blob: c36d9480335c6fc8a099aa1551910b0ce55a539f [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
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700102static void wake_up_inode(struct inode *inode)
103{
104 /*
105 * Prevent speculative execution through spin_unlock(&inode_lock);
106 */
107 smp_mb();
108 wake_up_bit(&inode->i_state, __I_LOCK);
109}
110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111static struct inode *alloc_inode(struct super_block *sb)
112{
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700113 static const struct address_space_operations empty_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 static struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800115 static const struct file_operations empty_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 struct inode *inode;
117
118 if (sb->s_op->alloc_inode)
119 inode = sb->s_op->alloc_inode(sb);
120 else
Christoph Lametere94b1762006-12-06 20:33:17 -0800121 inode = (struct inode *) kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
123 if (inode) {
124 struct address_space * const mapping = &inode->i_data;
125
126 inode->i_sb = sb;
127 inode->i_blkbits = sb->s_blocksize_bits;
128 inode->i_flags = 0;
129 atomic_set(&inode->i_count, 1);
130 inode->i_op = &empty_iops;
131 inode->i_fop = &empty_fops;
132 inode->i_nlink = 1;
133 atomic_set(&inode->i_writecount, 0);
134 inode->i_size = 0;
135 inode->i_blocks = 0;
136 inode->i_bytes = 0;
137 inode->i_generation = 0;
138#ifdef CONFIG_QUOTA
139 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
140#endif
141 inode->i_pipe = NULL;
142 inode->i_bdev = NULL;
143 inode->i_cdev = NULL;
144 inode->i_rdev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 inode->dirtied_when = 0;
146 if (security_inode_alloc(inode)) {
147 if (inode->i_sb->s_op->destroy_inode)
148 inode->i_sb->s_op->destroy_inode(inode);
149 else
150 kmem_cache_free(inode_cachep, (inode));
151 return NULL;
152 }
153
Peter Zijlstrad475fd42007-10-15 14:51:31 +0200154 spin_lock_init(&inode->i_lock);
155 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
156
157 mutex_init(&inode->i_mutex);
158 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
159
160 init_rwsem(&inode->i_alloc_sem);
161 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 mapping->a_ops = &empty_aops;
164 mapping->host = inode;
165 mapping->flags = 0;
Mel Gorman769848c2007-07-17 04:03:05 -0700166 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_PAGECACHE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 mapping->assoc_mapping = NULL;
168 mapping->backing_dev_info = &default_backing_dev_info;
169
170 /*
171 * If the block_device provides a backing_dev_info for client
172 * inodes then use that. Otherwise the inode share the bdev's
173 * backing_dev_info.
174 */
175 if (sb->s_bdev) {
176 struct backing_dev_info *bdi;
177
178 bdi = sb->s_bdev->bd_inode_backing_dev_info;
179 if (!bdi)
180 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
181 mapping->backing_dev_info = bdi;
182 }
Al Viroe6c6e642006-10-10 22:48:47 +0100183 inode->i_private = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 inode->i_mapping = mapping;
185 }
186 return inode;
187}
188
189void destroy_inode(struct inode *inode)
190{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200191 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 security_inode_free(inode);
193 if (inode->i_sb->s_op->destroy_inode)
194 inode->i_sb->s_op->destroy_inode(inode);
195 else
196 kmem_cache_free(inode_cachep, (inode));
197}
198
199
200/*
201 * These are initializations that only need to be done
202 * once, because the fields are idempotent across use
203 * of the inode, so let the slab aware of that.
204 */
205void inode_init_once(struct inode *inode)
206{
207 memset(inode, 0, sizeof(*inode));
208 INIT_HLIST_NODE(&inode->i_hash);
209 INIT_LIST_HEAD(&inode->i_dentry);
210 INIT_LIST_HEAD(&inode->i_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
212 rwlock_init(&inode->i_data.tree_lock);
213 spin_lock_init(&inode->i_data.i_mmap_lock);
214 INIT_LIST_HEAD(&inode->i_data.private_list);
215 spin_lock_init(&inode->i_data.private_lock);
216 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
217 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 i_size_ordered_init(inode);
Robert Love0eeca282005-07-12 17:06:03 -0400219#ifdef CONFIG_INOTIFY
220 INIT_LIST_HEAD(&inode->inotify_watches);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800221 mutex_init(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400222#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
225EXPORT_SYMBOL(inode_init_once);
226
Christoph Lameter4ba9b9d2007-10-16 23:25:51 -0700227static void init_once(struct kmem_cache * cachep, void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228{
229 struct inode * inode = (struct inode *) foo;
230
Christoph Lametera35afb82007-05-16 22:10:57 -0700231 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232}
233
234/*
235 * inode_lock must be held
236 */
237void __iget(struct inode * inode)
238{
239 if (atomic_read(&inode->i_count)) {
240 atomic_inc(&inode->i_count);
241 return;
242 }
243 atomic_inc(&inode->i_count);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700244 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 list_move(&inode->i_list, &inode_in_use);
246 inodes_stat.nr_unused--;
247}
248
249/**
250 * clear_inode - clear an inode
251 * @inode: inode to clear
252 *
253 * This is called by the filesystem to tell us
254 * that the inode is no longer useful. We just
255 * terminate it with extreme prejudice.
256 */
257void clear_inode(struct inode *inode)
258{
259 might_sleep();
260 invalidate_inode_buffers(inode);
261
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200262 BUG_ON(inode->i_data.nrpages);
263 BUG_ON(!(inode->i_state & I_FREEING));
264 BUG_ON(inode->i_state & I_CLEAR);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700265 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 DQUOT_DROP(inode);
Christoph Hellwigacb0c852007-05-08 00:25:52 -0700267 if (inode->i_sb->s_op->clear_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 inode->i_sb->s_op->clear_inode(inode);
Theodore Ts'oeaf796e2006-09-27 01:50:48 -0700269 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 bd_forget(inode);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700271 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 cd_forget(inode);
273 inode->i_state = I_CLEAR;
274}
275
276EXPORT_SYMBOL(clear_inode);
277
278/*
279 * dispose_list - dispose of the contents of a local list
280 * @head: the head of the list to free
281 *
282 * Dispose-list gets a local list with local inodes in it, so it doesn't
283 * need to worry about list corruption and SMP locks.
284 */
285static void dispose_list(struct list_head *head)
286{
287 int nr_disposed = 0;
288
289 while (!list_empty(head)) {
290 struct inode *inode;
291
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700292 inode = list_first_entry(head, struct inode, i_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 list_del(&inode->i_list);
294
295 if (inode->i_data.nrpages)
296 truncate_inode_pages(&inode->i_data, 0);
297 clear_inode(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700298
299 spin_lock(&inode_lock);
300 hlist_del_init(&inode->i_hash);
301 list_del_init(&inode->i_sb_list);
302 spin_unlock(&inode_lock);
303
304 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 destroy_inode(inode);
306 nr_disposed++;
307 }
308 spin_lock(&inode_lock);
309 inodes_stat.nr_inodes -= nr_disposed;
310 spin_unlock(&inode_lock);
311}
312
313/*
314 * Invalidate all inodes for a device.
315 */
316static int invalidate_list(struct list_head *head, struct list_head *dispose)
317{
318 struct list_head *next;
319 int busy = 0, count = 0;
320
321 next = head->next;
322 for (;;) {
323 struct list_head * tmp = next;
324 struct inode * inode;
325
326 /*
327 * We can reschedule here without worrying about the list's
328 * consistency because the per-sb list of inodes must not
Ingo Molnarf24075b2006-03-23 03:00:34 -0800329 * change during umount anymore, and because iprune_mutex keeps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 * shrink_icache_memory() away.
331 */
332 cond_resched_lock(&inode_lock);
333
334 next = next->next;
335 if (tmp == head)
336 break;
337 inode = list_entry(tmp, struct inode, i_sb_list);
338 invalidate_inode_buffers(inode);
339 if (!atomic_read(&inode->i_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 list_move(&inode->i_list, dispose);
341 inode->i_state |= I_FREEING;
342 count++;
343 continue;
344 }
345 busy = 1;
346 }
347 /* only unused inodes may be cached with i_count zero */
348 inodes_stat.nr_unused -= count;
349 return busy;
350}
351
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352/**
353 * invalidate_inodes - discard the inodes on a device
354 * @sb: superblock
355 *
356 * Discard all of the inodes for a given superblock. If the discard
357 * fails because there are busy inodes then a non zero value is returned.
358 * If the discard is successful all the inodes have been discarded.
359 */
360int invalidate_inodes(struct super_block * sb)
361{
362 int busy;
363 LIST_HEAD(throw_away);
364
Ingo Molnarf24075b2006-03-23 03:00:34 -0800365 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 spin_lock(&inode_lock);
Robert Love0eeca282005-07-12 17:06:03 -0400367 inotify_unmount_inodes(&sb->s_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 busy = invalidate_list(&sb->s_inodes, &throw_away);
369 spin_unlock(&inode_lock);
370
371 dispose_list(&throw_away);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800372 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
374 return busy;
375}
376
377EXPORT_SYMBOL(invalidate_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
379static int can_unuse(struct inode *inode)
380{
381 if (inode->i_state)
382 return 0;
383 if (inode_has_buffers(inode))
384 return 0;
385 if (atomic_read(&inode->i_count))
386 return 0;
387 if (inode->i_data.nrpages)
388 return 0;
389 return 1;
390}
391
392/*
393 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
394 * a temporary list and then are freed outside inode_lock by dispose_list().
395 *
396 * Any inodes which are pinned purely because of attached pagecache have their
397 * pagecache removed. We expect the final iput() on that inode to add it to
398 * the front of the inode_unused list. So look for it there and if the
399 * inode is still freeable, proceed. The right inode is found 99.9% of the
400 * time in testing on a 4-way.
401 *
402 * If the inode has metadata buffers attached to mapping->private_list then
403 * try to remove them.
404 */
405static void prune_icache(int nr_to_scan)
406{
407 LIST_HEAD(freeable);
408 int nr_pruned = 0;
409 int nr_scanned;
410 unsigned long reap = 0;
411
Ingo Molnarf24075b2006-03-23 03:00:34 -0800412 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 spin_lock(&inode_lock);
414 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
415 struct inode *inode;
416
417 if (list_empty(&inode_unused))
418 break;
419
420 inode = list_entry(inode_unused.prev, struct inode, i_list);
421
422 if (inode->i_state || atomic_read(&inode->i_count)) {
423 list_move(&inode->i_list, &inode_unused);
424 continue;
425 }
426 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
427 __iget(inode);
428 spin_unlock(&inode_lock);
429 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800430 reap += invalidate_mapping_pages(&inode->i_data,
431 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 iput(inode);
433 spin_lock(&inode_lock);
434
435 if (inode != list_entry(inode_unused.next,
436 struct inode, i_list))
437 continue; /* wrong inode or list_empty */
438 if (!can_unuse(inode))
439 continue;
440 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 list_move(&inode->i_list, &freeable);
442 inode->i_state |= I_FREEING;
443 nr_pruned++;
444 }
445 inodes_stat.nr_unused -= nr_pruned;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700446 if (current_is_kswapd())
447 __count_vm_events(KSWAPD_INODESTEAL, reap);
448 else
449 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 spin_unlock(&inode_lock);
451
452 dispose_list(&freeable);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800453 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454}
455
456/*
457 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
458 * "unused" means that no dentries are referring to the inodes: the files are
459 * not open and the dcache references to those inodes have already been
460 * reclaimed.
461 *
462 * This function is passed the number of inodes to scan, and it returns the
463 * total number of remaining possibly-reclaimable inodes.
464 */
Al Viro27496a82005-10-21 03:20:48 -0400465static int shrink_icache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466{
467 if (nr) {
468 /*
469 * Nasty deadlock avoidance. We may hold various FS locks,
470 * and we don't want to recurse into the FS that called us
471 * in clear_inode() and friends..
472 */
473 if (!(gfp_mask & __GFP_FS))
474 return -1;
475 prune_icache(nr);
476 }
477 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
478}
479
Rusty Russell8e1f9362007-07-17 04:03:17 -0700480static struct shrinker icache_shrinker = {
481 .shrink = shrink_icache_memory,
482 .seeks = DEFAULT_SEEKS,
483};
484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485static void __wait_on_freeing_inode(struct inode *inode);
486/*
487 * Called with the inode lock held.
488 * NOTE: we are not increasing the inode-refcount, you must call __iget()
489 * by hand after calling find_inode now! This simplifies iunique and won't
490 * add any additional branch in the common code.
491 */
492static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
493{
494 struct hlist_node *node;
495 struct inode * inode = NULL;
496
497repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700498 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 if (inode->i_sb != sb)
500 continue;
501 if (!test(inode, data))
502 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700503 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 __wait_on_freeing_inode(inode);
505 goto repeat;
506 }
507 break;
508 }
509 return node ? inode : NULL;
510}
511
512/*
513 * find_inode_fast is the fast path version of find_inode, see the comment at
514 * iget_locked for details.
515 */
516static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
517{
518 struct hlist_node *node;
519 struct inode * inode = NULL;
520
521repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700522 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 if (inode->i_ino != ino)
524 continue;
525 if (inode->i_sb != sb)
526 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700527 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 __wait_on_freeing_inode(inode);
529 goto repeat;
530 }
531 break;
532 }
533 return node ? inode : NULL;
534}
535
536/**
537 * new_inode - obtain an inode
538 * @sb: superblock
539 *
Mel Gorman769848c2007-07-17 04:03:05 -0700540 * Allocates a new inode for given superblock. The default gfp_mask
541 * for allocations related to inode->i_mapping is GFP_HIGHUSER_PAGECACHE.
542 * If HIGHMEM pages are unsuitable or it is known that pages allocated
543 * for the page cache are not reclaimable or migratable,
544 * mapping_set_gfp_mask() must be called with suitable flags on the
545 * newly created inode's mapping
546 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 */
548struct inode *new_inode(struct super_block *sb)
549{
Jeff Layton866b04f2007-05-08 00:32:29 -0700550 /*
551 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
552 * error if st_ino won't fit in target struct field. Use 32bit counter
553 * here to attempt to avoid that.
554 */
555 static unsigned int last_ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 struct inode * inode;
557
558 spin_lock_prefetch(&inode_lock);
559
560 inode = alloc_inode(sb);
561 if (inode) {
562 spin_lock(&inode_lock);
563 inodes_stat.nr_inodes++;
564 list_add(&inode->i_list, &inode_in_use);
565 list_add(&inode->i_sb_list, &sb->s_inodes);
566 inode->i_ino = ++last_ino;
567 inode->i_state = 0;
568 spin_unlock(&inode_lock);
569 }
570 return inode;
571}
572
573EXPORT_SYMBOL(new_inode);
574
575void unlock_new_inode(struct inode *inode)
576{
Peter Zijlstra14358e62007-10-14 01:38:33 +0200577#ifdef CONFIG_DEBUG_LOCK_ALLOC
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200578 if (inode->i_mode & S_IFDIR) {
579 struct file_system_type *type = inode->i_sb->s_type;
580
581 /*
582 * ensure nobody is actually holding i_mutex
583 */
584 mutex_destroy(&inode->i_mutex);
585 mutex_init(&inode->i_mutex);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200586 lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key);
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200587 }
Peter Zijlstra14358e62007-10-14 01:38:33 +0200588#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 /*
590 * This is special! We do not need the spinlock
591 * when clearing I_LOCK, because we're guaranteed
592 * that nobody else tries to do anything about the
593 * state of the inode when it is locked, as we
594 * just created it (so there can be no old holders
595 * that haven't tested I_LOCK).
596 */
597 inode->i_state &= ~(I_LOCK|I_NEW);
598 wake_up_inode(inode);
599}
600
601EXPORT_SYMBOL(unlock_new_inode);
602
603/*
604 * This is called without the inode lock held.. Be careful.
605 *
606 * We no longer cache the sb_flags in i_flags - see fs.h
607 * -- rmk@arm.uk.linux.org
608 */
609static 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)
610{
611 struct inode * inode;
612
613 inode = alloc_inode(sb);
614 if (inode) {
615 struct inode * old;
616
617 spin_lock(&inode_lock);
618 /* We released the lock, so.. */
619 old = find_inode(sb, head, test, data);
620 if (!old) {
621 if (set(inode, data))
622 goto set_failed;
623
624 inodes_stat.nr_inodes++;
625 list_add(&inode->i_list, &inode_in_use);
626 list_add(&inode->i_sb_list, &sb->s_inodes);
627 hlist_add_head(&inode->i_hash, head);
628 inode->i_state = I_LOCK|I_NEW;
629 spin_unlock(&inode_lock);
630
631 /* Return the locked inode with I_NEW set, the
632 * caller is responsible for filling in the contents
633 */
634 return inode;
635 }
636
637 /*
638 * Uhhuh, somebody else created the same inode under
639 * us. Use the old inode instead of the one we just
640 * allocated.
641 */
642 __iget(old);
643 spin_unlock(&inode_lock);
644 destroy_inode(inode);
645 inode = old;
646 wait_on_inode(inode);
647 }
648 return inode;
649
650set_failed:
651 spin_unlock(&inode_lock);
652 destroy_inode(inode);
653 return NULL;
654}
655
656/*
657 * get_new_inode_fast is the fast path version of get_new_inode, see the
658 * comment at iget_locked for details.
659 */
660static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
661{
662 struct inode * inode;
663
664 inode = alloc_inode(sb);
665 if (inode) {
666 struct inode * old;
667
668 spin_lock(&inode_lock);
669 /* We released the lock, so.. */
670 old = find_inode_fast(sb, head, ino);
671 if (!old) {
672 inode->i_ino = ino;
673 inodes_stat.nr_inodes++;
674 list_add(&inode->i_list, &inode_in_use);
675 list_add(&inode->i_sb_list, &sb->s_inodes);
676 hlist_add_head(&inode->i_hash, head);
677 inode->i_state = I_LOCK|I_NEW;
678 spin_unlock(&inode_lock);
679
680 /* Return the locked inode with I_NEW set, the
681 * caller is responsible for filling in the contents
682 */
683 return inode;
684 }
685
686 /*
687 * Uhhuh, somebody else created the same inode under
688 * us. Use the old inode instead of the one we just
689 * allocated.
690 */
691 __iget(old);
692 spin_unlock(&inode_lock);
693 destroy_inode(inode);
694 inode = old;
695 wait_on_inode(inode);
696 }
697 return inode;
698}
699
Andreas Mohred97bd32006-10-02 02:17:17 -0700700static unsigned long hash(struct super_block *sb, unsigned long hashval)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
702 unsigned long tmp;
703
704 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
705 L1_CACHE_BYTES;
706 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
707 return tmp & I_HASHMASK;
708}
709
710/**
711 * iunique - get a unique inode number
712 * @sb: superblock
713 * @max_reserved: highest reserved inode number
714 *
715 * Obtain an inode number that is unique on the system for a given
716 * superblock. This is used by file systems that have no natural
717 * permanent inode numbering system. An inode number is returned that
718 * is higher than the reserved limit but unique.
719 *
720 * BUGS:
721 * With a large number of inodes live on the file system this function
722 * currently becomes quite slow.
723 */
724ino_t iunique(struct super_block *sb, ino_t max_reserved)
725{
Jeff Layton866b04f2007-05-08 00:32:29 -0700726 /*
727 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
728 * error if st_ino won't fit in target struct field. Use 32bit counter
729 * here to attempt to avoid that.
730 */
731 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 struct inode *inode;
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700733 struct hlist_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700736 spin_lock(&inode_lock);
737 do {
738 if (counter <= max_reserved)
739 counter = max_reserved + 1;
740 res = counter++;
741 head = inode_hashtable + hash(sb, res);
742 inode = find_inode_fast(sb, head, res);
743 } while (inode != NULL);
744 spin_unlock(&inode_lock);
745
746 return res;
747}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748EXPORT_SYMBOL(iunique);
749
750struct inode *igrab(struct inode *inode)
751{
752 spin_lock(&inode_lock);
Jan Blunck4a3b0a42007-02-10 01:44:59 -0800753 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 __iget(inode);
755 else
756 /*
757 * Handle the case where s_op->clear_inode is not been
758 * called yet, and somebody is calling igrab
759 * while the inode is getting freed.
760 */
761 inode = NULL;
762 spin_unlock(&inode_lock);
763 return inode;
764}
765
766EXPORT_SYMBOL(igrab);
767
768/**
769 * ifind - internal function, you want ilookup5() or iget5().
770 * @sb: super block of file system to search
771 * @head: the head of the list to search
772 * @test: callback used for comparisons between inodes
773 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700774 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 *
776 * ifind() searches for the inode specified by @data in the inode
777 * cache. This is a generalized version of ifind_fast() for file systems where
778 * the inode number is not sufficient for unique identification 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 *
785 * Note, @test is called with the inode_lock held, so can't sleep.
786 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800787static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700789 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
791 struct inode *inode;
792
793 spin_lock(&inode_lock);
794 inode = find_inode(sb, head, test, data);
795 if (inode) {
796 __iget(inode);
797 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700798 if (likely(wait))
799 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 return inode;
801 }
802 spin_unlock(&inode_lock);
803 return NULL;
804}
805
806/**
807 * ifind_fast - internal function, you want ilookup() or iget().
808 * @sb: super block of file system to search
809 * @head: head of the list to search
810 * @ino: inode number to search for
811 *
812 * ifind_fast() searches for the inode @ino in the inode cache. This is for
813 * file systems where the inode number is sufficient for unique identification
814 * of an inode.
815 *
816 * If the inode is in the cache, the inode is returned with an incremented
817 * reference count.
818 *
819 * Otherwise NULL is returned.
820 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800821static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 struct hlist_head *head, unsigned long ino)
823{
824 struct inode *inode;
825
826 spin_lock(&inode_lock);
827 inode = find_inode_fast(sb, head, ino);
828 if (inode) {
829 __iget(inode);
830 spin_unlock(&inode_lock);
831 wait_on_inode(inode);
832 return inode;
833 }
834 spin_unlock(&inode_lock);
835 return NULL;
836}
837
838/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700839 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 * @sb: super block of file system to search
841 * @hashval: hash value (usually inode number) to search for
842 * @test: callback used for comparisons between inodes
843 * @data: opaque data pointer to pass to @test
844 *
845 * ilookup5() uses ifind() to search for the inode specified by @hashval and
846 * @data in the inode cache. This is a generalized version of ilookup() for
847 * file systems where the inode number is not sufficient for unique
848 * identification of an inode.
849 *
850 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700851 * reference count. Note, the inode lock is not waited upon so you have to be
852 * very careful what you do with the returned inode. You probably should be
853 * using ilookup5() instead.
854 *
855 * Otherwise NULL is returned.
856 *
857 * Note, @test is called with the inode_lock held, so can't sleep.
858 */
859struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
860 int (*test)(struct inode *, void *), void *data)
861{
862 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
863
864 return ifind(sb, head, test, data, 0);
865}
866
867EXPORT_SYMBOL(ilookup5_nowait);
868
869/**
870 * ilookup5 - search for an inode in the inode cache
871 * @sb: super block of file system to search
872 * @hashval: hash value (usually inode number) to search for
873 * @test: callback used for comparisons between inodes
874 * @data: opaque data pointer to pass to @test
875 *
876 * ilookup5() uses ifind() to search for the inode specified by @hashval and
877 * @data in the inode cache. This is a generalized version of ilookup() for
878 * file systems where the inode number is not sufficient for unique
879 * identification of an inode.
880 *
881 * If the inode is in the cache, the inode lock is waited upon and the inode is
882 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 *
884 * Otherwise NULL is returned.
885 *
886 * Note, @test is called with the inode_lock held, so can't sleep.
887 */
888struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
889 int (*test)(struct inode *, void *), void *data)
890{
891 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
892
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700893 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894}
895
896EXPORT_SYMBOL(ilookup5);
897
898/**
899 * ilookup - search for an inode in the inode cache
900 * @sb: super block of file system to search
901 * @ino: inode number to search for
902 *
903 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
904 * This is for file systems where the inode number is sufficient for unique
905 * identification of an inode.
906 *
907 * If the inode is in the cache, the inode is returned with an incremented
908 * reference count.
909 *
910 * Otherwise NULL is returned.
911 */
912struct inode *ilookup(struct super_block *sb, unsigned long ino)
913{
914 struct hlist_head *head = inode_hashtable + hash(sb, ino);
915
916 return ifind_fast(sb, head, ino);
917}
918
919EXPORT_SYMBOL(ilookup);
920
921/**
922 * iget5_locked - obtain an inode from a mounted file system
923 * @sb: super block of file system
924 * @hashval: hash value (usually inode number) to get
925 * @test: callback used for comparisons between inodes
926 * @set: callback used to initialize a new struct inode
927 * @data: opaque data pointer to pass to @test and @set
928 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 * iget5_locked() uses ifind() to search for the inode specified by @hashval
930 * and @data in the inode cache and if present it is returned with an increased
931 * reference count. This is a generalized version of iget_locked() for file
932 * systems where the inode number is not sufficient for unique identification
933 * of an inode.
934 *
935 * If the inode is not in cache, get_new_inode() is called to allocate a new
936 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
937 * file system gets to fill it in before unlocking it via unlock_new_inode().
938 *
939 * Note both @test and @set are called with the inode_lock held, so can't sleep.
940 */
941struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
942 int (*test)(struct inode *, void *),
943 int (*set)(struct inode *, void *), void *data)
944{
945 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
946 struct inode *inode;
947
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700948 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 if (inode)
950 return inode;
951 /*
952 * get_new_inode() 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(sb, head, test, set, data);
956}
957
958EXPORT_SYMBOL(iget5_locked);
959
960/**
961 * iget_locked - obtain an inode from a mounted file system
962 * @sb: super block of file system
963 * @ino: inode number to get
964 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
966 * the inode cache and if present it is returned with an increased reference
967 * count. This is for file systems where the inode number is sufficient for
968 * unique identification of an inode.
969 *
970 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
971 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
972 * The file system gets to fill it in before unlocking it via
973 * unlock_new_inode().
974 */
975struct inode *iget_locked(struct super_block *sb, unsigned long ino)
976{
977 struct hlist_head *head = inode_hashtable + hash(sb, ino);
978 struct inode *inode;
979
980 inode = ifind_fast(sb, head, ino);
981 if (inode)
982 return inode;
983 /*
984 * get_new_inode_fast() will do the right thing, re-trying the search
985 * in case it had to block at any point.
986 */
987 return get_new_inode_fast(sb, head, ino);
988}
989
990EXPORT_SYMBOL(iget_locked);
991
992/**
993 * __insert_inode_hash - hash an inode
994 * @inode: unhashed inode
995 * @hashval: unsigned long value used to locate this object in the
996 * inode_hashtable.
997 *
998 * Add an inode to the inode hash for this superblock.
999 */
1000void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1001{
1002 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1003 spin_lock(&inode_lock);
1004 hlist_add_head(&inode->i_hash, head);
1005 spin_unlock(&inode_lock);
1006}
1007
1008EXPORT_SYMBOL(__insert_inode_hash);
1009
1010/**
1011 * remove_inode_hash - remove an inode from the hash
1012 * @inode: inode to unhash
1013 *
1014 * Remove an inode from the superblock.
1015 */
1016void remove_inode_hash(struct inode *inode)
1017{
1018 spin_lock(&inode_lock);
1019 hlist_del_init(&inode->i_hash);
1020 spin_unlock(&inode_lock);
1021}
1022
1023EXPORT_SYMBOL(remove_inode_hash);
1024
1025/*
1026 * Tell the filesystem that this inode is no longer of any interest and should
1027 * be completely destroyed.
1028 *
1029 * We leave the inode in the inode hash table until *after* the filesystem's
1030 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1031 * instigate) will always find up-to-date information either in the hash or on
1032 * disk.
1033 *
1034 * I_FREEING is set so that no-one will take a new reference to the inode while
1035 * it is being deleted.
1036 */
1037void generic_delete_inode(struct inode *inode)
1038{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001039 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041 list_del_init(&inode->i_list);
1042 list_del_init(&inode->i_sb_list);
Andreas Mohred97bd32006-10-02 02:17:17 -07001043 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 inodes_stat.nr_inodes--;
1045 spin_unlock(&inode_lock);
1046
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 security_inode_delete(inode);
1048
1049 if (op->delete_inode) {
1050 void (*delete)(struct inode *) = op->delete_inode;
1051 if (!is_bad_inode(inode))
1052 DQUOT_INIT(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001053 /* Filesystems implementing their own
1054 * s_op->delete_inode are required to call
1055 * truncate_inode_pages and clear_inode()
1056 * internally */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 delete(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001058 } else {
1059 truncate_inode_pages(&inode->i_data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 clear_inode(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001061 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 spin_lock(&inode_lock);
1063 hlist_del_init(&inode->i_hash);
1064 spin_unlock(&inode_lock);
1065 wake_up_inode(inode);
Eric Sesterhennb7542f82006-04-02 13:38:18 +02001066 BUG_ON(inode->i_state != I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 destroy_inode(inode);
1068}
1069
1070EXPORT_SYMBOL(generic_delete_inode);
1071
1072static void generic_forget_inode(struct inode *inode)
1073{
1074 struct super_block *sb = inode->i_sb;
1075
1076 if (!hlist_unhashed(&inode->i_hash)) {
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001077 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 list_move(&inode->i_list, &inode_unused);
1079 inodes_stat.nr_unused++;
Christoph Hellwigacb0c852007-05-08 00:25:52 -07001080 if (sb->s_flags & MS_ACTIVE) {
Alexander Viro991114c2005-06-23 00:09:01 -07001081 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001083 }
1084 inode->i_state |= I_WILL_FREE;
1085 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 write_inode_now(inode, 1);
1087 spin_lock(&inode_lock);
Alexander Viro991114c2005-06-23 00:09:01 -07001088 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 inodes_stat.nr_unused--;
1090 hlist_del_init(&inode->i_hash);
1091 }
1092 list_del_init(&inode->i_list);
1093 list_del_init(&inode->i_sb_list);
Alexander Viro991114c2005-06-23 00:09:01 -07001094 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 inodes_stat.nr_inodes--;
1096 spin_unlock(&inode_lock);
1097 if (inode->i_data.nrpages)
1098 truncate_inode_pages(&inode->i_data, 0);
1099 clear_inode(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001100 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 destroy_inode(inode);
1102}
1103
1104/*
1105 * Normal UNIX filesystem behaviour: delete the
1106 * inode when the usage count drops to zero, and
1107 * i_nlink is zero.
1108 */
Mark Fashehcb2c0232005-07-07 17:56:03 -07001109void generic_drop_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110{
1111 if (!inode->i_nlink)
1112 generic_delete_inode(inode);
1113 else
1114 generic_forget_inode(inode);
1115}
1116
Mark Fashehcb2c0232005-07-07 17:56:03 -07001117EXPORT_SYMBOL_GPL(generic_drop_inode);
1118
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119/*
1120 * Called when we're dropping the last reference
1121 * to an inode.
1122 *
1123 * Call the FS "drop()" function, defaulting to
1124 * the legacy UNIX filesystem behaviour..
1125 *
1126 * NOTE! NOTE! NOTE! We're called with the inode lock
1127 * held, and the drop function is supposed to release
1128 * the lock!
1129 */
1130static inline void iput_final(struct inode *inode)
1131{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001132 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 void (*drop)(struct inode *) = generic_drop_inode;
1134
1135 if (op && op->drop_inode)
1136 drop = op->drop_inode;
1137 drop(inode);
1138}
1139
1140/**
1141 * iput - put an inode
1142 * @inode: inode to put
1143 *
1144 * Puts an inode, dropping its usage count. If the inode use count hits
1145 * zero, the inode is then freed and may also be destroyed.
1146 *
1147 * Consequently, iput() can sleep.
1148 */
1149void iput(struct inode *inode)
1150{
1151 if (inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 BUG_ON(inode->i_state == I_CLEAR);
1153
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1155 iput_final(inode);
1156 }
1157}
1158
1159EXPORT_SYMBOL(iput);
1160
1161/**
1162 * bmap - find a block number in a file
1163 * @inode: inode of file
1164 * @block: block to find
1165 *
1166 * Returns the block number on the device holding the inode that
1167 * is the disk block number for the block of the file requested.
1168 * That is, asked for block 4 of inode 1 the function will return the
1169 * disk block relative to the disk start that holds that block of the
1170 * file.
1171 */
1172sector_t bmap(struct inode * inode, sector_t block)
1173{
1174 sector_t res = 0;
1175 if (inode->i_mapping->a_ops->bmap)
1176 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1177 return res;
1178}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179EXPORT_SYMBOL(bmap);
1180
1181/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001182 * touch_atime - update the access time
1183 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001184 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 *
1186 * Update the accessed time on an inode and mark it for writeback.
1187 * This function automatically handles read only file systems and media,
1188 * as well as the "noatime" flag and inode specific "noatime" markers.
1189 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001190void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001192 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 struct timespec now;
1194
Dave Hansencdb70f32008-02-15 14:37:41 -08001195 if (mnt_want_write(mnt))
1196 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001197 if (inode->i_flags & S_NOATIME)
Dave Hansencdb70f32008-02-15 14:37:41 -08001198 goto out;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001199 if (IS_NOATIME(inode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001200 goto out;
Andrew Mortonb2276132006-12-13 00:34:33 -08001201 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001202 goto out;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001203
Dave Hansencdb70f32008-02-15 14:37:41 -08001204 if (mnt->mnt_flags & MNT_NOATIME)
1205 goto out;
1206 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1207 goto out;
1208 if (mnt->mnt_flags & MNT_RELATIME) {
1209 /*
1210 * With relative atime, only update atime if the previous
1211 * atime is earlier than either the ctime or mtime.
1212 */
1213 if (timespec_compare(&inode->i_mtime, &inode->i_atime) < 0 &&
1214 timespec_compare(&inode->i_ctime, &inode->i_atime) < 0)
1215 goto out;
Andrew Mortonb2276132006-12-13 00:34:33 -08001216 }
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001217
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 now = current_fs_time(inode->i_sb);
Valerie Henson47ae32d2006-12-13 00:34:34 -08001219 if (timespec_equal(&inode->i_atime, &now))
Dave Hansencdb70f32008-02-15 14:37:41 -08001220 goto out;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001221
1222 inode->i_atime = now;
1223 mark_inode_dirty_sync(inode);
Dave Hansencdb70f32008-02-15 14:37:41 -08001224out:
1225 mnt_drop_write(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001227EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
1229/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001230 * file_update_time - update mtime and ctime time
1231 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001233 * Update the mtime and ctime members of an inode and mark the inode
1234 * for writeback. Note that this function is meant exclusively for
1235 * usage in the file write path of filesystems, and filesystems may
1236 * choose to explicitly ignore update via this function with the
1237 * S_NOCTIME inode flag, e.g. for network filesystem where these
1238 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 */
1240
Christoph Hellwig870f4812006-01-09 20:52:01 -08001241void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001243 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 struct timespec now;
1245 int sync_it = 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001246 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
1248 if (IS_NOCMTIME(inode))
1249 return;
Dave Hansen20ddee22008-02-15 14:37:43 -08001250
1251 err = mnt_want_write(file->f_path.mnt);
1252 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 return;
1254
1255 now = current_fs_time(inode->i_sb);
Andreas Mohred97bd32006-10-02 02:17:17 -07001256 if (!timespec_equal(&inode->i_mtime, &now)) {
1257 inode->i_mtime = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001259 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
Andreas Mohred97bd32006-10-02 02:17:17 -07001261 if (!timespec_equal(&inode->i_ctime, &now)) {
1262 inode->i_ctime = now;
Christoph Hellwig870f4812006-01-09 20:52:01 -08001263 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001264 }
Christoph Hellwig870f4812006-01-09 20:52:01 -08001265
Jean Noel Cordenner7a224222008-01-28 23:58:27 -05001266 if (IS_I_VERSION(inode)) {
1267 inode_inc_iversion(inode);
1268 sync_it = 1;
1269 }
1270
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 if (sync_it)
1272 mark_inode_dirty_sync(inode);
Dave Hansen20ddee22008-02-15 14:37:43 -08001273 mnt_drop_write(file->f_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274}
1275
Christoph Hellwig870f4812006-01-09 20:52:01 -08001276EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
1278int inode_needs_sync(struct inode *inode)
1279{
1280 if (IS_SYNC(inode))
1281 return 1;
1282 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1283 return 1;
1284 return 0;
1285}
1286
1287EXPORT_SYMBOL(inode_needs_sync);
1288
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289int inode_wait(void *word)
1290{
1291 schedule();
1292 return 0;
1293}
1294
1295/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001296 * If we try to find an inode in the inode hash while it is being
1297 * deleted, we have to wait until the filesystem completes its
1298 * deletion before reporting that it isn't found. This function waits
1299 * until the deletion _might_ have completed. Callers are responsible
1300 * to recheck inode state.
1301 *
1302 * It doesn't matter if I_LOCK is not set initially, a call to
1303 * wake_up_inode() after removing from the hash list will DTRT.
1304 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 * This is called with inode_lock held.
1306 */
1307static void __wait_on_freeing_inode(struct inode *inode)
1308{
1309 wait_queue_head_t *wq;
1310 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1312 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1313 spin_unlock(&inode_lock);
1314 schedule();
1315 finish_wait(wq, &wait.wait);
1316 spin_lock(&inode_lock);
1317}
1318
Mark Fasheh62752ee2006-10-17 10:31:38 +02001319/*
1320 * We rarely want to lock two inodes that do not have a parent/child
1321 * relationship (such as directory, child inode) simultaneously. The
1322 * vast majority of file systems should be able to get along fine
1323 * without this. Do not use these functions except as a last resort.
1324 */
1325void inode_double_lock(struct inode *inode1, struct inode *inode2)
1326{
1327 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1328 if (inode1)
1329 mutex_lock(&inode1->i_mutex);
1330 else if (inode2)
1331 mutex_lock(&inode2->i_mutex);
1332 return;
1333 }
1334
1335 if (inode1 < inode2) {
1336 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1337 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1338 } else {
1339 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1340 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1341 }
1342}
1343EXPORT_SYMBOL(inode_double_lock);
1344
1345void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1346{
1347 if (inode1)
1348 mutex_unlock(&inode1->i_mutex);
1349
1350 if (inode2 && inode2 != inode1)
1351 mutex_unlock(&inode2->i_mutex);
1352}
1353EXPORT_SYMBOL(inode_double_unlock);
1354
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355static __initdata unsigned long ihash_entries;
1356static int __init set_ihash_entries(char *str)
1357{
1358 if (!str)
1359 return 0;
1360 ihash_entries = simple_strtoul(str, &str, 0);
1361 return 1;
1362}
1363__setup("ihash_entries=", set_ihash_entries);
1364
1365/*
1366 * Initialize the waitqueues and inode hash table.
1367 */
1368void __init inode_init_early(void)
1369{
1370 int loop;
1371
1372 /* If hashes are distributed across NUMA nodes, defer
1373 * hash allocation until vmalloc space is available.
1374 */
1375 if (hashdist)
1376 return;
1377
1378 inode_hashtable =
1379 alloc_large_system_hash("Inode-cache",
1380 sizeof(struct hlist_head),
1381 ihash_entries,
1382 14,
1383 HASH_EARLY,
1384 &i_hash_shift,
1385 &i_hash_mask,
1386 0);
1387
1388 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1389 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1390}
1391
Denis Cheng74bf17c2007-10-16 23:26:30 -07001392void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393{
1394 int loop;
1395
1396 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001397 inode_cachep = kmem_cache_create("inode_cache",
1398 sizeof(struct inode),
1399 0,
1400 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1401 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001402 init_once);
Rusty Russell8e1f9362007-07-17 04:03:17 -07001403 register_shrinker(&icache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
1405 /* Hash may have been set up in inode_init_early */
1406 if (!hashdist)
1407 return;
1408
1409 inode_hashtable =
1410 alloc_large_system_hash("Inode-cache",
1411 sizeof(struct hlist_head),
1412 ihash_entries,
1413 14,
1414 0,
1415 &i_hash_shift,
1416 &i_hash_mask,
1417 0);
1418
1419 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1420 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1421}
1422
1423void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1424{
1425 inode->i_mode = mode;
1426 if (S_ISCHR(mode)) {
1427 inode->i_fop = &def_chr_fops;
1428 inode->i_rdev = rdev;
1429 } else if (S_ISBLK(mode)) {
1430 inode->i_fop = &def_blk_fops;
1431 inode->i_rdev = rdev;
1432 } else if (S_ISFIFO(mode))
1433 inode->i_fop = &def_fifo_fops;
1434 else if (S_ISSOCK(mode))
1435 inode->i_fop = &bad_sock_fops;
1436 else
1437 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1438 mode);
1439}
1440EXPORT_SYMBOL(init_special_inode);