blob: 098a2443196f20cacd3c1a4c38964646a14c6e74 [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
David Chinner2cb15992008-10-30 17:32:23 +1100111/**
112 * inode_init_always - perform inode structure intialisation
113 * @sb - superblock inode belongs to.
114 * @inode - inode to initialise
115 *
116 * These are initializations that need to be done on every inode
117 * allocation as the fields are not initialised by slab allocation.
118 */
119struct inode *inode_init_always(struct super_block *sb, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120{
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700121 static const struct address_space_operations empty_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 static struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800123 static const struct file_operations empty_fops;
David Chinner2cb15992008-10-30 17:32:23 +1100124
125 struct address_space * const mapping = &inode->i_data;
126
127 inode->i_sb = sb;
128 inode->i_blkbits = sb->s_blocksize_bits;
129 inode->i_flags = 0;
130 atomic_set(&inode->i_count, 1);
131 inode->i_op = &empty_iops;
132 inode->i_fop = &empty_fops;
133 inode->i_nlink = 1;
134 atomic_set(&inode->i_writecount, 0);
135 inode->i_size = 0;
136 inode->i_blocks = 0;
137 inode->i_bytes = 0;
138 inode->i_generation = 0;
139#ifdef CONFIG_QUOTA
140 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
141#endif
142 inode->i_pipe = NULL;
143 inode->i_bdev = NULL;
144 inode->i_cdev = NULL;
145 inode->i_rdev = 0;
146 inode->dirtied_when = 0;
147 if (security_inode_alloc(inode)) {
148 if (inode->i_sb->s_op->destroy_inode)
149 inode->i_sb->s_op->destroy_inode(inode);
150 else
151 kmem_cache_free(inode_cachep, (inode));
152 return NULL;
153 }
154
155 spin_lock_init(&inode->i_lock);
156 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
157
158 mutex_init(&inode->i_mutex);
159 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
160
161 init_rwsem(&inode->i_alloc_sem);
162 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
163
164 mapping->a_ops = &empty_aops;
165 mapping->host = inode;
166 mapping->flags = 0;
167 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_PAGECACHE);
168 mapping->assoc_mapping = NULL;
169 mapping->backing_dev_info = &default_backing_dev_info;
170 mapping->writeback_index = 0;
171
172 /*
173 * If the block_device provides a backing_dev_info for client
174 * inodes then use that. Otherwise the inode share the bdev's
175 * backing_dev_info.
176 */
177 if (sb->s_bdev) {
178 struct backing_dev_info *bdi;
179
180 bdi = sb->s_bdev->bd_inode_backing_dev_info;
181 if (!bdi)
182 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
183 mapping->backing_dev_info = bdi;
184 }
185 inode->i_private = NULL;
186 inode->i_mapping = mapping;
187
188 return inode;
189}
190EXPORT_SYMBOL(inode_init_always);
191
192static struct inode *alloc_inode(struct super_block *sb)
193{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 struct inode *inode;
195
196 if (sb->s_op->alloc_inode)
197 inode = sb->s_op->alloc_inode(sb);
198 else
David Chinner2cb15992008-10-30 17:32:23 +1100199 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
David Chinner2cb15992008-10-30 17:32:23 +1100201 if (inode)
202 return inode_init_always(sb, inode);
203 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204}
205
206void destroy_inode(struct inode *inode)
207{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200208 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 security_inode_free(inode);
210 if (inode->i_sb->s_op->destroy_inode)
211 inode->i_sb->s_op->destroy_inode(inode);
212 else
213 kmem_cache_free(inode_cachep, (inode));
214}
Christoph Hellwig087e3b02008-10-30 18:24:37 +1100215EXPORT_SYMBOL(destroy_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
217
218/*
219 * These are initializations that only need to be done
220 * once, because the fields are idempotent across use
221 * of the inode, so let the slab aware of that.
222 */
223void inode_init_once(struct inode *inode)
224{
225 memset(inode, 0, sizeof(*inode));
226 INIT_HLIST_NODE(&inode->i_hash);
227 INIT_LIST_HEAD(&inode->i_dentry);
228 INIT_LIST_HEAD(&inode->i_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
Nick Piggin19fd6232008-07-25 19:45:32 -0700230 spin_lock_init(&inode->i_data.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 spin_lock_init(&inode->i_data.i_mmap_lock);
232 INIT_LIST_HEAD(&inode->i_data.private_list);
233 spin_lock_init(&inode->i_data.private_lock);
234 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
235 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 i_size_ordered_init(inode);
Robert Love0eeca282005-07-12 17:06:03 -0400237#ifdef CONFIG_INOTIFY
238 INIT_LIST_HEAD(&inode->inotify_watches);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800239 mutex_init(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400240#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
243EXPORT_SYMBOL(inode_init_once);
244
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700245static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
247 struct inode * inode = (struct inode *) foo;
248
Christoph Lametera35afb82007-05-16 22:10:57 -0700249 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250}
251
252/*
253 * inode_lock must be held
254 */
255void __iget(struct inode * inode)
256{
257 if (atomic_read(&inode->i_count)) {
258 atomic_inc(&inode->i_count);
259 return;
260 }
261 atomic_inc(&inode->i_count);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700262 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 list_move(&inode->i_list, &inode_in_use);
264 inodes_stat.nr_unused--;
265}
266
267/**
268 * clear_inode - clear an inode
269 * @inode: inode to clear
270 *
271 * This is called by the filesystem to tell us
272 * that the inode is no longer useful. We just
273 * terminate it with extreme prejudice.
274 */
275void clear_inode(struct inode *inode)
276{
277 might_sleep();
278 invalidate_inode_buffers(inode);
279
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200280 BUG_ON(inode->i_data.nrpages);
281 BUG_ON(!(inode->i_state & I_FREEING));
282 BUG_ON(inode->i_state & I_CLEAR);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700283 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 DQUOT_DROP(inode);
Christoph Hellwigacb0c852007-05-08 00:25:52 -0700285 if (inode->i_sb->s_op->clear_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 inode->i_sb->s_op->clear_inode(inode);
Theodore Ts'oeaf796e2006-09-27 01:50:48 -0700287 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 bd_forget(inode);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700289 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 cd_forget(inode);
291 inode->i_state = I_CLEAR;
292}
293
294EXPORT_SYMBOL(clear_inode);
295
296/*
297 * dispose_list - dispose of the contents of a local list
298 * @head: the head of the list to free
299 *
300 * Dispose-list gets a local list with local inodes in it, so it doesn't
301 * need to worry about list corruption and SMP locks.
302 */
303static void dispose_list(struct list_head *head)
304{
305 int nr_disposed = 0;
306
307 while (!list_empty(head)) {
308 struct inode *inode;
309
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700310 inode = list_first_entry(head, struct inode, i_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 list_del(&inode->i_list);
312
313 if (inode->i_data.nrpages)
314 truncate_inode_pages(&inode->i_data, 0);
315 clear_inode(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700316
317 spin_lock(&inode_lock);
318 hlist_del_init(&inode->i_hash);
319 list_del_init(&inode->i_sb_list);
320 spin_unlock(&inode_lock);
321
322 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 destroy_inode(inode);
324 nr_disposed++;
325 }
326 spin_lock(&inode_lock);
327 inodes_stat.nr_inodes -= nr_disposed;
328 spin_unlock(&inode_lock);
329}
330
331/*
332 * Invalidate all inodes for a device.
333 */
334static int invalidate_list(struct list_head *head, struct list_head *dispose)
335{
336 struct list_head *next;
337 int busy = 0, count = 0;
338
339 next = head->next;
340 for (;;) {
341 struct list_head * tmp = next;
342 struct inode * inode;
343
344 /*
345 * We can reschedule here without worrying about the list's
346 * consistency because the per-sb list of inodes must not
Ingo Molnarf24075b2006-03-23 03:00:34 -0800347 * change during umount anymore, and because iprune_mutex keeps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 * shrink_icache_memory() away.
349 */
350 cond_resched_lock(&inode_lock);
351
352 next = next->next;
353 if (tmp == head)
354 break;
355 inode = list_entry(tmp, struct inode, i_sb_list);
356 invalidate_inode_buffers(inode);
357 if (!atomic_read(&inode->i_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 list_move(&inode->i_list, dispose);
359 inode->i_state |= I_FREEING;
360 count++;
361 continue;
362 }
363 busy = 1;
364 }
365 /* only unused inodes may be cached with i_count zero */
366 inodes_stat.nr_unused -= count;
367 return busy;
368}
369
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370/**
371 * invalidate_inodes - discard the inodes on a device
372 * @sb: superblock
373 *
374 * Discard all of the inodes for a given superblock. If the discard
375 * fails because there are busy inodes then a non zero value is returned.
376 * If the discard is successful all the inodes have been discarded.
377 */
378int invalidate_inodes(struct super_block * sb)
379{
380 int busy;
381 LIST_HEAD(throw_away);
382
Ingo Molnarf24075b2006-03-23 03:00:34 -0800383 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 spin_lock(&inode_lock);
Robert Love0eeca282005-07-12 17:06:03 -0400385 inotify_unmount_inodes(&sb->s_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 busy = invalidate_list(&sb->s_inodes, &throw_away);
387 spin_unlock(&inode_lock);
388
389 dispose_list(&throw_away);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800390 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
392 return busy;
393}
394
395EXPORT_SYMBOL(invalidate_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
397static int can_unuse(struct inode *inode)
398{
399 if (inode->i_state)
400 return 0;
401 if (inode_has_buffers(inode))
402 return 0;
403 if (atomic_read(&inode->i_count))
404 return 0;
405 if (inode->i_data.nrpages)
406 return 0;
407 return 1;
408}
409
410/*
411 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
412 * a temporary list and then are freed outside inode_lock by dispose_list().
413 *
414 * Any inodes which are pinned purely because of attached pagecache have their
415 * pagecache removed. We expect the final iput() on that inode to add it to
416 * the front of the inode_unused list. So look for it there and if the
417 * inode is still freeable, proceed. The right inode is found 99.9% of the
418 * time in testing on a 4-way.
419 *
420 * If the inode has metadata buffers attached to mapping->private_list then
421 * try to remove them.
422 */
423static void prune_icache(int nr_to_scan)
424{
425 LIST_HEAD(freeable);
426 int nr_pruned = 0;
427 int nr_scanned;
428 unsigned long reap = 0;
429
Ingo Molnarf24075b2006-03-23 03:00:34 -0800430 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 spin_lock(&inode_lock);
432 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
433 struct inode *inode;
434
435 if (list_empty(&inode_unused))
436 break;
437
438 inode = list_entry(inode_unused.prev, struct inode, i_list);
439
440 if (inode->i_state || atomic_read(&inode->i_count)) {
441 list_move(&inode->i_list, &inode_unused);
442 continue;
443 }
444 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
445 __iget(inode);
446 spin_unlock(&inode_lock);
447 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800448 reap += invalidate_mapping_pages(&inode->i_data,
449 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 iput(inode);
451 spin_lock(&inode_lock);
452
453 if (inode != list_entry(inode_unused.next,
454 struct inode, i_list))
455 continue; /* wrong inode or list_empty */
456 if (!can_unuse(inode))
457 continue;
458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 list_move(&inode->i_list, &freeable);
460 inode->i_state |= I_FREEING;
461 nr_pruned++;
462 }
463 inodes_stat.nr_unused -= nr_pruned;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700464 if (current_is_kswapd())
465 __count_vm_events(KSWAPD_INODESTEAL, reap);
466 else
467 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 spin_unlock(&inode_lock);
469
470 dispose_list(&freeable);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800471 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472}
473
474/*
475 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
476 * "unused" means that no dentries are referring to the inodes: the files are
477 * not open and the dcache references to those inodes have already been
478 * reclaimed.
479 *
480 * This function is passed the number of inodes to scan, and it returns the
481 * total number of remaining possibly-reclaimable inodes.
482 */
Al Viro27496a82005-10-21 03:20:48 -0400483static int shrink_icache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484{
485 if (nr) {
486 /*
487 * Nasty deadlock avoidance. We may hold various FS locks,
488 * and we don't want to recurse into the FS that called us
489 * in clear_inode() and friends..
490 */
491 if (!(gfp_mask & __GFP_FS))
492 return -1;
493 prune_icache(nr);
494 }
495 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
496}
497
Rusty Russell8e1f9362007-07-17 04:03:17 -0700498static struct shrinker icache_shrinker = {
499 .shrink = shrink_icache_memory,
500 .seeks = DEFAULT_SEEKS,
501};
502
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503static void __wait_on_freeing_inode(struct inode *inode);
504/*
505 * Called with the inode lock held.
506 * NOTE: we are not increasing the inode-refcount, you must call __iget()
507 * by hand after calling find_inode now! This simplifies iunique and won't
508 * add any additional branch in the common code.
509 */
510static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
511{
512 struct hlist_node *node;
513 struct inode * inode = NULL;
514
515repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700516 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 if (inode->i_sb != sb)
518 continue;
519 if (!test(inode, data))
520 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700521 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 __wait_on_freeing_inode(inode);
523 goto repeat;
524 }
525 break;
526 }
527 return node ? inode : NULL;
528}
529
530/*
531 * find_inode_fast is the fast path version of find_inode, see the comment at
532 * iget_locked for details.
533 */
534static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
535{
536 struct hlist_node *node;
537 struct inode * inode = NULL;
538
539repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700540 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 if (inode->i_ino != ino)
542 continue;
543 if (inode->i_sb != sb)
544 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700545 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 __wait_on_freeing_inode(inode);
547 goto repeat;
548 }
549 break;
550 }
551 return node ? inode : NULL;
552}
553
David Chinner8290c352008-10-30 17:35:24 +1100554static unsigned long hash(struct super_block *sb, unsigned long hashval)
555{
556 unsigned long tmp;
557
558 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
559 L1_CACHE_BYTES;
560 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
561 return tmp & I_HASHMASK;
562}
563
564static inline void
565__inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
566 struct inode *inode)
567{
568 inodes_stat.nr_inodes++;
569 list_add(&inode->i_list, &inode_in_use);
570 list_add(&inode->i_sb_list, &sb->s_inodes);
571 if (head)
572 hlist_add_head(&inode->i_hash, head);
573}
574
575/**
576 * inode_add_to_lists - add a new inode to relevant lists
577 * @sb - superblock inode belongs to.
578 * @inode - inode to mark in use
579 *
580 * When an inode is allocated it needs to be accounted for, added to the in use
581 * list, the owning superblock and the inode hash. This needs to be done under
582 * the inode_lock, so export a function to do this rather than the inode lock
583 * itself. We calculate the hash list to add to here so it is all internal
584 * which requires the caller to have already set up the inode number in the
585 * inode to add.
586 */
587void inode_add_to_lists(struct super_block *sb, struct inode *inode)
588{
589 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
590
591 spin_lock(&inode_lock);
592 __inode_add_to_lists(sb, head, inode);
593 spin_unlock(&inode_lock);
594}
595EXPORT_SYMBOL_GPL(inode_add_to_lists);
596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597/**
598 * new_inode - obtain an inode
599 * @sb: superblock
600 *
Mel Gorman769848c2007-07-17 04:03:05 -0700601 * Allocates a new inode for given superblock. The default gfp_mask
602 * for allocations related to inode->i_mapping is GFP_HIGHUSER_PAGECACHE.
603 * If HIGHMEM pages are unsuitable or it is known that pages allocated
604 * for the page cache are not reclaimable or migratable,
605 * mapping_set_gfp_mask() must be called with suitable flags on the
606 * newly created inode's mapping
607 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 */
609struct inode *new_inode(struct super_block *sb)
610{
Jeff Layton866b04f2007-05-08 00:32:29 -0700611 /*
612 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
613 * error if st_ino won't fit in target struct field. Use 32bit counter
614 * here to attempt to avoid that.
615 */
616 static unsigned int last_ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 struct inode * inode;
618
619 spin_lock_prefetch(&inode_lock);
620
621 inode = alloc_inode(sb);
622 if (inode) {
623 spin_lock(&inode_lock);
David Chinner8290c352008-10-30 17:35:24 +1100624 __inode_add_to_lists(sb, NULL, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 inode->i_ino = ++last_ino;
626 inode->i_state = 0;
627 spin_unlock(&inode_lock);
628 }
629 return inode;
630}
631
632EXPORT_SYMBOL(new_inode);
633
634void unlock_new_inode(struct inode *inode)
635{
Peter Zijlstra14358e62007-10-14 01:38:33 +0200636#ifdef CONFIG_DEBUG_LOCK_ALLOC
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200637 if (inode->i_mode & S_IFDIR) {
638 struct file_system_type *type = inode->i_sb->s_type;
639
640 /*
641 * ensure nobody is actually holding i_mutex
642 */
643 mutex_destroy(&inode->i_mutex);
644 mutex_init(&inode->i_mutex);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200645 lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key);
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200646 }
Peter Zijlstra14358e62007-10-14 01:38:33 +0200647#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 /*
649 * This is special! We do not need the spinlock
650 * when clearing I_LOCK, because we're guaranteed
651 * that nobody else tries to do anything about the
652 * state of the inode when it is locked, as we
653 * just created it (so there can be no old holders
654 * that haven't tested I_LOCK).
655 */
656 inode->i_state &= ~(I_LOCK|I_NEW);
657 wake_up_inode(inode);
658}
659
660EXPORT_SYMBOL(unlock_new_inode);
661
662/*
663 * This is called without the inode lock held.. Be careful.
664 *
665 * We no longer cache the sb_flags in i_flags - see fs.h
666 * -- rmk@arm.uk.linux.org
667 */
668static 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)
669{
670 struct inode * inode;
671
672 inode = alloc_inode(sb);
673 if (inode) {
674 struct inode * old;
675
676 spin_lock(&inode_lock);
677 /* We released the lock, so.. */
678 old = find_inode(sb, head, test, data);
679 if (!old) {
680 if (set(inode, data))
681 goto set_failed;
682
David Chinner8290c352008-10-30 17:35:24 +1100683 __inode_add_to_lists(sb, head, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 inode->i_state = I_LOCK|I_NEW;
685 spin_unlock(&inode_lock);
686
687 /* Return the locked inode with I_NEW set, the
688 * caller is responsible for filling in the contents
689 */
690 return inode;
691 }
692
693 /*
694 * Uhhuh, somebody else created the same inode under
695 * us. Use the old inode instead of the one we just
696 * allocated.
697 */
698 __iget(old);
699 spin_unlock(&inode_lock);
700 destroy_inode(inode);
701 inode = old;
702 wait_on_inode(inode);
703 }
704 return inode;
705
706set_failed:
707 spin_unlock(&inode_lock);
708 destroy_inode(inode);
709 return NULL;
710}
711
712/*
713 * get_new_inode_fast is the fast path version of get_new_inode, see the
714 * comment at iget_locked for details.
715 */
716static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
717{
718 struct inode * inode;
719
720 inode = alloc_inode(sb);
721 if (inode) {
722 struct inode * old;
723
724 spin_lock(&inode_lock);
725 /* We released the lock, so.. */
726 old = find_inode_fast(sb, head, ino);
727 if (!old) {
728 inode->i_ino = ino;
David Chinner8290c352008-10-30 17:35:24 +1100729 __inode_add_to_lists(sb, head, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 inode->i_state = I_LOCK|I_NEW;
731 spin_unlock(&inode_lock);
732
733 /* Return the locked inode with I_NEW set, the
734 * caller is responsible for filling in the contents
735 */
736 return inode;
737 }
738
739 /*
740 * Uhhuh, somebody else created the same inode under
741 * us. Use the old inode instead of the one we just
742 * allocated.
743 */
744 __iget(old);
745 spin_unlock(&inode_lock);
746 destroy_inode(inode);
747 inode = old;
748 wait_on_inode(inode);
749 }
750 return inode;
751}
752
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753/**
754 * iunique - get a unique inode number
755 * @sb: superblock
756 * @max_reserved: highest reserved inode number
757 *
758 * Obtain an inode number that is unique on the system for a given
759 * superblock. This is used by file systems that have no natural
760 * permanent inode numbering system. An inode number is returned that
761 * is higher than the reserved limit but unique.
762 *
763 * BUGS:
764 * With a large number of inodes live on the file system this function
765 * currently becomes quite slow.
766 */
767ino_t iunique(struct super_block *sb, ino_t max_reserved)
768{
Jeff Layton866b04f2007-05-08 00:32:29 -0700769 /*
770 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
771 * error if st_ino won't fit in target struct field. Use 32bit counter
772 * here to attempt to avoid that.
773 */
774 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 struct inode *inode;
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700776 struct hlist_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700779 spin_lock(&inode_lock);
780 do {
781 if (counter <= max_reserved)
782 counter = max_reserved + 1;
783 res = counter++;
784 head = inode_hashtable + hash(sb, res);
785 inode = find_inode_fast(sb, head, res);
786 } while (inode != NULL);
787 spin_unlock(&inode_lock);
788
789 return res;
790}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791EXPORT_SYMBOL(iunique);
792
793struct inode *igrab(struct inode *inode)
794{
795 spin_lock(&inode_lock);
Jan Blunck4a3b0a42007-02-10 01:44:59 -0800796 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 __iget(inode);
798 else
799 /*
800 * Handle the case where s_op->clear_inode is not been
801 * called yet, and somebody is calling igrab
802 * while the inode is getting freed.
803 */
804 inode = NULL;
805 spin_unlock(&inode_lock);
806 return inode;
807}
808
809EXPORT_SYMBOL(igrab);
810
811/**
812 * ifind - internal function, you want ilookup5() or iget5().
813 * @sb: super block of file system to search
814 * @head: the head of the list to search
815 * @test: callback used for comparisons between inodes
816 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700817 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 *
819 * ifind() searches for the inode specified by @data in the inode
820 * cache. This is a generalized version of ifind_fast() for file systems where
821 * the inode number is not sufficient for unique identification of an inode.
822 *
823 * If the inode is in the cache, the inode is returned with an incremented
824 * reference count.
825 *
826 * Otherwise NULL is returned.
827 *
828 * Note, @test is called with the inode_lock held, so can't sleep.
829 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800830static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700832 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833{
834 struct inode *inode;
835
836 spin_lock(&inode_lock);
837 inode = find_inode(sb, head, test, data);
838 if (inode) {
839 __iget(inode);
840 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700841 if (likely(wait))
842 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 return inode;
844 }
845 spin_unlock(&inode_lock);
846 return NULL;
847}
848
849/**
850 * ifind_fast - internal function, you want ilookup() or iget().
851 * @sb: super block of file system to search
852 * @head: head of the list to search
853 * @ino: inode number to search for
854 *
855 * ifind_fast() searches for the inode @ino in the inode cache. This is for
856 * file systems where the inode number is sufficient for unique identification
857 * of an inode.
858 *
859 * If the inode is in the cache, the inode is returned with an incremented
860 * reference count.
861 *
862 * Otherwise NULL is returned.
863 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800864static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 struct hlist_head *head, unsigned long ino)
866{
867 struct inode *inode;
868
869 spin_lock(&inode_lock);
870 inode = find_inode_fast(sb, head, ino);
871 if (inode) {
872 __iget(inode);
873 spin_unlock(&inode_lock);
874 wait_on_inode(inode);
875 return inode;
876 }
877 spin_unlock(&inode_lock);
878 return NULL;
879}
880
881/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700882 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 * @sb: super block of file system to search
884 * @hashval: hash value (usually inode number) to search for
885 * @test: callback used for comparisons between inodes
886 * @data: opaque data pointer to pass to @test
887 *
888 * ilookup5() uses ifind() to search for the inode specified by @hashval and
889 * @data in the inode cache. This is a generalized version of ilookup() for
890 * file systems where the inode number is not sufficient for unique
891 * identification of an inode.
892 *
893 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700894 * reference count. Note, the inode lock is not waited upon so you have to be
895 * very careful what you do with the returned inode. You probably should be
896 * using ilookup5() instead.
897 *
898 * Otherwise NULL is returned.
899 *
900 * Note, @test is called with the inode_lock held, so can't sleep.
901 */
902struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
903 int (*test)(struct inode *, void *), void *data)
904{
905 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
906
907 return ifind(sb, head, test, data, 0);
908}
909
910EXPORT_SYMBOL(ilookup5_nowait);
911
912/**
913 * ilookup5 - search for an inode in the inode cache
914 * @sb: super block of file system to search
915 * @hashval: hash value (usually inode number) to search for
916 * @test: callback used for comparisons between inodes
917 * @data: opaque data pointer to pass to @test
918 *
919 * ilookup5() uses ifind() to search for the inode specified by @hashval and
920 * @data in the inode cache. This is a generalized version of ilookup() for
921 * file systems where the inode number is not sufficient for unique
922 * identification of an inode.
923 *
924 * If the inode is in the cache, the inode lock is waited upon and the inode is
925 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 *
927 * Otherwise NULL is returned.
928 *
929 * Note, @test is called with the inode_lock held, so can't sleep.
930 */
931struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
932 int (*test)(struct inode *, void *), void *data)
933{
934 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
935
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700936 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937}
938
939EXPORT_SYMBOL(ilookup5);
940
941/**
942 * ilookup - search for an inode in the inode cache
943 * @sb: super block of file system to search
944 * @ino: inode number to search for
945 *
946 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
947 * This is for file systems where the inode number is sufficient for unique
948 * identification of an inode.
949 *
950 * If the inode is in the cache, the inode is returned with an incremented
951 * reference count.
952 *
953 * Otherwise NULL is returned.
954 */
955struct inode *ilookup(struct super_block *sb, unsigned long ino)
956{
957 struct hlist_head *head = inode_hashtable + hash(sb, ino);
958
959 return ifind_fast(sb, head, ino);
960}
961
962EXPORT_SYMBOL(ilookup);
963
964/**
965 * iget5_locked - obtain an inode from a mounted file system
966 * @sb: super block of file system
967 * @hashval: hash value (usually inode number) to get
968 * @test: callback used for comparisons between inodes
969 * @set: callback used to initialize a new struct inode
970 * @data: opaque data pointer to pass to @test and @set
971 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 * iget5_locked() uses ifind() to search for the inode specified by @hashval
973 * and @data in the inode cache and if present it is returned with an increased
974 * reference count. This is a generalized version of iget_locked() for file
975 * systems where the inode number is not sufficient for unique identification
976 * of an inode.
977 *
978 * If the inode is not in cache, get_new_inode() is called to allocate a new
979 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
980 * file system gets to fill it in before unlocking it via unlock_new_inode().
981 *
982 * Note both @test and @set are called with the inode_lock held, so can't sleep.
983 */
984struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
985 int (*test)(struct inode *, void *),
986 int (*set)(struct inode *, void *), void *data)
987{
988 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
989 struct inode *inode;
990
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700991 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 if (inode)
993 return inode;
994 /*
995 * get_new_inode() will do the right thing, re-trying the search
996 * in case it had to block at any point.
997 */
998 return get_new_inode(sb, head, test, set, data);
999}
1000
1001EXPORT_SYMBOL(iget5_locked);
1002
1003/**
1004 * iget_locked - obtain an inode from a mounted file system
1005 * @sb: super block of file system
1006 * @ino: inode number to get
1007 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1009 * the inode cache and if present it is returned with an increased reference
1010 * count. This is for file systems where the inode number is sufficient for
1011 * unique identification of an inode.
1012 *
1013 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1014 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1015 * The file system gets to fill it in before unlocking it via
1016 * unlock_new_inode().
1017 */
1018struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1019{
1020 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1021 struct inode *inode;
1022
1023 inode = ifind_fast(sb, head, ino);
1024 if (inode)
1025 return inode;
1026 /*
1027 * get_new_inode_fast() will do the right thing, re-trying the search
1028 * in case it had to block at any point.
1029 */
1030 return get_new_inode_fast(sb, head, ino);
1031}
1032
1033EXPORT_SYMBOL(iget_locked);
1034
1035/**
1036 * __insert_inode_hash - hash an inode
1037 * @inode: unhashed inode
1038 * @hashval: unsigned long value used to locate this object in the
1039 * inode_hashtable.
1040 *
1041 * Add an inode to the inode hash for this superblock.
1042 */
1043void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1044{
1045 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1046 spin_lock(&inode_lock);
1047 hlist_add_head(&inode->i_hash, head);
1048 spin_unlock(&inode_lock);
1049}
1050
1051EXPORT_SYMBOL(__insert_inode_hash);
1052
1053/**
1054 * remove_inode_hash - remove an inode from the hash
1055 * @inode: inode to unhash
1056 *
1057 * Remove an inode from the superblock.
1058 */
1059void remove_inode_hash(struct inode *inode)
1060{
1061 spin_lock(&inode_lock);
1062 hlist_del_init(&inode->i_hash);
1063 spin_unlock(&inode_lock);
1064}
1065
1066EXPORT_SYMBOL(remove_inode_hash);
1067
1068/*
1069 * Tell the filesystem that this inode is no longer of any interest and should
1070 * be completely destroyed.
1071 *
1072 * We leave the inode in the inode hash table until *after* the filesystem's
1073 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1074 * instigate) will always find up-to-date information either in the hash or on
1075 * disk.
1076 *
1077 * I_FREEING is set so that no-one will take a new reference to the inode while
1078 * it is being deleted.
1079 */
1080void generic_delete_inode(struct inode *inode)
1081{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001082 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083
1084 list_del_init(&inode->i_list);
1085 list_del_init(&inode->i_sb_list);
Andreas Mohred97bd32006-10-02 02:17:17 -07001086 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 inodes_stat.nr_inodes--;
1088 spin_unlock(&inode_lock);
1089
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 security_inode_delete(inode);
1091
1092 if (op->delete_inode) {
1093 void (*delete)(struct inode *) = op->delete_inode;
1094 if (!is_bad_inode(inode))
1095 DQUOT_INIT(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001096 /* Filesystems implementing their own
1097 * s_op->delete_inode are required to call
1098 * truncate_inode_pages and clear_inode()
1099 * internally */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 delete(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001101 } else {
1102 truncate_inode_pages(&inode->i_data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 clear_inode(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001104 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 spin_lock(&inode_lock);
1106 hlist_del_init(&inode->i_hash);
1107 spin_unlock(&inode_lock);
1108 wake_up_inode(inode);
Eric Sesterhennb7542f82006-04-02 13:38:18 +02001109 BUG_ON(inode->i_state != I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 destroy_inode(inode);
1111}
1112
1113EXPORT_SYMBOL(generic_delete_inode);
1114
1115static void generic_forget_inode(struct inode *inode)
1116{
1117 struct super_block *sb = inode->i_sb;
1118
1119 if (!hlist_unhashed(&inode->i_hash)) {
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001120 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 list_move(&inode->i_list, &inode_unused);
1122 inodes_stat.nr_unused++;
Christoph Hellwigacb0c852007-05-08 00:25:52 -07001123 if (sb->s_flags & MS_ACTIVE) {
Alexander Viro991114c2005-06-23 00:09:01 -07001124 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001126 }
1127 inode->i_state |= I_WILL_FREE;
1128 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 write_inode_now(inode, 1);
1130 spin_lock(&inode_lock);
Alexander Viro991114c2005-06-23 00:09:01 -07001131 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 inodes_stat.nr_unused--;
1133 hlist_del_init(&inode->i_hash);
1134 }
1135 list_del_init(&inode->i_list);
1136 list_del_init(&inode->i_sb_list);
Alexander Viro991114c2005-06-23 00:09:01 -07001137 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 inodes_stat.nr_inodes--;
1139 spin_unlock(&inode_lock);
1140 if (inode->i_data.nrpages)
1141 truncate_inode_pages(&inode->i_data, 0);
1142 clear_inode(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001143 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 destroy_inode(inode);
1145}
1146
1147/*
1148 * Normal UNIX filesystem behaviour: delete the
1149 * inode when the usage count drops to zero, and
1150 * i_nlink is zero.
1151 */
Mark Fashehcb2c0232005-07-07 17:56:03 -07001152void generic_drop_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153{
1154 if (!inode->i_nlink)
1155 generic_delete_inode(inode);
1156 else
1157 generic_forget_inode(inode);
1158}
1159
Mark Fashehcb2c0232005-07-07 17:56:03 -07001160EXPORT_SYMBOL_GPL(generic_drop_inode);
1161
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162/*
1163 * Called when we're dropping the last reference
1164 * to an inode.
1165 *
1166 * Call the FS "drop()" function, defaulting to
1167 * the legacy UNIX filesystem behaviour..
1168 *
1169 * NOTE! NOTE! NOTE! We're called with the inode lock
1170 * held, and the drop function is supposed to release
1171 * the lock!
1172 */
1173static inline void iput_final(struct inode *inode)
1174{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001175 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 void (*drop)(struct inode *) = generic_drop_inode;
1177
1178 if (op && op->drop_inode)
1179 drop = op->drop_inode;
1180 drop(inode);
1181}
1182
1183/**
1184 * iput - put an inode
1185 * @inode: inode to put
1186 *
1187 * Puts an inode, dropping its usage count. If the inode use count hits
1188 * zero, the inode is then freed and may also be destroyed.
1189 *
1190 * Consequently, iput() can sleep.
1191 */
1192void iput(struct inode *inode)
1193{
1194 if (inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 BUG_ON(inode->i_state == I_CLEAR);
1196
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1198 iput_final(inode);
1199 }
1200}
1201
1202EXPORT_SYMBOL(iput);
1203
1204/**
1205 * bmap - find a block number in a file
1206 * @inode: inode of file
1207 * @block: block to find
1208 *
1209 * Returns the block number on the device holding the inode that
1210 * is the disk block number for the block of the file requested.
1211 * That is, asked for block 4 of inode 1 the function will return the
1212 * disk block relative to the disk start that holds that block of the
1213 * file.
1214 */
1215sector_t bmap(struct inode * inode, sector_t block)
1216{
1217 sector_t res = 0;
1218 if (inode->i_mapping->a_ops->bmap)
1219 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1220 return res;
1221}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222EXPORT_SYMBOL(bmap);
1223
1224/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001225 * touch_atime - update the access time
1226 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001227 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 *
1229 * Update the accessed time on an inode and mark it for writeback.
1230 * This function automatically handles read only file systems and media,
1231 * as well as the "noatime" flag and inode specific "noatime" markers.
1232 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001233void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001235 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 struct timespec now;
1237
Dave Hansencdb70f32008-02-15 14:37:41 -08001238 if (mnt_want_write(mnt))
1239 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001240 if (inode->i_flags & S_NOATIME)
Dave Hansencdb70f32008-02-15 14:37:41 -08001241 goto out;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001242 if (IS_NOATIME(inode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001243 goto out;
Andrew Mortonb2276132006-12-13 00:34:33 -08001244 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001245 goto out;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001246
Dave Hansencdb70f32008-02-15 14:37:41 -08001247 if (mnt->mnt_flags & MNT_NOATIME)
1248 goto out;
1249 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1250 goto out;
1251 if (mnt->mnt_flags & MNT_RELATIME) {
1252 /*
1253 * With relative atime, only update atime if the previous
1254 * atime is earlier than either the ctime or mtime.
1255 */
1256 if (timespec_compare(&inode->i_mtime, &inode->i_atime) < 0 &&
1257 timespec_compare(&inode->i_ctime, &inode->i_atime) < 0)
1258 goto out;
Andrew Mortonb2276132006-12-13 00:34:33 -08001259 }
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001260
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 now = current_fs_time(inode->i_sb);
Valerie Henson47ae32d2006-12-13 00:34:34 -08001262 if (timespec_equal(&inode->i_atime, &now))
Dave Hansencdb70f32008-02-15 14:37:41 -08001263 goto out;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001264
1265 inode->i_atime = now;
1266 mark_inode_dirty_sync(inode);
Dave Hansencdb70f32008-02-15 14:37:41 -08001267out:
1268 mnt_drop_write(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001270EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
1272/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001273 * file_update_time - update mtime and ctime time
1274 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001276 * Update the mtime and ctime members of an inode and mark the inode
1277 * for writeback. Note that this function is meant exclusively for
1278 * usage in the file write path of filesystems, and filesystems may
1279 * choose to explicitly ignore update via this function with the
1280 * S_NOCTIME inode flag, e.g. for network filesystem where these
1281 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 */
1283
Christoph Hellwig870f4812006-01-09 20:52:01 -08001284void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001286 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 struct timespec now;
1288 int sync_it = 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001289 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
1291 if (IS_NOCMTIME(inode))
1292 return;
Dave Hansen20ddee22008-02-15 14:37:43 -08001293
1294 err = mnt_want_write(file->f_path.mnt);
1295 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 return;
1297
1298 now = current_fs_time(inode->i_sb);
Andreas Mohred97bd32006-10-02 02:17:17 -07001299 if (!timespec_equal(&inode->i_mtime, &now)) {
1300 inode->i_mtime = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001302 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
Andreas Mohred97bd32006-10-02 02:17:17 -07001304 if (!timespec_equal(&inode->i_ctime, &now)) {
1305 inode->i_ctime = now;
Christoph Hellwig870f4812006-01-09 20:52:01 -08001306 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001307 }
Christoph Hellwig870f4812006-01-09 20:52:01 -08001308
Jean Noel Cordenner7a224222008-01-28 23:58:27 -05001309 if (IS_I_VERSION(inode)) {
1310 inode_inc_iversion(inode);
1311 sync_it = 1;
1312 }
1313
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 if (sync_it)
1315 mark_inode_dirty_sync(inode);
Dave Hansen20ddee22008-02-15 14:37:43 -08001316 mnt_drop_write(file->f_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317}
1318
Christoph Hellwig870f4812006-01-09 20:52:01 -08001319EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
1321int inode_needs_sync(struct inode *inode)
1322{
1323 if (IS_SYNC(inode))
1324 return 1;
1325 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1326 return 1;
1327 return 0;
1328}
1329
1330EXPORT_SYMBOL(inode_needs_sync);
1331
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332int inode_wait(void *word)
1333{
1334 schedule();
1335 return 0;
1336}
Stephen Rothwelld44dab82008-11-10 17:06:05 +11001337EXPORT_SYMBOL(inode_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
1339/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001340 * If we try to find an inode in the inode hash while it is being
1341 * deleted, we have to wait until the filesystem completes its
1342 * deletion before reporting that it isn't found. This function waits
1343 * until the deletion _might_ have completed. Callers are responsible
1344 * to recheck inode state.
1345 *
1346 * It doesn't matter if I_LOCK is not set initially, a call to
1347 * wake_up_inode() after removing from the hash list will DTRT.
1348 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 * This is called with inode_lock held.
1350 */
1351static void __wait_on_freeing_inode(struct inode *inode)
1352{
1353 wait_queue_head_t *wq;
1354 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1356 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1357 spin_unlock(&inode_lock);
1358 schedule();
1359 finish_wait(wq, &wait.wait);
1360 spin_lock(&inode_lock);
1361}
1362
Mark Fasheh62752ee2006-10-17 10:31:38 +02001363/*
1364 * We rarely want to lock two inodes that do not have a parent/child
1365 * relationship (such as directory, child inode) simultaneously. The
1366 * vast majority of file systems should be able to get along fine
1367 * without this. Do not use these functions except as a last resort.
1368 */
1369void inode_double_lock(struct inode *inode1, struct inode *inode2)
1370{
1371 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1372 if (inode1)
1373 mutex_lock(&inode1->i_mutex);
1374 else if (inode2)
1375 mutex_lock(&inode2->i_mutex);
1376 return;
1377 }
1378
1379 if (inode1 < inode2) {
1380 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1381 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1382 } else {
1383 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1384 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1385 }
1386}
1387EXPORT_SYMBOL(inode_double_lock);
1388
1389void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1390{
1391 if (inode1)
1392 mutex_unlock(&inode1->i_mutex);
1393
1394 if (inode2 && inode2 != inode1)
1395 mutex_unlock(&inode2->i_mutex);
1396}
1397EXPORT_SYMBOL(inode_double_unlock);
1398
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399static __initdata unsigned long ihash_entries;
1400static int __init set_ihash_entries(char *str)
1401{
1402 if (!str)
1403 return 0;
1404 ihash_entries = simple_strtoul(str, &str, 0);
1405 return 1;
1406}
1407__setup("ihash_entries=", set_ihash_entries);
1408
1409/*
1410 * Initialize the waitqueues and inode hash table.
1411 */
1412void __init inode_init_early(void)
1413{
1414 int loop;
1415
1416 /* If hashes are distributed across NUMA nodes, defer
1417 * hash allocation until vmalloc space is available.
1418 */
1419 if (hashdist)
1420 return;
1421
1422 inode_hashtable =
1423 alloc_large_system_hash("Inode-cache",
1424 sizeof(struct hlist_head),
1425 ihash_entries,
1426 14,
1427 HASH_EARLY,
1428 &i_hash_shift,
1429 &i_hash_mask,
1430 0);
1431
1432 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1433 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1434}
1435
Denis Cheng74bf17c2007-10-16 23:26:30 -07001436void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437{
1438 int loop;
1439
1440 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001441 inode_cachep = kmem_cache_create("inode_cache",
1442 sizeof(struct inode),
1443 0,
1444 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1445 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001446 init_once);
Rusty Russell8e1f9362007-07-17 04:03:17 -07001447 register_shrinker(&icache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
1449 /* Hash may have been set up in inode_init_early */
1450 if (!hashdist)
1451 return;
1452
1453 inode_hashtable =
1454 alloc_large_system_hash("Inode-cache",
1455 sizeof(struct hlist_head),
1456 ihash_entries,
1457 14,
1458 0,
1459 &i_hash_shift,
1460 &i_hash_mask,
1461 0);
1462
1463 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1464 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1465}
1466
1467void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1468{
1469 inode->i_mode = mode;
1470 if (S_ISCHR(mode)) {
1471 inode->i_fop = &def_chr_fops;
1472 inode->i_rdev = rdev;
1473 } else if (S_ISBLK(mode)) {
1474 inode->i_fop = &def_blk_fops;
1475 inode->i_rdev = rdev;
1476 } else if (S_ISFIFO(mode))
1477 inode->i_fop = &def_fifo_fops;
1478 else if (S_ISSOCK(mode))
1479 inode->i_fop = &bad_sock_fops;
1480 else
1481 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1482 mode);
1483}
1484EXPORT_SYMBOL(init_special_inode);