blob: 8b612813a6a74b2eeec407a65998191f5db9e8cf [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * (C) 1997 Linus Torvalds
Christoph Hellwig4b4563d2011-05-27 09:28:01 -04003 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005#include <linux/fs.h>
6#include <linux/mm.h>
7#include <linux/dcache.h>
8#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/slab.h>
10#include <linux/writeback.h>
11#include <linux/module.h>
12#include <linux/backing-dev.h>
13#include <linux/wait.h>
Nick Piggin88e0fbc2009-09-22 16:43:50 -070014#include <linux/rwsem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/hash.h>
16#include <linux/swap.h>
17#include <linux/security.h>
18#include <linux/pagemap.h>
19#include <linux/cdev.h>
20#include <linux/bootmem.h>
Eric Paris3be25f42009-05-21 17:01:26 -040021#include <linux/fsnotify.h>
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -080022#include <linux/mount.h>
Arjan van de Venefaee192009-01-06 07:20:54 -080023#include <linux/async.h>
Al Virof19d4a82009-06-08 19:50:45 -040024#include <linux/posix_acl.h>
Heiko Carstens9ce6e0b2011-05-22 18:54:21 +020025#include <linux/prefetch.h>
Eric Parisa178d202010-10-25 14:41:59 -040026#include <linux/ima.h>
Serge E. Hallyne795b712011-03-23 16:43:25 -070027#include <linux/cred.h>
Christoph Hellwig4b4563d2011-05-27 09:28:01 -040028#include <linux/buffer_head.h> /* for inode_has_buffers */
Miklos Szeredi7ada4db2011-11-21 12:11:32 +010029#include <linux/ratelimit.h>
Dave Chinnera66979a2011-03-22 22:23:41 +110030#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32/*
Christoph Hellwig4b4563d2011-05-27 09:28:01 -040033 * Inode locking rules:
Dave Chinner250df6e2011-03-22 22:23:36 +110034 *
35 * inode->i_lock protects:
36 * inode->i_state, inode->i_hash, __iget()
Dave Chinner09cc9fc2011-07-08 14:14:40 +100037 * inode->i_sb->s_inode_lru_lock protects:
Dave Chinner98b745c2011-07-08 14:14:39 +100038 * inode->i_sb->s_inode_lru, inode->i_lru
Dave Chinner55fa6092011-03-22 22:23:40 +110039 * inode_sb_list_lock protects:
40 * sb->s_inodes, inode->i_sb_list
Christoph Hellwigf758eea2011-04-21 18:19:44 -060041 * bdi->wb.list_lock protects:
Dave Chinnera66979a2011-03-22 22:23:41 +110042 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
Dave Chinner67a23c42011-03-22 22:23:42 +110043 * inode_hash_lock protects:
44 * inode_hashtable, inode->i_hash
Dave Chinner250df6e2011-03-22 22:23:36 +110045 *
46 * Lock ordering:
Dave Chinner55fa6092011-03-22 22:23:40 +110047 *
48 * inode_sb_list_lock
49 * inode->i_lock
Dave Chinner09cc9fc2011-07-08 14:14:40 +100050 * inode->i_sb->s_inode_lru_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110051 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060052 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110053 * inode->i_lock
Dave Chinner67a23c42011-03-22 22:23:42 +110054 *
55 * inode_hash_lock
56 * inode_sb_list_lock
57 * inode->i_lock
58 *
59 * iunique_lock
60 * inode_hash_lock
Dave Chinner250df6e2011-03-22 22:23:36 +110061 */
62
Eric Dumazetfa3536c2006-03-26 01:37:24 -080063static unsigned int i_hash_mask __read_mostly;
64static unsigned int i_hash_shift __read_mostly;
Dave Chinner67a23c42011-03-22 22:23:42 +110065static struct hlist_head *inode_hashtable __read_mostly;
66static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
Dave Chinner55fa6092011-03-22 22:23:40 +110068__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
69
Linus Torvalds1da177e2005-04-16 15:20:36 -070070/*
Jens Axboe7dcda1c2011-04-05 23:51:48 +020071 * Empty aops. Can be used for the cases where the user does not
72 * define any of the address_space operations.
73 */
74const struct address_space_operations empty_aops = {
75};
76EXPORT_SYMBOL(empty_aops);
77
78/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 * Statistics gathering..
80 */
81struct inodes_stat_t inodes_stat;
82
Nick Piggin3e880fb2011-01-07 17:49:19 +110083static DEFINE_PER_CPU(unsigned int, nr_inodes);
Dave Chinnerfcb94f72011-07-08 14:14:38 +100084static DEFINE_PER_CPU(unsigned int, nr_unused);
Dave Chinnercffbc8a2010-10-23 05:03:02 -040085
Manish Katiyar6b3304b2009-03-31 19:35:54 +053086static struct kmem_cache *inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
Nick Piggin3e880fb2011-01-07 17:49:19 +110088static int get_nr_inodes(void)
Dave Chinnercffbc8a2010-10-23 05:03:02 -040089{
Nick Piggin3e880fb2011-01-07 17:49:19 +110090 int i;
91 int sum = 0;
92 for_each_possible_cpu(i)
93 sum += per_cpu(nr_inodes, i);
94 return sum < 0 ? 0 : sum;
Dave Chinnercffbc8a2010-10-23 05:03:02 -040095}
96
97static inline int get_nr_inodes_unused(void)
98{
Dave Chinnerfcb94f72011-07-08 14:14:38 +100099 int i;
100 int sum = 0;
101 for_each_possible_cpu(i)
102 sum += per_cpu(nr_unused, i);
103 return sum < 0 ? 0 : sum;
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400104}
105
106int get_nr_dirty_inodes(void)
107{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100108 /* not actually dirty inodes, but a wild approximation */
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400109 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
110 return nr_dirty > 0 ? nr_dirty : 0;
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400111}
112
113/*
114 * Handle nr_inode sysctl
115 */
116#ifdef CONFIG_SYSCTL
117int proc_nr_inodes(ctl_table *table, int write,
118 void __user *buffer, size_t *lenp, loff_t *ppos)
119{
120 inodes_stat.nr_inodes = get_nr_inodes();
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000121 inodes_stat.nr_unused = get_nr_inodes_unused();
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400122 return proc_dointvec(table, write, buffer, lenp, ppos);
123}
124#endif
125
David Chinner2cb15992008-10-30 17:32:23 +1100126/**
127 * inode_init_always - perform inode structure intialisation
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800128 * @sb: superblock inode belongs to
129 * @inode: inode to initialise
David Chinner2cb15992008-10-30 17:32:23 +1100130 *
131 * These are initializations that need to be done on every inode
132 * allocation as the fields are not initialised by slab allocation.
133 */
Christoph Hellwig54e34622009-08-07 14:38:25 -0300134int inode_init_always(struct super_block *sb, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135{
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700136 static const struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800137 static const struct file_operations empty_fops;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530138 struct address_space *const mapping = &inode->i_data;
David Chinner2cb15992008-10-30 17:32:23 +1100139
140 inode->i_sb = sb;
141 inode->i_blkbits = sb->s_blocksize_bits;
142 inode->i_flags = 0;
143 atomic_set(&inode->i_count, 1);
144 inode->i_op = &empty_iops;
145 inode->i_fop = &empty_fops;
Miklos Szeredia78ef702011-10-28 14:13:30 +0200146 inode->__i_nlink = 1;
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700147 inode->i_opflags = 0;
Al Viro56ff5ef2008-12-09 09:34:39 -0500148 inode->i_uid = 0;
149 inode->i_gid = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100150 atomic_set(&inode->i_writecount, 0);
151 inode->i_size = 0;
152 inode->i_blocks = 0;
153 inode->i_bytes = 0;
154 inode->i_generation = 0;
155#ifdef CONFIG_QUOTA
156 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
157#endif
158 inode->i_pipe = NULL;
159 inode->i_bdev = NULL;
160 inode->i_cdev = NULL;
161 inode->i_rdev = 0;
162 inode->dirtied_when = 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500163
164 if (security_inode_alloc(inode))
Christoph Hellwig54e34622009-08-07 14:38:25 -0300165 goto out;
David Chinner2cb15992008-10-30 17:32:23 +1100166 spin_lock_init(&inode->i_lock);
167 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
168
169 mutex_init(&inode->i_mutex);
170 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
171
Christoph Hellwigbd5fe6c2011-06-24 14:29:43 -0400172 atomic_set(&inode->i_dio_count, 0);
David Chinner2cb15992008-10-30 17:32:23 +1100173
174 mapping->a_ops = &empty_aops;
175 mapping->host = inode;
176 mapping->flags = 0;
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800177 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
David Chinner2cb15992008-10-30 17:32:23 +1100178 mapping->assoc_mapping = NULL;
179 mapping->backing_dev_info = &default_backing_dev_info;
180 mapping->writeback_index = 0;
181
182 /*
183 * If the block_device provides a backing_dev_info for client
184 * inodes then use that. Otherwise the inode share the bdev's
185 * backing_dev_info.
186 */
187 if (sb->s_bdev) {
188 struct backing_dev_info *bdi;
189
Jens Axboe2c96ce92009-09-15 09:43:56 +0200190 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
David Chinner2cb15992008-10-30 17:32:23 +1100191 mapping->backing_dev_info = bdi;
192 }
193 inode->i_private = NULL;
194 inode->i_mapping = mapping;
Al Viro6b520e02011-12-12 15:51:45 -0500195 INIT_LIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
Al Virof19d4a82009-06-08 19:50:45 -0400196#ifdef CONFIG_FS_POSIX_ACL
197 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
198#endif
David Chinner2cb15992008-10-30 17:32:23 +1100199
Eric Paris3be25f42009-05-21 17:01:26 -0400200#ifdef CONFIG_FSNOTIFY
201 inode->i_fsnotify_mask = 0;
202#endif
203
Nick Piggin3e880fb2011-01-07 17:49:19 +1100204 this_cpu_inc(nr_inodes);
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400205
Christoph Hellwig54e34622009-08-07 14:38:25 -0300206 return 0;
Christoph Hellwig54e34622009-08-07 14:38:25 -0300207out:
208 return -ENOMEM;
David Chinner2cb15992008-10-30 17:32:23 +1100209}
210EXPORT_SYMBOL(inode_init_always);
211
212static struct inode *alloc_inode(struct super_block *sb)
213{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 struct inode *inode;
215
216 if (sb->s_op->alloc_inode)
217 inode = sb->s_op->alloc_inode(sb);
218 else
David Chinner2cb15992008-10-30 17:32:23 +1100219 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
Christoph Hellwig54e34622009-08-07 14:38:25 -0300221 if (!inode)
222 return NULL;
223
224 if (unlikely(inode_init_always(sb, inode))) {
225 if (inode->i_sb->s_op->destroy_inode)
226 inode->i_sb->s_op->destroy_inode(inode);
227 else
228 kmem_cache_free(inode_cachep, inode);
229 return NULL;
230 }
231
232 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
234
Nick Pigginff0c7d12011-01-07 17:49:50 +1100235void free_inode_nonrcu(struct inode *inode)
236{
237 kmem_cache_free(inode_cachep, inode);
238}
239EXPORT_SYMBOL(free_inode_nonrcu);
240
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300241void __destroy_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200243 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 security_inode_free(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400245 fsnotify_inode_delete(inode);
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100246 if (!inode->i_nlink) {
247 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
248 atomic_long_dec(&inode->i_sb->s_remove_count);
249 }
250
Al Virof19d4a82009-06-08 19:50:45 -0400251#ifdef CONFIG_FS_POSIX_ACL
252 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
253 posix_acl_release(inode->i_acl);
254 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
255 posix_acl_release(inode->i_default_acl);
256#endif
Nick Piggin3e880fb2011-01-07 17:49:19 +1100257 this_cpu_dec(nr_inodes);
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300258}
259EXPORT_SYMBOL(__destroy_inode);
260
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100261static void i_callback(struct rcu_head *head)
262{
263 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100264 kmem_cache_free(inode_cachep, inode);
265}
266
Christoph Hellwig56b0dac2010-10-06 10:48:55 +0200267static void destroy_inode(struct inode *inode)
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300268{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100269 BUG_ON(!list_empty(&inode->i_lru));
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300270 __destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 if (inode->i_sb->s_op->destroy_inode)
272 inode->i_sb->s_op->destroy_inode(inode);
273 else
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100274 call_rcu(&inode->i_rcu, i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100277/**
278 * drop_nlink - directly drop an inode's link count
279 * @inode: inode
280 *
281 * This is a low-level filesystem helper to replace any
282 * direct filesystem manipulation of i_nlink. In cases
283 * where we are attempting to track writes to the
284 * filesystem, a decrement to zero means an imminent
285 * write when the file is truncated and actually unlinked
286 * on the filesystem.
287 */
288void drop_nlink(struct inode *inode)
289{
290 WARN_ON(inode->i_nlink == 0);
291 inode->__i_nlink--;
292 if (!inode->i_nlink)
293 atomic_long_inc(&inode->i_sb->s_remove_count);
294}
295EXPORT_SYMBOL(drop_nlink);
296
297/**
298 * clear_nlink - directly zero an inode's link count
299 * @inode: inode
300 *
301 * This is a low-level filesystem helper to replace any
302 * direct filesystem manipulation of i_nlink. See
303 * drop_nlink() for why we care about i_nlink hitting zero.
304 */
305void clear_nlink(struct inode *inode)
306{
307 if (inode->i_nlink) {
308 inode->__i_nlink = 0;
309 atomic_long_inc(&inode->i_sb->s_remove_count);
310 }
311}
312EXPORT_SYMBOL(clear_nlink);
313
314/**
315 * set_nlink - directly set an inode's link count
316 * @inode: inode
317 * @nlink: new nlink (should be non-zero)
318 *
319 * This is a low-level filesystem helper to replace any
320 * direct filesystem manipulation of i_nlink.
321 */
322void set_nlink(struct inode *inode, unsigned int nlink)
323{
324 if (!nlink) {
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100325 clear_nlink(inode);
326 } else {
327 /* Yes, some filesystems do change nlink from zero to one */
328 if (inode->i_nlink == 0)
329 atomic_long_dec(&inode->i_sb->s_remove_count);
330
331 inode->__i_nlink = nlink;
332 }
333}
334EXPORT_SYMBOL(set_nlink);
335
336/**
337 * inc_nlink - directly increment an inode's link count
338 * @inode: inode
339 *
340 * This is a low-level filesystem helper to replace any
341 * direct filesystem manipulation of i_nlink. Currently,
342 * it is only here for parity with dec_nlink().
343 */
344void inc_nlink(struct inode *inode)
345{
346 if (WARN_ON(inode->i_nlink == 0))
347 atomic_long_dec(&inode->i_sb->s_remove_count);
348
349 inode->__i_nlink++;
350}
351EXPORT_SYMBOL(inc_nlink);
352
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100353void address_space_init_once(struct address_space *mapping)
354{
355 memset(mapping, 0, sizeof(*mapping));
356 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
357 spin_lock_init(&mapping->tree_lock);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700358 mutex_init(&mapping->i_mmap_mutex);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100359 INIT_LIST_HEAD(&mapping->private_list);
360 spin_lock_init(&mapping->private_lock);
361 INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
362 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100363}
364EXPORT_SYMBOL(address_space_init_once);
365
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366/*
367 * These are initializations that only need to be done
368 * once, because the fields are idempotent across use
369 * of the inode, so let the slab aware of that.
370 */
371void inode_init_once(struct inode *inode)
372{
373 memset(inode, 0, sizeof(*inode));
374 INIT_HLIST_NODE(&inode->i_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 INIT_LIST_HEAD(&inode->i_devices);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100376 INIT_LIST_HEAD(&inode->i_wb_list);
377 INIT_LIST_HEAD(&inode->i_lru);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100378 address_space_init_once(&inode->i_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 i_size_ordered_init(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400380#ifdef CONFIG_FSNOTIFY
Eric Parise61ce862009-12-17 21:24:24 -0500381 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
Eric Paris3be25f42009-05-21 17:01:26 -0400382#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384EXPORT_SYMBOL(inode_init_once);
385
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700386static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530388 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
Christoph Lametera35afb82007-05-16 22:10:57 -0700390 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
392
393/*
Dave Chinner250df6e2011-03-22 22:23:36 +1100394 * inode->i_lock must be held
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530396void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397{
Nick Piggin9e38d862010-10-23 06:55:17 -0400398 atomic_inc(&inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399}
400
Al Viro7de9c6ee2010-10-23 11:11:40 -0400401/*
402 * get additional reference to inode; caller must already hold one.
403 */
404void ihold(struct inode *inode)
405{
406 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
407}
408EXPORT_SYMBOL(ihold);
409
Nick Piggin9e38d862010-10-23 06:55:17 -0400410static void inode_lru_list_add(struct inode *inode)
411{
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000412 spin_lock(&inode->i_sb->s_inode_lru_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100413 if (list_empty(&inode->i_lru)) {
Dave Chinner98b745c2011-07-08 14:14:39 +1000414 list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
415 inode->i_sb->s_nr_inodes_unused++;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000416 this_cpu_inc(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400417 }
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000418 spin_unlock(&inode->i_sb->s_inode_lru_lock);
Nick Piggin9e38d862010-10-23 06:55:17 -0400419}
420
421static void inode_lru_list_del(struct inode *inode)
422{
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000423 spin_lock(&inode->i_sb->s_inode_lru_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100424 if (!list_empty(&inode->i_lru)) {
425 list_del_init(&inode->i_lru);
Dave Chinner98b745c2011-07-08 14:14:39 +1000426 inode->i_sb->s_nr_inodes_unused--;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000427 this_cpu_dec(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400428 }
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000429 spin_unlock(&inode->i_sb->s_inode_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430}
431
Christoph Hellwig646ec462010-10-23 07:15:32 -0400432/**
433 * inode_sb_list_add - add inode to the superblock list of inodes
434 * @inode: inode to add
435 */
436void inode_sb_list_add(struct inode *inode)
437{
Dave Chinner55fa6092011-03-22 22:23:40 +1100438 spin_lock(&inode_sb_list_lock);
439 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
440 spin_unlock(&inode_sb_list_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400441}
442EXPORT_SYMBOL_GPL(inode_sb_list_add);
443
Dave Chinner55fa6092011-03-22 22:23:40 +1100444static inline void inode_sb_list_del(struct inode *inode)
Christoph Hellwig646ec462010-10-23 07:15:32 -0400445{
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200446 if (!list_empty(&inode->i_sb_list)) {
447 spin_lock(&inode_sb_list_lock);
448 list_del_init(&inode->i_sb_list);
449 spin_unlock(&inode_sb_list_lock);
450 }
Christoph Hellwig646ec462010-10-23 07:15:32 -0400451}
452
Dave Chinner4c51acb2010-10-23 06:58:09 -0400453static unsigned long hash(struct super_block *sb, unsigned long hashval)
454{
455 unsigned long tmp;
456
457 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
458 L1_CACHE_BYTES;
Christoph Hellwig4b4563d2011-05-27 09:28:01 -0400459 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
460 return tmp & i_hash_mask;
Dave Chinner4c51acb2010-10-23 06:58:09 -0400461}
462
463/**
464 * __insert_inode_hash - hash an inode
465 * @inode: unhashed inode
466 * @hashval: unsigned long value used to locate this object in the
467 * inode_hashtable.
468 *
469 * Add an inode to the inode hash for this superblock.
470 */
471void __insert_inode_hash(struct inode *inode, unsigned long hashval)
472{
Christoph Hellwig646ec462010-10-23 07:15:32 -0400473 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
474
Dave Chinner67a23c42011-03-22 22:23:42 +1100475 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100476 spin_lock(&inode->i_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400477 hlist_add_head(&inode->i_hash, b);
Dave Chinner250df6e2011-03-22 22:23:36 +1100478 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100479 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400480}
481EXPORT_SYMBOL(__insert_inode_hash);
482
483/**
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200484 * __remove_inode_hash - remove an inode from the hash
Dave Chinner4c51acb2010-10-23 06:58:09 -0400485 * @inode: inode to unhash
486 *
487 * Remove an inode from the superblock.
488 */
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200489void __remove_inode_hash(struct inode *inode)
Dave Chinner4c51acb2010-10-23 06:58:09 -0400490{
Dave Chinner67a23c42011-03-22 22:23:42 +1100491 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100492 spin_lock(&inode->i_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400493 hlist_del_init(&inode->i_hash);
Dave Chinner250df6e2011-03-22 22:23:36 +1100494 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100495 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400496}
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200497EXPORT_SYMBOL(__remove_inode_hash);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400498
Al Virob0683aa2010-06-04 20:55:25 -0400499void end_writeback(struct inode *inode)
500{
501 might_sleep();
Jan Kara08142572011-06-27 16:18:10 -0700502 /*
503 * We have to cycle tree_lock here because reclaim can be still in the
504 * process of removing the last page (in __delete_from_page_cache())
505 * and we must not free mapping under it.
506 */
507 spin_lock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400508 BUG_ON(inode->i_data.nrpages);
Jan Kara08142572011-06-27 16:18:10 -0700509 spin_unlock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400510 BUG_ON(!list_empty(&inode->i_data.private_list));
511 BUG_ON(!(inode->i_state & I_FREEING));
512 BUG_ON(inode->i_state & I_CLEAR);
513 inode_sync_wait(inode);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100514 /* don't need i_lock here, no concurrent mods to i_state */
Al Virob0683aa2010-06-04 20:55:25 -0400515 inode->i_state = I_FREEING | I_CLEAR;
516}
517EXPORT_SYMBOL(end_writeback);
518
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100519/*
520 * Free the inode passed in, removing it from the lists it is still connected
521 * to. We remove any pages still attached to the inode and wait for any IO that
522 * is still in progress before finally destroying the inode.
523 *
524 * An inode must already be marked I_FREEING so that we avoid the inode being
525 * moved back onto lists if we race with other code that manipulates the lists
526 * (e.g. writeback_single_inode). The caller is responsible for setting this.
527 *
528 * An inode must already be removed from the LRU list before being evicted from
529 * the cache. This should occur atomically with setting the I_FREEING state
530 * flag, so no inodes here should ever be on the LRU when being evicted.
531 */
Al Viro644da592010-06-07 13:21:05 -0400532static void evict(struct inode *inode)
Al Virob4272d42010-06-04 19:33:20 -0400533{
534 const struct super_operations *op = inode->i_sb->s_op;
535
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100536 BUG_ON(!(inode->i_state & I_FREEING));
537 BUG_ON(!list_empty(&inode->i_lru));
538
Eric Dumazetb12362b2011-07-28 06:11:47 +0200539 if (!list_empty(&inode->i_wb_list))
540 inode_wb_list_del(inode);
541
Dave Chinner55fa6092011-03-22 22:23:40 +1100542 inode_sb_list_del(inode);
543
Al Virobe7ce412010-06-04 19:40:39 -0400544 if (op->evict_inode) {
545 op->evict_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400546 } else {
547 if (inode->i_data.nrpages)
548 truncate_inode_pages(&inode->i_data, 0);
Al Viro30140832010-06-07 13:23:20 -0400549 end_writeback(inode);
Al Virob4272d42010-06-04 19:33:20 -0400550 }
Al Viro661074e2010-06-04 20:19:55 -0400551 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
552 bd_forget(inode);
553 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
554 cd_forget(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100555
556 remove_inode_hash(inode);
557
558 spin_lock(&inode->i_lock);
559 wake_up_bit(&inode->i_state, __I_NEW);
560 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
561 spin_unlock(&inode->i_lock);
562
563 destroy_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400564}
565
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566/*
567 * dispose_list - dispose of the contents of a local list
568 * @head: the head of the list to free
569 *
570 * Dispose-list gets a local list with local inodes in it, so it doesn't
571 * need to worry about list corruption and SMP locks.
572 */
573static void dispose_list(struct list_head *head)
574{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 while (!list_empty(head)) {
576 struct inode *inode;
577
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100578 inode = list_first_entry(head, struct inode, i_lru);
579 list_del_init(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
Al Viro644da592010-06-07 13:21:05 -0400581 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583}
584
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585/**
Al Viro63997e92010-10-25 20:49:35 -0400586 * evict_inodes - evict all evictable inodes for a superblock
587 * @sb: superblock to operate on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 *
Al Viro63997e92010-10-25 20:49:35 -0400589 * Make sure that no inodes with zero refcount are retained. This is
590 * called by superblock shutdown after having MS_ACTIVE flag removed,
591 * so any inode reaching zero refcount during or after that call will
592 * be immediately evicted.
593 */
594void evict_inodes(struct super_block *sb)
595{
596 struct inode *inode, *next;
597 LIST_HEAD(dispose);
598
Dave Chinner55fa6092011-03-22 22:23:40 +1100599 spin_lock(&inode_sb_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400600 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
601 if (atomic_read(&inode->i_count))
602 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100603
604 spin_lock(&inode->i_lock);
605 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
606 spin_unlock(&inode->i_lock);
Al Viro63997e92010-10-25 20:49:35 -0400607 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100608 }
Al Viro63997e92010-10-25 20:49:35 -0400609
610 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100611 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100612 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100613 list_add(&inode->i_lru, &dispose);
Al Viro63997e92010-10-25 20:49:35 -0400614 }
Dave Chinner55fa6092011-03-22 22:23:40 +1100615 spin_unlock(&inode_sb_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400616
617 dispose_list(&dispose);
Al Viro63997e92010-10-25 20:49:35 -0400618}
619
620/**
Christoph Hellwiga0318782010-10-24 19:40:33 +0200621 * invalidate_inodes - attempt to free all inodes on a superblock
622 * @sb: superblock to operate on
NeilBrown93b270f2011-02-24 17:25:47 +1100623 * @kill_dirty: flag to guide handling of dirty inodes
Christoph Hellwiga0318782010-10-24 19:40:33 +0200624 *
625 * Attempts to free all inodes for a given superblock. If there were any
626 * busy inodes return a non-zero value, else zero.
NeilBrown93b270f2011-02-24 17:25:47 +1100627 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
628 * them as busy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 */
NeilBrown93b270f2011-02-24 17:25:47 +1100630int invalidate_inodes(struct super_block *sb, bool kill_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631{
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400632 int busy = 0;
Christoph Hellwiga0318782010-10-24 19:40:33 +0200633 struct inode *inode, *next;
634 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635
Dave Chinner55fa6092011-03-22 22:23:40 +1100636 spin_lock(&inode_sb_list_lock);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200637 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100638 spin_lock(&inode->i_lock);
639 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
640 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100642 }
NeilBrown93b270f2011-02-24 17:25:47 +1100643 if (inode->i_state & I_DIRTY && !kill_dirty) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100644 spin_unlock(&inode->i_lock);
NeilBrown93b270f2011-02-24 17:25:47 +1100645 busy = 1;
646 continue;
647 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200648 if (atomic_read(&inode->i_count)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100649 spin_unlock(&inode->i_lock);
Christoph Hellwig99a38912010-10-23 19:07:20 +0200650 busy = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 continue;
652 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200653
Christoph Hellwig99a38912010-10-23 19:07:20 +0200654 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100655 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100656 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100657 list_add(&inode->i_lru, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 }
Dave Chinner55fa6092011-03-22 22:23:40 +1100659 spin_unlock(&inode_sb_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
Christoph Hellwiga0318782010-10-24 19:40:33 +0200661 dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
663 return busy;
664}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
666static int can_unuse(struct inode *inode)
667{
Nick Piggin9e38d862010-10-23 06:55:17 -0400668 if (inode->i_state & ~I_REFERENCED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 return 0;
670 if (inode_has_buffers(inode))
671 return 0;
672 if (atomic_read(&inode->i_count))
673 return 0;
674 if (inode->i_data.nrpages)
675 return 0;
676 return 1;
677}
678
679/*
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000680 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
681 * This is called from the superblock shrinker function with a number of inodes
682 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
683 * then are freed outside inode_lock by dispose_list().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 *
685 * Any inodes which are pinned purely because of attached pagecache have their
Nick Piggin9e38d862010-10-23 06:55:17 -0400686 * pagecache removed. If the inode has metadata buffers attached to
687 * mapping->private_list then try to remove them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 *
Nick Piggin9e38d862010-10-23 06:55:17 -0400689 * If the inode has the I_REFERENCED flag set, then it means that it has been
690 * used recently - the flag is set in iput_final(). When we encounter such an
691 * inode, clear the flag and move it to the back of the LRU so it gets another
692 * pass through the LRU before it gets reclaimed. This is necessary because of
693 * the fact we are doing lazy LRU updates to minimise lock contention so the
694 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
695 * with this flag set because they are the inodes that are out of order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 */
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000697void prune_icache_sb(struct super_block *sb, int nr_to_scan)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
699 LIST_HEAD(freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 int nr_scanned;
701 unsigned long reap = 0;
702
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000703 spin_lock(&sb->s_inode_lru_lock);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000704 for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 struct inode *inode;
706
Dave Chinner98b745c2011-07-08 14:14:39 +1000707 if (list_empty(&sb->s_inode_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 break;
709
Dave Chinner98b745c2011-07-08 14:14:39 +1000710 inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
Nick Piggin9e38d862010-10-23 06:55:17 -0400712 /*
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000713 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
Dave Chinner02afc412011-03-22 22:23:38 +1100714 * so use a trylock. If we fail to get the lock, just move the
715 * inode to the back of the list so we don't spin on it.
716 */
717 if (!spin_trylock(&inode->i_lock)) {
Christoph Hellwig62a3dde2011-10-28 10:03:41 +0200718 list_move_tail(&inode->i_lru, &sb->s_inode_lru);
Dave Chinner02afc412011-03-22 22:23:38 +1100719 continue;
720 }
721
722 /*
Nick Piggin9e38d862010-10-23 06:55:17 -0400723 * Referenced or dirty inodes are still in use. Give them
724 * another pass through the LRU as we canot reclaim them now.
725 */
726 if (atomic_read(&inode->i_count) ||
727 (inode->i_state & ~I_REFERENCED)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100728 list_del_init(&inode->i_lru);
Dave Chinnerf283c862011-03-22 22:23:39 +1100729 spin_unlock(&inode->i_lock);
Dave Chinner98b745c2011-07-08 14:14:39 +1000730 sb->s_nr_inodes_unused--;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000731 this_cpu_dec(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400732 continue;
733 }
734
735 /* recently referenced inodes get one more pass */
736 if (inode->i_state & I_REFERENCED) {
Nick Piggin9e38d862010-10-23 06:55:17 -0400737 inode->i_state &= ~I_REFERENCED;
Dave Chinner98b745c2011-07-08 14:14:39 +1000738 list_move(&inode->i_lru, &sb->s_inode_lru);
Dave Chinnerf283c862011-03-22 22:23:39 +1100739 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 continue;
741 }
742 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
743 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100744 spin_unlock(&inode->i_lock);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000745 spin_unlock(&sb->s_inode_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800747 reap += invalidate_mapping_pages(&inode->i_data,
748 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 iput(inode);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000750 spin_lock(&sb->s_inode_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
Dave Chinner98b745c2011-07-08 14:14:39 +1000752 if (inode != list_entry(sb->s_inode_lru.next,
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100753 struct inode, i_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 continue; /* wrong inode or list_empty */
Dave Chinner02afc412011-03-22 22:23:38 +1100755 /* avoid lock inversions with trylock */
756 if (!spin_trylock(&inode->i_lock))
757 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100758 if (!can_unuse(inode)) {
759 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100761 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
Nick Piggin7ef0d732009-03-12 14:31:38 -0700763 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 inode->i_state |= I_FREEING;
Dave Chinner250df6e2011-03-22 22:23:36 +1100765 spin_unlock(&inode->i_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100766
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100767 list_move(&inode->i_lru, &freeable);
Dave Chinner98b745c2011-07-08 14:14:39 +1000768 sb->s_nr_inodes_unused--;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000769 this_cpu_dec(nr_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 }
Christoph Lameterf8891e52006-06-30 01:55:45 -0700771 if (current_is_kswapd())
772 __count_vm_events(KSWAPD_INODESTEAL, reap);
773 else
774 __count_vm_events(PGINODESTEAL, reap);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000775 spin_unlock(&sb->s_inode_lru_lock);
Konstantin Khlebnikov5f8aefd2012-01-10 15:07:18 -0800776 if (current->reclaim_state)
777 current->reclaim_state->reclaimed_slab += reap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
779 dispose_list(&freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780}
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782static void __wait_on_freeing_inode(struct inode *inode);
783/*
784 * Called with the inode lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530786static struct inode *find_inode(struct super_block *sb,
787 struct hlist_head *head,
788 int (*test)(struct inode *, void *),
789 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
791 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530792 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
794repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700795 hlist_for_each_entry(inode, node, head, i_hash) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100796 spin_lock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100797 if (inode->i_sb != sb) {
798 spin_unlock(&inode->i_lock);
799 continue;
800 }
801 if (!test(inode, data)) {
802 spin_unlock(&inode->i_lock);
803 continue;
804 }
Al Viroa4ffdde2010-06-02 17:38:30 -0400805 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 __wait_on_freeing_inode(inode);
807 goto repeat;
808 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400809 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100810 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400811 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400813 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814}
815
816/*
817 * find_inode_fast is the fast path version of find_inode, see the comment at
818 * iget_locked for details.
819 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530820static struct inode *find_inode_fast(struct super_block *sb,
821 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822{
823 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530824 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
826repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700827 hlist_for_each_entry(inode, node, head, i_hash) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100828 spin_lock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100829 if (inode->i_ino != ino) {
830 spin_unlock(&inode->i_lock);
831 continue;
832 }
833 if (inode->i_sb != sb) {
834 spin_unlock(&inode->i_lock);
835 continue;
836 }
Al Viroa4ffdde2010-06-02 17:38:30 -0400837 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 __wait_on_freeing_inode(inode);
839 goto repeat;
840 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400841 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100842 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400843 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400845 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846}
847
Eric Dumazetf991bd22010-10-23 11:18:01 -0400848/*
849 * Each cpu owns a range of LAST_INO_BATCH numbers.
850 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
851 * to renew the exhausted range.
David Chinner8290c352008-10-30 17:35:24 +1100852 *
Eric Dumazetf991bd22010-10-23 11:18:01 -0400853 * This does not significantly increase overflow rate because every CPU can
854 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
855 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
856 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
857 * overflow rate by 2x, which does not seem too significant.
858 *
859 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
860 * error if st_ino won't fit in target struct field. Use 32bit counter
861 * here to attempt to avoid that.
David Chinner8290c352008-10-30 17:35:24 +1100862 */
Eric Dumazetf991bd22010-10-23 11:18:01 -0400863#define LAST_INO_BATCH 1024
864static DEFINE_PER_CPU(unsigned int, last_ino);
David Chinner8290c352008-10-30 17:35:24 +1100865
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400866unsigned int get_next_ino(void)
Eric Dumazetf991bd22010-10-23 11:18:01 -0400867{
868 unsigned int *p = &get_cpu_var(last_ino);
869 unsigned int res = *p;
870
871#ifdef CONFIG_SMP
872 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
873 static atomic_t shared_last_ino;
874 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
875
876 res = next - LAST_INO_BATCH;
877 }
878#endif
879
880 *p = ++res;
881 put_cpu_var(last_ino);
882 return res;
David Chinner8290c352008-10-30 17:35:24 +1100883}
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400884EXPORT_SYMBOL(get_next_ino);
David Chinner8290c352008-10-30 17:35:24 +1100885
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886/**
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200887 * new_inode_pseudo - obtain an inode
888 * @sb: superblock
889 *
890 * Allocates a new inode for given superblock.
891 * Inode wont be chained in superblock s_inodes list
892 * This means :
893 * - fs can't be unmount
894 * - quotas, fsnotify, writeback can't work
895 */
896struct inode *new_inode_pseudo(struct super_block *sb)
897{
898 struct inode *inode = alloc_inode(sb);
899
900 if (inode) {
901 spin_lock(&inode->i_lock);
902 inode->i_state = 0;
903 spin_unlock(&inode->i_lock);
904 INIT_LIST_HEAD(&inode->i_sb_list);
905 }
906 return inode;
907}
908
909/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 * new_inode - obtain an inode
911 * @sb: superblock
912 *
Mel Gorman769848c2007-07-17 04:03:05 -0700913 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800914 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700915 * If HIGHMEM pages are unsuitable or it is known that pages allocated
916 * for the page cache are not reclaimable or migratable,
917 * mapping_set_gfp_mask() must be called with suitable flags on the
918 * newly created inode's mapping
919 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 */
921struct inode *new_inode(struct super_block *sb)
922{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530923 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
Dave Chinner55fa6092011-03-22 22:23:40 +1100925 spin_lock_prefetch(&inode_sb_list_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530926
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200927 inode = new_inode_pseudo(sb);
928 if (inode)
Dave Chinner55fa6092011-03-22 22:23:40 +1100929 inode_sb_list_add(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 return inode;
931}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932EXPORT_SYMBOL(new_inode);
933
Peter Zijlstra14358e62007-10-14 01:38:33 +0200934#ifdef CONFIG_DEBUG_LOCK_ALLOC
Josh Boyere096d0c2011-08-25 07:48:12 -0400935void lockdep_annotate_inode_mutex_key(struct inode *inode)
936{
Namhyung Kima3314a02010-10-11 22:38:00 +0900937 if (S_ISDIR(inode->i_mode)) {
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200938 struct file_system_type *type = inode->i_sb->s_type;
939
Jan Kara9a7aa122009-06-04 15:26:49 +0200940 /* Set new key only if filesystem hasn't already changed it */
Tyler Hicks978d6d82011-12-12 10:02:30 -0600941 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
Jan Kara9a7aa122009-06-04 15:26:49 +0200942 /*
943 * ensure nobody is actually holding i_mutex
944 */
945 mutex_destroy(&inode->i_mutex);
946 mutex_init(&inode->i_mutex);
947 lockdep_set_class(&inode->i_mutex,
948 &type->i_mutex_dir_key);
949 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200950 }
Josh Boyere096d0c2011-08-25 07:48:12 -0400951}
952EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200953#endif
Josh Boyere096d0c2011-08-25 07:48:12 -0400954
955/**
956 * unlock_new_inode - clear the I_NEW state and wake up any waiters
957 * @inode: new inode to unlock
958 *
959 * Called when the inode is fully initialised to clear the new state of the
960 * inode and wake up anyone waiting for the inode to finish initialisation.
961 */
962void unlock_new_inode(struct inode *inode)
963{
964 lockdep_annotate_inode_mutex_key(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100965 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100966 WARN_ON(!(inode->i_state & I_NEW));
967 inode->i_state &= ~I_NEW;
Al Viro310fa7a2012-03-10 17:07:28 -0500968 smp_mb();
Dave Chinner250df6e2011-03-22 22:23:36 +1100969 wake_up_bit(&inode->i_state, __I_NEW);
970 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972EXPORT_SYMBOL(unlock_new_inode);
973
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400974/**
975 * iget5_locked - obtain an inode from a mounted file system
976 * @sb: super block of file system
977 * @hashval: hash value (usually inode number) to get
978 * @test: callback used for comparisons between inodes
979 * @set: callback used to initialize a new struct inode
980 * @data: opaque data pointer to pass to @test and @set
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400982 * Search for the inode specified by @hashval and @data in the inode cache,
983 * and if present it is return it with an increased reference count. This is
984 * a generalized version of iget_locked() for file systems where the inode
985 * number is not sufficient for unique identification of an inode.
986 *
987 * If the inode is not in cache, allocate a new inode and return it locked,
988 * hashed, and with the I_NEW flag set. The file system gets to fill it in
989 * before unlocking it via unlock_new_inode().
990 *
991 * Note both @test and @set are called with the inode_hash_lock held, so can't
992 * sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400994struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
995 int (*test)(struct inode *, void *),
996 int (*set)(struct inode *, void *), void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400998 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530999 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001001 spin_lock(&inode_hash_lock);
1002 inode = find_inode(sb, head, test, data);
1003 spin_unlock(&inode_hash_lock);
1004
1005 if (inode) {
1006 wait_on_inode(inode);
1007 return inode;
1008 }
1009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 inode = alloc_inode(sb);
1011 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301012 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
Dave Chinner67a23c42011-03-22 22:23:42 +11001014 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 /* We released the lock, so.. */
1016 old = find_inode(sb, head, test, data);
1017 if (!old) {
1018 if (set(inode, data))
1019 goto set_failed;
1020
Dave Chinner250df6e2011-03-22 22:23:36 +11001021 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001022 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001023 hlist_add_head(&inode->i_hash, head);
1024 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001025 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001026 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
1028 /* Return the locked inode with I_NEW set, the
1029 * caller is responsible for filling in the contents
1030 */
1031 return inode;
1032 }
1033
1034 /*
1035 * Uhhuh, somebody else created the same inode under
1036 * us. Use the old inode instead of the one we just
1037 * allocated.
1038 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001039 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 destroy_inode(inode);
1041 inode = old;
1042 wait_on_inode(inode);
1043 }
1044 return inode;
1045
1046set_failed:
Dave Chinner67a23c42011-03-22 22:23:42 +11001047 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 destroy_inode(inode);
1049 return NULL;
1050}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001051EXPORT_SYMBOL(iget5_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001053/**
1054 * iget_locked - obtain an inode from a mounted file system
1055 * @sb: super block of file system
1056 * @ino: inode number to get
1057 *
1058 * Search for the inode specified by @ino in the inode cache and if present
1059 * return it with an increased reference count. This is for file systems
1060 * where the inode number is sufficient for unique identification of an inode.
1061 *
1062 * If the inode is not in cache, allocate a new inode and return it locked,
1063 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1064 * before unlocking it via unlock_new_inode().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001066struct inode *iget_locked(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001068 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301069 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001071 spin_lock(&inode_hash_lock);
1072 inode = find_inode_fast(sb, head, ino);
1073 spin_unlock(&inode_hash_lock);
1074 if (inode) {
1075 wait_on_inode(inode);
1076 return inode;
1077 }
1078
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 inode = alloc_inode(sb);
1080 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301081 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
Dave Chinner67a23c42011-03-22 22:23:42 +11001083 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 /* We released the lock, so.. */
1085 old = find_inode_fast(sb, head, ino);
1086 if (!old) {
1087 inode->i_ino = ino;
Dave Chinner250df6e2011-03-22 22:23:36 +11001088 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001089 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001090 hlist_add_head(&inode->i_hash, head);
1091 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001092 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001093 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
1095 /* Return the locked inode with I_NEW set, the
1096 * caller is responsible for filling in the contents
1097 */
1098 return inode;
1099 }
1100
1101 /*
1102 * Uhhuh, somebody else created the same inode under
1103 * us. Use the old inode instead of the one we just
1104 * allocated.
1105 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001106 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 destroy_inode(inode);
1108 inode = old;
1109 wait_on_inode(inode);
1110 }
1111 return inode;
1112}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001113EXPORT_SYMBOL(iget_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001115/*
1116 * search the inode cache for a matching inode number.
1117 * If we find one, then the inode number we are trying to
1118 * allocate is not unique and so we should not use it.
1119 *
1120 * Returns 1 if the inode number is unique, 0 if it is not.
1121 */
1122static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1123{
1124 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1125 struct hlist_node *node;
1126 struct inode *inode;
1127
Dave Chinner67a23c42011-03-22 22:23:42 +11001128 spin_lock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001129 hlist_for_each_entry(inode, node, b, i_hash) {
Dave Chinner67a23c42011-03-22 22:23:42 +11001130 if (inode->i_ino == ino && inode->i_sb == sb) {
1131 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001132 return 0;
Dave Chinner67a23c42011-03-22 22:23:42 +11001133 }
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001134 }
Dave Chinner67a23c42011-03-22 22:23:42 +11001135 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001136
1137 return 1;
1138}
1139
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140/**
1141 * iunique - get a unique inode number
1142 * @sb: superblock
1143 * @max_reserved: highest reserved inode number
1144 *
1145 * Obtain an inode number that is unique on the system for a given
1146 * superblock. This is used by file systems that have no natural
1147 * permanent inode numbering system. An inode number is returned that
1148 * is higher than the reserved limit but unique.
1149 *
1150 * BUGS:
1151 * With a large number of inodes live on the file system this function
1152 * currently becomes quite slow.
1153 */
1154ino_t iunique(struct super_block *sb, ino_t max_reserved)
1155{
Jeff Layton866b04f2007-05-08 00:32:29 -07001156 /*
1157 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1158 * error if st_ino won't fit in target struct field. Use 32bit counter
1159 * here to attempt to avoid that.
1160 */
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001161 static DEFINE_SPINLOCK(iunique_lock);
Jeff Layton866b04f2007-05-08 00:32:29 -07001162 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001165 spin_lock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001166 do {
1167 if (counter <= max_reserved)
1168 counter = max_reserved + 1;
1169 res = counter++;
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001170 } while (!test_inode_iunique(sb, res));
1171 spin_unlock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001172
1173 return res;
1174}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175EXPORT_SYMBOL(iunique);
1176
1177struct inode *igrab(struct inode *inode)
1178{
Dave Chinner250df6e2011-03-22 22:23:36 +11001179 spin_lock(&inode->i_lock);
1180 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001182 spin_unlock(&inode->i_lock);
1183 } else {
1184 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 /*
1186 * Handle the case where s_op->clear_inode is not been
1187 * called yet, and somebody is calling igrab
1188 * while the inode is getting freed.
1189 */
1190 inode = NULL;
Dave Chinner250df6e2011-03-22 22:23:36 +11001191 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 return inode;
1193}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194EXPORT_SYMBOL(igrab);
1195
1196/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001197 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 * @sb: super block of file system to search
1199 * @hashval: hash value (usually inode number) to search for
1200 * @test: callback used for comparisons between inodes
1201 * @data: opaque data pointer to pass to @test
1202 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001203 * Search for the inode specified by @hashval and @data in the inode cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 * If the inode is in the cache, the inode is returned with an incremented
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001205 * reference count.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001206 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001207 * Note: I_NEW is not waited upon so you have to be very careful what you do
1208 * with the returned inode. You probably should be using ilookup5() instead.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001209 *
Randy Dunlapb6d0ad62011-03-26 13:27:47 -07001210 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001211 */
1212struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1213 int (*test)(struct inode *, void *), void *data)
1214{
1215 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001216 struct inode *inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001217
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001218 spin_lock(&inode_hash_lock);
1219 inode = find_inode(sb, head, test, data);
1220 spin_unlock(&inode_hash_lock);
1221
1222 return inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001223}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001224EXPORT_SYMBOL(ilookup5_nowait);
1225
1226/**
1227 * ilookup5 - search for an inode in the inode cache
1228 * @sb: super block of file system to search
1229 * @hashval: hash value (usually inode number) to search for
1230 * @test: callback used for comparisons between inodes
1231 * @data: opaque data pointer to pass to @test
1232 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001233 * Search for the inode specified by @hashval and @data in the inode cache,
1234 * and if the inode is in the cache, return the inode with an incremented
1235 * reference count. Waits on I_NEW before returning the inode.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001236 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001238 * This is a generalized version of ilookup() for file systems where the
1239 * inode number is not sufficient for unique identification of an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001241 * Note: @test is called with the inode_hash_lock held, so can't sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 */
1243struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1244 int (*test)(struct inode *, void *), void *data)
1245{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001246 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001248 if (inode)
1249 wait_on_inode(inode);
1250 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252EXPORT_SYMBOL(ilookup5);
1253
1254/**
1255 * ilookup - search for an inode in the inode cache
1256 * @sb: super block of file system to search
1257 * @ino: inode number to search for
1258 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001259 * Search for the inode @ino in the inode cache, and if the inode is in the
1260 * cache, the inode is returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 */
1262struct inode *ilookup(struct super_block *sb, unsigned long ino)
1263{
1264 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001265 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001267 spin_lock(&inode_hash_lock);
1268 inode = find_inode_fast(sb, head, ino);
1269 spin_unlock(&inode_hash_lock);
1270
1271 if (inode)
1272 wait_on_inode(inode);
1273 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275EXPORT_SYMBOL(ilookup);
1276
Al Viro261bca82008-12-30 01:48:21 -05001277int insert_inode_locked(struct inode *inode)
1278{
1279 struct super_block *sb = inode->i_sb;
1280 ino_t ino = inode->i_ino;
1281 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001282
Al Viro261bca82008-12-30 01:48:21 -05001283 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001284 struct hlist_node *node;
1285 struct inode *old = NULL;
Dave Chinner67a23c42011-03-22 22:23:42 +11001286 spin_lock(&inode_hash_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001287 hlist_for_each_entry(old, node, head, i_hash) {
1288 if (old->i_ino != ino)
1289 continue;
1290 if (old->i_sb != sb)
1291 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001292 spin_lock(&old->i_lock);
1293 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1294 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001295 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001296 }
Al Viro72a43d62009-05-13 19:13:40 +01001297 break;
1298 }
1299 if (likely(!node)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001300 spin_lock(&inode->i_lock);
1301 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001302 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001303 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001304 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001305 return 0;
1306 }
1307 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001308 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001309 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001310 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001311 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001312 iput(old);
1313 return -EBUSY;
1314 }
1315 iput(old);
1316 }
1317}
Al Viro261bca82008-12-30 01:48:21 -05001318EXPORT_SYMBOL(insert_inode_locked);
1319
1320int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1321 int (*test)(struct inode *, void *), void *data)
1322{
1323 struct super_block *sb = inode->i_sb;
1324 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001325
Al Viro261bca82008-12-30 01:48:21 -05001326 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001327 struct hlist_node *node;
1328 struct inode *old = NULL;
1329
Dave Chinner67a23c42011-03-22 22:23:42 +11001330 spin_lock(&inode_hash_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001331 hlist_for_each_entry(old, node, head, i_hash) {
1332 if (old->i_sb != sb)
1333 continue;
1334 if (!test(old, data))
1335 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001336 spin_lock(&old->i_lock);
1337 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1338 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001339 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001340 }
Al Viro72a43d62009-05-13 19:13:40 +01001341 break;
1342 }
1343 if (likely(!node)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001344 spin_lock(&inode->i_lock);
1345 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001346 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001347 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001348 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001349 return 0;
1350 }
1351 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001352 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001353 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001354 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001355 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001356 iput(old);
1357 return -EBUSY;
1358 }
1359 iput(old);
1360 }
1361}
Al Viro261bca82008-12-30 01:48:21 -05001362EXPORT_SYMBOL(insert_inode_locked4);
1363
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Al Viro45321ac2010-06-07 13:43:19 -04001365int generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366{
Al Viro45321ac2010-06-07 13:43:19 -04001367 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369EXPORT_SYMBOL(generic_delete_inode);
1370
Al Viro45321ac2010-06-07 13:43:19 -04001371/*
Al Viro45321ac2010-06-07 13:43:19 -04001372 * Called when we're dropping the last reference
1373 * to an inode.
1374 *
1375 * Call the FS "drop_inode()" function, defaulting to
1376 * the legacy UNIX filesystem behaviour. If it tells
1377 * us to evict inode, do so. Otherwise, retain inode
1378 * in cache if fs is alive, sync and evict if fs is
1379 * shutting down.
1380 */
1381static void iput_final(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382{
1383 struct super_block *sb = inode->i_sb;
Al Viro45321ac2010-06-07 13:43:19 -04001384 const struct super_operations *op = inode->i_sb->s_op;
1385 int drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
Dave Chinner250df6e2011-03-22 22:23:36 +11001387 WARN_ON(inode->i_state & I_NEW);
1388
Al Viroe7f59092011-07-07 15:45:59 -04001389 if (op->drop_inode)
Al Viro45321ac2010-06-07 13:43:19 -04001390 drop = op->drop_inode(inode);
1391 else
1392 drop = generic_drop_inode(inode);
1393
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001394 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1395 inode->i_state |= I_REFERENCED;
1396 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1397 inode_lru_list_add(inode);
1398 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001399 return;
1400 }
1401
Al Viro45321ac2010-06-07 13:43:19 -04001402 if (!drop) {
Alexander Viro991114c2005-06-23 00:09:01 -07001403 inode->i_state |= I_WILL_FREE;
Dave Chinner250df6e2011-03-22 22:23:36 +11001404 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 write_inode_now(inode, 1);
Dave Chinner250df6e2011-03-22 22:23:36 +11001406 spin_lock(&inode->i_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001407 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001408 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001410
Alexander Viro991114c2005-06-23 00:09:01 -07001411 inode->i_state |= I_FREEING;
Eric Dumazetc4ae0c62011-07-28 06:55:13 +02001412 if (!list_empty(&inode->i_lru))
1413 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001414 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001415
1416 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417}
1418
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301420 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 * @inode: inode to put
1422 *
1423 * Puts an inode, dropping its usage count. If the inode use count hits
1424 * zero, the inode is then freed and may also be destroyed.
1425 *
1426 * Consequently, iput() can sleep.
1427 */
1428void iput(struct inode *inode)
1429{
1430 if (inode) {
Al Viroa4ffdde2010-06-02 17:38:30 -04001431 BUG_ON(inode->i_state & I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
Dave Chinnerf283c862011-03-22 22:23:39 +11001433 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 iput_final(inode);
1435 }
1436}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437EXPORT_SYMBOL(iput);
1438
1439/**
1440 * bmap - find a block number in a file
1441 * @inode: inode of file
1442 * @block: block to find
1443 *
1444 * Returns the block number on the device holding the inode that
1445 * is the disk block number for the block of the file requested.
1446 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301447 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 * file.
1449 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301450sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451{
1452 sector_t res = 0;
1453 if (inode->i_mapping->a_ops->bmap)
1454 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1455 return res;
1456}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457EXPORT_SYMBOL(bmap);
1458
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001459/*
1460 * With relative atime, only update atime if the previous atime is
1461 * earlier than either the ctime or mtime or if at least a day has
1462 * passed since the last atime update.
1463 */
1464static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1465 struct timespec now)
1466{
1467
1468 if (!(mnt->mnt_flags & MNT_RELATIME))
1469 return 1;
1470 /*
1471 * Is mtime younger than atime? If yes, update atime:
1472 */
1473 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1474 return 1;
1475 /*
1476 * Is ctime younger than atime? If yes, update atime:
1477 */
1478 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1479 return 1;
1480
1481 /*
1482 * Is the previous atime value older than a day? If yes,
1483 * update atime:
1484 */
1485 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1486 return 1;
1487 /*
1488 * Good, we can skip the atime update:
1489 */
1490 return 0;
1491}
1492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001494 * touch_atime - update the access time
1495 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001496 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 *
1498 * Update the accessed time on an inode and mark it for writeback.
1499 * This function automatically handles read only file systems and media,
1500 * as well as the "noatime" flag and inode specific "noatime" markers.
1501 */
Al Viro68ac1232012-03-15 08:21:57 -04001502void touch_atime(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503{
Al Viro68ac1232012-03-15 08:21:57 -04001504 struct vfsmount *mnt = path->mnt;
1505 struct inode *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 struct timespec now;
1507
Andrew Mortonb2276132006-12-13 00:34:33 -08001508 if (inode->i_flags & S_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001509 return;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001510 if (IS_NOATIME(inode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001511 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001512 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001513 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001514
Dave Hansencdb70f32008-02-15 14:37:41 -08001515 if (mnt->mnt_flags & MNT_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001516 return;
Dave Hansencdb70f32008-02-15 14:37:41 -08001517 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001518 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001519
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 now = current_fs_time(inode->i_sb);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001521
1522 if (!relatime_need_update(mnt, inode, now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001523 return;
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001524
Valerie Henson47ae32d2006-12-13 00:34:34 -08001525 if (timespec_equal(&inode->i_atime, &now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001526 return;
1527
1528 if (mnt_want_write(mnt))
1529 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001530
1531 inode->i_atime = now;
1532 mark_inode_dirty_sync(inode);
Dave Hansencdb70f32008-02-15 14:37:41 -08001533 mnt_drop_write(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001535EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536
1537/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001538 * file_update_time - update mtime and ctime time
1539 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001541 * Update the mtime and ctime members of an inode and mark the inode
1542 * for writeback. Note that this function is meant exclusively for
1543 * usage in the file write path of filesystems, and filesystems may
1544 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001545 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Christoph Hellwig870f4812006-01-09 20:52:01 -08001546 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 */
1548
Christoph Hellwig870f4812006-01-09 20:52:01 -08001549void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001551 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 struct timespec now;
Andi Kleence06e0b2009-09-18 13:05:48 -07001553 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
Andi Kleence06e0b2009-09-18 13:05:48 -07001555 /* First try to exhaust all avenues to not sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 if (IS_NOCMTIME(inode))
1557 return;
Dave Hansen20ddee22008-02-15 14:37:43 -08001558
Andi Kleence06e0b2009-09-18 13:05:48 -07001559 now = current_fs_time(inode->i_sb);
1560 if (!timespec_equal(&inode->i_mtime, &now))
1561 sync_it = S_MTIME;
1562
1563 if (!timespec_equal(&inode->i_ctime, &now))
1564 sync_it |= S_CTIME;
1565
1566 if (IS_I_VERSION(inode))
1567 sync_it |= S_VERSION;
1568
1569 if (!sync_it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 return;
1571
Andi Kleence06e0b2009-09-18 13:05:48 -07001572 /* Finally allowed to write? Takes lock. */
1573 if (mnt_want_write_file(file))
1574 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
Andi Kleence06e0b2009-09-18 13:05:48 -07001576 /* Only change inode inside the lock region */
1577 if (sync_it & S_VERSION)
Jean Noel Cordenner7a224222008-01-28 23:58:27 -05001578 inode_inc_iversion(inode);
Andi Kleence06e0b2009-09-18 13:05:48 -07001579 if (sync_it & S_CTIME)
1580 inode->i_ctime = now;
1581 if (sync_it & S_MTIME)
1582 inode->i_mtime = now;
1583 mark_inode_dirty_sync(inode);
Al Viro2a79f172011-12-09 08:06:57 -05001584 mnt_drop_write_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001586EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
1588int inode_needs_sync(struct inode *inode)
1589{
1590 if (IS_SYNC(inode))
1591 return 1;
1592 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1593 return 1;
1594 return 0;
1595}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596EXPORT_SYMBOL(inode_needs_sync);
1597
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598int inode_wait(void *word)
1599{
1600 schedule();
1601 return 0;
1602}
Stephen Rothwelld44dab82008-11-10 17:06:05 +11001603EXPORT_SYMBOL(inode_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604
1605/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001606 * If we try to find an inode in the inode hash while it is being
1607 * deleted, we have to wait until the filesystem completes its
1608 * deletion before reporting that it isn't found. This function waits
1609 * until the deletion _might_ have completed. Callers are responsible
1610 * to recheck inode state.
1611 *
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001612 * It doesn't matter if I_NEW is not set initially, a call to
Dave Chinner250df6e2011-03-22 22:23:36 +11001613 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1614 * will DTRT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 */
1616static void __wait_on_freeing_inode(struct inode *inode)
1617{
1618 wait_queue_head_t *wq;
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001619 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1620 wq = bit_waitqueue(&inode->i_state, __I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +11001622 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001623 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 schedule();
1625 finish_wait(wq, &wait.wait);
Dave Chinner67a23c42011-03-22 22:23:42 +11001626 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627}
1628
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629static __initdata unsigned long ihash_entries;
1630static int __init set_ihash_entries(char *str)
1631{
1632 if (!str)
1633 return 0;
1634 ihash_entries = simple_strtoul(str, &str, 0);
1635 return 1;
1636}
1637__setup("ihash_entries=", set_ihash_entries);
1638
1639/*
1640 * Initialize the waitqueues and inode hash table.
1641 */
1642void __init inode_init_early(void)
1643{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001644 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645
1646 /* If hashes are distributed across NUMA nodes, defer
1647 * hash allocation until vmalloc space is available.
1648 */
1649 if (hashdist)
1650 return;
1651
1652 inode_hashtable =
1653 alloc_large_system_hash("Inode-cache",
1654 sizeof(struct hlist_head),
1655 ihash_entries,
1656 14,
1657 HASH_EARLY,
1658 &i_hash_shift,
1659 &i_hash_mask,
1660 0);
1661
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001662 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1664}
1665
Denis Cheng74bf17c2007-10-16 23:26:30 -07001666void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001668 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669
1670 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001671 inode_cachep = kmem_cache_create("inode_cache",
1672 sizeof(struct inode),
1673 0,
1674 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1675 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001676 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677
1678 /* Hash may have been set up in inode_init_early */
1679 if (!hashdist)
1680 return;
1681
1682 inode_hashtable =
1683 alloc_large_system_hash("Inode-cache",
1684 sizeof(struct hlist_head),
1685 ihash_entries,
1686 14,
1687 0,
1688 &i_hash_shift,
1689 &i_hash_mask,
1690 0);
1691
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001692 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1694}
1695
1696void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1697{
1698 inode->i_mode = mode;
1699 if (S_ISCHR(mode)) {
1700 inode->i_fop = &def_chr_fops;
1701 inode->i_rdev = rdev;
1702 } else if (S_ISBLK(mode)) {
1703 inode->i_fop = &def_blk_fops;
1704 inode->i_rdev = rdev;
1705 } else if (S_ISFIFO(mode))
1706 inode->i_fop = &def_fifo_fops;
1707 else if (S_ISSOCK(mode))
1708 inode->i_fop = &bad_sock_fops;
1709 else
Manish Katiyaraf0d9ae2009-09-18 13:05:43 -07001710 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1711 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1712 inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713}
1714EXPORT_SYMBOL(init_special_inode);
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001715
1716/**
Ben Hutchingseaae668d2011-02-15 12:48:09 +00001717 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001718 * @inode: New inode
1719 * @dir: Directory inode
1720 * @mode: mode of the new inode
1721 */
1722void inode_init_owner(struct inode *inode, const struct inode *dir,
Al Viro62bb1092011-07-24 23:20:18 -04001723 umode_t mode)
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001724{
1725 inode->i_uid = current_fsuid();
1726 if (dir && dir->i_mode & S_ISGID) {
1727 inode->i_gid = dir->i_gid;
1728 if (S_ISDIR(mode))
1729 mode |= S_ISGID;
1730 } else
1731 inode->i_gid = current_fsgid();
1732 inode->i_mode = mode;
1733}
1734EXPORT_SYMBOL(inode_init_owner);
Serge E. Hallyne795b712011-03-23 16:43:25 -07001735
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001736/**
1737 * inode_owner_or_capable - check current task permissions to inode
1738 * @inode: inode being checked
1739 *
1740 * Return true if current either has CAP_FOWNER to the inode, or
1741 * owns the file.
Serge E. Hallyne795b712011-03-23 16:43:25 -07001742 */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001743bool inode_owner_or_capable(const struct inode *inode)
Serge E. Hallyne795b712011-03-23 16:43:25 -07001744{
1745 struct user_namespace *ns = inode_userns(inode);
1746
1747 if (current_user_ns() == ns && current_fsuid() == inode->i_uid)
1748 return true;
1749 if (ns_capable(ns, CAP_FOWNER))
1750 return true;
1751 return false;
1752}
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001753EXPORT_SYMBOL(inode_owner_or_capable);