blob: da85e56378f3e1d98c241901171e24e83882f189 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/slab.h>
12#include <linux/writeback.h>
13#include <linux/module.h>
14#include <linux/backing-dev.h>
15#include <linux/wait.h>
Nick Piggin88e0fbc2009-09-22 16:43:50 -070016#include <linux/rwsem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/hash.h>
18#include <linux/swap.h>
19#include <linux/security.h>
20#include <linux/pagemap.h>
21#include <linux/cdev.h>
22#include <linux/bootmem.h>
Eric Paris3be25f42009-05-21 17:01:26 -040023#include <linux/fsnotify.h>
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -080024#include <linux/mount.h>
Arjan van de Venefaee192009-01-06 07:20:54 -080025#include <linux/async.h>
Al Virof19d4a82009-06-08 19:50:45 -040026#include <linux/posix_acl.h>
Eric Parisa178d202010-10-25 14:41:59 -040027#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
29/*
30 * This is needed for the following functions:
31 * - inode_has_buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 * - invalidate_bdev
33 *
34 * FIXME: remove all knowledge of the buffer layer from this file
35 */
36#include <linux/buffer_head.h>
37
38/*
39 * New inode.c implementation.
40 *
41 * This implementation has the basic premise of trying
42 * to be extremely low-overhead and SMP-safe, yet be
43 * simple enough to be "obviously correct".
44 *
45 * Famous last words.
46 */
47
48/* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
49
50/* #define INODE_PARANOIA 1 */
51/* #define INODE_DEBUG 1 */
52
53/*
54 * Inode lookup is no longer as critical as it used to be:
55 * most of the lookups are going to be through the dcache.
56 */
57#define I_HASHBITS i_hash_shift
58#define I_HASHMASK i_hash_mask
59
Eric Dumazetfa3536c2006-03-26 01:37:24 -080060static unsigned int i_hash_mask __read_mostly;
61static unsigned int i_hash_shift __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
63/*
64 * Each inode can be on two separate lists. One is
65 * the hash list of the inode, used for lookups. The
66 * other linked list is the "type" list:
67 * "in_use" - valid inode, i_count > 0, i_nlink > 0
68 * "dirty" - as "in_use" but also dirty
69 * "unused" - valid inode, i_count = 0
70 *
71 * A "dirty" list is maintained for each super block,
72 * allowing for low-overhead inode sync() operations.
73 */
74
Nick Piggin7ccf19a2010-10-21 11:49:30 +110075static LIST_HEAD(inode_lru);
Eric Dumazetfa3536c2006-03-26 01:37:24 -080076static struct hlist_head *inode_hashtable __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
78/*
79 * A simple spinlock to protect the list manipulations.
80 *
81 * NOTE! You also have to own the lock if you change
82 * the i_state of an inode while it is in use..
83 */
84DEFINE_SPINLOCK(inode_lock);
85
86/*
Nick Piggin88e0fbc2009-09-22 16:43:50 -070087 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 * icache shrinking path, and the umount path. Without this exclusion,
89 * by the time prune_icache calls iput for the inode whose pages it has
90 * been invalidating, or by the time it calls clear_inode & destroy_inode
91 * from its final dispose_list, the struct super_block they refer to
92 * (for inode->i_sb->s_op) may already have been freed and reused.
Nick Piggin88e0fbc2009-09-22 16:43:50 -070093 *
94 * We make this an rwsem because the fastpath is icache shrinking. In
95 * some cases a filesystem may be doing a significant amount of work in
96 * its inode reclaim code, so this should improve parallelism.
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 */
Nick Piggin88e0fbc2009-09-22 16:43:50 -070098static DECLARE_RWSEM(iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
100/*
101 * Statistics gathering..
102 */
103struct inodes_stat_t inodes_stat;
104
Nick Piggin3e880fb2011-01-07 17:49:19 +1100105static DEFINE_PER_CPU(unsigned int, nr_inodes);
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400106
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530107static struct kmem_cache *inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
Nick Piggin3e880fb2011-01-07 17:49:19 +1100109static int get_nr_inodes(void)
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400110{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100111 int i;
112 int sum = 0;
113 for_each_possible_cpu(i)
114 sum += per_cpu(nr_inodes, i);
115 return sum < 0 ? 0 : sum;
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400116}
117
118static inline int get_nr_inodes_unused(void)
119{
Nick Piggin86c87492011-01-07 17:49:18 +1100120 return inodes_stat.nr_unused;
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400121}
122
123int get_nr_dirty_inodes(void)
124{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100125 /* not actually dirty inodes, but a wild approximation */
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400126 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
127 return nr_dirty > 0 ? nr_dirty : 0;
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400128}
129
130/*
131 * Handle nr_inode sysctl
132 */
133#ifdef CONFIG_SYSCTL
134int proc_nr_inodes(ctl_table *table, int write,
135 void __user *buffer, size_t *lenp, loff_t *ppos)
136{
137 inodes_stat.nr_inodes = get_nr_inodes();
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400138 return proc_dointvec(table, write, buffer, lenp, ppos);
139}
140#endif
141
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700142static void wake_up_inode(struct inode *inode)
143{
144 /*
145 * Prevent speculative execution through spin_unlock(&inode_lock);
146 */
147 smp_mb();
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100148 wake_up_bit(&inode->i_state, __I_NEW);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700149}
150
David Chinner2cb15992008-10-30 17:32:23 +1100151/**
152 * inode_init_always - perform inode structure intialisation
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800153 * @sb: superblock inode belongs to
154 * @inode: inode to initialise
David Chinner2cb15992008-10-30 17:32:23 +1100155 *
156 * These are initializations that need to be done on every inode
157 * allocation as the fields are not initialised by slab allocation.
158 */
Christoph Hellwig54e34622009-08-07 14:38:25 -0300159int inode_init_always(struct super_block *sb, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160{
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700161 static const struct address_space_operations empty_aops;
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700162 static const struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800163 static const struct file_operations empty_fops;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530164 struct address_space *const mapping = &inode->i_data;
David Chinner2cb15992008-10-30 17:32:23 +1100165
166 inode->i_sb = sb;
167 inode->i_blkbits = sb->s_blocksize_bits;
168 inode->i_flags = 0;
169 atomic_set(&inode->i_count, 1);
170 inode->i_op = &empty_iops;
171 inode->i_fop = &empty_fops;
172 inode->i_nlink = 1;
Al Viro56ff5ef2008-12-09 09:34:39 -0500173 inode->i_uid = 0;
174 inode->i_gid = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100175 atomic_set(&inode->i_writecount, 0);
176 inode->i_size = 0;
177 inode->i_blocks = 0;
178 inode->i_bytes = 0;
179 inode->i_generation = 0;
180#ifdef CONFIG_QUOTA
181 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
182#endif
183 inode->i_pipe = NULL;
184 inode->i_bdev = NULL;
185 inode->i_cdev = NULL;
186 inode->i_rdev = 0;
187 inode->dirtied_when = 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500188
189 if (security_inode_alloc(inode))
Christoph Hellwig54e34622009-08-07 14:38:25 -0300190 goto out;
David Chinner2cb15992008-10-30 17:32:23 +1100191 spin_lock_init(&inode->i_lock);
192 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
193
194 mutex_init(&inode->i_mutex);
195 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
196
197 init_rwsem(&inode->i_alloc_sem);
198 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
199
200 mapping->a_ops = &empty_aops;
201 mapping->host = inode;
202 mapping->flags = 0;
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800203 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
David Chinner2cb15992008-10-30 17:32:23 +1100204 mapping->assoc_mapping = NULL;
205 mapping->backing_dev_info = &default_backing_dev_info;
206 mapping->writeback_index = 0;
207
208 /*
209 * If the block_device provides a backing_dev_info for client
210 * inodes then use that. Otherwise the inode share the bdev's
211 * backing_dev_info.
212 */
213 if (sb->s_bdev) {
214 struct backing_dev_info *bdi;
215
Jens Axboe2c96ce92009-09-15 09:43:56 +0200216 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
David Chinner2cb15992008-10-30 17:32:23 +1100217 mapping->backing_dev_info = bdi;
218 }
219 inode->i_private = NULL;
220 inode->i_mapping = mapping;
Al Virof19d4a82009-06-08 19:50:45 -0400221#ifdef CONFIG_FS_POSIX_ACL
222 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
223#endif
David Chinner2cb15992008-10-30 17:32:23 +1100224
Eric Paris3be25f42009-05-21 17:01:26 -0400225#ifdef CONFIG_FSNOTIFY
226 inode->i_fsnotify_mask = 0;
227#endif
228
Nick Piggin3e880fb2011-01-07 17:49:19 +1100229 this_cpu_inc(nr_inodes);
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400230
Christoph Hellwig54e34622009-08-07 14:38:25 -0300231 return 0;
Christoph Hellwig54e34622009-08-07 14:38:25 -0300232out:
233 return -ENOMEM;
David Chinner2cb15992008-10-30 17:32:23 +1100234}
235EXPORT_SYMBOL(inode_init_always);
236
237static struct inode *alloc_inode(struct super_block *sb)
238{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 struct inode *inode;
240
241 if (sb->s_op->alloc_inode)
242 inode = sb->s_op->alloc_inode(sb);
243 else
David Chinner2cb15992008-10-30 17:32:23 +1100244 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Christoph Hellwig54e34622009-08-07 14:38:25 -0300246 if (!inode)
247 return NULL;
248
249 if (unlikely(inode_init_always(sb, inode))) {
250 if (inode->i_sb->s_op->destroy_inode)
251 inode->i_sb->s_op->destroy_inode(inode);
252 else
253 kmem_cache_free(inode_cachep, inode);
254 return NULL;
255 }
256
257 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
259
Nick Pigginff0c7d12011-01-07 17:49:50 +1100260void free_inode_nonrcu(struct inode *inode)
261{
262 kmem_cache_free(inode_cachep, inode);
263}
264EXPORT_SYMBOL(free_inode_nonrcu);
265
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300266void __destroy_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200268 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 security_inode_free(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400270 fsnotify_inode_delete(inode);
Al Virof19d4a82009-06-08 19:50:45 -0400271#ifdef CONFIG_FS_POSIX_ACL
272 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
273 posix_acl_release(inode->i_acl);
274 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
275 posix_acl_release(inode->i_default_acl);
276#endif
Nick Piggin3e880fb2011-01-07 17:49:19 +1100277 this_cpu_dec(nr_inodes);
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300278}
279EXPORT_SYMBOL(__destroy_inode);
280
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100281static void i_callback(struct rcu_head *head)
282{
283 struct inode *inode = container_of(head, struct inode, i_rcu);
284 INIT_LIST_HEAD(&inode->i_dentry);
285 kmem_cache_free(inode_cachep, inode);
286}
287
Christoph Hellwig56b0dac2010-10-06 10:48:55 +0200288static void destroy_inode(struct inode *inode)
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300289{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100290 BUG_ON(!list_empty(&inode->i_lru));
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300291 __destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 if (inode->i_sb->s_op->destroy_inode)
293 inode->i_sb->s_op->destroy_inode(inode);
294 else
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100295 call_rcu(&inode->i_rcu, i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
298/*
299 * These are initializations that only need to be done
300 * once, because the fields are idempotent across use
301 * of the inode, so let the slab aware of that.
302 */
303void inode_init_once(struct inode *inode)
304{
305 memset(inode, 0, sizeof(*inode));
306 INIT_HLIST_NODE(&inode->i_hash);
307 INIT_LIST_HEAD(&inode->i_dentry);
308 INIT_LIST_HEAD(&inode->i_devices);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100309 INIT_LIST_HEAD(&inode->i_wb_list);
310 INIT_LIST_HEAD(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
Nick Piggin19fd6232008-07-25 19:45:32 -0700312 spin_lock_init(&inode->i_data.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 spin_lock_init(&inode->i_data.i_mmap_lock);
314 INIT_LIST_HEAD(&inode->i_data.private_list);
315 spin_lock_init(&inode->i_data.private_lock);
316 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
317 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 i_size_ordered_init(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400319#ifdef CONFIG_FSNOTIFY
Eric Parise61ce862009-12-17 21:24:24 -0500320 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
Eric Paris3be25f42009-05-21 17:01:26 -0400321#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323EXPORT_SYMBOL(inode_init_once);
324
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700325static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530327 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
Christoph Lametera35afb82007-05-16 22:10:57 -0700329 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330}
331
332/*
333 * inode_lock must be held
334 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530335void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336{
Nick Piggin9e38d862010-10-23 06:55:17 -0400337 atomic_inc(&inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338}
339
Al Viro7de9c6ee2010-10-23 11:11:40 -0400340/*
341 * get additional reference to inode; caller must already hold one.
342 */
343void ihold(struct inode *inode)
344{
345 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
346}
347EXPORT_SYMBOL(ihold);
348
Nick Piggin9e38d862010-10-23 06:55:17 -0400349static void inode_lru_list_add(struct inode *inode)
350{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100351 if (list_empty(&inode->i_lru)) {
352 list_add(&inode->i_lru, &inode_lru);
Nick Piggin86c87492011-01-07 17:49:18 +1100353 inodes_stat.nr_unused++;
Nick Piggin9e38d862010-10-23 06:55:17 -0400354 }
355}
356
357static void inode_lru_list_del(struct inode *inode)
358{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100359 if (!list_empty(&inode->i_lru)) {
360 list_del_init(&inode->i_lru);
Nick Piggin86c87492011-01-07 17:49:18 +1100361 inodes_stat.nr_unused--;
Nick Piggin9e38d862010-10-23 06:55:17 -0400362 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363}
364
Christoph Hellwig646ec462010-10-23 07:15:32 -0400365static inline void __inode_sb_list_add(struct inode *inode)
366{
367 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
368}
369
370/**
371 * inode_sb_list_add - add inode to the superblock list of inodes
372 * @inode: inode to add
373 */
374void inode_sb_list_add(struct inode *inode)
375{
376 spin_lock(&inode_lock);
377 __inode_sb_list_add(inode);
378 spin_unlock(&inode_lock);
379}
380EXPORT_SYMBOL_GPL(inode_sb_list_add);
381
382static inline void __inode_sb_list_del(struct inode *inode)
383{
384 list_del_init(&inode->i_sb_list);
385}
386
Dave Chinner4c51acb2010-10-23 06:58:09 -0400387static unsigned long hash(struct super_block *sb, unsigned long hashval)
388{
389 unsigned long tmp;
390
391 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
392 L1_CACHE_BYTES;
393 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
394 return tmp & I_HASHMASK;
395}
396
397/**
398 * __insert_inode_hash - hash an inode
399 * @inode: unhashed inode
400 * @hashval: unsigned long value used to locate this object in the
401 * inode_hashtable.
402 *
403 * Add an inode to the inode hash for this superblock.
404 */
405void __insert_inode_hash(struct inode *inode, unsigned long hashval)
406{
Christoph Hellwig646ec462010-10-23 07:15:32 -0400407 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
408
Dave Chinner4c51acb2010-10-23 06:58:09 -0400409 spin_lock(&inode_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400410 hlist_add_head(&inode->i_hash, b);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400411 spin_unlock(&inode_lock);
412}
413EXPORT_SYMBOL(__insert_inode_hash);
414
415/**
416 * __remove_inode_hash - remove an inode from the hash
417 * @inode: inode to unhash
418 *
419 * Remove an inode from the superblock.
420 */
421static void __remove_inode_hash(struct inode *inode)
422{
423 hlist_del_init(&inode->i_hash);
424}
425
426/**
427 * remove_inode_hash - remove an inode from the hash
428 * @inode: inode to unhash
429 *
430 * Remove an inode from the superblock.
431 */
432void remove_inode_hash(struct inode *inode)
433{
434 spin_lock(&inode_lock);
435 hlist_del_init(&inode->i_hash);
436 spin_unlock(&inode_lock);
437}
438EXPORT_SYMBOL(remove_inode_hash);
439
Al Virob0683aa2010-06-04 20:55:25 -0400440void end_writeback(struct inode *inode)
441{
442 might_sleep();
443 BUG_ON(inode->i_data.nrpages);
444 BUG_ON(!list_empty(&inode->i_data.private_list));
445 BUG_ON(!(inode->i_state & I_FREEING));
446 BUG_ON(inode->i_state & I_CLEAR);
447 inode_sync_wait(inode);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100448 /* don't need i_lock here, no concurrent mods to i_state */
Al Virob0683aa2010-06-04 20:55:25 -0400449 inode->i_state = I_FREEING | I_CLEAR;
450}
451EXPORT_SYMBOL(end_writeback);
452
Al Viro644da592010-06-07 13:21:05 -0400453static void evict(struct inode *inode)
Al Virob4272d42010-06-04 19:33:20 -0400454{
455 const struct super_operations *op = inode->i_sb->s_op;
456
Al Virobe7ce412010-06-04 19:40:39 -0400457 if (op->evict_inode) {
458 op->evict_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400459 } else {
460 if (inode->i_data.nrpages)
461 truncate_inode_pages(&inode->i_data, 0);
Al Viro30140832010-06-07 13:23:20 -0400462 end_writeback(inode);
Al Virob4272d42010-06-04 19:33:20 -0400463 }
Al Viro661074e2010-06-04 20:19:55 -0400464 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
465 bd_forget(inode);
466 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
467 cd_forget(inode);
Al Virob4272d42010-06-04 19:33:20 -0400468}
469
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470/*
471 * dispose_list - dispose of the contents of a local list
472 * @head: the head of the list to free
473 *
474 * Dispose-list gets a local list with local inodes in it, so it doesn't
475 * need to worry about list corruption and SMP locks.
476 */
477static void dispose_list(struct list_head *head)
478{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 while (!list_empty(head)) {
480 struct inode *inode;
481
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100482 inode = list_first_entry(head, struct inode, i_lru);
483 list_del_init(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Al Viro644da592010-06-07 13:21:05 -0400485 evict(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700486
487 spin_lock(&inode_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400488 __remove_inode_hash(inode);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400489 __inode_sb_list_del(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700490 spin_unlock(&inode_lock);
491
492 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495}
496
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497/**
Al Viro63997e92010-10-25 20:49:35 -0400498 * evict_inodes - evict all evictable inodes for a superblock
499 * @sb: superblock to operate on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 *
Al Viro63997e92010-10-25 20:49:35 -0400501 * Make sure that no inodes with zero refcount are retained. This is
502 * called by superblock shutdown after having MS_ACTIVE flag removed,
503 * so any inode reaching zero refcount during or after that call will
504 * be immediately evicted.
505 */
506void evict_inodes(struct super_block *sb)
507{
508 struct inode *inode, *next;
509 LIST_HEAD(dispose);
510
511 down_write(&iprune_sem);
512
513 spin_lock(&inode_lock);
514 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
515 if (atomic_read(&inode->i_count))
516 continue;
517
518 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
519 WARN_ON(1);
520 continue;
521 }
522
523 inode->i_state |= I_FREEING;
524
525 /*
526 * Move the inode off the IO lists and LRU once I_FREEING is
527 * set so that it won't get moved back on there if it is dirty.
528 */
529 list_move(&inode->i_lru, &dispose);
530 list_del_init(&inode->i_wb_list);
531 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
Nick Piggin86c87492011-01-07 17:49:18 +1100532 inodes_stat.nr_unused--;
Al Viro63997e92010-10-25 20:49:35 -0400533 }
534 spin_unlock(&inode_lock);
535
536 dispose_list(&dispose);
537 up_write(&iprune_sem);
538}
539
540/**
Christoph Hellwiga0318782010-10-24 19:40:33 +0200541 * invalidate_inodes - attempt to free all inodes on a superblock
542 * @sb: superblock to operate on
543 *
544 * Attempts to free all inodes for a given superblock. If there were any
545 * busy inodes return a non-zero value, else zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530547int invalidate_inodes(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400549 int busy = 0;
Christoph Hellwiga0318782010-10-24 19:40:33 +0200550 struct inode *inode, *next;
551 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700553 down_write(&iprune_sem);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200554
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 spin_lock(&inode_lock);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200556 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
Al Viro63997e92010-10-25 20:49:35 -0400557 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 continue;
Christoph Hellwig99a38912010-10-23 19:07:20 +0200559 if (atomic_read(&inode->i_count)) {
560 busy = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 continue;
562 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200563
Christoph Hellwig99a38912010-10-23 19:07:20 +0200564 inode->i_state |= I_FREEING;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100565
566 /*
567 * Move the inode off the IO lists and LRU once I_FREEING is
568 * set so that it won't get moved back on there if it is dirty.
569 */
Christoph Hellwiga0318782010-10-24 19:40:33 +0200570 list_move(&inode->i_lru, &dispose);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100571 list_del_init(&inode->i_wb_list);
Christoph Hellwig99a38912010-10-23 19:07:20 +0200572 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
Nick Piggin86c87492011-01-07 17:49:18 +1100573 inodes_stat.nr_unused--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 spin_unlock(&inode_lock);
576
Christoph Hellwiga0318782010-10-24 19:40:33 +0200577 dispose_list(&dispose);
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700578 up_write(&iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580 return busy;
581}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
583static int can_unuse(struct inode *inode)
584{
Nick Piggin9e38d862010-10-23 06:55:17 -0400585 if (inode->i_state & ~I_REFERENCED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 return 0;
587 if (inode_has_buffers(inode))
588 return 0;
589 if (atomic_read(&inode->i_count))
590 return 0;
591 if (inode->i_data.nrpages)
592 return 0;
593 return 1;
594}
595
596/*
Nick Piggin9e38d862010-10-23 06:55:17 -0400597 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
598 * temporary list and then are freed outside inode_lock by dispose_list().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 *
600 * Any inodes which are pinned purely because of attached pagecache have their
Nick Piggin9e38d862010-10-23 06:55:17 -0400601 * pagecache removed. If the inode has metadata buffers attached to
602 * mapping->private_list then try to remove them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 *
Nick Piggin9e38d862010-10-23 06:55:17 -0400604 * If the inode has the I_REFERENCED flag set, then it means that it has been
605 * used recently - the flag is set in iput_final(). When we encounter such an
606 * inode, clear the flag and move it to the back of the LRU so it gets another
607 * pass through the LRU before it gets reclaimed. This is necessary because of
608 * the fact we are doing lazy LRU updates to minimise lock contention so the
609 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
610 * with this flag set because they are the inodes that are out of order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 */
612static void prune_icache(int nr_to_scan)
613{
614 LIST_HEAD(freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 int nr_scanned;
616 unsigned long reap = 0;
617
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700618 down_read(&iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 spin_lock(&inode_lock);
620 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
621 struct inode *inode;
622
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100623 if (list_empty(&inode_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 break;
625
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100626 inode = list_entry(inode_lru.prev, struct inode, i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Nick Piggin9e38d862010-10-23 06:55:17 -0400628 /*
629 * Referenced or dirty inodes are still in use. Give them
630 * another pass through the LRU as we canot reclaim them now.
631 */
632 if (atomic_read(&inode->i_count) ||
633 (inode->i_state & ~I_REFERENCED)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100634 list_del_init(&inode->i_lru);
Nick Piggin86c87492011-01-07 17:49:18 +1100635 inodes_stat.nr_unused--;
Nick Piggin9e38d862010-10-23 06:55:17 -0400636 continue;
637 }
638
639 /* recently referenced inodes get one more pass */
640 if (inode->i_state & I_REFERENCED) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100641 list_move(&inode->i_lru, &inode_lru);
Nick Piggin9e38d862010-10-23 06:55:17 -0400642 inode->i_state &= ~I_REFERENCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 continue;
644 }
645 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
646 __iget(inode);
647 spin_unlock(&inode_lock);
648 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800649 reap += invalidate_mapping_pages(&inode->i_data,
650 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 iput(inode);
652 spin_lock(&inode_lock);
653
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100654 if (inode != list_entry(inode_lru.next,
655 struct inode, i_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 continue; /* wrong inode or list_empty */
657 if (!can_unuse(inode))
658 continue;
659 }
Nick Piggin7ef0d732009-03-12 14:31:38 -0700660 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 inode->i_state |= I_FREEING;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100662
663 /*
664 * Move the inode off the IO lists and LRU once I_FREEING is
665 * set so that it won't get moved back on there if it is dirty.
666 */
667 list_move(&inode->i_lru, &freeable);
668 list_del_init(&inode->i_wb_list);
Nick Piggin86c87492011-01-07 17:49:18 +1100669 inodes_stat.nr_unused--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 }
Christoph Lameterf8891e52006-06-30 01:55:45 -0700671 if (current_is_kswapd())
672 __count_vm_events(KSWAPD_INODESTEAL, reap);
673 else
674 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 spin_unlock(&inode_lock);
676
677 dispose_list(&freeable);
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700678 up_read(&iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679}
680
681/*
682 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
683 * "unused" means that no dentries are referring to the inodes: the files are
684 * not open and the dcache references to those inodes have already been
685 * reclaimed.
686 *
687 * This function is passed the number of inodes to scan, and it returns the
688 * total number of remaining possibly-reclaimable inodes.
689 */
Dave Chinner7f8275d2010-07-19 14:56:17 +1000690static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691{
692 if (nr) {
693 /*
694 * Nasty deadlock avoidance. We may hold various FS locks,
695 * and we don't want to recurse into the FS that called us
696 * in clear_inode() and friends..
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530697 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 if (!(gfp_mask & __GFP_FS))
699 return -1;
700 prune_icache(nr);
701 }
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400702 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703}
704
Rusty Russell8e1f9362007-07-17 04:03:17 -0700705static struct shrinker icache_shrinker = {
706 .shrink = shrink_icache_memory,
707 .seeks = DEFAULT_SEEKS,
708};
709
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710static void __wait_on_freeing_inode(struct inode *inode);
711/*
712 * Called with the inode lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530714static struct inode *find_inode(struct super_block *sb,
715 struct hlist_head *head,
716 int (*test)(struct inode *, void *),
717 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718{
719 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530720 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
722repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700723 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 if (inode->i_sb != sb)
725 continue;
726 if (!test(inode, data))
727 continue;
Al Viroa4ffdde2010-06-02 17:38:30 -0400728 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 __wait_on_freeing_inode(inode);
730 goto repeat;
731 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400732 __iget(inode);
733 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400735 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736}
737
738/*
739 * find_inode_fast is the fast path version of find_inode, see the comment at
740 * iget_locked for details.
741 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530742static struct inode *find_inode_fast(struct super_block *sb,
743 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
745 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530746 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
748repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700749 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 if (inode->i_ino != ino)
751 continue;
752 if (inode->i_sb != sb)
753 continue;
Al Viroa4ffdde2010-06-02 17:38:30 -0400754 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 __wait_on_freeing_inode(inode);
756 goto repeat;
757 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400758 __iget(inode);
759 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400761 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762}
763
Eric Dumazetf991bd22010-10-23 11:18:01 -0400764/*
765 * Each cpu owns a range of LAST_INO_BATCH numbers.
766 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
767 * to renew the exhausted range.
David Chinner8290c352008-10-30 17:35:24 +1100768 *
Eric Dumazetf991bd22010-10-23 11:18:01 -0400769 * This does not significantly increase overflow rate because every CPU can
770 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
771 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
772 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
773 * overflow rate by 2x, which does not seem too significant.
774 *
775 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
776 * error if st_ino won't fit in target struct field. Use 32bit counter
777 * here to attempt to avoid that.
David Chinner8290c352008-10-30 17:35:24 +1100778 */
Eric Dumazetf991bd22010-10-23 11:18:01 -0400779#define LAST_INO_BATCH 1024
780static DEFINE_PER_CPU(unsigned int, last_ino);
David Chinner8290c352008-10-30 17:35:24 +1100781
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400782unsigned int get_next_ino(void)
Eric Dumazetf991bd22010-10-23 11:18:01 -0400783{
784 unsigned int *p = &get_cpu_var(last_ino);
785 unsigned int res = *p;
786
787#ifdef CONFIG_SMP
788 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
789 static atomic_t shared_last_ino;
790 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
791
792 res = next - LAST_INO_BATCH;
793 }
794#endif
795
796 *p = ++res;
797 put_cpu_var(last_ino);
798 return res;
David Chinner8290c352008-10-30 17:35:24 +1100799}
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400800EXPORT_SYMBOL(get_next_ino);
David Chinner8290c352008-10-30 17:35:24 +1100801
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802/**
803 * new_inode - obtain an inode
804 * @sb: superblock
805 *
Mel Gorman769848c2007-07-17 04:03:05 -0700806 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800807 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700808 * If HIGHMEM pages are unsuitable or it is known that pages allocated
809 * for the page cache are not reclaimable or migratable,
810 * mapping_set_gfp_mask() must be called with suitable flags on the
811 * newly created inode's mapping
812 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 */
814struct inode *new_inode(struct super_block *sb)
815{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530816 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
818 spin_lock_prefetch(&inode_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530819
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 inode = alloc_inode(sb);
821 if (inode) {
822 spin_lock(&inode_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400823 __inode_sb_list_add(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 inode->i_state = 0;
825 spin_unlock(&inode_lock);
826 }
827 return inode;
828}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829EXPORT_SYMBOL(new_inode);
830
831void unlock_new_inode(struct inode *inode)
832{
Peter Zijlstra14358e62007-10-14 01:38:33 +0200833#ifdef CONFIG_DEBUG_LOCK_ALLOC
Namhyung Kima3314a02010-10-11 22:38:00 +0900834 if (S_ISDIR(inode->i_mode)) {
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200835 struct file_system_type *type = inode->i_sb->s_type;
836
Jan Kara9a7aa122009-06-04 15:26:49 +0200837 /* Set new key only if filesystem hasn't already changed it */
838 if (!lockdep_match_class(&inode->i_mutex,
839 &type->i_mutex_key)) {
840 /*
841 * ensure nobody is actually holding i_mutex
842 */
843 mutex_destroy(&inode->i_mutex);
844 mutex_init(&inode->i_mutex);
845 lockdep_set_class(&inode->i_mutex,
846 &type->i_mutex_dir_key);
847 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200848 }
Peter Zijlstra14358e62007-10-14 01:38:33 +0200849#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 /*
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100851 * This is special! We do not need the spinlock when clearing I_NEW,
Jan Kara580be082009-09-21 17:01:06 -0700852 * because we're guaranteed that nobody else tries to do anything about
853 * the state of the inode when it is locked, as we just created it (so
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100854 * there can be no old holders that haven't tested I_NEW).
Jan Kara580be082009-09-21 17:01:06 -0700855 * However we must emit the memory barrier so that other CPUs reliably
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100856 * see the clearing of I_NEW after the other inode initialisation has
Jan Kara580be082009-09-21 17:01:06 -0700857 * completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 */
Jan Kara580be082009-09-21 17:01:06 -0700859 smp_mb();
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100860 WARN_ON(!(inode->i_state & I_NEW));
861 inode->i_state &= ~I_NEW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 wake_up_inode(inode);
863}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864EXPORT_SYMBOL(unlock_new_inode);
865
866/*
867 * This is called without the inode lock held.. Be careful.
868 *
869 * We no longer cache the sb_flags in i_flags - see fs.h
870 * -- rmk@arm.uk.linux.org
871 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530872static struct inode *get_new_inode(struct super_block *sb,
873 struct hlist_head *head,
874 int (*test)(struct inode *, void *),
875 int (*set)(struct inode *, void *),
876 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530878 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
880 inode = alloc_inode(sb);
881 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530882 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
884 spin_lock(&inode_lock);
885 /* We released the lock, so.. */
886 old = find_inode(sb, head, test, data);
887 if (!old) {
888 if (set(inode, data))
889 goto set_failed;
890
Christoph Hellwig646ec462010-10-23 07:15:32 -0400891 hlist_add_head(&inode->i_hash, head);
892 __inode_sb_list_add(inode);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100893 inode->i_state = I_NEW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 spin_unlock(&inode_lock);
895
896 /* Return the locked inode with I_NEW set, the
897 * caller is responsible for filling in the contents
898 */
899 return inode;
900 }
901
902 /*
903 * Uhhuh, somebody else created the same inode under
904 * us. Use the old inode instead of the one we just
905 * allocated.
906 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 spin_unlock(&inode_lock);
908 destroy_inode(inode);
909 inode = old;
910 wait_on_inode(inode);
911 }
912 return inode;
913
914set_failed:
915 spin_unlock(&inode_lock);
916 destroy_inode(inode);
917 return NULL;
918}
919
920/*
921 * get_new_inode_fast is the fast path version of get_new_inode, see the
922 * comment at iget_locked for details.
923 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530924static struct inode *get_new_inode_fast(struct super_block *sb,
925 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530927 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
929 inode = alloc_inode(sb);
930 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530931 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
933 spin_lock(&inode_lock);
934 /* We released the lock, so.. */
935 old = find_inode_fast(sb, head, ino);
936 if (!old) {
937 inode->i_ino = ino;
Christoph Hellwig646ec462010-10-23 07:15:32 -0400938 hlist_add_head(&inode->i_hash, head);
939 __inode_sb_list_add(inode);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100940 inode->i_state = I_NEW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 spin_unlock(&inode_lock);
942
943 /* Return the locked inode with I_NEW set, the
944 * caller is responsible for filling in the contents
945 */
946 return inode;
947 }
948
949 /*
950 * Uhhuh, somebody else created the same inode under
951 * us. Use the old inode instead of the one we just
952 * allocated.
953 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 spin_unlock(&inode_lock);
955 destroy_inode(inode);
956 inode = old;
957 wait_on_inode(inode);
958 }
959 return inode;
960}
961
Christoph Hellwigad5e1952010-10-23 07:00:16 -0400962/*
963 * search the inode cache for a matching inode number.
964 * If we find one, then the inode number we are trying to
965 * allocate is not unique and so we should not use it.
966 *
967 * Returns 1 if the inode number is unique, 0 if it is not.
968 */
969static int test_inode_iunique(struct super_block *sb, unsigned long ino)
970{
971 struct hlist_head *b = inode_hashtable + hash(sb, ino);
972 struct hlist_node *node;
973 struct inode *inode;
974
975 hlist_for_each_entry(inode, node, b, i_hash) {
976 if (inode->i_ino == ino && inode->i_sb == sb)
977 return 0;
978 }
979
980 return 1;
981}
982
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983/**
984 * iunique - get a unique inode number
985 * @sb: superblock
986 * @max_reserved: highest reserved inode number
987 *
988 * Obtain an inode number that is unique on the system for a given
989 * superblock. This is used by file systems that have no natural
990 * permanent inode numbering system. An inode number is returned that
991 * is higher than the reserved limit but unique.
992 *
993 * BUGS:
994 * With a large number of inodes live on the file system this function
995 * currently becomes quite slow.
996 */
997ino_t iunique(struct super_block *sb, ino_t max_reserved)
998{
Jeff Layton866b04f2007-05-08 00:32:29 -0700999 /*
1000 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1001 * error if st_ino won't fit in target struct field. Use 32bit counter
1002 * here to attempt to avoid that.
1003 */
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001004 static DEFINE_SPINLOCK(iunique_lock);
Jeff Layton866b04f2007-05-08 00:32:29 -07001005 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001008 spin_lock(&inode_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001009 spin_lock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001010 do {
1011 if (counter <= max_reserved)
1012 counter = max_reserved + 1;
1013 res = counter++;
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001014 } while (!test_inode_iunique(sb, res));
1015 spin_unlock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001016 spin_unlock(&inode_lock);
1017
1018 return res;
1019}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020EXPORT_SYMBOL(iunique);
1021
1022struct inode *igrab(struct inode *inode)
1023{
1024 spin_lock(&inode_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -04001025 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 __iget(inode);
1027 else
1028 /*
1029 * Handle the case where s_op->clear_inode is not been
1030 * called yet, and somebody is calling igrab
1031 * while the inode is getting freed.
1032 */
1033 inode = NULL;
1034 spin_unlock(&inode_lock);
1035 return inode;
1036}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037EXPORT_SYMBOL(igrab);
1038
1039/**
1040 * ifind - internal function, you want ilookup5() or iget5().
1041 * @sb: super block of file system to search
1042 * @head: the head of the list to search
1043 * @test: callback used for comparisons between inodes
1044 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001045 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 *
1047 * ifind() searches for the inode specified by @data in the inode
1048 * cache. This is a generalized version of ifind_fast() for file systems where
1049 * the inode number is not sufficient for unique identification of an inode.
1050 *
1051 * If the inode is in the cache, the inode is returned with an incremented
1052 * reference count.
1053 *
1054 * Otherwise NULL is returned.
1055 *
1056 * Note, @test is called with the inode_lock held, so can't sleep.
1057 */
Matt Mackall5d2bea42006-01-08 01:05:21 -08001058static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001060 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061{
1062 struct inode *inode;
1063
1064 spin_lock(&inode_lock);
1065 inode = find_inode(sb, head, test, data);
1066 if (inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001068 if (likely(wait))
1069 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 return inode;
1071 }
1072 spin_unlock(&inode_lock);
1073 return NULL;
1074}
1075
1076/**
1077 * ifind_fast - internal function, you want ilookup() or iget().
1078 * @sb: super block of file system to search
1079 * @head: head of the list to search
1080 * @ino: inode number to search for
1081 *
1082 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1083 * file systems where the inode number is sufficient for unique identification
1084 * of an inode.
1085 *
1086 * If the inode is in the cache, the inode is returned with an incremented
1087 * reference count.
1088 *
1089 * Otherwise NULL is returned.
1090 */
Matt Mackall5d2bea42006-01-08 01:05:21 -08001091static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 struct hlist_head *head, unsigned long ino)
1093{
1094 struct inode *inode;
1095
1096 spin_lock(&inode_lock);
1097 inode = find_inode_fast(sb, head, ino);
1098 if (inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 spin_unlock(&inode_lock);
1100 wait_on_inode(inode);
1101 return inode;
1102 }
1103 spin_unlock(&inode_lock);
1104 return NULL;
1105}
1106
1107/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001108 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 * @sb: super block of file system to search
1110 * @hashval: hash value (usually inode number) to search for
1111 * @test: callback used for comparisons between inodes
1112 * @data: opaque data pointer to pass to @test
1113 *
1114 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1115 * @data in the inode cache. This is a generalized version of ilookup() for
1116 * file systems where the inode number is not sufficient for unique
1117 * identification of an inode.
1118 *
1119 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001120 * reference count. Note, the inode lock is not waited upon so you have to be
1121 * very careful what you do with the returned inode. You probably should be
1122 * using ilookup5() instead.
1123 *
1124 * Otherwise NULL is returned.
1125 *
1126 * Note, @test is called with the inode_lock held, so can't sleep.
1127 */
1128struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1129 int (*test)(struct inode *, void *), void *data)
1130{
1131 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1132
1133 return ifind(sb, head, test, data, 0);
1134}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001135EXPORT_SYMBOL(ilookup5_nowait);
1136
1137/**
1138 * ilookup5 - search for an inode in the inode cache
1139 * @sb: super block of file system to search
1140 * @hashval: hash value (usually inode number) to search for
1141 * @test: callback used for comparisons between inodes
1142 * @data: opaque data pointer to pass to @test
1143 *
1144 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1145 * @data in the inode cache. This is a generalized version of ilookup() for
1146 * file systems where the inode number is not sufficient for unique
1147 * identification of an inode.
1148 *
1149 * If the inode is in the cache, the inode lock is waited upon and the inode is
1150 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 *
1152 * Otherwise NULL is returned.
1153 *
1154 * Note, @test is called with the inode_lock held, so can't sleep.
1155 */
1156struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1157 int (*test)(struct inode *, void *), void *data)
1158{
1159 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1160
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001161 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163EXPORT_SYMBOL(ilookup5);
1164
1165/**
1166 * ilookup - search for an inode in the inode cache
1167 * @sb: super block of file system to search
1168 * @ino: inode number to search for
1169 *
1170 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1171 * This is for file systems where the inode number is sufficient for unique
1172 * identification of an inode.
1173 *
1174 * If the inode is in the cache, the inode is returned with an incremented
1175 * reference count.
1176 *
1177 * Otherwise NULL is returned.
1178 */
1179struct inode *ilookup(struct super_block *sb, unsigned long ino)
1180{
1181 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1182
1183 return ifind_fast(sb, head, ino);
1184}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185EXPORT_SYMBOL(ilookup);
1186
1187/**
1188 * iget5_locked - obtain an inode from a mounted file system
1189 * @sb: super block of file system
1190 * @hashval: hash value (usually inode number) to get
1191 * @test: callback used for comparisons between inodes
1192 * @set: callback used to initialize a new struct inode
1193 * @data: opaque data pointer to pass to @test and @set
1194 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1196 * and @data in the inode cache and if present it is returned with an increased
1197 * reference count. This is a generalized version of iget_locked() for file
1198 * systems where the inode number is not sufficient for unique identification
1199 * of an inode.
1200 *
1201 * If the inode is not in cache, get_new_inode() is called to allocate a new
1202 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1203 * file system gets to fill it in before unlocking it via unlock_new_inode().
1204 *
1205 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1206 */
1207struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1208 int (*test)(struct inode *, void *),
1209 int (*set)(struct inode *, void *), void *data)
1210{
1211 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1212 struct inode *inode;
1213
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001214 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 if (inode)
1216 return inode;
1217 /*
1218 * get_new_inode() will do the right thing, re-trying the search
1219 * in case it had to block at any point.
1220 */
1221 return get_new_inode(sb, head, test, set, data);
1222}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223EXPORT_SYMBOL(iget5_locked);
1224
1225/**
1226 * iget_locked - obtain an inode from a mounted file system
1227 * @sb: super block of file system
1228 * @ino: inode number to get
1229 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1231 * the inode cache and if present it is returned with an increased reference
1232 * count. This is for file systems where the inode number is sufficient for
1233 * unique identification of an inode.
1234 *
1235 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1236 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1237 * The file system gets to fill it in before unlocking it via
1238 * unlock_new_inode().
1239 */
1240struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1241{
1242 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1243 struct inode *inode;
1244
1245 inode = ifind_fast(sb, head, ino);
1246 if (inode)
1247 return inode;
1248 /*
1249 * get_new_inode_fast() will do the right thing, re-trying the search
1250 * in case it had to block at any point.
1251 */
1252 return get_new_inode_fast(sb, head, ino);
1253}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254EXPORT_SYMBOL(iget_locked);
1255
Al Viro261bca82008-12-30 01:48:21 -05001256int insert_inode_locked(struct inode *inode)
1257{
1258 struct super_block *sb = inode->i_sb;
1259 ino_t ino = inode->i_ino;
1260 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001261
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001262 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001263 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001264 struct hlist_node *node;
1265 struct inode *old = NULL;
Al Viro261bca82008-12-30 01:48:21 -05001266 spin_lock(&inode_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001267 hlist_for_each_entry(old, node, head, i_hash) {
1268 if (old->i_ino != ino)
1269 continue;
1270 if (old->i_sb != sb)
1271 continue;
Al Viroa4ffdde2010-06-02 17:38:30 -04001272 if (old->i_state & (I_FREEING|I_WILL_FREE))
Al Viro72a43d62009-05-13 19:13:40 +01001273 continue;
1274 break;
1275 }
1276 if (likely(!node)) {
Al Viro261bca82008-12-30 01:48:21 -05001277 hlist_add_head(&inode->i_hash, head);
1278 spin_unlock(&inode_lock);
1279 return 0;
1280 }
1281 __iget(old);
1282 spin_unlock(&inode_lock);
1283 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001284 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001285 iput(old);
1286 return -EBUSY;
1287 }
1288 iput(old);
1289 }
1290}
Al Viro261bca82008-12-30 01:48:21 -05001291EXPORT_SYMBOL(insert_inode_locked);
1292
1293int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1294 int (*test)(struct inode *, void *), void *data)
1295{
1296 struct super_block *sb = inode->i_sb;
1297 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001298
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001299 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001300
1301 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001302 struct hlist_node *node;
1303 struct inode *old = NULL;
1304
Al Viro261bca82008-12-30 01:48:21 -05001305 spin_lock(&inode_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001306 hlist_for_each_entry(old, node, head, i_hash) {
1307 if (old->i_sb != sb)
1308 continue;
1309 if (!test(old, data))
1310 continue;
Al Viroa4ffdde2010-06-02 17:38:30 -04001311 if (old->i_state & (I_FREEING|I_WILL_FREE))
Al Viro72a43d62009-05-13 19:13:40 +01001312 continue;
1313 break;
1314 }
1315 if (likely(!node)) {
Al Viro261bca82008-12-30 01:48:21 -05001316 hlist_add_head(&inode->i_hash, head);
1317 spin_unlock(&inode_lock);
1318 return 0;
1319 }
1320 __iget(old);
1321 spin_unlock(&inode_lock);
1322 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001323 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001324 iput(old);
1325 return -EBUSY;
1326 }
1327 iput(old);
1328 }
1329}
Al Viro261bca82008-12-30 01:48:21 -05001330EXPORT_SYMBOL(insert_inode_locked4);
1331
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
Al Viro45321ac2010-06-07 13:43:19 -04001333int generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334{
Al Viro45321ac2010-06-07 13:43:19 -04001335 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337EXPORT_SYMBOL(generic_delete_inode);
1338
Al Viro45321ac2010-06-07 13:43:19 -04001339/*
1340 * Normal UNIX filesystem behaviour: delete the
1341 * inode when the usage count drops to zero, and
1342 * i_nlink is zero.
Jan Kara22fe40422009-09-18 13:05:44 -07001343 */
Al Viro45321ac2010-06-07 13:43:19 -04001344int generic_drop_inode(struct inode *inode)
1345{
Al Viro1d3382cb2010-10-23 15:19:20 -04001346 return !inode->i_nlink || inode_unhashed(inode);
Al Viro45321ac2010-06-07 13:43:19 -04001347}
1348EXPORT_SYMBOL_GPL(generic_drop_inode);
1349
1350/*
1351 * Called when we're dropping the last reference
1352 * to an inode.
1353 *
1354 * Call the FS "drop_inode()" function, defaulting to
1355 * the legacy UNIX filesystem behaviour. If it tells
1356 * us to evict inode, do so. Otherwise, retain inode
1357 * in cache if fs is alive, sync and evict if fs is
1358 * shutting down.
1359 */
1360static void iput_final(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361{
1362 struct super_block *sb = inode->i_sb;
Al Viro45321ac2010-06-07 13:43:19 -04001363 const struct super_operations *op = inode->i_sb->s_op;
1364 int drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
Al Viro45321ac2010-06-07 13:43:19 -04001366 if (op && op->drop_inode)
1367 drop = op->drop_inode(inode);
1368 else
1369 drop = generic_drop_inode(inode);
1370
1371 if (!drop) {
Christoph Hellwigacb0c852007-05-08 00:25:52 -07001372 if (sb->s_flags & MS_ACTIVE) {
Nick Piggin9e38d862010-10-23 06:55:17 -04001373 inode->i_state |= I_REFERENCED;
1374 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1375 inode_lru_list_add(inode);
1376 }
Alexander Viro991114c2005-06-23 00:09:01 -07001377 spin_unlock(&inode_lock);
Al Viro45321ac2010-06-07 13:43:19 -04001378 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001379 }
Nick Piggin7ef0d732009-03-12 14:31:38 -07001380 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001381 inode->i_state |= I_WILL_FREE;
1382 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 write_inode_now(inode, 1);
1384 spin_lock(&inode_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001385 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001386 inode->i_state &= ~I_WILL_FREE;
Dave Chinner4c51acb2010-10-23 06:58:09 -04001387 __remove_inode_hash(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001389
Nick Piggin7ef0d732009-03-12 14:31:38 -07001390 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001391 inode->i_state |= I_FREEING;
Nick Piggin9e38d862010-10-23 06:55:17 -04001392
1393 /*
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001394 * Move the inode off the IO lists and LRU once I_FREEING is
1395 * set so that it won't get moved back on there if it is dirty.
Nick Piggin9e38d862010-10-23 06:55:17 -04001396 */
1397 inode_lru_list_del(inode);
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001398 list_del_init(&inode->i_wb_list);
Nick Piggin9e38d862010-10-23 06:55:17 -04001399
Christoph Hellwig646ec462010-10-23 07:15:32 -04001400 __inode_sb_list_del(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 spin_unlock(&inode_lock);
Al Viro644da592010-06-07 13:21:05 -04001402 evict(inode);
Dave Chinner4c51acb2010-10-23 06:58:09 -04001403 remove_inode_hash(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001404 wake_up_inode(inode);
Al Viro45321ac2010-06-07 13:43:19 -04001405 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 destroy_inode(inode);
1407}
1408
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301410 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 * @inode: inode to put
1412 *
1413 * Puts an inode, dropping its usage count. If the inode use count hits
1414 * zero, the inode is then freed and may also be destroyed.
1415 *
1416 * Consequently, iput() can sleep.
1417 */
1418void iput(struct inode *inode)
1419{
1420 if (inode) {
Al Viroa4ffdde2010-06-02 17:38:30 -04001421 BUG_ON(inode->i_state & I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1424 iput_final(inode);
1425 }
1426}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427EXPORT_SYMBOL(iput);
1428
1429/**
1430 * bmap - find a block number in a file
1431 * @inode: inode of file
1432 * @block: block to find
1433 *
1434 * Returns the block number on the device holding the inode that
1435 * is the disk block number for the block of the file requested.
1436 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301437 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 * file.
1439 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301440sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441{
1442 sector_t res = 0;
1443 if (inode->i_mapping->a_ops->bmap)
1444 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1445 return res;
1446}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447EXPORT_SYMBOL(bmap);
1448
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001449/*
1450 * With relative atime, only update atime if the previous atime is
1451 * earlier than either the ctime or mtime or if at least a day has
1452 * passed since the last atime update.
1453 */
1454static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1455 struct timespec now)
1456{
1457
1458 if (!(mnt->mnt_flags & MNT_RELATIME))
1459 return 1;
1460 /*
1461 * Is mtime younger than atime? If yes, update atime:
1462 */
1463 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1464 return 1;
1465 /*
1466 * Is ctime younger than atime? If yes, update atime:
1467 */
1468 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1469 return 1;
1470
1471 /*
1472 * Is the previous atime value older than a day? If yes,
1473 * update atime:
1474 */
1475 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1476 return 1;
1477 /*
1478 * Good, we can skip the atime update:
1479 */
1480 return 0;
1481}
1482
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001484 * touch_atime - update the access time
1485 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001486 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 *
1488 * Update the accessed time on an inode and mark it for writeback.
1489 * This function automatically handles read only file systems and media,
1490 * as well as the "noatime" flag and inode specific "noatime" markers.
1491 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001492void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001494 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 struct timespec now;
1496
Andrew Mortonb2276132006-12-13 00:34:33 -08001497 if (inode->i_flags & S_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001498 return;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001499 if (IS_NOATIME(inode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001500 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001501 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001502 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001503
Dave Hansencdb70f32008-02-15 14:37:41 -08001504 if (mnt->mnt_flags & MNT_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001505 return;
Dave Hansencdb70f32008-02-15 14:37:41 -08001506 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001507 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001508
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 now = current_fs_time(inode->i_sb);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001510
1511 if (!relatime_need_update(mnt, inode, now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001512 return;
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001513
Valerie Henson47ae32d2006-12-13 00:34:34 -08001514 if (timespec_equal(&inode->i_atime, &now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001515 return;
1516
1517 if (mnt_want_write(mnt))
1518 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001519
1520 inode->i_atime = now;
1521 mark_inode_dirty_sync(inode);
Dave Hansencdb70f32008-02-15 14:37:41 -08001522 mnt_drop_write(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001524EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
1526/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001527 * file_update_time - update mtime and ctime time
1528 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001530 * Update the mtime and ctime members of an inode and mark the inode
1531 * for writeback. Note that this function is meant exclusively for
1532 * usage in the file write path of filesystems, and filesystems may
1533 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001534 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Christoph Hellwig870f4812006-01-09 20:52:01 -08001535 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 */
1537
Christoph Hellwig870f4812006-01-09 20:52:01 -08001538void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001540 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 struct timespec now;
Andi Kleence06e0b2009-09-18 13:05:48 -07001542 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
Andi Kleence06e0b2009-09-18 13:05:48 -07001544 /* First try to exhaust all avenues to not sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 if (IS_NOCMTIME(inode))
1546 return;
Dave Hansen20ddee22008-02-15 14:37:43 -08001547
Andi Kleence06e0b2009-09-18 13:05:48 -07001548 now = current_fs_time(inode->i_sb);
1549 if (!timespec_equal(&inode->i_mtime, &now))
1550 sync_it = S_MTIME;
1551
1552 if (!timespec_equal(&inode->i_ctime, &now))
1553 sync_it |= S_CTIME;
1554
1555 if (IS_I_VERSION(inode))
1556 sync_it |= S_VERSION;
1557
1558 if (!sync_it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 return;
1560
Andi Kleence06e0b2009-09-18 13:05:48 -07001561 /* Finally allowed to write? Takes lock. */
1562 if (mnt_want_write_file(file))
1563 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Andi Kleence06e0b2009-09-18 13:05:48 -07001565 /* Only change inode inside the lock region */
1566 if (sync_it & S_VERSION)
Jean Noel Cordenner7a224222008-01-28 23:58:27 -05001567 inode_inc_iversion(inode);
Andi Kleence06e0b2009-09-18 13:05:48 -07001568 if (sync_it & S_CTIME)
1569 inode->i_ctime = now;
1570 if (sync_it & S_MTIME)
1571 inode->i_mtime = now;
1572 mark_inode_dirty_sync(inode);
Dave Hansen20ddee22008-02-15 14:37:43 -08001573 mnt_drop_write(file->f_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001575EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576
1577int inode_needs_sync(struct inode *inode)
1578{
1579 if (IS_SYNC(inode))
1580 return 1;
1581 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1582 return 1;
1583 return 0;
1584}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585EXPORT_SYMBOL(inode_needs_sync);
1586
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587int inode_wait(void *word)
1588{
1589 schedule();
1590 return 0;
1591}
Stephen Rothwelld44dab82008-11-10 17:06:05 +11001592EXPORT_SYMBOL(inode_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
1594/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001595 * If we try to find an inode in the inode hash while it is being
1596 * deleted, we have to wait until the filesystem completes its
1597 * deletion before reporting that it isn't found. This function waits
1598 * until the deletion _might_ have completed. Callers are responsible
1599 * to recheck inode state.
1600 *
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001601 * It doesn't matter if I_NEW is not set initially, a call to
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001602 * wake_up_inode() after removing from the hash list will DTRT.
1603 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 * This is called with inode_lock held.
1605 */
1606static void __wait_on_freeing_inode(struct inode *inode)
1607{
1608 wait_queue_head_t *wq;
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001609 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1610 wq = bit_waitqueue(&inode->i_state, __I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1612 spin_unlock(&inode_lock);
1613 schedule();
1614 finish_wait(wq, &wait.wait);
1615 spin_lock(&inode_lock);
1616}
1617
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618static __initdata unsigned long ihash_entries;
1619static int __init set_ihash_entries(char *str)
1620{
1621 if (!str)
1622 return 0;
1623 ihash_entries = simple_strtoul(str, &str, 0);
1624 return 1;
1625}
1626__setup("ihash_entries=", set_ihash_entries);
1627
1628/*
1629 * Initialize the waitqueues and inode hash table.
1630 */
1631void __init inode_init_early(void)
1632{
1633 int loop;
1634
1635 /* If hashes are distributed across NUMA nodes, defer
1636 * hash allocation until vmalloc space is available.
1637 */
1638 if (hashdist)
1639 return;
1640
1641 inode_hashtable =
1642 alloc_large_system_hash("Inode-cache",
1643 sizeof(struct hlist_head),
1644 ihash_entries,
1645 14,
1646 HASH_EARLY,
1647 &i_hash_shift,
1648 &i_hash_mask,
1649 0);
1650
1651 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1652 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1653}
1654
Denis Cheng74bf17c2007-10-16 23:26:30 -07001655void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656{
1657 int loop;
1658
1659 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001660 inode_cachep = kmem_cache_create("inode_cache",
1661 sizeof(struct inode),
1662 0,
1663 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1664 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001665 init_once);
Rusty Russell8e1f9362007-07-17 04:03:17 -07001666 register_shrinker(&icache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
1668 /* Hash may have been set up in inode_init_early */
1669 if (!hashdist)
1670 return;
1671
1672 inode_hashtable =
1673 alloc_large_system_hash("Inode-cache",
1674 sizeof(struct hlist_head),
1675 ihash_entries,
1676 14,
1677 0,
1678 &i_hash_shift,
1679 &i_hash_mask,
1680 0);
1681
1682 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1683 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1684}
1685
1686void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1687{
1688 inode->i_mode = mode;
1689 if (S_ISCHR(mode)) {
1690 inode->i_fop = &def_chr_fops;
1691 inode->i_rdev = rdev;
1692 } else if (S_ISBLK(mode)) {
1693 inode->i_fop = &def_blk_fops;
1694 inode->i_rdev = rdev;
1695 } else if (S_ISFIFO(mode))
1696 inode->i_fop = &def_fifo_fops;
1697 else if (S_ISSOCK(mode))
1698 inode->i_fop = &bad_sock_fops;
1699 else
Manish Katiyaraf0d9ae2009-09-18 13:05:43 -07001700 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1701 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1702 inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703}
1704EXPORT_SYMBOL(init_special_inode);
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001705
1706/**
1707 * Init uid,gid,mode for new inode according to posix standards
1708 * @inode: New inode
1709 * @dir: Directory inode
1710 * @mode: mode of the new inode
1711 */
1712void inode_init_owner(struct inode *inode, const struct inode *dir,
1713 mode_t mode)
1714{
1715 inode->i_uid = current_fsuid();
1716 if (dir && dir->i_mode & S_ISGID) {
1717 inode->i_gid = dir->i_gid;
1718 if (S_ISDIR(mode))
1719 mode |= S_ISGID;
1720 } else
1721 inode->i_gid = current_fsgid();
1722 inode->i_mode = mode;
1723}
1724EXPORT_SYMBOL(inode_init_owner);