blob: 0647d80accf6fb51274fe3e8489075e582ba5c16 [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 Pigginfa0d7e32011-01-07 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 Pigginfa0d7e32011-01-07 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
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100298void address_space_init_once(struct address_space *mapping)
299{
300 memset(mapping, 0, sizeof(*mapping));
301 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
302 spin_lock_init(&mapping->tree_lock);
303 spin_lock_init(&mapping->i_mmap_lock);
304 INIT_LIST_HEAD(&mapping->private_list);
305 spin_lock_init(&mapping->private_lock);
306 INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
307 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
308 mutex_init(&mapping->unmap_mutex);
309}
310EXPORT_SYMBOL(address_space_init_once);
311
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312/*
313 * These are initializations that only need to be done
314 * once, because the fields are idempotent across use
315 * of the inode, so let the slab aware of that.
316 */
317void inode_init_once(struct inode *inode)
318{
319 memset(inode, 0, sizeof(*inode));
320 INIT_HLIST_NODE(&inode->i_hash);
321 INIT_LIST_HEAD(&inode->i_dentry);
322 INIT_LIST_HEAD(&inode->i_devices);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100323 INIT_LIST_HEAD(&inode->i_wb_list);
324 INIT_LIST_HEAD(&inode->i_lru);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100325 address_space_init_once(&inode->i_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 i_size_ordered_init(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400327#ifdef CONFIG_FSNOTIFY
Eric Parise61ce862009-12-17 21:24:24 -0500328 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
Eric Paris3be25f42009-05-21 17:01:26 -0400329#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331EXPORT_SYMBOL(inode_init_once);
332
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700333static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530335 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336
Christoph Lametera35afb82007-05-16 22:10:57 -0700337 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338}
339
340/*
341 * inode_lock must be held
342 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530343void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344{
Nick Piggin9e38d862010-10-23 06:55:17 -0400345 atomic_inc(&inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346}
347
Al Viro7de9c6e2010-10-23 11:11:40 -0400348/*
349 * get additional reference to inode; caller must already hold one.
350 */
351void ihold(struct inode *inode)
352{
353 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
354}
355EXPORT_SYMBOL(ihold);
356
Nick Piggin9e38d862010-10-23 06:55:17 -0400357static void inode_lru_list_add(struct inode *inode)
358{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100359 if (list_empty(&inode->i_lru)) {
360 list_add(&inode->i_lru, &inode_lru);
Nick Piggin86c87492011-01-07 17:49:18 +1100361 inodes_stat.nr_unused++;
Nick Piggin9e38d862010-10-23 06:55:17 -0400362 }
363}
364
365static void inode_lru_list_del(struct inode *inode)
366{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100367 if (!list_empty(&inode->i_lru)) {
368 list_del_init(&inode->i_lru);
Nick Piggin86c87492011-01-07 17:49:18 +1100369 inodes_stat.nr_unused--;
Nick Piggin9e38d862010-10-23 06:55:17 -0400370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371}
372
Christoph Hellwig646ec462010-10-23 07:15:32 -0400373static inline void __inode_sb_list_add(struct inode *inode)
374{
375 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
376}
377
378/**
379 * inode_sb_list_add - add inode to the superblock list of inodes
380 * @inode: inode to add
381 */
382void inode_sb_list_add(struct inode *inode)
383{
384 spin_lock(&inode_lock);
385 __inode_sb_list_add(inode);
386 spin_unlock(&inode_lock);
387}
388EXPORT_SYMBOL_GPL(inode_sb_list_add);
389
390static inline void __inode_sb_list_del(struct inode *inode)
391{
392 list_del_init(&inode->i_sb_list);
393}
394
Dave Chinner4c51acb2010-10-23 06:58:09 -0400395static unsigned long hash(struct super_block *sb, unsigned long hashval)
396{
397 unsigned long tmp;
398
399 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
400 L1_CACHE_BYTES;
401 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
402 return tmp & I_HASHMASK;
403}
404
405/**
406 * __insert_inode_hash - hash an inode
407 * @inode: unhashed inode
408 * @hashval: unsigned long value used to locate this object in the
409 * inode_hashtable.
410 *
411 * Add an inode to the inode hash for this superblock.
412 */
413void __insert_inode_hash(struct inode *inode, unsigned long hashval)
414{
Christoph Hellwig646ec462010-10-23 07:15:32 -0400415 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
416
Dave Chinner4c51acb2010-10-23 06:58:09 -0400417 spin_lock(&inode_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400418 hlist_add_head(&inode->i_hash, b);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400419 spin_unlock(&inode_lock);
420}
421EXPORT_SYMBOL(__insert_inode_hash);
422
423/**
424 * __remove_inode_hash - remove an inode from the hash
425 * @inode: inode to unhash
426 *
427 * Remove an inode from the superblock.
428 */
429static void __remove_inode_hash(struct inode *inode)
430{
431 hlist_del_init(&inode->i_hash);
432}
433
434/**
435 * remove_inode_hash - remove an inode from the hash
436 * @inode: inode to unhash
437 *
438 * Remove an inode from the superblock.
439 */
440void remove_inode_hash(struct inode *inode)
441{
442 spin_lock(&inode_lock);
443 hlist_del_init(&inode->i_hash);
444 spin_unlock(&inode_lock);
445}
446EXPORT_SYMBOL(remove_inode_hash);
447
Al Virob0683aa2010-06-04 20:55:25 -0400448void end_writeback(struct inode *inode)
449{
450 might_sleep();
451 BUG_ON(inode->i_data.nrpages);
452 BUG_ON(!list_empty(&inode->i_data.private_list));
453 BUG_ON(!(inode->i_state & I_FREEING));
454 BUG_ON(inode->i_state & I_CLEAR);
455 inode_sync_wait(inode);
Nick Pigginfa0d7e32011-01-07 17:49:49 +1100456 /* don't need i_lock here, no concurrent mods to i_state */
Al Virob0683aa2010-06-04 20:55:25 -0400457 inode->i_state = I_FREEING | I_CLEAR;
458}
459EXPORT_SYMBOL(end_writeback);
460
Al Viro644da592010-06-07 13:21:05 -0400461static void evict(struct inode *inode)
Al Virob4272d42010-06-04 19:33:20 -0400462{
463 const struct super_operations *op = inode->i_sb->s_op;
464
Al Virobe7ce412010-06-04 19:40:39 -0400465 if (op->evict_inode) {
466 op->evict_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400467 } else {
468 if (inode->i_data.nrpages)
469 truncate_inode_pages(&inode->i_data, 0);
Al Viro30140832010-06-07 13:23:20 -0400470 end_writeback(inode);
Al Virob4272d42010-06-04 19:33:20 -0400471 }
Al Viro661074e2010-06-04 20:19:55 -0400472 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
473 bd_forget(inode);
474 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
475 cd_forget(inode);
Al Virob4272d42010-06-04 19:33:20 -0400476}
477
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478/*
479 * dispose_list - dispose of the contents of a local list
480 * @head: the head of the list to free
481 *
482 * Dispose-list gets a local list with local inodes in it, so it doesn't
483 * need to worry about list corruption and SMP locks.
484 */
485static void dispose_list(struct list_head *head)
486{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 while (!list_empty(head)) {
488 struct inode *inode;
489
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100490 inode = list_first_entry(head, struct inode, i_lru);
491 list_del_init(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
Al Viro644da592010-06-07 13:21:05 -0400493 evict(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700494
495 spin_lock(&inode_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400496 __remove_inode_hash(inode);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400497 __inode_sb_list_del(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700498 spin_unlock(&inode_lock);
499
500 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503}
504
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505/**
Al Viro63997e92010-10-25 20:49:35 -0400506 * evict_inodes - evict all evictable inodes for a superblock
507 * @sb: superblock to operate on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 *
Al Viro63997e92010-10-25 20:49:35 -0400509 * Make sure that no inodes with zero refcount are retained. This is
510 * called by superblock shutdown after having MS_ACTIVE flag removed,
511 * so any inode reaching zero refcount during or after that call will
512 * be immediately evicted.
513 */
514void evict_inodes(struct super_block *sb)
515{
516 struct inode *inode, *next;
517 LIST_HEAD(dispose);
518
519 down_write(&iprune_sem);
520
521 spin_lock(&inode_lock);
522 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
523 if (atomic_read(&inode->i_count))
524 continue;
525
526 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
527 WARN_ON(1);
528 continue;
529 }
530
531 inode->i_state |= I_FREEING;
532
533 /*
534 * Move the inode off the IO lists and LRU once I_FREEING is
535 * set so that it won't get moved back on there if it is dirty.
536 */
537 list_move(&inode->i_lru, &dispose);
538 list_del_init(&inode->i_wb_list);
539 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
Nick Piggin86c87492011-01-07 17:49:18 +1100540 inodes_stat.nr_unused--;
Al Viro63997e92010-10-25 20:49:35 -0400541 }
542 spin_unlock(&inode_lock);
543
544 dispose_list(&dispose);
545 up_write(&iprune_sem);
546}
547
548/**
Christoph Hellwiga0318782010-10-24 19:40:33 +0200549 * invalidate_inodes - attempt to free all inodes on a superblock
550 * @sb: superblock to operate on
NeilBrown93b270f2011-02-24 17:25:47 +1100551 * @kill_dirty: flag to guide handling of dirty inodes
Christoph Hellwiga0318782010-10-24 19:40:33 +0200552 *
553 * Attempts to free all inodes for a given superblock. If there were any
554 * busy inodes return a non-zero value, else zero.
NeilBrown93b270f2011-02-24 17:25:47 +1100555 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
556 * them as busy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 */
NeilBrown93b270f2011-02-24 17:25:47 +1100558int invalidate_inodes(struct super_block *sb, bool kill_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559{
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400560 int busy = 0;
Christoph Hellwiga0318782010-10-24 19:40:33 +0200561 struct inode *inode, *next;
562 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700564 down_write(&iprune_sem);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200565
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 spin_lock(&inode_lock);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200567 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
Al Viro63997e92010-10-25 20:49:35 -0400568 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 continue;
NeilBrown93b270f2011-02-24 17:25:47 +1100570 if (inode->i_state & I_DIRTY && !kill_dirty) {
571 busy = 1;
572 continue;
573 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200574 if (atomic_read(&inode->i_count)) {
575 busy = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 continue;
577 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200578
Christoph Hellwig99a38912010-10-23 19:07:20 +0200579 inode->i_state |= I_FREEING;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100580
581 /*
582 * Move the inode off the IO lists and LRU once I_FREEING is
583 * set so that it won't get moved back on there if it is dirty.
584 */
Christoph Hellwiga0318782010-10-24 19:40:33 +0200585 list_move(&inode->i_lru, &dispose);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100586 list_del_init(&inode->i_wb_list);
Christoph Hellwig99a38912010-10-23 19:07:20 +0200587 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
Nick Piggin86c87492011-01-07 17:49:18 +1100588 inodes_stat.nr_unused--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 spin_unlock(&inode_lock);
591
Christoph Hellwiga0318782010-10-24 19:40:33 +0200592 dispose_list(&dispose);
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700593 up_write(&iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
595 return busy;
596}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
598static int can_unuse(struct inode *inode)
599{
Nick Piggin9e38d862010-10-23 06:55:17 -0400600 if (inode->i_state & ~I_REFERENCED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 return 0;
602 if (inode_has_buffers(inode))
603 return 0;
604 if (atomic_read(&inode->i_count))
605 return 0;
606 if (inode->i_data.nrpages)
607 return 0;
608 return 1;
609}
610
611/*
Nick Piggin9e38d862010-10-23 06:55:17 -0400612 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
613 * temporary list and then are freed outside inode_lock by dispose_list().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 *
615 * Any inodes which are pinned purely because of attached pagecache have their
Nick Piggin9e38d862010-10-23 06:55:17 -0400616 * pagecache removed. If the inode has metadata buffers attached to
617 * mapping->private_list then try to remove them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 *
Nick Piggin9e38d862010-10-23 06:55:17 -0400619 * If the inode has the I_REFERENCED flag set, then it means that it has been
620 * used recently - the flag is set in iput_final(). When we encounter such an
621 * inode, clear the flag and move it to the back of the LRU so it gets another
622 * pass through the LRU before it gets reclaimed. This is necessary because of
623 * the fact we are doing lazy LRU updates to minimise lock contention so the
624 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
625 * with this flag set because they are the inodes that are out of order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 */
627static void prune_icache(int nr_to_scan)
628{
629 LIST_HEAD(freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 int nr_scanned;
631 unsigned long reap = 0;
632
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700633 down_read(&iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 spin_lock(&inode_lock);
635 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
636 struct inode *inode;
637
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100638 if (list_empty(&inode_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 break;
640
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100641 inode = list_entry(inode_lru.prev, struct inode, i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
Nick Piggin9e38d862010-10-23 06:55:17 -0400643 /*
644 * Referenced or dirty inodes are still in use. Give them
645 * another pass through the LRU as we canot reclaim them now.
646 */
647 if (atomic_read(&inode->i_count) ||
648 (inode->i_state & ~I_REFERENCED)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100649 list_del_init(&inode->i_lru);
Nick Piggin86c87492011-01-07 17:49:18 +1100650 inodes_stat.nr_unused--;
Nick Piggin9e38d862010-10-23 06:55:17 -0400651 continue;
652 }
653
654 /* recently referenced inodes get one more pass */
655 if (inode->i_state & I_REFERENCED) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100656 list_move(&inode->i_lru, &inode_lru);
Nick Piggin9e38d862010-10-23 06:55:17 -0400657 inode->i_state &= ~I_REFERENCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 continue;
659 }
660 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
661 __iget(inode);
662 spin_unlock(&inode_lock);
663 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800664 reap += invalidate_mapping_pages(&inode->i_data,
665 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 iput(inode);
667 spin_lock(&inode_lock);
668
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100669 if (inode != list_entry(inode_lru.next,
670 struct inode, i_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 continue; /* wrong inode or list_empty */
672 if (!can_unuse(inode))
673 continue;
674 }
Nick Piggin7ef0d732009-03-12 14:31:38 -0700675 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 inode->i_state |= I_FREEING;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100677
678 /*
679 * Move the inode off the IO lists and LRU once I_FREEING is
680 * set so that it won't get moved back on there if it is dirty.
681 */
682 list_move(&inode->i_lru, &freeable);
683 list_del_init(&inode->i_wb_list);
Nick Piggin86c87492011-01-07 17:49:18 +1100684 inodes_stat.nr_unused--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 }
Christoph Lameterf8891e52006-06-30 01:55:45 -0700686 if (current_is_kswapd())
687 __count_vm_events(KSWAPD_INODESTEAL, reap);
688 else
689 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 spin_unlock(&inode_lock);
691
692 dispose_list(&freeable);
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700693 up_read(&iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694}
695
696/*
697 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
698 * "unused" means that no dentries are referring to the inodes: the files are
699 * not open and the dcache references to those inodes have already been
700 * reclaimed.
701 *
702 * This function is passed the number of inodes to scan, and it returns the
703 * total number of remaining possibly-reclaimable inodes.
704 */
Dave Chinner7f8275d2010-07-19 14:56:17 +1000705static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
707 if (nr) {
708 /*
709 * Nasty deadlock avoidance. We may hold various FS locks,
710 * and we don't want to recurse into the FS that called us
711 * in clear_inode() and friends..
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530712 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 if (!(gfp_mask & __GFP_FS))
714 return -1;
715 prune_icache(nr);
716 }
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400717 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718}
719
Rusty Russell8e1f9362007-07-17 04:03:17 -0700720static struct shrinker icache_shrinker = {
721 .shrink = shrink_icache_memory,
722 .seeks = DEFAULT_SEEKS,
723};
724
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725static void __wait_on_freeing_inode(struct inode *inode);
726/*
727 * Called with the inode lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530729static struct inode *find_inode(struct super_block *sb,
730 struct hlist_head *head,
731 int (*test)(struct inode *, void *),
732 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733{
734 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530735 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
737repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700738 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 if (inode->i_sb != sb)
740 continue;
741 if (!test(inode, data))
742 continue;
Al Viroa4ffdde2010-06-02 17:38:30 -0400743 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 __wait_on_freeing_inode(inode);
745 goto repeat;
746 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400747 __iget(inode);
748 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400750 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751}
752
753/*
754 * find_inode_fast is the fast path version of find_inode, see the comment at
755 * iget_locked for details.
756 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530757static struct inode *find_inode_fast(struct super_block *sb,
758 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759{
760 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530761 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
763repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700764 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 if (inode->i_ino != ino)
766 continue;
767 if (inode->i_sb != sb)
768 continue;
Al Viroa4ffdde2010-06-02 17:38:30 -0400769 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 __wait_on_freeing_inode(inode);
771 goto repeat;
772 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400773 __iget(inode);
774 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400776 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777}
778
Eric Dumazetf991bd22010-10-23 11:18:01 -0400779/*
780 * Each cpu owns a range of LAST_INO_BATCH numbers.
781 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
782 * to renew the exhausted range.
David Chinner8290c352008-10-30 17:35:24 +1100783 *
Eric Dumazetf991bd22010-10-23 11:18:01 -0400784 * This does not significantly increase overflow rate because every CPU can
785 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
786 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
787 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
788 * overflow rate by 2x, which does not seem too significant.
789 *
790 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
791 * error if st_ino won't fit in target struct field. Use 32bit counter
792 * here to attempt to avoid that.
David Chinner8290c352008-10-30 17:35:24 +1100793 */
Eric Dumazetf991bd22010-10-23 11:18:01 -0400794#define LAST_INO_BATCH 1024
795static DEFINE_PER_CPU(unsigned int, last_ino);
David Chinner8290c352008-10-30 17:35:24 +1100796
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400797unsigned int get_next_ino(void)
Eric Dumazetf991bd22010-10-23 11:18:01 -0400798{
799 unsigned int *p = &get_cpu_var(last_ino);
800 unsigned int res = *p;
801
802#ifdef CONFIG_SMP
803 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
804 static atomic_t shared_last_ino;
805 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
806
807 res = next - LAST_INO_BATCH;
808 }
809#endif
810
811 *p = ++res;
812 put_cpu_var(last_ino);
813 return res;
David Chinner8290c352008-10-30 17:35:24 +1100814}
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400815EXPORT_SYMBOL(get_next_ino);
David Chinner8290c352008-10-30 17:35:24 +1100816
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817/**
818 * new_inode - obtain an inode
819 * @sb: superblock
820 *
Mel Gorman769848c2007-07-17 04:03:05 -0700821 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800822 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700823 * If HIGHMEM pages are unsuitable or it is known that pages allocated
824 * for the page cache are not reclaimable or migratable,
825 * mapping_set_gfp_mask() must be called with suitable flags on the
826 * newly created inode's mapping
827 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 */
829struct inode *new_inode(struct super_block *sb)
830{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530831 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832
833 spin_lock_prefetch(&inode_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530834
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 inode = alloc_inode(sb);
836 if (inode) {
837 spin_lock(&inode_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400838 __inode_sb_list_add(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 inode->i_state = 0;
840 spin_unlock(&inode_lock);
841 }
842 return inode;
843}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844EXPORT_SYMBOL(new_inode);
845
846void unlock_new_inode(struct inode *inode)
847{
Peter Zijlstra14358e62007-10-14 01:38:33 +0200848#ifdef CONFIG_DEBUG_LOCK_ALLOC
Namhyung Kima3314a02010-10-11 22:38:00 +0900849 if (S_ISDIR(inode->i_mode)) {
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200850 struct file_system_type *type = inode->i_sb->s_type;
851
Jan Kara9a7aa122009-06-04 15:26:49 +0200852 /* Set new key only if filesystem hasn't already changed it */
853 if (!lockdep_match_class(&inode->i_mutex,
854 &type->i_mutex_key)) {
855 /*
856 * ensure nobody is actually holding i_mutex
857 */
858 mutex_destroy(&inode->i_mutex);
859 mutex_init(&inode->i_mutex);
860 lockdep_set_class(&inode->i_mutex,
861 &type->i_mutex_dir_key);
862 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200863 }
Peter Zijlstra14358e62007-10-14 01:38:33 +0200864#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 /*
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100866 * This is special! We do not need the spinlock when clearing I_NEW,
Jan Kara580be082009-09-21 17:01:06 -0700867 * because we're guaranteed that nobody else tries to do anything about
868 * the state of the inode when it is locked, as we just created it (so
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100869 * there can be no old holders that haven't tested I_NEW).
Jan Kara580be082009-09-21 17:01:06 -0700870 * However we must emit the memory barrier so that other CPUs reliably
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100871 * see the clearing of I_NEW after the other inode initialisation has
Jan Kara580be082009-09-21 17:01:06 -0700872 * completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 */
Jan Kara580be082009-09-21 17:01:06 -0700874 smp_mb();
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100875 WARN_ON(!(inode->i_state & I_NEW));
876 inode->i_state &= ~I_NEW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 wake_up_inode(inode);
878}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879EXPORT_SYMBOL(unlock_new_inode);
880
881/*
882 * This is called without the inode lock held.. Be careful.
883 *
884 * We no longer cache the sb_flags in i_flags - see fs.h
885 * -- rmk@arm.uk.linux.org
886 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530887static struct inode *get_new_inode(struct super_block *sb,
888 struct hlist_head *head,
889 int (*test)(struct inode *, void *),
890 int (*set)(struct inode *, void *),
891 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530893 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
895 inode = alloc_inode(sb);
896 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530897 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
899 spin_lock(&inode_lock);
900 /* We released the lock, so.. */
901 old = find_inode(sb, head, test, data);
902 if (!old) {
903 if (set(inode, data))
904 goto set_failed;
905
Christoph Hellwig646ec462010-10-23 07:15:32 -0400906 hlist_add_head(&inode->i_hash, head);
907 __inode_sb_list_add(inode);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100908 inode->i_state = I_NEW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 spin_unlock(&inode_lock);
910
911 /* Return the locked inode with I_NEW set, the
912 * caller is responsible for filling in the contents
913 */
914 return inode;
915 }
916
917 /*
918 * Uhhuh, somebody else created the same inode under
919 * us. Use the old inode instead of the one we just
920 * allocated.
921 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 spin_unlock(&inode_lock);
923 destroy_inode(inode);
924 inode = old;
925 wait_on_inode(inode);
926 }
927 return inode;
928
929set_failed:
930 spin_unlock(&inode_lock);
931 destroy_inode(inode);
932 return NULL;
933}
934
935/*
936 * get_new_inode_fast is the fast path version of get_new_inode, see the
937 * comment at iget_locked for details.
938 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530939static struct inode *get_new_inode_fast(struct super_block *sb,
940 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530942 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
944 inode = alloc_inode(sb);
945 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530946 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
948 spin_lock(&inode_lock);
949 /* We released the lock, so.. */
950 old = find_inode_fast(sb, head, ino);
951 if (!old) {
952 inode->i_ino = ino;
Christoph Hellwig646ec462010-10-23 07:15:32 -0400953 hlist_add_head(&inode->i_hash, head);
954 __inode_sb_list_add(inode);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100955 inode->i_state = I_NEW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 spin_unlock(&inode_lock);
957
958 /* Return the locked inode with I_NEW set, the
959 * caller is responsible for filling in the contents
960 */
961 return inode;
962 }
963
964 /*
965 * Uhhuh, somebody else created the same inode under
966 * us. Use the old inode instead of the one we just
967 * allocated.
968 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 spin_unlock(&inode_lock);
970 destroy_inode(inode);
971 inode = old;
972 wait_on_inode(inode);
973 }
974 return inode;
975}
976
Christoph Hellwigad5e1952010-10-23 07:00:16 -0400977/*
978 * search the inode cache for a matching inode number.
979 * If we find one, then the inode number we are trying to
980 * allocate is not unique and so we should not use it.
981 *
982 * Returns 1 if the inode number is unique, 0 if it is not.
983 */
984static int test_inode_iunique(struct super_block *sb, unsigned long ino)
985{
986 struct hlist_head *b = inode_hashtable + hash(sb, ino);
987 struct hlist_node *node;
988 struct inode *inode;
989
990 hlist_for_each_entry(inode, node, b, i_hash) {
991 if (inode->i_ino == ino && inode->i_sb == sb)
992 return 0;
993 }
994
995 return 1;
996}
997
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998/**
999 * iunique - get a unique inode number
1000 * @sb: superblock
1001 * @max_reserved: highest reserved inode number
1002 *
1003 * Obtain an inode number that is unique on the system for a given
1004 * superblock. This is used by file systems that have no natural
1005 * permanent inode numbering system. An inode number is returned that
1006 * is higher than the reserved limit but unique.
1007 *
1008 * BUGS:
1009 * With a large number of inodes live on the file system this function
1010 * currently becomes quite slow.
1011 */
1012ino_t iunique(struct super_block *sb, ino_t max_reserved)
1013{
Jeff Layton866b04f2007-05-08 00:32:29 -07001014 /*
1015 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1016 * error if st_ino won't fit in target struct field. Use 32bit counter
1017 * here to attempt to avoid that.
1018 */
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001019 static DEFINE_SPINLOCK(iunique_lock);
Jeff Layton866b04f2007-05-08 00:32:29 -07001020 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001023 spin_lock(&inode_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001024 spin_lock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001025 do {
1026 if (counter <= max_reserved)
1027 counter = max_reserved + 1;
1028 res = counter++;
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001029 } while (!test_inode_iunique(sb, res));
1030 spin_unlock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001031 spin_unlock(&inode_lock);
1032
1033 return res;
1034}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035EXPORT_SYMBOL(iunique);
1036
1037struct inode *igrab(struct inode *inode)
1038{
1039 spin_lock(&inode_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -04001040 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 __iget(inode);
1042 else
1043 /*
1044 * Handle the case where s_op->clear_inode is not been
1045 * called yet, and somebody is calling igrab
1046 * while the inode is getting freed.
1047 */
1048 inode = NULL;
1049 spin_unlock(&inode_lock);
1050 return inode;
1051}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052EXPORT_SYMBOL(igrab);
1053
1054/**
1055 * ifind - internal function, you want ilookup5() or iget5().
1056 * @sb: super block of file system to search
1057 * @head: the head of the list to search
1058 * @test: callback used for comparisons between inodes
1059 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001060 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 *
1062 * ifind() searches for the inode specified by @data in the inode
1063 * cache. This is a generalized version of ifind_fast() for file systems where
1064 * the inode number is not sufficient for unique identification of an inode.
1065 *
1066 * If the inode is in the cache, the inode is returned with an incremented
1067 * reference count.
1068 *
1069 * Otherwise NULL is returned.
1070 *
1071 * Note, @test is called with the inode_lock held, so can't sleep.
1072 */
Matt Mackall5d2bea42006-01-08 01:05:21 -08001073static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001075 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076{
1077 struct inode *inode;
1078
1079 spin_lock(&inode_lock);
1080 inode = find_inode(sb, head, test, data);
1081 if (inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001083 if (likely(wait))
1084 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 return inode;
1086 }
1087 spin_unlock(&inode_lock);
1088 return NULL;
1089}
1090
1091/**
1092 * ifind_fast - internal function, you want ilookup() or iget().
1093 * @sb: super block of file system to search
1094 * @head: head of the list to search
1095 * @ino: inode number to search for
1096 *
1097 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1098 * file systems where the inode number is sufficient for unique identification
1099 * of an inode.
1100 *
1101 * If the inode is in the cache, the inode is returned with an incremented
1102 * reference count.
1103 *
1104 * Otherwise NULL is returned.
1105 */
Matt Mackall5d2bea42006-01-08 01:05:21 -08001106static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 struct hlist_head *head, unsigned long ino)
1108{
1109 struct inode *inode;
1110
1111 spin_lock(&inode_lock);
1112 inode = find_inode_fast(sb, head, ino);
1113 if (inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 spin_unlock(&inode_lock);
1115 wait_on_inode(inode);
1116 return inode;
1117 }
1118 spin_unlock(&inode_lock);
1119 return NULL;
1120}
1121
1122/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001123 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 * @sb: super block of file system to search
1125 * @hashval: hash value (usually inode number) to search for
1126 * @test: callback used for comparisons between inodes
1127 * @data: opaque data pointer to pass to @test
1128 *
1129 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1130 * @data in the inode cache. This is a generalized version of ilookup() for
1131 * file systems where the inode number is not sufficient for unique
1132 * identification of an inode.
1133 *
1134 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001135 * reference count. Note, the inode lock is not waited upon so you have to be
1136 * very careful what you do with the returned inode. You probably should be
1137 * using ilookup5() instead.
1138 *
1139 * Otherwise NULL is returned.
1140 *
1141 * Note, @test is called with the inode_lock held, so can't sleep.
1142 */
1143struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1144 int (*test)(struct inode *, void *), void *data)
1145{
1146 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1147
1148 return ifind(sb, head, test, data, 0);
1149}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001150EXPORT_SYMBOL(ilookup5_nowait);
1151
1152/**
1153 * ilookup5 - search for an inode in the inode cache
1154 * @sb: super block of file system to search
1155 * @hashval: hash value (usually inode number) to search for
1156 * @test: callback used for comparisons between inodes
1157 * @data: opaque data pointer to pass to @test
1158 *
1159 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1160 * @data in the inode cache. This is a generalized version of ilookup() for
1161 * file systems where the inode number is not sufficient for unique
1162 * identification of an inode.
1163 *
1164 * If the inode is in the cache, the inode lock is waited upon and the inode is
1165 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 *
1167 * Otherwise NULL is returned.
1168 *
1169 * Note, @test is called with the inode_lock held, so can't sleep.
1170 */
1171struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1172 int (*test)(struct inode *, void *), void *data)
1173{
1174 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1175
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001176 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178EXPORT_SYMBOL(ilookup5);
1179
1180/**
1181 * ilookup - search for an inode in the inode cache
1182 * @sb: super block of file system to search
1183 * @ino: inode number to search for
1184 *
1185 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1186 * This is for file systems where the inode number is sufficient for unique
1187 * identification of an inode.
1188 *
1189 * If the inode is in the cache, the inode is returned with an incremented
1190 * reference count.
1191 *
1192 * Otherwise NULL is returned.
1193 */
1194struct inode *ilookup(struct super_block *sb, unsigned long ino)
1195{
1196 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1197
1198 return ifind_fast(sb, head, ino);
1199}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200EXPORT_SYMBOL(ilookup);
1201
1202/**
1203 * iget5_locked - obtain an inode from a mounted file system
1204 * @sb: super block of file system
1205 * @hashval: hash value (usually inode number) to get
1206 * @test: callback used for comparisons between inodes
1207 * @set: callback used to initialize a new struct inode
1208 * @data: opaque data pointer to pass to @test and @set
1209 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1211 * and @data in the inode cache and if present it is returned with an increased
1212 * reference count. This is a generalized version of iget_locked() for file
1213 * systems where the inode number is not sufficient for unique identification
1214 * of an inode.
1215 *
1216 * If the inode is not in cache, get_new_inode() is called to allocate a new
1217 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1218 * file system gets to fill it in before unlocking it via unlock_new_inode().
1219 *
1220 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1221 */
1222struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1223 int (*test)(struct inode *, void *),
1224 int (*set)(struct inode *, void *), void *data)
1225{
1226 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1227 struct inode *inode;
1228
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001229 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 if (inode)
1231 return inode;
1232 /*
1233 * get_new_inode() 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(sb, head, test, set, data);
1237}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238EXPORT_SYMBOL(iget5_locked);
1239
1240/**
1241 * iget_locked - obtain an inode from a mounted file system
1242 * @sb: super block of file system
1243 * @ino: inode number to get
1244 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1246 * the inode cache and if present it is returned with an increased reference
1247 * count. This is for file systems where the inode number is sufficient for
1248 * unique identification of an inode.
1249 *
1250 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1251 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1252 * The file system gets to fill it in before unlocking it via
1253 * unlock_new_inode().
1254 */
1255struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1256{
1257 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1258 struct inode *inode;
1259
1260 inode = ifind_fast(sb, head, ino);
1261 if (inode)
1262 return inode;
1263 /*
1264 * get_new_inode_fast() will do the right thing, re-trying the search
1265 * in case it had to block at any point.
1266 */
1267 return get_new_inode_fast(sb, head, ino);
1268}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269EXPORT_SYMBOL(iget_locked);
1270
Al Viro261bca82008-12-30 01:48:21 -05001271int insert_inode_locked(struct inode *inode)
1272{
1273 struct super_block *sb = inode->i_sb;
1274 ino_t ino = inode->i_ino;
1275 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001276
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001277 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001278 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001279 struct hlist_node *node;
1280 struct inode *old = NULL;
Al Viro261bca82008-12-30 01:48:21 -05001281 spin_lock(&inode_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001282 hlist_for_each_entry(old, node, head, i_hash) {
1283 if (old->i_ino != ino)
1284 continue;
1285 if (old->i_sb != sb)
1286 continue;
Al Viroa4ffdde2010-06-02 17:38:30 -04001287 if (old->i_state & (I_FREEING|I_WILL_FREE))
Al Viro72a43d62009-05-13 19:13:40 +01001288 continue;
1289 break;
1290 }
1291 if (likely(!node)) {
Al Viro261bca82008-12-30 01:48:21 -05001292 hlist_add_head(&inode->i_hash, head);
1293 spin_unlock(&inode_lock);
1294 return 0;
1295 }
1296 __iget(old);
1297 spin_unlock(&inode_lock);
1298 wait_on_inode(old);
Al Viro1d3382c2010-10-23 15:19:20 -04001299 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001300 iput(old);
1301 return -EBUSY;
1302 }
1303 iput(old);
1304 }
1305}
Al Viro261bca82008-12-30 01:48:21 -05001306EXPORT_SYMBOL(insert_inode_locked);
1307
1308int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1309 int (*test)(struct inode *, void *), void *data)
1310{
1311 struct super_block *sb = inode->i_sb;
1312 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001313
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001314 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001315
1316 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001317 struct hlist_node *node;
1318 struct inode *old = NULL;
1319
Al Viro261bca82008-12-30 01:48:21 -05001320 spin_lock(&inode_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001321 hlist_for_each_entry(old, node, head, i_hash) {
1322 if (old->i_sb != sb)
1323 continue;
1324 if (!test(old, data))
1325 continue;
Al Viroa4ffdde2010-06-02 17:38:30 -04001326 if (old->i_state & (I_FREEING|I_WILL_FREE))
Al Viro72a43d62009-05-13 19:13:40 +01001327 continue;
1328 break;
1329 }
1330 if (likely(!node)) {
Al Viro261bca82008-12-30 01:48:21 -05001331 hlist_add_head(&inode->i_hash, head);
1332 spin_unlock(&inode_lock);
1333 return 0;
1334 }
1335 __iget(old);
1336 spin_unlock(&inode_lock);
1337 wait_on_inode(old);
Al Viro1d3382c2010-10-23 15:19:20 -04001338 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001339 iput(old);
1340 return -EBUSY;
1341 }
1342 iput(old);
1343 }
1344}
Al Viro261bca82008-12-30 01:48:21 -05001345EXPORT_SYMBOL(insert_inode_locked4);
1346
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
Al Viro45321ac2010-06-07 13:43:19 -04001348int generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349{
Al Viro45321ac2010-06-07 13:43:19 -04001350 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352EXPORT_SYMBOL(generic_delete_inode);
1353
Al Viro45321ac2010-06-07 13:43:19 -04001354/*
1355 * Normal UNIX filesystem behaviour: delete the
1356 * inode when the usage count drops to zero, and
1357 * i_nlink is zero.
Jan Kara22fe40422009-09-18 13:05:44 -07001358 */
Al Viro45321ac2010-06-07 13:43:19 -04001359int generic_drop_inode(struct inode *inode)
1360{
Al Viro1d3382c2010-10-23 15:19:20 -04001361 return !inode->i_nlink || inode_unhashed(inode);
Al Viro45321ac2010-06-07 13:43:19 -04001362}
1363EXPORT_SYMBOL_GPL(generic_drop_inode);
1364
1365/*
1366 * Called when we're dropping the last reference
1367 * to an inode.
1368 *
1369 * Call the FS "drop_inode()" function, defaulting to
1370 * the legacy UNIX filesystem behaviour. If it tells
1371 * us to evict inode, do so. Otherwise, retain inode
1372 * in cache if fs is alive, sync and evict if fs is
1373 * shutting down.
1374 */
1375static void iput_final(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376{
1377 struct super_block *sb = inode->i_sb;
Al Viro45321ac2010-06-07 13:43:19 -04001378 const struct super_operations *op = inode->i_sb->s_op;
1379 int drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
Al Viro45321ac2010-06-07 13:43:19 -04001381 if (op && op->drop_inode)
1382 drop = op->drop_inode(inode);
1383 else
1384 drop = generic_drop_inode(inode);
1385
1386 if (!drop) {
Christoph Hellwigacb0c852007-05-08 00:25:52 -07001387 if (sb->s_flags & MS_ACTIVE) {
Nick Piggin9e38d862010-10-23 06:55:17 -04001388 inode->i_state |= I_REFERENCED;
1389 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1390 inode_lru_list_add(inode);
1391 }
Alexander Viro991114c2005-06-23 00:09:01 -07001392 spin_unlock(&inode_lock);
Al Viro45321ac2010-06-07 13:43:19 -04001393 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001394 }
Nick Piggin7ef0d732009-03-12 14:31:38 -07001395 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001396 inode->i_state |= I_WILL_FREE;
1397 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 write_inode_now(inode, 1);
1399 spin_lock(&inode_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001400 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001401 inode->i_state &= ~I_WILL_FREE;
Dave Chinner4c51acb2010-10-23 06:58:09 -04001402 __remove_inode_hash(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001404
Nick Piggin7ef0d732009-03-12 14:31:38 -07001405 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001406 inode->i_state |= I_FREEING;
Nick Piggin9e38d862010-10-23 06:55:17 -04001407
1408 /*
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001409 * Move the inode off the IO lists and LRU once I_FREEING is
1410 * set so that it won't get moved back on there if it is dirty.
Nick Piggin9e38d862010-10-23 06:55:17 -04001411 */
1412 inode_lru_list_del(inode);
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001413 list_del_init(&inode->i_wb_list);
Nick Piggin9e38d862010-10-23 06:55:17 -04001414
Christoph Hellwig646ec462010-10-23 07:15:32 -04001415 __inode_sb_list_del(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 spin_unlock(&inode_lock);
Al Viro644da592010-06-07 13:21:05 -04001417 evict(inode);
Dave Chinner4c51acb2010-10-23 06:58:09 -04001418 remove_inode_hash(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001419 wake_up_inode(inode);
Al Viro45321ac2010-06-07 13:43:19 -04001420 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 destroy_inode(inode);
1422}
1423
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301425 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 * @inode: inode to put
1427 *
1428 * Puts an inode, dropping its usage count. If the inode use count hits
1429 * zero, the inode is then freed and may also be destroyed.
1430 *
1431 * Consequently, iput() can sleep.
1432 */
1433void iput(struct inode *inode)
1434{
1435 if (inode) {
Al Viroa4ffdde2010-06-02 17:38:30 -04001436 BUG_ON(inode->i_state & I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1439 iput_final(inode);
1440 }
1441}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442EXPORT_SYMBOL(iput);
1443
1444/**
1445 * bmap - find a block number in a file
1446 * @inode: inode of file
1447 * @block: block to find
1448 *
1449 * Returns the block number on the device holding the inode that
1450 * is the disk block number for the block of the file requested.
1451 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301452 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 * file.
1454 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301455sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456{
1457 sector_t res = 0;
1458 if (inode->i_mapping->a_ops->bmap)
1459 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1460 return res;
1461}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462EXPORT_SYMBOL(bmap);
1463
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001464/*
1465 * With relative atime, only update atime if the previous atime is
1466 * earlier than either the ctime or mtime or if at least a day has
1467 * passed since the last atime update.
1468 */
1469static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1470 struct timespec now)
1471{
1472
1473 if (!(mnt->mnt_flags & MNT_RELATIME))
1474 return 1;
1475 /*
1476 * Is mtime younger than atime? If yes, update atime:
1477 */
1478 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1479 return 1;
1480 /*
1481 * Is ctime younger than atime? If yes, update atime:
1482 */
1483 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1484 return 1;
1485
1486 /*
1487 * Is the previous atime value older than a day? If yes,
1488 * update atime:
1489 */
1490 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1491 return 1;
1492 /*
1493 * Good, we can skip the atime update:
1494 */
1495 return 0;
1496}
1497
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001499 * touch_atime - update the access time
1500 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001501 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 *
1503 * Update the accessed time on an inode and mark it for writeback.
1504 * This function automatically handles read only file systems and media,
1505 * as well as the "noatime" flag and inode specific "noatime" markers.
1506 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001507void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001509 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 struct timespec now;
1511
Andrew Mortonb2276132006-12-13 00:34:33 -08001512 if (inode->i_flags & S_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001513 return;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001514 if (IS_NOATIME(inode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001515 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001516 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001517 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001518
Dave Hansencdb70f32008-02-15 14:37:41 -08001519 if (mnt->mnt_flags & MNT_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001520 return;
Dave Hansencdb70f32008-02-15 14:37:41 -08001521 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001522 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001523
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 now = current_fs_time(inode->i_sb);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001525
1526 if (!relatime_need_update(mnt, inode, now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001527 return;
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001528
Valerie Henson47ae32d2006-12-13 00:34:34 -08001529 if (timespec_equal(&inode->i_atime, &now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001530 return;
1531
1532 if (mnt_want_write(mnt))
1533 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001534
1535 inode->i_atime = now;
1536 mark_inode_dirty_sync(inode);
Dave Hansencdb70f32008-02-15 14:37:41 -08001537 mnt_drop_write(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001539EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
1541/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001542 * file_update_time - update mtime and ctime time
1543 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001545 * Update the mtime and ctime members of an inode and mark the inode
1546 * for writeback. Note that this function is meant exclusively for
1547 * usage in the file write path of filesystems, and filesystems may
1548 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001549 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Christoph Hellwig870f4812006-01-09 20:52:01 -08001550 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 */
1552
Christoph Hellwig870f4812006-01-09 20:52:01 -08001553void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001555 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 struct timespec now;
Andi Kleence06e0b2009-09-18 13:05:48 -07001557 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558
Andi Kleence06e0b2009-09-18 13:05:48 -07001559 /* First try to exhaust all avenues to not sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 if (IS_NOCMTIME(inode))
1561 return;
Dave Hansen20ddee22008-02-15 14:37:43 -08001562
Andi Kleence06e0b2009-09-18 13:05:48 -07001563 now = current_fs_time(inode->i_sb);
1564 if (!timespec_equal(&inode->i_mtime, &now))
1565 sync_it = S_MTIME;
1566
1567 if (!timespec_equal(&inode->i_ctime, &now))
1568 sync_it |= S_CTIME;
1569
1570 if (IS_I_VERSION(inode))
1571 sync_it |= S_VERSION;
1572
1573 if (!sync_it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 return;
1575
Andi Kleence06e0b2009-09-18 13:05:48 -07001576 /* Finally allowed to write? Takes lock. */
1577 if (mnt_want_write_file(file))
1578 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
Andi Kleence06e0b2009-09-18 13:05:48 -07001580 /* Only change inode inside the lock region */
1581 if (sync_it & S_VERSION)
Jean Noel Cordenner7a224222008-01-28 23:58:27 -05001582 inode_inc_iversion(inode);
Andi Kleence06e0b2009-09-18 13:05:48 -07001583 if (sync_it & S_CTIME)
1584 inode->i_ctime = now;
1585 if (sync_it & S_MTIME)
1586 inode->i_mtime = now;
1587 mark_inode_dirty_sync(inode);
Dave Hansen20ddee22008-02-15 14:37:43 -08001588 mnt_drop_write(file->f_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001590EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591
1592int inode_needs_sync(struct inode *inode)
1593{
1594 if (IS_SYNC(inode))
1595 return 1;
1596 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1597 return 1;
1598 return 0;
1599}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600EXPORT_SYMBOL(inode_needs_sync);
1601
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602int inode_wait(void *word)
1603{
1604 schedule();
1605 return 0;
1606}
Stephen Rothwelld44dab82008-11-10 17:06:05 +11001607EXPORT_SYMBOL(inode_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608
1609/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001610 * If we try to find an inode in the inode hash while it is being
1611 * deleted, we have to wait until the filesystem completes its
1612 * deletion before reporting that it isn't found. This function waits
1613 * until the deletion _might_ have completed. Callers are responsible
1614 * to recheck inode state.
1615 *
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001616 * It doesn't matter if I_NEW is not set initially, a call to
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001617 * wake_up_inode() after removing from the hash list will DTRT.
1618 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 * This is called with inode_lock held.
1620 */
1621static void __wait_on_freeing_inode(struct inode *inode)
1622{
1623 wait_queue_head_t *wq;
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001624 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1625 wq = bit_waitqueue(&inode->i_state, __I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1627 spin_unlock(&inode_lock);
1628 schedule();
1629 finish_wait(wq, &wait.wait);
1630 spin_lock(&inode_lock);
1631}
1632
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633static __initdata unsigned long ihash_entries;
1634static int __init set_ihash_entries(char *str)
1635{
1636 if (!str)
1637 return 0;
1638 ihash_entries = simple_strtoul(str, &str, 0);
1639 return 1;
1640}
1641__setup("ihash_entries=", set_ihash_entries);
1642
1643/*
1644 * Initialize the waitqueues and inode hash table.
1645 */
1646void __init inode_init_early(void)
1647{
1648 int loop;
1649
1650 /* If hashes are distributed across NUMA nodes, defer
1651 * hash allocation until vmalloc space is available.
1652 */
1653 if (hashdist)
1654 return;
1655
1656 inode_hashtable =
1657 alloc_large_system_hash("Inode-cache",
1658 sizeof(struct hlist_head),
1659 ihash_entries,
1660 14,
1661 HASH_EARLY,
1662 &i_hash_shift,
1663 &i_hash_mask,
1664 0);
1665
1666 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1667 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1668}
1669
Denis Cheng74bf17c2007-10-16 23:26:30 -07001670void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671{
1672 int loop;
1673
1674 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001675 inode_cachep = kmem_cache_create("inode_cache",
1676 sizeof(struct inode),
1677 0,
1678 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1679 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001680 init_once);
Rusty Russell8e1f9362007-07-17 04:03:17 -07001681 register_shrinker(&icache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
1683 /* Hash may have been set up in inode_init_early */
1684 if (!hashdist)
1685 return;
1686
1687 inode_hashtable =
1688 alloc_large_system_hash("Inode-cache",
1689 sizeof(struct hlist_head),
1690 ihash_entries,
1691 14,
1692 0,
1693 &i_hash_shift,
1694 &i_hash_mask,
1695 0);
1696
1697 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1698 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1699}
1700
1701void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1702{
1703 inode->i_mode = mode;
1704 if (S_ISCHR(mode)) {
1705 inode->i_fop = &def_chr_fops;
1706 inode->i_rdev = rdev;
1707 } else if (S_ISBLK(mode)) {
1708 inode->i_fop = &def_blk_fops;
1709 inode->i_rdev = rdev;
1710 } else if (S_ISFIFO(mode))
1711 inode->i_fop = &def_fifo_fops;
1712 else if (S_ISSOCK(mode))
1713 inode->i_fop = &bad_sock_fops;
1714 else
Manish Katiyaraf0d9ae2009-09-18 13:05:43 -07001715 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1716 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1717 inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718}
1719EXPORT_SYMBOL(init_special_inode);
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001720
1721/**
1722 * Init uid,gid,mode for new inode according to posix standards
1723 * @inode: New inode
1724 * @dir: Directory inode
1725 * @mode: mode of the new inode
1726 */
1727void inode_init_owner(struct inode *inode, const struct inode *dir,
1728 mode_t mode)
1729{
1730 inode->i_uid = current_fsuid();
1731 if (dir && dir->i_mode & S_ISGID) {
1732 inode->i_gid = dir->i_gid;
1733 if (S_ISDIR(mode))
1734 mode |= S_ISGID;
1735 } else
1736 inode->i_gid = current_fsgid();
1737 inode->i_mode = mode;
1738}
1739EXPORT_SYMBOL(inode_init_owner);