blob: 0487ddba139780a4e0bfabc8f4ef2103358d9713 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/inode.c
3 *
4 * (C) 1997 Linus Torvalds
5 */
6
Linus Torvalds1da177e2005-04-16 15:20:36 -07007#include <linux/fs.h>
8#include <linux/mm.h>
9#include <linux/dcache.h>
10#include <linux/init.h>
11#include <linux/quotaops.h>
12#include <linux/slab.h>
13#include <linux/writeback.h>
14#include <linux/module.h>
15#include <linux/backing-dev.h>
16#include <linux/wait.h>
17#include <linux/hash.h>
18#include <linux/swap.h>
19#include <linux/security.h>
20#include <linux/pagemap.h>
21#include <linux/cdev.h>
22#include <linux/bootmem.h>
Robert Love0eeca282005-07-12 17:06:03 -040023#include <linux/inotify.h>
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -080024#include <linux/mount.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26/*
27 * This is needed for the following functions:
28 * - inode_has_buffers
29 * - invalidate_inode_buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 * - invalidate_bdev
31 *
32 * FIXME: remove all knowledge of the buffer layer from this file
33 */
34#include <linux/buffer_head.h>
35
36/*
37 * New inode.c implementation.
38 *
39 * This implementation has the basic premise of trying
40 * to be extremely low-overhead and SMP-safe, yet be
41 * simple enough to be "obviously correct".
42 *
43 * Famous last words.
44 */
45
46/* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
47
48/* #define INODE_PARANOIA 1 */
49/* #define INODE_DEBUG 1 */
50
51/*
52 * Inode lookup is no longer as critical as it used to be:
53 * most of the lookups are going to be through the dcache.
54 */
55#define I_HASHBITS i_hash_shift
56#define I_HASHMASK i_hash_mask
57
Eric Dumazetfa3536c2006-03-26 01:37:24 -080058static unsigned int i_hash_mask __read_mostly;
59static unsigned int i_hash_shift __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61/*
62 * Each inode can be on two separate lists. One is
63 * the hash list of the inode, used for lookups. The
64 * other linked list is the "type" list:
65 * "in_use" - valid inode, i_count > 0, i_nlink > 0
66 * "dirty" - as "in_use" but also dirty
67 * "unused" - valid inode, i_count = 0
68 *
69 * A "dirty" list is maintained for each super block,
70 * allowing for low-overhead inode sync() operations.
71 */
72
73LIST_HEAD(inode_in_use);
74LIST_HEAD(inode_unused);
Eric Dumazetfa3536c2006-03-26 01:37:24 -080075static struct hlist_head *inode_hashtable __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
77/*
78 * A simple spinlock to protect the list manipulations.
79 *
80 * NOTE! You also have to own the lock if you change
81 * the i_state of an inode while it is in use..
82 */
83DEFINE_SPINLOCK(inode_lock);
84
85/*
Ingo Molnarf24075b2006-03-23 03:00:34 -080086 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 * icache shrinking path, and the umount path. Without this exclusion,
88 * by the time prune_icache calls iput for the inode whose pages it has
89 * been invalidating, or by the time it calls clear_inode & destroy_inode
90 * from its final dispose_list, the struct super_block they refer to
91 * (for inode->i_sb->s_op) may already have been freed and reused.
92 */
Adrian Bunkbdfc3262006-03-25 03:06:56 -080093static DEFINE_MUTEX(iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
95/*
96 * Statistics gathering..
97 */
98struct inodes_stat_t inodes_stat;
99
Christoph Lametere18b8902006-12-06 20:33:20 -0800100static struct kmem_cache * inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700102static void wake_up_inode(struct inode *inode)
103{
104 /*
105 * Prevent speculative execution through spin_unlock(&inode_lock);
106 */
107 smp_mb();
108 wake_up_bit(&inode->i_state, __I_LOCK);
109}
110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111static struct inode *alloc_inode(struct super_block *sb)
112{
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700113 static const struct address_space_operations empty_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 static struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800115 static const struct file_operations empty_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 struct inode *inode;
117
118 if (sb->s_op->alloc_inode)
119 inode = sb->s_op->alloc_inode(sb);
120 else
Christoph Lametere94b1762006-12-06 20:33:17 -0800121 inode = (struct inode *) kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
123 if (inode) {
124 struct address_space * const mapping = &inode->i_data;
125
126 inode->i_sb = sb;
127 inode->i_blkbits = sb->s_blocksize_bits;
128 inode->i_flags = 0;
129 atomic_set(&inode->i_count, 1);
130 inode->i_op = &empty_iops;
131 inode->i_fop = &empty_fops;
132 inode->i_nlink = 1;
133 atomic_set(&inode->i_writecount, 0);
134 inode->i_size = 0;
135 inode->i_blocks = 0;
136 inode->i_bytes = 0;
137 inode->i_generation = 0;
138#ifdef CONFIG_QUOTA
139 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
140#endif
141 inode->i_pipe = NULL;
142 inode->i_bdev = NULL;
143 inode->i_cdev = NULL;
144 inode->i_rdev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 inode->dirtied_when = 0;
146 if (security_inode_alloc(inode)) {
147 if (inode->i_sb->s_op->destroy_inode)
148 inode->i_sb->s_op->destroy_inode(inode);
149 else
150 kmem_cache_free(inode_cachep, (inode));
151 return NULL;
152 }
153
Peter Zijlstrad475fd42007-10-15 14:51:31 +0200154 spin_lock_init(&inode->i_lock);
155 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
156
157 mutex_init(&inode->i_mutex);
158 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
159
160 init_rwsem(&inode->i_alloc_sem);
161 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 mapping->a_ops = &empty_aops;
164 mapping->host = inode;
165 mapping->flags = 0;
Mel Gorman769848c2007-07-17 04:03:05 -0700166 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_PAGECACHE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 mapping->assoc_mapping = NULL;
168 mapping->backing_dev_info = &default_backing_dev_info;
Chris Mason7d455e02008-08-15 00:40:43 -0700169 mapping->writeback_index = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
171 /*
172 * If the block_device provides a backing_dev_info for client
173 * inodes then use that. Otherwise the inode share the bdev's
174 * backing_dev_info.
175 */
176 if (sb->s_bdev) {
177 struct backing_dev_info *bdi;
178
179 bdi = sb->s_bdev->bd_inode_backing_dev_info;
180 if (!bdi)
181 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
182 mapping->backing_dev_info = bdi;
183 }
Al Viroe6c6e642006-10-10 22:48:47 +0100184 inode->i_private = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 inode->i_mapping = mapping;
186 }
187 return inode;
188}
189
190void destroy_inode(struct inode *inode)
191{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200192 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 security_inode_free(inode);
194 if (inode->i_sb->s_op->destroy_inode)
195 inode->i_sb->s_op->destroy_inode(inode);
196 else
197 kmem_cache_free(inode_cachep, (inode));
198}
199
200
201/*
202 * These are initializations that only need to be done
203 * once, because the fields are idempotent across use
204 * of the inode, so let the slab aware of that.
205 */
206void inode_init_once(struct inode *inode)
207{
208 memset(inode, 0, sizeof(*inode));
209 INIT_HLIST_NODE(&inode->i_hash);
210 INIT_LIST_HEAD(&inode->i_dentry);
211 INIT_LIST_HEAD(&inode->i_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
Nick Piggin19fd6232008-07-25 19:45:32 -0700213 spin_lock_init(&inode->i_data.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 spin_lock_init(&inode->i_data.i_mmap_lock);
215 INIT_LIST_HEAD(&inode->i_data.private_list);
216 spin_lock_init(&inode->i_data.private_lock);
217 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
218 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 i_size_ordered_init(inode);
Robert Love0eeca282005-07-12 17:06:03 -0400220#ifdef CONFIG_INOTIFY
221 INIT_LIST_HEAD(&inode->inotify_watches);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800222 mutex_init(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400223#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225
226EXPORT_SYMBOL(inode_init_once);
227
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700228static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
230 struct inode * inode = (struct inode *) foo;
231
Christoph Lametera35afb82007-05-16 22:10:57 -0700232 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
234
235/*
236 * inode_lock must be held
237 */
238void __iget(struct inode * inode)
239{
240 if (atomic_read(&inode->i_count)) {
241 atomic_inc(&inode->i_count);
242 return;
243 }
244 atomic_inc(&inode->i_count);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700245 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 list_move(&inode->i_list, &inode_in_use);
247 inodes_stat.nr_unused--;
248}
249
250/**
251 * clear_inode - clear an inode
252 * @inode: inode to clear
253 *
254 * This is called by the filesystem to tell us
255 * that the inode is no longer useful. We just
256 * terminate it with extreme prejudice.
257 */
258void clear_inode(struct inode *inode)
259{
260 might_sleep();
261 invalidate_inode_buffers(inode);
262
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200263 BUG_ON(inode->i_data.nrpages);
264 BUG_ON(!(inode->i_state & I_FREEING));
265 BUG_ON(inode->i_state & I_CLEAR);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700266 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 DQUOT_DROP(inode);
Christoph Hellwigacb0c852007-05-08 00:25:52 -0700268 if (inode->i_sb->s_op->clear_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 inode->i_sb->s_op->clear_inode(inode);
Theodore Ts'oeaf796e2006-09-27 01:50:48 -0700270 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 bd_forget(inode);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700272 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 cd_forget(inode);
274 inode->i_state = I_CLEAR;
275}
276
277EXPORT_SYMBOL(clear_inode);
278
279/*
280 * dispose_list - dispose of the contents of a local list
281 * @head: the head of the list to free
282 *
283 * Dispose-list gets a local list with local inodes in it, so it doesn't
284 * need to worry about list corruption and SMP locks.
285 */
286static void dispose_list(struct list_head *head)
287{
288 int nr_disposed = 0;
289
290 while (!list_empty(head)) {
291 struct inode *inode;
292
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700293 inode = list_first_entry(head, struct inode, i_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 list_del(&inode->i_list);
295
296 if (inode->i_data.nrpages)
297 truncate_inode_pages(&inode->i_data, 0);
298 clear_inode(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700299
300 spin_lock(&inode_lock);
301 hlist_del_init(&inode->i_hash);
302 list_del_init(&inode->i_sb_list);
303 spin_unlock(&inode_lock);
304
305 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 destroy_inode(inode);
307 nr_disposed++;
308 }
309 spin_lock(&inode_lock);
310 inodes_stat.nr_inodes -= nr_disposed;
311 spin_unlock(&inode_lock);
312}
313
314/*
315 * Invalidate all inodes for a device.
316 */
317static int invalidate_list(struct list_head *head, struct list_head *dispose)
318{
319 struct list_head *next;
320 int busy = 0, count = 0;
321
322 next = head->next;
323 for (;;) {
324 struct list_head * tmp = next;
325 struct inode * inode;
326
327 /*
328 * We can reschedule here without worrying about the list's
329 * consistency because the per-sb list of inodes must not
Ingo Molnarf24075b2006-03-23 03:00:34 -0800330 * change during umount anymore, and because iprune_mutex keeps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 * shrink_icache_memory() away.
332 */
333 cond_resched_lock(&inode_lock);
334
335 next = next->next;
336 if (tmp == head)
337 break;
338 inode = list_entry(tmp, struct inode, i_sb_list);
339 invalidate_inode_buffers(inode);
340 if (!atomic_read(&inode->i_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 list_move(&inode->i_list, dispose);
342 inode->i_state |= I_FREEING;
343 count++;
344 continue;
345 }
346 busy = 1;
347 }
348 /* only unused inodes may be cached with i_count zero */
349 inodes_stat.nr_unused -= count;
350 return busy;
351}
352
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353/**
354 * invalidate_inodes - discard the inodes on a device
355 * @sb: superblock
356 *
357 * Discard all of the inodes for a given superblock. If the discard
358 * fails because there are busy inodes then a non zero value is returned.
359 * If the discard is successful all the inodes have been discarded.
360 */
361int invalidate_inodes(struct super_block * sb)
362{
363 int busy;
364 LIST_HEAD(throw_away);
365
Ingo Molnarf24075b2006-03-23 03:00:34 -0800366 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 spin_lock(&inode_lock);
Robert Love0eeca282005-07-12 17:06:03 -0400368 inotify_unmount_inodes(&sb->s_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 busy = invalidate_list(&sb->s_inodes, &throw_away);
370 spin_unlock(&inode_lock);
371
372 dispose_list(&throw_away);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800373 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374
375 return busy;
376}
377
378EXPORT_SYMBOL(invalidate_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
380static int can_unuse(struct inode *inode)
381{
382 if (inode->i_state)
383 return 0;
384 if (inode_has_buffers(inode))
385 return 0;
386 if (atomic_read(&inode->i_count))
387 return 0;
388 if (inode->i_data.nrpages)
389 return 0;
390 return 1;
391}
392
393/*
394 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
395 * a temporary list and then are freed outside inode_lock by dispose_list().
396 *
397 * Any inodes which are pinned purely because of attached pagecache have their
398 * pagecache removed. We expect the final iput() on that inode to add it to
399 * the front of the inode_unused list. So look for it there and if the
400 * inode is still freeable, proceed. The right inode is found 99.9% of the
401 * time in testing on a 4-way.
402 *
403 * If the inode has metadata buffers attached to mapping->private_list then
404 * try to remove them.
405 */
406static void prune_icache(int nr_to_scan)
407{
408 LIST_HEAD(freeable);
409 int nr_pruned = 0;
410 int nr_scanned;
411 unsigned long reap = 0;
412
Ingo Molnarf24075b2006-03-23 03:00:34 -0800413 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 spin_lock(&inode_lock);
415 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
416 struct inode *inode;
417
418 if (list_empty(&inode_unused))
419 break;
420
421 inode = list_entry(inode_unused.prev, struct inode, i_list);
422
423 if (inode->i_state || atomic_read(&inode->i_count)) {
424 list_move(&inode->i_list, &inode_unused);
425 continue;
426 }
427 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
428 __iget(inode);
429 spin_unlock(&inode_lock);
430 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800431 reap += invalidate_mapping_pages(&inode->i_data,
432 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 iput(inode);
434 spin_lock(&inode_lock);
435
436 if (inode != list_entry(inode_unused.next,
437 struct inode, i_list))
438 continue; /* wrong inode or list_empty */
439 if (!can_unuse(inode))
440 continue;
441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 list_move(&inode->i_list, &freeable);
443 inode->i_state |= I_FREEING;
444 nr_pruned++;
445 }
446 inodes_stat.nr_unused -= nr_pruned;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700447 if (current_is_kswapd())
448 __count_vm_events(KSWAPD_INODESTEAL, reap);
449 else
450 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 spin_unlock(&inode_lock);
452
453 dispose_list(&freeable);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800454 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455}
456
457/*
458 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
459 * "unused" means that no dentries are referring to the inodes: the files are
460 * not open and the dcache references to those inodes have already been
461 * reclaimed.
462 *
463 * This function is passed the number of inodes to scan, and it returns the
464 * total number of remaining possibly-reclaimable inodes.
465 */
Al Viro27496a82005-10-21 03:20:48 -0400466static int shrink_icache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467{
468 if (nr) {
469 /*
470 * Nasty deadlock avoidance. We may hold various FS locks,
471 * and we don't want to recurse into the FS that called us
472 * in clear_inode() and friends..
473 */
474 if (!(gfp_mask & __GFP_FS))
475 return -1;
476 prune_icache(nr);
477 }
478 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
479}
480
Rusty Russell8e1f9362007-07-17 04:03:17 -0700481static struct shrinker icache_shrinker = {
482 .shrink = shrink_icache_memory,
483 .seeks = DEFAULT_SEEKS,
484};
485
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486static void __wait_on_freeing_inode(struct inode *inode);
487/*
488 * Called with the inode lock held.
489 * NOTE: we are not increasing the inode-refcount, you must call __iget()
490 * by hand after calling find_inode now! This simplifies iunique and won't
491 * add any additional branch in the common code.
492 */
493static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
494{
495 struct hlist_node *node;
496 struct inode * inode = NULL;
497
498repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700499 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 if (inode->i_sb != sb)
501 continue;
502 if (!test(inode, data))
503 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700504 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 __wait_on_freeing_inode(inode);
506 goto repeat;
507 }
508 break;
509 }
510 return node ? inode : NULL;
511}
512
513/*
514 * find_inode_fast is the fast path version of find_inode, see the comment at
515 * iget_locked for details.
516 */
517static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
518{
519 struct hlist_node *node;
520 struct inode * inode = NULL;
521
522repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700523 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 if (inode->i_ino != ino)
525 continue;
526 if (inode->i_sb != sb)
527 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700528 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 __wait_on_freeing_inode(inode);
530 goto repeat;
531 }
532 break;
533 }
534 return node ? inode : NULL;
535}
536
537/**
538 * new_inode - obtain an inode
539 * @sb: superblock
540 *
Mel Gorman769848c2007-07-17 04:03:05 -0700541 * Allocates a new inode for given superblock. The default gfp_mask
542 * for allocations related to inode->i_mapping is GFP_HIGHUSER_PAGECACHE.
543 * If HIGHMEM pages are unsuitable or it is known that pages allocated
544 * for the page cache are not reclaimable or migratable,
545 * mapping_set_gfp_mask() must be called with suitable flags on the
546 * newly created inode's mapping
547 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 */
549struct inode *new_inode(struct super_block *sb)
550{
Jeff Layton866b04f2007-05-08 00:32:29 -0700551 /*
552 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
553 * error if st_ino won't fit in target struct field. Use 32bit counter
554 * here to attempt to avoid that.
555 */
556 static unsigned int last_ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 struct inode * inode;
558
559 spin_lock_prefetch(&inode_lock);
560
561 inode = alloc_inode(sb);
562 if (inode) {
563 spin_lock(&inode_lock);
564 inodes_stat.nr_inodes++;
565 list_add(&inode->i_list, &inode_in_use);
566 list_add(&inode->i_sb_list, &sb->s_inodes);
567 inode->i_ino = ++last_ino;
568 inode->i_state = 0;
569 spin_unlock(&inode_lock);
570 }
571 return inode;
572}
573
574EXPORT_SYMBOL(new_inode);
575
576void unlock_new_inode(struct inode *inode)
577{
Peter Zijlstra14358e62007-10-14 01:38:33 +0200578#ifdef CONFIG_DEBUG_LOCK_ALLOC
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200579 if (inode->i_mode & S_IFDIR) {
580 struct file_system_type *type = inode->i_sb->s_type;
581
582 /*
583 * ensure nobody is actually holding i_mutex
584 */
585 mutex_destroy(&inode->i_mutex);
586 mutex_init(&inode->i_mutex);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200587 lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key);
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200588 }
Peter Zijlstra14358e62007-10-14 01:38:33 +0200589#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 /*
591 * This is special! We do not need the spinlock
592 * when clearing I_LOCK, because we're guaranteed
593 * that nobody else tries to do anything about the
594 * state of the inode when it is locked, as we
595 * just created it (so there can be no old holders
596 * that haven't tested I_LOCK).
597 */
598 inode->i_state &= ~(I_LOCK|I_NEW);
599 wake_up_inode(inode);
600}
601
602EXPORT_SYMBOL(unlock_new_inode);
603
604/*
605 * This is called without the inode lock held.. Be careful.
606 *
607 * We no longer cache the sb_flags in i_flags - see fs.h
608 * -- rmk@arm.uk.linux.org
609 */
610static 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)
611{
612 struct inode * inode;
613
614 inode = alloc_inode(sb);
615 if (inode) {
616 struct inode * old;
617
618 spin_lock(&inode_lock);
619 /* We released the lock, so.. */
620 old = find_inode(sb, head, test, data);
621 if (!old) {
622 if (set(inode, data))
623 goto set_failed;
624
625 inodes_stat.nr_inodes++;
626 list_add(&inode->i_list, &inode_in_use);
627 list_add(&inode->i_sb_list, &sb->s_inodes);
628 hlist_add_head(&inode->i_hash, head);
629 inode->i_state = I_LOCK|I_NEW;
630 spin_unlock(&inode_lock);
631
632 /* Return the locked inode with I_NEW set, the
633 * caller is responsible for filling in the contents
634 */
635 return inode;
636 }
637
638 /*
639 * Uhhuh, somebody else created the same inode under
640 * us. Use the old inode instead of the one we just
641 * allocated.
642 */
643 __iget(old);
644 spin_unlock(&inode_lock);
645 destroy_inode(inode);
646 inode = old;
647 wait_on_inode(inode);
648 }
649 return inode;
650
651set_failed:
652 spin_unlock(&inode_lock);
653 destroy_inode(inode);
654 return NULL;
655}
656
657/*
658 * get_new_inode_fast is the fast path version of get_new_inode, see the
659 * comment at iget_locked for details.
660 */
661static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
662{
663 struct inode * inode;
664
665 inode = alloc_inode(sb);
666 if (inode) {
667 struct inode * old;
668
669 spin_lock(&inode_lock);
670 /* We released the lock, so.. */
671 old = find_inode_fast(sb, head, ino);
672 if (!old) {
673 inode->i_ino = ino;
674 inodes_stat.nr_inodes++;
675 list_add(&inode->i_list, &inode_in_use);
676 list_add(&inode->i_sb_list, &sb->s_inodes);
677 hlist_add_head(&inode->i_hash, head);
678 inode->i_state = I_LOCK|I_NEW;
679 spin_unlock(&inode_lock);
680
681 /* Return the locked inode with I_NEW set, the
682 * caller is responsible for filling in the contents
683 */
684 return inode;
685 }
686
687 /*
688 * Uhhuh, somebody else created the same inode under
689 * us. Use the old inode instead of the one we just
690 * allocated.
691 */
692 __iget(old);
693 spin_unlock(&inode_lock);
694 destroy_inode(inode);
695 inode = old;
696 wait_on_inode(inode);
697 }
698 return inode;
699}
700
Andreas Mohred97bd32006-10-02 02:17:17 -0700701static unsigned long hash(struct super_block *sb, unsigned long hashval)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
703 unsigned long tmp;
704
705 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
706 L1_CACHE_BYTES;
707 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
708 return tmp & I_HASHMASK;
709}
710
711/**
712 * iunique - get a unique inode number
713 * @sb: superblock
714 * @max_reserved: highest reserved inode number
715 *
716 * Obtain an inode number that is unique on the system for a given
717 * superblock. This is used by file systems that have no natural
718 * permanent inode numbering system. An inode number is returned that
719 * is higher than the reserved limit but unique.
720 *
721 * BUGS:
722 * With a large number of inodes live on the file system this function
723 * currently becomes quite slow.
724 */
725ino_t iunique(struct super_block *sb, ino_t max_reserved)
726{
Jeff Layton866b04f2007-05-08 00:32:29 -0700727 /*
728 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
729 * error if st_ino won't fit in target struct field. Use 32bit counter
730 * here to attempt to avoid that.
731 */
732 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 struct inode *inode;
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700734 struct hlist_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700737 spin_lock(&inode_lock);
738 do {
739 if (counter <= max_reserved)
740 counter = max_reserved + 1;
741 res = counter++;
742 head = inode_hashtable + hash(sb, res);
743 inode = find_inode_fast(sb, head, res);
744 } while (inode != NULL);
745 spin_unlock(&inode_lock);
746
747 return res;
748}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749EXPORT_SYMBOL(iunique);
750
751struct inode *igrab(struct inode *inode)
752{
753 spin_lock(&inode_lock);
Jan Blunck4a3b0a42007-02-10 01:44:59 -0800754 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 __iget(inode);
756 else
757 /*
758 * Handle the case where s_op->clear_inode is not been
759 * called yet, and somebody is calling igrab
760 * while the inode is getting freed.
761 */
762 inode = NULL;
763 spin_unlock(&inode_lock);
764 return inode;
765}
766
767EXPORT_SYMBOL(igrab);
768
769/**
770 * ifind - internal function, you want ilookup5() or iget5().
771 * @sb: super block of file system to search
772 * @head: the head of the list to search
773 * @test: callback used for comparisons between inodes
774 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700775 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 *
777 * ifind() searches for the inode specified by @data in the inode
778 * cache. This is a generalized version of ifind_fast() for file systems where
779 * the inode number is not sufficient for unique identification of an inode.
780 *
781 * If the inode is in the cache, the inode is returned with an incremented
782 * reference count.
783 *
784 * Otherwise NULL is returned.
785 *
786 * Note, @test is called with the inode_lock held, so can't sleep.
787 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800788static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700790 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791{
792 struct inode *inode;
793
794 spin_lock(&inode_lock);
795 inode = find_inode(sb, head, test, data);
796 if (inode) {
797 __iget(inode);
798 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700799 if (likely(wait))
800 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 return inode;
802 }
803 spin_unlock(&inode_lock);
804 return NULL;
805}
806
807/**
808 * ifind_fast - internal function, you want ilookup() or iget().
809 * @sb: super block of file system to search
810 * @head: head of the list to search
811 * @ino: inode number to search for
812 *
813 * ifind_fast() searches for the inode @ino in the inode cache. This is for
814 * file systems where the inode number is sufficient for unique identification
815 * of an inode.
816 *
817 * If the inode is in the cache, the inode is returned with an incremented
818 * reference count.
819 *
820 * Otherwise NULL is returned.
821 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800822static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 struct hlist_head *head, unsigned long ino)
824{
825 struct inode *inode;
826
827 spin_lock(&inode_lock);
828 inode = find_inode_fast(sb, head, ino);
829 if (inode) {
830 __iget(inode);
831 spin_unlock(&inode_lock);
832 wait_on_inode(inode);
833 return inode;
834 }
835 spin_unlock(&inode_lock);
836 return NULL;
837}
838
839/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700840 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 * @sb: super block of file system to search
842 * @hashval: hash value (usually inode number) to search for
843 * @test: callback used for comparisons between inodes
844 * @data: opaque data pointer to pass to @test
845 *
846 * ilookup5() uses ifind() to search for the inode specified by @hashval and
847 * @data in the inode cache. This is a generalized version of ilookup() for
848 * file systems where the inode number is not sufficient for unique
849 * identification of an inode.
850 *
851 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700852 * reference count. Note, the inode lock is not waited upon so you have to be
853 * very careful what you do with the returned inode. You probably should be
854 * using ilookup5() instead.
855 *
856 * Otherwise NULL is returned.
857 *
858 * Note, @test is called with the inode_lock held, so can't sleep.
859 */
860struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
861 int (*test)(struct inode *, void *), void *data)
862{
863 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
864
865 return ifind(sb, head, test, data, 0);
866}
867
868EXPORT_SYMBOL(ilookup5_nowait);
869
870/**
871 * ilookup5 - search for an inode in the inode cache
872 * @sb: super block of file system to search
873 * @hashval: hash value (usually inode number) to search for
874 * @test: callback used for comparisons between inodes
875 * @data: opaque data pointer to pass to @test
876 *
877 * ilookup5() uses ifind() to search for the inode specified by @hashval and
878 * @data in the inode cache. This is a generalized version of ilookup() for
879 * file systems where the inode number is not sufficient for unique
880 * identification of an inode.
881 *
882 * If the inode is in the cache, the inode lock is waited upon and the inode is
883 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 *
885 * Otherwise NULL is returned.
886 *
887 * Note, @test is called with the inode_lock held, so can't sleep.
888 */
889struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
890 int (*test)(struct inode *, void *), void *data)
891{
892 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
893
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700894 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895}
896
897EXPORT_SYMBOL(ilookup5);
898
899/**
900 * ilookup - search for an inode in the inode cache
901 * @sb: super block of file system to search
902 * @ino: inode number to search for
903 *
904 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
905 * This is for file systems where the inode number is sufficient for unique
906 * identification of an inode.
907 *
908 * If the inode is in the cache, the inode is returned with an incremented
909 * reference count.
910 *
911 * Otherwise NULL is returned.
912 */
913struct inode *ilookup(struct super_block *sb, unsigned long ino)
914{
915 struct hlist_head *head = inode_hashtable + hash(sb, ino);
916
917 return ifind_fast(sb, head, ino);
918}
919
920EXPORT_SYMBOL(ilookup);
921
922/**
923 * iget5_locked - obtain an inode from a mounted file system
924 * @sb: super block of file system
925 * @hashval: hash value (usually inode number) to get
926 * @test: callback used for comparisons between inodes
927 * @set: callback used to initialize a new struct inode
928 * @data: opaque data pointer to pass to @test and @set
929 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 * iget5_locked() uses ifind() to search for the inode specified by @hashval
931 * and @data in the inode cache and if present it is returned with an increased
932 * reference count. This is a generalized version of iget_locked() for file
933 * systems where the inode number is not sufficient for unique identification
934 * of an inode.
935 *
936 * If the inode is not in cache, get_new_inode() is called to allocate a new
937 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
938 * file system gets to fill it in before unlocking it via unlock_new_inode().
939 *
940 * Note both @test and @set are called with the inode_lock held, so can't sleep.
941 */
942struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
943 int (*test)(struct inode *, void *),
944 int (*set)(struct inode *, void *), void *data)
945{
946 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
947 struct inode *inode;
948
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700949 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 if (inode)
951 return inode;
952 /*
953 * get_new_inode() will do the right thing, re-trying the search
954 * in case it had to block at any point.
955 */
956 return get_new_inode(sb, head, test, set, data);
957}
958
959EXPORT_SYMBOL(iget5_locked);
960
961/**
962 * iget_locked - obtain an inode from a mounted file system
963 * @sb: super block of file system
964 * @ino: inode number to get
965 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
967 * the inode cache and if present it is returned with an increased reference
968 * count. This is for file systems where the inode number is sufficient for
969 * unique identification of an inode.
970 *
971 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
972 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
973 * The file system gets to fill it in before unlocking it via
974 * unlock_new_inode().
975 */
976struct inode *iget_locked(struct super_block *sb, unsigned long ino)
977{
978 struct hlist_head *head = inode_hashtable + hash(sb, ino);
979 struct inode *inode;
980
981 inode = ifind_fast(sb, head, ino);
982 if (inode)
983 return inode;
984 /*
985 * get_new_inode_fast() will do the right thing, re-trying the search
986 * in case it had to block at any point.
987 */
988 return get_new_inode_fast(sb, head, ino);
989}
990
991EXPORT_SYMBOL(iget_locked);
992
993/**
994 * __insert_inode_hash - hash an inode
995 * @inode: unhashed inode
996 * @hashval: unsigned long value used to locate this object in the
997 * inode_hashtable.
998 *
999 * Add an inode to the inode hash for this superblock.
1000 */
1001void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1002{
1003 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1004 spin_lock(&inode_lock);
1005 hlist_add_head(&inode->i_hash, head);
1006 spin_unlock(&inode_lock);
1007}
1008
1009EXPORT_SYMBOL(__insert_inode_hash);
1010
1011/**
1012 * remove_inode_hash - remove an inode from the hash
1013 * @inode: inode to unhash
1014 *
1015 * Remove an inode from the superblock.
1016 */
1017void remove_inode_hash(struct inode *inode)
1018{
1019 spin_lock(&inode_lock);
1020 hlist_del_init(&inode->i_hash);
1021 spin_unlock(&inode_lock);
1022}
1023
1024EXPORT_SYMBOL(remove_inode_hash);
1025
1026/*
1027 * Tell the filesystem that this inode is no longer of any interest and should
1028 * be completely destroyed.
1029 *
1030 * We leave the inode in the inode hash table until *after* the filesystem's
1031 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1032 * instigate) will always find up-to-date information either in the hash or on
1033 * disk.
1034 *
1035 * I_FREEING is set so that no-one will take a new reference to the inode while
1036 * it is being deleted.
1037 */
1038void generic_delete_inode(struct inode *inode)
1039{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001040 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
1042 list_del_init(&inode->i_list);
1043 list_del_init(&inode->i_sb_list);
Andreas Mohred97bd32006-10-02 02:17:17 -07001044 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 inodes_stat.nr_inodes--;
1046 spin_unlock(&inode_lock);
1047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 security_inode_delete(inode);
1049
1050 if (op->delete_inode) {
1051 void (*delete)(struct inode *) = op->delete_inode;
1052 if (!is_bad_inode(inode))
1053 DQUOT_INIT(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001054 /* Filesystems implementing their own
1055 * s_op->delete_inode are required to call
1056 * truncate_inode_pages and clear_inode()
1057 * internally */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 delete(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001059 } else {
1060 truncate_inode_pages(&inode->i_data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 clear_inode(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001062 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 spin_lock(&inode_lock);
1064 hlist_del_init(&inode->i_hash);
1065 spin_unlock(&inode_lock);
1066 wake_up_inode(inode);
Eric Sesterhennb7542f82006-04-02 13:38:18 +02001067 BUG_ON(inode->i_state != I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 destroy_inode(inode);
1069}
1070
1071EXPORT_SYMBOL(generic_delete_inode);
1072
1073static void generic_forget_inode(struct inode *inode)
1074{
1075 struct super_block *sb = inode->i_sb;
1076
1077 if (!hlist_unhashed(&inode->i_hash)) {
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001078 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 list_move(&inode->i_list, &inode_unused);
1080 inodes_stat.nr_unused++;
Christoph Hellwigacb0c852007-05-08 00:25:52 -07001081 if (sb->s_flags & MS_ACTIVE) {
Alexander Viro991114c2005-06-23 00:09:01 -07001082 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001084 }
1085 inode->i_state |= I_WILL_FREE;
1086 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 write_inode_now(inode, 1);
1088 spin_lock(&inode_lock);
Alexander Viro991114c2005-06-23 00:09:01 -07001089 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 inodes_stat.nr_unused--;
1091 hlist_del_init(&inode->i_hash);
1092 }
1093 list_del_init(&inode->i_list);
1094 list_del_init(&inode->i_sb_list);
Alexander Viro991114c2005-06-23 00:09:01 -07001095 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 inodes_stat.nr_inodes--;
1097 spin_unlock(&inode_lock);
1098 if (inode->i_data.nrpages)
1099 truncate_inode_pages(&inode->i_data, 0);
1100 clear_inode(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001101 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 destroy_inode(inode);
1103}
1104
1105/*
1106 * Normal UNIX filesystem behaviour: delete the
1107 * inode when the usage count drops to zero, and
1108 * i_nlink is zero.
1109 */
Mark Fashehcb2c0232005-07-07 17:56:03 -07001110void generic_drop_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111{
1112 if (!inode->i_nlink)
1113 generic_delete_inode(inode);
1114 else
1115 generic_forget_inode(inode);
1116}
1117
Mark Fashehcb2c0232005-07-07 17:56:03 -07001118EXPORT_SYMBOL_GPL(generic_drop_inode);
1119
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120/*
1121 * Called when we're dropping the last reference
1122 * to an inode.
1123 *
1124 * Call the FS "drop()" function, defaulting to
1125 * the legacy UNIX filesystem behaviour..
1126 *
1127 * NOTE! NOTE! NOTE! We're called with the inode lock
1128 * held, and the drop function is supposed to release
1129 * the lock!
1130 */
1131static inline void iput_final(struct inode *inode)
1132{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001133 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 void (*drop)(struct inode *) = generic_drop_inode;
1135
1136 if (op && op->drop_inode)
1137 drop = op->drop_inode;
1138 drop(inode);
1139}
1140
1141/**
1142 * iput - put an inode
1143 * @inode: inode to put
1144 *
1145 * Puts an inode, dropping its usage count. If the inode use count hits
1146 * zero, the inode is then freed and may also be destroyed.
1147 *
1148 * Consequently, iput() can sleep.
1149 */
1150void iput(struct inode *inode)
1151{
1152 if (inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 BUG_ON(inode->i_state == I_CLEAR);
1154
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1156 iput_final(inode);
1157 }
1158}
1159
1160EXPORT_SYMBOL(iput);
1161
1162/**
1163 * bmap - find a block number in a file
1164 * @inode: inode of file
1165 * @block: block to find
1166 *
1167 * Returns the block number on the device holding the inode that
1168 * is the disk block number for the block of the file requested.
1169 * That is, asked for block 4 of inode 1 the function will return the
1170 * disk block relative to the disk start that holds that block of the
1171 * file.
1172 */
1173sector_t bmap(struct inode * inode, sector_t block)
1174{
1175 sector_t res = 0;
1176 if (inode->i_mapping->a_ops->bmap)
1177 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1178 return res;
1179}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180EXPORT_SYMBOL(bmap);
1181
1182/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001183 * touch_atime - update the access time
1184 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001185 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 *
1187 * Update the accessed time on an inode and mark it for writeback.
1188 * This function automatically handles read only file systems and media,
1189 * as well as the "noatime" flag and inode specific "noatime" markers.
1190 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001191void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001193 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 struct timespec now;
1195
Dave Hansencdb70f32008-02-15 14:37:41 -08001196 if (mnt_want_write(mnt))
1197 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001198 if (inode->i_flags & S_NOATIME)
Dave Hansencdb70f32008-02-15 14:37:41 -08001199 goto out;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001200 if (IS_NOATIME(inode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001201 goto out;
Andrew Mortonb2276132006-12-13 00:34:33 -08001202 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001203 goto out;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001204
Dave Hansencdb70f32008-02-15 14:37:41 -08001205 if (mnt->mnt_flags & MNT_NOATIME)
1206 goto out;
1207 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1208 goto out;
1209 if (mnt->mnt_flags & MNT_RELATIME) {
1210 /*
1211 * With relative atime, only update atime if the previous
1212 * atime is earlier than either the ctime or mtime.
1213 */
1214 if (timespec_compare(&inode->i_mtime, &inode->i_atime) < 0 &&
1215 timespec_compare(&inode->i_ctime, &inode->i_atime) < 0)
1216 goto out;
Andrew Mortonb2276132006-12-13 00:34:33 -08001217 }
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001218
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 now = current_fs_time(inode->i_sb);
Valerie Henson47ae32d2006-12-13 00:34:34 -08001220 if (timespec_equal(&inode->i_atime, &now))
Dave Hansencdb70f32008-02-15 14:37:41 -08001221 goto out;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001222
1223 inode->i_atime = now;
1224 mark_inode_dirty_sync(inode);
Dave Hansencdb70f32008-02-15 14:37:41 -08001225out:
1226 mnt_drop_write(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001228EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
1230/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001231 * file_update_time - update mtime and ctime time
1232 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001234 * Update the mtime and ctime members of an inode and mark the inode
1235 * for writeback. Note that this function is meant exclusively for
1236 * usage in the file write path of filesystems, and filesystems may
1237 * choose to explicitly ignore update via this function with the
1238 * S_NOCTIME inode flag, e.g. for network filesystem where these
1239 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 */
1241
Christoph Hellwig870f4812006-01-09 20:52:01 -08001242void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001244 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 struct timespec now;
1246 int sync_it = 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001247 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
1249 if (IS_NOCMTIME(inode))
1250 return;
Dave Hansen20ddee22008-02-15 14:37:43 -08001251
1252 err = mnt_want_write(file->f_path.mnt);
1253 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 return;
1255
1256 now = current_fs_time(inode->i_sb);
Andreas Mohred97bd32006-10-02 02:17:17 -07001257 if (!timespec_equal(&inode->i_mtime, &now)) {
1258 inode->i_mtime = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001260 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
Andreas Mohred97bd32006-10-02 02:17:17 -07001262 if (!timespec_equal(&inode->i_ctime, &now)) {
1263 inode->i_ctime = now;
Christoph Hellwig870f4812006-01-09 20:52:01 -08001264 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001265 }
Christoph Hellwig870f4812006-01-09 20:52:01 -08001266
Jean Noel Cordenner7a224222008-01-28 23:58:27 -05001267 if (IS_I_VERSION(inode)) {
1268 inode_inc_iversion(inode);
1269 sync_it = 1;
1270 }
1271
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 if (sync_it)
1273 mark_inode_dirty_sync(inode);
Dave Hansen20ddee22008-02-15 14:37:43 -08001274 mnt_drop_write(file->f_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275}
1276
Christoph Hellwig870f4812006-01-09 20:52:01 -08001277EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
1279int inode_needs_sync(struct inode *inode)
1280{
1281 if (IS_SYNC(inode))
1282 return 1;
1283 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1284 return 1;
1285 return 0;
1286}
1287
1288EXPORT_SYMBOL(inode_needs_sync);
1289
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290int inode_wait(void *word)
1291{
1292 schedule();
1293 return 0;
1294}
1295
1296/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001297 * If we try to find an inode in the inode hash while it is being
1298 * deleted, we have to wait until the filesystem completes its
1299 * deletion before reporting that it isn't found. This function waits
1300 * until the deletion _might_ have completed. Callers are responsible
1301 * to recheck inode state.
1302 *
1303 * It doesn't matter if I_LOCK is not set initially, a call to
1304 * wake_up_inode() after removing from the hash list will DTRT.
1305 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 * This is called with inode_lock held.
1307 */
1308static void __wait_on_freeing_inode(struct inode *inode)
1309{
1310 wait_queue_head_t *wq;
1311 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1313 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1314 spin_unlock(&inode_lock);
1315 schedule();
1316 finish_wait(wq, &wait.wait);
1317 spin_lock(&inode_lock);
1318}
1319
Mark Fasheh62752ee2006-10-17 10:31:38 +02001320/*
1321 * We rarely want to lock two inodes that do not have a parent/child
1322 * relationship (such as directory, child inode) simultaneously. The
1323 * vast majority of file systems should be able to get along fine
1324 * without this. Do not use these functions except as a last resort.
1325 */
1326void inode_double_lock(struct inode *inode1, struct inode *inode2)
1327{
1328 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1329 if (inode1)
1330 mutex_lock(&inode1->i_mutex);
1331 else if (inode2)
1332 mutex_lock(&inode2->i_mutex);
1333 return;
1334 }
1335
1336 if (inode1 < inode2) {
1337 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1338 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1339 } else {
1340 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1341 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1342 }
1343}
1344EXPORT_SYMBOL(inode_double_lock);
1345
1346void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1347{
1348 if (inode1)
1349 mutex_unlock(&inode1->i_mutex);
1350
1351 if (inode2 && inode2 != inode1)
1352 mutex_unlock(&inode2->i_mutex);
1353}
1354EXPORT_SYMBOL(inode_double_unlock);
1355
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356static __initdata unsigned long ihash_entries;
1357static int __init set_ihash_entries(char *str)
1358{
1359 if (!str)
1360 return 0;
1361 ihash_entries = simple_strtoul(str, &str, 0);
1362 return 1;
1363}
1364__setup("ihash_entries=", set_ihash_entries);
1365
1366/*
1367 * Initialize the waitqueues and inode hash table.
1368 */
1369void __init inode_init_early(void)
1370{
1371 int loop;
1372
1373 /* If hashes are distributed across NUMA nodes, defer
1374 * hash allocation until vmalloc space is available.
1375 */
1376 if (hashdist)
1377 return;
1378
1379 inode_hashtable =
1380 alloc_large_system_hash("Inode-cache",
1381 sizeof(struct hlist_head),
1382 ihash_entries,
1383 14,
1384 HASH_EARLY,
1385 &i_hash_shift,
1386 &i_hash_mask,
1387 0);
1388
1389 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1390 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1391}
1392
Denis Cheng74bf17c2007-10-16 23:26:30 -07001393void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394{
1395 int loop;
1396
1397 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001398 inode_cachep = kmem_cache_create("inode_cache",
1399 sizeof(struct inode),
1400 0,
1401 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1402 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001403 init_once);
Rusty Russell8e1f9362007-07-17 04:03:17 -07001404 register_shrinker(&icache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
1406 /* Hash may have been set up in inode_init_early */
1407 if (!hashdist)
1408 return;
1409
1410 inode_hashtable =
1411 alloc_large_system_hash("Inode-cache",
1412 sizeof(struct hlist_head),
1413 ihash_entries,
1414 14,
1415 0,
1416 &i_hash_shift,
1417 &i_hash_mask,
1418 0);
1419
1420 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1421 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1422}
1423
1424void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1425{
1426 inode->i_mode = mode;
1427 if (S_ISCHR(mode)) {
1428 inode->i_fop = &def_chr_fops;
1429 inode->i_rdev = rdev;
1430 } else if (S_ISBLK(mode)) {
1431 inode->i_fop = &def_blk_fops;
1432 inode->i_rdev = rdev;
1433 } else if (S_ISFIFO(mode))
1434 inode->i_fop = &def_fifo_fops;
1435 else if (S_ISSOCK(mode))
1436 inode->i_fop = &bad_sock_fops;
1437 else
1438 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1439 mode);
1440}
1441EXPORT_SYMBOL(init_special_inode);