
452 up(&iprune_sem);
453
454 if (current_is_kswapd())
455 mod_page_state(kswapd_inodesteal, reap);
456 else
457 mod_page_state(pginodesteal, reap);
458}
459
460/*
461 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
462 * "unused" means that no dentries are referring to the inodes: the files are
463 * not open and the dcache references to those inodes have already been
464 * reclaimed.
465 *
466 * This function is passed the number of inodes to scan, and it returns the
467 * total number of remaining possibly-reclaimable inodes.
468 */
469static int shrink_icache_memory(int nr, unsigned int gfp_mask)
470{
471 if (nr) {
472 /*
473 * Nasty deadlock avoidance. We may hold various FS locks,
474 * and we don't want to recurse into the FS that called us
475 * in clear_inode() and friends..
476 */
477 if (!(gfp_mask & __GFP_FS))
478 return -1;
479 prune_icache(nr);
480 }
481 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
482}
483
484static void __wait_on_freeing_inode(struct inode *inode);
485/*
486 * Called with the inode lock held.
487 * NOTE: we are not increasing the inode-refcount, you must call __iget()
488 * by hand after calling find_inode now! This simplifies iunique and won't
489 * add any additional branch in the common code.
490 */
491static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
492{
493 struct hlist_node *node;
494 struct inode * inode = NULL;
495
496repeat:
497 hlist_for_each (node, head) {
498 inode = hlist_entry(node, struct inode, i_hash);
499 if (inode->i_sb != sb)
500 continue;
501 if (!test(inode, data))
502 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700503 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 __wait_on_freeing_inode(inode);
505 goto repeat;
506 }
507 break;
508 }
509 return node ? inode : NULL;
510}
511
512/*
513 * find_inode_fast is the fast path version of find_inode, see the comment at
514 * iget_locked for details.
515 */
516static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
517{
518 struct hlist_node *node;
519 struct inode * inode = NULL;
520
521repeat:
522 hlist_for_each (node, head) {
523 inode = hlist_entry(node, struct inode, i_hash);
524 if (inode->i_ino != ino)
525 continue;
526 if (inode->i_sb != sb)
527 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700528 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 __wait_on_freeing_inode(inode);
530 goto repeat;
531 }
532 break;
533 }
534 return node ? inode : NULL;
535}
536
537/**
538 * new_inode - obtain an inode
539 * @sb: superblock
540 *
541 * Allocates a new inode for given superblock.
542 */
543struct inode *new_inode(struct super_block *sb)
544{
545 static unsigned long last_ino;
546 struct inode * inode;
547
548 spin_lock_prefetch(&inode_lock);
549
550 inode = alloc_inode(sb);
551 if (inode) {
552 spin_lock(&inode_lock);
553 inodes_stat.nr_inodes++;
554 list_add(&inode->i_list, &inode_in_use);
555 list_add(&inode->i_sb_list, &sb->s_inodes);
556 inode->i_ino = ++last_ino;
557 inode->i_state = 0;
558 spin_unlock(&inode_lock);
559 }
560 return inode;
561}
562
563EXPORT_SYMBOL(new_inode);
564
565void unlock_new_inode(struct inode *inode)
566{
567 /*
568 * This is special! We do not need the spinlock
569 * when clearing I_LOCK, because we're guaranteed
570 * that nobody else tries to do anything about the
571 * state of the inode when it is locked, as we
572 * just created it (so there can be no old holders
573 * that haven't tested I_LOCK).
574 */
575 inode->i_state &= ~(I_LOCK|I_NEW);
576 wake_up_inode(inode);
577}
578
579EXPORT_SYMBOL(unlock_new_inode);
580
581/*
582 * This is called without the inode lock held.. Be careful.
583 *
584 * We no longer cache the sb_flags in i_flags - see fs.h
585 * -- rmk@arm.uk.linux.org
586 */
587static 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)
588{
589 struct inode * inode;
590
591 inode = alloc_inode(sb);
592 if (inode) {
593 struct inode * old;
594
595 spin_lock(&inode_lock);
596 /* We released the lock, so.. */
597 old = find_inode(sb, head, test, data);
598 if (!old) {
599 if (set(inode, data))
600 goto set_failed;
601
602 inodes_stat.nr_inodes++;
603 list_add(&inode->i_list, &inode_in_use);
604 list_add(&inode->i_sb_list, &sb->s_inodes);
605 hlist_add_head(&inode->i_hash, head);
606 inode->i_state = I_LOCK|I_NEW;
607 spin_unlock(&inode_lock);
608
609 /* Return the locked inode with I_NEW set, the
610 * caller is responsible for filling in the contents
611 */
612 return inode;
613 }
614
615 /*
616 * Uhhuh, somebody else created the same inode under
617 * us. Use the old inode instead of the one we just
618 * allocated.
619 */
620 __iget(old);
621 spin_unlock(&inode_lock);
622 destroy_inode(inode);
623 inode = old;
624 wait_on_inode(inode);
625 }
626 return inode;
627
628set_failed:
629 spin_unlock(&inode_lock);
630 destroy_inode(inode);
631 return NULL;
632}
633
634/*
635 * get_new_inode_fast is the fast path version of get_new_inode, see the
636 * comment at iget_locked for details.
637 */
638static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
639{
640 struct inode * inode;
641
642 inode = alloc_inode(sb);
643 if (inode) {
644 struct inode * old;
645
646 spin_lock(&inode_lock);
647 /* We released the lock, so.. */
648 old = find_inode_fast(sb, head, ino);
649 if (!old) {
650 inode->i_ino = ino;
651 inodes_stat.nr_inodes++;
652 list_add(&inode->i_list, &inode_in_use);
653 list_add(&inode->i_sb_list, &sb->s_inodes);
654 hlist_add_head(&inode->i_hash, head);
655 inode->i_state = I_LOCK|I_NEW;
656 spin_unlock(&inode_lock);
657
658 /* Return the locked inode with I_NEW set, the
659 * caller is responsible for filling in the contents
660 */
661 return inode;
662 }
663
664 /*
665 * Uhhuh, somebody else created the same inode under
666 * us. Use the old inode instead of the one we just
667 * allocated.
668 */
669 __iget(old);
670 spin_unlock(&inode_lock);
671 destroy_inode(inode);
672 inode = old;
673 wait_on_inode(inode);
674 }
675 return inode;
676}
677
678static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
679{
680 unsigned long tmp;
681
682 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
683 L1_CACHE_BYTES;
684 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
685 return tmp & I_HASHMASK;
686}
687
688/**
689 * iunique - get a unique inode number
690 * @sb: superblock
691 * @max_reserved: highest reserved inode number
692 *
693 * Obtain an inode number that is unique on the system for a given
694 * superblock. This is used by file systems that have no natural
695 * permanent inode numbering system. An inode number is returned that
696 * is higher than the reserved limit but unique.
697 *
698 * BUGS:
699 * With a large number of inodes live on the file system this function
700 * currently becomes quite slow.
701 */
702ino_t iunique(struct super_block *sb, ino_t max_reserved)
703{
704 static ino_t counter;
705 struct inode *inode;
706 struct hlist_head * head;
707 ino_t res;
708 spin_lock(&inode_lock);
709retry:
710 if (counter > max_reserved) {
711 head = inode_hashtable + hash(sb,counter);
712 res = counter++;
713 inode = find_inode_fast(sb, head, res);
714 if (!inode) {
715 spin_unlock(&inode_lock);
716 return res;
717 }
718 } else {
719 counter = max_reserved + 1;
720 }
721 goto retry;
722
723}
724
725EXPORT_SYMBOL(iunique);
726
727struct inode *igrab(struct inode *inode)
728{
729 spin_lock(&inode_lock);
Alexander Viro991114c2005-06-23 00:09:01 -0700730 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 __iget(inode);
732 else
733 /*
734 * Handle the case where s_op->clear_inode is not been
735 * called yet, and somebody is calling igrab
736 * while the inode is getting freed.
737 */
738 inode = NULL;
739 spin_unlock(&inode_lock);
740 return inode;
741}
742
743EXPORT_SYMBOL(igrab);
744
745/**
746 * ifind - internal function, you want ilookup5() or iget5().
747 * @sb: super block of file system to search
748 * @head: the head of the list to search
749 * @test: callback used for comparisons between inodes
750 * @data: opaque data pointer to pass to @test
751 *
752 * ifind() searches for the inode specified by @data in the inode
753 * cache. This is a generalized version of ifind_fast() for file systems where
754 * the inode number is not sufficient for unique identification of an inode.
755 *
756 * If the inode is in the cache, the inode is returned with an incremented
757 * reference count.
758 *
759 * Otherwise NULL is returned.
760 *
761 * Note, @test is called with the inode_lock held, so can't sleep.
762 */
763static inline struct inode *ifind(struct super_block *sb,
764 struct hlist_head *head, int (*test)(struct inode *, void *),
765 void *data)
766{
767 struct inode *inode;
768
769 spin_lock(&inode_lock);
770 inode = find_inode(sb, head, test, data);
771 if (inode) {
772 __iget(inode);
773 spin_unlock(&inode_lock);
774 wait_on_inode(inode);
775 return inode;
776 }
777 spin_unlock(&inode_lock);
778 return NULL;
779}
780
781/**
782 * ifind_fast - internal function, you want ilookup() or iget().
783 * @sb: super block of file system to search
784 * @head: head of the list to search
785 * @ino: inode number to search for
786 *
787 * ifind_fast() searches for the inode @ino in the inode cache. This is for
788 * file systems where the inode number is sufficient for unique identification
789 * of an inode.
790 *
791 * If the inode is in the cache, the inode is returned with an incremented
792 * reference count.
793 *
794 * Otherwise NULL is returned.
795 */
796static inline struct inode *ifind_fast(struct super_block *sb,
797 struct hlist_head *head, unsigned long ino)
798{
799 struct inode *inode;
800
801 spin_lock(&inode_lock);
802 inode = find_inode_fast(sb, head, ino);
803 if (inode) {
804 __iget(inode);
805 spin_unlock(&inode_lock);
806 wait_on_inode(inode);
807 return inode;
808 }
809 spin_unlock(&inode_lock);
810 return NULL;
811}
812
813/**
814 * ilookup5 - search for an inode in the inode cache
815 * @sb: super block of file system to search
816 * @hashval: hash value (usually inode number) to search for
817 * @test: callback used for comparisons between inodes
818 * @data: opaque data pointer to pass to @test
819 *
820 * ilookup5() uses ifind() to search for the inode specified by @hashval and
821 * @data in the inode cache. This is a generalized version of ilookup() for
822 * file systems where the inode number is not sufficient for unique
823 * identification of an inode.
824 *
825 * If the inode is in the cache, the inode is returned with an incremented
826 * reference count.
827 *
828 * Otherwise NULL is returned.
829 *
830 * Note, @test is called with the inode_lock held, so can't sleep.
831 */
832struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
833 int (*test)(struct inode *, void *), void *data)
834{
835 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
836
837 return ifind(sb, head, test, data);
838}
839
840EXPORT_SYMBOL(ilookup5);
841
842/**
843 * ilookup - search for an inode in the inode cache
844 * @sb: super block of file system to search
845 * @ino: inode number to search for
846 *
847 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
848 * This is for file systems where the inode number is sufficient for unique
849 * identification of an inode.
850 *
851 * If the inode is in the cache, the inode is returned with an incremented
852 * reference count.
853 *
854 * Otherwise NULL is returned.
855 */
856struct inode *ilookup(struct super_block *sb, unsigned long ino)
857{
858 struct hlist_head *head = inode_hashtable + hash(sb, ino);
859
860 return ifind_fast(sb, head, ino);
861}
862
863EXPORT_SYMBOL(ilookup);
864
865/**
866 * iget5_locked - obtain an inode from a mounted file system
867 * @sb: super block of file system
868 * @hashval: hash value (usually inode number) to get
869 * @test: callback used for comparisons between inodes
870 * @set: callback used to initialize a new struct inode
871 * @data: opaque data pointer to pass to @test and @set
872 *
873 * This is iget() without the read_inode() portion of get_new_inode().
874 *
875 * iget5_locked() uses ifind() to search for the inode specified by @hashval
876 * and @data in the inode cache and if present it is returned with an increased
877 * reference count. This is a generalized version of iget_locked() for file
878 * systems where the inode number is not sufficient for unique identification
879 * of an inode.
880 *
881 * If the inode is not in cache, get_new_inode() is called to allocate a new
882 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
883 * file system gets to fill it in before unlocking it via unlock_new_inode().
884 *
885 * Note both @test and @set are called with the inode_lock held, so can't sleep.
886 */
887struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
888 int (*test)(struct inode *, void *),
889 int (*set)(struct inode *, void *), void *data)
890{
891 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
892 struct inode *inode;
893
894 inode = ifind(sb, head, test, data);
895 if (inode)
896 return inode;
897 /*
898 * get_new_inode() will do the right thing, re-trying the search
899 * in case it had to block at any point.
900 */
901 return get_new_inode(sb, head, test, set, data);
902}
903
904EXPORT_SYMBOL(iget5_locked);
905
906/**
907 * iget_locked - obtain an inode from a mounted file system
908 * @sb: super block of file system
909 * @ino: inode number to get
910 *
911 * This is iget() without the read_inode() portion of get_new_inode_fast().
912 *
913 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
914 * the inode cache and if present it is returned with an increased reference
915 * count. This is for file systems where the inode number is sufficient for
916 * unique identification of an inode.
917 *
918 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
919 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
920 * The file system gets to fill it in before unlocking it via
921 * unlock_new_inode().
922 */
923struct inode *iget_locked(struct super_block *sb, unsigned long ino)
924{
925 struct hlist_head *head = inode_hashtable + hash(sb, ino);
926 struct inode *inode;
927
928 inode = ifind_fast(sb, head, ino);
929 if (inode)
930 return inode;
931 /*
932 * get_new_inode_fast() will do the right thing, re-trying the search
933 * in case it had to block at any point.
934 */
935 return get_new_inode_fast(sb, head, ino);
936}
937
938EXPORT_SYMBOL(iget_locked);
939
940/**
941 * __insert_inode_hash - hash an inode
942 * @inode: unhashed inode
943 * @hashval: unsigned long value used to locate this object in the
944 * inode_hashtable.
945 *
946 * Add an inode to the inode hash for this superblock.
947 */
948void __insert_inode_hash(struct inode *inode, unsigned long hashval)
949{
950 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
951 spin_lock(&inode_lock);
952 hlist_add_head(&inode->i_hash, head);
953 spin_unlock(&inode_lock);
954}
955
956EXPORT_SYMBOL(__insert_inode_hash);
957
958/**
959 * remove_inode_hash - remove an inode from the hash
960 * @inode: inode to unhash
961 *
962 * Remove an inode from the superblock.
963 */
964void remove_inode_hash(struct inode *inode)
965{
966 spin_lock(&inode_lock);
967 hlist_del_init(&inode->i_hash);
968 spin_unlock(&inode_lock);
969}
970
971EXPORT_SYMBOL(remove_inode_hash);
972
973/*
974 * Tell the filesystem that this inode is no longer of any interest and should
975 * be completely destroyed.
976 *
977 * We leave the inode in the inode hash table until *after* the filesystem's
978 * ->delete_inode completes. This ensures that an iget (such as nfsd might
979 * instigate) will always find up-to-date information either in the hash or on
980 * disk.
981 *
982 * I_FREEING is set so that no-one will take a new reference to the inode while
983 * it is being deleted.
984 */
985void generic_delete_inode(struct inode *inode)
986{
987 struct super_operations *op = inode->i_sb->s_op;
988
989 list_del_init(&inode->i_list);
990 list_del_init(&inode->i_sb_list);
991 inode->i_state|=I_FREEING;
992 inodes_stat.nr_inodes--;
993 spin_unlock(&inode_lock);
994
995 if (inode->i_data.nrpages)
996 truncate_inode_pages(&inode->i_data, 0);
997
998 security_inode_delete(inode);
999
1000 if (op->delete_inode) {
1001 void (*delete)(struct inode *) = op->delete_inode;
1002 if (!is_bad_inode(inode))
1003 DQUOT_INIT(inode);
1004 /* s_op->delete_inode internally recalls clear_inode() */
1005 delete(inode);
1006 } else
1007 clear_inode(inode);
1008 spin_lock(&inode_lock);
1009 hlist_del_init(&inode->i_hash);
1010 spin_unlock(&inode_lock);
1011 wake_up_inode(inode);
1012 if (inode->i_state != I_CLEAR)
1013 BUG();
1014 destroy_inode(inode);
1015}
1016
1017EXPORT_SYMBOL(generic_delete_inode);
1018
1019static void generic_forget_inode(struct inode *inode)
1020{
1021 struct super_block *sb = inode->i_sb;
1022
1023 if (!hlist_unhashed(&inode->i_hash)) {
1024 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1025 list_move(&inode->i_list, &inode_unused);
1026 inodes_stat.nr_unused++;
Alexander Viro991114c2005-06-23 00:09:01 -07001027 if (!sb || (sb->s_flags & MS_ACTIVE)) {
1028 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001030 }
1031 inode->i_state |= I_WILL_FREE;
1032 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 write_inode_now(inode, 1);
1034 spin_lock(&inode_lock);
Alexander Viro991114c2005-06-23 00:09:01 -07001035 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 inodes_stat.nr_unused--;
1037 hlist_del_init(&inode->i_hash);
1038 }
1039 list_del_init(&inode->i_list);
1040 list_del_init(&inode->i_sb_list);
Alexander Viro991114c2005-06-23 00:09:01 -07001041 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 inodes_stat.nr_inodes--;
1043 spin_unlock(&inode_lock);
1044 if (inode->i_data.nrpages)
1045 truncate_inode_pages(&inode->i_data, 0);
1046 clear_inode(inode);
1047 destroy_inode(inode);
1048}
1049
1050/*
1051 * Normal UNIX filesystem behaviour: delete the
1052 * inode when the usage count drops to zero, and
1053 * i_nlink is zero.
1054 */
1055static void generic_drop_inode(struct inode *inode)
1056{
1057 if (!inode->i_nlink)
1058 generic_delete_inode(inode);
1059 else
1060 generic_forget_inode(inode);
1061}
1062
1063/*
1064 * Called when we're dropping the last reference
1065 * to an inode.
1066 *
1067 * Call the FS "drop()" function, defaulting to
1068 * the legacy UNIX filesystem behaviour..
1069 *
1070 * NOTE! NOTE! NOTE! We're called with the inode lock
1071 * held, and the drop function is supposed to release
1072 * the lock!
1073 */
1074static inline void iput_final(struct inode *inode)
1075{
1076 struct super_operations *op = inode->i_sb->s_op;
1077 void (*drop)(struct inode *) = generic_drop_inode;
1078
1079 if (op && op->drop_inode)
1080 drop = op->drop_inode;
1081 drop(inode);
1082}
1083
1084/**
1085 * iput - put an inode
1086 * @inode: inode to put
1087 *
1088 * Puts an inode, dropping its usage count. If the inode use count hits
1089 * zero, the inode is then freed and may also be destroyed.
1090 *
1091 * Consequently, iput() can sleep.
1092 */
1093void iput(struct inode *inode)
1094{
1095 if (inode) {
1096 struct super_operations *op = inode->i_sb->s_op;
1097
1098 BUG_ON(inode->i_state == I_CLEAR);
1099
1100 if (op && op->put_inode)
1101 op->put_inode(inode);
1102
1103 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1104 iput_final(inode);
1105 }
1106}
1107
1108EXPORT_SYMBOL(iput);
1109
1110/**
1111 * bmap - find a block number in a file
1112 * @inode: inode of file
1113 * @block: block to find
1114 *
1115 * Returns the block number on the device holding the inode that
1116 * is the disk block number for the block of the file requested.
1117 * That is, asked for block 4 of inode 1 the function will return the
1118 * disk block relative to the disk start that holds that block of the
1119 * file.
1120 */
1121sector_t bmap(struct inode * inode, sector_t block)
1122{
1123 sector_t res = 0;
1124 if (inode->i_mapping->a_ops->bmap)
1125 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1126 return res;
1127}
1128
1129EXPORT_SYMBOL(bmap);
1130
1131/**
1132 * update_atime - update the access time
1133 * @inode: inode accessed
1134 *
1135 * Update the accessed time on an inode and mark it for writeback.
1136 * This function automatically handles read only file systems and media,
1137 * as well as the "noatime" flag and inode specific "noatime" markers.
1138 */
1139void update_atime(struct inode *inode)
1140{
1141 struct timespec now;
1142
1143 if (IS_NOATIME(inode))
1144 return;
1145 if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1146 return;
1147 if (IS_RDONLY(inode))
1148 return;
1149
1150 now = current_fs_time(inode->i_sb);
1151 if (!timespec_equal(&inode->i_atime, &now)) {
1152 inode->i_atime = now;
1153 mark_inode_dirty_sync(inode);
1154 } else {
1155 if (!timespec_equal(&inode->i_atime, &now))
1156 inode->i_atime = now;
1157 }
1158}
1159
1160EXPORT_SYMBOL(update_atime);
1161
1162/**
1163 * inode_update_time - update mtime and ctime time
1164 * @inode: inode accessed
1165 * @ctime_too: update ctime too
1166 *
1167 * Update the mtime time on an inode and mark it for writeback.
1168 * When ctime_too is specified update the ctime too.
1169 */
1170
1171void inode_update_time(struct inode *inode, int ctime_too)
1172{
1173 struct timespec now;
1174 int sync_it = 0;
1175
1176 if (IS_NOCMTIME(inode))
1177 return;
1178 if (IS_RDONLY(inode))
1179 return;
1180
1181 now = current_fs_time(inode->i_sb);
1182 if (!timespec_equal(&inode->i_mtime, &now))
1183 sync_it = 1;
1184 inode->i_mtime = now;
1185
1186 if (ctime_too) {
1187 if (!timespec_equal(&inode->i_ctime, &now))
1188 sync_it = 1;
1189 inode->i_ctime = now;
1190 }
1191 if (sync_it)
1192 mark_inode_dirty_sync(inode);
1193}
1194
1195EXPORT_SYMBOL(inode_update_time);
1196
1197int inode_needs_sync(struct inode *inode)
1198{
1199 if (IS_SYNC(inode))
1200 return 1;
1201 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1202 return 1;
1203 return 0;
1204}
1205
1206EXPORT_SYMBOL(inode_needs_sync);
1207
1208/*
1209 * Quota functions that want to walk the inode lists..
1210 */
1211#ifdef CONFIG_QUOTA
1212
1213/* Function back in dquot.c */
1214int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1215
1216void remove_dquot_ref(struct super_block *sb, int type,
1217 struct list_head *tofree_head)
1218{
1219 struct inode *inode;
1220
1221 if (!sb->dq_op)
1222 return; /* nothing to do */
1223 spin_lock(&inode_lock); /* This lock is for inodes code */
1224
1225 /*
1226 * We don't have to lock against quota code - test IS_QUOTAINIT is
1227 * just for speedup...
1228 */
1229 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1230 if (!IS_NOQUOTA(inode))
1231 remove_inode_dquot_ref(inode, type, tofree_head);
1232
1233 spin_unlock(&inode_lock);
1234}
1235
1236#endif
1237
1238int inode_wait(void *word)
1239{
1240 schedule();
1241 return 0;
1242}
1243
1244/*
1245 * If we try to find an inode in the inode hash while it is being deleted, we
1246 * have to wait until the filesystem completes its deletion before reporting
1247 * that it isn't found. This is because iget will immediately call
1248 * ->read_inode, and we want to be sure that evidence of the deletion is found
1249 * by ->read_inode.
1250 * This is called with inode_lock held.
1251 */
1252static void __wait_on_freeing_inode(struct inode *inode)
1253{
1254 wait_queue_head_t *wq;
1255 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1256
1257 /*
1258 * I_FREEING and I_CLEAR are cleared in process context under
1259 * inode_lock, so we have to give the tasks who would clear them
1260 * a chance to run and acquire inode_lock.
1261 */
1262 if (!(inode->i_state & I_LOCK)) {
1263 spin_unlock(&inode_lock);
1264 yield();
1265 spin_lock(&inode_lock);
1266 return;
1267 }
1268 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1269 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1270 spin_unlock(&inode_lock);
1271 schedule();
1272 finish_wait(wq, &wait.wait);
1273 spin_lock(&inode_lock);
1274}
1275
1276void wake_up_inode(struct inode *inode)
1277{
1278 /*
1279 * Prevent speculative execution through spin_unlock(&inode_lock);
1280 */
1281 smp_mb();
1282 wake_up_bit(&inode->i_state, __I_LOCK);
1283}
1284
1285static __initdata unsigned long ihash_entries;
1286static int __init set_ihash_entries(char *str)
1287{
1288 if (!str)
1289 return 0;
1290 ihash_entries = simple_strtoul(str, &str, 0);
1291 return 1;
1292}
1293__setup("ihash_entries=", set_ihash_entries);
1294
1295/*
1296 * Initialize the waitqueues and inode hash table.
1297 */
1298void __init inode_init_early(void)
1299{
1300 int loop;
1301
1302 /* If hashes are distributed across NUMA nodes, defer
1303 * hash allocation until vmalloc space is available.
1304 */
1305 if (hashdist)
1306 return;
1307
1308 inode_hashtable =
1309 alloc_large_system_hash("Inode-cache",
1310 sizeof(struct hlist_head),
1311 ihash_entries,
1312 14,
1313 HASH_EARLY,
1314 &i_hash_shift,
1315 &i_hash_mask,
1316 0);
1317
1318 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1319 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1320}
1321
1322void __init inode_init(unsigned long mempages)
1323{
1324 int loop;
1325
1326 /* inode slab cache */
1327 inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
Andrea Arcangelie422fd22005-05-05 16:15:04 -07001328 0, SLAB_RECLAIM_ACCOUNT|SLAB_PANIC, init_once, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1330
1331 /* Hash may have been set up in inode_init_early */
1332 if (!hashdist)
1333 return;
1334
1335 inode_hashtable =
1336 alloc_large_system_hash("Inode-cache",
1337 sizeof(struct hlist_head),
1338 ihash_entries,
1339 14,
1340 0,
1341 &i_hash_shift,
1342 &i_hash_mask,
1343 0);
1344
1345 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1346 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1347}
1348
1349void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1350{
1351 inode->i_mode = mode;
1352 if (S_ISCHR(mode)) {
1353 inode->i_fop = &def_chr_fops;
1354 inode->i_rdev = rdev;
1355 } else if (S_ISBLK(mode)) {
1356 inode->i_fop = &def_blk_fops;
1357 inode->i_rdev = rdev;
1358 } else if (S_ISFIFO(mode))
1359 inode->i_fop = &def_fifo_fops;
1360 else if (S_ISSOCK(mode))
1361 inode->i_fop = &bad_sock_fops;
1362 else
1363 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1364 mode);
1365}
1366EXPORT_SYMBOL(init_special_inode);