blob: 3a2446a27d2c29c9bd6172f8f4e17b9871453d3e [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>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/inotify.h>
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -080025#include <linux/mount.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
27/*
28 * This is needed for the following functions:
29 * - inode_has_buffers
30 * - invalidate_inode_buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 * - invalidate_bdev
32 *
33 * FIXME: remove all knowledge of the buffer layer from this file
34 */
35#include <linux/buffer_head.h>
36
37/*
38 * New inode.c implementation.
39 *
40 * This implementation has the basic premise of trying
41 * to be extremely low-overhead and SMP-safe, yet be
42 * simple enough to be "obviously correct".
43 *
44 * Famous last words.
45 */
46
47/* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
48
49/* #define INODE_PARANOIA 1 */
50/* #define INODE_DEBUG 1 */
51
52/*
53 * Inode lookup is no longer as critical as it used to be:
54 * most of the lookups are going to be through the dcache.
55 */
56#define I_HASHBITS i_hash_shift
57#define I_HASHMASK i_hash_mask
58
Eric Dumazetfa3536c2006-03-26 01:37:24 -080059static unsigned int i_hash_mask __read_mostly;
60static unsigned int i_hash_shift __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
62/*
63 * Each inode can be on two separate lists. One is
64 * the hash list of the inode, used for lookups. The
65 * other linked list is the "type" list:
66 * "in_use" - valid inode, i_count > 0, i_nlink > 0
67 * "dirty" - as "in_use" but also dirty
68 * "unused" - valid inode, i_count = 0
69 *
70 * A "dirty" list is maintained for each super block,
71 * allowing for low-overhead inode sync() operations.
72 */
73
74LIST_HEAD(inode_in_use);
75LIST_HEAD(inode_unused);
Eric Dumazetfa3536c2006-03-26 01:37:24 -080076static struct hlist_head *inode_hashtable __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
78/*
79 * A simple spinlock to protect the list manipulations.
80 *
81 * NOTE! You also have to own the lock if you change
82 * the i_state of an inode while it is in use..
83 */
84DEFINE_SPINLOCK(inode_lock);
85
86/*
Ingo Molnarf24075b2006-03-23 03:00:34 -080087 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 * icache shrinking path, and the umount path. Without this exclusion,
89 * by the time prune_icache calls iput for the inode whose pages it has
90 * been invalidating, or by the time it calls clear_inode & destroy_inode
91 * from its final dispose_list, the struct super_block they refer to
92 * (for inode->i_sb->s_op) may already have been freed and reused.
93 */
Adrian Bunkbdfc3262006-03-25 03:06:56 -080094static DEFINE_MUTEX(iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
96/*
97 * Statistics gathering..
98 */
99struct inodes_stat_t inodes_stat;
100
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800101static kmem_cache_t * inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
103static struct inode *alloc_inode(struct super_block *sb)
104{
105 static struct address_space_operations empty_aops;
106 static struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800107 static const struct file_operations empty_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 struct inode *inode;
109
110 if (sb->s_op->alloc_inode)
111 inode = sb->s_op->alloc_inode(sb);
112 else
113 inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
114
115 if (inode) {
116 struct address_space * const mapping = &inode->i_data;
117
118 inode->i_sb = sb;
119 inode->i_blkbits = sb->s_blocksize_bits;
120 inode->i_flags = 0;
121 atomic_set(&inode->i_count, 1);
122 inode->i_op = &empty_iops;
123 inode->i_fop = &empty_fops;
124 inode->i_nlink = 1;
125 atomic_set(&inode->i_writecount, 0);
126 inode->i_size = 0;
127 inode->i_blocks = 0;
128 inode->i_bytes = 0;
129 inode->i_generation = 0;
130#ifdef CONFIG_QUOTA
131 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
132#endif
133 inode->i_pipe = NULL;
134 inode->i_bdev = NULL;
135 inode->i_cdev = NULL;
136 inode->i_rdev = 0;
137 inode->i_security = NULL;
138 inode->dirtied_when = 0;
139 if (security_inode_alloc(inode)) {
140 if (inode->i_sb->s_op->destroy_inode)
141 inode->i_sb->s_op->destroy_inode(inode);
142 else
143 kmem_cache_free(inode_cachep, (inode));
144 return NULL;
145 }
146
147 mapping->a_ops = &empty_aops;
148 mapping->host = inode;
149 mapping->flags = 0;
150 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
151 mapping->assoc_mapping = NULL;
152 mapping->backing_dev_info = &default_backing_dev_info;
153
154 /*
155 * If the block_device provides a backing_dev_info for client
156 * inodes then use that. Otherwise the inode share the bdev's
157 * backing_dev_info.
158 */
159 if (sb->s_bdev) {
160 struct backing_dev_info *bdi;
161
162 bdi = sb->s_bdev->bd_inode_backing_dev_info;
163 if (!bdi)
164 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
165 mapping->backing_dev_info = bdi;
166 }
167 memset(&inode->u, 0, sizeof(inode->u));
168 inode->i_mapping = mapping;
169 }
170 return inode;
171}
172
173void destroy_inode(struct inode *inode)
174{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200175 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 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);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800195 mutex_init(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 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);
Robert Love0eeca282005-07-12 17:06:03 -0400206#ifdef CONFIG_INOTIFY
207 INIT_LIST_HEAD(&inode->inotify_watches);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800208 mutex_init(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400209#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210}
211
212EXPORT_SYMBOL(inode_init_once);
213
214static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
215{
216 struct inode * inode = (struct inode *) foo;
217
218 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
219 SLAB_CTOR_CONSTRUCTOR)
220 inode_init_once(inode);
221}
222
223/*
224 * inode_lock must be held
225 */
226void __iget(struct inode * inode)
227{
228 if (atomic_read(&inode->i_count)) {
229 atomic_inc(&inode->i_count);
230 return;
231 }
232 atomic_inc(&inode->i_count);
233 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
234 list_move(&inode->i_list, &inode_in_use);
235 inodes_stat.nr_unused--;
236}
237
238/**
239 * clear_inode - clear an inode
240 * @inode: inode to clear
241 *
242 * This is called by the filesystem to tell us
243 * that the inode is no longer useful. We just
244 * terminate it with extreme prejudice.
245 */
246void clear_inode(struct inode *inode)
247{
248 might_sleep();
249 invalidate_inode_buffers(inode);
250
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200251 BUG_ON(inode->i_data.nrpages);
252 BUG_ON(!(inode->i_state & I_FREEING));
253 BUG_ON(inode->i_state & I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 wait_on_inode(inode);
255 DQUOT_DROP(inode);
256 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
257 inode->i_sb->s_op->clear_inode(inode);
258 if (inode->i_bdev)
259 bd_forget(inode);
260 if (inode->i_cdev)
261 cd_forget(inode);
262 inode->i_state = I_CLEAR;
263}
264
265EXPORT_SYMBOL(clear_inode);
266
267/*
268 * dispose_list - dispose of the contents of a local list
269 * @head: the head of the list to free
270 *
271 * Dispose-list gets a local list with local inodes in it, so it doesn't
272 * need to worry about list corruption and SMP locks.
273 */
274static void dispose_list(struct list_head *head)
275{
276 int nr_disposed = 0;
277
278 while (!list_empty(head)) {
279 struct inode *inode;
280
281 inode = list_entry(head->next, struct inode, i_list);
282 list_del(&inode->i_list);
283
284 if (inode->i_data.nrpages)
285 truncate_inode_pages(&inode->i_data, 0);
286 clear_inode(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700287
288 spin_lock(&inode_lock);
289 hlist_del_init(&inode->i_hash);
290 list_del_init(&inode->i_sb_list);
291 spin_unlock(&inode_lock);
292
293 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 destroy_inode(inode);
295 nr_disposed++;
296 }
297 spin_lock(&inode_lock);
298 inodes_stat.nr_inodes -= nr_disposed;
299 spin_unlock(&inode_lock);
300}
301
302/*
303 * Invalidate all inodes for a device.
304 */
305static int invalidate_list(struct list_head *head, struct list_head *dispose)
306{
307 struct list_head *next;
308 int busy = 0, count = 0;
309
310 next = head->next;
311 for (;;) {
312 struct list_head * tmp = next;
313 struct inode * inode;
314
315 /*
316 * We can reschedule here without worrying about the list's
317 * consistency because the per-sb list of inodes must not
Ingo Molnarf24075b2006-03-23 03:00:34 -0800318 * change during umount anymore, and because iprune_mutex keeps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 * shrink_icache_memory() away.
320 */
321 cond_resched_lock(&inode_lock);
322
323 next = next->next;
324 if (tmp == head)
325 break;
326 inode = list_entry(tmp, struct inode, i_sb_list);
327 invalidate_inode_buffers(inode);
328 if (!atomic_read(&inode->i_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 list_move(&inode->i_list, dispose);
330 inode->i_state |= I_FREEING;
331 count++;
332 continue;
333 }
334 busy = 1;
335 }
336 /* only unused inodes may be cached with i_count zero */
337 inodes_stat.nr_unused -= count;
338 return busy;
339}
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341/**
342 * invalidate_inodes - discard the inodes on a device
343 * @sb: superblock
344 *
345 * Discard all of the inodes for a given superblock. If the discard
346 * fails because there are busy inodes then a non zero value is returned.
347 * If the discard is successful all the inodes have been discarded.
348 */
349int invalidate_inodes(struct super_block * sb)
350{
351 int busy;
352 LIST_HEAD(throw_away);
353
Ingo Molnarf24075b2006-03-23 03:00:34 -0800354 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 spin_lock(&inode_lock);
Robert Love0eeca282005-07-12 17:06:03 -0400356 inotify_unmount_inodes(&sb->s_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 busy = invalidate_list(&sb->s_inodes, &throw_away);
358 spin_unlock(&inode_lock);
359
360 dispose_list(&throw_away);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800361 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
363 return busy;
364}
365
366EXPORT_SYMBOL(invalidate_inodes);
367
Christoph Hellwig2ef41632005-05-05 16:15:59 -0700368int __invalidate_device(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369{
Christoph Hellwig2ef41632005-05-05 16:15:59 -0700370 struct super_block *sb = get_super(bdev);
371 int res = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 if (sb) {
374 /*
375 * no need to lock the super, get_super holds the
Ingo Molnarf24075b2006-03-23 03:00:34 -0800376 * read mutex so the filesystem cannot go away
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 * under us (->put_super runs with the write lock
378 * hold).
379 */
380 shrink_dcache_sb(sb);
381 res = invalidate_inodes(sb);
382 drop_super(sb);
383 }
384 invalidate_bdev(bdev, 0);
385 return res;
386}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387EXPORT_SYMBOL(__invalidate_device);
388
389static int can_unuse(struct inode *inode)
390{
391 if (inode->i_state)
392 return 0;
393 if (inode_has_buffers(inode))
394 return 0;
395 if (atomic_read(&inode->i_count))
396 return 0;
397 if (inode->i_data.nrpages)
398 return 0;
399 return 1;
400}
401
402/*
403 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
404 * a temporary list and then are freed outside inode_lock by dispose_list().
405 *
406 * Any inodes which are pinned purely because of attached pagecache have their
407 * pagecache removed. We expect the final iput() on that inode to add it to
408 * the front of the inode_unused list. So look for it there and if the
409 * inode is still freeable, proceed. The right inode is found 99.9% of the
410 * time in testing on a 4-way.
411 *
412 * If the inode has metadata buffers attached to mapping->private_list then
413 * try to remove them.
414 */
415static void prune_icache(int nr_to_scan)
416{
417 LIST_HEAD(freeable);
418 int nr_pruned = 0;
419 int nr_scanned;
420 unsigned long reap = 0;
421
Ingo Molnarf24075b2006-03-23 03:00:34 -0800422 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 spin_lock(&inode_lock);
424 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
425 struct inode *inode;
426
427 if (list_empty(&inode_unused))
428 break;
429
430 inode = list_entry(inode_unused.prev, struct inode, i_list);
431
432 if (inode->i_state || atomic_read(&inode->i_count)) {
433 list_move(&inode->i_list, &inode_unused);
434 continue;
435 }
436 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
437 __iget(inode);
438 spin_unlock(&inode_lock);
439 if (remove_inode_buffers(inode))
440 reap += invalidate_inode_pages(&inode->i_data);
441 iput(inode);
442 spin_lock(&inode_lock);
443
444 if (inode != list_entry(inode_unused.next,
445 struct inode, i_list))
446 continue; /* wrong inode or list_empty */
447 if (!can_unuse(inode))
448 continue;
449 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 list_move(&inode->i_list, &freeable);
451 inode->i_state |= I_FREEING;
452 nr_pruned++;
453 }
454 inodes_stat.nr_unused -= nr_pruned;
455 spin_unlock(&inode_lock);
456
457 dispose_list(&freeable);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800458 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
460 if (current_is_kswapd())
461 mod_page_state(kswapd_inodesteal, reap);
462 else
463 mod_page_state(pginodesteal, reap);
464}
465
466/*
467 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
468 * "unused" means that no dentries are referring to the inodes: the files are
469 * not open and the dcache references to those inodes have already been
470 * reclaimed.
471 *
472 * This function is passed the number of inodes to scan, and it returns the
473 * total number of remaining possibly-reclaimable inodes.
474 */
Al Viro27496a82005-10-21 03:20:48 -0400475static int shrink_icache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
477 if (nr) {
478 /*
479 * Nasty deadlock avoidance. We may hold various FS locks,
480 * and we don't want to recurse into the FS that called us
481 * in clear_inode() and friends..
482 */
483 if (!(gfp_mask & __GFP_FS))
484 return -1;
485 prune_icache(nr);
486 }
487 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
488}
489
490static void __wait_on_freeing_inode(struct inode *inode);
491/*
492 * Called with the inode lock held.
493 * NOTE: we are not increasing the inode-refcount, you must call __iget()
494 * by hand after calling find_inode now! This simplifies iunique and won't
495 * add any additional branch in the common code.
496 */
497static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
498{
499 struct hlist_node *node;
500 struct inode * inode = NULL;
501
502repeat:
503 hlist_for_each (node, head) {
504 inode = hlist_entry(node, struct inode, i_hash);
505 if (inode->i_sb != sb)
506 continue;
507 if (!test(inode, data))
508 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700509 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 __wait_on_freeing_inode(inode);
511 goto repeat;
512 }
513 break;
514 }
515 return node ? inode : NULL;
516}
517
518/*
519 * find_inode_fast is the fast path version of find_inode, see the comment at
520 * iget_locked for details.
521 */
522static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
523{
524 struct hlist_node *node;
525 struct inode * inode = NULL;
526
527repeat:
528 hlist_for_each (node, head) {
529 inode = hlist_entry(node, struct inode, i_hash);
530 if (inode->i_ino != ino)
531 continue;
532 if (inode->i_sb != sb)
533 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700534 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 __wait_on_freeing_inode(inode);
536 goto repeat;
537 }
538 break;
539 }
540 return node ? inode : NULL;
541}
542
543/**
544 * new_inode - obtain an inode
545 * @sb: superblock
546 *
547 * Allocates a new inode for given superblock.
548 */
549struct inode *new_inode(struct super_block *sb)
550{
551 static unsigned long last_ino;
552 struct inode * inode;
553
554 spin_lock_prefetch(&inode_lock);
555
556 inode = alloc_inode(sb);
557 if (inode) {
558 spin_lock(&inode_lock);
559 inodes_stat.nr_inodes++;
560 list_add(&inode->i_list, &inode_in_use);
561 list_add(&inode->i_sb_list, &sb->s_inodes);
562 inode->i_ino = ++last_ino;
563 inode->i_state = 0;
564 spin_unlock(&inode_lock);
565 }
566 return inode;
567}
568
569EXPORT_SYMBOL(new_inode);
570
571void unlock_new_inode(struct inode *inode)
572{
573 /*
574 * This is special! We do not need the spinlock
575 * when clearing I_LOCK, because we're guaranteed
576 * that nobody else tries to do anything about the
577 * state of the inode when it is locked, as we
578 * just created it (so there can be no old holders
579 * that haven't tested I_LOCK).
580 */
581 inode->i_state &= ~(I_LOCK|I_NEW);
582 wake_up_inode(inode);
583}
584
585EXPORT_SYMBOL(unlock_new_inode);
586
587/*
588 * This is called without the inode lock held.. Be careful.
589 *
590 * We no longer cache the sb_flags in i_flags - see fs.h
591 * -- rmk@arm.uk.linux.org
592 */
593static 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)
594{
595 struct inode * inode;
596
597 inode = alloc_inode(sb);
598 if (inode) {
599 struct inode * old;
600
601 spin_lock(&inode_lock);
602 /* We released the lock, so.. */
603 old = find_inode(sb, head, test, data);
604 if (!old) {
605 if (set(inode, data))
606 goto set_failed;
607
608 inodes_stat.nr_inodes++;
609 list_add(&inode->i_list, &inode_in_use);
610 list_add(&inode->i_sb_list, &sb->s_inodes);
611 hlist_add_head(&inode->i_hash, head);
612 inode->i_state = I_LOCK|I_NEW;
613 spin_unlock(&inode_lock);
614
615 /* Return the locked inode with I_NEW set, the
616 * caller is responsible for filling in the contents
617 */
618 return inode;
619 }
620
621 /*
622 * Uhhuh, somebody else created the same inode under
623 * us. Use the old inode instead of the one we just
624 * allocated.
625 */
626 __iget(old);
627 spin_unlock(&inode_lock);
628 destroy_inode(inode);
629 inode = old;
630 wait_on_inode(inode);
631 }
632 return inode;
633
634set_failed:
635 spin_unlock(&inode_lock);
636 destroy_inode(inode);
637 return NULL;
638}
639
640/*
641 * get_new_inode_fast is the fast path version of get_new_inode, see the
642 * comment at iget_locked for details.
643 */
644static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
645{
646 struct inode * inode;
647
648 inode = alloc_inode(sb);
649 if (inode) {
650 struct inode * old;
651
652 spin_lock(&inode_lock);
653 /* We released the lock, so.. */
654 old = find_inode_fast(sb, head, ino);
655 if (!old) {
656 inode->i_ino = ino;
657 inodes_stat.nr_inodes++;
658 list_add(&inode->i_list, &inode_in_use);
659 list_add(&inode->i_sb_list, &sb->s_inodes);
660 hlist_add_head(&inode->i_hash, head);
661 inode->i_state = I_LOCK|I_NEW;
662 spin_unlock(&inode_lock);
663
664 /* Return the locked inode with I_NEW set, the
665 * caller is responsible for filling in the contents
666 */
667 return inode;
668 }
669
670 /*
671 * Uhhuh, somebody else created the same inode under
672 * us. Use the old inode instead of the one we just
673 * allocated.
674 */
675 __iget(old);
676 spin_unlock(&inode_lock);
677 destroy_inode(inode);
678 inode = old;
679 wait_on_inode(inode);
680 }
681 return inode;
682}
683
684static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
685{
686 unsigned long tmp;
687
688 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
689 L1_CACHE_BYTES;
690 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
691 return tmp & I_HASHMASK;
692}
693
694/**
695 * iunique - get a unique inode number
696 * @sb: superblock
697 * @max_reserved: highest reserved inode number
698 *
699 * Obtain an inode number that is unique on the system for a given
700 * superblock. This is used by file systems that have no natural
701 * permanent inode numbering system. An inode number is returned that
702 * is higher than the reserved limit but unique.
703 *
704 * BUGS:
705 * With a large number of inodes live on the file system this function
706 * currently becomes quite slow.
707 */
708ino_t iunique(struct super_block *sb, ino_t max_reserved)
709{
710 static ino_t counter;
711 struct inode *inode;
712 struct hlist_head * head;
713 ino_t res;
714 spin_lock(&inode_lock);
715retry:
716 if (counter > max_reserved) {
717 head = inode_hashtable + hash(sb,counter);
718 res = counter++;
719 inode = find_inode_fast(sb, head, res);
720 if (!inode) {
721 spin_unlock(&inode_lock);
722 return res;
723 }
724 } else {
725 counter = max_reserved + 1;
726 }
727 goto retry;
728
729}
730
731EXPORT_SYMBOL(iunique);
732
733struct inode *igrab(struct inode *inode)
734{
735 spin_lock(&inode_lock);
Alexander Viro991114c2005-06-23 00:09:01 -0700736 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 __iget(inode);
738 else
739 /*
740 * Handle the case where s_op->clear_inode is not been
741 * called yet, and somebody is calling igrab
742 * while the inode is getting freed.
743 */
744 inode = NULL;
745 spin_unlock(&inode_lock);
746 return inode;
747}
748
749EXPORT_SYMBOL(igrab);
750
751/**
752 * ifind - internal function, you want ilookup5() or iget5().
753 * @sb: super block of file system to search
754 * @head: the head of the list to search
755 * @test: callback used for comparisons between inodes
756 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700757 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 *
759 * ifind() searches for the inode specified by @data in the inode
760 * cache. This is a generalized version of ifind_fast() for file systems where
761 * the inode number is not sufficient for unique identification of an inode.
762 *
763 * If the inode is in the cache, the inode is returned with an incremented
764 * reference count.
765 *
766 * Otherwise NULL is returned.
767 *
768 * Note, @test is called with the inode_lock held, so can't sleep.
769 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800770static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700772 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773{
774 struct inode *inode;
775
776 spin_lock(&inode_lock);
777 inode = find_inode(sb, head, test, data);
778 if (inode) {
779 __iget(inode);
780 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700781 if (likely(wait))
782 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 return inode;
784 }
785 spin_unlock(&inode_lock);
786 return NULL;
787}
788
789/**
790 * ifind_fast - internal function, you want ilookup() or iget().
791 * @sb: super block of file system to search
792 * @head: head of the list to search
793 * @ino: inode number to search for
794 *
795 * ifind_fast() searches for the inode @ino in the inode cache. This is for
796 * file systems where the inode number is sufficient for unique identification
797 * of an inode.
798 *
799 * If the inode is in the cache, the inode is returned with an incremented
800 * reference count.
801 *
802 * Otherwise NULL is returned.
803 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800804static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 struct hlist_head *head, unsigned long ino)
806{
807 struct inode *inode;
808
809 spin_lock(&inode_lock);
810 inode = find_inode_fast(sb, head, ino);
811 if (inode) {
812 __iget(inode);
813 spin_unlock(&inode_lock);
814 wait_on_inode(inode);
815 return inode;
816 }
817 spin_unlock(&inode_lock);
818 return NULL;
819}
820
821/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700822 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 * @sb: super block of file system to search
824 * @hashval: hash value (usually inode number) to search for
825 * @test: callback used for comparisons between inodes
826 * @data: opaque data pointer to pass to @test
827 *
828 * ilookup5() uses ifind() to search for the inode specified by @hashval and
829 * @data in the inode cache. This is a generalized version of ilookup() for
830 * file systems where the inode number is not sufficient for unique
831 * identification of an inode.
832 *
833 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700834 * reference count. Note, the inode lock is not waited upon so you have to be
835 * very careful what you do with the returned inode. You probably should be
836 * using ilookup5() instead.
837 *
838 * Otherwise NULL is returned.
839 *
840 * Note, @test is called with the inode_lock held, so can't sleep.
841 */
842struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
843 int (*test)(struct inode *, void *), void *data)
844{
845 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
846
847 return ifind(sb, head, test, data, 0);
848}
849
850EXPORT_SYMBOL(ilookup5_nowait);
851
852/**
853 * ilookup5 - search for an inode in the inode cache
854 * @sb: super block of file system to search
855 * @hashval: hash value (usually inode number) to search for
856 * @test: callback used for comparisons between inodes
857 * @data: opaque data pointer to pass to @test
858 *
859 * ilookup5() uses ifind() to search for the inode specified by @hashval and
860 * @data in the inode cache. This is a generalized version of ilookup() for
861 * file systems where the inode number is not sufficient for unique
862 * identification of an inode.
863 *
864 * If the inode is in the cache, the inode lock is waited upon and the inode is
865 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 *
867 * Otherwise NULL is returned.
868 *
869 * Note, @test is called with the inode_lock held, so can't sleep.
870 */
871struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
872 int (*test)(struct inode *, void *), void *data)
873{
874 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
875
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700876 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877}
878
879EXPORT_SYMBOL(ilookup5);
880
881/**
882 * ilookup - search for an inode in the inode cache
883 * @sb: super block of file system to search
884 * @ino: inode number to search for
885 *
886 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
887 * This is for file systems where the inode number is sufficient for unique
888 * identification of an inode.
889 *
890 * If the inode is in the cache, the inode is returned with an incremented
891 * reference count.
892 *
893 * Otherwise NULL is returned.
894 */
895struct inode *ilookup(struct super_block *sb, unsigned long ino)
896{
897 struct hlist_head *head = inode_hashtable + hash(sb, ino);
898
899 return ifind_fast(sb, head, ino);
900}
901
902EXPORT_SYMBOL(ilookup);
903
904/**
905 * iget5_locked - obtain an inode from a mounted file system
906 * @sb: super block of file system
907 * @hashval: hash value (usually inode number) to get
908 * @test: callback used for comparisons between inodes
909 * @set: callback used to initialize a new struct inode
910 * @data: opaque data pointer to pass to @test and @set
911 *
912 * This is iget() without the read_inode() portion of get_new_inode().
913 *
914 * iget5_locked() uses ifind() to search for the inode specified by @hashval
915 * and @data in the inode cache and if present it is returned with an increased
916 * reference count. This is a generalized version of iget_locked() for file
917 * systems where the inode number is not sufficient for unique identification
918 * of an inode.
919 *
920 * If the inode is not in cache, get_new_inode() is called to allocate a new
921 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
922 * file system gets to fill it in before unlocking it via unlock_new_inode().
923 *
924 * Note both @test and @set are called with the inode_lock held, so can't sleep.
925 */
926struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
927 int (*test)(struct inode *, void *),
928 int (*set)(struct inode *, void *), void *data)
929{
930 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
931 struct inode *inode;
932
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700933 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 if (inode)
935 return inode;
936 /*
937 * get_new_inode() will do the right thing, re-trying the search
938 * in case it had to block at any point.
939 */
940 return get_new_inode(sb, head, test, set, data);
941}
942
943EXPORT_SYMBOL(iget5_locked);
944
945/**
946 * iget_locked - obtain an inode from a mounted file system
947 * @sb: super block of file system
948 * @ino: inode number to get
949 *
950 * This is iget() without the read_inode() portion of get_new_inode_fast().
951 *
952 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
953 * the inode cache and if present it is returned with an increased reference
954 * count. This is for file systems where the inode number is sufficient for
955 * unique identification of an inode.
956 *
957 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
958 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
959 * The file system gets to fill it in before unlocking it via
960 * unlock_new_inode().
961 */
962struct inode *iget_locked(struct super_block *sb, unsigned long ino)
963{
964 struct hlist_head *head = inode_hashtable + hash(sb, ino);
965 struct inode *inode;
966
967 inode = ifind_fast(sb, head, ino);
968 if (inode)
969 return inode;
970 /*
971 * get_new_inode_fast() will do the right thing, re-trying the search
972 * in case it had to block at any point.
973 */
974 return get_new_inode_fast(sb, head, ino);
975}
976
977EXPORT_SYMBOL(iget_locked);
978
979/**
980 * __insert_inode_hash - hash an inode
981 * @inode: unhashed inode
982 * @hashval: unsigned long value used to locate this object in the
983 * inode_hashtable.
984 *
985 * Add an inode to the inode hash for this superblock.
986 */
987void __insert_inode_hash(struct inode *inode, unsigned long hashval)
988{
989 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
990 spin_lock(&inode_lock);
991 hlist_add_head(&inode->i_hash, head);
992 spin_unlock(&inode_lock);
993}
994
995EXPORT_SYMBOL(__insert_inode_hash);
996
997/**
998 * remove_inode_hash - remove an inode from the hash
999 * @inode: inode to unhash
1000 *
1001 * Remove an inode from the superblock.
1002 */
1003void remove_inode_hash(struct inode *inode)
1004{
1005 spin_lock(&inode_lock);
1006 hlist_del_init(&inode->i_hash);
1007 spin_unlock(&inode_lock);
1008}
1009
1010EXPORT_SYMBOL(remove_inode_hash);
1011
1012/*
1013 * Tell the filesystem that this inode is no longer of any interest and should
1014 * be completely destroyed.
1015 *
1016 * We leave the inode in the inode hash table until *after* the filesystem's
1017 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1018 * instigate) will always find up-to-date information either in the hash or on
1019 * disk.
1020 *
1021 * I_FREEING is set so that no-one will take a new reference to the inode while
1022 * it is being deleted.
1023 */
1024void generic_delete_inode(struct inode *inode)
1025{
1026 struct super_operations *op = inode->i_sb->s_op;
1027
1028 list_del_init(&inode->i_list);
1029 list_del_init(&inode->i_sb_list);
1030 inode->i_state|=I_FREEING;
1031 inodes_stat.nr_inodes--;
1032 spin_unlock(&inode_lock);
1033
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 security_inode_delete(inode);
1035
1036 if (op->delete_inode) {
1037 void (*delete)(struct inode *) = op->delete_inode;
1038 if (!is_bad_inode(inode))
1039 DQUOT_INIT(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001040 /* Filesystems implementing their own
1041 * s_op->delete_inode are required to call
1042 * truncate_inode_pages and clear_inode()
1043 * internally */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 delete(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001045 } else {
1046 truncate_inode_pages(&inode->i_data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 clear_inode(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001048 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 spin_lock(&inode_lock);
1050 hlist_del_init(&inode->i_hash);
1051 spin_unlock(&inode_lock);
1052 wake_up_inode(inode);
Eric Sesterhennb7542f82006-04-02 13:38:18 +02001053 BUG_ON(inode->i_state != I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 destroy_inode(inode);
1055}
1056
1057EXPORT_SYMBOL(generic_delete_inode);
1058
1059static void generic_forget_inode(struct inode *inode)
1060{
1061 struct super_block *sb = inode->i_sb;
1062
1063 if (!hlist_unhashed(&inode->i_hash)) {
1064 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1065 list_move(&inode->i_list, &inode_unused);
1066 inodes_stat.nr_unused++;
Alexander Viro991114c2005-06-23 00:09:01 -07001067 if (!sb || (sb->s_flags & MS_ACTIVE)) {
1068 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001070 }
1071 inode->i_state |= I_WILL_FREE;
1072 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 write_inode_now(inode, 1);
1074 spin_lock(&inode_lock);
Alexander Viro991114c2005-06-23 00:09:01 -07001075 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 inodes_stat.nr_unused--;
1077 hlist_del_init(&inode->i_hash);
1078 }
1079 list_del_init(&inode->i_list);
1080 list_del_init(&inode->i_sb_list);
Alexander Viro991114c2005-06-23 00:09:01 -07001081 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 inodes_stat.nr_inodes--;
1083 spin_unlock(&inode_lock);
1084 if (inode->i_data.nrpages)
1085 truncate_inode_pages(&inode->i_data, 0);
1086 clear_inode(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001087 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 destroy_inode(inode);
1089}
1090
1091/*
1092 * Normal UNIX filesystem behaviour: delete the
1093 * inode when the usage count drops to zero, and
1094 * i_nlink is zero.
1095 */
Mark Fashehcb2c0232005-07-07 17:56:03 -07001096void generic_drop_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097{
1098 if (!inode->i_nlink)
1099 generic_delete_inode(inode);
1100 else
1101 generic_forget_inode(inode);
1102}
1103
Mark Fashehcb2c0232005-07-07 17:56:03 -07001104EXPORT_SYMBOL_GPL(generic_drop_inode);
1105
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106/*
1107 * Called when we're dropping the last reference
1108 * to an inode.
1109 *
1110 * Call the FS "drop()" function, defaulting to
1111 * the legacy UNIX filesystem behaviour..
1112 *
1113 * NOTE! NOTE! NOTE! We're called with the inode lock
1114 * held, and the drop function is supposed to release
1115 * the lock!
1116 */
1117static inline void iput_final(struct inode *inode)
1118{
1119 struct super_operations *op = inode->i_sb->s_op;
1120 void (*drop)(struct inode *) = generic_drop_inode;
1121
1122 if (op && op->drop_inode)
1123 drop = op->drop_inode;
1124 drop(inode);
1125}
1126
1127/**
1128 * iput - put an inode
1129 * @inode: inode to put
1130 *
1131 * Puts an inode, dropping its usage count. If the inode use count hits
1132 * zero, the inode is then freed and may also be destroyed.
1133 *
1134 * Consequently, iput() can sleep.
1135 */
1136void iput(struct inode *inode)
1137{
1138 if (inode) {
1139 struct super_operations *op = inode->i_sb->s_op;
1140
1141 BUG_ON(inode->i_state == I_CLEAR);
1142
1143 if (op && op->put_inode)
1144 op->put_inode(inode);
1145
1146 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1147 iput_final(inode);
1148 }
1149}
1150
1151EXPORT_SYMBOL(iput);
1152
1153/**
1154 * bmap - find a block number in a file
1155 * @inode: inode of file
1156 * @block: block to find
1157 *
1158 * Returns the block number on the device holding the inode that
1159 * is the disk block number for the block of the file requested.
1160 * That is, asked for block 4 of inode 1 the function will return the
1161 * disk block relative to the disk start that holds that block of the
1162 * file.
1163 */
1164sector_t bmap(struct inode * inode, sector_t block)
1165{
1166 sector_t res = 0;
1167 if (inode->i_mapping->a_ops->bmap)
1168 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1169 return res;
1170}
1171
1172EXPORT_SYMBOL(bmap);
1173
1174/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001175 * touch_atime - update the access time
1176 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001177 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 *
1179 * Update the accessed time on an inode and mark it for writeback.
1180 * This function automatically handles read only file systems and media,
1181 * as well as the "noatime" flag and inode specific "noatime" markers.
1182 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001183void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001185 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 struct timespec now;
1187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 if (IS_RDONLY(inode))
1189 return;
1190
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001191 if ((inode->i_flags & S_NOATIME) ||
1192 (inode->i_sb->s_flags & MS_NOATIME) ||
1193 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
1194 return;
1195
1196 /*
1197 * We may have a NULL vfsmount when coming from NFSD
1198 */
1199 if (mnt &&
1200 ((mnt->mnt_flags & MNT_NOATIME) ||
1201 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))))
1202 return;
1203
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 now = current_fs_time(inode->i_sb);
1205 if (!timespec_equal(&inode->i_atime, &now)) {
1206 inode->i_atime = now;
1207 mark_inode_dirty_sync(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 }
1209}
1210
Christoph Hellwig869243a2006-01-09 20:52:03 -08001211EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
1213/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001214 * file_update_time - update mtime and ctime time
1215 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001217 * Update the mtime and ctime members of an inode and mark the inode
1218 * for writeback. Note that this function is meant exclusively for
1219 * usage in the file write path of filesystems, and filesystems may
1220 * choose to explicitly ignore update via this function with the
1221 * S_NOCTIME inode flag, e.g. for network filesystem where these
1222 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 */
1224
Christoph Hellwig870f4812006-01-09 20:52:01 -08001225void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226{
Christoph Hellwig870f4812006-01-09 20:52:01 -08001227 struct inode *inode = file->f_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 struct timespec now;
1229 int sync_it = 0;
1230
1231 if (IS_NOCMTIME(inode))
1232 return;
1233 if (IS_RDONLY(inode))
1234 return;
1235
1236 now = current_fs_time(inode->i_sb);
1237 if (!timespec_equal(&inode->i_mtime, &now))
1238 sync_it = 1;
1239 inode->i_mtime = now;
1240
Christoph Hellwig870f4812006-01-09 20:52:01 -08001241 if (!timespec_equal(&inode->i_ctime, &now))
1242 sync_it = 1;
1243 inode->i_ctime = now;
1244
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 if (sync_it)
1246 mark_inode_dirty_sync(inode);
1247}
1248
Christoph Hellwig870f4812006-01-09 20:52:01 -08001249EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
1251int inode_needs_sync(struct inode *inode)
1252{
1253 if (IS_SYNC(inode))
1254 return 1;
1255 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1256 return 1;
1257 return 0;
1258}
1259
1260EXPORT_SYMBOL(inode_needs_sync);
1261
1262/*
1263 * Quota functions that want to walk the inode lists..
1264 */
1265#ifdef CONFIG_QUOTA
1266
1267/* Function back in dquot.c */
1268int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1269
1270void remove_dquot_ref(struct super_block *sb, int type,
1271 struct list_head *tofree_head)
1272{
1273 struct inode *inode;
1274
1275 if (!sb->dq_op)
1276 return; /* nothing to do */
1277 spin_lock(&inode_lock); /* This lock is for inodes code */
1278
1279 /*
1280 * We don't have to lock against quota code - test IS_QUOTAINIT is
1281 * just for speedup...
1282 */
1283 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1284 if (!IS_NOQUOTA(inode))
1285 remove_inode_dquot_ref(inode, type, tofree_head);
1286
1287 spin_unlock(&inode_lock);
1288}
1289
1290#endif
1291
1292int inode_wait(void *word)
1293{
1294 schedule();
1295 return 0;
1296}
1297
1298/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001299 * If we try to find an inode in the inode hash while it is being
1300 * deleted, we have to wait until the filesystem completes its
1301 * deletion before reporting that it isn't found. This function waits
1302 * until the deletion _might_ have completed. Callers are responsible
1303 * to recheck inode state.
1304 *
1305 * It doesn't matter if I_LOCK is not set initially, a call to
1306 * wake_up_inode() after removing from the hash list will DTRT.
1307 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 * This is called with inode_lock held.
1309 */
1310static void __wait_on_freeing_inode(struct inode *inode)
1311{
1312 wait_queue_head_t *wq;
1313 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1315 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1316 spin_unlock(&inode_lock);
1317 schedule();
1318 finish_wait(wq, &wait.wait);
1319 spin_lock(&inode_lock);
1320}
1321
1322void wake_up_inode(struct inode *inode)
1323{
1324 /*
1325 * Prevent speculative execution through spin_unlock(&inode_lock);
1326 */
1327 smp_mb();
1328 wake_up_bit(&inode->i_state, __I_LOCK);
1329}
1330
1331static __initdata unsigned long ihash_entries;
1332static int __init set_ihash_entries(char *str)
1333{
1334 if (!str)
1335 return 0;
1336 ihash_entries = simple_strtoul(str, &str, 0);
1337 return 1;
1338}
1339__setup("ihash_entries=", set_ihash_entries);
1340
1341/*
1342 * Initialize the waitqueues and inode hash table.
1343 */
1344void __init inode_init_early(void)
1345{
1346 int loop;
1347
1348 /* If hashes are distributed across NUMA nodes, defer
1349 * hash allocation until vmalloc space is available.
1350 */
1351 if (hashdist)
1352 return;
1353
1354 inode_hashtable =
1355 alloc_large_system_hash("Inode-cache",
1356 sizeof(struct hlist_head),
1357 ihash_entries,
1358 14,
1359 HASH_EARLY,
1360 &i_hash_shift,
1361 &i_hash_mask,
1362 0);
1363
1364 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1365 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1366}
1367
1368void __init inode_init(unsigned long mempages)
1369{
1370 int loop;
1371
1372 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001373 inode_cachep = kmem_cache_create("inode_cache",
1374 sizeof(struct inode),
1375 0,
1376 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1377 SLAB_MEM_SPREAD),
1378 init_once,
1379 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1381
1382 /* Hash may have been set up in inode_init_early */
1383 if (!hashdist)
1384 return;
1385
1386 inode_hashtable =
1387 alloc_large_system_hash("Inode-cache",
1388 sizeof(struct hlist_head),
1389 ihash_entries,
1390 14,
1391 0,
1392 &i_hash_shift,
1393 &i_hash_mask,
1394 0);
1395
1396 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1397 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1398}
1399
1400void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1401{
1402 inode->i_mode = mode;
1403 if (S_ISCHR(mode)) {
1404 inode->i_fop = &def_chr_fops;
1405 inode->i_rdev = rdev;
1406 } else if (S_ISBLK(mode)) {
1407 inode->i_fop = &def_blk_fops;
1408 inode->i_rdev = rdev;
1409 } else if (S_ISFIFO(mode))
1410 inode->i_fop = &def_fifo_fops;
1411 else if (S_ISSOCK(mode))
1412 inode->i_fop = &bad_sock_fops;
1413 else
1414 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1415 mode);
1416}
1417EXPORT_SYMBOL(init_special_inode);