blob: 9a012cc5b6cd6e060977a6b13565c8f11d98dfca [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/inode.c
3 *
4 * (C) 1997 Linus Torvalds
5 */
6
Linus Torvalds1da177e2005-04-16 15:20:36 -07007#include <linux/fs.h>
8#include <linux/mm.h>
9#include <linux/dcache.h>
10#include <linux/init.h>
11#include <linux/quotaops.h>
12#include <linux/slab.h>
13#include <linux/writeback.h>
14#include <linux/module.h>
15#include <linux/backing-dev.h>
16#include <linux/wait.h>
17#include <linux/hash.h>
18#include <linux/swap.h>
19#include <linux/security.h>
20#include <linux/pagemap.h>
21#include <linux/cdev.h>
22#include <linux/bootmem.h>
Robert Love0eeca282005-07-12 17:06:03 -040023#include <linux/inotify.h>
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -080024#include <linux/mount.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26/*
27 * This is needed for the following functions:
28 * - inode_has_buffers
29 * - invalidate_inode_buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 * - 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
Eric Dumazetfa3536c2006-03-26 01:37:24 -080058static unsigned int i_hash_mask __read_mostly;
59static unsigned int i_hash_shift __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
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);
Eric Dumazetfa3536c2006-03-26 01:37:24 -080075static struct hlist_head *inode_hashtable __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
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/*
Ingo Molnarf24075b2006-03-23 03:00:34 -080086 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 * 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 */
Adrian Bunkbdfc3262006-03-25 03:06:56 -080093static DEFINE_MUTEX(iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
95/*
96 * Statistics gathering..
97 */
98struct inodes_stat_t inodes_stat;
99
Christoph Lametere18b8902006-12-06 20:33:20 -0800100static struct kmem_cache * inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
102static struct inode *alloc_inode(struct super_block *sb)
103{
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700104 static const struct address_space_operations empty_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 static struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800106 static const struct file_operations empty_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 struct inode *inode;
108
109 if (sb->s_op->alloc_inode)
110 inode = sb->s_op->alloc_inode(sb);
111 else
Christoph Lametere94b1762006-12-06 20:33:17 -0800112 inode = (struct inode *) kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 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 }
Al Viroe6c6e642006-10-10 22:48:47 +0100165 inode->i_private = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 inode->i_mapping = mapping;
167 }
168 return inode;
169}
170
171void destroy_inode(struct inode *inode)
172{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200173 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 security_inode_free(inode);
175 if (inode->i_sb->s_op->destroy_inode)
176 inode->i_sb->s_op->destroy_inode(inode);
177 else
178 kmem_cache_free(inode_cachep, (inode));
179}
180
181
182/*
183 * These are initializations that only need to be done
184 * once, because the fields are idempotent across use
185 * of the inode, so let the slab aware of that.
186 */
187void inode_init_once(struct inode *inode)
188{
189 memset(inode, 0, sizeof(*inode));
190 INIT_HLIST_NODE(&inode->i_hash);
191 INIT_LIST_HEAD(&inode->i_dentry);
192 INIT_LIST_HEAD(&inode->i_devices);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800193 mutex_init(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 init_rwsem(&inode->i_alloc_sem);
195 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
196 rwlock_init(&inode->i_data.tree_lock);
197 spin_lock_init(&inode->i_data.i_mmap_lock);
198 INIT_LIST_HEAD(&inode->i_data.private_list);
199 spin_lock_init(&inode->i_data.private_lock);
200 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
201 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
202 spin_lock_init(&inode->i_lock);
203 i_size_ordered_init(inode);
Robert Love0eeca282005-07-12 17:06:03 -0400204#ifdef CONFIG_INOTIFY
205 INIT_LIST_HEAD(&inode->inotify_watches);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800206 mutex_init(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400207#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
209
210EXPORT_SYMBOL(inode_init_once);
211
Christoph Lametere18b8902006-12-06 20:33:20 -0800212static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213{
214 struct inode * inode = (struct inode *) foo;
215
Christoph Lametera35afb82007-05-16 22:10:57 -0700216 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217}
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
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200247 BUG_ON(inode->i_data.nrpages);
248 BUG_ON(!(inode->i_state & I_FREEING));
249 BUG_ON(inode->i_state & I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 wait_on_inode(inode);
251 DQUOT_DROP(inode);
Christoph Hellwigacb0c852007-05-08 00:25:52 -0700252 if (inode->i_sb->s_op->clear_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 inode->i_sb->s_op->clear_inode(inode);
Theodore Ts'oeaf796e2006-09-27 01:50:48 -0700254 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 bd_forget(inode);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700256 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 cd_forget(inode);
258 inode->i_state = I_CLEAR;
259}
260
261EXPORT_SYMBOL(clear_inode);
262
263/*
264 * dispose_list - dispose of the contents of a local list
265 * @head: the head of the list to free
266 *
267 * Dispose-list gets a local list with local inodes in it, so it doesn't
268 * need to worry about list corruption and SMP locks.
269 */
270static void dispose_list(struct list_head *head)
271{
272 int nr_disposed = 0;
273
274 while (!list_empty(head)) {
275 struct inode *inode;
276
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700277 inode = list_first_entry(head, struct inode, i_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 list_del(&inode->i_list);
279
280 if (inode->i_data.nrpages)
281 truncate_inode_pages(&inode->i_data, 0);
282 clear_inode(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700283
284 spin_lock(&inode_lock);
285 hlist_del_init(&inode->i_hash);
286 list_del_init(&inode->i_sb_list);
287 spin_unlock(&inode_lock);
288
289 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 destroy_inode(inode);
291 nr_disposed++;
292 }
293 spin_lock(&inode_lock);
294 inodes_stat.nr_inodes -= nr_disposed;
295 spin_unlock(&inode_lock);
296}
297
298/*
299 * Invalidate all inodes for a device.
300 */
301static int invalidate_list(struct list_head *head, struct list_head *dispose)
302{
303 struct list_head *next;
304 int busy = 0, count = 0;
305
306 next = head->next;
307 for (;;) {
308 struct list_head * tmp = next;
309 struct inode * inode;
310
311 /*
312 * We can reschedule here without worrying about the list's
313 * consistency because the per-sb list of inodes must not
Ingo Molnarf24075b2006-03-23 03:00:34 -0800314 * change during umount anymore, and because iprune_mutex keeps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 * shrink_icache_memory() away.
316 */
317 cond_resched_lock(&inode_lock);
318
319 next = next->next;
320 if (tmp == head)
321 break;
322 inode = list_entry(tmp, struct inode, i_sb_list);
323 invalidate_inode_buffers(inode);
324 if (!atomic_read(&inode->i_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 list_move(&inode->i_list, dispose);
326 inode->i_state |= I_FREEING;
327 count++;
328 continue;
329 }
330 busy = 1;
331 }
332 /* only unused inodes may be cached with i_count zero */
333 inodes_stat.nr_unused -= count;
334 return busy;
335}
336
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337/**
338 * invalidate_inodes - discard the inodes on a device
339 * @sb: superblock
340 *
341 * Discard all of the inodes for a given superblock. If the discard
342 * fails because there are busy inodes then a non zero value is returned.
343 * If the discard is successful all the inodes have been discarded.
344 */
345int invalidate_inodes(struct super_block * sb)
346{
347 int busy;
348 LIST_HEAD(throw_away);
349
Ingo Molnarf24075b2006-03-23 03:00:34 -0800350 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 spin_lock(&inode_lock);
Robert Love0eeca282005-07-12 17:06:03 -0400352 inotify_unmount_inodes(&sb->s_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 busy = invalidate_list(&sb->s_inodes, &throw_away);
354 spin_unlock(&inode_lock);
355
356 dispose_list(&throw_away);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800357 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
359 return busy;
360}
361
362EXPORT_SYMBOL(invalidate_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363
364static int can_unuse(struct inode *inode)
365{
366 if (inode->i_state)
367 return 0;
368 if (inode_has_buffers(inode))
369 return 0;
370 if (atomic_read(&inode->i_count))
371 return 0;
372 if (inode->i_data.nrpages)
373 return 0;
374 return 1;
375}
376
377/*
378 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
379 * a temporary list and then are freed outside inode_lock by dispose_list().
380 *
381 * Any inodes which are pinned purely because of attached pagecache have their
382 * pagecache removed. We expect the final iput() on that inode to add it to
383 * the front of the inode_unused list. So look for it there and if the
384 * inode is still freeable, proceed. The right inode is found 99.9% of the
385 * time in testing on a 4-way.
386 *
387 * If the inode has metadata buffers attached to mapping->private_list then
388 * try to remove them.
389 */
390static void prune_icache(int nr_to_scan)
391{
392 LIST_HEAD(freeable);
393 int nr_pruned = 0;
394 int nr_scanned;
395 unsigned long reap = 0;
396
Ingo Molnarf24075b2006-03-23 03:00:34 -0800397 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 spin_lock(&inode_lock);
399 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
400 struct inode *inode;
401
402 if (list_empty(&inode_unused))
403 break;
404
405 inode = list_entry(inode_unused.prev, struct inode, i_list);
406
407 if (inode->i_state || atomic_read(&inode->i_count)) {
408 list_move(&inode->i_list, &inode_unused);
409 continue;
410 }
411 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
412 __iget(inode);
413 spin_unlock(&inode_lock);
414 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800415 reap += invalidate_mapping_pages(&inode->i_data,
416 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 iput(inode);
418 spin_lock(&inode_lock);
419
420 if (inode != list_entry(inode_unused.next,
421 struct inode, i_list))
422 continue; /* wrong inode or list_empty */
423 if (!can_unuse(inode))
424 continue;
425 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 list_move(&inode->i_list, &freeable);
427 inode->i_state |= I_FREEING;
428 nr_pruned++;
429 }
430 inodes_stat.nr_unused -= nr_pruned;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700431 if (current_is_kswapd())
432 __count_vm_events(KSWAPD_INODESTEAL, reap);
433 else
434 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 spin_unlock(&inode_lock);
436
437 dispose_list(&freeable);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800438 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439}
440
441/*
442 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
443 * "unused" means that no dentries are referring to the inodes: the files are
444 * not open and the dcache references to those inodes have already been
445 * reclaimed.
446 *
447 * This function is passed the number of inodes to scan, and it returns the
448 * total number of remaining possibly-reclaimable inodes.
449 */
Al Viro27496a82005-10-21 03:20:48 -0400450static int shrink_icache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451{
452 if (nr) {
453 /*
454 * Nasty deadlock avoidance. We may hold various FS locks,
455 * and we don't want to recurse into the FS that called us
456 * in clear_inode() and friends..
457 */
458 if (!(gfp_mask & __GFP_FS))
459 return -1;
460 prune_icache(nr);
461 }
462 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
463}
464
465static void __wait_on_freeing_inode(struct inode *inode);
466/*
467 * Called with the inode lock held.
468 * NOTE: we are not increasing the inode-refcount, you must call __iget()
469 * by hand after calling find_inode now! This simplifies iunique and won't
470 * add any additional branch in the common code.
471 */
472static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
473{
474 struct hlist_node *node;
475 struct inode * inode = NULL;
476
477repeat:
478 hlist_for_each (node, head) {
479 inode = hlist_entry(node, struct inode, i_hash);
480 if (inode->i_sb != sb)
481 continue;
482 if (!test(inode, data))
483 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700484 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 __wait_on_freeing_inode(inode);
486 goto repeat;
487 }
488 break;
489 }
490 return node ? inode : NULL;
491}
492
493/*
494 * find_inode_fast is the fast path version of find_inode, see the comment at
495 * iget_locked for details.
496 */
497static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
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_ino != ino)
506 continue;
507 if (inode->i_sb != sb)
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 * new_inode - obtain an inode
520 * @sb: superblock
521 *
522 * Allocates a new inode for given superblock.
523 */
524struct inode *new_inode(struct super_block *sb)
525{
Jeff Layton866b04f2007-05-08 00:32:29 -0700526 /*
527 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
528 * error if st_ino won't fit in target struct field. Use 32bit counter
529 * here to attempt to avoid that.
530 */
531 static unsigned int last_ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 struct inode * inode;
533
534 spin_lock_prefetch(&inode_lock);
535
536 inode = alloc_inode(sb);
537 if (inode) {
538 spin_lock(&inode_lock);
539 inodes_stat.nr_inodes++;
540 list_add(&inode->i_list, &inode_in_use);
541 list_add(&inode->i_sb_list, &sb->s_inodes);
542 inode->i_ino = ++last_ino;
543 inode->i_state = 0;
544 spin_unlock(&inode_lock);
545 }
546 return inode;
547}
548
549EXPORT_SYMBOL(new_inode);
550
551void unlock_new_inode(struct inode *inode)
552{
553 /*
554 * This is special! We do not need the spinlock
555 * when clearing I_LOCK, because we're guaranteed
556 * that nobody else tries to do anything about the
557 * state of the inode when it is locked, as we
558 * just created it (so there can be no old holders
559 * that haven't tested I_LOCK).
560 */
561 inode->i_state &= ~(I_LOCK|I_NEW);
562 wake_up_inode(inode);
563}
564
565EXPORT_SYMBOL(unlock_new_inode);
566
567/*
568 * This is called without the inode lock held.. Be careful.
569 *
570 * We no longer cache the sb_flags in i_flags - see fs.h
571 * -- rmk@arm.uk.linux.org
572 */
573static 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)
574{
575 struct inode * inode;
576
577 inode = alloc_inode(sb);
578 if (inode) {
579 struct inode * old;
580
581 spin_lock(&inode_lock);
582 /* We released the lock, so.. */
583 old = find_inode(sb, head, test, data);
584 if (!old) {
585 if (set(inode, data))
586 goto set_failed;
587
588 inodes_stat.nr_inodes++;
589 list_add(&inode->i_list, &inode_in_use);
590 list_add(&inode->i_sb_list, &sb->s_inodes);
591 hlist_add_head(&inode->i_hash, head);
592 inode->i_state = I_LOCK|I_NEW;
593 spin_unlock(&inode_lock);
594
595 /* Return the locked inode with I_NEW set, the
596 * caller is responsible for filling in the contents
597 */
598 return inode;
599 }
600
601 /*
602 * Uhhuh, somebody else created the same inode under
603 * us. Use the old inode instead of the one we just
604 * allocated.
605 */
606 __iget(old);
607 spin_unlock(&inode_lock);
608 destroy_inode(inode);
609 inode = old;
610 wait_on_inode(inode);
611 }
612 return inode;
613
614set_failed:
615 spin_unlock(&inode_lock);
616 destroy_inode(inode);
617 return NULL;
618}
619
620/*
621 * get_new_inode_fast is the fast path version of get_new_inode, see the
622 * comment at iget_locked for details.
623 */
624static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
625{
626 struct inode * inode;
627
628 inode = alloc_inode(sb);
629 if (inode) {
630 struct inode * old;
631
632 spin_lock(&inode_lock);
633 /* We released the lock, so.. */
634 old = find_inode_fast(sb, head, ino);
635 if (!old) {
636 inode->i_ino = ino;
637 inodes_stat.nr_inodes++;
638 list_add(&inode->i_list, &inode_in_use);
639 list_add(&inode->i_sb_list, &sb->s_inodes);
640 hlist_add_head(&inode->i_hash, head);
641 inode->i_state = I_LOCK|I_NEW;
642 spin_unlock(&inode_lock);
643
644 /* Return the locked inode with I_NEW set, the
645 * caller is responsible for filling in the contents
646 */
647 return inode;
648 }
649
650 /*
651 * Uhhuh, somebody else created the same inode under
652 * us. Use the old inode instead of the one we just
653 * allocated.
654 */
655 __iget(old);
656 spin_unlock(&inode_lock);
657 destroy_inode(inode);
658 inode = old;
659 wait_on_inode(inode);
660 }
661 return inode;
662}
663
Andreas Mohred97bd32006-10-02 02:17:17 -0700664static unsigned long hash(struct super_block *sb, unsigned long hashval)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
666 unsigned long tmp;
667
668 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
669 L1_CACHE_BYTES;
670 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
671 return tmp & I_HASHMASK;
672}
673
674/**
675 * iunique - get a unique inode number
676 * @sb: superblock
677 * @max_reserved: highest reserved inode number
678 *
679 * Obtain an inode number that is unique on the system for a given
680 * superblock. This is used by file systems that have no natural
681 * permanent inode numbering system. An inode number is returned that
682 * is higher than the reserved limit but unique.
683 *
684 * BUGS:
685 * With a large number of inodes live on the file system this function
686 * currently becomes quite slow.
687 */
688ino_t iunique(struct super_block *sb, ino_t max_reserved)
689{
Jeff Layton866b04f2007-05-08 00:32:29 -0700690 /*
691 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
692 * error if st_ino won't fit in target struct field. Use 32bit counter
693 * here to attempt to avoid that.
694 */
695 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 struct inode *inode;
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700697 struct hlist_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700700 spin_lock(&inode_lock);
701 do {
702 if (counter <= max_reserved)
703 counter = max_reserved + 1;
704 res = counter++;
705 head = inode_hashtable + hash(sb, res);
706 inode = find_inode_fast(sb, head, res);
707 } while (inode != NULL);
708 spin_unlock(&inode_lock);
709
710 return res;
711}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712EXPORT_SYMBOL(iunique);
713
714struct inode *igrab(struct inode *inode)
715{
716 spin_lock(&inode_lock);
Jan Blunck4a3b0a42007-02-10 01:44:59 -0800717 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 __iget(inode);
719 else
720 /*
721 * Handle the case where s_op->clear_inode is not been
722 * called yet, and somebody is calling igrab
723 * while the inode is getting freed.
724 */
725 inode = NULL;
726 spin_unlock(&inode_lock);
727 return inode;
728}
729
730EXPORT_SYMBOL(igrab);
731
732/**
733 * ifind - internal function, you want ilookup5() or iget5().
734 * @sb: super block of file system to search
735 * @head: the head of the list to search
736 * @test: callback used for comparisons between inodes
737 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700738 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 *
740 * ifind() searches for the inode specified by @data in the inode
741 * cache. This is a generalized version of ifind_fast() for file systems where
742 * the inode number is not sufficient for unique identification of an inode.
743 *
744 * If the inode is in the cache, the inode is returned with an incremented
745 * reference count.
746 *
747 * Otherwise NULL is returned.
748 *
749 * Note, @test is called with the inode_lock held, so can't sleep.
750 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800751static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700753 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754{
755 struct inode *inode;
756
757 spin_lock(&inode_lock);
758 inode = find_inode(sb, head, test, data);
759 if (inode) {
760 __iget(inode);
761 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700762 if (likely(wait))
763 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 return inode;
765 }
766 spin_unlock(&inode_lock);
767 return NULL;
768}
769
770/**
771 * ifind_fast - internal function, you want ilookup() or iget().
772 * @sb: super block of file system to search
773 * @head: head of the list to search
774 * @ino: inode number to search for
775 *
776 * ifind_fast() searches for the inode @ino in the inode cache. This is for
777 * file systems where the inode number is sufficient for unique identification
778 * of an inode.
779 *
780 * If the inode is in the cache, the inode is returned with an incremented
781 * reference count.
782 *
783 * Otherwise NULL is returned.
784 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800785static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 struct hlist_head *head, unsigned long ino)
787{
788 struct inode *inode;
789
790 spin_lock(&inode_lock);
791 inode = find_inode_fast(sb, head, ino);
792 if (inode) {
793 __iget(inode);
794 spin_unlock(&inode_lock);
795 wait_on_inode(inode);
796 return inode;
797 }
798 spin_unlock(&inode_lock);
799 return NULL;
800}
801
802/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700803 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 * @sb: super block of file system to search
805 * @hashval: hash value (usually inode number) to search for
806 * @test: callback used for comparisons between inodes
807 * @data: opaque data pointer to pass to @test
808 *
809 * ilookup5() uses ifind() to search for the inode specified by @hashval and
810 * @data in the inode cache. This is a generalized version of ilookup() for
811 * file systems where the inode number is not sufficient for unique
812 * identification of an inode.
813 *
814 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700815 * reference count. Note, the inode lock is not waited upon so you have to be
816 * very careful what you do with the returned inode. You probably should be
817 * using ilookup5() instead.
818 *
819 * Otherwise NULL is returned.
820 *
821 * Note, @test is called with the inode_lock held, so can't sleep.
822 */
823struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
824 int (*test)(struct inode *, void *), void *data)
825{
826 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
827
828 return ifind(sb, head, test, data, 0);
829}
830
831EXPORT_SYMBOL(ilookup5_nowait);
832
833/**
834 * ilookup5 - search for an inode in the inode cache
835 * @sb: super block of file system to search
836 * @hashval: hash value (usually inode number) to search for
837 * @test: callback used for comparisons between inodes
838 * @data: opaque data pointer to pass to @test
839 *
840 * ilookup5() uses ifind() to search for the inode specified by @hashval and
841 * @data in the inode cache. This is a generalized version of ilookup() for
842 * file systems where the inode number is not sufficient for unique
843 * identification of an inode.
844 *
845 * If the inode is in the cache, the inode lock is waited upon and the inode is
846 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 *
848 * Otherwise NULL is returned.
849 *
850 * Note, @test is called with the inode_lock held, so can't sleep.
851 */
852struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
853 int (*test)(struct inode *, void *), void *data)
854{
855 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
856
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700857 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858}
859
860EXPORT_SYMBOL(ilookup5);
861
862/**
863 * ilookup - search for an inode in the inode cache
864 * @sb: super block of file system to search
865 * @ino: inode number to search for
866 *
867 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
868 * This is for file systems where the inode number is sufficient for unique
869 * identification of an inode.
870 *
871 * If the inode is in the cache, the inode is returned with an incremented
872 * reference count.
873 *
874 * Otherwise NULL is returned.
875 */
876struct inode *ilookup(struct super_block *sb, unsigned long ino)
877{
878 struct hlist_head *head = inode_hashtable + hash(sb, ino);
879
880 return ifind_fast(sb, head, ino);
881}
882
883EXPORT_SYMBOL(ilookup);
884
885/**
886 * iget5_locked - obtain an inode from a mounted file system
887 * @sb: super block of file system
888 * @hashval: hash value (usually inode number) to get
889 * @test: callback used for comparisons between inodes
890 * @set: callback used to initialize a new struct inode
891 * @data: opaque data pointer to pass to @test and @set
892 *
893 * This is iget() without the read_inode() portion of get_new_inode().
894 *
895 * iget5_locked() uses ifind() to search for the inode specified by @hashval
896 * and @data in the inode cache and if present it is returned with an increased
897 * reference count. This is a generalized version of iget_locked() for file
898 * systems where the inode number is not sufficient for unique identification
899 * of an inode.
900 *
901 * If the inode is not in cache, get_new_inode() is called to allocate a new
902 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
903 * file system gets to fill it in before unlocking it via unlock_new_inode().
904 *
905 * Note both @test and @set are called with the inode_lock held, so can't sleep.
906 */
907struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
908 int (*test)(struct inode *, void *),
909 int (*set)(struct inode *, void *), void *data)
910{
911 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
912 struct inode *inode;
913
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700914 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 if (inode)
916 return inode;
917 /*
918 * get_new_inode() will do the right thing, re-trying the search
919 * in case it had to block at any point.
920 */
921 return get_new_inode(sb, head, test, set, data);
922}
923
924EXPORT_SYMBOL(iget5_locked);
925
926/**
927 * iget_locked - obtain an inode from a mounted file system
928 * @sb: super block of file system
929 * @ino: inode number to get
930 *
931 * This is iget() without the read_inode() portion of get_new_inode_fast().
932 *
933 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
934 * the inode cache and if present it is returned with an increased reference
935 * count. This is for file systems where the inode number is sufficient for
936 * unique identification of an inode.
937 *
938 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
939 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
940 * The file system gets to fill it in before unlocking it via
941 * unlock_new_inode().
942 */
943struct inode *iget_locked(struct super_block *sb, unsigned long ino)
944{
945 struct hlist_head *head = inode_hashtable + hash(sb, ino);
946 struct inode *inode;
947
948 inode = ifind_fast(sb, head, ino);
949 if (inode)
950 return inode;
951 /*
952 * get_new_inode_fast() will do the right thing, re-trying the search
953 * in case it had to block at any point.
954 */
955 return get_new_inode_fast(sb, head, ino);
956}
957
958EXPORT_SYMBOL(iget_locked);
959
960/**
961 * __insert_inode_hash - hash an inode
962 * @inode: unhashed inode
963 * @hashval: unsigned long value used to locate this object in the
964 * inode_hashtable.
965 *
966 * Add an inode to the inode hash for this superblock.
967 */
968void __insert_inode_hash(struct inode *inode, unsigned long hashval)
969{
970 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
971 spin_lock(&inode_lock);
972 hlist_add_head(&inode->i_hash, head);
973 spin_unlock(&inode_lock);
974}
975
976EXPORT_SYMBOL(__insert_inode_hash);
977
978/**
979 * remove_inode_hash - remove an inode from the hash
980 * @inode: inode to unhash
981 *
982 * Remove an inode from the superblock.
983 */
984void remove_inode_hash(struct inode *inode)
985{
986 spin_lock(&inode_lock);
987 hlist_del_init(&inode->i_hash);
988 spin_unlock(&inode_lock);
989}
990
991EXPORT_SYMBOL(remove_inode_hash);
992
993/*
994 * Tell the filesystem that this inode is no longer of any interest and should
995 * be completely destroyed.
996 *
997 * We leave the inode in the inode hash table until *after* the filesystem's
998 * ->delete_inode completes. This ensures that an iget (such as nfsd might
999 * instigate) will always find up-to-date information either in the hash or on
1000 * disk.
1001 *
1002 * I_FREEING is set so that no-one will take a new reference to the inode while
1003 * it is being deleted.
1004 */
1005void generic_delete_inode(struct inode *inode)
1006{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001007 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
1009 list_del_init(&inode->i_list);
1010 list_del_init(&inode->i_sb_list);
Andreas Mohred97bd32006-10-02 02:17:17 -07001011 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 inodes_stat.nr_inodes--;
1013 spin_unlock(&inode_lock);
1014
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 security_inode_delete(inode);
1016
1017 if (op->delete_inode) {
1018 void (*delete)(struct inode *) = op->delete_inode;
1019 if (!is_bad_inode(inode))
1020 DQUOT_INIT(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001021 /* Filesystems implementing their own
1022 * s_op->delete_inode are required to call
1023 * truncate_inode_pages and clear_inode()
1024 * internally */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 delete(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001026 } else {
1027 truncate_inode_pages(&inode->i_data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 clear_inode(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001029 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 spin_lock(&inode_lock);
1031 hlist_del_init(&inode->i_hash);
1032 spin_unlock(&inode_lock);
1033 wake_up_inode(inode);
Eric Sesterhennb7542f82006-04-02 13:38:18 +02001034 BUG_ON(inode->i_state != I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 destroy_inode(inode);
1036}
1037
1038EXPORT_SYMBOL(generic_delete_inode);
1039
1040static void generic_forget_inode(struct inode *inode)
1041{
1042 struct super_block *sb = inode->i_sb;
1043
1044 if (!hlist_unhashed(&inode->i_hash)) {
1045 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1046 list_move(&inode->i_list, &inode_unused);
1047 inodes_stat.nr_unused++;
Christoph Hellwigacb0c852007-05-08 00:25:52 -07001048 if (sb->s_flags & MS_ACTIVE) {
Alexander Viro991114c2005-06-23 00:09:01 -07001049 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001051 }
1052 inode->i_state |= I_WILL_FREE;
1053 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 write_inode_now(inode, 1);
1055 spin_lock(&inode_lock);
Alexander Viro991114c2005-06-23 00:09:01 -07001056 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 inodes_stat.nr_unused--;
1058 hlist_del_init(&inode->i_hash);
1059 }
1060 list_del_init(&inode->i_list);
1061 list_del_init(&inode->i_sb_list);
Alexander Viro991114c2005-06-23 00:09:01 -07001062 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 inodes_stat.nr_inodes--;
1064 spin_unlock(&inode_lock);
1065 if (inode->i_data.nrpages)
1066 truncate_inode_pages(&inode->i_data, 0);
1067 clear_inode(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001068 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 destroy_inode(inode);
1070}
1071
1072/*
1073 * Normal UNIX filesystem behaviour: delete the
1074 * inode when the usage count drops to zero, and
1075 * i_nlink is zero.
1076 */
Mark Fashehcb2c0232005-07-07 17:56:03 -07001077void generic_drop_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078{
1079 if (!inode->i_nlink)
1080 generic_delete_inode(inode);
1081 else
1082 generic_forget_inode(inode);
1083}
1084
Mark Fashehcb2c0232005-07-07 17:56:03 -07001085EXPORT_SYMBOL_GPL(generic_drop_inode);
1086
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087/*
1088 * Called when we're dropping the last reference
1089 * to an inode.
1090 *
1091 * Call the FS "drop()" function, defaulting to
1092 * the legacy UNIX filesystem behaviour..
1093 *
1094 * NOTE! NOTE! NOTE! We're called with the inode lock
1095 * held, and the drop function is supposed to release
1096 * the lock!
1097 */
1098static inline void iput_final(struct inode *inode)
1099{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001100 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 void (*drop)(struct inode *) = generic_drop_inode;
1102
1103 if (op && op->drop_inode)
1104 drop = op->drop_inode;
1105 drop(inode);
1106}
1107
1108/**
1109 * iput - put an inode
1110 * @inode: inode to put
1111 *
1112 * Puts an inode, dropping its usage count. If the inode use count hits
1113 * zero, the inode is then freed and may also be destroyed.
1114 *
1115 * Consequently, iput() can sleep.
1116 */
1117void iput(struct inode *inode)
1118{
1119 if (inode) {
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001120 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
1122 BUG_ON(inode->i_state == I_CLEAR);
1123
1124 if (op && op->put_inode)
1125 op->put_inode(inode);
1126
1127 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1128 iput_final(inode);
1129 }
1130}
1131
1132EXPORT_SYMBOL(iput);
1133
1134/**
1135 * bmap - find a block number in a file
1136 * @inode: inode of file
1137 * @block: block to find
1138 *
1139 * Returns the block number on the device holding the inode that
1140 * is the disk block number for the block of the file requested.
1141 * That is, asked for block 4 of inode 1 the function will return the
1142 * disk block relative to the disk start that holds that block of the
1143 * file.
1144 */
1145sector_t bmap(struct inode * inode, sector_t block)
1146{
1147 sector_t res = 0;
1148 if (inode->i_mapping->a_ops->bmap)
1149 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1150 return res;
1151}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152EXPORT_SYMBOL(bmap);
1153
1154/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001155 * touch_atime - update the access time
1156 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001157 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 *
1159 * Update the accessed time on an inode and mark it for writeback.
1160 * This function automatically handles read only file systems and media,
1161 * as well as the "noatime" flag and inode specific "noatime" markers.
1162 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001163void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001165 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 struct timespec now;
1167
Andrew Mortonb2276132006-12-13 00:34:33 -08001168 if (inode->i_flags & S_NOATIME)
1169 return;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001170 if (IS_NOATIME(inode))
Andrew Mortonb2276132006-12-13 00:34:33 -08001171 return;
1172 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001173 return;
1174
1175 /*
1176 * We may have a NULL vfsmount when coming from NFSD
1177 */
Andrew Mortonb2276132006-12-13 00:34:33 -08001178 if (mnt) {
1179 if (mnt->mnt_flags & MNT_NOATIME)
1180 return;
1181 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1182 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001183
1184 if (mnt->mnt_flags & MNT_RELATIME) {
1185 /*
1186 * With relative atime, only update atime if the
1187 * previous atime is earlier than either the ctime or
1188 * mtime.
1189 */
1190 if (timespec_compare(&inode->i_mtime,
1191 &inode->i_atime) < 0 &&
1192 timespec_compare(&inode->i_ctime,
1193 &inode->i_atime) < 0)
1194 return;
1195 }
Andrew Mortonb2276132006-12-13 00:34:33 -08001196 }
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001197
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 now = current_fs_time(inode->i_sb);
Valerie Henson47ae32d2006-12-13 00:34:34 -08001199 if (timespec_equal(&inode->i_atime, &now))
1200 return;
1201
1202 inode->i_atime = now;
1203 mark_inode_dirty_sync(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001205EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
1207/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001208 * file_update_time - update mtime and ctime time
1209 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001211 * Update the mtime and ctime members of an inode and mark the inode
1212 * for writeback. Note that this function is meant exclusively for
1213 * usage in the file write path of filesystems, and filesystems may
1214 * choose to explicitly ignore update via this function with the
1215 * S_NOCTIME inode flag, e.g. for network filesystem where these
1216 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 */
1218
Christoph Hellwig870f4812006-01-09 20:52:01 -08001219void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001221 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 struct timespec now;
1223 int sync_it = 0;
1224
1225 if (IS_NOCMTIME(inode))
1226 return;
1227 if (IS_RDONLY(inode))
1228 return;
1229
1230 now = current_fs_time(inode->i_sb);
Andreas Mohred97bd32006-10-02 02:17:17 -07001231 if (!timespec_equal(&inode->i_mtime, &now)) {
1232 inode->i_mtime = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001234 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235
Andreas Mohred97bd32006-10-02 02:17:17 -07001236 if (!timespec_equal(&inode->i_ctime, &now)) {
1237 inode->i_ctime = now;
Christoph Hellwig870f4812006-01-09 20:52:01 -08001238 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001239 }
Christoph Hellwig870f4812006-01-09 20:52:01 -08001240
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 if (sync_it)
1242 mark_inode_dirty_sync(inode);
1243}
1244
Christoph Hellwig870f4812006-01-09 20:52:01 -08001245EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
1247int inode_needs_sync(struct inode *inode)
1248{
1249 if (IS_SYNC(inode))
1250 return 1;
1251 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1252 return 1;
1253 return 0;
1254}
1255
1256EXPORT_SYMBOL(inode_needs_sync);
1257
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258int inode_wait(void *word)
1259{
1260 schedule();
1261 return 0;
1262}
1263
1264/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001265 * If we try to find an inode in the inode hash while it is being
1266 * deleted, we have to wait until the filesystem completes its
1267 * deletion before reporting that it isn't found. This function waits
1268 * until the deletion _might_ have completed. Callers are responsible
1269 * to recheck inode state.
1270 *
1271 * It doesn't matter if I_LOCK is not set initially, a call to
1272 * wake_up_inode() after removing from the hash list will DTRT.
1273 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 * This is called with inode_lock held.
1275 */
1276static void __wait_on_freeing_inode(struct inode *inode)
1277{
1278 wait_queue_head_t *wq;
1279 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1281 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1282 spin_unlock(&inode_lock);
1283 schedule();
1284 finish_wait(wq, &wait.wait);
1285 spin_lock(&inode_lock);
1286}
1287
1288void wake_up_inode(struct inode *inode)
1289{
1290 /*
1291 * Prevent speculative execution through spin_unlock(&inode_lock);
1292 */
1293 smp_mb();
1294 wake_up_bit(&inode->i_state, __I_LOCK);
1295}
1296
Mark Fasheh62752ee2006-10-17 10:31:38 +02001297/*
1298 * We rarely want to lock two inodes that do not have a parent/child
1299 * relationship (such as directory, child inode) simultaneously. The
1300 * vast majority of file systems should be able to get along fine
1301 * without this. Do not use these functions except as a last resort.
1302 */
1303void inode_double_lock(struct inode *inode1, struct inode *inode2)
1304{
1305 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1306 if (inode1)
1307 mutex_lock(&inode1->i_mutex);
1308 else if (inode2)
1309 mutex_lock(&inode2->i_mutex);
1310 return;
1311 }
1312
1313 if (inode1 < inode2) {
1314 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1315 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1316 } else {
1317 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1318 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1319 }
1320}
1321EXPORT_SYMBOL(inode_double_lock);
1322
1323void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1324{
1325 if (inode1)
1326 mutex_unlock(&inode1->i_mutex);
1327
1328 if (inode2 && inode2 != inode1)
1329 mutex_unlock(&inode2->i_mutex);
1330}
1331EXPORT_SYMBOL(inode_double_unlock);
1332
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333static __initdata unsigned long ihash_entries;
1334static int __init set_ihash_entries(char *str)
1335{
1336 if (!str)
1337 return 0;
1338 ihash_entries = simple_strtoul(str, &str, 0);
1339 return 1;
1340}
1341__setup("ihash_entries=", set_ihash_entries);
1342
1343/*
1344 * Initialize the waitqueues and inode hash table.
1345 */
1346void __init inode_init_early(void)
1347{
1348 int loop;
1349
1350 /* If hashes are distributed across NUMA nodes, defer
1351 * hash allocation until vmalloc space is available.
1352 */
1353 if (hashdist)
1354 return;
1355
1356 inode_hashtable =
1357 alloc_large_system_hash("Inode-cache",
1358 sizeof(struct hlist_head),
1359 ihash_entries,
1360 14,
1361 HASH_EARLY,
1362 &i_hash_shift,
1363 &i_hash_mask,
1364 0);
1365
1366 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1367 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1368}
1369
1370void __init inode_init(unsigned long mempages)
1371{
1372 int loop;
1373
1374 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001375 inode_cachep = kmem_cache_create("inode_cache",
1376 sizeof(struct inode),
1377 0,
1378 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1379 SLAB_MEM_SPREAD),
1380 init_once,
1381 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1383
1384 /* Hash may have been set up in inode_init_early */
1385 if (!hashdist)
1386 return;
1387
1388 inode_hashtable =
1389 alloc_large_system_hash("Inode-cache",
1390 sizeof(struct hlist_head),
1391 ihash_entries,
1392 14,
1393 0,
1394 &i_hash_shift,
1395 &i_hash_mask,
1396 0);
1397
1398 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1399 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1400}
1401
1402void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1403{
1404 inode->i_mode = mode;
1405 if (S_ISCHR(mode)) {
1406 inode->i_fop = &def_chr_fops;
1407 inode->i_rdev = rdev;
1408 } else if (S_ISBLK(mode)) {
1409 inode->i_fop = &def_blk_fops;
1410 inode->i_rdev = rdev;
1411 } else if (S_ISFIFO(mode))
1412 inode->i_fop = &def_fifo_fops;
1413 else if (S_ISSOCK(mode))
1414 inode->i_fop = &bad_sock_fops;
1415 else
1416 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1417 mode);
1418}
1419EXPORT_SYMBOL(init_special_inode);