blob: f5ff71cb3e2b3bcd1ba92512022c41f2f9571208 [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 */
Christoph Hellwig54e34622009-08-07 14:38:25 -0300123int 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;
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700126 static const struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800127 static const struct file_operations empty_fops;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530128 struct address_space *const mapping = &inode->i_data;
David Chinner2cb15992008-10-30 17:32:23 +1100129
130 inode->i_sb = sb;
131 inode->i_blkbits = sb->s_blocksize_bits;
132 inode->i_flags = 0;
133 atomic_set(&inode->i_count, 1);
134 inode->i_op = &empty_iops;
135 inode->i_fop = &empty_fops;
136 inode->i_nlink = 1;
Al Viro56ff5ef2008-12-09 09:34:39 -0500137 inode->i_uid = 0;
138 inode->i_gid = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100139 atomic_set(&inode->i_writecount, 0);
140 inode->i_size = 0;
141 inode->i_blocks = 0;
142 inode->i_bytes = 0;
143 inode->i_generation = 0;
144#ifdef CONFIG_QUOTA
145 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
146#endif
147 inode->i_pipe = NULL;
148 inode->i_bdev = NULL;
149 inode->i_cdev = NULL;
150 inode->i_rdev = 0;
151 inode->dirtied_when = 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500152
153 if (security_inode_alloc(inode))
Christoph Hellwig54e34622009-08-07 14:38:25 -0300154 goto out;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500155
156 /* allocate and initialize an i_integrity */
157 if (ima_inode_alloc(inode))
158 goto out_free_security;
David Chinner2cb15992008-10-30 17:32:23 +1100159
160 spin_lock_init(&inode->i_lock);
161 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
162
163 mutex_init(&inode->i_mutex);
164 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
165
166 init_rwsem(&inode->i_alloc_sem);
167 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
168
169 mapping->a_ops = &empty_aops;
170 mapping->host = inode;
171 mapping->flags = 0;
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800172 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
David Chinner2cb15992008-10-30 17:32:23 +1100173 mapping->assoc_mapping = NULL;
174 mapping->backing_dev_info = &default_backing_dev_info;
175 mapping->writeback_index = 0;
176
177 /*
178 * If the block_device provides a backing_dev_info for client
179 * inodes then use that. Otherwise the inode share the bdev's
180 * backing_dev_info.
181 */
182 if (sb->s_bdev) {
183 struct backing_dev_info *bdi;
184
Jens Axboe2c96ce92009-09-15 09:43:56 +0200185 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
David Chinner2cb15992008-10-30 17:32:23 +1100186 mapping->backing_dev_info = bdi;
187 }
188 inode->i_private = NULL;
189 inode->i_mapping = mapping;
Al Virof19d4a82009-06-08 19:50:45 -0400190#ifdef CONFIG_FS_POSIX_ACL
191 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
192#endif
David Chinner2cb15992008-10-30 17:32:23 +1100193
Eric Paris3be25f42009-05-21 17:01:26 -0400194#ifdef CONFIG_FSNOTIFY
195 inode->i_fsnotify_mask = 0;
196#endif
197
Christoph Hellwig54e34622009-08-07 14:38:25 -0300198 return 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500199
200out_free_security:
201 security_inode_free(inode);
Christoph Hellwig54e34622009-08-07 14:38:25 -0300202out:
203 return -ENOMEM;
David Chinner2cb15992008-10-30 17:32:23 +1100204}
205EXPORT_SYMBOL(inode_init_always);
206
207static struct inode *alloc_inode(struct super_block *sb)
208{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 struct inode *inode;
210
211 if (sb->s_op->alloc_inode)
212 inode = sb->s_op->alloc_inode(sb);
213 else
David Chinner2cb15992008-10-30 17:32:23 +1100214 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
Christoph Hellwig54e34622009-08-07 14:38:25 -0300216 if (!inode)
217 return NULL;
218
219 if (unlikely(inode_init_always(sb, inode))) {
220 if (inode->i_sb->s_op->destroy_inode)
221 inode->i_sb->s_op->destroy_inode(inode);
222 else
223 kmem_cache_free(inode_cachep, inode);
224 return NULL;
225 }
226
227 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228}
229
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300230void __destroy_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200232 BUG_ON(inode_has_buffers(inode));
Hugh Dickinsf07502d2009-06-06 21:18:09 +0100233 ima_inode_free(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 security_inode_free(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400235 fsnotify_inode_delete(inode);
Al Virof19d4a82009-06-08 19:50:45 -0400236#ifdef CONFIG_FS_POSIX_ACL
237 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
238 posix_acl_release(inode->i_acl);
239 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
240 posix_acl_release(inode->i_default_acl);
241#endif
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300242}
243EXPORT_SYMBOL(__destroy_inode);
244
245void destroy_inode(struct inode *inode)
246{
247 __destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 if (inode->i_sb->s_op->destroy_inode)
249 inode->i_sb->s_op->destroy_inode(inode);
250 else
251 kmem_cache_free(inode_cachep, (inode));
252}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
254/*
255 * These are initializations that only need to be done
256 * once, because the fields are idempotent across use
257 * of the inode, so let the slab aware of that.
258 */
259void inode_init_once(struct inode *inode)
260{
261 memset(inode, 0, sizeof(*inode));
262 INIT_HLIST_NODE(&inode->i_hash);
263 INIT_LIST_HEAD(&inode->i_dentry);
264 INIT_LIST_HEAD(&inode->i_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
Nick Piggin19fd6232008-07-25 19:45:32 -0700266 spin_lock_init(&inode->i_data.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 spin_lock_init(&inode->i_data.i_mmap_lock);
268 INIT_LIST_HEAD(&inode->i_data.private_list);
269 spin_lock_init(&inode->i_data.private_lock);
270 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
271 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 i_size_ordered_init(inode);
Robert Love0eeca282005-07-12 17:06:03 -0400273#ifdef CONFIG_INOTIFY
274 INIT_LIST_HEAD(&inode->inotify_watches);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800275 mutex_init(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400276#endif
Eric Paris3be25f42009-05-21 17:01:26 -0400277#ifdef CONFIG_FSNOTIFY
278 INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries);
279#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281EXPORT_SYMBOL(inode_init_once);
282
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700283static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530285 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
Christoph Lametera35afb82007-05-16 22:10:57 -0700287 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288}
289
290/*
291 * inode_lock must be held
292 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530293void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
295 if (atomic_read(&inode->i_count)) {
296 atomic_inc(&inode->i_count);
297 return;
298 }
299 atomic_inc(&inode->i_count);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700300 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 list_move(&inode->i_list, &inode_in_use);
302 inodes_stat.nr_unused--;
303}
304
305/**
306 * clear_inode - clear an inode
307 * @inode: inode to clear
308 *
309 * This is called by the filesystem to tell us
310 * that the inode is no longer useful. We just
311 * terminate it with extreme prejudice.
312 */
313void clear_inode(struct inode *inode)
314{
315 might_sleep();
316 invalidate_inode_buffers(inode);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530317
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200318 BUG_ON(inode->i_data.nrpages);
319 BUG_ON(!(inode->i_state & I_FREEING));
320 BUG_ON(inode->i_state & I_CLEAR);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700321 inode_sync_wait(inode);
Jan Kara9e3509e2009-01-26 16:45:12 +0100322 vfs_dq_drop(inode);
Christoph Hellwigacb0c852007-05-08 00:25:52 -0700323 if (inode->i_sb->s_op->clear_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 inode->i_sb->s_op->clear_inode(inode);
Theodore Ts'oeaf796e2006-09-27 01:50:48 -0700325 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 bd_forget(inode);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700327 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 cd_forget(inode);
329 inode->i_state = I_CLEAR;
330}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331EXPORT_SYMBOL(clear_inode);
332
333/*
334 * dispose_list - dispose of the contents of a local list
335 * @head: the head of the list to free
336 *
337 * Dispose-list gets a local list with local inodes in it, so it doesn't
338 * need to worry about list corruption and SMP locks.
339 */
340static void dispose_list(struct list_head *head)
341{
342 int nr_disposed = 0;
343
344 while (!list_empty(head)) {
345 struct inode *inode;
346
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700347 inode = list_first_entry(head, struct inode, i_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 list_del(&inode->i_list);
349
350 if (inode->i_data.nrpages)
351 truncate_inode_pages(&inode->i_data, 0);
352 clear_inode(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700353
354 spin_lock(&inode_lock);
355 hlist_del_init(&inode->i_hash);
356 list_del_init(&inode->i_sb_list);
357 spin_unlock(&inode_lock);
358
359 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 destroy_inode(inode);
361 nr_disposed++;
362 }
363 spin_lock(&inode_lock);
364 inodes_stat.nr_inodes -= nr_disposed;
365 spin_unlock(&inode_lock);
366}
367
368/*
369 * Invalidate all inodes for a device.
370 */
371static int invalidate_list(struct list_head *head, struct list_head *dispose)
372{
373 struct list_head *next;
374 int busy = 0, count = 0;
375
376 next = head->next;
377 for (;;) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530378 struct list_head *tmp = next;
379 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
381 /*
382 * We can reschedule here without worrying about the list's
383 * consistency because the per-sb list of inodes must not
Ingo Molnarf24075b2006-03-23 03:00:34 -0800384 * change during umount anymore, and because iprune_mutex keeps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 * shrink_icache_memory() away.
386 */
387 cond_resched_lock(&inode_lock);
388
389 next = next->next;
390 if (tmp == head)
391 break;
392 inode = list_entry(tmp, struct inode, i_sb_list);
Nick Pigginaabb8fd2009-03-11 13:17:36 -0700393 if (inode->i_state & I_NEW)
394 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 invalidate_inode_buffers(inode);
396 if (!atomic_read(&inode->i_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 list_move(&inode->i_list, dispose);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700398 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 inode->i_state |= I_FREEING;
400 count++;
401 continue;
402 }
403 busy = 1;
404 }
405 /* only unused inodes may be cached with i_count zero */
406 inodes_stat.nr_unused -= count;
407 return busy;
408}
409
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410/**
411 * invalidate_inodes - discard the inodes on a device
412 * @sb: superblock
413 *
414 * Discard all of the inodes for a given superblock. If the discard
415 * fails because there are busy inodes then a non zero value is returned.
416 * If the discard is successful all the inodes have been discarded.
417 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530418int invalidate_inodes(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419{
420 int busy;
421 LIST_HEAD(throw_away);
422
Ingo Molnarf24075b2006-03-23 03:00:34 -0800423 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 spin_lock(&inode_lock);
Robert Love0eeca282005-07-12 17:06:03 -0400425 inotify_unmount_inodes(&sb->s_inodes);
Eric Paris164bc612009-05-21 17:01:58 -0400426 fsnotify_unmount_inodes(&sb->s_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 busy = invalidate_list(&sb->s_inodes, &throw_away);
428 spin_unlock(&inode_lock);
429
430 dispose_list(&throw_away);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800431 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
433 return busy;
434}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435EXPORT_SYMBOL(invalidate_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
437static int can_unuse(struct inode *inode)
438{
439 if (inode->i_state)
440 return 0;
441 if (inode_has_buffers(inode))
442 return 0;
443 if (atomic_read(&inode->i_count))
444 return 0;
445 if (inode->i_data.nrpages)
446 return 0;
447 return 1;
448}
449
450/*
451 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
452 * a temporary list and then are freed outside inode_lock by dispose_list().
453 *
454 * Any inodes which are pinned purely because of attached pagecache have their
455 * pagecache removed. We expect the final iput() on that inode to add it to
456 * the front of the inode_unused list. So look for it there and if the
457 * inode is still freeable, proceed. The right inode is found 99.9% of the
458 * time in testing on a 4-way.
459 *
460 * If the inode has metadata buffers attached to mapping->private_list then
461 * try to remove them.
462 */
463static void prune_icache(int nr_to_scan)
464{
465 LIST_HEAD(freeable);
466 int nr_pruned = 0;
467 int nr_scanned;
468 unsigned long reap = 0;
469
Ingo Molnarf24075b2006-03-23 03:00:34 -0800470 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 spin_lock(&inode_lock);
472 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
473 struct inode *inode;
474
475 if (list_empty(&inode_unused))
476 break;
477
478 inode = list_entry(inode_unused.prev, struct inode, i_list);
479
480 if (inode->i_state || atomic_read(&inode->i_count)) {
481 list_move(&inode->i_list, &inode_unused);
482 continue;
483 }
484 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
485 __iget(inode);
486 spin_unlock(&inode_lock);
487 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800488 reap += invalidate_mapping_pages(&inode->i_data,
489 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 iput(inode);
491 spin_lock(&inode_lock);
492
493 if (inode != list_entry(inode_unused.next,
494 struct inode, i_list))
495 continue; /* wrong inode or list_empty */
496 if (!can_unuse(inode))
497 continue;
498 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 list_move(&inode->i_list, &freeable);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700500 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 inode->i_state |= I_FREEING;
502 nr_pruned++;
503 }
504 inodes_stat.nr_unused -= nr_pruned;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700505 if (current_is_kswapd())
506 __count_vm_events(KSWAPD_INODESTEAL, reap);
507 else
508 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 spin_unlock(&inode_lock);
510
511 dispose_list(&freeable);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800512 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513}
514
515/*
516 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
517 * "unused" means that no dentries are referring to the inodes: the files are
518 * not open and the dcache references to those inodes have already been
519 * reclaimed.
520 *
521 * This function is passed the number of inodes to scan, and it returns the
522 * total number of remaining possibly-reclaimable inodes.
523 */
Al Viro27496a82005-10-21 03:20:48 -0400524static int shrink_icache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
526 if (nr) {
527 /*
528 * Nasty deadlock avoidance. We may hold various FS locks,
529 * and we don't want to recurse into the FS that called us
530 * in clear_inode() and friends..
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530531 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 if (!(gfp_mask & __GFP_FS))
533 return -1;
534 prune_icache(nr);
535 }
536 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
537}
538
Rusty Russell8e1f9362007-07-17 04:03:17 -0700539static struct shrinker icache_shrinker = {
540 .shrink = shrink_icache_memory,
541 .seeks = DEFAULT_SEEKS,
542};
543
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544static void __wait_on_freeing_inode(struct inode *inode);
545/*
546 * Called with the inode lock held.
547 * NOTE: we are not increasing the inode-refcount, you must call __iget()
548 * by hand after calling find_inode now! This simplifies iunique and won't
549 * add any additional branch in the common code.
550 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530551static struct inode *find_inode(struct super_block *sb,
552 struct hlist_head *head,
553 int (*test)(struct inode *, void *),
554 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555{
556 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530557 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
559repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700560 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 if (inode->i_sb != sb)
562 continue;
563 if (!test(inode, data))
564 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700565 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 __wait_on_freeing_inode(inode);
567 goto repeat;
568 }
569 break;
570 }
571 return node ? inode : NULL;
572}
573
574/*
575 * find_inode_fast is the fast path version of find_inode, see the comment at
576 * iget_locked for details.
577 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530578static struct inode *find_inode_fast(struct super_block *sb,
579 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580{
581 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530582 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
584repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700585 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 if (inode->i_ino != ino)
587 continue;
588 if (inode->i_sb != sb)
589 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700590 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 __wait_on_freeing_inode(inode);
592 goto repeat;
593 }
594 break;
595 }
596 return node ? inode : NULL;
597}
598
David Chinner8290c352008-10-30 17:35:24 +1100599static unsigned long hash(struct super_block *sb, unsigned long hashval)
600{
601 unsigned long tmp;
602
603 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
604 L1_CACHE_BYTES;
605 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
606 return tmp & I_HASHMASK;
607}
608
609static inline void
610__inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
611 struct inode *inode)
612{
613 inodes_stat.nr_inodes++;
614 list_add(&inode->i_list, &inode_in_use);
615 list_add(&inode->i_sb_list, &sb->s_inodes);
616 if (head)
617 hlist_add_head(&inode->i_hash, head);
618}
619
620/**
621 * inode_add_to_lists - add a new inode to relevant lists
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800622 * @sb: superblock inode belongs to
623 * @inode: inode to mark in use
David Chinner8290c352008-10-30 17:35:24 +1100624 *
625 * When an inode is allocated it needs to be accounted for, added to the in use
626 * list, the owning superblock and the inode hash. This needs to be done under
627 * the inode_lock, so export a function to do this rather than the inode lock
628 * itself. We calculate the hash list to add to here so it is all internal
629 * which requires the caller to have already set up the inode number in the
630 * inode to add.
631 */
632void inode_add_to_lists(struct super_block *sb, struct inode *inode)
633{
634 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
635
636 spin_lock(&inode_lock);
637 __inode_add_to_lists(sb, head, inode);
638 spin_unlock(&inode_lock);
639}
640EXPORT_SYMBOL_GPL(inode_add_to_lists);
641
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642/**
643 * new_inode - obtain an inode
644 * @sb: superblock
645 *
Mel Gorman769848c2007-07-17 04:03:05 -0700646 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800647 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700648 * If HIGHMEM pages are unsuitable or it is known that pages allocated
649 * for the page cache are not reclaimable or migratable,
650 * mapping_set_gfp_mask() must be called with suitable flags on the
651 * newly created inode's mapping
652 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 */
654struct inode *new_inode(struct super_block *sb)
655{
Jeff Layton866b04f2007-05-08 00:32:29 -0700656 /*
657 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
658 * error if st_ino won't fit in target struct field. Use 32bit counter
659 * here to attempt to avoid that.
660 */
661 static unsigned int last_ino;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530662 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
664 spin_lock_prefetch(&inode_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 inode = alloc_inode(sb);
667 if (inode) {
668 spin_lock(&inode_lock);
David Chinner8290c352008-10-30 17:35:24 +1100669 __inode_add_to_lists(sb, NULL, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 inode->i_ino = ++last_ino;
671 inode->i_state = 0;
672 spin_unlock(&inode_lock);
673 }
674 return inode;
675}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676EXPORT_SYMBOL(new_inode);
677
678void unlock_new_inode(struct inode *inode)
679{
Peter Zijlstra14358e62007-10-14 01:38:33 +0200680#ifdef CONFIG_DEBUG_LOCK_ALLOC
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200681 if (inode->i_mode & S_IFDIR) {
682 struct file_system_type *type = inode->i_sb->s_type;
683
Jan Kara9a7aa122009-06-04 15:26:49 +0200684 /* Set new key only if filesystem hasn't already changed it */
685 if (!lockdep_match_class(&inode->i_mutex,
686 &type->i_mutex_key)) {
687 /*
688 * ensure nobody is actually holding i_mutex
689 */
690 mutex_destroy(&inode->i_mutex);
691 mutex_init(&inode->i_mutex);
692 lockdep_set_class(&inode->i_mutex,
693 &type->i_mutex_dir_key);
694 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200695 }
Peter Zijlstra14358e62007-10-14 01:38:33 +0200696#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 /*
Jan Kara580be082009-09-21 17:01:06 -0700698 * This is special! We do not need the spinlock when clearing I_LOCK,
699 * because we're guaranteed that nobody else tries to do anything about
700 * the state of the inode when it is locked, as we just created it (so
701 * there can be no old holders that haven't tested I_LOCK).
702 * However we must emit the memory barrier so that other CPUs reliably
703 * see the clearing of I_LOCK after the other inode initialisation has
704 * completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 */
Jan Kara580be082009-09-21 17:01:06 -0700706 smp_mb();
Nick Piggin7ef0d732009-03-12 14:31:38 -0700707 WARN_ON((inode->i_state & (I_LOCK|I_NEW)) != (I_LOCK|I_NEW));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 inode->i_state &= ~(I_LOCK|I_NEW);
709 wake_up_inode(inode);
710}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711EXPORT_SYMBOL(unlock_new_inode);
712
713/*
714 * This is called without the inode lock held.. Be careful.
715 *
716 * We no longer cache the sb_flags in i_flags - see fs.h
717 * -- rmk@arm.uk.linux.org
718 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530719static struct inode *get_new_inode(struct super_block *sb,
720 struct hlist_head *head,
721 int (*test)(struct inode *, void *),
722 int (*set)(struct inode *, void *),
723 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530725 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
727 inode = alloc_inode(sb);
728 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530729 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
731 spin_lock(&inode_lock);
732 /* We released the lock, so.. */
733 old = find_inode(sb, head, test, data);
734 if (!old) {
735 if (set(inode, data))
736 goto set_failed;
737
David Chinner8290c352008-10-30 17:35:24 +1100738 __inode_add_to_lists(sb, head, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 inode->i_state = I_LOCK|I_NEW;
740 spin_unlock(&inode_lock);
741
742 /* Return the locked inode with I_NEW set, the
743 * caller is responsible for filling in the contents
744 */
745 return inode;
746 }
747
748 /*
749 * Uhhuh, somebody else created the same inode under
750 * us. Use the old inode instead of the one we just
751 * allocated.
752 */
753 __iget(old);
754 spin_unlock(&inode_lock);
755 destroy_inode(inode);
756 inode = old;
757 wait_on_inode(inode);
758 }
759 return inode;
760
761set_failed:
762 spin_unlock(&inode_lock);
763 destroy_inode(inode);
764 return NULL;
765}
766
767/*
768 * get_new_inode_fast is the fast path version of get_new_inode, see the
769 * comment at iget_locked for details.
770 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530771static struct inode *get_new_inode_fast(struct super_block *sb,
772 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530774 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
776 inode = alloc_inode(sb);
777 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530778 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779
780 spin_lock(&inode_lock);
781 /* We released the lock, so.. */
782 old = find_inode_fast(sb, head, ino);
783 if (!old) {
784 inode->i_ino = ino;
David Chinner8290c352008-10-30 17:35:24 +1100785 __inode_add_to_lists(sb, head, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 inode->i_state = I_LOCK|I_NEW;
787 spin_unlock(&inode_lock);
788
789 /* Return the locked inode with I_NEW set, the
790 * caller is responsible for filling in the contents
791 */
792 return inode;
793 }
794
795 /*
796 * Uhhuh, somebody else created the same inode under
797 * us. Use the old inode instead of the one we just
798 * allocated.
799 */
800 __iget(old);
801 spin_unlock(&inode_lock);
802 destroy_inode(inode);
803 inode = old;
804 wait_on_inode(inode);
805 }
806 return inode;
807}
808
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809/**
810 * iunique - get a unique inode number
811 * @sb: superblock
812 * @max_reserved: highest reserved inode number
813 *
814 * Obtain an inode number that is unique on the system for a given
815 * superblock. This is used by file systems that have no natural
816 * permanent inode numbering system. An inode number is returned that
817 * is higher than the reserved limit but unique.
818 *
819 * BUGS:
820 * With a large number of inodes live on the file system this function
821 * currently becomes quite slow.
822 */
823ino_t iunique(struct super_block *sb, ino_t max_reserved)
824{
Jeff Layton866b04f2007-05-08 00:32:29 -0700825 /*
826 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
827 * error if st_ino won't fit in target struct field. Use 32bit counter
828 * here to attempt to avoid that.
829 */
830 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 struct inode *inode;
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700832 struct hlist_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700835 spin_lock(&inode_lock);
836 do {
837 if (counter <= max_reserved)
838 counter = max_reserved + 1;
839 res = counter++;
840 head = inode_hashtable + hash(sb, res);
841 inode = find_inode_fast(sb, head, res);
842 } while (inode != NULL);
843 spin_unlock(&inode_lock);
844
845 return res;
846}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847EXPORT_SYMBOL(iunique);
848
849struct inode *igrab(struct inode *inode)
850{
851 spin_lock(&inode_lock);
Jan Blunck4a3b0a42007-02-10 01:44:59 -0800852 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 __iget(inode);
854 else
855 /*
856 * Handle the case where s_op->clear_inode is not been
857 * called yet, and somebody is calling igrab
858 * while the inode is getting freed.
859 */
860 inode = NULL;
861 spin_unlock(&inode_lock);
862 return inode;
863}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864EXPORT_SYMBOL(igrab);
865
866/**
867 * ifind - internal function, you want ilookup5() or iget5().
868 * @sb: super block of file system to search
869 * @head: the head of the list to search
870 * @test: callback used for comparisons between inodes
871 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700872 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 *
874 * ifind() searches for the inode specified by @data in the inode
875 * cache. This is a generalized version of ifind_fast() for file systems where
876 * the inode number is not sufficient for unique identification of an inode.
877 *
878 * If the inode is in the cache, the inode is returned with an incremented
879 * reference count.
880 *
881 * Otherwise NULL is returned.
882 *
883 * Note, @test is called with the inode_lock held, so can't sleep.
884 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800885static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700887 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888{
889 struct inode *inode;
890
891 spin_lock(&inode_lock);
892 inode = find_inode(sb, head, test, data);
893 if (inode) {
894 __iget(inode);
895 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700896 if (likely(wait))
897 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 return inode;
899 }
900 spin_unlock(&inode_lock);
901 return NULL;
902}
903
904/**
905 * ifind_fast - internal function, you want ilookup() or iget().
906 * @sb: super block of file system to search
907 * @head: head of the list to search
908 * @ino: inode number to search for
909 *
910 * ifind_fast() searches for the inode @ino in the inode cache. This is for
911 * file systems where the inode number is sufficient for unique identification
912 * of an inode.
913 *
914 * If the inode is in the cache, the inode is returned with an incremented
915 * reference count.
916 *
917 * Otherwise NULL is returned.
918 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800919static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 struct hlist_head *head, unsigned long ino)
921{
922 struct inode *inode;
923
924 spin_lock(&inode_lock);
925 inode = find_inode_fast(sb, head, ino);
926 if (inode) {
927 __iget(inode);
928 spin_unlock(&inode_lock);
929 wait_on_inode(inode);
930 return inode;
931 }
932 spin_unlock(&inode_lock);
933 return NULL;
934}
935
936/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700937 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 * @sb: super block of file system to search
939 * @hashval: hash value (usually inode number) to search for
940 * @test: callback used for comparisons between inodes
941 * @data: opaque data pointer to pass to @test
942 *
943 * ilookup5() uses ifind() to search for the inode specified by @hashval and
944 * @data in the inode cache. This is a generalized version of ilookup() for
945 * file systems where the inode number is not sufficient for unique
946 * identification of an inode.
947 *
948 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700949 * reference count. Note, the inode lock is not waited upon so you have to be
950 * very careful what you do with the returned inode. You probably should be
951 * using ilookup5() instead.
952 *
953 * Otherwise NULL is returned.
954 *
955 * Note, @test is called with the inode_lock held, so can't sleep.
956 */
957struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
958 int (*test)(struct inode *, void *), void *data)
959{
960 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
961
962 return ifind(sb, head, test, data, 0);
963}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700964EXPORT_SYMBOL(ilookup5_nowait);
965
966/**
967 * ilookup5 - search for an inode in the inode cache
968 * @sb: super block of file system to search
969 * @hashval: hash value (usually inode number) to search for
970 * @test: callback used for comparisons between inodes
971 * @data: opaque data pointer to pass to @test
972 *
973 * ilookup5() uses ifind() to search for the inode specified by @hashval and
974 * @data in the inode cache. This is a generalized version of ilookup() for
975 * file systems where the inode number is not sufficient for unique
976 * identification of an inode.
977 *
978 * If the inode is in the cache, the inode lock is waited upon and the inode is
979 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 *
981 * Otherwise NULL is returned.
982 *
983 * Note, @test is called with the inode_lock held, so can't sleep.
984 */
985struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
986 int (*test)(struct inode *, void *), void *data)
987{
988 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
989
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700990 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992EXPORT_SYMBOL(ilookup5);
993
994/**
995 * ilookup - search for an inode in the inode cache
996 * @sb: super block of file system to search
997 * @ino: inode number to search for
998 *
999 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1000 * This is for file systems where the inode number is sufficient for unique
1001 * identification of an inode.
1002 *
1003 * If the inode is in the cache, the inode is returned with an incremented
1004 * reference count.
1005 *
1006 * Otherwise NULL is returned.
1007 */
1008struct inode *ilookup(struct super_block *sb, unsigned long ino)
1009{
1010 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1011
1012 return ifind_fast(sb, head, ino);
1013}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014EXPORT_SYMBOL(ilookup);
1015
1016/**
1017 * iget5_locked - obtain an inode from a mounted file system
1018 * @sb: super block of file system
1019 * @hashval: hash value (usually inode number) to get
1020 * @test: callback used for comparisons between inodes
1021 * @set: callback used to initialize a new struct inode
1022 * @data: opaque data pointer to pass to @test and @set
1023 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1025 * and @data in the inode cache and if present it is returned with an increased
1026 * reference count. This is a generalized version of iget_locked() for file
1027 * systems where the inode number is not sufficient for unique identification
1028 * of an inode.
1029 *
1030 * If the inode is not in cache, get_new_inode() is called to allocate a new
1031 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1032 * file system gets to fill it in before unlocking it via unlock_new_inode().
1033 *
1034 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1035 */
1036struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1037 int (*test)(struct inode *, void *),
1038 int (*set)(struct inode *, void *), void *data)
1039{
1040 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1041 struct inode *inode;
1042
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001043 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 if (inode)
1045 return inode;
1046 /*
1047 * get_new_inode() will do the right thing, re-trying the search
1048 * in case it had to block at any point.
1049 */
1050 return get_new_inode(sb, head, test, set, data);
1051}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052EXPORT_SYMBOL(iget5_locked);
1053
1054/**
1055 * iget_locked - obtain an inode from a mounted file system
1056 * @sb: super block of file system
1057 * @ino: inode number to get
1058 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1060 * the inode cache and if present it is returned with an increased reference
1061 * count. This is for file systems where the inode number is sufficient for
1062 * unique identification of an inode.
1063 *
1064 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1065 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1066 * The file system gets to fill it in before unlocking it via
1067 * unlock_new_inode().
1068 */
1069struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1070{
1071 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1072 struct inode *inode;
1073
1074 inode = ifind_fast(sb, head, ino);
1075 if (inode)
1076 return inode;
1077 /*
1078 * get_new_inode_fast() will do the right thing, re-trying the search
1079 * in case it had to block at any point.
1080 */
1081 return get_new_inode_fast(sb, head, ino);
1082}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083EXPORT_SYMBOL(iget_locked);
1084
Al Viro261bca82008-12-30 01:48:21 -05001085int insert_inode_locked(struct inode *inode)
1086{
1087 struct super_block *sb = inode->i_sb;
1088 ino_t ino = inode->i_ino;
1089 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001090
1091 inode->i_state |= I_LOCK|I_NEW;
1092 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001093 struct hlist_node *node;
1094 struct inode *old = NULL;
Al Viro261bca82008-12-30 01:48:21 -05001095 spin_lock(&inode_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001096 hlist_for_each_entry(old, node, head, i_hash) {
1097 if (old->i_ino != ino)
1098 continue;
1099 if (old->i_sb != sb)
1100 continue;
1101 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1102 continue;
1103 break;
1104 }
1105 if (likely(!node)) {
Al Viro261bca82008-12-30 01:48:21 -05001106 hlist_add_head(&inode->i_hash, head);
1107 spin_unlock(&inode_lock);
1108 return 0;
1109 }
1110 __iget(old);
1111 spin_unlock(&inode_lock);
1112 wait_on_inode(old);
1113 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1114 iput(old);
1115 return -EBUSY;
1116 }
1117 iput(old);
1118 }
1119}
Al Viro261bca82008-12-30 01:48:21 -05001120EXPORT_SYMBOL(insert_inode_locked);
1121
1122int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1123 int (*test)(struct inode *, void *), void *data)
1124{
1125 struct super_block *sb = inode->i_sb;
1126 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001127
1128 inode->i_state |= I_LOCK|I_NEW;
1129
1130 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001131 struct hlist_node *node;
1132 struct inode *old = NULL;
1133
Al Viro261bca82008-12-30 01:48:21 -05001134 spin_lock(&inode_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001135 hlist_for_each_entry(old, node, head, i_hash) {
1136 if (old->i_sb != sb)
1137 continue;
1138 if (!test(old, data))
1139 continue;
1140 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1141 continue;
1142 break;
1143 }
1144 if (likely(!node)) {
Al Viro261bca82008-12-30 01:48:21 -05001145 hlist_add_head(&inode->i_hash, head);
1146 spin_unlock(&inode_lock);
1147 return 0;
1148 }
1149 __iget(old);
1150 spin_unlock(&inode_lock);
1151 wait_on_inode(old);
1152 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1153 iput(old);
1154 return -EBUSY;
1155 }
1156 iput(old);
1157 }
1158}
Al Viro261bca82008-12-30 01:48:21 -05001159EXPORT_SYMBOL(insert_inode_locked4);
1160
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161/**
1162 * __insert_inode_hash - hash an inode
1163 * @inode: unhashed inode
1164 * @hashval: unsigned long value used to locate this object in the
1165 * inode_hashtable.
1166 *
1167 * Add an inode to the inode hash for this superblock.
1168 */
1169void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1170{
1171 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1172 spin_lock(&inode_lock);
1173 hlist_add_head(&inode->i_hash, head);
1174 spin_unlock(&inode_lock);
1175}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176EXPORT_SYMBOL(__insert_inode_hash);
1177
1178/**
1179 * remove_inode_hash - remove an inode from the hash
1180 * @inode: inode to unhash
1181 *
1182 * Remove an inode from the superblock.
1183 */
1184void remove_inode_hash(struct inode *inode)
1185{
1186 spin_lock(&inode_lock);
1187 hlist_del_init(&inode->i_hash);
1188 spin_unlock(&inode_lock);
1189}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190EXPORT_SYMBOL(remove_inode_hash);
1191
1192/*
1193 * Tell the filesystem that this inode is no longer of any interest and should
1194 * be completely destroyed.
1195 *
1196 * We leave the inode in the inode hash table until *after* the filesystem's
1197 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1198 * instigate) will always find up-to-date information either in the hash or on
1199 * disk.
1200 *
1201 * I_FREEING is set so that no-one will take a new reference to the inode while
1202 * it is being deleted.
1203 */
Arjan van de Venb32714b2009-01-09 07:04:15 -08001204void generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001206 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207
Arjan van de Venb32714b2009-01-09 07:04:15 -08001208 list_del_init(&inode->i_list);
1209 list_del_init(&inode->i_sb_list);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001210 WARN_ON(inode->i_state & I_NEW);
Arjan van de Venb32714b2009-01-09 07:04:15 -08001211 inode->i_state |= I_FREEING;
1212 inodes_stat.nr_inodes--;
1213 spin_unlock(&inode_lock);
1214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 security_inode_delete(inode);
1216
1217 if (op->delete_inode) {
1218 void (*delete)(struct inode *) = op->delete_inode;
1219 if (!is_bad_inode(inode))
Jan Kara9e3509e2009-01-26 16:45:12 +01001220 vfs_dq_init(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001221 /* Filesystems implementing their own
1222 * s_op->delete_inode are required to call
1223 * truncate_inode_pages and clear_inode()
1224 * internally */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 delete(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001226 } else {
1227 truncate_inode_pages(&inode->i_data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 clear_inode(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001229 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 spin_lock(&inode_lock);
1231 hlist_del_init(&inode->i_hash);
1232 spin_unlock(&inode_lock);
1233 wake_up_inode(inode);
Eric Sesterhennb7542f82006-04-02 13:38:18 +02001234 BUG_ON(inode->i_state != I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 destroy_inode(inode);
1236}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237EXPORT_SYMBOL(generic_delete_inode);
1238
1239static void generic_forget_inode(struct inode *inode)
1240{
1241 struct super_block *sb = inode->i_sb;
1242
1243 if (!hlist_unhashed(&inode->i_hash)) {
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001244 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 list_move(&inode->i_list, &inode_unused);
1246 inodes_stat.nr_unused++;
Christoph Hellwigacb0c852007-05-08 00:25:52 -07001247 if (sb->s_flags & MS_ACTIVE) {
Alexander Viro991114c2005-06-23 00:09:01 -07001248 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001250 }
Nick Piggin7ef0d732009-03-12 14:31:38 -07001251 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001252 inode->i_state |= I_WILL_FREE;
1253 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 write_inode_now(inode, 1);
1255 spin_lock(&inode_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001256 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001257 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 inodes_stat.nr_unused--;
1259 hlist_del_init(&inode->i_hash);
1260 }
1261 list_del_init(&inode->i_list);
1262 list_del_init(&inode->i_sb_list);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001263 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001264 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 inodes_stat.nr_inodes--;
1266 spin_unlock(&inode_lock);
1267 if (inode->i_data.nrpages)
1268 truncate_inode_pages(&inode->i_data, 0);
1269 clear_inode(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001270 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 destroy_inode(inode);
1272}
1273
1274/*
1275 * Normal UNIX filesystem behaviour: delete the
1276 * inode when the usage count drops to zero, and
1277 * i_nlink is zero.
1278 */
Mark Fashehcb2c0232005-07-07 17:56:03 -07001279void generic_drop_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280{
1281 if (!inode->i_nlink)
1282 generic_delete_inode(inode);
1283 else
1284 generic_forget_inode(inode);
1285}
Mark Fashehcb2c0232005-07-07 17:56:03 -07001286EXPORT_SYMBOL_GPL(generic_drop_inode);
1287
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288/*
1289 * Called when we're dropping the last reference
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301290 * to an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 *
1292 * Call the FS "drop()" function, defaulting to
1293 * the legacy UNIX filesystem behaviour..
1294 *
1295 * NOTE! NOTE! NOTE! We're called with the inode lock
1296 * held, and the drop function is supposed to release
1297 * the lock!
1298 */
1299static inline void iput_final(struct inode *inode)
1300{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001301 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 void (*drop)(struct inode *) = generic_drop_inode;
1303
1304 if (op && op->drop_inode)
1305 drop = op->drop_inode;
1306 drop(inode);
1307}
1308
1309/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301310 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 * @inode: inode to put
1312 *
1313 * Puts an inode, dropping its usage count. If the inode use count hits
1314 * zero, the inode is then freed and may also be destroyed.
1315 *
1316 * Consequently, iput() can sleep.
1317 */
1318void iput(struct inode *inode)
1319{
1320 if (inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 BUG_ON(inode->i_state == I_CLEAR);
1322
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1324 iput_final(inode);
1325 }
1326}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327EXPORT_SYMBOL(iput);
1328
1329/**
1330 * bmap - find a block number in a file
1331 * @inode: inode of file
1332 * @block: block to find
1333 *
1334 * Returns the block number on the device holding the inode that
1335 * is the disk block number for the block of the file requested.
1336 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301337 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 * file.
1339 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301340sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341{
1342 sector_t res = 0;
1343 if (inode->i_mapping->a_ops->bmap)
1344 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1345 return res;
1346}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347EXPORT_SYMBOL(bmap);
1348
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001349/*
1350 * With relative atime, only update atime if the previous atime is
1351 * earlier than either the ctime or mtime or if at least a day has
1352 * passed since the last atime update.
1353 */
1354static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1355 struct timespec now)
1356{
1357
1358 if (!(mnt->mnt_flags & MNT_RELATIME))
1359 return 1;
1360 /*
1361 * Is mtime younger than atime? If yes, update atime:
1362 */
1363 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1364 return 1;
1365 /*
1366 * Is ctime younger than atime? If yes, update atime:
1367 */
1368 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1369 return 1;
1370
1371 /*
1372 * Is the previous atime value older than a day? If yes,
1373 * update atime:
1374 */
1375 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1376 return 1;
1377 /*
1378 * Good, we can skip the atime update:
1379 */
1380 return 0;
1381}
1382
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001384 * touch_atime - update the access time
1385 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001386 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 *
1388 * Update the accessed time on an inode and mark it for writeback.
1389 * This function automatically handles read only file systems and media,
1390 * as well as the "noatime" flag and inode specific "noatime" markers.
1391 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001392void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001394 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 struct timespec now;
1396
Dave Hansencdb70f32008-02-15 14:37:41 -08001397 if (mnt_want_write(mnt))
1398 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001399 if (inode->i_flags & S_NOATIME)
Dave Hansencdb70f32008-02-15 14:37:41 -08001400 goto out;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001401 if (IS_NOATIME(inode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001402 goto out;
Andrew Mortonb2276132006-12-13 00:34:33 -08001403 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001404 goto out;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001405
Dave Hansencdb70f32008-02-15 14:37:41 -08001406 if (mnt->mnt_flags & MNT_NOATIME)
1407 goto out;
1408 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1409 goto out;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001410
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 now = current_fs_time(inode->i_sb);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001412
1413 if (!relatime_need_update(mnt, inode, now))
1414 goto out;
1415
Valerie Henson47ae32d2006-12-13 00:34:34 -08001416 if (timespec_equal(&inode->i_atime, &now))
Dave Hansencdb70f32008-02-15 14:37:41 -08001417 goto out;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001418
1419 inode->i_atime = now;
1420 mark_inode_dirty_sync(inode);
Dave Hansencdb70f32008-02-15 14:37:41 -08001421out:
1422 mnt_drop_write(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001424EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425
1426/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001427 * file_update_time - update mtime and ctime time
1428 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001430 * Update the mtime and ctime members of an inode and mark the inode
1431 * for writeback. Note that this function is meant exclusively for
1432 * usage in the file write path of filesystems, and filesystems may
1433 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001434 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Christoph Hellwig870f4812006-01-09 20:52:01 -08001435 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 */
1437
Christoph Hellwig870f4812006-01-09 20:52:01 -08001438void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001440 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 struct timespec now;
1442 int sync_it = 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001443 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444
1445 if (IS_NOCMTIME(inode))
1446 return;
Dave Hansen20ddee22008-02-15 14:37:43 -08001447
npiggin@suse.de96029c42009-04-26 20:25:55 +10001448 err = mnt_want_write_file(file);
Dave Hansen20ddee22008-02-15 14:37:43 -08001449 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 return;
1451
1452 now = current_fs_time(inode->i_sb);
Andreas Mohred97bd32006-10-02 02:17:17 -07001453 if (!timespec_equal(&inode->i_mtime, &now)) {
1454 inode->i_mtime = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001456 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
Andreas Mohred97bd32006-10-02 02:17:17 -07001458 if (!timespec_equal(&inode->i_ctime, &now)) {
1459 inode->i_ctime = now;
Christoph Hellwig870f4812006-01-09 20:52:01 -08001460 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001461 }
Christoph Hellwig870f4812006-01-09 20:52:01 -08001462
Jean Noel Cordenner7a224222008-01-28 23:58:27 -05001463 if (IS_I_VERSION(inode)) {
1464 inode_inc_iversion(inode);
1465 sync_it = 1;
1466 }
1467
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 if (sync_it)
1469 mark_inode_dirty_sync(inode);
Dave Hansen20ddee22008-02-15 14:37:43 -08001470 mnt_drop_write(file->f_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001472EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473
1474int inode_needs_sync(struct inode *inode)
1475{
1476 if (IS_SYNC(inode))
1477 return 1;
1478 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1479 return 1;
1480 return 0;
1481}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482EXPORT_SYMBOL(inode_needs_sync);
1483
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484int inode_wait(void *word)
1485{
1486 schedule();
1487 return 0;
1488}
Stephen Rothwelld44dab82008-11-10 17:06:05 +11001489EXPORT_SYMBOL(inode_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490
1491/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001492 * If we try to find an inode in the inode hash while it is being
1493 * deleted, we have to wait until the filesystem completes its
1494 * deletion before reporting that it isn't found. This function waits
1495 * until the deletion _might_ have completed. Callers are responsible
1496 * to recheck inode state.
1497 *
1498 * It doesn't matter if I_LOCK is not set initially, a call to
1499 * wake_up_inode() after removing from the hash list will DTRT.
1500 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 * This is called with inode_lock held.
1502 */
1503static void __wait_on_freeing_inode(struct inode *inode)
1504{
1505 wait_queue_head_t *wq;
1506 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1508 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1509 spin_unlock(&inode_lock);
1510 schedule();
1511 finish_wait(wq, &wait.wait);
1512 spin_lock(&inode_lock);
1513}
1514
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515static __initdata unsigned long ihash_entries;
1516static int __init set_ihash_entries(char *str)
1517{
1518 if (!str)
1519 return 0;
1520 ihash_entries = simple_strtoul(str, &str, 0);
1521 return 1;
1522}
1523__setup("ihash_entries=", set_ihash_entries);
1524
1525/*
1526 * Initialize the waitqueues and inode hash table.
1527 */
1528void __init inode_init_early(void)
1529{
1530 int loop;
1531
1532 /* If hashes are distributed across NUMA nodes, defer
1533 * hash allocation until vmalloc space is available.
1534 */
1535 if (hashdist)
1536 return;
1537
1538 inode_hashtable =
1539 alloc_large_system_hash("Inode-cache",
1540 sizeof(struct hlist_head),
1541 ihash_entries,
1542 14,
1543 HASH_EARLY,
1544 &i_hash_shift,
1545 &i_hash_mask,
1546 0);
1547
1548 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1549 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1550}
1551
Denis Cheng74bf17c2007-10-16 23:26:30 -07001552void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553{
1554 int loop;
1555
1556 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001557 inode_cachep = kmem_cache_create("inode_cache",
1558 sizeof(struct inode),
1559 0,
1560 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1561 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001562 init_once);
Rusty Russell8e1f9362007-07-17 04:03:17 -07001563 register_shrinker(&icache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
1565 /* Hash may have been set up in inode_init_early */
1566 if (!hashdist)
1567 return;
1568
1569 inode_hashtable =
1570 alloc_large_system_hash("Inode-cache",
1571 sizeof(struct hlist_head),
1572 ihash_entries,
1573 14,
1574 0,
1575 &i_hash_shift,
1576 &i_hash_mask,
1577 0);
1578
1579 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1580 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1581}
1582
1583void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1584{
1585 inode->i_mode = mode;
1586 if (S_ISCHR(mode)) {
1587 inode->i_fop = &def_chr_fops;
1588 inode->i_rdev = rdev;
1589 } else if (S_ISBLK(mode)) {
1590 inode->i_fop = &def_blk_fops;
1591 inode->i_rdev = rdev;
1592 } else if (S_ISFIFO(mode))
1593 inode->i_fop = &def_fifo_fops;
1594 else if (S_ISSOCK(mode))
1595 inode->i_fop = &bad_sock_fops;
1596 else
1597 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1598 mode);
1599}
1600EXPORT_SYMBOL(init_special_inode);