blob: bf6adf122c688b337abf5d7c00288857b74a5927 [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
Peter Zijlstrad475fd42007-10-15 14:51:31 +0200145 spin_lock_init(&inode->i_lock);
146 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
147
148 mutex_init(&inode->i_mutex);
149 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
150
151 init_rwsem(&inode->i_alloc_sem);
152 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 mapping->a_ops = &empty_aops;
155 mapping->host = inode;
156 mapping->flags = 0;
Mel Gorman769848c2007-07-17 04:03:05 -0700157 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_PAGECACHE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 mapping->assoc_mapping = NULL;
159 mapping->backing_dev_info = &default_backing_dev_info;
160
161 /*
162 * If the block_device provides a backing_dev_info for client
163 * inodes then use that. Otherwise the inode share the bdev's
164 * backing_dev_info.
165 */
166 if (sb->s_bdev) {
167 struct backing_dev_info *bdi;
168
169 bdi = sb->s_bdev->bd_inode_backing_dev_info;
170 if (!bdi)
171 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
172 mapping->backing_dev_info = bdi;
173 }
Al Viroe6c6e642006-10-10 22:48:47 +0100174 inode->i_private = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 inode->i_mapping = mapping;
176 }
177 return inode;
178}
179
180void destroy_inode(struct inode *inode)
181{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200182 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 security_inode_free(inode);
184 if (inode->i_sb->s_op->destroy_inode)
185 inode->i_sb->s_op->destroy_inode(inode);
186 else
187 kmem_cache_free(inode_cachep, (inode));
188}
189
190
191/*
192 * These are initializations that only need to be done
193 * once, because the fields are idempotent across use
194 * of the inode, so let the slab aware of that.
195 */
196void inode_init_once(struct inode *inode)
197{
198 memset(inode, 0, sizeof(*inode));
199 INIT_HLIST_NODE(&inode->i_hash);
200 INIT_LIST_HEAD(&inode->i_dentry);
201 INIT_LIST_HEAD(&inode->i_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
203 rwlock_init(&inode->i_data.tree_lock);
204 spin_lock_init(&inode->i_data.i_mmap_lock);
205 INIT_LIST_HEAD(&inode->i_data.private_list);
206 spin_lock_init(&inode->i_data.private_lock);
207 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
208 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 i_size_ordered_init(inode);
Robert Love0eeca282005-07-12 17:06:03 -0400210#ifdef CONFIG_INOTIFY
211 INIT_LIST_HEAD(&inode->inotify_watches);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800212 mutex_init(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400213#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214}
215
216EXPORT_SYMBOL(inode_init_once);
217
Christoph Lametere18b8902006-12-06 20:33:20 -0800218static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
220 struct inode * inode = (struct inode *) foo;
221
Christoph Lametera35afb82007-05-16 22:10:57 -0700222 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
225/*
226 * inode_lock must be held
227 */
228void __iget(struct inode * inode)
229{
230 if (atomic_read(&inode->i_count)) {
231 atomic_inc(&inode->i_count);
232 return;
233 }
234 atomic_inc(&inode->i_count);
235 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
236 list_move(&inode->i_list, &inode_in_use);
237 inodes_stat.nr_unused--;
238}
239
240/**
241 * clear_inode - clear an inode
242 * @inode: inode to clear
243 *
244 * This is called by the filesystem to tell us
245 * that the inode is no longer useful. We just
246 * terminate it with extreme prejudice.
247 */
248void clear_inode(struct inode *inode)
249{
250 might_sleep();
251 invalidate_inode_buffers(inode);
252
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200253 BUG_ON(inode->i_data.nrpages);
254 BUG_ON(!(inode->i_state & I_FREEING));
255 BUG_ON(inode->i_state & I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 wait_on_inode(inode);
257 DQUOT_DROP(inode);
Christoph Hellwigacb0c852007-05-08 00:25:52 -0700258 if (inode->i_sb->s_op->clear_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 inode->i_sb->s_op->clear_inode(inode);
Theodore Ts'oeaf796e2006-09-27 01:50:48 -0700260 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 bd_forget(inode);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700262 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 cd_forget(inode);
264 inode->i_state = I_CLEAR;
265}
266
267EXPORT_SYMBOL(clear_inode);
268
269/*
270 * dispose_list - dispose of the contents of a local list
271 * @head: the head of the list to free
272 *
273 * Dispose-list gets a local list with local inodes in it, so it doesn't
274 * need to worry about list corruption and SMP locks.
275 */
276static void dispose_list(struct list_head *head)
277{
278 int nr_disposed = 0;
279
280 while (!list_empty(head)) {
281 struct inode *inode;
282
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700283 inode = list_first_entry(head, struct inode, i_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 list_del(&inode->i_list);
285
286 if (inode->i_data.nrpages)
287 truncate_inode_pages(&inode->i_data, 0);
288 clear_inode(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700289
290 spin_lock(&inode_lock);
291 hlist_del_init(&inode->i_hash);
292 list_del_init(&inode->i_sb_list);
293 spin_unlock(&inode_lock);
294
295 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 destroy_inode(inode);
297 nr_disposed++;
298 }
299 spin_lock(&inode_lock);
300 inodes_stat.nr_inodes -= nr_disposed;
301 spin_unlock(&inode_lock);
302}
303
304/*
305 * Invalidate all inodes for a device.
306 */
307static int invalidate_list(struct list_head *head, struct list_head *dispose)
308{
309 struct list_head *next;
310 int busy = 0, count = 0;
311
312 next = head->next;
313 for (;;) {
314 struct list_head * tmp = next;
315 struct inode * inode;
316
317 /*
318 * We can reschedule here without worrying about the list's
319 * consistency because the per-sb list of inodes must not
Ingo Molnarf24075b2006-03-23 03:00:34 -0800320 * change during umount anymore, and because iprune_mutex keeps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 * shrink_icache_memory() away.
322 */
323 cond_resched_lock(&inode_lock);
324
325 next = next->next;
326 if (tmp == head)
327 break;
328 inode = list_entry(tmp, struct inode, i_sb_list);
329 invalidate_inode_buffers(inode);
330 if (!atomic_read(&inode->i_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 list_move(&inode->i_list, dispose);
332 inode->i_state |= I_FREEING;
333 count++;
334 continue;
335 }
336 busy = 1;
337 }
338 /* only unused inodes may be cached with i_count zero */
339 inodes_stat.nr_unused -= count;
340 return busy;
341}
342
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343/**
344 * invalidate_inodes - discard the inodes on a device
345 * @sb: superblock
346 *
347 * Discard all of the inodes for a given superblock. If the discard
348 * fails because there are busy inodes then a non zero value is returned.
349 * If the discard is successful all the inodes have been discarded.
350 */
351int invalidate_inodes(struct super_block * sb)
352{
353 int busy;
354 LIST_HEAD(throw_away);
355
Ingo Molnarf24075b2006-03-23 03:00:34 -0800356 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 spin_lock(&inode_lock);
Robert Love0eeca282005-07-12 17:06:03 -0400358 inotify_unmount_inodes(&sb->s_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 busy = invalidate_list(&sb->s_inodes, &throw_away);
360 spin_unlock(&inode_lock);
361
362 dispose_list(&throw_away);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800363 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
365 return busy;
366}
367
368EXPORT_SYMBOL(invalidate_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369
370static int can_unuse(struct inode *inode)
371{
372 if (inode->i_state)
373 return 0;
374 if (inode_has_buffers(inode))
375 return 0;
376 if (atomic_read(&inode->i_count))
377 return 0;
378 if (inode->i_data.nrpages)
379 return 0;
380 return 1;
381}
382
383/*
384 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
385 * a temporary list and then are freed outside inode_lock by dispose_list().
386 *
387 * Any inodes which are pinned purely because of attached pagecache have their
388 * pagecache removed. We expect the final iput() on that inode to add it to
389 * the front of the inode_unused list. So look for it there and if the
390 * inode is still freeable, proceed. The right inode is found 99.9% of the
391 * time in testing on a 4-way.
392 *
393 * If the inode has metadata buffers attached to mapping->private_list then
394 * try to remove them.
395 */
396static void prune_icache(int nr_to_scan)
397{
398 LIST_HEAD(freeable);
399 int nr_pruned = 0;
400 int nr_scanned;
401 unsigned long reap = 0;
402
Ingo Molnarf24075b2006-03-23 03:00:34 -0800403 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 spin_lock(&inode_lock);
405 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
406 struct inode *inode;
407
408 if (list_empty(&inode_unused))
409 break;
410
411 inode = list_entry(inode_unused.prev, struct inode, i_list);
412
413 if (inode->i_state || atomic_read(&inode->i_count)) {
414 list_move(&inode->i_list, &inode_unused);
415 continue;
416 }
417 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
418 __iget(inode);
419 spin_unlock(&inode_lock);
420 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800421 reap += invalidate_mapping_pages(&inode->i_data,
422 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 iput(inode);
424 spin_lock(&inode_lock);
425
426 if (inode != list_entry(inode_unused.next,
427 struct inode, i_list))
428 continue; /* wrong inode or list_empty */
429 if (!can_unuse(inode))
430 continue;
431 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 list_move(&inode->i_list, &freeable);
433 inode->i_state |= I_FREEING;
434 nr_pruned++;
435 }
436 inodes_stat.nr_unused -= nr_pruned;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700437 if (current_is_kswapd())
438 __count_vm_events(KSWAPD_INODESTEAL, reap);
439 else
440 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 spin_unlock(&inode_lock);
442
443 dispose_list(&freeable);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800444 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445}
446
447/*
448 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
449 * "unused" means that no dentries are referring to the inodes: the files are
450 * not open and the dcache references to those inodes have already been
451 * reclaimed.
452 *
453 * This function is passed the number of inodes to scan, and it returns the
454 * total number of remaining possibly-reclaimable inodes.
455 */
Al Viro27496a82005-10-21 03:20:48 -0400456static int shrink_icache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
458 if (nr) {
459 /*
460 * Nasty deadlock avoidance. We may hold various FS locks,
461 * and we don't want to recurse into the FS that called us
462 * in clear_inode() and friends..
463 */
464 if (!(gfp_mask & __GFP_FS))
465 return -1;
466 prune_icache(nr);
467 }
468 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
469}
470
Rusty Russell8e1f9362007-07-17 04:03:17 -0700471static struct shrinker icache_shrinker = {
472 .shrink = shrink_icache_memory,
473 .seeks = DEFAULT_SEEKS,
474};
475
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476static void __wait_on_freeing_inode(struct inode *inode);
477/*
478 * Called with the inode lock held.
479 * NOTE: we are not increasing the inode-refcount, you must call __iget()
480 * by hand after calling find_inode now! This simplifies iunique and won't
481 * add any additional branch in the common code.
482 */
483static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
484{
485 struct hlist_node *node;
486 struct inode * inode = NULL;
487
488repeat:
489 hlist_for_each (node, head) {
490 inode = hlist_entry(node, struct inode, i_hash);
491 if (inode->i_sb != sb)
492 continue;
493 if (!test(inode, data))
494 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700495 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 __wait_on_freeing_inode(inode);
497 goto repeat;
498 }
499 break;
500 }
501 return node ? inode : NULL;
502}
503
504/*
505 * find_inode_fast is the fast path version of find_inode, see the comment at
506 * iget_locked for details.
507 */
508static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
509{
510 struct hlist_node *node;
511 struct inode * inode = NULL;
512
513repeat:
514 hlist_for_each (node, head) {
515 inode = hlist_entry(node, struct inode, i_hash);
516 if (inode->i_ino != ino)
517 continue;
518 if (inode->i_sb != sb)
519 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700520 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 __wait_on_freeing_inode(inode);
522 goto repeat;
523 }
524 break;
525 }
526 return node ? inode : NULL;
527}
528
529/**
530 * new_inode - obtain an inode
531 * @sb: superblock
532 *
Mel Gorman769848c2007-07-17 04:03:05 -0700533 * Allocates a new inode for given superblock. The default gfp_mask
534 * for allocations related to inode->i_mapping is GFP_HIGHUSER_PAGECACHE.
535 * If HIGHMEM pages are unsuitable or it is known that pages allocated
536 * for the page cache are not reclaimable or migratable,
537 * mapping_set_gfp_mask() must be called with suitable flags on the
538 * newly created inode's mapping
539 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 */
541struct inode *new_inode(struct super_block *sb)
542{
Jeff Layton866b04f2007-05-08 00:32:29 -0700543 /*
544 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
545 * error if st_ino won't fit in target struct field. Use 32bit counter
546 * here to attempt to avoid that.
547 */
548 static unsigned int last_ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 struct inode * inode;
550
551 spin_lock_prefetch(&inode_lock);
552
553 inode = alloc_inode(sb);
554 if (inode) {
555 spin_lock(&inode_lock);
556 inodes_stat.nr_inodes++;
557 list_add(&inode->i_list, &inode_in_use);
558 list_add(&inode->i_sb_list, &sb->s_inodes);
559 inode->i_ino = ++last_ino;
560 inode->i_state = 0;
561 spin_unlock(&inode_lock);
562 }
563 return inode;
564}
565
566EXPORT_SYMBOL(new_inode);
567
568void unlock_new_inode(struct inode *inode)
569{
570 /*
571 * This is special! We do not need the spinlock
572 * when clearing I_LOCK, because we're guaranteed
573 * that nobody else tries to do anything about the
574 * state of the inode when it is locked, as we
575 * just created it (so there can be no old holders
576 * that haven't tested I_LOCK).
577 */
578 inode->i_state &= ~(I_LOCK|I_NEW);
579 wake_up_inode(inode);
580}
581
582EXPORT_SYMBOL(unlock_new_inode);
583
584/*
585 * This is called without the inode lock held.. Be careful.
586 *
587 * We no longer cache the sb_flags in i_flags - see fs.h
588 * -- rmk@arm.uk.linux.org
589 */
590static 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)
591{
592 struct inode * inode;
593
594 inode = alloc_inode(sb);
595 if (inode) {
596 struct inode * old;
597
598 spin_lock(&inode_lock);
599 /* We released the lock, so.. */
600 old = find_inode(sb, head, test, data);
601 if (!old) {
602 if (set(inode, data))
603 goto set_failed;
604
605 inodes_stat.nr_inodes++;
606 list_add(&inode->i_list, &inode_in_use);
607 list_add(&inode->i_sb_list, &sb->s_inodes);
608 hlist_add_head(&inode->i_hash, head);
609 inode->i_state = I_LOCK|I_NEW;
610 spin_unlock(&inode_lock);
611
612 /* Return the locked inode with I_NEW set, the
613 * caller is responsible for filling in the contents
614 */
615 return inode;
616 }
617
618 /*
619 * Uhhuh, somebody else created the same inode under
620 * us. Use the old inode instead of the one we just
621 * allocated.
622 */
623 __iget(old);
624 spin_unlock(&inode_lock);
625 destroy_inode(inode);
626 inode = old;
627 wait_on_inode(inode);
628 }
629 return inode;
630
631set_failed:
632 spin_unlock(&inode_lock);
633 destroy_inode(inode);
634 return NULL;
635}
636
637/*
638 * get_new_inode_fast is the fast path version of get_new_inode, see the
639 * comment at iget_locked for details.
640 */
641static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
642{
643 struct inode * inode;
644
645 inode = alloc_inode(sb);
646 if (inode) {
647 struct inode * old;
648
649 spin_lock(&inode_lock);
650 /* We released the lock, so.. */
651 old = find_inode_fast(sb, head, ino);
652 if (!old) {
653 inode->i_ino = ino;
654 inodes_stat.nr_inodes++;
655 list_add(&inode->i_list, &inode_in_use);
656 list_add(&inode->i_sb_list, &sb->s_inodes);
657 hlist_add_head(&inode->i_hash, head);
658 inode->i_state = I_LOCK|I_NEW;
659 spin_unlock(&inode_lock);
660
661 /* Return the locked inode with I_NEW set, the
662 * caller is responsible for filling in the contents
663 */
664 return inode;
665 }
666
667 /*
668 * Uhhuh, somebody else created the same inode under
669 * us. Use the old inode instead of the one we just
670 * allocated.
671 */
672 __iget(old);
673 spin_unlock(&inode_lock);
674 destroy_inode(inode);
675 inode = old;
676 wait_on_inode(inode);
677 }
678 return inode;
679}
680
Andreas Mohred97bd32006-10-02 02:17:17 -0700681static unsigned long hash(struct super_block *sb, unsigned long hashval)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
683 unsigned long tmp;
684
685 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
686 L1_CACHE_BYTES;
687 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
688 return tmp & I_HASHMASK;
689}
690
691/**
692 * iunique - get a unique inode number
693 * @sb: superblock
694 * @max_reserved: highest reserved inode number
695 *
696 * Obtain an inode number that is unique on the system for a given
697 * superblock. This is used by file systems that have no natural
698 * permanent inode numbering system. An inode number is returned that
699 * is higher than the reserved limit but unique.
700 *
701 * BUGS:
702 * With a large number of inodes live on the file system this function
703 * currently becomes quite slow.
704 */
705ino_t iunique(struct super_block *sb, ino_t max_reserved)
706{
Jeff Layton866b04f2007-05-08 00:32:29 -0700707 /*
708 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
709 * error if st_ino won't fit in target struct field. Use 32bit counter
710 * here to attempt to avoid that.
711 */
712 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 struct inode *inode;
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700714 struct hlist_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700717 spin_lock(&inode_lock);
718 do {
719 if (counter <= max_reserved)
720 counter = max_reserved + 1;
721 res = counter++;
722 head = inode_hashtable + hash(sb, res);
723 inode = find_inode_fast(sb, head, res);
724 } while (inode != NULL);
725 spin_unlock(&inode_lock);
726
727 return res;
728}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729EXPORT_SYMBOL(iunique);
730
731struct inode *igrab(struct inode *inode)
732{
733 spin_lock(&inode_lock);
Jan Blunck4a3b0a42007-02-10 01:44:59 -0800734 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 __iget(inode);
736 else
737 /*
738 * Handle the case where s_op->clear_inode is not been
739 * called yet, and somebody is calling igrab
740 * while the inode is getting freed.
741 */
742 inode = NULL;
743 spin_unlock(&inode_lock);
744 return inode;
745}
746
747EXPORT_SYMBOL(igrab);
748
749/**
750 * ifind - internal function, you want ilookup5() or iget5().
751 * @sb: super block of file system to search
752 * @head: the head of the list to search
753 * @test: callback used for comparisons between inodes
754 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700755 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 *
757 * ifind() searches for the inode specified by @data in the inode
758 * cache. This is a generalized version of ifind_fast() for file systems where
759 * the inode number is not sufficient for unique identification of an inode.
760 *
761 * If the inode is in the cache, the inode is returned with an incremented
762 * reference count.
763 *
764 * Otherwise NULL is returned.
765 *
766 * Note, @test is called with the inode_lock held, so can't sleep.
767 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800768static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700770 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771{
772 struct inode *inode;
773
774 spin_lock(&inode_lock);
775 inode = find_inode(sb, head, test, data);
776 if (inode) {
777 __iget(inode);
778 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700779 if (likely(wait))
780 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 return inode;
782 }
783 spin_unlock(&inode_lock);
784 return NULL;
785}
786
787/**
788 * ifind_fast - internal function, you want ilookup() or iget().
789 * @sb: super block of file system to search
790 * @head: head of the list to search
791 * @ino: inode number to search for
792 *
793 * ifind_fast() searches for the inode @ino in the inode cache. This is for
794 * file systems where the inode number is sufficient for unique identification
795 * of an inode.
796 *
797 * If the inode is in the cache, the inode is returned with an incremented
798 * reference count.
799 *
800 * Otherwise NULL is returned.
801 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800802static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 struct hlist_head *head, unsigned long ino)
804{
805 struct inode *inode;
806
807 spin_lock(&inode_lock);
808 inode = find_inode_fast(sb, head, ino);
809 if (inode) {
810 __iget(inode);
811 spin_unlock(&inode_lock);
812 wait_on_inode(inode);
813 return inode;
814 }
815 spin_unlock(&inode_lock);
816 return NULL;
817}
818
819/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700820 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 * @sb: super block of file system to search
822 * @hashval: hash value (usually inode number) to search for
823 * @test: callback used for comparisons between inodes
824 * @data: opaque data pointer to pass to @test
825 *
826 * ilookup5() uses ifind() to search for the inode specified by @hashval and
827 * @data in the inode cache. This is a generalized version of ilookup() for
828 * file systems where the inode number is not sufficient for unique
829 * identification of an inode.
830 *
831 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700832 * reference count. Note, the inode lock is not waited upon so you have to be
833 * very careful what you do with the returned inode. You probably should be
834 * using ilookup5() instead.
835 *
836 * Otherwise NULL is returned.
837 *
838 * Note, @test is called with the inode_lock held, so can't sleep.
839 */
840struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
841 int (*test)(struct inode *, void *), void *data)
842{
843 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
844
845 return ifind(sb, head, test, data, 0);
846}
847
848EXPORT_SYMBOL(ilookup5_nowait);
849
850/**
851 * ilookup5 - search for an inode in the inode cache
852 * @sb: super block of file system to search
853 * @hashval: hash value (usually inode number) to search for
854 * @test: callback used for comparisons between inodes
855 * @data: opaque data pointer to pass to @test
856 *
857 * ilookup5() uses ifind() to search for the inode specified by @hashval and
858 * @data in the inode cache. This is a generalized version of ilookup() for
859 * file systems where the inode number is not sufficient for unique
860 * identification of an inode.
861 *
862 * If the inode is in the cache, the inode lock is waited upon and the inode is
863 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 *
865 * Otherwise NULL is returned.
866 *
867 * Note, @test is called with the inode_lock held, so can't sleep.
868 */
869struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
870 int (*test)(struct inode *, void *), void *data)
871{
872 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
873
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700874 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
877EXPORT_SYMBOL(ilookup5);
878
879/**
880 * ilookup - search for an inode in the inode cache
881 * @sb: super block of file system to search
882 * @ino: inode number to search for
883 *
884 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
885 * This is for file systems where the inode number is sufficient for unique
886 * identification of an inode.
887 *
888 * If the inode is in the cache, the inode is returned with an incremented
889 * reference count.
890 *
891 * Otherwise NULL is returned.
892 */
893struct inode *ilookup(struct super_block *sb, unsigned long ino)
894{
895 struct hlist_head *head = inode_hashtable + hash(sb, ino);
896
897 return ifind_fast(sb, head, ino);
898}
899
900EXPORT_SYMBOL(ilookup);
901
902/**
903 * iget5_locked - obtain an inode from a mounted file system
904 * @sb: super block of file system
905 * @hashval: hash value (usually inode number) to get
906 * @test: callback used for comparisons between inodes
907 * @set: callback used to initialize a new struct inode
908 * @data: opaque data pointer to pass to @test and @set
909 *
910 * This is iget() without the read_inode() portion of get_new_inode().
911 *
912 * iget5_locked() uses ifind() to search for the inode specified by @hashval
913 * and @data in the inode cache and if present it is returned with an increased
914 * reference count. This is a generalized version of iget_locked() for file
915 * systems where the inode number is not sufficient for unique identification
916 * of an inode.
917 *
918 * If the inode is not in cache, get_new_inode() is called to allocate a new
919 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
920 * file system gets to fill it in before unlocking it via unlock_new_inode().
921 *
922 * Note both @test and @set are called with the inode_lock held, so can't sleep.
923 */
924struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
925 int (*test)(struct inode *, void *),
926 int (*set)(struct inode *, void *), void *data)
927{
928 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
929 struct inode *inode;
930
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700931 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 if (inode)
933 return inode;
934 /*
935 * get_new_inode() will do the right thing, re-trying the search
936 * in case it had to block at any point.
937 */
938 return get_new_inode(sb, head, test, set, data);
939}
940
941EXPORT_SYMBOL(iget5_locked);
942
943/**
944 * iget_locked - obtain an inode from a mounted file system
945 * @sb: super block of file system
946 * @ino: inode number to get
947 *
948 * This is iget() without the read_inode() portion of get_new_inode_fast().
949 *
950 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
951 * the inode cache and if present it is returned with an increased reference
952 * count. This is for file systems where the inode number is sufficient for
953 * unique identification of an inode.
954 *
955 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
956 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
957 * The file system gets to fill it in before unlocking it via
958 * unlock_new_inode().
959 */
960struct inode *iget_locked(struct super_block *sb, unsigned long ino)
961{
962 struct hlist_head *head = inode_hashtable + hash(sb, ino);
963 struct inode *inode;
964
965 inode = ifind_fast(sb, head, ino);
966 if (inode)
967 return inode;
968 /*
969 * get_new_inode_fast() will do the right thing, re-trying the search
970 * in case it had to block at any point.
971 */
972 return get_new_inode_fast(sb, head, ino);
973}
974
975EXPORT_SYMBOL(iget_locked);
976
977/**
978 * __insert_inode_hash - hash an inode
979 * @inode: unhashed inode
980 * @hashval: unsigned long value used to locate this object in the
981 * inode_hashtable.
982 *
983 * Add an inode to the inode hash for this superblock.
984 */
985void __insert_inode_hash(struct inode *inode, unsigned long hashval)
986{
987 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
988 spin_lock(&inode_lock);
989 hlist_add_head(&inode->i_hash, head);
990 spin_unlock(&inode_lock);
991}
992
993EXPORT_SYMBOL(__insert_inode_hash);
994
995/**
996 * remove_inode_hash - remove an inode from the hash
997 * @inode: inode to unhash
998 *
999 * Remove an inode from the superblock.
1000 */
1001void remove_inode_hash(struct inode *inode)
1002{
1003 spin_lock(&inode_lock);
1004 hlist_del_init(&inode->i_hash);
1005 spin_unlock(&inode_lock);
1006}
1007
1008EXPORT_SYMBOL(remove_inode_hash);
1009
1010/*
1011 * Tell the filesystem that this inode is no longer of any interest and should
1012 * be completely destroyed.
1013 *
1014 * We leave the inode in the inode hash table until *after* the filesystem's
1015 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1016 * instigate) will always find up-to-date information either in the hash or on
1017 * disk.
1018 *
1019 * I_FREEING is set so that no-one will take a new reference to the inode while
1020 * it is being deleted.
1021 */
1022void generic_delete_inode(struct inode *inode)
1023{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001024 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
1026 list_del_init(&inode->i_list);
1027 list_del_init(&inode->i_sb_list);
Andreas Mohred97bd32006-10-02 02:17:17 -07001028 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 inodes_stat.nr_inodes--;
1030 spin_unlock(&inode_lock);
1031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 security_inode_delete(inode);
1033
1034 if (op->delete_inode) {
1035 void (*delete)(struct inode *) = op->delete_inode;
1036 if (!is_bad_inode(inode))
1037 DQUOT_INIT(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001038 /* Filesystems implementing their own
1039 * s_op->delete_inode are required to call
1040 * truncate_inode_pages and clear_inode()
1041 * internally */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 delete(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001043 } else {
1044 truncate_inode_pages(&inode->i_data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 clear_inode(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001046 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 spin_lock(&inode_lock);
1048 hlist_del_init(&inode->i_hash);
1049 spin_unlock(&inode_lock);
1050 wake_up_inode(inode);
Eric Sesterhennb7542f82006-04-02 13:38:18 +02001051 BUG_ON(inode->i_state != I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 destroy_inode(inode);
1053}
1054
1055EXPORT_SYMBOL(generic_delete_inode);
1056
1057static void generic_forget_inode(struct inode *inode)
1058{
1059 struct super_block *sb = inode->i_sb;
1060
1061 if (!hlist_unhashed(&inode->i_hash)) {
1062 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1063 list_move(&inode->i_list, &inode_unused);
1064 inodes_stat.nr_unused++;
Christoph Hellwigacb0c852007-05-08 00:25:52 -07001065 if (sb->s_flags & MS_ACTIVE) {
Alexander Viro991114c2005-06-23 00:09:01 -07001066 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001068 }
1069 inode->i_state |= I_WILL_FREE;
1070 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 write_inode_now(inode, 1);
1072 spin_lock(&inode_lock);
Alexander Viro991114c2005-06-23 00:09:01 -07001073 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 inodes_stat.nr_unused--;
1075 hlist_del_init(&inode->i_hash);
1076 }
1077 list_del_init(&inode->i_list);
1078 list_del_init(&inode->i_sb_list);
Alexander Viro991114c2005-06-23 00:09:01 -07001079 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 inodes_stat.nr_inodes--;
1081 spin_unlock(&inode_lock);
1082 if (inode->i_data.nrpages)
1083 truncate_inode_pages(&inode->i_data, 0);
1084 clear_inode(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001085 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 destroy_inode(inode);
1087}
1088
1089/*
1090 * Normal UNIX filesystem behaviour: delete the
1091 * inode when the usage count drops to zero, and
1092 * i_nlink is zero.
1093 */
Mark Fashehcb2c0232005-07-07 17:56:03 -07001094void generic_drop_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095{
1096 if (!inode->i_nlink)
1097 generic_delete_inode(inode);
1098 else
1099 generic_forget_inode(inode);
1100}
1101
Mark Fashehcb2c0232005-07-07 17:56:03 -07001102EXPORT_SYMBOL_GPL(generic_drop_inode);
1103
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104/*
1105 * Called when we're dropping the last reference
1106 * to an inode.
1107 *
1108 * Call the FS "drop()" function, defaulting to
1109 * the legacy UNIX filesystem behaviour..
1110 *
1111 * NOTE! NOTE! NOTE! We're called with the inode lock
1112 * held, and the drop function is supposed to release
1113 * the lock!
1114 */
1115static inline void iput_final(struct inode *inode)
1116{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001117 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 void (*drop)(struct inode *) = generic_drop_inode;
1119
1120 if (op && op->drop_inode)
1121 drop = op->drop_inode;
1122 drop(inode);
1123}
1124
1125/**
1126 * iput - put an inode
1127 * @inode: inode to put
1128 *
1129 * Puts an inode, dropping its usage count. If the inode use count hits
1130 * zero, the inode is then freed and may also be destroyed.
1131 *
1132 * Consequently, iput() can sleep.
1133 */
1134void iput(struct inode *inode)
1135{
1136 if (inode) {
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001137 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
1139 BUG_ON(inode->i_state == I_CLEAR);
1140
1141 if (op && op->put_inode)
1142 op->put_inode(inode);
1143
1144 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1145 iput_final(inode);
1146 }
1147}
1148
1149EXPORT_SYMBOL(iput);
1150
1151/**
1152 * bmap - find a block number in a file
1153 * @inode: inode of file
1154 * @block: block to find
1155 *
1156 * Returns the block number on the device holding the inode that
1157 * is the disk block number for the block of the file requested.
1158 * That is, asked for block 4 of inode 1 the function will return the
1159 * disk block relative to the disk start that holds that block of the
1160 * file.
1161 */
1162sector_t bmap(struct inode * inode, sector_t block)
1163{
1164 sector_t res = 0;
1165 if (inode->i_mapping->a_ops->bmap)
1166 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1167 return res;
1168}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169EXPORT_SYMBOL(bmap);
1170
1171/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001172 * touch_atime - update the access time
1173 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001174 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 *
1176 * Update the accessed time on an inode and mark it for writeback.
1177 * This function automatically handles read only file systems and media,
1178 * as well as the "noatime" flag and inode specific "noatime" markers.
1179 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001180void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001182 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 struct timespec now;
1184
Andrew Mortonb2276132006-12-13 00:34:33 -08001185 if (inode->i_flags & S_NOATIME)
1186 return;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001187 if (IS_NOATIME(inode))
Andrew Mortonb2276132006-12-13 00:34:33 -08001188 return;
1189 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001190 return;
1191
1192 /*
1193 * We may have a NULL vfsmount when coming from NFSD
1194 */
Andrew Mortonb2276132006-12-13 00:34:33 -08001195 if (mnt) {
1196 if (mnt->mnt_flags & MNT_NOATIME)
1197 return;
1198 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1199 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001200
1201 if (mnt->mnt_flags & MNT_RELATIME) {
1202 /*
1203 * With relative atime, only update atime if the
1204 * previous atime is earlier than either the ctime or
1205 * mtime.
1206 */
1207 if (timespec_compare(&inode->i_mtime,
1208 &inode->i_atime) < 0 &&
1209 timespec_compare(&inode->i_ctime,
1210 &inode->i_atime) < 0)
1211 return;
1212 }
Andrew Mortonb2276132006-12-13 00:34:33 -08001213 }
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 now = current_fs_time(inode->i_sb);
Valerie Henson47ae32d2006-12-13 00:34:34 -08001216 if (timespec_equal(&inode->i_atime, &now))
1217 return;
1218
1219 inode->i_atime = now;
1220 mark_inode_dirty_sync(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001222EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223
1224/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001225 * file_update_time - update mtime and ctime time
1226 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001228 * Update the mtime and ctime members of an inode and mark the inode
1229 * for writeback. Note that this function is meant exclusively for
1230 * usage in the file write path of filesystems, and filesystems may
1231 * choose to explicitly ignore update via this function with the
1232 * S_NOCTIME inode flag, e.g. for network filesystem where these
1233 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 */
1235
Christoph Hellwig870f4812006-01-09 20:52:01 -08001236void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001238 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 struct timespec now;
1240 int sync_it = 0;
1241
1242 if (IS_NOCMTIME(inode))
1243 return;
1244 if (IS_RDONLY(inode))
1245 return;
1246
1247 now = current_fs_time(inode->i_sb);
Andreas Mohred97bd32006-10-02 02:17:17 -07001248 if (!timespec_equal(&inode->i_mtime, &now)) {
1249 inode->i_mtime = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001251 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
Andreas Mohred97bd32006-10-02 02:17:17 -07001253 if (!timespec_equal(&inode->i_ctime, &now)) {
1254 inode->i_ctime = now;
Christoph Hellwig870f4812006-01-09 20:52:01 -08001255 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001256 }
Christoph Hellwig870f4812006-01-09 20:52:01 -08001257
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 if (sync_it)
1259 mark_inode_dirty_sync(inode);
1260}
1261
Christoph Hellwig870f4812006-01-09 20:52:01 -08001262EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
1264int inode_needs_sync(struct inode *inode)
1265{
1266 if (IS_SYNC(inode))
1267 return 1;
1268 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1269 return 1;
1270 return 0;
1271}
1272
1273EXPORT_SYMBOL(inode_needs_sync);
1274
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275int inode_wait(void *word)
1276{
1277 schedule();
1278 return 0;
1279}
1280
1281/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001282 * If we try to find an inode in the inode hash while it is being
1283 * deleted, we have to wait until the filesystem completes its
1284 * deletion before reporting that it isn't found. This function waits
1285 * until the deletion _might_ have completed. Callers are responsible
1286 * to recheck inode state.
1287 *
1288 * It doesn't matter if I_LOCK is not set initially, a call to
1289 * wake_up_inode() after removing from the hash list will DTRT.
1290 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 * This is called with inode_lock held.
1292 */
1293static void __wait_on_freeing_inode(struct inode *inode)
1294{
1295 wait_queue_head_t *wq;
1296 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1298 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1299 spin_unlock(&inode_lock);
1300 schedule();
1301 finish_wait(wq, &wait.wait);
1302 spin_lock(&inode_lock);
1303}
1304
1305void wake_up_inode(struct inode *inode)
1306{
1307 /*
1308 * Prevent speculative execution through spin_unlock(&inode_lock);
1309 */
1310 smp_mb();
1311 wake_up_bit(&inode->i_state, __I_LOCK);
1312}
1313
Mark Fasheh62752ee2006-10-17 10:31:38 +02001314/*
1315 * We rarely want to lock two inodes that do not have a parent/child
1316 * relationship (such as directory, child inode) simultaneously. The
1317 * vast majority of file systems should be able to get along fine
1318 * without this. Do not use these functions except as a last resort.
1319 */
1320void inode_double_lock(struct inode *inode1, struct inode *inode2)
1321{
1322 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1323 if (inode1)
1324 mutex_lock(&inode1->i_mutex);
1325 else if (inode2)
1326 mutex_lock(&inode2->i_mutex);
1327 return;
1328 }
1329
1330 if (inode1 < inode2) {
1331 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1332 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1333 } else {
1334 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1335 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1336 }
1337}
1338EXPORT_SYMBOL(inode_double_lock);
1339
1340void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1341{
1342 if (inode1)
1343 mutex_unlock(&inode1->i_mutex);
1344
1345 if (inode2 && inode2 != inode1)
1346 mutex_unlock(&inode2->i_mutex);
1347}
1348EXPORT_SYMBOL(inode_double_unlock);
1349
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350static __initdata unsigned long ihash_entries;
1351static int __init set_ihash_entries(char *str)
1352{
1353 if (!str)
1354 return 0;
1355 ihash_entries = simple_strtoul(str, &str, 0);
1356 return 1;
1357}
1358__setup("ihash_entries=", set_ihash_entries);
1359
1360/*
1361 * Initialize the waitqueues and inode hash table.
1362 */
1363void __init inode_init_early(void)
1364{
1365 int loop;
1366
1367 /* If hashes are distributed across NUMA nodes, defer
1368 * hash allocation until vmalloc space is available.
1369 */
1370 if (hashdist)
1371 return;
1372
1373 inode_hashtable =
1374 alloc_large_system_hash("Inode-cache",
1375 sizeof(struct hlist_head),
1376 ihash_entries,
1377 14,
1378 HASH_EARLY,
1379 &i_hash_shift,
1380 &i_hash_mask,
1381 0);
1382
1383 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1384 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1385}
1386
1387void __init inode_init(unsigned long mempages)
1388{
1389 int loop;
1390
1391 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001392 inode_cachep = kmem_cache_create("inode_cache",
1393 sizeof(struct inode),
1394 0,
1395 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1396 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001397 init_once);
Rusty Russell8e1f9362007-07-17 04:03:17 -07001398 register_shrinker(&icache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
1400 /* Hash may have been set up in inode_init_early */
1401 if (!hashdist)
1402 return;
1403
1404 inode_hashtable =
1405 alloc_large_system_hash("Inode-cache",
1406 sizeof(struct hlist_head),
1407 ihash_entries,
1408 14,
1409 0,
1410 &i_hash_shift,
1411 &i_hash_mask,
1412 0);
1413
1414 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1415 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1416}
1417
1418void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1419{
1420 inode->i_mode = mode;
1421 if (S_ISCHR(mode)) {
1422 inode->i_fop = &def_chr_fops;
1423 inode->i_rdev = rdev;
1424 } else if (S_ISBLK(mode)) {
1425 inode->i_fop = &def_blk_fops;
1426 inode->i_rdev = rdev;
1427 } else if (S_ISFIFO(mode))
1428 inode->i_fop = &def_fifo_fops;
1429 else if (S_ISSOCK(mode))
1430 inode->i_fop = &bad_sock_fops;
1431 else
1432 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1433 mode);
1434}
1435EXPORT_SYMBOL(init_special_inode);