blob: ae2727ab0c3ab7695b14f256da0bccdfd3668d81 [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
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400105static struct percpu_counter nr_inodes __cacheline_aligned_in_smp;
106static struct percpu_counter nr_inodes_unused __cacheline_aligned_in_smp;
107
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530108static struct kmem_cache *inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400110static inline int get_nr_inodes(void)
111{
112 return percpu_counter_sum_positive(&nr_inodes);
113}
114
115static inline int get_nr_inodes_unused(void)
116{
117 return percpu_counter_sum_positive(&nr_inodes_unused);
118}
119
120int get_nr_dirty_inodes(void)
121{
122 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
123 return nr_dirty > 0 ? nr_dirty : 0;
124
125}
126
127/*
128 * Handle nr_inode sysctl
129 */
130#ifdef CONFIG_SYSCTL
131int proc_nr_inodes(ctl_table *table, int write,
132 void __user *buffer, size_t *lenp, loff_t *ppos)
133{
134 inodes_stat.nr_inodes = get_nr_inodes();
135 inodes_stat.nr_unused = get_nr_inodes_unused();
136 return proc_dointvec(table, write, buffer, lenp, ppos);
137}
138#endif
139
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700140static void wake_up_inode(struct inode *inode)
141{
142 /*
143 * Prevent speculative execution through spin_unlock(&inode_lock);
144 */
145 smp_mb();
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100146 wake_up_bit(&inode->i_state, __I_NEW);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700147}
148
David Chinner2cb15992008-10-30 17:32:23 +1100149/**
150 * inode_init_always - perform inode structure intialisation
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800151 * @sb: superblock inode belongs to
152 * @inode: inode to initialise
David Chinner2cb15992008-10-30 17:32:23 +1100153 *
154 * These are initializations that need to be done on every inode
155 * allocation as the fields are not initialised by slab allocation.
156 */
Christoph Hellwig54e34622009-08-07 14:38:25 -0300157int inode_init_always(struct super_block *sb, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158{
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700159 static const struct address_space_operations empty_aops;
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700160 static const struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800161 static const struct file_operations empty_fops;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530162 struct address_space *const mapping = &inode->i_data;
David Chinner2cb15992008-10-30 17:32:23 +1100163
164 inode->i_sb = sb;
165 inode->i_blkbits = sb->s_blocksize_bits;
166 inode->i_flags = 0;
167 atomic_set(&inode->i_count, 1);
168 inode->i_op = &empty_iops;
169 inode->i_fop = &empty_fops;
170 inode->i_nlink = 1;
Al Viro56ff5ef2008-12-09 09:34:39 -0500171 inode->i_uid = 0;
172 inode->i_gid = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100173 atomic_set(&inode->i_writecount, 0);
174 inode->i_size = 0;
175 inode->i_blocks = 0;
176 inode->i_bytes = 0;
177 inode->i_generation = 0;
178#ifdef CONFIG_QUOTA
179 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
180#endif
181 inode->i_pipe = NULL;
182 inode->i_bdev = NULL;
183 inode->i_cdev = NULL;
184 inode->i_rdev = 0;
185 inode->dirtied_when = 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500186
187 if (security_inode_alloc(inode))
Christoph Hellwig54e34622009-08-07 14:38:25 -0300188 goto out;
David Chinner2cb15992008-10-30 17:32:23 +1100189 spin_lock_init(&inode->i_lock);
190 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
191
192 mutex_init(&inode->i_mutex);
193 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
194
195 init_rwsem(&inode->i_alloc_sem);
196 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
197
198 mapping->a_ops = &empty_aops;
199 mapping->host = inode;
200 mapping->flags = 0;
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800201 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
David Chinner2cb15992008-10-30 17:32:23 +1100202 mapping->assoc_mapping = NULL;
203 mapping->backing_dev_info = &default_backing_dev_info;
204 mapping->writeback_index = 0;
205
206 /*
207 * If the block_device provides a backing_dev_info for client
208 * inodes then use that. Otherwise the inode share the bdev's
209 * backing_dev_info.
210 */
211 if (sb->s_bdev) {
212 struct backing_dev_info *bdi;
213
Jens Axboe2c96ce92009-09-15 09:43:56 +0200214 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
David Chinner2cb15992008-10-30 17:32:23 +1100215 mapping->backing_dev_info = bdi;
216 }
217 inode->i_private = NULL;
218 inode->i_mapping = mapping;
Al Virof19d4a82009-06-08 19:50:45 -0400219#ifdef CONFIG_FS_POSIX_ACL
220 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
221#endif
David Chinner2cb15992008-10-30 17:32:23 +1100222
Eric Paris3be25f42009-05-21 17:01:26 -0400223#ifdef CONFIG_FSNOTIFY
224 inode->i_fsnotify_mask = 0;
225#endif
226
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400227 percpu_counter_inc(&nr_inodes);
228
Christoph Hellwig54e34622009-08-07 14:38:25 -0300229 return 0;
Christoph Hellwig54e34622009-08-07 14:38:25 -0300230out:
231 return -ENOMEM;
David Chinner2cb15992008-10-30 17:32:23 +1100232}
233EXPORT_SYMBOL(inode_init_always);
234
235static struct inode *alloc_inode(struct super_block *sb)
236{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 struct inode *inode;
238
239 if (sb->s_op->alloc_inode)
240 inode = sb->s_op->alloc_inode(sb);
241 else
David Chinner2cb15992008-10-30 17:32:23 +1100242 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Christoph Hellwig54e34622009-08-07 14:38:25 -0300244 if (!inode)
245 return NULL;
246
247 if (unlikely(inode_init_always(sb, inode))) {
248 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 return NULL;
253 }
254
255 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
257
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300258void __destroy_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200260 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 security_inode_free(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400262 fsnotify_inode_delete(inode);
Al Virof19d4a82009-06-08 19:50:45 -0400263#ifdef CONFIG_FS_POSIX_ACL
264 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
265 posix_acl_release(inode->i_acl);
266 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
267 posix_acl_release(inode->i_default_acl);
268#endif
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400269 percpu_counter_dec(&nr_inodes);
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300270}
271EXPORT_SYMBOL(__destroy_inode);
272
Christoph Hellwig56b0dac2010-10-06 10:48:55 +0200273static void destroy_inode(struct inode *inode)
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300274{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100275 BUG_ON(!list_empty(&inode->i_lru));
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300276 __destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 if (inode->i_sb->s_op->destroy_inode)
278 inode->i_sb->s_op->destroy_inode(inode);
279 else
280 kmem_cache_free(inode_cachep, (inode));
281}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282
283/*
284 * These are initializations that only need to be done
285 * once, because the fields are idempotent across use
286 * of the inode, so let the slab aware of that.
287 */
288void inode_init_once(struct inode *inode)
289{
290 memset(inode, 0, sizeof(*inode));
291 INIT_HLIST_NODE(&inode->i_hash);
292 INIT_LIST_HEAD(&inode->i_dentry);
293 INIT_LIST_HEAD(&inode->i_devices);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100294 INIT_LIST_HEAD(&inode->i_wb_list);
295 INIT_LIST_HEAD(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
Nick Piggin19fd6232008-07-25 19:45:32 -0700297 spin_lock_init(&inode->i_data.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 spin_lock_init(&inode->i_data.i_mmap_lock);
299 INIT_LIST_HEAD(&inode->i_data.private_list);
300 spin_lock_init(&inode->i_data.private_lock);
301 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
302 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 i_size_ordered_init(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400304#ifdef CONFIG_FSNOTIFY
Eric Parise61ce862009-12-17 21:24:24 -0500305 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
Eric Paris3be25f42009-05-21 17:01:26 -0400306#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308EXPORT_SYMBOL(inode_init_once);
309
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700310static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530312 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
Christoph Lametera35afb82007-05-16 22:10:57 -0700314 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315}
316
317/*
318 * inode_lock must be held
319 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530320void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321{
Nick Piggin9e38d862010-10-23 06:55:17 -0400322 atomic_inc(&inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323}
324
Al Viro7de9c6ee2010-10-23 11:11:40 -0400325/*
326 * get additional reference to inode; caller must already hold one.
327 */
328void ihold(struct inode *inode)
329{
330 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
331}
332EXPORT_SYMBOL(ihold);
333
Nick Piggin9e38d862010-10-23 06:55:17 -0400334static void inode_lru_list_add(struct inode *inode)
335{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100336 if (list_empty(&inode->i_lru)) {
337 list_add(&inode->i_lru, &inode_lru);
Nick Piggin9e38d862010-10-23 06:55:17 -0400338 percpu_counter_inc(&nr_inodes_unused);
339 }
340}
341
342static void inode_lru_list_del(struct inode *inode)
343{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100344 if (!list_empty(&inode->i_lru)) {
345 list_del_init(&inode->i_lru);
Nick Piggin9e38d862010-10-23 06:55:17 -0400346 percpu_counter_dec(&nr_inodes_unused);
347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348}
349
Christoph Hellwig646ec462010-10-23 07:15:32 -0400350static inline void __inode_sb_list_add(struct inode *inode)
351{
352 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
353}
354
355/**
356 * inode_sb_list_add - add inode to the superblock list of inodes
357 * @inode: inode to add
358 */
359void inode_sb_list_add(struct inode *inode)
360{
361 spin_lock(&inode_lock);
362 __inode_sb_list_add(inode);
363 spin_unlock(&inode_lock);
364}
365EXPORT_SYMBOL_GPL(inode_sb_list_add);
366
367static inline void __inode_sb_list_del(struct inode *inode)
368{
369 list_del_init(&inode->i_sb_list);
370}
371
Dave Chinner4c51acb2010-10-23 06:58:09 -0400372static unsigned long hash(struct super_block *sb, unsigned long hashval)
373{
374 unsigned long tmp;
375
376 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
377 L1_CACHE_BYTES;
378 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
379 return tmp & I_HASHMASK;
380}
381
382/**
383 * __insert_inode_hash - hash an inode
384 * @inode: unhashed inode
385 * @hashval: unsigned long value used to locate this object in the
386 * inode_hashtable.
387 *
388 * Add an inode to the inode hash for this superblock.
389 */
390void __insert_inode_hash(struct inode *inode, unsigned long hashval)
391{
Christoph Hellwig646ec462010-10-23 07:15:32 -0400392 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
393
Dave Chinner4c51acb2010-10-23 06:58:09 -0400394 spin_lock(&inode_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400395 hlist_add_head(&inode->i_hash, b);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400396 spin_unlock(&inode_lock);
397}
398EXPORT_SYMBOL(__insert_inode_hash);
399
400/**
401 * __remove_inode_hash - remove an inode from the hash
402 * @inode: inode to unhash
403 *
404 * Remove an inode from the superblock.
405 */
406static void __remove_inode_hash(struct inode *inode)
407{
408 hlist_del_init(&inode->i_hash);
409}
410
411/**
412 * remove_inode_hash - remove an inode from the hash
413 * @inode: inode to unhash
414 *
415 * Remove an inode from the superblock.
416 */
417void remove_inode_hash(struct inode *inode)
418{
419 spin_lock(&inode_lock);
420 hlist_del_init(&inode->i_hash);
421 spin_unlock(&inode_lock);
422}
423EXPORT_SYMBOL(remove_inode_hash);
424
Al Virob0683aa2010-06-04 20:55:25 -0400425void end_writeback(struct inode *inode)
426{
427 might_sleep();
428 BUG_ON(inode->i_data.nrpages);
429 BUG_ON(!list_empty(&inode->i_data.private_list));
430 BUG_ON(!(inode->i_state & I_FREEING));
431 BUG_ON(inode->i_state & I_CLEAR);
432 inode_sync_wait(inode);
433 inode->i_state = I_FREEING | I_CLEAR;
434}
435EXPORT_SYMBOL(end_writeback);
436
Al Viro644da592010-06-07 13:21:05 -0400437static void evict(struct inode *inode)
Al Virob4272d42010-06-04 19:33:20 -0400438{
439 const struct super_operations *op = inode->i_sb->s_op;
440
Al Virobe7ce412010-06-04 19:40:39 -0400441 if (op->evict_inode) {
442 op->evict_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400443 } else {
444 if (inode->i_data.nrpages)
445 truncate_inode_pages(&inode->i_data, 0);
Al Viro30140832010-06-07 13:23:20 -0400446 end_writeback(inode);
Al Virob4272d42010-06-04 19:33:20 -0400447 }
Al Viro661074e2010-06-04 20:19:55 -0400448 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
449 bd_forget(inode);
450 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
451 cd_forget(inode);
Al Virob4272d42010-06-04 19:33:20 -0400452}
453
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454/*
455 * dispose_list - dispose of the contents of a local list
456 * @head: the head of the list to free
457 *
458 * Dispose-list gets a local list with local inodes in it, so it doesn't
459 * need to worry about list corruption and SMP locks.
460 */
461static void dispose_list(struct list_head *head)
462{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 while (!list_empty(head)) {
464 struct inode *inode;
465
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100466 inode = list_first_entry(head, struct inode, i_lru);
467 list_del_init(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
Al Viro644da592010-06-07 13:21:05 -0400469 evict(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700470
471 spin_lock(&inode_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400472 __remove_inode_hash(inode);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400473 __inode_sb_list_del(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700474 spin_unlock(&inode_lock);
475
476 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479}
480
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481/**
Al Viro63997e92010-10-25 20:49:35 -0400482 * evict_inodes - evict all evictable inodes for a superblock
483 * @sb: superblock to operate on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 *
Al Viro63997e92010-10-25 20:49:35 -0400485 * Make sure that no inodes with zero refcount are retained. This is
486 * called by superblock shutdown after having MS_ACTIVE flag removed,
487 * so any inode reaching zero refcount during or after that call will
488 * be immediately evicted.
489 */
490void evict_inodes(struct super_block *sb)
491{
492 struct inode *inode, *next;
493 LIST_HEAD(dispose);
494
495 down_write(&iprune_sem);
496
497 spin_lock(&inode_lock);
498 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
499 if (atomic_read(&inode->i_count))
500 continue;
501
502 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
503 WARN_ON(1);
504 continue;
505 }
506
507 inode->i_state |= I_FREEING;
508
509 /*
510 * Move the inode off the IO lists and LRU once I_FREEING is
511 * set so that it won't get moved back on there if it is dirty.
512 */
513 list_move(&inode->i_lru, &dispose);
514 list_del_init(&inode->i_wb_list);
515 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
516 percpu_counter_dec(&nr_inodes_unused);
517 }
518 spin_unlock(&inode_lock);
519
520 dispose_list(&dispose);
521 up_write(&iprune_sem);
522}
523
524/**
Christoph Hellwiga0318782010-10-24 19:40:33 +0200525 * invalidate_inodes - attempt to free all inodes on a superblock
526 * @sb: superblock to operate on
527 *
528 * Attempts to free all inodes for a given superblock. If there were any
529 * busy inodes return a non-zero value, else zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530531int invalidate_inodes(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532{
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400533 int busy = 0;
Christoph Hellwiga0318782010-10-24 19:40:33 +0200534 struct inode *inode, *next;
535 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700537 down_write(&iprune_sem);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200538
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 spin_lock(&inode_lock);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200540 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
Al Viro63997e92010-10-25 20:49:35 -0400541 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 continue;
Christoph Hellwig99a38912010-10-23 19:07:20 +0200543 if (atomic_read(&inode->i_count)) {
544 busy = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 continue;
546 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200547
Christoph Hellwig99a38912010-10-23 19:07:20 +0200548 inode->i_state |= I_FREEING;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100549
550 /*
551 * Move the inode off the IO lists and LRU once I_FREEING is
552 * set so that it won't get moved back on there if it is dirty.
553 */
Christoph Hellwiga0318782010-10-24 19:40:33 +0200554 list_move(&inode->i_lru, &dispose);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100555 list_del_init(&inode->i_wb_list);
Christoph Hellwig99a38912010-10-23 19:07:20 +0200556 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
557 percpu_counter_dec(&nr_inodes_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 spin_unlock(&inode_lock);
560
Christoph Hellwiga0318782010-10-24 19:40:33 +0200561 dispose_list(&dispose);
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700562 up_write(&iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
564 return busy;
565}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
567static int can_unuse(struct inode *inode)
568{
Nick Piggin9e38d862010-10-23 06:55:17 -0400569 if (inode->i_state & ~I_REFERENCED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 return 0;
571 if (inode_has_buffers(inode))
572 return 0;
573 if (atomic_read(&inode->i_count))
574 return 0;
575 if (inode->i_data.nrpages)
576 return 0;
577 return 1;
578}
579
580/*
Nick Piggin9e38d862010-10-23 06:55:17 -0400581 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
582 * temporary list and then are freed outside inode_lock by dispose_list().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 *
584 * Any inodes which are pinned purely because of attached pagecache have their
Nick Piggin9e38d862010-10-23 06:55:17 -0400585 * pagecache removed. If the inode has metadata buffers attached to
586 * mapping->private_list then try to remove them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 *
Nick Piggin9e38d862010-10-23 06:55:17 -0400588 * If the inode has the I_REFERENCED flag set, then it means that it has been
589 * used recently - the flag is set in iput_final(). When we encounter such an
590 * inode, clear the flag and move it to the back of the LRU so it gets another
591 * pass through the LRU before it gets reclaimed. This is necessary because of
592 * the fact we are doing lazy LRU updates to minimise lock contention so the
593 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
594 * with this flag set because they are the inodes that are out of order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 */
596static void prune_icache(int nr_to_scan)
597{
598 LIST_HEAD(freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 int nr_scanned;
600 unsigned long reap = 0;
601
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700602 down_read(&iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 spin_lock(&inode_lock);
604 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
605 struct inode *inode;
606
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100607 if (list_empty(&inode_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 break;
609
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100610 inode = list_entry(inode_lru.prev, struct inode, i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
Nick Piggin9e38d862010-10-23 06:55:17 -0400612 /*
613 * Referenced or dirty inodes are still in use. Give them
614 * another pass through the LRU as we canot reclaim them now.
615 */
616 if (atomic_read(&inode->i_count) ||
617 (inode->i_state & ~I_REFERENCED)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100618 list_del_init(&inode->i_lru);
Nick Piggin9e38d862010-10-23 06:55:17 -0400619 percpu_counter_dec(&nr_inodes_unused);
620 continue;
621 }
622
623 /* recently referenced inodes get one more pass */
624 if (inode->i_state & I_REFERENCED) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100625 list_move(&inode->i_lru, &inode_lru);
Nick Piggin9e38d862010-10-23 06:55:17 -0400626 inode->i_state &= ~I_REFERENCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 continue;
628 }
629 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
630 __iget(inode);
631 spin_unlock(&inode_lock);
632 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800633 reap += invalidate_mapping_pages(&inode->i_data,
634 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 iput(inode);
636 spin_lock(&inode_lock);
637
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100638 if (inode != list_entry(inode_lru.next,
639 struct inode, i_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 continue; /* wrong inode or list_empty */
641 if (!can_unuse(inode))
642 continue;
643 }
Nick Piggin7ef0d732009-03-12 14:31:38 -0700644 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 inode->i_state |= I_FREEING;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100646
647 /*
648 * Move the inode off the IO lists and LRU once I_FREEING is
649 * set so that it won't get moved back on there if it is dirty.
650 */
651 list_move(&inode->i_lru, &freeable);
652 list_del_init(&inode->i_wb_list);
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400653 percpu_counter_dec(&nr_inodes_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 }
Christoph Lameterf8891e52006-06-30 01:55:45 -0700655 if (current_is_kswapd())
656 __count_vm_events(KSWAPD_INODESTEAL, reap);
657 else
658 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 spin_unlock(&inode_lock);
660
661 dispose_list(&freeable);
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700662 up_read(&iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
664
665/*
666 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
667 * "unused" means that no dentries are referring to the inodes: the files are
668 * not open and the dcache references to those inodes have already been
669 * reclaimed.
670 *
671 * This function is passed the number of inodes to scan, and it returns the
672 * total number of remaining possibly-reclaimable inodes.
673 */
Dave Chinner7f8275d2010-07-19 14:56:17 +1000674static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675{
676 if (nr) {
677 /*
678 * Nasty deadlock avoidance. We may hold various FS locks,
679 * and we don't want to recurse into the FS that called us
680 * in clear_inode() and friends..
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530681 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 if (!(gfp_mask & __GFP_FS))
683 return -1;
684 prune_icache(nr);
685 }
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400686 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687}
688
Rusty Russell8e1f9362007-07-17 04:03:17 -0700689static struct shrinker icache_shrinker = {
690 .shrink = shrink_icache_memory,
691 .seeks = DEFAULT_SEEKS,
692};
693
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694static void __wait_on_freeing_inode(struct inode *inode);
695/*
696 * Called with the inode lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530698static struct inode *find_inode(struct super_block *sb,
699 struct hlist_head *head,
700 int (*test)(struct inode *, void *),
701 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
703 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530704 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
706repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700707 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 if (inode->i_sb != sb)
709 continue;
710 if (!test(inode, data))
711 continue;
Al Viroa4ffdde2010-06-02 17:38:30 -0400712 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 __wait_on_freeing_inode(inode);
714 goto repeat;
715 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400716 __iget(inode);
717 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400719 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720}
721
722/*
723 * find_inode_fast is the fast path version of find_inode, see the comment at
724 * iget_locked for details.
725 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530726static struct inode *find_inode_fast(struct super_block *sb,
727 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728{
729 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530730 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
732repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700733 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 if (inode->i_ino != ino)
735 continue;
736 if (inode->i_sb != sb)
737 continue;
Al Viroa4ffdde2010-06-02 17:38:30 -0400738 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 __wait_on_freeing_inode(inode);
740 goto repeat;
741 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400742 __iget(inode);
743 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400745 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746}
747
Eric Dumazetf991bd22010-10-23 11:18:01 -0400748/*
749 * Each cpu owns a range of LAST_INO_BATCH numbers.
750 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
751 * to renew the exhausted range.
David Chinner8290c352008-10-30 17:35:24 +1100752 *
Eric Dumazetf991bd22010-10-23 11:18:01 -0400753 * This does not significantly increase overflow rate because every CPU can
754 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
755 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
756 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
757 * overflow rate by 2x, which does not seem too significant.
758 *
759 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
760 * error if st_ino won't fit in target struct field. Use 32bit counter
761 * here to attempt to avoid that.
David Chinner8290c352008-10-30 17:35:24 +1100762 */
Eric Dumazetf991bd22010-10-23 11:18:01 -0400763#define LAST_INO_BATCH 1024
764static DEFINE_PER_CPU(unsigned int, last_ino);
David Chinner8290c352008-10-30 17:35:24 +1100765
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400766unsigned int get_next_ino(void)
Eric Dumazetf991bd22010-10-23 11:18:01 -0400767{
768 unsigned int *p = &get_cpu_var(last_ino);
769 unsigned int res = *p;
770
771#ifdef CONFIG_SMP
772 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
773 static atomic_t shared_last_ino;
774 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
775
776 res = next - LAST_INO_BATCH;
777 }
778#endif
779
780 *p = ++res;
781 put_cpu_var(last_ino);
782 return res;
David Chinner8290c352008-10-30 17:35:24 +1100783}
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400784EXPORT_SYMBOL(get_next_ino);
David Chinner8290c352008-10-30 17:35:24 +1100785
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786/**
787 * new_inode - obtain an inode
788 * @sb: superblock
789 *
Mel Gorman769848c2007-07-17 04:03:05 -0700790 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800791 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700792 * If HIGHMEM pages are unsuitable or it is known that pages allocated
793 * for the page cache are not reclaimable or migratable,
794 * mapping_set_gfp_mask() must be called with suitable flags on the
795 * newly created inode's mapping
796 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 */
798struct inode *new_inode(struct super_block *sb)
799{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530800 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
802 spin_lock_prefetch(&inode_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530803
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 inode = alloc_inode(sb);
805 if (inode) {
806 spin_lock(&inode_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400807 __inode_sb_list_add(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 inode->i_state = 0;
809 spin_unlock(&inode_lock);
810 }
811 return inode;
812}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813EXPORT_SYMBOL(new_inode);
814
815void unlock_new_inode(struct inode *inode)
816{
Peter Zijlstra14358e62007-10-14 01:38:33 +0200817#ifdef CONFIG_DEBUG_LOCK_ALLOC
Namhyung Kima3314a02010-10-11 22:38:00 +0900818 if (S_ISDIR(inode->i_mode)) {
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200819 struct file_system_type *type = inode->i_sb->s_type;
820
Jan Kara9a7aa122009-06-04 15:26:49 +0200821 /* Set new key only if filesystem hasn't already changed it */
822 if (!lockdep_match_class(&inode->i_mutex,
823 &type->i_mutex_key)) {
824 /*
825 * ensure nobody is actually holding i_mutex
826 */
827 mutex_destroy(&inode->i_mutex);
828 mutex_init(&inode->i_mutex);
829 lockdep_set_class(&inode->i_mutex,
830 &type->i_mutex_dir_key);
831 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200832 }
Peter Zijlstra14358e62007-10-14 01:38:33 +0200833#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 /*
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100835 * This is special! We do not need the spinlock when clearing I_NEW,
Jan Kara580be082009-09-21 17:01:06 -0700836 * because we're guaranteed that nobody else tries to do anything about
837 * the state of the inode when it is locked, as we just created it (so
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100838 * there can be no old holders that haven't tested I_NEW).
Jan Kara580be082009-09-21 17:01:06 -0700839 * However we must emit the memory barrier so that other CPUs reliably
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100840 * see the clearing of I_NEW after the other inode initialisation has
Jan Kara580be082009-09-21 17:01:06 -0700841 * completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 */
Jan Kara580be082009-09-21 17:01:06 -0700843 smp_mb();
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100844 WARN_ON(!(inode->i_state & I_NEW));
845 inode->i_state &= ~I_NEW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 wake_up_inode(inode);
847}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848EXPORT_SYMBOL(unlock_new_inode);
849
850/*
851 * This is called without the inode lock held.. Be careful.
852 *
853 * We no longer cache the sb_flags in i_flags - see fs.h
854 * -- rmk@arm.uk.linux.org
855 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530856static struct inode *get_new_inode(struct super_block *sb,
857 struct hlist_head *head,
858 int (*test)(struct inode *, void *),
859 int (*set)(struct inode *, void *),
860 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530862 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
864 inode = alloc_inode(sb);
865 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530866 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
868 spin_lock(&inode_lock);
869 /* We released the lock, so.. */
870 old = find_inode(sb, head, test, data);
871 if (!old) {
872 if (set(inode, data))
873 goto set_failed;
874
Christoph Hellwig646ec462010-10-23 07:15:32 -0400875 hlist_add_head(&inode->i_hash, head);
876 __inode_sb_list_add(inode);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100877 inode->i_state = I_NEW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 spin_unlock(&inode_lock);
879
880 /* Return the locked inode with I_NEW set, the
881 * caller is responsible for filling in the contents
882 */
883 return inode;
884 }
885
886 /*
887 * Uhhuh, somebody else created the same inode under
888 * us. Use the old inode instead of the one we just
889 * allocated.
890 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 spin_unlock(&inode_lock);
892 destroy_inode(inode);
893 inode = old;
894 wait_on_inode(inode);
895 }
896 return inode;
897
898set_failed:
899 spin_unlock(&inode_lock);
900 destroy_inode(inode);
901 return NULL;
902}
903
904/*
905 * get_new_inode_fast is the fast path version of get_new_inode, see the
906 * comment at iget_locked for details.
907 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530908static struct inode *get_new_inode_fast(struct super_block *sb,
909 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530911 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
913 inode = alloc_inode(sb);
914 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530915 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
917 spin_lock(&inode_lock);
918 /* We released the lock, so.. */
919 old = find_inode_fast(sb, head, ino);
920 if (!old) {
921 inode->i_ino = ino;
Christoph Hellwig646ec462010-10-23 07:15:32 -0400922 hlist_add_head(&inode->i_hash, head);
923 __inode_sb_list_add(inode);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100924 inode->i_state = I_NEW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 spin_unlock(&inode_lock);
926
927 /* Return the locked inode with I_NEW set, the
928 * caller is responsible for filling in the contents
929 */
930 return inode;
931 }
932
933 /*
934 * Uhhuh, somebody else created the same inode under
935 * us. Use the old inode instead of the one we just
936 * allocated.
937 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 spin_unlock(&inode_lock);
939 destroy_inode(inode);
940 inode = old;
941 wait_on_inode(inode);
942 }
943 return inode;
944}
945
Christoph Hellwigad5e1952010-10-23 07:00:16 -0400946/*
947 * search the inode cache for a matching inode number.
948 * If we find one, then the inode number we are trying to
949 * allocate is not unique and so we should not use it.
950 *
951 * Returns 1 if the inode number is unique, 0 if it is not.
952 */
953static int test_inode_iunique(struct super_block *sb, unsigned long ino)
954{
955 struct hlist_head *b = inode_hashtable + hash(sb, ino);
956 struct hlist_node *node;
957 struct inode *inode;
958
959 hlist_for_each_entry(inode, node, b, i_hash) {
960 if (inode->i_ino == ino && inode->i_sb == sb)
961 return 0;
962 }
963
964 return 1;
965}
966
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967/**
968 * iunique - get a unique inode number
969 * @sb: superblock
970 * @max_reserved: highest reserved inode number
971 *
972 * Obtain an inode number that is unique on the system for a given
973 * superblock. This is used by file systems that have no natural
974 * permanent inode numbering system. An inode number is returned that
975 * is higher than the reserved limit but unique.
976 *
977 * BUGS:
978 * With a large number of inodes live on the file system this function
979 * currently becomes quite slow.
980 */
981ino_t iunique(struct super_block *sb, ino_t max_reserved)
982{
Jeff Layton866b04f2007-05-08 00:32:29 -0700983 /*
984 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
985 * error if st_ino won't fit in target struct field. Use 32bit counter
986 * here to attempt to avoid that.
987 */
Christoph Hellwigad5e1952010-10-23 07:00:16 -0400988 static DEFINE_SPINLOCK(iunique_lock);
Jeff Layton866b04f2007-05-08 00:32:29 -0700989 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700992 spin_lock(&inode_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -0400993 spin_lock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700994 do {
995 if (counter <= max_reserved)
996 counter = max_reserved + 1;
997 res = counter++;
Christoph Hellwigad5e1952010-10-23 07:00:16 -0400998 } while (!test_inode_iunique(sb, res));
999 spin_unlock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001000 spin_unlock(&inode_lock);
1001
1002 return res;
1003}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004EXPORT_SYMBOL(iunique);
1005
1006struct inode *igrab(struct inode *inode)
1007{
1008 spin_lock(&inode_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -04001009 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 __iget(inode);
1011 else
1012 /*
1013 * Handle the case where s_op->clear_inode is not been
1014 * called yet, and somebody is calling igrab
1015 * while the inode is getting freed.
1016 */
1017 inode = NULL;
1018 spin_unlock(&inode_lock);
1019 return inode;
1020}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021EXPORT_SYMBOL(igrab);
1022
1023/**
1024 * ifind - internal function, you want ilookup5() or iget5().
1025 * @sb: super block of file system to search
1026 * @head: the head of the list to search
1027 * @test: callback used for comparisons between inodes
1028 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001029 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 *
1031 * ifind() searches for the inode specified by @data in the inode
1032 * cache. This is a generalized version of ifind_fast() for file systems where
1033 * the inode number is not sufficient for unique identification of an inode.
1034 *
1035 * If the inode is in the cache, the inode is returned with an incremented
1036 * reference count.
1037 *
1038 * Otherwise NULL is returned.
1039 *
1040 * Note, @test is called with the inode_lock held, so can't sleep.
1041 */
Matt Mackall5d2bea42006-01-08 01:05:21 -08001042static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001044 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045{
1046 struct inode *inode;
1047
1048 spin_lock(&inode_lock);
1049 inode = find_inode(sb, head, test, data);
1050 if (inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001052 if (likely(wait))
1053 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 return inode;
1055 }
1056 spin_unlock(&inode_lock);
1057 return NULL;
1058}
1059
1060/**
1061 * ifind_fast - internal function, you want ilookup() or iget().
1062 * @sb: super block of file system to search
1063 * @head: head of the list to search
1064 * @ino: inode number to search for
1065 *
1066 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1067 * file systems where the inode number is sufficient for unique identification
1068 * of an inode.
1069 *
1070 * If the inode is in the cache, the inode is returned with an incremented
1071 * reference count.
1072 *
1073 * Otherwise NULL is returned.
1074 */
Matt Mackall5d2bea42006-01-08 01:05:21 -08001075static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 struct hlist_head *head, unsigned long ino)
1077{
1078 struct inode *inode;
1079
1080 spin_lock(&inode_lock);
1081 inode = find_inode_fast(sb, head, ino);
1082 if (inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 spin_unlock(&inode_lock);
1084 wait_on_inode(inode);
1085 return inode;
1086 }
1087 spin_unlock(&inode_lock);
1088 return NULL;
1089}
1090
1091/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001092 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 * @sb: super block of file system to search
1094 * @hashval: hash value (usually inode number) to search for
1095 * @test: callback used for comparisons between inodes
1096 * @data: opaque data pointer to pass to @test
1097 *
1098 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1099 * @data in the inode cache. This is a generalized version of ilookup() for
1100 * file systems where the inode number is not sufficient for unique
1101 * identification of an inode.
1102 *
1103 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001104 * reference count. Note, the inode lock is not waited upon so you have to be
1105 * very careful what you do with the returned inode. You probably should be
1106 * using ilookup5() instead.
1107 *
1108 * Otherwise NULL is returned.
1109 *
1110 * Note, @test is called with the inode_lock held, so can't sleep.
1111 */
1112struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1113 int (*test)(struct inode *, void *), void *data)
1114{
1115 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1116
1117 return ifind(sb, head, test, data, 0);
1118}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001119EXPORT_SYMBOL(ilookup5_nowait);
1120
1121/**
1122 * ilookup5 - search for an inode in the inode cache
1123 * @sb: super block of file system to search
1124 * @hashval: hash value (usually inode number) to search for
1125 * @test: callback used for comparisons between inodes
1126 * @data: opaque data pointer to pass to @test
1127 *
1128 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1129 * @data in the inode cache. This is a generalized version of ilookup() for
1130 * file systems where the inode number is not sufficient for unique
1131 * identification of an inode.
1132 *
1133 * If the inode is in the cache, the inode lock is waited upon and the inode is
1134 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 *
1136 * Otherwise NULL is returned.
1137 *
1138 * Note, @test is called with the inode_lock held, so can't sleep.
1139 */
1140struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1141 int (*test)(struct inode *, void *), void *data)
1142{
1143 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1144
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001145 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147EXPORT_SYMBOL(ilookup5);
1148
1149/**
1150 * ilookup - search for an inode in the inode cache
1151 * @sb: super block of file system to search
1152 * @ino: inode number to search for
1153 *
1154 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1155 * This is for file systems where the inode number is sufficient for unique
1156 * identification of an inode.
1157 *
1158 * If the inode is in the cache, the inode is returned with an incremented
1159 * reference count.
1160 *
1161 * Otherwise NULL is returned.
1162 */
1163struct inode *ilookup(struct super_block *sb, unsigned long ino)
1164{
1165 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1166
1167 return ifind_fast(sb, head, ino);
1168}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169EXPORT_SYMBOL(ilookup);
1170
1171/**
1172 * iget5_locked - obtain an inode from a mounted file system
1173 * @sb: super block of file system
1174 * @hashval: hash value (usually inode number) to get
1175 * @test: callback used for comparisons between inodes
1176 * @set: callback used to initialize a new struct inode
1177 * @data: opaque data pointer to pass to @test and @set
1178 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1180 * and @data in the inode cache and if present it is returned with an increased
1181 * reference count. This is a generalized version of iget_locked() for file
1182 * systems where the inode number is not sufficient for unique identification
1183 * of an inode.
1184 *
1185 * If the inode is not in cache, get_new_inode() is called to allocate a new
1186 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1187 * file system gets to fill it in before unlocking it via unlock_new_inode().
1188 *
1189 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1190 */
1191struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1192 int (*test)(struct inode *, void *),
1193 int (*set)(struct inode *, void *), void *data)
1194{
1195 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1196 struct inode *inode;
1197
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001198 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 if (inode)
1200 return inode;
1201 /*
1202 * get_new_inode() will do the right thing, re-trying the search
1203 * in case it had to block at any point.
1204 */
1205 return get_new_inode(sb, head, test, set, data);
1206}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207EXPORT_SYMBOL(iget5_locked);
1208
1209/**
1210 * iget_locked - obtain an inode from a mounted file system
1211 * @sb: super block of file system
1212 * @ino: inode number to get
1213 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1215 * the inode cache and if present it is returned with an increased reference
1216 * count. This is for file systems where the inode number is sufficient for
1217 * unique identification of an inode.
1218 *
1219 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1220 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1221 * The file system gets to fill it in before unlocking it via
1222 * unlock_new_inode().
1223 */
1224struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1225{
1226 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1227 struct inode *inode;
1228
1229 inode = ifind_fast(sb, head, ino);
1230 if (inode)
1231 return inode;
1232 /*
1233 * get_new_inode_fast() will do the right thing, re-trying the search
1234 * in case it had to block at any point.
1235 */
1236 return get_new_inode_fast(sb, head, ino);
1237}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238EXPORT_SYMBOL(iget_locked);
1239
Al Viro261bca82008-12-30 01:48:21 -05001240int insert_inode_locked(struct inode *inode)
1241{
1242 struct super_block *sb = inode->i_sb;
1243 ino_t ino = inode->i_ino;
1244 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001245
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001246 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001247 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001248 struct hlist_node *node;
1249 struct inode *old = NULL;
Al Viro261bca82008-12-30 01:48:21 -05001250 spin_lock(&inode_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001251 hlist_for_each_entry(old, node, head, i_hash) {
1252 if (old->i_ino != ino)
1253 continue;
1254 if (old->i_sb != sb)
1255 continue;
Al Viroa4ffdde2010-06-02 17:38:30 -04001256 if (old->i_state & (I_FREEING|I_WILL_FREE))
Al Viro72a43d62009-05-13 19:13:40 +01001257 continue;
1258 break;
1259 }
1260 if (likely(!node)) {
Al Viro261bca82008-12-30 01:48:21 -05001261 hlist_add_head(&inode->i_hash, head);
1262 spin_unlock(&inode_lock);
1263 return 0;
1264 }
1265 __iget(old);
1266 spin_unlock(&inode_lock);
1267 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001268 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001269 iput(old);
1270 return -EBUSY;
1271 }
1272 iput(old);
1273 }
1274}
Al Viro261bca82008-12-30 01:48:21 -05001275EXPORT_SYMBOL(insert_inode_locked);
1276
1277int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1278 int (*test)(struct inode *, void *), void *data)
1279{
1280 struct super_block *sb = inode->i_sb;
1281 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001282
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001283 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001284
1285 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001286 struct hlist_node *node;
1287 struct inode *old = NULL;
1288
Al Viro261bca82008-12-30 01:48:21 -05001289 spin_lock(&inode_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001290 hlist_for_each_entry(old, node, head, i_hash) {
1291 if (old->i_sb != sb)
1292 continue;
1293 if (!test(old, data))
1294 continue;
Al Viroa4ffdde2010-06-02 17:38:30 -04001295 if (old->i_state & (I_FREEING|I_WILL_FREE))
Al Viro72a43d62009-05-13 19:13:40 +01001296 continue;
1297 break;
1298 }
1299 if (likely(!node)) {
Al Viro261bca82008-12-30 01:48:21 -05001300 hlist_add_head(&inode->i_hash, head);
1301 spin_unlock(&inode_lock);
1302 return 0;
1303 }
1304 __iget(old);
1305 spin_unlock(&inode_lock);
1306 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001307 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001308 iput(old);
1309 return -EBUSY;
1310 }
1311 iput(old);
1312 }
1313}
Al Viro261bca82008-12-30 01:48:21 -05001314EXPORT_SYMBOL(insert_inode_locked4);
1315
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
Al Viro45321ac2010-06-07 13:43:19 -04001317int generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318{
Al Viro45321ac2010-06-07 13:43:19 -04001319 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321EXPORT_SYMBOL(generic_delete_inode);
1322
Al Viro45321ac2010-06-07 13:43:19 -04001323/*
1324 * Normal UNIX filesystem behaviour: delete the
1325 * inode when the usage count drops to zero, and
1326 * i_nlink is zero.
Jan Kara22fe40422009-09-18 13:05:44 -07001327 */
Al Viro45321ac2010-06-07 13:43:19 -04001328int generic_drop_inode(struct inode *inode)
1329{
Al Viro1d3382cb2010-10-23 15:19:20 -04001330 return !inode->i_nlink || inode_unhashed(inode);
Al Viro45321ac2010-06-07 13:43:19 -04001331}
1332EXPORT_SYMBOL_GPL(generic_drop_inode);
1333
1334/*
1335 * Called when we're dropping the last reference
1336 * to an inode.
1337 *
1338 * Call the FS "drop_inode()" function, defaulting to
1339 * the legacy UNIX filesystem behaviour. If it tells
1340 * us to evict inode, do so. Otherwise, retain inode
1341 * in cache if fs is alive, sync and evict if fs is
1342 * shutting down.
1343 */
1344static void iput_final(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345{
1346 struct super_block *sb = inode->i_sb;
Al Viro45321ac2010-06-07 13:43:19 -04001347 const struct super_operations *op = inode->i_sb->s_op;
1348 int drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
Al Viro45321ac2010-06-07 13:43:19 -04001350 if (op && op->drop_inode)
1351 drop = op->drop_inode(inode);
1352 else
1353 drop = generic_drop_inode(inode);
1354
1355 if (!drop) {
Christoph Hellwigacb0c852007-05-08 00:25:52 -07001356 if (sb->s_flags & MS_ACTIVE) {
Nick Piggin9e38d862010-10-23 06:55:17 -04001357 inode->i_state |= I_REFERENCED;
1358 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1359 inode_lru_list_add(inode);
1360 }
Alexander Viro991114c2005-06-23 00:09:01 -07001361 spin_unlock(&inode_lock);
Al Viro45321ac2010-06-07 13:43:19 -04001362 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001363 }
Nick Piggin7ef0d732009-03-12 14:31:38 -07001364 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001365 inode->i_state |= I_WILL_FREE;
1366 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 write_inode_now(inode, 1);
1368 spin_lock(&inode_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001369 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001370 inode->i_state &= ~I_WILL_FREE;
Dave Chinner4c51acb2010-10-23 06:58:09 -04001371 __remove_inode_hash(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001373
Nick Piggin7ef0d732009-03-12 14:31:38 -07001374 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001375 inode->i_state |= I_FREEING;
Nick Piggin9e38d862010-10-23 06:55:17 -04001376
1377 /*
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001378 * Move the inode off the IO lists and LRU once I_FREEING is
1379 * set so that it won't get moved back on there if it is dirty.
Nick Piggin9e38d862010-10-23 06:55:17 -04001380 */
1381 inode_lru_list_del(inode);
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001382 list_del_init(&inode->i_wb_list);
Nick Piggin9e38d862010-10-23 06:55:17 -04001383
Christoph Hellwig646ec462010-10-23 07:15:32 -04001384 __inode_sb_list_del(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 spin_unlock(&inode_lock);
Al Viro644da592010-06-07 13:21:05 -04001386 evict(inode);
Dave Chinner4c51acb2010-10-23 06:58:09 -04001387 remove_inode_hash(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001388 wake_up_inode(inode);
Al Viro45321ac2010-06-07 13:43:19 -04001389 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 destroy_inode(inode);
1391}
1392
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301394 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 * @inode: inode to put
1396 *
1397 * Puts an inode, dropping its usage count. If the inode use count hits
1398 * zero, the inode is then freed and may also be destroyed.
1399 *
1400 * Consequently, iput() can sleep.
1401 */
1402void iput(struct inode *inode)
1403{
1404 if (inode) {
Al Viroa4ffdde2010-06-02 17:38:30 -04001405 BUG_ON(inode->i_state & I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1408 iput_final(inode);
1409 }
1410}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411EXPORT_SYMBOL(iput);
1412
1413/**
1414 * bmap - find a block number in a file
1415 * @inode: inode of file
1416 * @block: block to find
1417 *
1418 * Returns the block number on the device holding the inode that
1419 * is the disk block number for the block of the file requested.
1420 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301421 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 * file.
1423 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301424sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425{
1426 sector_t res = 0;
1427 if (inode->i_mapping->a_ops->bmap)
1428 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1429 return res;
1430}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431EXPORT_SYMBOL(bmap);
1432
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001433/*
1434 * With relative atime, only update atime if the previous atime is
1435 * earlier than either the ctime or mtime or if at least a day has
1436 * passed since the last atime update.
1437 */
1438static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1439 struct timespec now)
1440{
1441
1442 if (!(mnt->mnt_flags & MNT_RELATIME))
1443 return 1;
1444 /*
1445 * Is mtime younger than atime? If yes, update atime:
1446 */
1447 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1448 return 1;
1449 /*
1450 * Is ctime younger than atime? If yes, update atime:
1451 */
1452 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1453 return 1;
1454
1455 /*
1456 * Is the previous atime value older than a day? If yes,
1457 * update atime:
1458 */
1459 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1460 return 1;
1461 /*
1462 * Good, we can skip the atime update:
1463 */
1464 return 0;
1465}
1466
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001468 * touch_atime - update the access time
1469 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001470 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 *
1472 * Update the accessed time on an inode and mark it for writeback.
1473 * This function automatically handles read only file systems and media,
1474 * as well as the "noatime" flag and inode specific "noatime" markers.
1475 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001476void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001478 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 struct timespec now;
1480
Andrew Mortonb2276132006-12-13 00:34:33 -08001481 if (inode->i_flags & S_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001482 return;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001483 if (IS_NOATIME(inode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001484 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001485 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001486 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001487
Dave Hansencdb70f32008-02-15 14:37:41 -08001488 if (mnt->mnt_flags & MNT_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001489 return;
Dave Hansencdb70f32008-02-15 14:37:41 -08001490 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001491 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 now = current_fs_time(inode->i_sb);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001494
1495 if (!relatime_need_update(mnt, inode, now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001496 return;
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001497
Valerie Henson47ae32d2006-12-13 00:34:34 -08001498 if (timespec_equal(&inode->i_atime, &now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001499 return;
1500
1501 if (mnt_want_write(mnt))
1502 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001503
1504 inode->i_atime = now;
1505 mark_inode_dirty_sync(inode);
Dave Hansencdb70f32008-02-15 14:37:41 -08001506 mnt_drop_write(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001508EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
1510/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001511 * file_update_time - update mtime and ctime time
1512 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001514 * Update the mtime and ctime members of an inode and mark the inode
1515 * for writeback. Note that this function is meant exclusively for
1516 * usage in the file write path of filesystems, and filesystems may
1517 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001518 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Christoph Hellwig870f4812006-01-09 20:52:01 -08001519 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 */
1521
Christoph Hellwig870f4812006-01-09 20:52:01 -08001522void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001524 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 struct timespec now;
Andi Kleence06e0b2009-09-18 13:05:48 -07001526 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
Andi Kleence06e0b2009-09-18 13:05:48 -07001528 /* First try to exhaust all avenues to not sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 if (IS_NOCMTIME(inode))
1530 return;
Dave Hansen20ddee22008-02-15 14:37:43 -08001531
Andi Kleence06e0b2009-09-18 13:05:48 -07001532 now = current_fs_time(inode->i_sb);
1533 if (!timespec_equal(&inode->i_mtime, &now))
1534 sync_it = S_MTIME;
1535
1536 if (!timespec_equal(&inode->i_ctime, &now))
1537 sync_it |= S_CTIME;
1538
1539 if (IS_I_VERSION(inode))
1540 sync_it |= S_VERSION;
1541
1542 if (!sync_it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 return;
1544
Andi Kleence06e0b2009-09-18 13:05:48 -07001545 /* Finally allowed to write? Takes lock. */
1546 if (mnt_want_write_file(file))
1547 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
Andi Kleence06e0b2009-09-18 13:05:48 -07001549 /* Only change inode inside the lock region */
1550 if (sync_it & S_VERSION)
Jean Noel Cordenner7a224222008-01-28 23:58:27 -05001551 inode_inc_iversion(inode);
Andi Kleence06e0b2009-09-18 13:05:48 -07001552 if (sync_it & S_CTIME)
1553 inode->i_ctime = now;
1554 if (sync_it & S_MTIME)
1555 inode->i_mtime = now;
1556 mark_inode_dirty_sync(inode);
Dave Hansen20ddee22008-02-15 14:37:43 -08001557 mnt_drop_write(file->f_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001559EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560
1561int inode_needs_sync(struct inode *inode)
1562{
1563 if (IS_SYNC(inode))
1564 return 1;
1565 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1566 return 1;
1567 return 0;
1568}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569EXPORT_SYMBOL(inode_needs_sync);
1570
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571int inode_wait(void *word)
1572{
1573 schedule();
1574 return 0;
1575}
Stephen Rothwelld44dab82008-11-10 17:06:05 +11001576EXPORT_SYMBOL(inode_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
1578/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001579 * If we try to find an inode in the inode hash while it is being
1580 * deleted, we have to wait until the filesystem completes its
1581 * deletion before reporting that it isn't found. This function waits
1582 * until the deletion _might_ have completed. Callers are responsible
1583 * to recheck inode state.
1584 *
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001585 * It doesn't matter if I_NEW is not set initially, a call to
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001586 * wake_up_inode() after removing from the hash list will DTRT.
1587 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 * This is called with inode_lock held.
1589 */
1590static void __wait_on_freeing_inode(struct inode *inode)
1591{
1592 wait_queue_head_t *wq;
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001593 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1594 wq = bit_waitqueue(&inode->i_state, __I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1596 spin_unlock(&inode_lock);
1597 schedule();
1598 finish_wait(wq, &wait.wait);
1599 spin_lock(&inode_lock);
1600}
1601
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602static __initdata unsigned long ihash_entries;
1603static int __init set_ihash_entries(char *str)
1604{
1605 if (!str)
1606 return 0;
1607 ihash_entries = simple_strtoul(str, &str, 0);
1608 return 1;
1609}
1610__setup("ihash_entries=", set_ihash_entries);
1611
1612/*
1613 * Initialize the waitqueues and inode hash table.
1614 */
1615void __init inode_init_early(void)
1616{
1617 int loop;
1618
1619 /* If hashes are distributed across NUMA nodes, defer
1620 * hash allocation until vmalloc space is available.
1621 */
1622 if (hashdist)
1623 return;
1624
1625 inode_hashtable =
1626 alloc_large_system_hash("Inode-cache",
1627 sizeof(struct hlist_head),
1628 ihash_entries,
1629 14,
1630 HASH_EARLY,
1631 &i_hash_shift,
1632 &i_hash_mask,
1633 0);
1634
1635 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1636 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1637}
1638
Denis Cheng74bf17c2007-10-16 23:26:30 -07001639void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640{
1641 int loop;
1642
1643 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001644 inode_cachep = kmem_cache_create("inode_cache",
1645 sizeof(struct inode),
1646 0,
1647 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1648 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001649 init_once);
Rusty Russell8e1f9362007-07-17 04:03:17 -07001650 register_shrinker(&icache_shrinker);
Dave Chinnercffbc8a2010-10-23 05:03:02 -04001651 percpu_counter_init(&nr_inodes, 0);
1652 percpu_counter_init(&nr_inodes_unused, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
1654 /* Hash may have been set up in inode_init_early */
1655 if (!hashdist)
1656 return;
1657
1658 inode_hashtable =
1659 alloc_large_system_hash("Inode-cache",
1660 sizeof(struct hlist_head),
1661 ihash_entries,
1662 14,
1663 0,
1664 &i_hash_shift,
1665 &i_hash_mask,
1666 0);
1667
1668 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1669 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1670}
1671
1672void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1673{
1674 inode->i_mode = mode;
1675 if (S_ISCHR(mode)) {
1676 inode->i_fop = &def_chr_fops;
1677 inode->i_rdev = rdev;
1678 } else if (S_ISBLK(mode)) {
1679 inode->i_fop = &def_blk_fops;
1680 inode->i_rdev = rdev;
1681 } else if (S_ISFIFO(mode))
1682 inode->i_fop = &def_fifo_fops;
1683 else if (S_ISSOCK(mode))
1684 inode->i_fop = &bad_sock_fops;
1685 else
Manish Katiyaraf0d9ae2009-09-18 13:05:43 -07001686 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1687 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1688 inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689}
1690EXPORT_SYMBOL(init_special_inode);
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001691
1692/**
1693 * Init uid,gid,mode for new inode according to posix standards
1694 * @inode: New inode
1695 * @dir: Directory inode
1696 * @mode: mode of the new inode
1697 */
1698void inode_init_owner(struct inode *inode, const struct inode *dir,
1699 mode_t mode)
1700{
1701 inode->i_uid = current_fsuid();
1702 if (dir && dir->i_mode & S_ISGID) {
1703 inode->i_gid = dir->i_gid;
1704 if (S_ISDIR(mode))
1705 mode |= S_ISGID;
1706 } else
1707 inode->i_gid = current_fsgid();
1708 inode->i_mode = mode;
1709}
1710EXPORT_SYMBOL(inode_init_owner);