blob: 67880e6043994fdbb1baa8e83c3008dac3e49b58 [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 */
Al Viroe59cc472011-12-07 13:17:19 -05005#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07006#include <linux/fs.h>
7#include <linux/mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/hash.h>
10#include <linux/swap.h>
11#include <linux/security.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/cdev.h>
13#include <linux/bootmem.h>
Eric Paris3be25f42009-05-21 17:01:26 -040014#include <linux/fsnotify.h>
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -080015#include <linux/mount.h>
Al Virof19d4a82009-06-08 19:50:45 -040016#include <linux/posix_acl.h>
Heiko Carstens9ce6e0b2011-05-22 18:54:21 +020017#include <linux/prefetch.h>
Christoph Hellwig4b4563d2011-05-27 09:28:01 -040018#include <linux/buffer_head.h> /* for inode_has_buffers */
Miklos Szeredi7ada4db2011-11-21 12:11:32 +010019#include <linux/ratelimit.h>
Dave Chinnera66979a2011-03-22 22:23:41 +110020#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
22/*
Christoph Hellwig4b4563d2011-05-27 09:28:01 -040023 * Inode locking rules:
Dave Chinner250df6e2011-03-22 22:23:36 +110024 *
25 * inode->i_lock protects:
26 * inode->i_state, inode->i_hash, __iget()
Dave Chinner09cc9fc2011-07-08 14:14:40 +100027 * inode->i_sb->s_inode_lru_lock protects:
Dave Chinner98b745c2011-07-08 14:14:39 +100028 * inode->i_sb->s_inode_lru, inode->i_lru
Dave Chinner55fa6092011-03-22 22:23:40 +110029 * inode_sb_list_lock protects:
30 * sb->s_inodes, inode->i_sb_list
Christoph Hellwigf758eea2011-04-21 18:19:44 -060031 * bdi->wb.list_lock protects:
Dave Chinnera66979a2011-03-22 22:23:41 +110032 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
Dave Chinner67a23c42011-03-22 22:23:42 +110033 * inode_hash_lock protects:
34 * inode_hashtable, inode->i_hash
Dave Chinner250df6e2011-03-22 22:23:36 +110035 *
36 * Lock ordering:
Dave Chinner55fa6092011-03-22 22:23:40 +110037 *
38 * inode_sb_list_lock
39 * inode->i_lock
Dave Chinner09cc9fc2011-07-08 14:14:40 +100040 * inode->i_sb->s_inode_lru_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110041 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060042 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110043 * inode->i_lock
Dave Chinner67a23c42011-03-22 22:23:42 +110044 *
45 * inode_hash_lock
46 * inode_sb_list_lock
47 * inode->i_lock
48 *
49 * iunique_lock
50 * inode_hash_lock
Dave Chinner250df6e2011-03-22 22:23:36 +110051 */
52
Eric Dumazetfa3536c2006-03-26 01:37:24 -080053static unsigned int i_hash_mask __read_mostly;
54static unsigned int i_hash_shift __read_mostly;
Dave Chinner67a23c42011-03-22 22:23:42 +110055static struct hlist_head *inode_hashtable __read_mostly;
56static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Dave Chinner55fa6092011-03-22 22:23:40 +110058__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
59
Linus Torvalds1da177e2005-04-16 15:20:36 -070060/*
Jens Axboe7dcda1c2011-04-05 23:51:48 +020061 * Empty aops. Can be used for the cases where the user does not
62 * define any of the address_space operations.
63 */
64const struct address_space_operations empty_aops = {
65};
66EXPORT_SYMBOL(empty_aops);
67
68/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 * Statistics gathering..
70 */
71struct inodes_stat_t inodes_stat;
72
Nick Piggin3e880fb2011-01-07 17:49:19 +110073static DEFINE_PER_CPU(unsigned int, nr_inodes);
Dave Chinnerfcb94f72011-07-08 14:14:38 +100074static DEFINE_PER_CPU(unsigned int, nr_unused);
Dave Chinnercffbc8a2010-10-23 05:03:02 -040075
Manish Katiyar6b3304b2009-03-31 19:35:54 +053076static struct kmem_cache *inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
Nick Piggin3e880fb2011-01-07 17:49:19 +110078static int get_nr_inodes(void)
Dave Chinnercffbc8a2010-10-23 05:03:02 -040079{
Nick Piggin3e880fb2011-01-07 17:49:19 +110080 int i;
81 int sum = 0;
82 for_each_possible_cpu(i)
83 sum += per_cpu(nr_inodes, i);
84 return sum < 0 ? 0 : sum;
Dave Chinnercffbc8a2010-10-23 05:03:02 -040085}
86
87static inline int get_nr_inodes_unused(void)
88{
Dave Chinnerfcb94f72011-07-08 14:14:38 +100089 int i;
90 int sum = 0;
91 for_each_possible_cpu(i)
92 sum += per_cpu(nr_unused, i);
93 return sum < 0 ? 0 : sum;
Dave Chinnercffbc8a2010-10-23 05:03:02 -040094}
95
96int get_nr_dirty_inodes(void)
97{
Nick Piggin3e880fb2011-01-07 17:49:19 +110098 /* not actually dirty inodes, but a wild approximation */
Dave Chinnercffbc8a2010-10-23 05:03:02 -040099 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
100 return nr_dirty > 0 ? nr_dirty : 0;
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400101}
102
103/*
104 * Handle nr_inode sysctl
105 */
106#ifdef CONFIG_SYSCTL
107int proc_nr_inodes(ctl_table *table, int write,
108 void __user *buffer, size_t *lenp, loff_t *ppos)
109{
110 inodes_stat.nr_inodes = get_nr_inodes();
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000111 inodes_stat.nr_unused = get_nr_inodes_unused();
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400112 return proc_dointvec(table, write, buffer, lenp, ppos);
113}
114#endif
115
David Chinner2cb15992008-10-30 17:32:23 +1100116/**
117 * inode_init_always - perform inode structure intialisation
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800118 * @sb: superblock inode belongs to
119 * @inode: inode to initialise
David Chinner2cb15992008-10-30 17:32:23 +1100120 *
121 * These are initializations that need to be done on every inode
122 * allocation as the fields are not initialised by slab allocation.
123 */
Christoph Hellwig54e34622009-08-07 14:38:25 -0300124int inode_init_always(struct super_block *sb, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125{
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700126 static const struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800127 static const struct file_operations empty_fops;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530128 struct address_space *const mapping = &inode->i_data;
David Chinner2cb15992008-10-30 17:32:23 +1100129
130 inode->i_sb = sb;
131 inode->i_blkbits = sb->s_blocksize_bits;
132 inode->i_flags = 0;
133 atomic_set(&inode->i_count, 1);
134 inode->i_op = &empty_iops;
135 inode->i_fop = &empty_fops;
Miklos Szeredia78ef702011-10-28 14:13:30 +0200136 inode->__i_nlink = 1;
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700137 inode->i_opflags = 0;
Eric W. Biederman92361632012-02-08 07:07:50 -0800138 i_uid_write(inode, 0);
139 i_gid_write(inode, 0);
David Chinner2cb15992008-10-30 17:32:23 +1100140 atomic_set(&inode->i_writecount, 0);
141 inode->i_size = 0;
142 inode->i_blocks = 0;
143 inode->i_bytes = 0;
144 inode->i_generation = 0;
145#ifdef CONFIG_QUOTA
146 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
147#endif
148 inode->i_pipe = NULL;
149 inode->i_bdev = NULL;
150 inode->i_cdev = NULL;
151 inode->i_rdev = 0;
152 inode->dirtied_when = 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500153
154 if (security_inode_alloc(inode))
Christoph Hellwig54e34622009-08-07 14:38:25 -0300155 goto out;
David Chinner2cb15992008-10-30 17:32:23 +1100156 spin_lock_init(&inode->i_lock);
157 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
158
159 mutex_init(&inode->i_mutex);
160 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
161
Christoph Hellwigbd5fe6c2011-06-24 14:29:43 -0400162 atomic_set(&inode->i_dio_count, 0);
David Chinner2cb15992008-10-30 17:32:23 +1100163
164 mapping->a_ops = &empty_aops;
165 mapping->host = inode;
166 mapping->flags = 0;
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800167 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800168 mapping->private_data = NULL;
David Chinner2cb15992008-10-30 17:32:23 +1100169 mapping->backing_dev_info = &default_backing_dev_info;
170 mapping->writeback_index = 0;
171
172 /*
173 * If the block_device provides a backing_dev_info for client
174 * inodes then use that. Otherwise the inode share the bdev's
175 * backing_dev_info.
176 */
177 if (sb->s_bdev) {
178 struct backing_dev_info *bdi;
179
Jens Axboe2c96ce92009-09-15 09:43:56 +0200180 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
David Chinner2cb15992008-10-30 17:32:23 +1100181 mapping->backing_dev_info = bdi;
182 }
183 inode->i_private = NULL;
184 inode->i_mapping = mapping;
Al Virob3d9b7a2012-06-09 13:51:19 -0400185 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
Al Virof19d4a82009-06-08 19:50:45 -0400186#ifdef CONFIG_FS_POSIX_ACL
187 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
188#endif
David Chinner2cb15992008-10-30 17:32:23 +1100189
Eric Paris3be25f42009-05-21 17:01:26 -0400190#ifdef CONFIG_FSNOTIFY
191 inode->i_fsnotify_mask = 0;
192#endif
193
Nick Piggin3e880fb2011-01-07 17:49:19 +1100194 this_cpu_inc(nr_inodes);
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400195
Christoph Hellwig54e34622009-08-07 14:38:25 -0300196 return 0;
Christoph Hellwig54e34622009-08-07 14:38:25 -0300197out:
198 return -ENOMEM;
David Chinner2cb15992008-10-30 17:32:23 +1100199}
200EXPORT_SYMBOL(inode_init_always);
201
202static struct inode *alloc_inode(struct super_block *sb)
203{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 struct inode *inode;
205
206 if (sb->s_op->alloc_inode)
207 inode = sb->s_op->alloc_inode(sb);
208 else
David Chinner2cb15992008-10-30 17:32:23 +1100209 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Christoph Hellwig54e34622009-08-07 14:38:25 -0300211 if (!inode)
212 return NULL;
213
214 if (unlikely(inode_init_always(sb, inode))) {
215 if (inode->i_sb->s_op->destroy_inode)
216 inode->i_sb->s_op->destroy_inode(inode);
217 else
218 kmem_cache_free(inode_cachep, inode);
219 return NULL;
220 }
221
222 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
Nick Pigginff0c7d12011-01-07 17:49:50 +1100225void free_inode_nonrcu(struct inode *inode)
226{
227 kmem_cache_free(inode_cachep, inode);
228}
229EXPORT_SYMBOL(free_inode_nonrcu);
230
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300231void __destroy_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200233 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 security_inode_free(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400235 fsnotify_inode_delete(inode);
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100236 if (!inode->i_nlink) {
237 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
238 atomic_long_dec(&inode->i_sb->s_remove_count);
239 }
240
Al Virof19d4a82009-06-08 19:50:45 -0400241#ifdef CONFIG_FS_POSIX_ACL
242 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
243 posix_acl_release(inode->i_acl);
244 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
245 posix_acl_release(inode->i_default_acl);
246#endif
Nick Piggin3e880fb2011-01-07 17:49:19 +1100247 this_cpu_dec(nr_inodes);
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300248}
249EXPORT_SYMBOL(__destroy_inode);
250
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100251static void i_callback(struct rcu_head *head)
252{
253 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100254 kmem_cache_free(inode_cachep, inode);
255}
256
Christoph Hellwig56b0dac2010-10-06 10:48:55 +0200257static void destroy_inode(struct inode *inode)
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300258{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100259 BUG_ON(!list_empty(&inode->i_lru));
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300260 __destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 if (inode->i_sb->s_op->destroy_inode)
262 inode->i_sb->s_op->destroy_inode(inode);
263 else
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100264 call_rcu(&inode->i_rcu, i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100267/**
268 * drop_nlink - directly drop an inode's link count
269 * @inode: inode
270 *
271 * This is a low-level filesystem helper to replace any
272 * direct filesystem manipulation of i_nlink. In cases
273 * where we are attempting to track writes to the
274 * filesystem, a decrement to zero means an imminent
275 * write when the file is truncated and actually unlinked
276 * on the filesystem.
277 */
278void drop_nlink(struct inode *inode)
279{
280 WARN_ON(inode->i_nlink == 0);
281 inode->__i_nlink--;
282 if (!inode->i_nlink)
283 atomic_long_inc(&inode->i_sb->s_remove_count);
284}
285EXPORT_SYMBOL(drop_nlink);
286
287/**
288 * clear_nlink - directly zero an inode's link count
289 * @inode: inode
290 *
291 * This is a low-level filesystem helper to replace any
292 * direct filesystem manipulation of i_nlink. See
293 * drop_nlink() for why we care about i_nlink hitting zero.
294 */
295void clear_nlink(struct inode *inode)
296{
297 if (inode->i_nlink) {
298 inode->__i_nlink = 0;
299 atomic_long_inc(&inode->i_sb->s_remove_count);
300 }
301}
302EXPORT_SYMBOL(clear_nlink);
303
304/**
305 * set_nlink - directly set an inode's link count
306 * @inode: inode
307 * @nlink: new nlink (should be non-zero)
308 *
309 * This is a low-level filesystem helper to replace any
310 * direct filesystem manipulation of i_nlink.
311 */
312void set_nlink(struct inode *inode, unsigned int nlink)
313{
314 if (!nlink) {
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100315 clear_nlink(inode);
316 } else {
317 /* Yes, some filesystems do change nlink from zero to one */
318 if (inode->i_nlink == 0)
319 atomic_long_dec(&inode->i_sb->s_remove_count);
320
321 inode->__i_nlink = nlink;
322 }
323}
324EXPORT_SYMBOL(set_nlink);
325
326/**
327 * inc_nlink - directly increment an inode's link count
328 * @inode: inode
329 *
330 * This is a low-level filesystem helper to replace any
331 * direct filesystem manipulation of i_nlink. Currently,
332 * it is only here for parity with dec_nlink().
333 */
334void inc_nlink(struct inode *inode)
335{
336 if (WARN_ON(inode->i_nlink == 0))
337 atomic_long_dec(&inode->i_sb->s_remove_count);
338
339 inode->__i_nlink++;
340}
341EXPORT_SYMBOL(inc_nlink);
342
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100343void address_space_init_once(struct address_space *mapping)
344{
345 memset(mapping, 0, sizeof(*mapping));
346 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
347 spin_lock_init(&mapping->tree_lock);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700348 mutex_init(&mapping->i_mmap_mutex);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100349 INIT_LIST_HEAD(&mapping->private_list);
350 spin_lock_init(&mapping->private_lock);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700351 mapping->i_mmap = RB_ROOT;
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100352 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100353}
354EXPORT_SYMBOL(address_space_init_once);
355
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356/*
357 * These are initializations that only need to be done
358 * once, because the fields are idempotent across use
359 * of the inode, so let the slab aware of that.
360 */
361void inode_init_once(struct inode *inode)
362{
363 memset(inode, 0, sizeof(*inode));
364 INIT_HLIST_NODE(&inode->i_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 INIT_LIST_HEAD(&inode->i_devices);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100366 INIT_LIST_HEAD(&inode->i_wb_list);
367 INIT_LIST_HEAD(&inode->i_lru);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100368 address_space_init_once(&inode->i_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 i_size_ordered_init(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400370#ifdef CONFIG_FSNOTIFY
Eric Parise61ce862009-12-17 21:24:24 -0500371 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
Eric Paris3be25f42009-05-21 17:01:26 -0400372#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374EXPORT_SYMBOL(inode_init_once);
375
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700376static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530378 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
Christoph Lametera35afb82007-05-16 22:10:57 -0700380 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381}
382
383/*
Dave Chinner250df6e2011-03-22 22:23:36 +1100384 * inode->i_lock must be held
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530386void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387{
Nick Piggin9e38d862010-10-23 06:55:17 -0400388 atomic_inc(&inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389}
390
Al Viro7de9c6ee2010-10-23 11:11:40 -0400391/*
392 * get additional reference to inode; caller must already hold one.
393 */
394void ihold(struct inode *inode)
395{
396 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
397}
398EXPORT_SYMBOL(ihold);
399
Nick Piggin9e38d862010-10-23 06:55:17 -0400400static void inode_lru_list_add(struct inode *inode)
401{
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000402 spin_lock(&inode->i_sb->s_inode_lru_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100403 if (list_empty(&inode->i_lru)) {
Dave Chinner98b745c2011-07-08 14:14:39 +1000404 list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
405 inode->i_sb->s_nr_inodes_unused++;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000406 this_cpu_inc(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400407 }
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000408 spin_unlock(&inode->i_sb->s_inode_lru_lock);
Nick Piggin9e38d862010-10-23 06:55:17 -0400409}
410
Jan Kara4eff96d2012-11-26 16:29:51 -0800411/*
412 * Add inode to LRU if needed (inode is unused and clean).
413 *
414 * Needs inode->i_lock held.
415 */
416void inode_add_lru(struct inode *inode)
417{
418 if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) &&
419 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
420 inode_lru_list_add(inode);
421}
422
423
Nick Piggin9e38d862010-10-23 06:55:17 -0400424static void inode_lru_list_del(struct inode *inode)
425{
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000426 spin_lock(&inode->i_sb->s_inode_lru_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100427 if (!list_empty(&inode->i_lru)) {
428 list_del_init(&inode->i_lru);
Dave Chinner98b745c2011-07-08 14:14:39 +1000429 inode->i_sb->s_nr_inodes_unused--;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000430 this_cpu_dec(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400431 }
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000432 spin_unlock(&inode->i_sb->s_inode_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
434
Christoph Hellwig646ec462010-10-23 07:15:32 -0400435/**
436 * inode_sb_list_add - add inode to the superblock list of inodes
437 * @inode: inode to add
438 */
439void inode_sb_list_add(struct inode *inode)
440{
Dave Chinner55fa6092011-03-22 22:23:40 +1100441 spin_lock(&inode_sb_list_lock);
442 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
443 spin_unlock(&inode_sb_list_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400444}
445EXPORT_SYMBOL_GPL(inode_sb_list_add);
446
Dave Chinner55fa6092011-03-22 22:23:40 +1100447static inline void inode_sb_list_del(struct inode *inode)
Christoph Hellwig646ec462010-10-23 07:15:32 -0400448{
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200449 if (!list_empty(&inode->i_sb_list)) {
450 spin_lock(&inode_sb_list_lock);
451 list_del_init(&inode->i_sb_list);
452 spin_unlock(&inode_sb_list_lock);
453 }
Christoph Hellwig646ec462010-10-23 07:15:32 -0400454}
455
Dave Chinner4c51acb2010-10-23 06:58:09 -0400456static unsigned long hash(struct super_block *sb, unsigned long hashval)
457{
458 unsigned long tmp;
459
460 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
461 L1_CACHE_BYTES;
Christoph Hellwig4b4563d2011-05-27 09:28:01 -0400462 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
463 return tmp & i_hash_mask;
Dave Chinner4c51acb2010-10-23 06:58:09 -0400464}
465
466/**
467 * __insert_inode_hash - hash an inode
468 * @inode: unhashed inode
469 * @hashval: unsigned long value used to locate this object in the
470 * inode_hashtable.
471 *
472 * Add an inode to the inode hash for this superblock.
473 */
474void __insert_inode_hash(struct inode *inode, unsigned long hashval)
475{
Christoph Hellwig646ec462010-10-23 07:15:32 -0400476 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
477
Dave Chinner67a23c42011-03-22 22:23:42 +1100478 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100479 spin_lock(&inode->i_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400480 hlist_add_head(&inode->i_hash, b);
Dave Chinner250df6e2011-03-22 22:23:36 +1100481 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100482 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400483}
484EXPORT_SYMBOL(__insert_inode_hash);
485
486/**
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200487 * __remove_inode_hash - remove an inode from the hash
Dave Chinner4c51acb2010-10-23 06:58:09 -0400488 * @inode: inode to unhash
489 *
490 * Remove an inode from the superblock.
491 */
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200492void __remove_inode_hash(struct inode *inode)
Dave Chinner4c51acb2010-10-23 06:58:09 -0400493{
Dave Chinner67a23c42011-03-22 22:23:42 +1100494 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100495 spin_lock(&inode->i_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400496 hlist_del_init(&inode->i_hash);
Dave Chinner250df6e2011-03-22 22:23:36 +1100497 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100498 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400499}
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200500EXPORT_SYMBOL(__remove_inode_hash);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400501
Jan Karadbd57682012-05-03 14:48:02 +0200502void clear_inode(struct inode *inode)
Al Virob0683aa2010-06-04 20:55:25 -0400503{
504 might_sleep();
Jan Kara08142572011-06-27 16:18:10 -0700505 /*
506 * We have to cycle tree_lock here because reclaim can be still in the
507 * process of removing the last page (in __delete_from_page_cache())
508 * and we must not free mapping under it.
509 */
510 spin_lock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400511 BUG_ON(inode->i_data.nrpages);
Jan Kara08142572011-06-27 16:18:10 -0700512 spin_unlock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400513 BUG_ON(!list_empty(&inode->i_data.private_list));
514 BUG_ON(!(inode->i_state & I_FREEING));
515 BUG_ON(inode->i_state & I_CLEAR);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100516 /* don't need i_lock here, no concurrent mods to i_state */
Al Virob0683aa2010-06-04 20:55:25 -0400517 inode->i_state = I_FREEING | I_CLEAR;
518}
Jan Karadbd57682012-05-03 14:48:02 +0200519EXPORT_SYMBOL(clear_inode);
Al Virob0683aa2010-06-04 20:55:25 -0400520
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100521/*
522 * Free the inode passed in, removing it from the lists it is still connected
523 * to. We remove any pages still attached to the inode and wait for any IO that
524 * is still in progress before finally destroying the inode.
525 *
526 * An inode must already be marked I_FREEING so that we avoid the inode being
527 * moved back onto lists if we race with other code that manipulates the lists
528 * (e.g. writeback_single_inode). The caller is responsible for setting this.
529 *
530 * An inode must already be removed from the LRU list before being evicted from
531 * the cache. This should occur atomically with setting the I_FREEING state
532 * flag, so no inodes here should ever be on the LRU when being evicted.
533 */
Al Viro644da592010-06-07 13:21:05 -0400534static void evict(struct inode *inode)
Al Virob4272d42010-06-04 19:33:20 -0400535{
536 const struct super_operations *op = inode->i_sb->s_op;
537
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100538 BUG_ON(!(inode->i_state & I_FREEING));
539 BUG_ON(!list_empty(&inode->i_lru));
540
Eric Dumazetb12362b2011-07-28 06:11:47 +0200541 if (!list_empty(&inode->i_wb_list))
542 inode_wb_list_del(inode);
543
Dave Chinner55fa6092011-03-22 22:23:40 +1100544 inode_sb_list_del(inode);
545
Jan Kara169ebd92012-05-03 14:48:03 +0200546 /*
547 * Wait for flusher thread to be done with the inode so that filesystem
548 * does not start destroying it while writeback is still running. Since
549 * the inode has I_FREEING set, flusher thread won't start new work on
550 * the inode. We just have to wait for running writeback to finish.
551 */
552 inode_wait_for_writeback(inode);
Jan Kara7994e6f2012-05-03 14:48:01 +0200553
Al Virobe7ce412010-06-04 19:40:39 -0400554 if (op->evict_inode) {
555 op->evict_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400556 } else {
557 if (inode->i_data.nrpages)
558 truncate_inode_pages(&inode->i_data, 0);
Jan Karadbd57682012-05-03 14:48:02 +0200559 clear_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400560 }
Al Viro661074e2010-06-04 20:19:55 -0400561 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
562 bd_forget(inode);
563 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
564 cd_forget(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100565
566 remove_inode_hash(inode);
567
568 spin_lock(&inode->i_lock);
569 wake_up_bit(&inode->i_state, __I_NEW);
570 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
571 spin_unlock(&inode->i_lock);
572
573 destroy_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400574}
575
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576/*
577 * dispose_list - dispose of the contents of a local list
578 * @head: the head of the list to free
579 *
580 * Dispose-list gets a local list with local inodes in it, so it doesn't
581 * need to worry about list corruption and SMP locks.
582 */
583static void dispose_list(struct list_head *head)
584{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 while (!list_empty(head)) {
586 struct inode *inode;
587
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100588 inode = list_first_entry(head, struct inode, i_lru);
589 list_del_init(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
Al Viro644da592010-06-07 13:21:05 -0400591 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593}
594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595/**
Al Viro63997e92010-10-25 20:49:35 -0400596 * evict_inodes - evict all evictable inodes for a superblock
597 * @sb: superblock to operate on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 *
Al Viro63997e92010-10-25 20:49:35 -0400599 * Make sure that no inodes with zero refcount are retained. This is
600 * called by superblock shutdown after having MS_ACTIVE flag removed,
601 * so any inode reaching zero refcount during or after that call will
602 * be immediately evicted.
603 */
604void evict_inodes(struct super_block *sb)
605{
606 struct inode *inode, *next;
607 LIST_HEAD(dispose);
608
Dave Chinner55fa6092011-03-22 22:23:40 +1100609 spin_lock(&inode_sb_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400610 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
611 if (atomic_read(&inode->i_count))
612 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100613
614 spin_lock(&inode->i_lock);
615 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
616 spin_unlock(&inode->i_lock);
Al Viro63997e92010-10-25 20:49:35 -0400617 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100618 }
Al Viro63997e92010-10-25 20:49:35 -0400619
620 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100621 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100622 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100623 list_add(&inode->i_lru, &dispose);
Al Viro63997e92010-10-25 20:49:35 -0400624 }
Dave Chinner55fa6092011-03-22 22:23:40 +1100625 spin_unlock(&inode_sb_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400626
627 dispose_list(&dispose);
Al Viro63997e92010-10-25 20:49:35 -0400628}
629
630/**
Christoph Hellwiga0318782010-10-24 19:40:33 +0200631 * invalidate_inodes - attempt to free all inodes on a superblock
632 * @sb: superblock to operate on
NeilBrown93b270f2011-02-24 17:25:47 +1100633 * @kill_dirty: flag to guide handling of dirty inodes
Christoph Hellwiga0318782010-10-24 19:40:33 +0200634 *
635 * Attempts to free all inodes for a given superblock. If there were any
636 * busy inodes return a non-zero value, else zero.
NeilBrown93b270f2011-02-24 17:25:47 +1100637 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
638 * them as busy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 */
NeilBrown93b270f2011-02-24 17:25:47 +1100640int invalidate_inodes(struct super_block *sb, bool kill_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641{
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400642 int busy = 0;
Christoph Hellwiga0318782010-10-24 19:40:33 +0200643 struct inode *inode, *next;
644 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Dave Chinner55fa6092011-03-22 22:23:40 +1100646 spin_lock(&inode_sb_list_lock);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200647 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100648 spin_lock(&inode->i_lock);
649 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
650 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100652 }
NeilBrown93b270f2011-02-24 17:25:47 +1100653 if (inode->i_state & I_DIRTY && !kill_dirty) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100654 spin_unlock(&inode->i_lock);
NeilBrown93b270f2011-02-24 17:25:47 +1100655 busy = 1;
656 continue;
657 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200658 if (atomic_read(&inode->i_count)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100659 spin_unlock(&inode->i_lock);
Christoph Hellwig99a38912010-10-23 19:07:20 +0200660 busy = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 continue;
662 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200663
Christoph Hellwig99a38912010-10-23 19:07:20 +0200664 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100665 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100666 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100667 list_add(&inode->i_lru, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 }
Dave Chinner55fa6092011-03-22 22:23:40 +1100669 spin_unlock(&inode_sb_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
Christoph Hellwiga0318782010-10-24 19:40:33 +0200671 dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
673 return busy;
674}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
676static int can_unuse(struct inode *inode)
677{
Nick Piggin9e38d862010-10-23 06:55:17 -0400678 if (inode->i_state & ~I_REFERENCED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 return 0;
680 if (inode_has_buffers(inode))
681 return 0;
682 if (atomic_read(&inode->i_count))
683 return 0;
684 if (inode->i_data.nrpages)
685 return 0;
686 return 1;
687}
688
689/*
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000690 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
691 * This is called from the superblock shrinker function with a number of inodes
692 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
693 * then are freed outside inode_lock by dispose_list().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 *
695 * Any inodes which are pinned purely because of attached pagecache have their
Nick Piggin9e38d862010-10-23 06:55:17 -0400696 * pagecache removed. If the inode has metadata buffers attached to
697 * mapping->private_list then try to remove them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 *
Nick Piggin9e38d862010-10-23 06:55:17 -0400699 * If the inode has the I_REFERENCED flag set, then it means that it has been
700 * used recently - the flag is set in iput_final(). When we encounter such an
701 * inode, clear the flag and move it to the back of the LRU so it gets another
702 * pass through the LRU before it gets reclaimed. This is necessary because of
703 * the fact we are doing lazy LRU updates to minimise lock contention so the
704 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
705 * with this flag set because they are the inodes that are out of order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 */
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000707void prune_icache_sb(struct super_block *sb, int nr_to_scan)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708{
709 LIST_HEAD(freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 int nr_scanned;
711 unsigned long reap = 0;
712
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000713 spin_lock(&sb->s_inode_lru_lock);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000714 for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 struct inode *inode;
716
Dave Chinner98b745c2011-07-08 14:14:39 +1000717 if (list_empty(&sb->s_inode_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 break;
719
Dave Chinner98b745c2011-07-08 14:14:39 +1000720 inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
Nick Piggin9e38d862010-10-23 06:55:17 -0400722 /*
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000723 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
Dave Chinner02afc412011-03-22 22:23:38 +1100724 * so use a trylock. If we fail to get the lock, just move the
725 * inode to the back of the list so we don't spin on it.
726 */
727 if (!spin_trylock(&inode->i_lock)) {
Christoph Hellwig62a3dde2011-10-28 10:03:41 +0200728 list_move_tail(&inode->i_lru, &sb->s_inode_lru);
Dave Chinner02afc412011-03-22 22:23:38 +1100729 continue;
730 }
731
732 /*
Nick Piggin9e38d862010-10-23 06:55:17 -0400733 * Referenced or dirty inodes are still in use. Give them
734 * another pass through the LRU as we canot reclaim them now.
735 */
736 if (atomic_read(&inode->i_count) ||
737 (inode->i_state & ~I_REFERENCED)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100738 list_del_init(&inode->i_lru);
Dave Chinnerf283c862011-03-22 22:23:39 +1100739 spin_unlock(&inode->i_lock);
Dave Chinner98b745c2011-07-08 14:14:39 +1000740 sb->s_nr_inodes_unused--;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000741 this_cpu_dec(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400742 continue;
743 }
744
745 /* recently referenced inodes get one more pass */
746 if (inode->i_state & I_REFERENCED) {
Nick Piggin9e38d862010-10-23 06:55:17 -0400747 inode->i_state &= ~I_REFERENCED;
Dave Chinner98b745c2011-07-08 14:14:39 +1000748 list_move(&inode->i_lru, &sb->s_inode_lru);
Dave Chinnerf283c862011-03-22 22:23:39 +1100749 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 continue;
751 }
752 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
753 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100754 spin_unlock(&inode->i_lock);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000755 spin_unlock(&sb->s_inode_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800757 reap += invalidate_mapping_pages(&inode->i_data,
758 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 iput(inode);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000760 spin_lock(&sb->s_inode_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
Dave Chinner98b745c2011-07-08 14:14:39 +1000762 if (inode != list_entry(sb->s_inode_lru.next,
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100763 struct inode, i_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 continue; /* wrong inode or list_empty */
Dave Chinner02afc412011-03-22 22:23:38 +1100765 /* avoid lock inversions with trylock */
766 if (!spin_trylock(&inode->i_lock))
767 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100768 if (!can_unuse(inode)) {
769 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100771 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 }
Nick Piggin7ef0d732009-03-12 14:31:38 -0700773 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 inode->i_state |= I_FREEING;
Dave Chinner250df6e2011-03-22 22:23:36 +1100775 spin_unlock(&inode->i_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100776
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100777 list_move(&inode->i_lru, &freeable);
Dave Chinner98b745c2011-07-08 14:14:39 +1000778 sb->s_nr_inodes_unused--;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000779 this_cpu_dec(nr_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 }
Christoph Lameterf8891e52006-06-30 01:55:45 -0700781 if (current_is_kswapd())
782 __count_vm_events(KSWAPD_INODESTEAL, reap);
783 else
784 __count_vm_events(PGINODESTEAL, reap);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000785 spin_unlock(&sb->s_inode_lru_lock);
Konstantin Khlebnikov5f8aefd2012-01-10 15:07:18 -0800786 if (current->reclaim_state)
787 current->reclaim_state->reclaimed_slab += reap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
789 dispose_list(&freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790}
791
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792static void __wait_on_freeing_inode(struct inode *inode);
793/*
794 * Called with the inode lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530796static struct inode *find_inode(struct super_block *sb,
797 struct hlist_head *head,
798 int (*test)(struct inode *, void *),
799 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800{
801 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530802 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
804repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700805 hlist_for_each_entry(inode, node, head, i_hash) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100806 spin_lock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100807 if (inode->i_sb != sb) {
808 spin_unlock(&inode->i_lock);
809 continue;
810 }
811 if (!test(inode, data)) {
812 spin_unlock(&inode->i_lock);
813 continue;
814 }
Al Viroa4ffdde2010-06-02 17:38:30 -0400815 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 __wait_on_freeing_inode(inode);
817 goto repeat;
818 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400819 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100820 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400821 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400823 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824}
825
826/*
827 * find_inode_fast is the fast path version of find_inode, see the comment at
828 * iget_locked for details.
829 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530830static struct inode *find_inode_fast(struct super_block *sb,
831 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832{
833 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530834 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
836repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700837 hlist_for_each_entry(inode, node, head, i_hash) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100838 spin_lock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100839 if (inode->i_ino != ino) {
840 spin_unlock(&inode->i_lock);
841 continue;
842 }
843 if (inode->i_sb != sb) {
844 spin_unlock(&inode->i_lock);
845 continue;
846 }
Al Viroa4ffdde2010-06-02 17:38:30 -0400847 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 __wait_on_freeing_inode(inode);
849 goto repeat;
850 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400851 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100852 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400853 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400855 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856}
857
Eric Dumazetf991bd22010-10-23 11:18:01 -0400858/*
859 * Each cpu owns a range of LAST_INO_BATCH numbers.
860 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
861 * to renew the exhausted range.
David Chinner8290c352008-10-30 17:35:24 +1100862 *
Eric Dumazetf991bd22010-10-23 11:18:01 -0400863 * This does not significantly increase overflow rate because every CPU can
864 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
865 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
866 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
867 * overflow rate by 2x, which does not seem too significant.
868 *
869 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
870 * error if st_ino won't fit in target struct field. Use 32bit counter
871 * here to attempt to avoid that.
David Chinner8290c352008-10-30 17:35:24 +1100872 */
Eric Dumazetf991bd22010-10-23 11:18:01 -0400873#define LAST_INO_BATCH 1024
874static DEFINE_PER_CPU(unsigned int, last_ino);
David Chinner8290c352008-10-30 17:35:24 +1100875
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400876unsigned int get_next_ino(void)
Eric Dumazetf991bd22010-10-23 11:18:01 -0400877{
878 unsigned int *p = &get_cpu_var(last_ino);
879 unsigned int res = *p;
880
881#ifdef CONFIG_SMP
882 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
883 static atomic_t shared_last_ino;
884 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
885
886 res = next - LAST_INO_BATCH;
887 }
888#endif
889
890 *p = ++res;
891 put_cpu_var(last_ino);
892 return res;
David Chinner8290c352008-10-30 17:35:24 +1100893}
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400894EXPORT_SYMBOL(get_next_ino);
David Chinner8290c352008-10-30 17:35:24 +1100895
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896/**
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200897 * new_inode_pseudo - obtain an inode
898 * @sb: superblock
899 *
900 * Allocates a new inode for given superblock.
901 * Inode wont be chained in superblock s_inodes list
902 * This means :
903 * - fs can't be unmount
904 * - quotas, fsnotify, writeback can't work
905 */
906struct inode *new_inode_pseudo(struct super_block *sb)
907{
908 struct inode *inode = alloc_inode(sb);
909
910 if (inode) {
911 spin_lock(&inode->i_lock);
912 inode->i_state = 0;
913 spin_unlock(&inode->i_lock);
914 INIT_LIST_HEAD(&inode->i_sb_list);
915 }
916 return inode;
917}
918
919/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 * new_inode - obtain an inode
921 * @sb: superblock
922 *
Mel Gorman769848c2007-07-17 04:03:05 -0700923 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800924 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700925 * If HIGHMEM pages are unsuitable or it is known that pages allocated
926 * for the page cache are not reclaimable or migratable,
927 * mapping_set_gfp_mask() must be called with suitable flags on the
928 * newly created inode's mapping
929 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 */
931struct inode *new_inode(struct super_block *sb)
932{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530933 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
Dave Chinner55fa6092011-03-22 22:23:40 +1100935 spin_lock_prefetch(&inode_sb_list_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530936
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200937 inode = new_inode_pseudo(sb);
938 if (inode)
Dave Chinner55fa6092011-03-22 22:23:40 +1100939 inode_sb_list_add(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 return inode;
941}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942EXPORT_SYMBOL(new_inode);
943
Peter Zijlstra14358e62007-10-14 01:38:33 +0200944#ifdef CONFIG_DEBUG_LOCK_ALLOC
Josh Boyere096d0c2011-08-25 07:48:12 -0400945void lockdep_annotate_inode_mutex_key(struct inode *inode)
946{
Namhyung Kima3314a02010-10-11 22:38:00 +0900947 if (S_ISDIR(inode->i_mode)) {
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200948 struct file_system_type *type = inode->i_sb->s_type;
949
Jan Kara9a7aa122009-06-04 15:26:49 +0200950 /* Set new key only if filesystem hasn't already changed it */
Tyler Hicks978d6d82011-12-12 10:02:30 -0600951 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
Jan Kara9a7aa122009-06-04 15:26:49 +0200952 /*
953 * ensure nobody is actually holding i_mutex
954 */
955 mutex_destroy(&inode->i_mutex);
956 mutex_init(&inode->i_mutex);
957 lockdep_set_class(&inode->i_mutex,
958 &type->i_mutex_dir_key);
959 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200960 }
Josh Boyere096d0c2011-08-25 07:48:12 -0400961}
962EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200963#endif
Josh Boyere096d0c2011-08-25 07:48:12 -0400964
965/**
966 * unlock_new_inode - clear the I_NEW state and wake up any waiters
967 * @inode: new inode to unlock
968 *
969 * Called when the inode is fully initialised to clear the new state of the
970 * inode and wake up anyone waiting for the inode to finish initialisation.
971 */
972void unlock_new_inode(struct inode *inode)
973{
974 lockdep_annotate_inode_mutex_key(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100975 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100976 WARN_ON(!(inode->i_state & I_NEW));
977 inode->i_state &= ~I_NEW;
Al Viro310fa7a2012-03-10 17:07:28 -0500978 smp_mb();
Dave Chinner250df6e2011-03-22 22:23:36 +1100979 wake_up_bit(&inode->i_state, __I_NEW);
980 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982EXPORT_SYMBOL(unlock_new_inode);
983
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400984/**
985 * iget5_locked - obtain an inode from a mounted file system
986 * @sb: super block of file system
987 * @hashval: hash value (usually inode number) to get
988 * @test: callback used for comparisons between inodes
989 * @set: callback used to initialize a new struct inode
990 * @data: opaque data pointer to pass to @test and @set
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400992 * Search for the inode specified by @hashval and @data in the inode cache,
993 * and if present it is return it with an increased reference count. This is
994 * a generalized version of iget_locked() for file systems where the inode
995 * number is not sufficient for unique identification of an inode.
996 *
997 * If the inode is not in cache, allocate a new inode and return it locked,
998 * hashed, and with the I_NEW flag set. The file system gets to fill it in
999 * before unlocking it via unlock_new_inode().
1000 *
1001 * Note both @test and @set are called with the inode_hash_lock held, so can't
1002 * sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001004struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1005 int (*test)(struct inode *, void *),
1006 int (*set)(struct inode *, void *), void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001008 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301009 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001011 spin_lock(&inode_hash_lock);
1012 inode = find_inode(sb, head, test, data);
1013 spin_unlock(&inode_hash_lock);
1014
1015 if (inode) {
1016 wait_on_inode(inode);
1017 return inode;
1018 }
1019
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 inode = alloc_inode(sb);
1021 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301022 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
Dave Chinner67a23c42011-03-22 22:23:42 +11001024 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 /* We released the lock, so.. */
1026 old = find_inode(sb, head, test, data);
1027 if (!old) {
1028 if (set(inode, data))
1029 goto set_failed;
1030
Dave Chinner250df6e2011-03-22 22:23:36 +11001031 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001032 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001033 hlist_add_head(&inode->i_hash, head);
1034 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001035 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001036 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
1038 /* Return the locked inode with I_NEW set, the
1039 * caller is responsible for filling in the contents
1040 */
1041 return inode;
1042 }
1043
1044 /*
1045 * Uhhuh, somebody else created the same inode under
1046 * us. Use the old inode instead of the one we just
1047 * allocated.
1048 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001049 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 destroy_inode(inode);
1051 inode = old;
1052 wait_on_inode(inode);
1053 }
1054 return inode;
1055
1056set_failed:
Dave Chinner67a23c42011-03-22 22:23:42 +11001057 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 destroy_inode(inode);
1059 return NULL;
1060}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001061EXPORT_SYMBOL(iget5_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001063/**
1064 * iget_locked - obtain an inode from a mounted file system
1065 * @sb: super block of file system
1066 * @ino: inode number to get
1067 *
1068 * Search for the inode specified by @ino in the inode cache and if present
1069 * return it with an increased reference count. This is for file systems
1070 * where the inode number is sufficient for unique identification of an inode.
1071 *
1072 * If the inode is not in cache, allocate a new inode and return it locked,
1073 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1074 * before unlocking it via unlock_new_inode().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001076struct inode *iget_locked(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001078 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301079 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001081 spin_lock(&inode_hash_lock);
1082 inode = find_inode_fast(sb, head, ino);
1083 spin_unlock(&inode_hash_lock);
1084 if (inode) {
1085 wait_on_inode(inode);
1086 return inode;
1087 }
1088
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 inode = alloc_inode(sb);
1090 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301091 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Dave Chinner67a23c42011-03-22 22:23:42 +11001093 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 /* We released the lock, so.. */
1095 old = find_inode_fast(sb, head, ino);
1096 if (!old) {
1097 inode->i_ino = ino;
Dave Chinner250df6e2011-03-22 22:23:36 +11001098 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001099 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001100 hlist_add_head(&inode->i_hash, head);
1101 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001102 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001103 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104
1105 /* Return the locked inode with I_NEW set, the
1106 * caller is responsible for filling in the contents
1107 */
1108 return inode;
1109 }
1110
1111 /*
1112 * Uhhuh, somebody else created the same inode under
1113 * us. Use the old inode instead of the one we just
1114 * allocated.
1115 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001116 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 destroy_inode(inode);
1118 inode = old;
1119 wait_on_inode(inode);
1120 }
1121 return inode;
1122}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001123EXPORT_SYMBOL(iget_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001125/*
1126 * search the inode cache for a matching inode number.
1127 * If we find one, then the inode number we are trying to
1128 * allocate is not unique and so we should not use it.
1129 *
1130 * Returns 1 if the inode number is unique, 0 if it is not.
1131 */
1132static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1133{
1134 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1135 struct hlist_node *node;
1136 struct inode *inode;
1137
Dave Chinner67a23c42011-03-22 22:23:42 +11001138 spin_lock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001139 hlist_for_each_entry(inode, node, b, i_hash) {
Dave Chinner67a23c42011-03-22 22:23:42 +11001140 if (inode->i_ino == ino && inode->i_sb == sb) {
1141 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001142 return 0;
Dave Chinner67a23c42011-03-22 22:23:42 +11001143 }
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001144 }
Dave Chinner67a23c42011-03-22 22:23:42 +11001145 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001146
1147 return 1;
1148}
1149
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150/**
1151 * iunique - get a unique inode number
1152 * @sb: superblock
1153 * @max_reserved: highest reserved inode number
1154 *
1155 * Obtain an inode number that is unique on the system for a given
1156 * superblock. This is used by file systems that have no natural
1157 * permanent inode numbering system. An inode number is returned that
1158 * is higher than the reserved limit but unique.
1159 *
1160 * BUGS:
1161 * With a large number of inodes live on the file system this function
1162 * currently becomes quite slow.
1163 */
1164ino_t iunique(struct super_block *sb, ino_t max_reserved)
1165{
Jeff Layton866b04f2007-05-08 00:32:29 -07001166 /*
1167 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1168 * error if st_ino won't fit in target struct field. Use 32bit counter
1169 * here to attempt to avoid that.
1170 */
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001171 static DEFINE_SPINLOCK(iunique_lock);
Jeff Layton866b04f2007-05-08 00:32:29 -07001172 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001175 spin_lock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001176 do {
1177 if (counter <= max_reserved)
1178 counter = max_reserved + 1;
1179 res = counter++;
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001180 } while (!test_inode_iunique(sb, res));
1181 spin_unlock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001182
1183 return res;
1184}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185EXPORT_SYMBOL(iunique);
1186
1187struct inode *igrab(struct inode *inode)
1188{
Dave Chinner250df6e2011-03-22 22:23:36 +11001189 spin_lock(&inode->i_lock);
1190 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001192 spin_unlock(&inode->i_lock);
1193 } else {
1194 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 /*
1196 * Handle the case where s_op->clear_inode is not been
1197 * called yet, and somebody is calling igrab
1198 * while the inode is getting freed.
1199 */
1200 inode = NULL;
Dave Chinner250df6e2011-03-22 22:23:36 +11001201 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 return inode;
1203}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204EXPORT_SYMBOL(igrab);
1205
1206/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001207 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 * @sb: super block of file system to search
1209 * @hashval: hash value (usually inode number) to search for
1210 * @test: callback used for comparisons between inodes
1211 * @data: opaque data pointer to pass to @test
1212 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001213 * Search for the inode specified by @hashval and @data in the inode cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 * If the inode is in the cache, the inode is returned with an incremented
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001215 * reference count.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001216 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001217 * Note: I_NEW is not waited upon so you have to be very careful what you do
1218 * with the returned inode. You probably should be using ilookup5() instead.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001219 *
Randy Dunlapb6d0ad62011-03-26 13:27:47 -07001220 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001221 */
1222struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1223 int (*test)(struct inode *, void *), void *data)
1224{
1225 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001226 struct inode *inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001227
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001228 spin_lock(&inode_hash_lock);
1229 inode = find_inode(sb, head, test, data);
1230 spin_unlock(&inode_hash_lock);
1231
1232 return inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001233}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001234EXPORT_SYMBOL(ilookup5_nowait);
1235
1236/**
1237 * ilookup5 - search for an inode in the inode cache
1238 * @sb: super block of file system to search
1239 * @hashval: hash value (usually inode number) to search for
1240 * @test: callback used for comparisons between inodes
1241 * @data: opaque data pointer to pass to @test
1242 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001243 * Search for the inode specified by @hashval and @data in the inode cache,
1244 * and if the inode is in the cache, return the inode with an incremented
1245 * reference count. Waits on I_NEW before returning the inode.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001246 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001248 * This is a generalized version of ilookup() for file systems where the
1249 * inode number is not sufficient for unique identification of an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001251 * Note: @test is called with the inode_hash_lock held, so can't sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 */
1253struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1254 int (*test)(struct inode *, void *), void *data)
1255{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001256 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001258 if (inode)
1259 wait_on_inode(inode);
1260 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262EXPORT_SYMBOL(ilookup5);
1263
1264/**
1265 * ilookup - search for an inode in the inode cache
1266 * @sb: super block of file system to search
1267 * @ino: inode number to search for
1268 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001269 * Search for the inode @ino in the inode cache, and if the inode is in the
1270 * cache, the inode is returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 */
1272struct inode *ilookup(struct super_block *sb, unsigned long ino)
1273{
1274 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001275 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001277 spin_lock(&inode_hash_lock);
1278 inode = find_inode_fast(sb, head, ino);
1279 spin_unlock(&inode_hash_lock);
1280
1281 if (inode)
1282 wait_on_inode(inode);
1283 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285EXPORT_SYMBOL(ilookup);
1286
Al Viro261bca82008-12-30 01:48:21 -05001287int insert_inode_locked(struct inode *inode)
1288{
1289 struct super_block *sb = inode->i_sb;
1290 ino_t ino = inode->i_ino;
1291 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001292
Al Viro261bca82008-12-30 01:48:21 -05001293 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001294 struct hlist_node *node;
1295 struct inode *old = NULL;
Dave Chinner67a23c42011-03-22 22:23:42 +11001296 spin_lock(&inode_hash_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001297 hlist_for_each_entry(old, node, head, i_hash) {
1298 if (old->i_ino != ino)
1299 continue;
1300 if (old->i_sb != sb)
1301 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001302 spin_lock(&old->i_lock);
1303 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1304 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001305 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001306 }
Al Viro72a43d62009-05-13 19:13:40 +01001307 break;
1308 }
1309 if (likely(!node)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001310 spin_lock(&inode->i_lock);
1311 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001312 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001313 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001314 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001315 return 0;
1316 }
1317 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001318 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001319 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001320 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001321 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001322 iput(old);
1323 return -EBUSY;
1324 }
1325 iput(old);
1326 }
1327}
Al Viro261bca82008-12-30 01:48:21 -05001328EXPORT_SYMBOL(insert_inode_locked);
1329
1330int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1331 int (*test)(struct inode *, void *), void *data)
1332{
1333 struct super_block *sb = inode->i_sb;
1334 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001335
Al Viro261bca82008-12-30 01:48:21 -05001336 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001337 struct hlist_node *node;
1338 struct inode *old = NULL;
1339
Dave Chinner67a23c42011-03-22 22:23:42 +11001340 spin_lock(&inode_hash_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001341 hlist_for_each_entry(old, node, head, i_hash) {
1342 if (old->i_sb != sb)
1343 continue;
1344 if (!test(old, data))
1345 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001346 spin_lock(&old->i_lock);
1347 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1348 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001349 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001350 }
Al Viro72a43d62009-05-13 19:13:40 +01001351 break;
1352 }
1353 if (likely(!node)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001354 spin_lock(&inode->i_lock);
1355 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001356 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001357 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001358 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001359 return 0;
1360 }
1361 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001362 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001363 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001364 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001365 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001366 iput(old);
1367 return -EBUSY;
1368 }
1369 iput(old);
1370 }
1371}
Al Viro261bca82008-12-30 01:48:21 -05001372EXPORT_SYMBOL(insert_inode_locked4);
1373
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
Al Viro45321ac2010-06-07 13:43:19 -04001375int generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376{
Al Viro45321ac2010-06-07 13:43:19 -04001377 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379EXPORT_SYMBOL(generic_delete_inode);
1380
Al Viro45321ac2010-06-07 13:43:19 -04001381/*
Al Viro45321ac2010-06-07 13:43:19 -04001382 * Called when we're dropping the last reference
1383 * to an inode.
1384 *
1385 * Call the FS "drop_inode()" function, defaulting to
1386 * the legacy UNIX filesystem behaviour. If it tells
1387 * us to evict inode, do so. Otherwise, retain inode
1388 * in cache if fs is alive, sync and evict if fs is
1389 * shutting down.
1390 */
1391static void iput_final(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392{
1393 struct super_block *sb = inode->i_sb;
Al Viro45321ac2010-06-07 13:43:19 -04001394 const struct super_operations *op = inode->i_sb->s_op;
1395 int drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396
Dave Chinner250df6e2011-03-22 22:23:36 +11001397 WARN_ON(inode->i_state & I_NEW);
1398
Al Viroe7f59092011-07-07 15:45:59 -04001399 if (op->drop_inode)
Al Viro45321ac2010-06-07 13:43:19 -04001400 drop = op->drop_inode(inode);
1401 else
1402 drop = generic_drop_inode(inode);
1403
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001404 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1405 inode->i_state |= I_REFERENCED;
Jan Kara4eff96d2012-11-26 16:29:51 -08001406 inode_add_lru(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001407 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001408 return;
1409 }
1410
Al Viro45321ac2010-06-07 13:43:19 -04001411 if (!drop) {
Alexander Viro991114c2005-06-23 00:09:01 -07001412 inode->i_state |= I_WILL_FREE;
Dave Chinner250df6e2011-03-22 22:23:36 +11001413 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 write_inode_now(inode, 1);
Dave Chinner250df6e2011-03-22 22:23:36 +11001415 spin_lock(&inode->i_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001416 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001417 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001419
Alexander Viro991114c2005-06-23 00:09:01 -07001420 inode->i_state |= I_FREEING;
Eric Dumazetc4ae0c62011-07-28 06:55:13 +02001421 if (!list_empty(&inode->i_lru))
1422 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001423 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001424
1425 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426}
1427
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301429 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 * @inode: inode to put
1431 *
1432 * Puts an inode, dropping its usage count. If the inode use count hits
1433 * zero, the inode is then freed and may also be destroyed.
1434 *
1435 * Consequently, iput() can sleep.
1436 */
1437void iput(struct inode *inode)
1438{
1439 if (inode) {
Al Viroa4ffdde2010-06-02 17:38:30 -04001440 BUG_ON(inode->i_state & I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
Dave Chinnerf283c862011-03-22 22:23:39 +11001442 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 iput_final(inode);
1444 }
1445}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446EXPORT_SYMBOL(iput);
1447
1448/**
1449 * bmap - find a block number in a file
1450 * @inode: inode of file
1451 * @block: block to find
1452 *
1453 * Returns the block number on the device holding the inode that
1454 * is the disk block number for the block of the file requested.
1455 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301456 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 * file.
1458 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301459sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460{
1461 sector_t res = 0;
1462 if (inode->i_mapping->a_ops->bmap)
1463 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1464 return res;
1465}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466EXPORT_SYMBOL(bmap);
1467
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001468/*
1469 * With relative atime, only update atime if the previous atime is
1470 * earlier than either the ctime or mtime or if at least a day has
1471 * passed since the last atime update.
1472 */
1473static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1474 struct timespec now)
1475{
1476
1477 if (!(mnt->mnt_flags & MNT_RELATIME))
1478 return 1;
1479 /*
1480 * Is mtime younger than atime? If yes, update atime:
1481 */
1482 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1483 return 1;
1484 /*
1485 * Is ctime younger than atime? If yes, update atime:
1486 */
1487 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1488 return 1;
1489
1490 /*
1491 * Is the previous atime value older than a day? If yes,
1492 * update atime:
1493 */
1494 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1495 return 1;
1496 /*
1497 * Good, we can skip the atime update:
1498 */
1499 return 0;
1500}
1501
Josef Bacikc3b2da32012-03-26 09:59:21 -04001502/*
1503 * This does the actual work of updating an inodes time or version. Must have
1504 * had called mnt_want_write() before calling this.
1505 */
1506static int update_time(struct inode *inode, struct timespec *time, int flags)
1507{
1508 if (inode->i_op->update_time)
1509 return inode->i_op->update_time(inode, time, flags);
1510
1511 if (flags & S_ATIME)
1512 inode->i_atime = *time;
1513 if (flags & S_VERSION)
1514 inode_inc_iversion(inode);
1515 if (flags & S_CTIME)
1516 inode->i_ctime = *time;
1517 if (flags & S_MTIME)
1518 inode->i_mtime = *time;
1519 mark_inode_dirty_sync(inode);
1520 return 0;
1521}
1522
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001524 * touch_atime - update the access time
Randy Dunlap185553b2012-04-17 17:03:25 -07001525 * @path: the &struct path to update
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 *
1527 * Update the accessed time on an inode and mark it for writeback.
1528 * This function automatically handles read only file systems and media,
1529 * as well as the "noatime" flag and inode specific "noatime" markers.
1530 */
Al Viro68ac1232012-03-15 08:21:57 -04001531void touch_atime(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532{
Al Viro68ac1232012-03-15 08:21:57 -04001533 struct vfsmount *mnt = path->mnt;
1534 struct inode *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 struct timespec now;
1536
Andrew Mortonb2276132006-12-13 00:34:33 -08001537 if (inode->i_flags & S_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001538 return;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001539 if (IS_NOATIME(inode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001540 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001541 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001542 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001543
Dave Hansencdb70f32008-02-15 14:37:41 -08001544 if (mnt->mnt_flags & MNT_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001545 return;
Dave Hansencdb70f32008-02-15 14:37:41 -08001546 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001547 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 now = current_fs_time(inode->i_sb);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001550
1551 if (!relatime_need_update(mnt, inode, now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001552 return;
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001553
Valerie Henson47ae32d2006-12-13 00:34:34 -08001554 if (timespec_equal(&inode->i_atime, &now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001555 return;
1556
Jan Kara5d37e9e2012-06-12 16:20:36 +02001557 if (!sb_start_write_trylock(inode->i_sb))
Andi Kleenb12536c2009-09-18 13:05:47 -07001558 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001559
Jan Kara5d37e9e2012-06-12 16:20:36 +02001560 if (__mnt_want_write(mnt))
1561 goto skip_update;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001562 /*
1563 * File systems can error out when updating inodes if they need to
1564 * allocate new space to modify an inode (such is the case for
1565 * Btrfs), but since we touch atime while walking down the path we
1566 * really don't care if we failed to update the atime of the file,
1567 * so just ignore the return value.
Alexander Block2bc5565282012-06-15 09:49:33 +02001568 * We may also fail on filesystems that have the ability to make parts
1569 * of the fs read only, e.g. subvolumes in Btrfs.
Josef Bacikc3b2da32012-03-26 09:59:21 -04001570 */
1571 update_time(inode, &now, S_ATIME);
Jan Kara5d37e9e2012-06-12 16:20:36 +02001572 __mnt_drop_write(mnt);
1573skip_update:
1574 sb_end_write(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001576EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
Cong Wang3ed37642012-05-15 14:57:33 +08001578/*
1579 * The logic we want is
1580 *
1581 * if suid or (sgid and xgrp)
1582 * remove privs
1583 */
1584int should_remove_suid(struct dentry *dentry)
1585{
1586 umode_t mode = dentry->d_inode->i_mode;
1587 int kill = 0;
1588
1589 /* suid always must be killed */
1590 if (unlikely(mode & S_ISUID))
1591 kill = ATTR_KILL_SUID;
1592
1593 /*
1594 * sgid without any exec bits is just a mandatory locking mark; leave
1595 * it alone. If some exec bits are set, it's a real sgid; kill it.
1596 */
1597 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1598 kill |= ATTR_KILL_SGID;
1599
1600 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1601 return kill;
1602
1603 return 0;
1604}
1605EXPORT_SYMBOL(should_remove_suid);
1606
1607static int __remove_suid(struct dentry *dentry, int kill)
1608{
1609 struct iattr newattrs;
1610
1611 newattrs.ia_valid = ATTR_FORCE | kill;
1612 return notify_change(dentry, &newattrs);
1613}
1614
1615int file_remove_suid(struct file *file)
1616{
1617 struct dentry *dentry = file->f_path.dentry;
1618 struct inode *inode = dentry->d_inode;
1619 int killsuid;
1620 int killpriv;
1621 int error = 0;
1622
1623 /* Fast path for nothing security related */
1624 if (IS_NOSEC(inode))
1625 return 0;
1626
1627 killsuid = should_remove_suid(dentry);
1628 killpriv = security_inode_need_killpriv(dentry);
1629
1630 if (killpriv < 0)
1631 return killpriv;
1632 if (killpriv)
1633 error = security_inode_killpriv(dentry);
1634 if (!error && killsuid)
1635 error = __remove_suid(dentry, killsuid);
1636 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1637 inode->i_flags |= S_NOSEC;
1638
1639 return error;
1640}
1641EXPORT_SYMBOL(file_remove_suid);
1642
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001644 * file_update_time - update mtime and ctime time
1645 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001647 * Update the mtime and ctime members of an inode and mark the inode
1648 * for writeback. Note that this function is meant exclusively for
1649 * usage in the file write path of filesystems, and filesystems may
1650 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001651 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Josef Bacikc3b2da32012-03-26 09:59:21 -04001652 * timestamps are handled by the server. This can return an error for
1653 * file systems who need to allocate space in order to update an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 */
1655
Josef Bacikc3b2da32012-03-26 09:59:21 -04001656int file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657{
Al Viro496ad9a2013-01-23 17:07:38 -05001658 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 struct timespec now;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001660 int sync_it = 0;
1661 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662
Andi Kleence06e0b2009-09-18 13:05:48 -07001663 /* First try to exhaust all avenues to not sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 if (IS_NOCMTIME(inode))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001665 return 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001666
Andi Kleence06e0b2009-09-18 13:05:48 -07001667 now = current_fs_time(inode->i_sb);
1668 if (!timespec_equal(&inode->i_mtime, &now))
1669 sync_it = S_MTIME;
1670
1671 if (!timespec_equal(&inode->i_ctime, &now))
1672 sync_it |= S_CTIME;
1673
1674 if (IS_I_VERSION(inode))
1675 sync_it |= S_VERSION;
1676
1677 if (!sync_it)
Josef Bacikc3b2da32012-03-26 09:59:21 -04001678 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
Andi Kleence06e0b2009-09-18 13:05:48 -07001680 /* Finally allowed to write? Takes lock. */
Jan Karaeb04c282012-06-12 16:20:35 +02001681 if (__mnt_want_write_file(file))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001682 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
Josef Bacikc3b2da32012-03-26 09:59:21 -04001684 ret = update_time(inode, &now, sync_it);
Jan Karaeb04c282012-06-12 16:20:35 +02001685 __mnt_drop_write_file(file);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001686
1687 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001689EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
1691int inode_needs_sync(struct inode *inode)
1692{
1693 if (IS_SYNC(inode))
1694 return 1;
1695 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1696 return 1;
1697 return 0;
1698}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699EXPORT_SYMBOL(inode_needs_sync);
1700
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701int inode_wait(void *word)
1702{
1703 schedule();
1704 return 0;
1705}
Stephen Rothwelld44dab82008-11-10 17:06:05 +11001706EXPORT_SYMBOL(inode_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707
1708/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001709 * If we try to find an inode in the inode hash while it is being
1710 * deleted, we have to wait until the filesystem completes its
1711 * deletion before reporting that it isn't found. This function waits
1712 * until the deletion _might_ have completed. Callers are responsible
1713 * to recheck inode state.
1714 *
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001715 * It doesn't matter if I_NEW is not set initially, a call to
Dave Chinner250df6e2011-03-22 22:23:36 +11001716 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1717 * will DTRT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 */
1719static void __wait_on_freeing_inode(struct inode *inode)
1720{
1721 wait_queue_head_t *wq;
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001722 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1723 wq = bit_waitqueue(&inode->i_state, __I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +11001725 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001726 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 schedule();
1728 finish_wait(wq, &wait.wait);
Dave Chinner67a23c42011-03-22 22:23:42 +11001729 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730}
1731
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732static __initdata unsigned long ihash_entries;
1733static int __init set_ihash_entries(char *str)
1734{
1735 if (!str)
1736 return 0;
1737 ihash_entries = simple_strtoul(str, &str, 0);
1738 return 1;
1739}
1740__setup("ihash_entries=", set_ihash_entries);
1741
1742/*
1743 * Initialize the waitqueues and inode hash table.
1744 */
1745void __init inode_init_early(void)
1746{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001747 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748
1749 /* If hashes are distributed across NUMA nodes, defer
1750 * hash allocation until vmalloc space is available.
1751 */
1752 if (hashdist)
1753 return;
1754
1755 inode_hashtable =
1756 alloc_large_system_hash("Inode-cache",
1757 sizeof(struct hlist_head),
1758 ihash_entries,
1759 14,
1760 HASH_EARLY,
1761 &i_hash_shift,
1762 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001763 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 0);
1765
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001766 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1768}
1769
Denis Cheng74bf17c2007-10-16 23:26:30 -07001770void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001772 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
1774 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001775 inode_cachep = kmem_cache_create("inode_cache",
1776 sizeof(struct inode),
1777 0,
1778 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1779 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001780 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781
1782 /* Hash may have been set up in inode_init_early */
1783 if (!hashdist)
1784 return;
1785
1786 inode_hashtable =
1787 alloc_large_system_hash("Inode-cache",
1788 sizeof(struct hlist_head),
1789 ihash_entries,
1790 14,
1791 0,
1792 &i_hash_shift,
1793 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001794 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 0);
1796
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001797 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1799}
1800
1801void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1802{
1803 inode->i_mode = mode;
1804 if (S_ISCHR(mode)) {
1805 inode->i_fop = &def_chr_fops;
1806 inode->i_rdev = rdev;
1807 } else if (S_ISBLK(mode)) {
1808 inode->i_fop = &def_blk_fops;
1809 inode->i_rdev = rdev;
1810 } else if (S_ISFIFO(mode))
1811 inode->i_fop = &def_fifo_fops;
1812 else if (S_ISSOCK(mode))
1813 inode->i_fop = &bad_sock_fops;
1814 else
Manish Katiyaraf0d9ae2009-09-18 13:05:43 -07001815 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1816 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1817 inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818}
1819EXPORT_SYMBOL(init_special_inode);
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001820
1821/**
Ben Hutchingseaae668d2011-02-15 12:48:09 +00001822 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001823 * @inode: New inode
1824 * @dir: Directory inode
1825 * @mode: mode of the new inode
1826 */
1827void inode_init_owner(struct inode *inode, const struct inode *dir,
Al Viro62bb1092011-07-24 23:20:18 -04001828 umode_t mode)
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001829{
1830 inode->i_uid = current_fsuid();
1831 if (dir && dir->i_mode & S_ISGID) {
1832 inode->i_gid = dir->i_gid;
1833 if (S_ISDIR(mode))
1834 mode |= S_ISGID;
1835 } else
1836 inode->i_gid = current_fsgid();
1837 inode->i_mode = mode;
1838}
1839EXPORT_SYMBOL(inode_init_owner);
Serge E. Hallyne795b712011-03-23 16:43:25 -07001840
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001841/**
1842 * inode_owner_or_capable - check current task permissions to inode
1843 * @inode: inode being checked
1844 *
1845 * Return true if current either has CAP_FOWNER to the inode, or
1846 * owns the file.
Serge E. Hallyne795b712011-03-23 16:43:25 -07001847 */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001848bool inode_owner_or_capable(const struct inode *inode)
Serge E. Hallyne795b712011-03-23 16:43:25 -07001849{
Eric W. Biederman92361632012-02-08 07:07:50 -08001850 if (uid_eq(current_fsuid(), inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001851 return true;
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -08001852 if (inode_capable(inode, CAP_FOWNER))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001853 return true;
1854 return false;
1855}
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001856EXPORT_SYMBOL(inode_owner_or_capable);
Trond Myklebust1d59d612012-05-31 12:22:33 -04001857
1858/*
1859 * Direct i/o helper functions
1860 */
1861static void __inode_dio_wait(struct inode *inode)
1862{
1863 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1864 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1865
1866 do {
1867 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1868 if (atomic_read(&inode->i_dio_count))
1869 schedule();
1870 } while (atomic_read(&inode->i_dio_count));
1871 finish_wait(wq, &q.wait);
1872}
1873
1874/**
1875 * inode_dio_wait - wait for outstanding DIO requests to finish
1876 * @inode: inode to wait for
1877 *
1878 * Waits for all pending direct I/O requests to finish so that we can
1879 * proceed with a truncate or equivalent operation.
1880 *
1881 * Must be called under a lock that serializes taking new references
1882 * to i_dio_count, usually by inode->i_mutex.
1883 */
1884void inode_dio_wait(struct inode *inode)
1885{
1886 if (atomic_read(&inode->i_dio_count))
1887 __inode_dio_wait(inode);
1888}
1889EXPORT_SYMBOL(inode_dio_wait);
1890
1891/*
1892 * inode_dio_done - signal finish of a direct I/O requests
1893 * @inode: inode the direct I/O happens on
1894 *
1895 * This is called once we've finished processing a direct I/O request,
1896 * and is used to wake up callers waiting for direct I/O to be quiesced.
1897 */
1898void inode_dio_done(struct inode *inode)
1899{
1900 if (atomic_dec_and_test(&inode->i_dio_count))
1901 wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1902}
1903EXPORT_SYMBOL(inode_dio_done);