blob: d6dfb09c828083b8ba64d3e8aacb2b6a64581038 [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{
Al Virof4e0c302013-06-11 08:34:36 +0400336 if (unlikely(inode->i_nlink == 0)) {
337 WARN_ON(!(inode->i_state & I_LINKABLE));
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100338 atomic_long_dec(&inode->i_sb->s_remove_count);
Al Virof4e0c302013-06-11 08:34:36 +0400339 }
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100340
341 inode->__i_nlink++;
342}
343EXPORT_SYMBOL(inc_nlink);
344
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100345void address_space_init_once(struct address_space *mapping)
346{
347 memset(mapping, 0, sizeof(*mapping));
348 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
349 spin_lock_init(&mapping->tree_lock);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700350 mutex_init(&mapping->i_mmap_mutex);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100351 INIT_LIST_HEAD(&mapping->private_list);
352 spin_lock_init(&mapping->private_lock);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700353 mapping->i_mmap = RB_ROOT;
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100354 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100355}
356EXPORT_SYMBOL(address_space_init_once);
357
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358/*
359 * These are initializations that only need to be done
360 * once, because the fields are idempotent across use
361 * of the inode, so let the slab aware of that.
362 */
363void inode_init_once(struct inode *inode)
364{
365 memset(inode, 0, sizeof(*inode));
366 INIT_HLIST_NODE(&inode->i_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 INIT_LIST_HEAD(&inode->i_devices);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100368 INIT_LIST_HEAD(&inode->i_wb_list);
369 INIT_LIST_HEAD(&inode->i_lru);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100370 address_space_init_once(&inode->i_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 i_size_ordered_init(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400372#ifdef CONFIG_FSNOTIFY
Eric Parise61ce862009-12-17 21:24:24 -0500373 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
Eric Paris3be25f42009-05-21 17:01:26 -0400374#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376EXPORT_SYMBOL(inode_init_once);
377
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700378static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530380 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
Christoph Lametera35afb82007-05-16 22:10:57 -0700382 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
384
385/*
Dave Chinner250df6e2011-03-22 22:23:36 +1100386 * inode->i_lock must be held
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530388void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389{
Nick Piggin9e38d862010-10-23 06:55:17 -0400390 atomic_inc(&inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
392
Al Viro7de9c6ee2010-10-23 11:11:40 -0400393/*
394 * get additional reference to inode; caller must already hold one.
395 */
396void ihold(struct inode *inode)
397{
398 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
399}
400EXPORT_SYMBOL(ihold);
401
Nick Piggin9e38d862010-10-23 06:55:17 -0400402static void inode_lru_list_add(struct inode *inode)
403{
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000404 spin_lock(&inode->i_sb->s_inode_lru_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100405 if (list_empty(&inode->i_lru)) {
Dave Chinner98b745c2011-07-08 14:14:39 +1000406 list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
407 inode->i_sb->s_nr_inodes_unused++;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000408 this_cpu_inc(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400409 }
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000410 spin_unlock(&inode->i_sb->s_inode_lru_lock);
Nick Piggin9e38d862010-10-23 06:55:17 -0400411}
412
Jan Kara4eff96d2012-11-26 16:29:51 -0800413/*
414 * Add inode to LRU if needed (inode is unused and clean).
415 *
416 * Needs inode->i_lock held.
417 */
418void inode_add_lru(struct inode *inode)
419{
420 if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) &&
421 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
422 inode_lru_list_add(inode);
423}
424
425
Nick Piggin9e38d862010-10-23 06:55:17 -0400426static void inode_lru_list_del(struct inode *inode)
427{
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000428 spin_lock(&inode->i_sb->s_inode_lru_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100429 if (!list_empty(&inode->i_lru)) {
430 list_del_init(&inode->i_lru);
Dave Chinner98b745c2011-07-08 14:14:39 +1000431 inode->i_sb->s_nr_inodes_unused--;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000432 this_cpu_dec(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400433 }
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000434 spin_unlock(&inode->i_sb->s_inode_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435}
436
Christoph Hellwig646ec462010-10-23 07:15:32 -0400437/**
438 * inode_sb_list_add - add inode to the superblock list of inodes
439 * @inode: inode to add
440 */
441void inode_sb_list_add(struct inode *inode)
442{
Dave Chinner55fa6092011-03-22 22:23:40 +1100443 spin_lock(&inode_sb_list_lock);
444 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
445 spin_unlock(&inode_sb_list_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400446}
447EXPORT_SYMBOL_GPL(inode_sb_list_add);
448
Dave Chinner55fa6092011-03-22 22:23:40 +1100449static inline void inode_sb_list_del(struct inode *inode)
Christoph Hellwig646ec462010-10-23 07:15:32 -0400450{
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200451 if (!list_empty(&inode->i_sb_list)) {
452 spin_lock(&inode_sb_list_lock);
453 list_del_init(&inode->i_sb_list);
454 spin_unlock(&inode_sb_list_lock);
455 }
Christoph Hellwig646ec462010-10-23 07:15:32 -0400456}
457
Dave Chinner4c51acb2010-10-23 06:58:09 -0400458static unsigned long hash(struct super_block *sb, unsigned long hashval)
459{
460 unsigned long tmp;
461
462 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
463 L1_CACHE_BYTES;
Christoph Hellwig4b4563d2011-05-27 09:28:01 -0400464 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
465 return tmp & i_hash_mask;
Dave Chinner4c51acb2010-10-23 06:58:09 -0400466}
467
468/**
469 * __insert_inode_hash - hash an inode
470 * @inode: unhashed inode
471 * @hashval: unsigned long value used to locate this object in the
472 * inode_hashtable.
473 *
474 * Add an inode to the inode hash for this superblock.
475 */
476void __insert_inode_hash(struct inode *inode, unsigned long hashval)
477{
Christoph Hellwig646ec462010-10-23 07:15:32 -0400478 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
479
Dave Chinner67a23c42011-03-22 22:23:42 +1100480 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100481 spin_lock(&inode->i_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400482 hlist_add_head(&inode->i_hash, b);
Dave Chinner250df6e2011-03-22 22:23:36 +1100483 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100484 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400485}
486EXPORT_SYMBOL(__insert_inode_hash);
487
488/**
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200489 * __remove_inode_hash - remove an inode from the hash
Dave Chinner4c51acb2010-10-23 06:58:09 -0400490 * @inode: inode to unhash
491 *
492 * Remove an inode from the superblock.
493 */
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200494void __remove_inode_hash(struct inode *inode)
Dave Chinner4c51acb2010-10-23 06:58:09 -0400495{
Dave Chinner67a23c42011-03-22 22:23:42 +1100496 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100497 spin_lock(&inode->i_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400498 hlist_del_init(&inode->i_hash);
Dave Chinner250df6e2011-03-22 22:23:36 +1100499 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100500 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400501}
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200502EXPORT_SYMBOL(__remove_inode_hash);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400503
Jan Karadbd57682012-05-03 14:48:02 +0200504void clear_inode(struct inode *inode)
Al Virob0683aa2010-06-04 20:55:25 -0400505{
506 might_sleep();
Jan Kara08142572011-06-27 16:18:10 -0700507 /*
508 * We have to cycle tree_lock here because reclaim can be still in the
509 * process of removing the last page (in __delete_from_page_cache())
510 * and we must not free mapping under it.
511 */
512 spin_lock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400513 BUG_ON(inode->i_data.nrpages);
Jan Kara08142572011-06-27 16:18:10 -0700514 spin_unlock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400515 BUG_ON(!list_empty(&inode->i_data.private_list));
516 BUG_ON(!(inode->i_state & I_FREEING));
517 BUG_ON(inode->i_state & I_CLEAR);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100518 /* don't need i_lock here, no concurrent mods to i_state */
Al Virob0683aa2010-06-04 20:55:25 -0400519 inode->i_state = I_FREEING | I_CLEAR;
520}
Jan Karadbd57682012-05-03 14:48:02 +0200521EXPORT_SYMBOL(clear_inode);
Al Virob0683aa2010-06-04 20:55:25 -0400522
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100523/*
524 * Free the inode passed in, removing it from the lists it is still connected
525 * to. We remove any pages still attached to the inode and wait for any IO that
526 * is still in progress before finally destroying the inode.
527 *
528 * An inode must already be marked I_FREEING so that we avoid the inode being
529 * moved back onto lists if we race with other code that manipulates the lists
530 * (e.g. writeback_single_inode). The caller is responsible for setting this.
531 *
532 * An inode must already be removed from the LRU list before being evicted from
533 * the cache. This should occur atomically with setting the I_FREEING state
534 * flag, so no inodes here should ever be on the LRU when being evicted.
535 */
Al Viro644da592010-06-07 13:21:05 -0400536static void evict(struct inode *inode)
Al Virob4272d42010-06-04 19:33:20 -0400537{
538 const struct super_operations *op = inode->i_sb->s_op;
539
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100540 BUG_ON(!(inode->i_state & I_FREEING));
541 BUG_ON(!list_empty(&inode->i_lru));
542
Eric Dumazetb12362b2011-07-28 06:11:47 +0200543 if (!list_empty(&inode->i_wb_list))
544 inode_wb_list_del(inode);
545
Dave Chinner55fa6092011-03-22 22:23:40 +1100546 inode_sb_list_del(inode);
547
Jan Kara169ebd92012-05-03 14:48:03 +0200548 /*
549 * Wait for flusher thread to be done with the inode so that filesystem
550 * does not start destroying it while writeback is still running. Since
551 * the inode has I_FREEING set, flusher thread won't start new work on
552 * the inode. We just have to wait for running writeback to finish.
553 */
554 inode_wait_for_writeback(inode);
Jan Kara7994e6f2012-05-03 14:48:01 +0200555
Al Virobe7ce412010-06-04 19:40:39 -0400556 if (op->evict_inode) {
557 op->evict_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400558 } else {
559 if (inode->i_data.nrpages)
560 truncate_inode_pages(&inode->i_data, 0);
Jan Karadbd57682012-05-03 14:48:02 +0200561 clear_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400562 }
Al Viro661074e2010-06-04 20:19:55 -0400563 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
564 bd_forget(inode);
565 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
566 cd_forget(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100567
568 remove_inode_hash(inode);
569
570 spin_lock(&inode->i_lock);
571 wake_up_bit(&inode->i_state, __I_NEW);
572 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
573 spin_unlock(&inode->i_lock);
574
575 destroy_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400576}
577
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578/*
579 * dispose_list - dispose of the contents of a local list
580 * @head: the head of the list to free
581 *
582 * Dispose-list gets a local list with local inodes in it, so it doesn't
583 * need to worry about list corruption and SMP locks.
584 */
585static void dispose_list(struct list_head *head)
586{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 while (!list_empty(head)) {
588 struct inode *inode;
589
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100590 inode = list_first_entry(head, struct inode, i_lru);
591 list_del_init(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
Al Viro644da592010-06-07 13:21:05 -0400593 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595}
596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597/**
Al Viro63997e92010-10-25 20:49:35 -0400598 * evict_inodes - evict all evictable inodes for a superblock
599 * @sb: superblock to operate on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 *
Al Viro63997e92010-10-25 20:49:35 -0400601 * Make sure that no inodes with zero refcount are retained. This is
602 * called by superblock shutdown after having MS_ACTIVE flag removed,
603 * so any inode reaching zero refcount during or after that call will
604 * be immediately evicted.
605 */
606void evict_inodes(struct super_block *sb)
607{
608 struct inode *inode, *next;
609 LIST_HEAD(dispose);
610
Dave Chinner55fa6092011-03-22 22:23:40 +1100611 spin_lock(&inode_sb_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400612 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
613 if (atomic_read(&inode->i_count))
614 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100615
616 spin_lock(&inode->i_lock);
617 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
618 spin_unlock(&inode->i_lock);
Al Viro63997e92010-10-25 20:49:35 -0400619 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100620 }
Al Viro63997e92010-10-25 20:49:35 -0400621
622 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100623 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100624 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100625 list_add(&inode->i_lru, &dispose);
Al Viro63997e92010-10-25 20:49:35 -0400626 }
Dave Chinner55fa6092011-03-22 22:23:40 +1100627 spin_unlock(&inode_sb_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400628
629 dispose_list(&dispose);
Al Viro63997e92010-10-25 20:49:35 -0400630}
631
632/**
Christoph Hellwiga0318782010-10-24 19:40:33 +0200633 * invalidate_inodes - attempt to free all inodes on a superblock
634 * @sb: superblock to operate on
NeilBrown93b270f2011-02-24 17:25:47 +1100635 * @kill_dirty: flag to guide handling of dirty inodes
Christoph Hellwiga0318782010-10-24 19:40:33 +0200636 *
637 * Attempts to free all inodes for a given superblock. If there were any
638 * busy inodes return a non-zero value, else zero.
NeilBrown93b270f2011-02-24 17:25:47 +1100639 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
640 * them as busy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 */
NeilBrown93b270f2011-02-24 17:25:47 +1100642int invalidate_inodes(struct super_block *sb, bool kill_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400644 int busy = 0;
Christoph Hellwiga0318782010-10-24 19:40:33 +0200645 struct inode *inode, *next;
646 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
Dave Chinner55fa6092011-03-22 22:23:40 +1100648 spin_lock(&inode_sb_list_lock);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200649 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100650 spin_lock(&inode->i_lock);
651 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
652 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100654 }
NeilBrown93b270f2011-02-24 17:25:47 +1100655 if (inode->i_state & I_DIRTY && !kill_dirty) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100656 spin_unlock(&inode->i_lock);
NeilBrown93b270f2011-02-24 17:25:47 +1100657 busy = 1;
658 continue;
659 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200660 if (atomic_read(&inode->i_count)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100661 spin_unlock(&inode->i_lock);
Christoph Hellwig99a38912010-10-23 19:07:20 +0200662 busy = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 continue;
664 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200665
Christoph Hellwig99a38912010-10-23 19:07:20 +0200666 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100667 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100668 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100669 list_add(&inode->i_lru, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 }
Dave Chinner55fa6092011-03-22 22:23:40 +1100671 spin_unlock(&inode_sb_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
Christoph Hellwiga0318782010-10-24 19:40:33 +0200673 dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
675 return busy;
676}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
678static int can_unuse(struct inode *inode)
679{
Nick Piggin9e38d862010-10-23 06:55:17 -0400680 if (inode->i_state & ~I_REFERENCED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 return 0;
682 if (inode_has_buffers(inode))
683 return 0;
684 if (atomic_read(&inode->i_count))
685 return 0;
686 if (inode->i_data.nrpages)
687 return 0;
688 return 1;
689}
690
691/*
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000692 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
693 * This is called from the superblock shrinker function with a number of inodes
694 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
695 * then are freed outside inode_lock by dispose_list().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 *
697 * Any inodes which are pinned purely because of attached pagecache have their
Nick Piggin9e38d862010-10-23 06:55:17 -0400698 * pagecache removed. If the inode has metadata buffers attached to
699 * mapping->private_list then try to remove them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 *
Nick Piggin9e38d862010-10-23 06:55:17 -0400701 * If the inode has the I_REFERENCED flag set, then it means that it has been
702 * used recently - the flag is set in iput_final(). When we encounter such an
703 * inode, clear the flag and move it to the back of the LRU so it gets another
704 * pass through the LRU before it gets reclaimed. This is necessary because of
705 * the fact we are doing lazy LRU updates to minimise lock contention so the
706 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
707 * with this flag set because they are the inodes that are out of order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 */
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000709void prune_icache_sb(struct super_block *sb, int nr_to_scan)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710{
711 LIST_HEAD(freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 int nr_scanned;
713 unsigned long reap = 0;
714
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000715 spin_lock(&sb->s_inode_lru_lock);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000716 for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 struct inode *inode;
718
Dave Chinner98b745c2011-07-08 14:14:39 +1000719 if (list_empty(&sb->s_inode_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 break;
721
Dave Chinner98b745c2011-07-08 14:14:39 +1000722 inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Nick Piggin9e38d862010-10-23 06:55:17 -0400724 /*
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000725 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
Dave Chinner02afc412011-03-22 22:23:38 +1100726 * so use a trylock. If we fail to get the lock, just move the
727 * inode to the back of the list so we don't spin on it.
728 */
729 if (!spin_trylock(&inode->i_lock)) {
Suleiman Souhlal5b55d702013-04-13 16:03:06 -0700730 list_move(&inode->i_lru, &sb->s_inode_lru);
Dave Chinner02afc412011-03-22 22:23:38 +1100731 continue;
732 }
733
734 /*
Nick Piggin9e38d862010-10-23 06:55:17 -0400735 * Referenced or dirty inodes are still in use. Give them
736 * another pass through the LRU as we canot reclaim them now.
737 */
738 if (atomic_read(&inode->i_count) ||
739 (inode->i_state & ~I_REFERENCED)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100740 list_del_init(&inode->i_lru);
Dave Chinnerf283c862011-03-22 22:23:39 +1100741 spin_unlock(&inode->i_lock);
Dave Chinner98b745c2011-07-08 14:14:39 +1000742 sb->s_nr_inodes_unused--;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000743 this_cpu_dec(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400744 continue;
745 }
746
747 /* recently referenced inodes get one more pass */
748 if (inode->i_state & I_REFERENCED) {
Nick Piggin9e38d862010-10-23 06:55:17 -0400749 inode->i_state &= ~I_REFERENCED;
Dave Chinner98b745c2011-07-08 14:14:39 +1000750 list_move(&inode->i_lru, &sb->s_inode_lru);
Dave Chinnerf283c862011-03-22 22:23:39 +1100751 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 continue;
753 }
754 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
755 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100756 spin_unlock(&inode->i_lock);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000757 spin_unlock(&sb->s_inode_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800759 reap += invalidate_mapping_pages(&inode->i_data,
760 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 iput(inode);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000762 spin_lock(&sb->s_inode_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
Dave Chinner98b745c2011-07-08 14:14:39 +1000764 if (inode != list_entry(sb->s_inode_lru.next,
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100765 struct inode, i_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 continue; /* wrong inode or list_empty */
Dave Chinner02afc412011-03-22 22:23:38 +1100767 /* avoid lock inversions with trylock */
768 if (!spin_trylock(&inode->i_lock))
769 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100770 if (!can_unuse(inode)) {
771 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100773 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 }
Nick Piggin7ef0d732009-03-12 14:31:38 -0700775 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 inode->i_state |= I_FREEING;
Dave Chinner250df6e2011-03-22 22:23:36 +1100777 spin_unlock(&inode->i_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100778
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100779 list_move(&inode->i_lru, &freeable);
Dave Chinner98b745c2011-07-08 14:14:39 +1000780 sb->s_nr_inodes_unused--;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000781 this_cpu_dec(nr_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 }
Christoph Lameterf8891e52006-06-30 01:55:45 -0700783 if (current_is_kswapd())
784 __count_vm_events(KSWAPD_INODESTEAL, reap);
785 else
786 __count_vm_events(PGINODESTEAL, reap);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000787 spin_unlock(&sb->s_inode_lru_lock);
Konstantin Khlebnikov5f8aefd2012-01-10 15:07:18 -0800788 if (current->reclaim_state)
789 current->reclaim_state->reclaimed_slab += reap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
791 dispose_list(&freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792}
793
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794static void __wait_on_freeing_inode(struct inode *inode);
795/*
796 * Called with the inode lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530798static struct inode *find_inode(struct super_block *sb,
799 struct hlist_head *head,
800 int (*test)(struct inode *, void *),
801 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530803 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
805repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800806 hlist_for_each_entry(inode, head, i_hash) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100807 spin_lock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100808 if (inode->i_sb != sb) {
809 spin_unlock(&inode->i_lock);
810 continue;
811 }
812 if (!test(inode, data)) {
813 spin_unlock(&inode->i_lock);
814 continue;
815 }
Al Viroa4ffdde2010-06-02 17:38:30 -0400816 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 __wait_on_freeing_inode(inode);
818 goto repeat;
819 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400820 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100821 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400822 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400824 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825}
826
827/*
828 * find_inode_fast is the fast path version of find_inode, see the comment at
829 * iget_locked for details.
830 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530831static struct inode *find_inode_fast(struct super_block *sb,
832 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530834 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
836repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800837 hlist_for_each_entry(inode, 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);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001135 struct inode *inode;
1136
Dave Chinner67a23c42011-03-22 22:23:42 +11001137 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001138 hlist_for_each_entry(inode, b, i_hash) {
Dave Chinner67a23c42011-03-22 22:23:42 +11001139 if (inode->i_ino == ino && inode->i_sb == sb) {
1140 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001141 return 0;
Dave Chinner67a23c42011-03-22 22:23:42 +11001142 }
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001143 }
Dave Chinner67a23c42011-03-22 22:23:42 +11001144 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001145
1146 return 1;
1147}
1148
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149/**
1150 * iunique - get a unique inode number
1151 * @sb: superblock
1152 * @max_reserved: highest reserved inode number
1153 *
1154 * Obtain an inode number that is unique on the system for a given
1155 * superblock. This is used by file systems that have no natural
1156 * permanent inode numbering system. An inode number is returned that
1157 * is higher than the reserved limit but unique.
1158 *
1159 * BUGS:
1160 * With a large number of inodes live on the file system this function
1161 * currently becomes quite slow.
1162 */
1163ino_t iunique(struct super_block *sb, ino_t max_reserved)
1164{
Jeff Layton866b04f2007-05-08 00:32:29 -07001165 /*
1166 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1167 * error if st_ino won't fit in target struct field. Use 32bit counter
1168 * here to attempt to avoid that.
1169 */
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001170 static DEFINE_SPINLOCK(iunique_lock);
Jeff Layton866b04f2007-05-08 00:32:29 -07001171 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001174 spin_lock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001175 do {
1176 if (counter <= max_reserved)
1177 counter = max_reserved + 1;
1178 res = counter++;
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001179 } while (!test_inode_iunique(sb, res));
1180 spin_unlock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001181
1182 return res;
1183}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184EXPORT_SYMBOL(iunique);
1185
1186struct inode *igrab(struct inode *inode)
1187{
Dave Chinner250df6e2011-03-22 22:23:36 +11001188 spin_lock(&inode->i_lock);
1189 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001191 spin_unlock(&inode->i_lock);
1192 } else {
1193 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 /*
1195 * Handle the case where s_op->clear_inode is not been
1196 * called yet, and somebody is calling igrab
1197 * while the inode is getting freed.
1198 */
1199 inode = NULL;
Dave Chinner250df6e2011-03-22 22:23:36 +11001200 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 return inode;
1202}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203EXPORT_SYMBOL(igrab);
1204
1205/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001206 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 * @sb: super block of file system to search
1208 * @hashval: hash value (usually inode number) to search for
1209 * @test: callback used for comparisons between inodes
1210 * @data: opaque data pointer to pass to @test
1211 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001212 * Search for the inode specified by @hashval and @data in the inode cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 * If the inode is in the cache, the inode is returned with an incremented
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001214 * reference count.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001215 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001216 * Note: I_NEW is not waited upon so you have to be very careful what you do
1217 * with the returned inode. You probably should be using ilookup5() instead.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001218 *
Randy Dunlapb6d0ad62011-03-26 13:27:47 -07001219 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001220 */
1221struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1222 int (*test)(struct inode *, void *), void *data)
1223{
1224 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001225 struct inode *inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001226
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001227 spin_lock(&inode_hash_lock);
1228 inode = find_inode(sb, head, test, data);
1229 spin_unlock(&inode_hash_lock);
1230
1231 return inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001232}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001233EXPORT_SYMBOL(ilookup5_nowait);
1234
1235/**
1236 * ilookup5 - search for an inode in the inode cache
1237 * @sb: super block of file system to search
1238 * @hashval: hash value (usually inode number) to search for
1239 * @test: callback used for comparisons between inodes
1240 * @data: opaque data pointer to pass to @test
1241 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001242 * Search for the inode specified by @hashval and @data in the inode cache,
1243 * and if the inode is in the cache, return the inode with an incremented
1244 * reference count. Waits on I_NEW before returning the inode.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001245 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001247 * This is a generalized version of ilookup() for file systems where the
1248 * inode number is not sufficient for unique identification of an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001250 * Note: @test is called with the inode_hash_lock held, so can't sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 */
1252struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1253 int (*test)(struct inode *, void *), void *data)
1254{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001255 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001257 if (inode)
1258 wait_on_inode(inode);
1259 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261EXPORT_SYMBOL(ilookup5);
1262
1263/**
1264 * ilookup - search for an inode in the inode cache
1265 * @sb: super block of file system to search
1266 * @ino: inode number to search for
1267 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001268 * Search for the inode @ino in the inode cache, and if the inode is in the
1269 * cache, the inode is returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 */
1271struct inode *ilookup(struct super_block *sb, unsigned long ino)
1272{
1273 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001274 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001276 spin_lock(&inode_hash_lock);
1277 inode = find_inode_fast(sb, head, ino);
1278 spin_unlock(&inode_hash_lock);
1279
1280 if (inode)
1281 wait_on_inode(inode);
1282 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284EXPORT_SYMBOL(ilookup);
1285
Al Viro261bca82008-12-30 01:48:21 -05001286int insert_inode_locked(struct inode *inode)
1287{
1288 struct super_block *sb = inode->i_sb;
1289 ino_t ino = inode->i_ino;
1290 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001291
Al Viro261bca82008-12-30 01:48:21 -05001292 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001293 struct inode *old = NULL;
Dave Chinner67a23c42011-03-22 22:23:42 +11001294 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001295 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001296 if (old->i_ino != ino)
1297 continue;
1298 if (old->i_sb != sb)
1299 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001300 spin_lock(&old->i_lock);
1301 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1302 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001303 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001304 }
Al Viro72a43d62009-05-13 19:13:40 +01001305 break;
1306 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001307 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001308 spin_lock(&inode->i_lock);
1309 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001310 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001311 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001312 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001313 return 0;
1314 }
1315 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001316 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001317 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001318 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001319 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001320 iput(old);
1321 return -EBUSY;
1322 }
1323 iput(old);
1324 }
1325}
Al Viro261bca82008-12-30 01:48:21 -05001326EXPORT_SYMBOL(insert_inode_locked);
1327
1328int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1329 int (*test)(struct inode *, void *), void *data)
1330{
1331 struct super_block *sb = inode->i_sb;
1332 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001333
Al Viro261bca82008-12-30 01:48:21 -05001334 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001335 struct inode *old = NULL;
1336
Dave Chinner67a23c42011-03-22 22:23:42 +11001337 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001338 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001339 if (old->i_sb != sb)
1340 continue;
1341 if (!test(old, data))
1342 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001343 spin_lock(&old->i_lock);
1344 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1345 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001346 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001347 }
Al Viro72a43d62009-05-13 19:13:40 +01001348 break;
1349 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001350 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001351 spin_lock(&inode->i_lock);
1352 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001353 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001354 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001355 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001356 return 0;
1357 }
1358 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001359 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001360 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001361 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001362 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001363 iput(old);
1364 return -EBUSY;
1365 }
1366 iput(old);
1367 }
1368}
Al Viro261bca82008-12-30 01:48:21 -05001369EXPORT_SYMBOL(insert_inode_locked4);
1370
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
Al Viro45321ac2010-06-07 13:43:19 -04001372int generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373{
Al Viro45321ac2010-06-07 13:43:19 -04001374 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376EXPORT_SYMBOL(generic_delete_inode);
1377
Al Viro45321ac2010-06-07 13:43:19 -04001378/*
Al Viro45321ac2010-06-07 13:43:19 -04001379 * Called when we're dropping the last reference
1380 * to an inode.
1381 *
1382 * Call the FS "drop_inode()" function, defaulting to
1383 * the legacy UNIX filesystem behaviour. If it tells
1384 * us to evict inode, do so. Otherwise, retain inode
1385 * in cache if fs is alive, sync and evict if fs is
1386 * shutting down.
1387 */
1388static void iput_final(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389{
1390 struct super_block *sb = inode->i_sb;
Al Viro45321ac2010-06-07 13:43:19 -04001391 const struct super_operations *op = inode->i_sb->s_op;
1392 int drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
Dave Chinner250df6e2011-03-22 22:23:36 +11001394 WARN_ON(inode->i_state & I_NEW);
1395
Al Viroe7f59092011-07-07 15:45:59 -04001396 if (op->drop_inode)
Al Viro45321ac2010-06-07 13:43:19 -04001397 drop = op->drop_inode(inode);
1398 else
1399 drop = generic_drop_inode(inode);
1400
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001401 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1402 inode->i_state |= I_REFERENCED;
Jan Kara4eff96d2012-11-26 16:29:51 -08001403 inode_add_lru(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001404 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001405 return;
1406 }
1407
Al Viro45321ac2010-06-07 13:43:19 -04001408 if (!drop) {
Alexander Viro991114c2005-06-23 00:09:01 -07001409 inode->i_state |= I_WILL_FREE;
Dave Chinner250df6e2011-03-22 22:23:36 +11001410 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 write_inode_now(inode, 1);
Dave Chinner250df6e2011-03-22 22:23:36 +11001412 spin_lock(&inode->i_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001413 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001414 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001416
Alexander Viro991114c2005-06-23 00:09:01 -07001417 inode->i_state |= I_FREEING;
Eric Dumazetc4ae0c62011-07-28 06:55:13 +02001418 if (!list_empty(&inode->i_lru))
1419 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001420 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001421
1422 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423}
1424
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301426 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 * @inode: inode to put
1428 *
1429 * Puts an inode, dropping its usage count. If the inode use count hits
1430 * zero, the inode is then freed and may also be destroyed.
1431 *
1432 * Consequently, iput() can sleep.
1433 */
1434void iput(struct inode *inode)
1435{
1436 if (inode) {
Al Viroa4ffdde2010-06-02 17:38:30 -04001437 BUG_ON(inode->i_state & I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
Dave Chinnerf283c862011-03-22 22:23:39 +11001439 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 iput_final(inode);
1441 }
1442}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443EXPORT_SYMBOL(iput);
1444
1445/**
1446 * bmap - find a block number in a file
1447 * @inode: inode of file
1448 * @block: block to find
1449 *
1450 * Returns the block number on the device holding the inode that
1451 * is the disk block number for the block of the file requested.
1452 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301453 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 * file.
1455 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301456sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457{
1458 sector_t res = 0;
1459 if (inode->i_mapping->a_ops->bmap)
1460 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1461 return res;
1462}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463EXPORT_SYMBOL(bmap);
1464
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001465/*
1466 * With relative atime, only update atime if the previous atime is
1467 * earlier than either the ctime or mtime or if at least a day has
1468 * passed since the last atime update.
1469 */
1470static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1471 struct timespec now)
1472{
1473
1474 if (!(mnt->mnt_flags & MNT_RELATIME))
1475 return 1;
1476 /*
1477 * Is mtime younger than atime? If yes, update atime:
1478 */
1479 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1480 return 1;
1481 /*
1482 * Is ctime younger than atime? If yes, update atime:
1483 */
1484 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1485 return 1;
1486
1487 /*
1488 * Is the previous atime value older than a day? If yes,
1489 * update atime:
1490 */
1491 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1492 return 1;
1493 /*
1494 * Good, we can skip the atime update:
1495 */
1496 return 0;
1497}
1498
Josef Bacikc3b2da32012-03-26 09:59:21 -04001499/*
1500 * This does the actual work of updating an inodes time or version. Must have
1501 * had called mnt_want_write() before calling this.
1502 */
1503static int update_time(struct inode *inode, struct timespec *time, int flags)
1504{
1505 if (inode->i_op->update_time)
1506 return inode->i_op->update_time(inode, time, flags);
1507
1508 if (flags & S_ATIME)
1509 inode->i_atime = *time;
1510 if (flags & S_VERSION)
1511 inode_inc_iversion(inode);
1512 if (flags & S_CTIME)
1513 inode->i_ctime = *time;
1514 if (flags & S_MTIME)
1515 inode->i_mtime = *time;
1516 mark_inode_dirty_sync(inode);
1517 return 0;
1518}
1519
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001521 * touch_atime - update the access time
Randy Dunlap185553b2012-04-17 17:03:25 -07001522 * @path: the &struct path to update
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 *
1524 * Update the accessed time on an inode and mark it for writeback.
1525 * This function automatically handles read only file systems and media,
1526 * as well as the "noatime" flag and inode specific "noatime" markers.
1527 */
Al Viro68ac1232012-03-15 08:21:57 -04001528void touch_atime(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529{
Al Viro68ac1232012-03-15 08:21:57 -04001530 struct vfsmount *mnt = path->mnt;
1531 struct inode *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 struct timespec now;
1533
Andrew Mortonb2276132006-12-13 00:34:33 -08001534 if (inode->i_flags & S_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001535 return;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001536 if (IS_NOATIME(inode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001537 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001538 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001539 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001540
Dave Hansencdb70f32008-02-15 14:37:41 -08001541 if (mnt->mnt_flags & MNT_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001542 return;
Dave Hansencdb70f32008-02-15 14:37:41 -08001543 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001544 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001545
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 now = current_fs_time(inode->i_sb);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001547
1548 if (!relatime_need_update(mnt, inode, now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001549 return;
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001550
Valerie Henson47ae32d2006-12-13 00:34:34 -08001551 if (timespec_equal(&inode->i_atime, &now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001552 return;
1553
Jan Kara5d37e9e2012-06-12 16:20:36 +02001554 if (!sb_start_write_trylock(inode->i_sb))
Andi Kleenb12536c2009-09-18 13:05:47 -07001555 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001556
Jan Kara5d37e9e2012-06-12 16:20:36 +02001557 if (__mnt_want_write(mnt))
1558 goto skip_update;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001559 /*
1560 * File systems can error out when updating inodes if they need to
1561 * allocate new space to modify an inode (such is the case for
1562 * Btrfs), but since we touch atime while walking down the path we
1563 * really don't care if we failed to update the atime of the file,
1564 * so just ignore the return value.
Alexander Block2bc5565282012-06-15 09:49:33 +02001565 * We may also fail on filesystems that have the ability to make parts
1566 * of the fs read only, e.g. subvolumes in Btrfs.
Josef Bacikc3b2da32012-03-26 09:59:21 -04001567 */
1568 update_time(inode, &now, S_ATIME);
Jan Kara5d37e9e2012-06-12 16:20:36 +02001569 __mnt_drop_write(mnt);
1570skip_update:
1571 sb_end_write(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001573EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
Cong Wang3ed37642012-05-15 14:57:33 +08001575/*
1576 * The logic we want is
1577 *
1578 * if suid or (sgid and xgrp)
1579 * remove privs
1580 */
1581int should_remove_suid(struct dentry *dentry)
1582{
1583 umode_t mode = dentry->d_inode->i_mode;
1584 int kill = 0;
1585
1586 /* suid always must be killed */
1587 if (unlikely(mode & S_ISUID))
1588 kill = ATTR_KILL_SUID;
1589
1590 /*
1591 * sgid without any exec bits is just a mandatory locking mark; leave
1592 * it alone. If some exec bits are set, it's a real sgid; kill it.
1593 */
1594 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1595 kill |= ATTR_KILL_SGID;
1596
1597 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1598 return kill;
1599
1600 return 0;
1601}
1602EXPORT_SYMBOL(should_remove_suid);
1603
1604static int __remove_suid(struct dentry *dentry, int kill)
1605{
1606 struct iattr newattrs;
1607
1608 newattrs.ia_valid = ATTR_FORCE | kill;
1609 return notify_change(dentry, &newattrs);
1610}
1611
1612int file_remove_suid(struct file *file)
1613{
1614 struct dentry *dentry = file->f_path.dentry;
1615 struct inode *inode = dentry->d_inode;
1616 int killsuid;
1617 int killpriv;
1618 int error = 0;
1619
1620 /* Fast path for nothing security related */
1621 if (IS_NOSEC(inode))
1622 return 0;
1623
1624 killsuid = should_remove_suid(dentry);
1625 killpriv = security_inode_need_killpriv(dentry);
1626
1627 if (killpriv < 0)
1628 return killpriv;
1629 if (killpriv)
1630 error = security_inode_killpriv(dentry);
1631 if (!error && killsuid)
1632 error = __remove_suid(dentry, killsuid);
1633 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1634 inode->i_flags |= S_NOSEC;
1635
1636 return error;
1637}
1638EXPORT_SYMBOL(file_remove_suid);
1639
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001641 * file_update_time - update mtime and ctime time
1642 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001644 * Update the mtime and ctime members of an inode and mark the inode
1645 * for writeback. Note that this function is meant exclusively for
1646 * usage in the file write path of filesystems, and filesystems may
1647 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001648 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Josef Bacikc3b2da32012-03-26 09:59:21 -04001649 * timestamps are handled by the server. This can return an error for
1650 * file systems who need to allocate space in order to update an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 */
1652
Josef Bacikc3b2da32012-03-26 09:59:21 -04001653int file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654{
Al Viro496ad9a2013-01-23 17:07:38 -05001655 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 struct timespec now;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001657 int sync_it = 0;
1658 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659
Andi Kleence06e0b2009-09-18 13:05:48 -07001660 /* First try to exhaust all avenues to not sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 if (IS_NOCMTIME(inode))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001662 return 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001663
Andi Kleence06e0b2009-09-18 13:05:48 -07001664 now = current_fs_time(inode->i_sb);
1665 if (!timespec_equal(&inode->i_mtime, &now))
1666 sync_it = S_MTIME;
1667
1668 if (!timespec_equal(&inode->i_ctime, &now))
1669 sync_it |= S_CTIME;
1670
1671 if (IS_I_VERSION(inode))
1672 sync_it |= S_VERSION;
1673
1674 if (!sync_it)
Josef Bacikc3b2da32012-03-26 09:59:21 -04001675 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
Andi Kleence06e0b2009-09-18 13:05:48 -07001677 /* Finally allowed to write? Takes lock. */
Jan Karaeb04c282012-06-12 16:20:35 +02001678 if (__mnt_want_write_file(file))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001679 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
Josef Bacikc3b2da32012-03-26 09:59:21 -04001681 ret = update_time(inode, &now, sync_it);
Jan Karaeb04c282012-06-12 16:20:35 +02001682 __mnt_drop_write_file(file);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001683
1684 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001686EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687
1688int inode_needs_sync(struct inode *inode)
1689{
1690 if (IS_SYNC(inode))
1691 return 1;
1692 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1693 return 1;
1694 return 0;
1695}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696EXPORT_SYMBOL(inode_needs_sync);
1697
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698int inode_wait(void *word)
1699{
1700 schedule();
1701 return 0;
1702}
Stephen Rothwelld44dab82008-11-10 17:06:05 +11001703EXPORT_SYMBOL(inode_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
1705/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001706 * If we try to find an inode in the inode hash while it is being
1707 * deleted, we have to wait until the filesystem completes its
1708 * deletion before reporting that it isn't found. This function waits
1709 * until the deletion _might_ have completed. Callers are responsible
1710 * to recheck inode state.
1711 *
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001712 * It doesn't matter if I_NEW is not set initially, a call to
Dave Chinner250df6e2011-03-22 22:23:36 +11001713 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1714 * will DTRT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 */
1716static void __wait_on_freeing_inode(struct inode *inode)
1717{
1718 wait_queue_head_t *wq;
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001719 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1720 wq = bit_waitqueue(&inode->i_state, __I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +11001722 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001723 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 schedule();
1725 finish_wait(wq, &wait.wait);
Dave Chinner67a23c42011-03-22 22:23:42 +11001726 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727}
1728
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729static __initdata unsigned long ihash_entries;
1730static int __init set_ihash_entries(char *str)
1731{
1732 if (!str)
1733 return 0;
1734 ihash_entries = simple_strtoul(str, &str, 0);
1735 return 1;
1736}
1737__setup("ihash_entries=", set_ihash_entries);
1738
1739/*
1740 * Initialize the waitqueues and inode hash table.
1741 */
1742void __init inode_init_early(void)
1743{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001744 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745
1746 /* If hashes are distributed across NUMA nodes, defer
1747 * hash allocation until vmalloc space is available.
1748 */
1749 if (hashdist)
1750 return;
1751
1752 inode_hashtable =
1753 alloc_large_system_hash("Inode-cache",
1754 sizeof(struct hlist_head),
1755 ihash_entries,
1756 14,
1757 HASH_EARLY,
1758 &i_hash_shift,
1759 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001760 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 0);
1762
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001763 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1765}
1766
Denis Cheng74bf17c2007-10-16 23:26:30 -07001767void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001769 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
1771 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001772 inode_cachep = kmem_cache_create("inode_cache",
1773 sizeof(struct inode),
1774 0,
1775 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1776 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001777 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
1779 /* Hash may have been set up in inode_init_early */
1780 if (!hashdist)
1781 return;
1782
1783 inode_hashtable =
1784 alloc_large_system_hash("Inode-cache",
1785 sizeof(struct hlist_head),
1786 ihash_entries,
1787 14,
1788 0,
1789 &i_hash_shift,
1790 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001791 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 0);
1793
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001794 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1796}
1797
1798void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1799{
1800 inode->i_mode = mode;
1801 if (S_ISCHR(mode)) {
1802 inode->i_fop = &def_chr_fops;
1803 inode->i_rdev = rdev;
1804 } else if (S_ISBLK(mode)) {
1805 inode->i_fop = &def_blk_fops;
1806 inode->i_rdev = rdev;
1807 } else if (S_ISFIFO(mode))
Al Viro599a0ac2013-03-12 09:58:10 -04001808 inode->i_fop = &pipefifo_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 else if (S_ISSOCK(mode))
1810 inode->i_fop = &bad_sock_fops;
1811 else
Manish Katiyaraf0d9ae2009-09-18 13:05:43 -07001812 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1813 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1814 inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815}
1816EXPORT_SYMBOL(init_special_inode);
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001817
1818/**
Ben Hutchingseaae668d2011-02-15 12:48:09 +00001819 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001820 * @inode: New inode
1821 * @dir: Directory inode
1822 * @mode: mode of the new inode
1823 */
1824void inode_init_owner(struct inode *inode, const struct inode *dir,
Al Viro62bb1092011-07-24 23:20:18 -04001825 umode_t mode)
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001826{
1827 inode->i_uid = current_fsuid();
1828 if (dir && dir->i_mode & S_ISGID) {
1829 inode->i_gid = dir->i_gid;
1830 if (S_ISDIR(mode))
1831 mode |= S_ISGID;
1832 } else
1833 inode->i_gid = current_fsgid();
1834 inode->i_mode = mode;
1835}
1836EXPORT_SYMBOL(inode_init_owner);
Serge E. Hallyne795b712011-03-23 16:43:25 -07001837
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001838/**
1839 * inode_owner_or_capable - check current task permissions to inode
1840 * @inode: inode being checked
1841 *
1842 * Return true if current either has CAP_FOWNER to the inode, or
1843 * owns the file.
Serge E. Hallyne795b712011-03-23 16:43:25 -07001844 */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001845bool inode_owner_or_capable(const struct inode *inode)
Serge E. Hallyne795b712011-03-23 16:43:25 -07001846{
Eric W. Biederman92361632012-02-08 07:07:50 -08001847 if (uid_eq(current_fsuid(), inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001848 return true;
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -08001849 if (inode_capable(inode, CAP_FOWNER))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001850 return true;
1851 return false;
1852}
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001853EXPORT_SYMBOL(inode_owner_or_capable);
Trond Myklebust1d59d612012-05-31 12:22:33 -04001854
1855/*
1856 * Direct i/o helper functions
1857 */
1858static void __inode_dio_wait(struct inode *inode)
1859{
1860 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1861 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1862
1863 do {
1864 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1865 if (atomic_read(&inode->i_dio_count))
1866 schedule();
1867 } while (atomic_read(&inode->i_dio_count));
1868 finish_wait(wq, &q.wait);
1869}
1870
1871/**
1872 * inode_dio_wait - wait for outstanding DIO requests to finish
1873 * @inode: inode to wait for
1874 *
1875 * Waits for all pending direct I/O requests to finish so that we can
1876 * proceed with a truncate or equivalent operation.
1877 *
1878 * Must be called under a lock that serializes taking new references
1879 * to i_dio_count, usually by inode->i_mutex.
1880 */
1881void inode_dio_wait(struct inode *inode)
1882{
1883 if (atomic_read(&inode->i_dio_count))
1884 __inode_dio_wait(inode);
1885}
1886EXPORT_SYMBOL(inode_dio_wait);
1887
1888/*
1889 * inode_dio_done - signal finish of a direct I/O requests
1890 * @inode: inode the direct I/O happens on
1891 *
1892 * This is called once we've finished processing a direct I/O request,
1893 * and is used to wake up callers waiting for direct I/O to be quiesced.
1894 */
1895void inode_dio_done(struct inode *inode)
1896{
1897 if (atomic_dec_and_test(&inode->i_dio_count))
1898 wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1899}
1900EXPORT_SYMBOL(inode_dio_done);