blob: 901bad1e5f1210cce8f2bc2a474345f5b1a4072a [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>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050020#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/pagemap.h>
22#include <linux/cdev.h>
23#include <linux/bootmem.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/inotify.h>
Eric Paris3be25f42009-05-21 17:01:26 -040025#include <linux/fsnotify.h>
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -080026#include <linux/mount.h>
Arjan van de Venefaee192009-01-06 07:20:54 -080027#include <linux/async.h>
Al Virof19d4a82009-06-08 19:50:45 -040028#include <linux/posix_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
30/*
31 * This is needed for the following functions:
32 * - inode_has_buffers
33 * - invalidate_inode_buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 * - invalidate_bdev
35 *
36 * FIXME: remove all knowledge of the buffer layer from this file
37 */
38#include <linux/buffer_head.h>
39
40/*
41 * New inode.c implementation.
42 *
43 * This implementation has the basic premise of trying
44 * to be extremely low-overhead and SMP-safe, yet be
45 * simple enough to be "obviously correct".
46 *
47 * Famous last words.
48 */
49
50/* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
51
52/* #define INODE_PARANOIA 1 */
53/* #define INODE_DEBUG 1 */
54
55/*
56 * Inode lookup is no longer as critical as it used to be:
57 * most of the lookups are going to be through the dcache.
58 */
59#define I_HASHBITS i_hash_shift
60#define I_HASHMASK i_hash_mask
61
Eric Dumazetfa3536c2006-03-26 01:37:24 -080062static unsigned int i_hash_mask __read_mostly;
63static unsigned int i_hash_shift __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
65/*
66 * Each inode can be on two separate lists. One is
67 * the hash list of the inode, used for lookups. The
68 * other linked list is the "type" list:
69 * "in_use" - valid inode, i_count > 0, i_nlink > 0
70 * "dirty" - as "in_use" but also dirty
71 * "unused" - valid inode, i_count = 0
72 *
73 * A "dirty" list is maintained for each super block,
74 * allowing for low-overhead inode sync() operations.
75 */
76
77LIST_HEAD(inode_in_use);
78LIST_HEAD(inode_unused);
Eric Dumazetfa3536c2006-03-26 01:37:24 -080079static struct hlist_head *inode_hashtable __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
81/*
82 * A simple spinlock to protect the list manipulations.
83 *
84 * NOTE! You also have to own the lock if you change
85 * the i_state of an inode while it is in use..
86 */
87DEFINE_SPINLOCK(inode_lock);
88
89/*
Ingo Molnarf24075b2006-03-23 03:00:34 -080090 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 * icache shrinking path, and the umount path. Without this exclusion,
92 * by the time prune_icache calls iput for the inode whose pages it has
93 * been invalidating, or by the time it calls clear_inode & destroy_inode
94 * from its final dispose_list, the struct super_block they refer to
95 * (for inode->i_sb->s_op) may already have been freed and reused.
96 */
Adrian Bunkbdfc3262006-03-25 03:06:56 -080097static DEFINE_MUTEX(iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
99/*
100 * Statistics gathering..
101 */
102struct inodes_stat_t inodes_stat;
103
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530104static struct kmem_cache *inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700106static void wake_up_inode(struct inode *inode)
107{
108 /*
109 * Prevent speculative execution through spin_unlock(&inode_lock);
110 */
111 smp_mb();
112 wake_up_bit(&inode->i_state, __I_LOCK);
113}
114
David Chinner2cb15992008-10-30 17:32:23 +1100115/**
116 * inode_init_always - perform inode structure intialisation
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800117 * @sb: superblock inode belongs to
118 * @inode: inode to initialise
David Chinner2cb15992008-10-30 17:32:23 +1100119 *
120 * These are initializations that need to be done on every inode
121 * allocation as the fields are not initialised by slab allocation.
122 */
123struct inode *inode_init_always(struct super_block *sb, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124{
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700125 static const struct address_space_operations empty_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 static struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800127 static const struct file_operations empty_fops;
David Chinner2cb15992008-10-30 17:32:23 +1100128
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530129 struct address_space *const mapping = &inode->i_data;
David Chinner2cb15992008-10-30 17:32:23 +1100130
131 inode->i_sb = sb;
132 inode->i_blkbits = sb->s_blocksize_bits;
133 inode->i_flags = 0;
134 atomic_set(&inode->i_count, 1);
135 inode->i_op = &empty_iops;
136 inode->i_fop = &empty_fops;
137 inode->i_nlink = 1;
Al Viro56ff5ef2008-12-09 09:34:39 -0500138 inode->i_uid = 0;
139 inode->i_gid = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100140 atomic_set(&inode->i_writecount, 0);
141 inode->i_size = 0;
142 inode->i_blocks = 0;
143 inode->i_bytes = 0;
144 inode->i_generation = 0;
145#ifdef CONFIG_QUOTA
146 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
147#endif
148 inode->i_pipe = NULL;
149 inode->i_bdev = NULL;
150 inode->i_cdev = NULL;
151 inode->i_rdev = 0;
152 inode->dirtied_when = 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500153
154 if (security_inode_alloc(inode))
155 goto out_free_inode;
156
157 /* allocate and initialize an i_integrity */
158 if (ima_inode_alloc(inode))
159 goto out_free_security;
David Chinner2cb15992008-10-30 17:32:23 +1100160
161 spin_lock_init(&inode->i_lock);
162 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
163
164 mutex_init(&inode->i_mutex);
165 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
166
167 init_rwsem(&inode->i_alloc_sem);
168 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
169
170 mapping->a_ops = &empty_aops;
171 mapping->host = inode;
172 mapping->flags = 0;
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800173 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
David Chinner2cb15992008-10-30 17:32:23 +1100174 mapping->assoc_mapping = NULL;
175 mapping->backing_dev_info = &default_backing_dev_info;
176 mapping->writeback_index = 0;
177
178 /*
179 * If the block_device provides a backing_dev_info for client
180 * inodes then use that. Otherwise the inode share the bdev's
181 * backing_dev_info.
182 */
183 if (sb->s_bdev) {
184 struct backing_dev_info *bdi;
185
186 bdi = sb->s_bdev->bd_inode_backing_dev_info;
187 if (!bdi)
188 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
189 mapping->backing_dev_info = bdi;
190 }
191 inode->i_private = NULL;
192 inode->i_mapping = mapping;
Al Virof19d4a82009-06-08 19:50:45 -0400193#ifdef CONFIG_FS_POSIX_ACL
194 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
195#endif
David Chinner2cb15992008-10-30 17:32:23 +1100196
Eric Paris3be25f42009-05-21 17:01:26 -0400197#ifdef CONFIG_FSNOTIFY
198 inode->i_fsnotify_mask = 0;
199#endif
200
David Chinner2cb15992008-10-30 17:32:23 +1100201 return inode;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500202
203out_free_security:
204 security_inode_free(inode);
205out_free_inode:
206 if (inode->i_sb->s_op->destroy_inode)
207 inode->i_sb->s_op->destroy_inode(inode);
208 else
209 kmem_cache_free(inode_cachep, (inode));
210 return NULL;
David Chinner2cb15992008-10-30 17:32:23 +1100211}
212EXPORT_SYMBOL(inode_init_always);
213
214static struct inode *alloc_inode(struct super_block *sb)
215{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 struct inode *inode;
217
218 if (sb->s_op->alloc_inode)
219 inode = sb->s_op->alloc_inode(sb);
220 else
David Chinner2cb15992008-10-30 17:32:23 +1100221 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
David Chinner2cb15992008-10-30 17:32:23 +1100223 if (inode)
224 return inode_init_always(sb, inode);
225 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226}
227
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530228void destroy_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200230 BUG_ON(inode_has_buffers(inode));
Hugh Dickinsf07502d2009-06-06 21:18:09 +0100231 ima_inode_free(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 security_inode_free(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400233 fsnotify_inode_delete(inode);
Al Virof19d4a82009-06-08 19:50:45 -0400234#ifdef CONFIG_FS_POSIX_ACL
235 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
236 posix_acl_release(inode->i_acl);
237 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
238 posix_acl_release(inode->i_default_acl);
239#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 if (inode->i_sb->s_op->destroy_inode)
241 inode->i_sb->s_op->destroy_inode(inode);
242 else
243 kmem_cache_free(inode_cachep, (inode));
244}
Christoph Hellwig087e3b02008-10-30 18:24:37 +1100245EXPORT_SYMBOL(destroy_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
247
248/*
249 * These are initializations that only need to be done
250 * once, because the fields are idempotent across use
251 * of the inode, so let the slab aware of that.
252 */
253void inode_init_once(struct inode *inode)
254{
255 memset(inode, 0, sizeof(*inode));
256 INIT_HLIST_NODE(&inode->i_hash);
257 INIT_LIST_HEAD(&inode->i_dentry);
258 INIT_LIST_HEAD(&inode->i_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
Nick Piggin19fd6232008-07-25 19:45:32 -0700260 spin_lock_init(&inode->i_data.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 spin_lock_init(&inode->i_data.i_mmap_lock);
262 INIT_LIST_HEAD(&inode->i_data.private_list);
263 spin_lock_init(&inode->i_data.private_lock);
264 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
265 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 i_size_ordered_init(inode);
Robert Love0eeca282005-07-12 17:06:03 -0400267#ifdef CONFIG_INOTIFY
268 INIT_LIST_HEAD(&inode->inotify_watches);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800269 mutex_init(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400270#endif
Eric Paris3be25f42009-05-21 17:01:26 -0400271#ifdef CONFIG_FSNOTIFY
272 INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries);
273#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275EXPORT_SYMBOL(inode_init_once);
276
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700277static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530279 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
Christoph Lametera35afb82007-05-16 22:10:57 -0700281 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282}
283
284/*
285 * inode_lock must be held
286 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530287void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288{
289 if (atomic_read(&inode->i_count)) {
290 atomic_inc(&inode->i_count);
291 return;
292 }
293 atomic_inc(&inode->i_count);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700294 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 list_move(&inode->i_list, &inode_in_use);
296 inodes_stat.nr_unused--;
297}
298
299/**
300 * clear_inode - clear an inode
301 * @inode: inode to clear
302 *
303 * This is called by the filesystem to tell us
304 * that the inode is no longer useful. We just
305 * terminate it with extreme prejudice.
306 */
307void clear_inode(struct inode *inode)
308{
309 might_sleep();
310 invalidate_inode_buffers(inode);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530311
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200312 BUG_ON(inode->i_data.nrpages);
313 BUG_ON(!(inode->i_state & I_FREEING));
314 BUG_ON(inode->i_state & I_CLEAR);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700315 inode_sync_wait(inode);
Jan Kara9e3509e2009-01-26 16:45:12 +0100316 vfs_dq_drop(inode);
Christoph Hellwigacb0c852007-05-08 00:25:52 -0700317 if (inode->i_sb->s_op->clear_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 inode->i_sb->s_op->clear_inode(inode);
Theodore Ts'oeaf796e2006-09-27 01:50:48 -0700319 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 bd_forget(inode);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700321 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 cd_forget(inode);
323 inode->i_state = I_CLEAR;
324}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325EXPORT_SYMBOL(clear_inode);
326
327/*
328 * dispose_list - dispose of the contents of a local list
329 * @head: the head of the list to free
330 *
331 * Dispose-list gets a local list with local inodes in it, so it doesn't
332 * need to worry about list corruption and SMP locks.
333 */
334static void dispose_list(struct list_head *head)
335{
336 int nr_disposed = 0;
337
338 while (!list_empty(head)) {
339 struct inode *inode;
340
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700341 inode = list_first_entry(head, struct inode, i_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 list_del(&inode->i_list);
343
344 if (inode->i_data.nrpages)
345 truncate_inode_pages(&inode->i_data, 0);
346 clear_inode(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700347
348 spin_lock(&inode_lock);
349 hlist_del_init(&inode->i_hash);
350 list_del_init(&inode->i_sb_list);
351 spin_unlock(&inode_lock);
352
353 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 destroy_inode(inode);
355 nr_disposed++;
356 }
357 spin_lock(&inode_lock);
358 inodes_stat.nr_inodes -= nr_disposed;
359 spin_unlock(&inode_lock);
360}
361
362/*
363 * Invalidate all inodes for a device.
364 */
365static int invalidate_list(struct list_head *head, struct list_head *dispose)
366{
367 struct list_head *next;
368 int busy = 0, count = 0;
369
370 next = head->next;
371 for (;;) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530372 struct list_head *tmp = next;
373 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374
375 /*
376 * We can reschedule here without worrying about the list's
377 * consistency because the per-sb list of inodes must not
Ingo Molnarf24075b2006-03-23 03:00:34 -0800378 * change during umount anymore, and because iprune_mutex keeps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 * shrink_icache_memory() away.
380 */
381 cond_resched_lock(&inode_lock);
382
383 next = next->next;
384 if (tmp == head)
385 break;
386 inode = list_entry(tmp, struct inode, i_sb_list);
Nick Pigginaabb8fd2009-03-11 13:17:36 -0700387 if (inode->i_state & I_NEW)
388 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 invalidate_inode_buffers(inode);
390 if (!atomic_read(&inode->i_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 list_move(&inode->i_list, dispose);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700392 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 inode->i_state |= I_FREEING;
394 count++;
395 continue;
396 }
397 busy = 1;
398 }
399 /* only unused inodes may be cached with i_count zero */
400 inodes_stat.nr_unused -= count;
401 return busy;
402}
403
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404/**
405 * invalidate_inodes - discard the inodes on a device
406 * @sb: superblock
407 *
408 * Discard all of the inodes for a given superblock. If the discard
409 * fails because there are busy inodes then a non zero value is returned.
410 * If the discard is successful all the inodes have been discarded.
411 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530412int invalidate_inodes(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413{
414 int busy;
415 LIST_HEAD(throw_away);
416
Ingo Molnarf24075b2006-03-23 03:00:34 -0800417 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 spin_lock(&inode_lock);
Robert Love0eeca282005-07-12 17:06:03 -0400419 inotify_unmount_inodes(&sb->s_inodes);
Eric Paris164bc612009-05-21 17:01:58 -0400420 fsnotify_unmount_inodes(&sb->s_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 busy = invalidate_list(&sb->s_inodes, &throw_away);
422 spin_unlock(&inode_lock);
423
424 dispose_list(&throw_away);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800425 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
427 return busy;
428}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429EXPORT_SYMBOL(invalidate_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
431static int can_unuse(struct inode *inode)
432{
433 if (inode->i_state)
434 return 0;
435 if (inode_has_buffers(inode))
436 return 0;
437 if (atomic_read(&inode->i_count))
438 return 0;
439 if (inode->i_data.nrpages)
440 return 0;
441 return 1;
442}
443
444/*
445 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
446 * a temporary list and then are freed outside inode_lock by dispose_list().
447 *
448 * Any inodes which are pinned purely because of attached pagecache have their
449 * pagecache removed. We expect the final iput() on that inode to add it to
450 * the front of the inode_unused list. So look for it there and if the
451 * inode is still freeable, proceed. The right inode is found 99.9% of the
452 * time in testing on a 4-way.
453 *
454 * If the inode has metadata buffers attached to mapping->private_list then
455 * try to remove them.
456 */
457static void prune_icache(int nr_to_scan)
458{
459 LIST_HEAD(freeable);
460 int nr_pruned = 0;
461 int nr_scanned;
462 unsigned long reap = 0;
463
Ingo Molnarf24075b2006-03-23 03:00:34 -0800464 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 spin_lock(&inode_lock);
466 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
467 struct inode *inode;
468
469 if (list_empty(&inode_unused))
470 break;
471
472 inode = list_entry(inode_unused.prev, struct inode, i_list);
473
474 if (inode->i_state || atomic_read(&inode->i_count)) {
475 list_move(&inode->i_list, &inode_unused);
476 continue;
477 }
478 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
479 __iget(inode);
480 spin_unlock(&inode_lock);
481 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800482 reap += invalidate_mapping_pages(&inode->i_data,
483 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 iput(inode);
485 spin_lock(&inode_lock);
486
487 if (inode != list_entry(inode_unused.next,
488 struct inode, i_list))
489 continue; /* wrong inode or list_empty */
490 if (!can_unuse(inode))
491 continue;
492 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 list_move(&inode->i_list, &freeable);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700494 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 inode->i_state |= I_FREEING;
496 nr_pruned++;
497 }
498 inodes_stat.nr_unused -= nr_pruned;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700499 if (current_is_kswapd())
500 __count_vm_events(KSWAPD_INODESTEAL, reap);
501 else
502 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 spin_unlock(&inode_lock);
504
505 dispose_list(&freeable);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800506 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507}
508
509/*
510 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
511 * "unused" means that no dentries are referring to the inodes: the files are
512 * not open and the dcache references to those inodes have already been
513 * reclaimed.
514 *
515 * This function is passed the number of inodes to scan, and it returns the
516 * total number of remaining possibly-reclaimable inodes.
517 */
Al Viro27496a82005-10-21 03:20:48 -0400518static int shrink_icache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519{
520 if (nr) {
521 /*
522 * Nasty deadlock avoidance. We may hold various FS locks,
523 * and we don't want to recurse into the FS that called us
524 * in clear_inode() and friends..
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530525 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 if (!(gfp_mask & __GFP_FS))
527 return -1;
528 prune_icache(nr);
529 }
530 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
531}
532
Rusty Russell8e1f9362007-07-17 04:03:17 -0700533static struct shrinker icache_shrinker = {
534 .shrink = shrink_icache_memory,
535 .seeks = DEFAULT_SEEKS,
536};
537
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538static void __wait_on_freeing_inode(struct inode *inode);
539/*
540 * Called with the inode lock held.
541 * NOTE: we are not increasing the inode-refcount, you must call __iget()
542 * by hand after calling find_inode now! This simplifies iunique and won't
543 * add any additional branch in the common code.
544 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530545static struct inode *find_inode(struct super_block *sb,
546 struct hlist_head *head,
547 int (*test)(struct inode *, void *),
548 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549{
550 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530551 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
553repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700554 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 if (inode->i_sb != sb)
556 continue;
557 if (!test(inode, data))
558 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700559 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 __wait_on_freeing_inode(inode);
561 goto repeat;
562 }
563 break;
564 }
565 return node ? inode : NULL;
566}
567
568/*
569 * find_inode_fast is the fast path version of find_inode, see the comment at
570 * iget_locked for details.
571 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530572static struct inode *find_inode_fast(struct super_block *sb,
573 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
575 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530576 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700579 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 if (inode->i_ino != ino)
581 continue;
582 if (inode->i_sb != sb)
583 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700584 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 __wait_on_freeing_inode(inode);
586 goto repeat;
587 }
588 break;
589 }
590 return node ? inode : NULL;
591}
592
David Chinner8290c352008-10-30 17:35:24 +1100593static unsigned long hash(struct super_block *sb, unsigned long hashval)
594{
595 unsigned long tmp;
596
597 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
598 L1_CACHE_BYTES;
599 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
600 return tmp & I_HASHMASK;
601}
602
603static inline void
604__inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
605 struct inode *inode)
606{
607 inodes_stat.nr_inodes++;
608 list_add(&inode->i_list, &inode_in_use);
609 list_add(&inode->i_sb_list, &sb->s_inodes);
610 if (head)
611 hlist_add_head(&inode->i_hash, head);
612}
613
614/**
615 * inode_add_to_lists - add a new inode to relevant lists
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800616 * @sb: superblock inode belongs to
617 * @inode: inode to mark in use
David Chinner8290c352008-10-30 17:35:24 +1100618 *
619 * When an inode is allocated it needs to be accounted for, added to the in use
620 * list, the owning superblock and the inode hash. This needs to be done under
621 * the inode_lock, so export a function to do this rather than the inode lock
622 * itself. We calculate the hash list to add to here so it is all internal
623 * which requires the caller to have already set up the inode number in the
624 * inode to add.
625 */
626void inode_add_to_lists(struct super_block *sb, struct inode *inode)
627{
628 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
629
630 spin_lock(&inode_lock);
631 __inode_add_to_lists(sb, head, inode);
632 spin_unlock(&inode_lock);
633}
634EXPORT_SYMBOL_GPL(inode_add_to_lists);
635
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636/**
637 * new_inode - obtain an inode
638 * @sb: superblock
639 *
Mel Gorman769848c2007-07-17 04:03:05 -0700640 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800641 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700642 * If HIGHMEM pages are unsuitable or it is known that pages allocated
643 * for the page cache are not reclaimable or migratable,
644 * mapping_set_gfp_mask() must be called with suitable flags on the
645 * newly created inode's mapping
646 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 */
648struct inode *new_inode(struct super_block *sb)
649{
Jeff Layton866b04f2007-05-08 00:32:29 -0700650 /*
651 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
652 * error if st_ino won't fit in target struct field. Use 32bit counter
653 * here to attempt to avoid that.
654 */
655 static unsigned int last_ino;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530656 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
658 spin_lock_prefetch(&inode_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530659
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 inode = alloc_inode(sb);
661 if (inode) {
662 spin_lock(&inode_lock);
David Chinner8290c352008-10-30 17:35:24 +1100663 __inode_add_to_lists(sb, NULL, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 inode->i_ino = ++last_ino;
665 inode->i_state = 0;
666 spin_unlock(&inode_lock);
667 }
668 return inode;
669}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670EXPORT_SYMBOL(new_inode);
671
672void unlock_new_inode(struct inode *inode)
673{
Peter Zijlstra14358e62007-10-14 01:38:33 +0200674#ifdef CONFIG_DEBUG_LOCK_ALLOC
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200675 if (inode->i_mode & S_IFDIR) {
676 struct file_system_type *type = inode->i_sb->s_type;
677
Jan Kara9a7aa122009-06-04 15:26:49 +0200678 /* Set new key only if filesystem hasn't already changed it */
679 if (!lockdep_match_class(&inode->i_mutex,
680 &type->i_mutex_key)) {
681 /*
682 * ensure nobody is actually holding i_mutex
683 */
684 mutex_destroy(&inode->i_mutex);
685 mutex_init(&inode->i_mutex);
686 lockdep_set_class(&inode->i_mutex,
687 &type->i_mutex_dir_key);
688 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200689 }
Peter Zijlstra14358e62007-10-14 01:38:33 +0200690#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 /*
692 * This is special! We do not need the spinlock
693 * when clearing I_LOCK, because we're guaranteed
694 * that nobody else tries to do anything about the
695 * state of the inode when it is locked, as we
696 * just created it (so there can be no old holders
697 * that haven't tested I_LOCK).
698 */
Nick Piggin7ef0d732009-03-12 14:31:38 -0700699 WARN_ON((inode->i_state & (I_LOCK|I_NEW)) != (I_LOCK|I_NEW));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 inode->i_state &= ~(I_LOCK|I_NEW);
701 wake_up_inode(inode);
702}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703EXPORT_SYMBOL(unlock_new_inode);
704
705/*
706 * This is called without the inode lock held.. Be careful.
707 *
708 * We no longer cache the sb_flags in i_flags - see fs.h
709 * -- rmk@arm.uk.linux.org
710 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530711static struct inode *get_new_inode(struct super_block *sb,
712 struct hlist_head *head,
713 int (*test)(struct inode *, void *),
714 int (*set)(struct inode *, void *),
715 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530717 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
719 inode = alloc_inode(sb);
720 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530721 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
723 spin_lock(&inode_lock);
724 /* We released the lock, so.. */
725 old = find_inode(sb, head, test, data);
726 if (!old) {
727 if (set(inode, data))
728 goto set_failed;
729
David Chinner8290c352008-10-30 17:35:24 +1100730 __inode_add_to_lists(sb, head, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 inode->i_state = I_LOCK|I_NEW;
732 spin_unlock(&inode_lock);
733
734 /* Return the locked inode with I_NEW set, the
735 * caller is responsible for filling in the contents
736 */
737 return inode;
738 }
739
740 /*
741 * Uhhuh, somebody else created the same inode under
742 * us. Use the old inode instead of the one we just
743 * allocated.
744 */
745 __iget(old);
746 spin_unlock(&inode_lock);
747 destroy_inode(inode);
748 inode = old;
749 wait_on_inode(inode);
750 }
751 return inode;
752
753set_failed:
754 spin_unlock(&inode_lock);
755 destroy_inode(inode);
756 return NULL;
757}
758
759/*
760 * get_new_inode_fast is the fast path version of get_new_inode, see the
761 * comment at iget_locked for details.
762 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530763static struct inode *get_new_inode_fast(struct super_block *sb,
764 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530766 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
768 inode = alloc_inode(sb);
769 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530770 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
772 spin_lock(&inode_lock);
773 /* We released the lock, so.. */
774 old = find_inode_fast(sb, head, ino);
775 if (!old) {
776 inode->i_ino = ino;
David Chinner8290c352008-10-30 17:35:24 +1100777 __inode_add_to_lists(sb, head, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 inode->i_state = I_LOCK|I_NEW;
779 spin_unlock(&inode_lock);
780
781 /* Return the locked inode with I_NEW set, the
782 * caller is responsible for filling in the contents
783 */
784 return inode;
785 }
786
787 /*
788 * Uhhuh, somebody else created the same inode under
789 * us. Use the old inode instead of the one we just
790 * allocated.
791 */
792 __iget(old);
793 spin_unlock(&inode_lock);
794 destroy_inode(inode);
795 inode = old;
796 wait_on_inode(inode);
797 }
798 return inode;
799}
800
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801/**
802 * iunique - get a unique inode number
803 * @sb: superblock
804 * @max_reserved: highest reserved inode number
805 *
806 * Obtain an inode number that is unique on the system for a given
807 * superblock. This is used by file systems that have no natural
808 * permanent inode numbering system. An inode number is returned that
809 * is higher than the reserved limit but unique.
810 *
811 * BUGS:
812 * With a large number of inodes live on the file system this function
813 * currently becomes quite slow.
814 */
815ino_t iunique(struct super_block *sb, ino_t max_reserved)
816{
Jeff Layton866b04f2007-05-08 00:32:29 -0700817 /*
818 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
819 * error if st_ino won't fit in target struct field. Use 32bit counter
820 * here to attempt to avoid that.
821 */
822 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 struct inode *inode;
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700824 struct hlist_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700827 spin_lock(&inode_lock);
828 do {
829 if (counter <= max_reserved)
830 counter = max_reserved + 1;
831 res = counter++;
832 head = inode_hashtable + hash(sb, res);
833 inode = find_inode_fast(sb, head, res);
834 } while (inode != NULL);
835 spin_unlock(&inode_lock);
836
837 return res;
838}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839EXPORT_SYMBOL(iunique);
840
841struct inode *igrab(struct inode *inode)
842{
843 spin_lock(&inode_lock);
Jan Blunck4a3b0a42007-02-10 01:44:59 -0800844 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 __iget(inode);
846 else
847 /*
848 * Handle the case where s_op->clear_inode is not been
849 * called yet, and somebody is calling igrab
850 * while the inode is getting freed.
851 */
852 inode = NULL;
853 spin_unlock(&inode_lock);
854 return inode;
855}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856EXPORT_SYMBOL(igrab);
857
858/**
859 * ifind - internal function, you want ilookup5() or iget5().
860 * @sb: super block of file system to search
861 * @head: the head of the list to search
862 * @test: callback used for comparisons between inodes
863 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700864 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 *
866 * ifind() searches for the inode specified by @data in the inode
867 * cache. This is a generalized version of ifind_fast() for file systems where
868 * the inode number is not sufficient for unique identification of an inode.
869 *
870 * If the inode is in the cache, the inode is returned with an incremented
871 * reference count.
872 *
873 * Otherwise NULL is returned.
874 *
875 * Note, @test is called with the inode_lock held, so can't sleep.
876 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800877static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700879 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880{
881 struct inode *inode;
882
883 spin_lock(&inode_lock);
884 inode = find_inode(sb, head, test, data);
885 if (inode) {
886 __iget(inode);
887 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700888 if (likely(wait))
889 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 return inode;
891 }
892 spin_unlock(&inode_lock);
893 return NULL;
894}
895
896/**
897 * ifind_fast - internal function, you want ilookup() or iget().
898 * @sb: super block of file system to search
899 * @head: head of the list to search
900 * @ino: inode number to search for
901 *
902 * ifind_fast() searches for the inode @ino in the inode cache. This is for
903 * file systems where the inode number is sufficient for unique identification
904 * of an inode.
905 *
906 * If the inode is in the cache, the inode is returned with an incremented
907 * reference count.
908 *
909 * Otherwise NULL is returned.
910 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800911static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 struct hlist_head *head, unsigned long ino)
913{
914 struct inode *inode;
915
916 spin_lock(&inode_lock);
917 inode = find_inode_fast(sb, head, ino);
918 if (inode) {
919 __iget(inode);
920 spin_unlock(&inode_lock);
921 wait_on_inode(inode);
922 return inode;
923 }
924 spin_unlock(&inode_lock);
925 return NULL;
926}
927
928/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700929 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 * @sb: super block of file system to search
931 * @hashval: hash value (usually inode number) to search for
932 * @test: callback used for comparisons between inodes
933 * @data: opaque data pointer to pass to @test
934 *
935 * ilookup5() uses ifind() to search for the inode specified by @hashval and
936 * @data in the inode cache. This is a generalized version of ilookup() for
937 * file systems where the inode number is not sufficient for unique
938 * identification of an inode.
939 *
940 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700941 * reference count. Note, the inode lock is not waited upon so you have to be
942 * very careful what you do with the returned inode. You probably should be
943 * using ilookup5() instead.
944 *
945 * Otherwise NULL is returned.
946 *
947 * Note, @test is called with the inode_lock held, so can't sleep.
948 */
949struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
950 int (*test)(struct inode *, void *), void *data)
951{
952 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
953
954 return ifind(sb, head, test, data, 0);
955}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700956EXPORT_SYMBOL(ilookup5_nowait);
957
958/**
959 * ilookup5 - search for an inode in the inode cache
960 * @sb: super block of file system to search
961 * @hashval: hash value (usually inode number) to search for
962 * @test: callback used for comparisons between inodes
963 * @data: opaque data pointer to pass to @test
964 *
965 * ilookup5() uses ifind() to search for the inode specified by @hashval and
966 * @data in the inode cache. This is a generalized version of ilookup() for
967 * file systems where the inode number is not sufficient for unique
968 * identification of an inode.
969 *
970 * If the inode is in the cache, the inode lock is waited upon and the inode is
971 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 *
973 * Otherwise NULL is returned.
974 *
975 * Note, @test is called with the inode_lock held, so can't sleep.
976 */
977struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
978 int (*test)(struct inode *, void *), void *data)
979{
980 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
981
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700982 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984EXPORT_SYMBOL(ilookup5);
985
986/**
987 * ilookup - search for an inode in the inode cache
988 * @sb: super block of file system to search
989 * @ino: inode number to search for
990 *
991 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
992 * This is for file systems where the inode number is sufficient for unique
993 * identification of an inode.
994 *
995 * If the inode is in the cache, the inode is returned with an incremented
996 * reference count.
997 *
998 * Otherwise NULL is returned.
999 */
1000struct inode *ilookup(struct super_block *sb, unsigned long ino)
1001{
1002 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1003
1004 return ifind_fast(sb, head, ino);
1005}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006EXPORT_SYMBOL(ilookup);
1007
1008/**
1009 * iget5_locked - obtain an inode from a mounted file system
1010 * @sb: super block of file system
1011 * @hashval: hash value (usually inode number) to get
1012 * @test: callback used for comparisons between inodes
1013 * @set: callback used to initialize a new struct inode
1014 * @data: opaque data pointer to pass to @test and @set
1015 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1017 * and @data in the inode cache and if present it is returned with an increased
1018 * reference count. This is a generalized version of iget_locked() for file
1019 * systems where the inode number is not sufficient for unique identification
1020 * of an inode.
1021 *
1022 * If the inode is not in cache, get_new_inode() is called to allocate a new
1023 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1024 * file system gets to fill it in before unlocking it via unlock_new_inode().
1025 *
1026 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1027 */
1028struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1029 int (*test)(struct inode *, void *),
1030 int (*set)(struct inode *, void *), void *data)
1031{
1032 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1033 struct inode *inode;
1034
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001035 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 if (inode)
1037 return inode;
1038 /*
1039 * get_new_inode() will do the right thing, re-trying the search
1040 * in case it had to block at any point.
1041 */
1042 return get_new_inode(sb, head, test, set, data);
1043}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044EXPORT_SYMBOL(iget5_locked);
1045
1046/**
1047 * iget_locked - obtain an inode from a mounted file system
1048 * @sb: super block of file system
1049 * @ino: inode number to get
1050 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1052 * the inode cache and if present it is returned with an increased reference
1053 * count. This is for file systems where the inode number is sufficient for
1054 * unique identification of an inode.
1055 *
1056 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1057 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1058 * The file system gets to fill it in before unlocking it via
1059 * unlock_new_inode().
1060 */
1061struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1062{
1063 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1064 struct inode *inode;
1065
1066 inode = ifind_fast(sb, head, ino);
1067 if (inode)
1068 return inode;
1069 /*
1070 * get_new_inode_fast() will do the right thing, re-trying the search
1071 * in case it had to block at any point.
1072 */
1073 return get_new_inode_fast(sb, head, ino);
1074}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075EXPORT_SYMBOL(iget_locked);
1076
Al Viro261bca82008-12-30 01:48:21 -05001077int insert_inode_locked(struct inode *inode)
1078{
1079 struct super_block *sb = inode->i_sb;
1080 ino_t ino = inode->i_ino;
1081 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001082
1083 inode->i_state |= I_LOCK|I_NEW;
1084 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001085 struct hlist_node *node;
1086 struct inode *old = NULL;
Al Viro261bca82008-12-30 01:48:21 -05001087 spin_lock(&inode_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001088 hlist_for_each_entry(old, node, head, i_hash) {
1089 if (old->i_ino != ino)
1090 continue;
1091 if (old->i_sb != sb)
1092 continue;
1093 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1094 continue;
1095 break;
1096 }
1097 if (likely(!node)) {
Al Viro261bca82008-12-30 01:48:21 -05001098 hlist_add_head(&inode->i_hash, head);
1099 spin_unlock(&inode_lock);
1100 return 0;
1101 }
1102 __iget(old);
1103 spin_unlock(&inode_lock);
1104 wait_on_inode(old);
1105 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1106 iput(old);
1107 return -EBUSY;
1108 }
1109 iput(old);
1110 }
1111}
Al Viro261bca82008-12-30 01:48:21 -05001112EXPORT_SYMBOL(insert_inode_locked);
1113
1114int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1115 int (*test)(struct inode *, void *), void *data)
1116{
1117 struct super_block *sb = inode->i_sb;
1118 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001119
1120 inode->i_state |= I_LOCK|I_NEW;
1121
1122 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001123 struct hlist_node *node;
1124 struct inode *old = NULL;
1125
Al Viro261bca82008-12-30 01:48:21 -05001126 spin_lock(&inode_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001127 hlist_for_each_entry(old, node, head, i_hash) {
1128 if (old->i_sb != sb)
1129 continue;
1130 if (!test(old, data))
1131 continue;
1132 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1133 continue;
1134 break;
1135 }
1136 if (likely(!node)) {
Al Viro261bca82008-12-30 01:48:21 -05001137 hlist_add_head(&inode->i_hash, head);
1138 spin_unlock(&inode_lock);
1139 return 0;
1140 }
1141 __iget(old);
1142 spin_unlock(&inode_lock);
1143 wait_on_inode(old);
1144 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1145 iput(old);
1146 return -EBUSY;
1147 }
1148 iput(old);
1149 }
1150}
Al Viro261bca82008-12-30 01:48:21 -05001151EXPORT_SYMBOL(insert_inode_locked4);
1152
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153/**
1154 * __insert_inode_hash - hash an inode
1155 * @inode: unhashed inode
1156 * @hashval: unsigned long value used to locate this object in the
1157 * inode_hashtable.
1158 *
1159 * Add an inode to the inode hash for this superblock.
1160 */
1161void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1162{
1163 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1164 spin_lock(&inode_lock);
1165 hlist_add_head(&inode->i_hash, head);
1166 spin_unlock(&inode_lock);
1167}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168EXPORT_SYMBOL(__insert_inode_hash);
1169
1170/**
1171 * remove_inode_hash - remove an inode from the hash
1172 * @inode: inode to unhash
1173 *
1174 * Remove an inode from the superblock.
1175 */
1176void remove_inode_hash(struct inode *inode)
1177{
1178 spin_lock(&inode_lock);
1179 hlist_del_init(&inode->i_hash);
1180 spin_unlock(&inode_lock);
1181}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182EXPORT_SYMBOL(remove_inode_hash);
1183
1184/*
1185 * Tell the filesystem that this inode is no longer of any interest and should
1186 * be completely destroyed.
1187 *
1188 * We leave the inode in the inode hash table until *after* the filesystem's
1189 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1190 * instigate) will always find up-to-date information either in the hash or on
1191 * disk.
1192 *
1193 * I_FREEING is set so that no-one will take a new reference to the inode while
1194 * it is being deleted.
1195 */
Arjan van de Venb32714b2009-01-09 07:04:15 -08001196void generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001198 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199
Arjan van de Venb32714b2009-01-09 07:04:15 -08001200 list_del_init(&inode->i_list);
1201 list_del_init(&inode->i_sb_list);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001202 WARN_ON(inode->i_state & I_NEW);
Arjan van de Venb32714b2009-01-09 07:04:15 -08001203 inode->i_state |= I_FREEING;
1204 inodes_stat.nr_inodes--;
1205 spin_unlock(&inode_lock);
1206
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 security_inode_delete(inode);
1208
1209 if (op->delete_inode) {
1210 void (*delete)(struct inode *) = op->delete_inode;
1211 if (!is_bad_inode(inode))
Jan Kara9e3509e2009-01-26 16:45:12 +01001212 vfs_dq_init(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001213 /* Filesystems implementing their own
1214 * s_op->delete_inode are required to call
1215 * truncate_inode_pages and clear_inode()
1216 * internally */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 delete(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001218 } else {
1219 truncate_inode_pages(&inode->i_data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 clear_inode(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001221 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 spin_lock(&inode_lock);
1223 hlist_del_init(&inode->i_hash);
1224 spin_unlock(&inode_lock);
1225 wake_up_inode(inode);
Eric Sesterhennb7542f82006-04-02 13:38:18 +02001226 BUG_ON(inode->i_state != I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 destroy_inode(inode);
1228}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229EXPORT_SYMBOL(generic_delete_inode);
1230
1231static void generic_forget_inode(struct inode *inode)
1232{
1233 struct super_block *sb = inode->i_sb;
1234
1235 if (!hlist_unhashed(&inode->i_hash)) {
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001236 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 list_move(&inode->i_list, &inode_unused);
1238 inodes_stat.nr_unused++;
Christoph Hellwigacb0c852007-05-08 00:25:52 -07001239 if (sb->s_flags & MS_ACTIVE) {
Alexander Viro991114c2005-06-23 00:09:01 -07001240 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001242 }
Nick Piggin7ef0d732009-03-12 14:31:38 -07001243 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001244 inode->i_state |= I_WILL_FREE;
1245 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 write_inode_now(inode, 1);
1247 spin_lock(&inode_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001248 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001249 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 inodes_stat.nr_unused--;
1251 hlist_del_init(&inode->i_hash);
1252 }
1253 list_del_init(&inode->i_list);
1254 list_del_init(&inode->i_sb_list);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001255 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001256 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 inodes_stat.nr_inodes--;
1258 spin_unlock(&inode_lock);
1259 if (inode->i_data.nrpages)
1260 truncate_inode_pages(&inode->i_data, 0);
1261 clear_inode(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001262 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 destroy_inode(inode);
1264}
1265
1266/*
1267 * Normal UNIX filesystem behaviour: delete the
1268 * inode when the usage count drops to zero, and
1269 * i_nlink is zero.
1270 */
Mark Fashehcb2c0232005-07-07 17:56:03 -07001271void generic_drop_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272{
1273 if (!inode->i_nlink)
1274 generic_delete_inode(inode);
1275 else
1276 generic_forget_inode(inode);
1277}
Mark Fashehcb2c0232005-07-07 17:56:03 -07001278EXPORT_SYMBOL_GPL(generic_drop_inode);
1279
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280/*
1281 * Called when we're dropping the last reference
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301282 * to an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 *
1284 * Call the FS "drop()" function, defaulting to
1285 * the legacy UNIX filesystem behaviour..
1286 *
1287 * NOTE! NOTE! NOTE! We're called with the inode lock
1288 * held, and the drop function is supposed to release
1289 * the lock!
1290 */
1291static inline void iput_final(struct inode *inode)
1292{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001293 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 void (*drop)(struct inode *) = generic_drop_inode;
1295
1296 if (op && op->drop_inode)
1297 drop = op->drop_inode;
1298 drop(inode);
1299}
1300
1301/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301302 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 * @inode: inode to put
1304 *
1305 * Puts an inode, dropping its usage count. If the inode use count hits
1306 * zero, the inode is then freed and may also be destroyed.
1307 *
1308 * Consequently, iput() can sleep.
1309 */
1310void iput(struct inode *inode)
1311{
1312 if (inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 BUG_ON(inode->i_state == I_CLEAR);
1314
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1316 iput_final(inode);
1317 }
1318}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319EXPORT_SYMBOL(iput);
1320
1321/**
1322 * bmap - find a block number in a file
1323 * @inode: inode of file
1324 * @block: block to find
1325 *
1326 * Returns the block number on the device holding the inode that
1327 * is the disk block number for the block of the file requested.
1328 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301329 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 * file.
1331 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301332sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333{
1334 sector_t res = 0;
1335 if (inode->i_mapping->a_ops->bmap)
1336 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1337 return res;
1338}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339EXPORT_SYMBOL(bmap);
1340
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001341/*
1342 * With relative atime, only update atime if the previous atime is
1343 * earlier than either the ctime or mtime or if at least a day has
1344 * passed since the last atime update.
1345 */
1346static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1347 struct timespec now)
1348{
1349
1350 if (!(mnt->mnt_flags & MNT_RELATIME))
1351 return 1;
1352 /*
1353 * Is mtime younger than atime? If yes, update atime:
1354 */
1355 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1356 return 1;
1357 /*
1358 * Is ctime younger than atime? If yes, update atime:
1359 */
1360 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1361 return 1;
1362
1363 /*
1364 * Is the previous atime value older than a day? If yes,
1365 * update atime:
1366 */
1367 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1368 return 1;
1369 /*
1370 * Good, we can skip the atime update:
1371 */
1372 return 0;
1373}
1374
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001376 * touch_atime - update the access time
1377 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001378 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 *
1380 * Update the accessed time on an inode and mark it for writeback.
1381 * This function automatically handles read only file systems and media,
1382 * as well as the "noatime" flag and inode specific "noatime" markers.
1383 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001384void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001386 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 struct timespec now;
1388
Dave Hansencdb70f32008-02-15 14:37:41 -08001389 if (mnt_want_write(mnt))
1390 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001391 if (inode->i_flags & S_NOATIME)
Dave Hansencdb70f32008-02-15 14:37:41 -08001392 goto out;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001393 if (IS_NOATIME(inode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001394 goto out;
Andrew Mortonb2276132006-12-13 00:34:33 -08001395 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001396 goto out;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001397
Dave Hansencdb70f32008-02-15 14:37:41 -08001398 if (mnt->mnt_flags & MNT_NOATIME)
1399 goto out;
1400 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1401 goto out;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001402
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 now = current_fs_time(inode->i_sb);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001404
1405 if (!relatime_need_update(mnt, inode, now))
1406 goto out;
1407
Valerie Henson47ae32d2006-12-13 00:34:34 -08001408 if (timespec_equal(&inode->i_atime, &now))
Dave Hansencdb70f32008-02-15 14:37:41 -08001409 goto out;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001410
1411 inode->i_atime = now;
1412 mark_inode_dirty_sync(inode);
Dave Hansencdb70f32008-02-15 14:37:41 -08001413out:
1414 mnt_drop_write(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001416EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
1418/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001419 * file_update_time - update mtime and ctime time
1420 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001422 * Update the mtime and ctime members of an inode and mark the inode
1423 * for writeback. Note that this function is meant exclusively for
1424 * usage in the file write path of filesystems, and filesystems may
1425 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001426 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Christoph Hellwig870f4812006-01-09 20:52:01 -08001427 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 */
1429
Christoph Hellwig870f4812006-01-09 20:52:01 -08001430void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001432 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 struct timespec now;
1434 int sync_it = 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001435 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
1437 if (IS_NOCMTIME(inode))
1438 return;
Dave Hansen20ddee22008-02-15 14:37:43 -08001439
npiggin@suse.de96029c42009-04-26 20:25:55 +10001440 err = mnt_want_write_file(file);
Dave Hansen20ddee22008-02-15 14:37:43 -08001441 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 return;
1443
1444 now = current_fs_time(inode->i_sb);
Andreas Mohred97bd32006-10-02 02:17:17 -07001445 if (!timespec_equal(&inode->i_mtime, &now)) {
1446 inode->i_mtime = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001448 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
Andreas Mohred97bd32006-10-02 02:17:17 -07001450 if (!timespec_equal(&inode->i_ctime, &now)) {
1451 inode->i_ctime = now;
Christoph Hellwig870f4812006-01-09 20:52:01 -08001452 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001453 }
Christoph Hellwig870f4812006-01-09 20:52:01 -08001454
Jean Noel Cordenner7a224222008-01-28 23:58:27 -05001455 if (IS_I_VERSION(inode)) {
1456 inode_inc_iversion(inode);
1457 sync_it = 1;
1458 }
1459
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 if (sync_it)
1461 mark_inode_dirty_sync(inode);
Dave Hansen20ddee22008-02-15 14:37:43 -08001462 mnt_drop_write(file->f_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001464EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465
1466int inode_needs_sync(struct inode *inode)
1467{
1468 if (IS_SYNC(inode))
1469 return 1;
1470 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1471 return 1;
1472 return 0;
1473}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474EXPORT_SYMBOL(inode_needs_sync);
1475
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476int inode_wait(void *word)
1477{
1478 schedule();
1479 return 0;
1480}
Stephen Rothwelld44dab82008-11-10 17:06:05 +11001481EXPORT_SYMBOL(inode_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482
1483/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001484 * If we try to find an inode in the inode hash while it is being
1485 * deleted, we have to wait until the filesystem completes its
1486 * deletion before reporting that it isn't found. This function waits
1487 * until the deletion _might_ have completed. Callers are responsible
1488 * to recheck inode state.
1489 *
1490 * It doesn't matter if I_LOCK is not set initially, a call to
1491 * wake_up_inode() after removing from the hash list will DTRT.
1492 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 * This is called with inode_lock held.
1494 */
1495static void __wait_on_freeing_inode(struct inode *inode)
1496{
1497 wait_queue_head_t *wq;
1498 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1500 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1501 spin_unlock(&inode_lock);
1502 schedule();
1503 finish_wait(wq, &wait.wait);
1504 spin_lock(&inode_lock);
1505}
1506
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507static __initdata unsigned long ihash_entries;
1508static int __init set_ihash_entries(char *str)
1509{
1510 if (!str)
1511 return 0;
1512 ihash_entries = simple_strtoul(str, &str, 0);
1513 return 1;
1514}
1515__setup("ihash_entries=", set_ihash_entries);
1516
1517/*
1518 * Initialize the waitqueues and inode hash table.
1519 */
1520void __init inode_init_early(void)
1521{
1522 int loop;
1523
1524 /* If hashes are distributed across NUMA nodes, defer
1525 * hash allocation until vmalloc space is available.
1526 */
1527 if (hashdist)
1528 return;
1529
1530 inode_hashtable =
1531 alloc_large_system_hash("Inode-cache",
1532 sizeof(struct hlist_head),
1533 ihash_entries,
1534 14,
1535 HASH_EARLY,
1536 &i_hash_shift,
1537 &i_hash_mask,
1538 0);
1539
1540 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1541 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1542}
1543
Denis Cheng74bf17c2007-10-16 23:26:30 -07001544void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545{
1546 int loop;
1547
1548 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001549 inode_cachep = kmem_cache_create("inode_cache",
1550 sizeof(struct inode),
1551 0,
1552 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1553 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001554 init_once);
Rusty Russell8e1f9362007-07-17 04:03:17 -07001555 register_shrinker(&icache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
1557 /* Hash may have been set up in inode_init_early */
1558 if (!hashdist)
1559 return;
1560
1561 inode_hashtable =
1562 alloc_large_system_hash("Inode-cache",
1563 sizeof(struct hlist_head),
1564 ihash_entries,
1565 14,
1566 0,
1567 &i_hash_shift,
1568 &i_hash_mask,
1569 0);
1570
1571 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1572 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1573}
1574
1575void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1576{
1577 inode->i_mode = mode;
1578 if (S_ISCHR(mode)) {
1579 inode->i_fop = &def_chr_fops;
1580 inode->i_rdev = rdev;
1581 } else if (S_ISBLK(mode)) {
1582 inode->i_fop = &def_blk_fops;
1583 inode->i_rdev = rdev;
1584 } else if (S_ISFIFO(mode))
1585 inode->i_fop = &def_fifo_fops;
1586 else if (S_ISSOCK(mode))
1587 inode->i_fop = &bad_sock_fops;
1588 else
1589 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1590 mode);
1591}
1592EXPORT_SYMBOL(init_special_inode);