blob: c6165771e00e7443618176cc7156bbf14dd3387b [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 Lameter4ba9b9d2007-10-16 23:25:51 -0700218static void init_once(struct kmem_cache * cachep, void *foo)
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{
Peter Zijlstra14358e62007-10-14 01:38:33 +0200570#ifdef CONFIG_DEBUG_LOCK_ALLOC
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200571 if (inode->i_mode & S_IFDIR) {
572 struct file_system_type *type = inode->i_sb->s_type;
573
574 /*
575 * ensure nobody is actually holding i_mutex
576 */
577 mutex_destroy(&inode->i_mutex);
578 mutex_init(&inode->i_mutex);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200579 lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key);
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200580 }
Peter Zijlstra14358e62007-10-14 01:38:33 +0200581#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 /*
583 * This is special! We do not need the spinlock
584 * when clearing I_LOCK, because we're guaranteed
585 * that nobody else tries to do anything about the
586 * state of the inode when it is locked, as we
587 * just created it (so there can be no old holders
588 * that haven't tested I_LOCK).
589 */
590 inode->i_state &= ~(I_LOCK|I_NEW);
591 wake_up_inode(inode);
592}
593
594EXPORT_SYMBOL(unlock_new_inode);
595
596/*
597 * This is called without the inode lock held.. Be careful.
598 *
599 * We no longer cache the sb_flags in i_flags - see fs.h
600 * -- rmk@arm.uk.linux.org
601 */
602static 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)
603{
604 struct inode * inode;
605
606 inode = alloc_inode(sb);
607 if (inode) {
608 struct inode * old;
609
610 spin_lock(&inode_lock);
611 /* We released the lock, so.. */
612 old = find_inode(sb, head, test, data);
613 if (!old) {
614 if (set(inode, data))
615 goto set_failed;
616
617 inodes_stat.nr_inodes++;
618 list_add(&inode->i_list, &inode_in_use);
619 list_add(&inode->i_sb_list, &sb->s_inodes);
620 hlist_add_head(&inode->i_hash, head);
621 inode->i_state = I_LOCK|I_NEW;
622 spin_unlock(&inode_lock);
623
624 /* Return the locked inode with I_NEW set, the
625 * caller is responsible for filling in the contents
626 */
627 return inode;
628 }
629
630 /*
631 * Uhhuh, somebody else created the same inode under
632 * us. Use the old inode instead of the one we just
633 * allocated.
634 */
635 __iget(old);
636 spin_unlock(&inode_lock);
637 destroy_inode(inode);
638 inode = old;
639 wait_on_inode(inode);
640 }
641 return inode;
642
643set_failed:
644 spin_unlock(&inode_lock);
645 destroy_inode(inode);
646 return NULL;
647}
648
649/*
650 * get_new_inode_fast is the fast path version of get_new_inode, see the
651 * comment at iget_locked for details.
652 */
653static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
654{
655 struct inode * inode;
656
657 inode = alloc_inode(sb);
658 if (inode) {
659 struct inode * old;
660
661 spin_lock(&inode_lock);
662 /* We released the lock, so.. */
663 old = find_inode_fast(sb, head, ino);
664 if (!old) {
665 inode->i_ino = ino;
666 inodes_stat.nr_inodes++;
667 list_add(&inode->i_list, &inode_in_use);
668 list_add(&inode->i_sb_list, &sb->s_inodes);
669 hlist_add_head(&inode->i_hash, head);
670 inode->i_state = I_LOCK|I_NEW;
671 spin_unlock(&inode_lock);
672
673 /* Return the locked inode with I_NEW set, the
674 * caller is responsible for filling in the contents
675 */
676 return inode;
677 }
678
679 /*
680 * Uhhuh, somebody else created the same inode under
681 * us. Use the old inode instead of the one we just
682 * allocated.
683 */
684 __iget(old);
685 spin_unlock(&inode_lock);
686 destroy_inode(inode);
687 inode = old;
688 wait_on_inode(inode);
689 }
690 return inode;
691}
692
Andreas Mohred97bd32006-10-02 02:17:17 -0700693static unsigned long hash(struct super_block *sb, unsigned long hashval)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694{
695 unsigned long tmp;
696
697 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
698 L1_CACHE_BYTES;
699 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
700 return tmp & I_HASHMASK;
701}
702
703/**
704 * iunique - get a unique inode number
705 * @sb: superblock
706 * @max_reserved: highest reserved inode number
707 *
708 * Obtain an inode number that is unique on the system for a given
709 * superblock. This is used by file systems that have no natural
710 * permanent inode numbering system. An inode number is returned that
711 * is higher than the reserved limit but unique.
712 *
713 * BUGS:
714 * With a large number of inodes live on the file system this function
715 * currently becomes quite slow.
716 */
717ino_t iunique(struct super_block *sb, ino_t max_reserved)
718{
Jeff Layton866b04f2007-05-08 00:32:29 -0700719 /*
720 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
721 * error if st_ino won't fit in target struct field. Use 32bit counter
722 * here to attempt to avoid that.
723 */
724 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 struct inode *inode;
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700726 struct hlist_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700729 spin_lock(&inode_lock);
730 do {
731 if (counter <= max_reserved)
732 counter = max_reserved + 1;
733 res = counter++;
734 head = inode_hashtable + hash(sb, res);
735 inode = find_inode_fast(sb, head, res);
736 } while (inode != NULL);
737 spin_unlock(&inode_lock);
738
739 return res;
740}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741EXPORT_SYMBOL(iunique);
742
743struct inode *igrab(struct inode *inode)
744{
745 spin_lock(&inode_lock);
Jan Blunck4a3b0a42007-02-10 01:44:59 -0800746 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 __iget(inode);
748 else
749 /*
750 * Handle the case where s_op->clear_inode is not been
751 * called yet, and somebody is calling igrab
752 * while the inode is getting freed.
753 */
754 inode = NULL;
755 spin_unlock(&inode_lock);
756 return inode;
757}
758
759EXPORT_SYMBOL(igrab);
760
761/**
762 * ifind - internal function, you want ilookup5() or iget5().
763 * @sb: super block of file system to search
764 * @head: the head of the list to search
765 * @test: callback used for comparisons between inodes
766 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700767 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 *
769 * ifind() searches for the inode specified by @data in the inode
770 * cache. This is a generalized version of ifind_fast() for file systems where
771 * the inode number is not sufficient for unique identification of an inode.
772 *
773 * If the inode is in the cache, the inode is returned with an incremented
774 * reference count.
775 *
776 * Otherwise NULL is returned.
777 *
778 * Note, @test is called with the inode_lock held, so can't sleep.
779 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800780static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700782 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783{
784 struct inode *inode;
785
786 spin_lock(&inode_lock);
787 inode = find_inode(sb, head, test, data);
788 if (inode) {
789 __iget(inode);
790 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700791 if (likely(wait))
792 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 return inode;
794 }
795 spin_unlock(&inode_lock);
796 return NULL;
797}
798
799/**
800 * ifind_fast - internal function, you want ilookup() or iget().
801 * @sb: super block of file system to search
802 * @head: head of the list to search
803 * @ino: inode number to search for
804 *
805 * ifind_fast() searches for the inode @ino in the inode cache. This is for
806 * file systems where the inode number is sufficient for unique identification
807 * of an inode.
808 *
809 * If the inode is in the cache, the inode is returned with an incremented
810 * reference count.
811 *
812 * Otherwise NULL is returned.
813 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800814static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 struct hlist_head *head, unsigned long ino)
816{
817 struct inode *inode;
818
819 spin_lock(&inode_lock);
820 inode = find_inode_fast(sb, head, ino);
821 if (inode) {
822 __iget(inode);
823 spin_unlock(&inode_lock);
824 wait_on_inode(inode);
825 return inode;
826 }
827 spin_unlock(&inode_lock);
828 return NULL;
829}
830
831/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700832 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 * @sb: super block of file system to search
834 * @hashval: hash value (usually inode number) to search for
835 * @test: callback used for comparisons between inodes
836 * @data: opaque data pointer to pass to @test
837 *
838 * ilookup5() uses ifind() to search for the inode specified by @hashval and
839 * @data in the inode cache. This is a generalized version of ilookup() for
840 * file systems where the inode number is not sufficient for unique
841 * identification of an inode.
842 *
843 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700844 * reference count. Note, the inode lock is not waited upon so you have to be
845 * very careful what you do with the returned inode. You probably should be
846 * using ilookup5() instead.
847 *
848 * Otherwise NULL is returned.
849 *
850 * Note, @test is called with the inode_lock held, so can't sleep.
851 */
852struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
853 int (*test)(struct inode *, void *), void *data)
854{
855 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
856
857 return ifind(sb, head, test, data, 0);
858}
859
860EXPORT_SYMBOL(ilookup5_nowait);
861
862/**
863 * ilookup5 - search for an inode in the inode cache
864 * @sb: super block of file system to search
865 * @hashval: hash value (usually inode number) to search for
866 * @test: callback used for comparisons between inodes
867 * @data: opaque data pointer to pass to @test
868 *
869 * ilookup5() uses ifind() to search for the inode specified by @hashval and
870 * @data in the inode cache. This is a generalized version of ilookup() for
871 * file systems where the inode number is not sufficient for unique
872 * identification of an inode.
873 *
874 * If the inode is in the cache, the inode lock is waited upon and the inode is
875 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 *
877 * Otherwise NULL is returned.
878 *
879 * Note, @test is called with the inode_lock held, so can't sleep.
880 */
881struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
882 int (*test)(struct inode *, void *), void *data)
883{
884 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
885
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700886 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887}
888
889EXPORT_SYMBOL(ilookup5);
890
891/**
892 * ilookup - search for an inode in the inode cache
893 * @sb: super block of file system to search
894 * @ino: inode number to search for
895 *
896 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
897 * This is for file systems where the inode number is sufficient for unique
898 * identification of an inode.
899 *
900 * If the inode is in the cache, the inode is returned with an incremented
901 * reference count.
902 *
903 * Otherwise NULL is returned.
904 */
905struct inode *ilookup(struct super_block *sb, unsigned long ino)
906{
907 struct hlist_head *head = inode_hashtable + hash(sb, ino);
908
909 return ifind_fast(sb, head, ino);
910}
911
912EXPORT_SYMBOL(ilookup);
913
914/**
915 * iget5_locked - obtain an inode from a mounted file system
916 * @sb: super block of file system
917 * @hashval: hash value (usually inode number) to get
918 * @test: callback used for comparisons between inodes
919 * @set: callback used to initialize a new struct inode
920 * @data: opaque data pointer to pass to @test and @set
921 *
922 * This is iget() without the read_inode() portion of get_new_inode().
923 *
924 * iget5_locked() uses ifind() to search for the inode specified by @hashval
925 * and @data in the inode cache and if present it is returned with an increased
926 * reference count. This is a generalized version of iget_locked() for file
927 * systems where the inode number is not sufficient for unique identification
928 * of an inode.
929 *
930 * If the inode is not in cache, get_new_inode() is called to allocate a new
931 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
932 * file system gets to fill it in before unlocking it via unlock_new_inode().
933 *
934 * Note both @test and @set are called with the inode_lock held, so can't sleep.
935 */
936struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
937 int (*test)(struct inode *, void *),
938 int (*set)(struct inode *, void *), void *data)
939{
940 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
941 struct inode *inode;
942
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700943 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 if (inode)
945 return inode;
946 /*
947 * get_new_inode() will do the right thing, re-trying the search
948 * in case it had to block at any point.
949 */
950 return get_new_inode(sb, head, test, set, data);
951}
952
953EXPORT_SYMBOL(iget5_locked);
954
955/**
956 * iget_locked - obtain an inode from a mounted file system
957 * @sb: super block of file system
958 * @ino: inode number to get
959 *
960 * This is iget() without the read_inode() portion of get_new_inode_fast().
961 *
962 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
963 * the inode cache and if present it is returned with an increased reference
964 * count. This is for file systems where the inode number is sufficient for
965 * unique identification of an inode.
966 *
967 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
968 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
969 * The file system gets to fill it in before unlocking it via
970 * unlock_new_inode().
971 */
972struct inode *iget_locked(struct super_block *sb, unsigned long ino)
973{
974 struct hlist_head *head = inode_hashtable + hash(sb, ino);
975 struct inode *inode;
976
977 inode = ifind_fast(sb, head, ino);
978 if (inode)
979 return inode;
980 /*
981 * get_new_inode_fast() will do the right thing, re-trying the search
982 * in case it had to block at any point.
983 */
984 return get_new_inode_fast(sb, head, ino);
985}
986
987EXPORT_SYMBOL(iget_locked);
988
989/**
990 * __insert_inode_hash - hash an inode
991 * @inode: unhashed inode
992 * @hashval: unsigned long value used to locate this object in the
993 * inode_hashtable.
994 *
995 * Add an inode to the inode hash for this superblock.
996 */
997void __insert_inode_hash(struct inode *inode, unsigned long hashval)
998{
999 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1000 spin_lock(&inode_lock);
1001 hlist_add_head(&inode->i_hash, head);
1002 spin_unlock(&inode_lock);
1003}
1004
1005EXPORT_SYMBOL(__insert_inode_hash);
1006
1007/**
1008 * remove_inode_hash - remove an inode from the hash
1009 * @inode: inode to unhash
1010 *
1011 * Remove an inode from the superblock.
1012 */
1013void remove_inode_hash(struct inode *inode)
1014{
1015 spin_lock(&inode_lock);
1016 hlist_del_init(&inode->i_hash);
1017 spin_unlock(&inode_lock);
1018}
1019
1020EXPORT_SYMBOL(remove_inode_hash);
1021
1022/*
1023 * Tell the filesystem that this inode is no longer of any interest and should
1024 * be completely destroyed.
1025 *
1026 * We leave the inode in the inode hash table until *after* the filesystem's
1027 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1028 * instigate) will always find up-to-date information either in the hash or on
1029 * disk.
1030 *
1031 * I_FREEING is set so that no-one will take a new reference to the inode while
1032 * it is being deleted.
1033 */
1034void generic_delete_inode(struct inode *inode)
1035{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001036 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
1038 list_del_init(&inode->i_list);
1039 list_del_init(&inode->i_sb_list);
Andreas Mohred97bd32006-10-02 02:17:17 -07001040 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 inodes_stat.nr_inodes--;
1042 spin_unlock(&inode_lock);
1043
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 security_inode_delete(inode);
1045
1046 if (op->delete_inode) {
1047 void (*delete)(struct inode *) = op->delete_inode;
1048 if (!is_bad_inode(inode))
1049 DQUOT_INIT(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001050 /* Filesystems implementing their own
1051 * s_op->delete_inode are required to call
1052 * truncate_inode_pages and clear_inode()
1053 * internally */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 delete(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001055 } else {
1056 truncate_inode_pages(&inode->i_data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 clear_inode(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001058 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 spin_lock(&inode_lock);
1060 hlist_del_init(&inode->i_hash);
1061 spin_unlock(&inode_lock);
1062 wake_up_inode(inode);
Eric Sesterhennb7542f82006-04-02 13:38:18 +02001063 BUG_ON(inode->i_state != I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 destroy_inode(inode);
1065}
1066
1067EXPORT_SYMBOL(generic_delete_inode);
1068
1069static void generic_forget_inode(struct inode *inode)
1070{
1071 struct super_block *sb = inode->i_sb;
1072
1073 if (!hlist_unhashed(&inode->i_hash)) {
1074 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1075 list_move(&inode->i_list, &inode_unused);
1076 inodes_stat.nr_unused++;
Christoph Hellwigacb0c852007-05-08 00:25:52 -07001077 if (sb->s_flags & MS_ACTIVE) {
Alexander Viro991114c2005-06-23 00:09:01 -07001078 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001080 }
1081 inode->i_state |= I_WILL_FREE;
1082 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 write_inode_now(inode, 1);
1084 spin_lock(&inode_lock);
Alexander Viro991114c2005-06-23 00:09:01 -07001085 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 inodes_stat.nr_unused--;
1087 hlist_del_init(&inode->i_hash);
1088 }
1089 list_del_init(&inode->i_list);
1090 list_del_init(&inode->i_sb_list);
Alexander Viro991114c2005-06-23 00:09:01 -07001091 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 inodes_stat.nr_inodes--;
1093 spin_unlock(&inode_lock);
1094 if (inode->i_data.nrpages)
1095 truncate_inode_pages(&inode->i_data, 0);
1096 clear_inode(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001097 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 destroy_inode(inode);
1099}
1100
1101/*
1102 * Normal UNIX filesystem behaviour: delete the
1103 * inode when the usage count drops to zero, and
1104 * i_nlink is zero.
1105 */
Mark Fashehcb2c0232005-07-07 17:56:03 -07001106void generic_drop_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107{
1108 if (!inode->i_nlink)
1109 generic_delete_inode(inode);
1110 else
1111 generic_forget_inode(inode);
1112}
1113
Mark Fashehcb2c0232005-07-07 17:56:03 -07001114EXPORT_SYMBOL_GPL(generic_drop_inode);
1115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116/*
1117 * Called when we're dropping the last reference
1118 * to an inode.
1119 *
1120 * Call the FS "drop()" function, defaulting to
1121 * the legacy UNIX filesystem behaviour..
1122 *
1123 * NOTE! NOTE! NOTE! We're called with the inode lock
1124 * held, and the drop function is supposed to release
1125 * the lock!
1126 */
1127static inline void iput_final(struct inode *inode)
1128{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001129 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 void (*drop)(struct inode *) = generic_drop_inode;
1131
1132 if (op && op->drop_inode)
1133 drop = op->drop_inode;
1134 drop(inode);
1135}
1136
1137/**
1138 * iput - put an inode
1139 * @inode: inode to put
1140 *
1141 * Puts an inode, dropping its usage count. If the inode use count hits
1142 * zero, the inode is then freed and may also be destroyed.
1143 *
1144 * Consequently, iput() can sleep.
1145 */
1146void iput(struct inode *inode)
1147{
1148 if (inode) {
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001149 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
1151 BUG_ON(inode->i_state == I_CLEAR);
1152
1153 if (op && op->put_inode)
1154 op->put_inode(inode);
1155
1156 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1157 iput_final(inode);
1158 }
1159}
1160
1161EXPORT_SYMBOL(iput);
1162
1163/**
1164 * bmap - find a block number in a file
1165 * @inode: inode of file
1166 * @block: block to find
1167 *
1168 * Returns the block number on the device holding the inode that
1169 * is the disk block number for the block of the file requested.
1170 * That is, asked for block 4 of inode 1 the function will return the
1171 * disk block relative to the disk start that holds that block of the
1172 * file.
1173 */
1174sector_t bmap(struct inode * inode, sector_t block)
1175{
1176 sector_t res = 0;
1177 if (inode->i_mapping->a_ops->bmap)
1178 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1179 return res;
1180}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181EXPORT_SYMBOL(bmap);
1182
1183/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001184 * touch_atime - update the access time
1185 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001186 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 *
1188 * Update the accessed time on an inode and mark it for writeback.
1189 * This function automatically handles read only file systems and media,
1190 * as well as the "noatime" flag and inode specific "noatime" markers.
1191 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001192void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001194 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 struct timespec now;
1196
Andrew Mortonb2276132006-12-13 00:34:33 -08001197 if (inode->i_flags & S_NOATIME)
1198 return;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001199 if (IS_NOATIME(inode))
Andrew Mortonb2276132006-12-13 00:34:33 -08001200 return;
1201 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001202 return;
1203
1204 /*
1205 * We may have a NULL vfsmount when coming from NFSD
1206 */
Andrew Mortonb2276132006-12-13 00:34:33 -08001207 if (mnt) {
1208 if (mnt->mnt_flags & MNT_NOATIME)
1209 return;
1210 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1211 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001212
1213 if (mnt->mnt_flags & MNT_RELATIME) {
1214 /*
1215 * With relative atime, only update atime if the
1216 * previous atime is earlier than either the ctime or
1217 * mtime.
1218 */
1219 if (timespec_compare(&inode->i_mtime,
1220 &inode->i_atime) < 0 &&
1221 timespec_compare(&inode->i_ctime,
1222 &inode->i_atime) < 0)
1223 return;
1224 }
Andrew Mortonb2276132006-12-13 00:34:33 -08001225 }
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001226
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 now = current_fs_time(inode->i_sb);
Valerie Henson47ae32d2006-12-13 00:34:34 -08001228 if (timespec_equal(&inode->i_atime, &now))
1229 return;
1230
1231 inode->i_atime = now;
1232 mark_inode_dirty_sync(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001234EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235
1236/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001237 * file_update_time - update mtime and ctime time
1238 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001240 * Update the mtime and ctime members of an inode and mark the inode
1241 * for writeback. Note that this function is meant exclusively for
1242 * usage in the file write path of filesystems, and filesystems may
1243 * choose to explicitly ignore update via this function with the
1244 * S_NOCTIME inode flag, e.g. for network filesystem where these
1245 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 */
1247
Christoph Hellwig870f4812006-01-09 20:52:01 -08001248void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001250 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 struct timespec now;
1252 int sync_it = 0;
1253
1254 if (IS_NOCMTIME(inode))
1255 return;
1256 if (IS_RDONLY(inode))
1257 return;
1258
1259 now = current_fs_time(inode->i_sb);
Andreas Mohred97bd32006-10-02 02:17:17 -07001260 if (!timespec_equal(&inode->i_mtime, &now)) {
1261 inode->i_mtime = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001263 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
Andreas Mohred97bd32006-10-02 02:17:17 -07001265 if (!timespec_equal(&inode->i_ctime, &now)) {
1266 inode->i_ctime = now;
Christoph Hellwig870f4812006-01-09 20:52:01 -08001267 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001268 }
Christoph Hellwig870f4812006-01-09 20:52:01 -08001269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 if (sync_it)
1271 mark_inode_dirty_sync(inode);
1272}
1273
Christoph Hellwig870f4812006-01-09 20:52:01 -08001274EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
1276int inode_needs_sync(struct inode *inode)
1277{
1278 if (IS_SYNC(inode))
1279 return 1;
1280 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1281 return 1;
1282 return 0;
1283}
1284
1285EXPORT_SYMBOL(inode_needs_sync);
1286
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287int inode_wait(void *word)
1288{
1289 schedule();
1290 return 0;
1291}
1292
1293/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001294 * If we try to find an inode in the inode hash while it is being
1295 * deleted, we have to wait until the filesystem completes its
1296 * deletion before reporting that it isn't found. This function waits
1297 * until the deletion _might_ have completed. Callers are responsible
1298 * to recheck inode state.
1299 *
1300 * It doesn't matter if I_LOCK is not set initially, a call to
1301 * wake_up_inode() after removing from the hash list will DTRT.
1302 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 * This is called with inode_lock held.
1304 */
1305static void __wait_on_freeing_inode(struct inode *inode)
1306{
1307 wait_queue_head_t *wq;
1308 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1310 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1311 spin_unlock(&inode_lock);
1312 schedule();
1313 finish_wait(wq, &wait.wait);
1314 spin_lock(&inode_lock);
1315}
1316
1317void wake_up_inode(struct inode *inode)
1318{
1319 /*
1320 * Prevent speculative execution through spin_unlock(&inode_lock);
1321 */
1322 smp_mb();
1323 wake_up_bit(&inode->i_state, __I_LOCK);
1324}
1325
Mark Fasheh62752ee2006-10-17 10:31:38 +02001326/*
1327 * We rarely want to lock two inodes that do not have a parent/child
1328 * relationship (such as directory, child inode) simultaneously. The
1329 * vast majority of file systems should be able to get along fine
1330 * without this. Do not use these functions except as a last resort.
1331 */
1332void inode_double_lock(struct inode *inode1, struct inode *inode2)
1333{
1334 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1335 if (inode1)
1336 mutex_lock(&inode1->i_mutex);
1337 else if (inode2)
1338 mutex_lock(&inode2->i_mutex);
1339 return;
1340 }
1341
1342 if (inode1 < inode2) {
1343 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1344 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1345 } else {
1346 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1347 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1348 }
1349}
1350EXPORT_SYMBOL(inode_double_lock);
1351
1352void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1353{
1354 if (inode1)
1355 mutex_unlock(&inode1->i_mutex);
1356
1357 if (inode2 && inode2 != inode1)
1358 mutex_unlock(&inode2->i_mutex);
1359}
1360EXPORT_SYMBOL(inode_double_unlock);
1361
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362static __initdata unsigned long ihash_entries;
1363static int __init set_ihash_entries(char *str)
1364{
1365 if (!str)
1366 return 0;
1367 ihash_entries = simple_strtoul(str, &str, 0);
1368 return 1;
1369}
1370__setup("ihash_entries=", set_ihash_entries);
1371
1372/*
1373 * Initialize the waitqueues and inode hash table.
1374 */
1375void __init inode_init_early(void)
1376{
1377 int loop;
1378
1379 /* If hashes are distributed across NUMA nodes, defer
1380 * hash allocation until vmalloc space is available.
1381 */
1382 if (hashdist)
1383 return;
1384
1385 inode_hashtable =
1386 alloc_large_system_hash("Inode-cache",
1387 sizeof(struct hlist_head),
1388 ihash_entries,
1389 14,
1390 HASH_EARLY,
1391 &i_hash_shift,
1392 &i_hash_mask,
1393 0);
1394
1395 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1396 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1397}
1398
Denis Cheng74bf17c2007-10-16 23:26:30 -07001399void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
1401 int loop;
1402
1403 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001404 inode_cachep = kmem_cache_create("inode_cache",
1405 sizeof(struct inode),
1406 0,
1407 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1408 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001409 init_once);
Rusty Russell8e1f9362007-07-17 04:03:17 -07001410 register_shrinker(&icache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
1412 /* Hash may have been set up in inode_init_early */
1413 if (!hashdist)
1414 return;
1415
1416 inode_hashtable =
1417 alloc_large_system_hash("Inode-cache",
1418 sizeof(struct hlist_head),
1419 ihash_entries,
1420 14,
1421 0,
1422 &i_hash_shift,
1423 &i_hash_mask,
1424 0);
1425
1426 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1427 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1428}
1429
1430void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1431{
1432 inode->i_mode = mode;
1433 if (S_ISCHR(mode)) {
1434 inode->i_fop = &def_chr_fops;
1435 inode->i_rdev = rdev;
1436 } else if (S_ISBLK(mode)) {
1437 inode->i_fop = &def_blk_fops;
1438 inode->i_rdev = rdev;
1439 } else if (S_ISFIFO(mode))
1440 inode->i_fop = &def_fifo_fops;
1441 else if (S_ISSOCK(mode))
1442 inode->i_fop = &bad_sock_fops;
1443 else
1444 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1445 mode);
1446}
1447EXPORT_SYMBOL(init_special_inode);