blob: ac8d904b3f1624bfa945ba088d2909e740e16a6c [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);
David Chinner2cb15992008-10-30 17:32:23 +1100168 mapping->assoc_mapping = NULL;
169 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);
351 INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
352 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
411static void inode_lru_list_del(struct inode *inode)
412{
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000413 spin_lock(&inode->i_sb->s_inode_lru_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100414 if (!list_empty(&inode->i_lru)) {
415 list_del_init(&inode->i_lru);
Dave Chinner98b745c2011-07-08 14:14:39 +1000416 inode->i_sb->s_nr_inodes_unused--;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000417 this_cpu_dec(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400418 }
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000419 spin_unlock(&inode->i_sb->s_inode_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420}
421
Christoph Hellwig646ec462010-10-23 07:15:32 -0400422/**
423 * inode_sb_list_add - add inode to the superblock list of inodes
424 * @inode: inode to add
425 */
426void inode_sb_list_add(struct inode *inode)
427{
Dave Chinner55fa6092011-03-22 22:23:40 +1100428 spin_lock(&inode_sb_list_lock);
429 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
430 spin_unlock(&inode_sb_list_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400431}
432EXPORT_SYMBOL_GPL(inode_sb_list_add);
433
Dave Chinner55fa6092011-03-22 22:23:40 +1100434static inline void inode_sb_list_del(struct inode *inode)
Christoph Hellwig646ec462010-10-23 07:15:32 -0400435{
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200436 if (!list_empty(&inode->i_sb_list)) {
437 spin_lock(&inode_sb_list_lock);
438 list_del_init(&inode->i_sb_list);
439 spin_unlock(&inode_sb_list_lock);
440 }
Christoph Hellwig646ec462010-10-23 07:15:32 -0400441}
442
Dave Chinner4c51acb2010-10-23 06:58:09 -0400443static unsigned long hash(struct super_block *sb, unsigned long hashval)
444{
445 unsigned long tmp;
446
447 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
448 L1_CACHE_BYTES;
Christoph Hellwig4b4563d2011-05-27 09:28:01 -0400449 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
450 return tmp & i_hash_mask;
Dave Chinner4c51acb2010-10-23 06:58:09 -0400451}
452
453/**
454 * __insert_inode_hash - hash an inode
455 * @inode: unhashed inode
456 * @hashval: unsigned long value used to locate this object in the
457 * inode_hashtable.
458 *
459 * Add an inode to the inode hash for this superblock.
460 */
461void __insert_inode_hash(struct inode *inode, unsigned long hashval)
462{
Christoph Hellwig646ec462010-10-23 07:15:32 -0400463 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
464
Dave Chinner67a23c42011-03-22 22:23:42 +1100465 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100466 spin_lock(&inode->i_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400467 hlist_add_head(&inode->i_hash, b);
Dave Chinner250df6e2011-03-22 22:23:36 +1100468 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100469 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400470}
471EXPORT_SYMBOL(__insert_inode_hash);
472
473/**
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200474 * __remove_inode_hash - remove an inode from the hash
Dave Chinner4c51acb2010-10-23 06:58:09 -0400475 * @inode: inode to unhash
476 *
477 * Remove an inode from the superblock.
478 */
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200479void __remove_inode_hash(struct inode *inode)
Dave Chinner4c51acb2010-10-23 06:58:09 -0400480{
Dave Chinner67a23c42011-03-22 22:23:42 +1100481 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100482 spin_lock(&inode->i_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400483 hlist_del_init(&inode->i_hash);
Dave Chinner250df6e2011-03-22 22:23:36 +1100484 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100485 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400486}
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200487EXPORT_SYMBOL(__remove_inode_hash);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400488
Jan Karadbd57682012-05-03 14:48:02 +0200489void clear_inode(struct inode *inode)
Al Virob0683aa2010-06-04 20:55:25 -0400490{
491 might_sleep();
Jan Kara08142572011-06-27 16:18:10 -0700492 /*
493 * We have to cycle tree_lock here because reclaim can be still in the
494 * process of removing the last page (in __delete_from_page_cache())
495 * and we must not free mapping under it.
496 */
497 spin_lock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400498 BUG_ON(inode->i_data.nrpages);
Jan Kara08142572011-06-27 16:18:10 -0700499 spin_unlock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400500 BUG_ON(!list_empty(&inode->i_data.private_list));
501 BUG_ON(!(inode->i_state & I_FREEING));
502 BUG_ON(inode->i_state & I_CLEAR);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100503 /* don't need i_lock here, no concurrent mods to i_state */
Al Virob0683aa2010-06-04 20:55:25 -0400504 inode->i_state = I_FREEING | I_CLEAR;
505}
Jan Karadbd57682012-05-03 14:48:02 +0200506EXPORT_SYMBOL(clear_inode);
Al Virob0683aa2010-06-04 20:55:25 -0400507
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100508/*
509 * Free the inode passed in, removing it from the lists it is still connected
510 * to. We remove any pages still attached to the inode and wait for any IO that
511 * is still in progress before finally destroying the inode.
512 *
513 * An inode must already be marked I_FREEING so that we avoid the inode being
514 * moved back onto lists if we race with other code that manipulates the lists
515 * (e.g. writeback_single_inode). The caller is responsible for setting this.
516 *
517 * An inode must already be removed from the LRU list before being evicted from
518 * the cache. This should occur atomically with setting the I_FREEING state
519 * flag, so no inodes here should ever be on the LRU when being evicted.
520 */
Al Viro644da592010-06-07 13:21:05 -0400521static void evict(struct inode *inode)
Al Virob4272d42010-06-04 19:33:20 -0400522{
523 const struct super_operations *op = inode->i_sb->s_op;
524
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100525 BUG_ON(!(inode->i_state & I_FREEING));
526 BUG_ON(!list_empty(&inode->i_lru));
527
Eric Dumazetb12362b2011-07-28 06:11:47 +0200528 if (!list_empty(&inode->i_wb_list))
529 inode_wb_list_del(inode);
530
Dave Chinner55fa6092011-03-22 22:23:40 +1100531 inode_sb_list_del(inode);
532
Jan Kara169ebd92012-05-03 14:48:03 +0200533 /*
534 * Wait for flusher thread to be done with the inode so that filesystem
535 * does not start destroying it while writeback is still running. Since
536 * the inode has I_FREEING set, flusher thread won't start new work on
537 * the inode. We just have to wait for running writeback to finish.
538 */
539 inode_wait_for_writeback(inode);
Jan Kara7994e6f2012-05-03 14:48:01 +0200540
Al Virobe7ce412010-06-04 19:40:39 -0400541 if (op->evict_inode) {
542 op->evict_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400543 } else {
544 if (inode->i_data.nrpages)
545 truncate_inode_pages(&inode->i_data, 0);
Jan Karadbd57682012-05-03 14:48:02 +0200546 clear_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400547 }
Al Viro661074e2010-06-04 20:19:55 -0400548 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
549 bd_forget(inode);
550 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
551 cd_forget(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100552
553 remove_inode_hash(inode);
554
555 spin_lock(&inode->i_lock);
556 wake_up_bit(&inode->i_state, __I_NEW);
557 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
558 spin_unlock(&inode->i_lock);
559
560 destroy_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400561}
562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563/*
564 * dispose_list - dispose of the contents of a local list
565 * @head: the head of the list to free
566 *
567 * Dispose-list gets a local list with local inodes in it, so it doesn't
568 * need to worry about list corruption and SMP locks.
569 */
570static void dispose_list(struct list_head *head)
571{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 while (!list_empty(head)) {
573 struct inode *inode;
574
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100575 inode = list_first_entry(head, struct inode, i_lru);
576 list_del_init(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
Al Viro644da592010-06-07 13:21:05 -0400578 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580}
581
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582/**
Al Viro63997e92010-10-25 20:49:35 -0400583 * evict_inodes - evict all evictable inodes for a superblock
584 * @sb: superblock to operate on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 *
Al Viro63997e92010-10-25 20:49:35 -0400586 * Make sure that no inodes with zero refcount are retained. This is
587 * called by superblock shutdown after having MS_ACTIVE flag removed,
588 * so any inode reaching zero refcount during or after that call will
589 * be immediately evicted.
590 */
591void evict_inodes(struct super_block *sb)
592{
593 struct inode *inode, *next;
594 LIST_HEAD(dispose);
595
Dave Chinner55fa6092011-03-22 22:23:40 +1100596 spin_lock(&inode_sb_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400597 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
598 if (atomic_read(&inode->i_count))
599 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100600
601 spin_lock(&inode->i_lock);
602 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
603 spin_unlock(&inode->i_lock);
Al Viro63997e92010-10-25 20:49:35 -0400604 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100605 }
Al Viro63997e92010-10-25 20:49:35 -0400606
607 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100608 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100609 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100610 list_add(&inode->i_lru, &dispose);
Al Viro63997e92010-10-25 20:49:35 -0400611 }
Dave Chinner55fa6092011-03-22 22:23:40 +1100612 spin_unlock(&inode_sb_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400613
614 dispose_list(&dispose);
Al Viro63997e92010-10-25 20:49:35 -0400615}
616
617/**
Christoph Hellwiga0318782010-10-24 19:40:33 +0200618 * invalidate_inodes - attempt to free all inodes on a superblock
619 * @sb: superblock to operate on
NeilBrown93b270f2011-02-24 17:25:47 +1100620 * @kill_dirty: flag to guide handling of dirty inodes
Christoph Hellwiga0318782010-10-24 19:40:33 +0200621 *
622 * Attempts to free all inodes for a given superblock. If there were any
623 * busy inodes return a non-zero value, else zero.
NeilBrown93b270f2011-02-24 17:25:47 +1100624 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
625 * them as busy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 */
NeilBrown93b270f2011-02-24 17:25:47 +1100627int invalidate_inodes(struct super_block *sb, bool kill_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400629 int busy = 0;
Christoph Hellwiga0318782010-10-24 19:40:33 +0200630 struct inode *inode, *next;
631 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
Dave Chinner55fa6092011-03-22 22:23:40 +1100633 spin_lock(&inode_sb_list_lock);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200634 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100635 spin_lock(&inode->i_lock);
636 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
637 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100639 }
NeilBrown93b270f2011-02-24 17:25:47 +1100640 if (inode->i_state & I_DIRTY && !kill_dirty) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100641 spin_unlock(&inode->i_lock);
NeilBrown93b270f2011-02-24 17:25:47 +1100642 busy = 1;
643 continue;
644 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200645 if (atomic_read(&inode->i_count)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100646 spin_unlock(&inode->i_lock);
Christoph Hellwig99a38912010-10-23 19:07:20 +0200647 busy = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 continue;
649 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200650
Christoph Hellwig99a38912010-10-23 19:07:20 +0200651 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100652 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100653 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100654 list_add(&inode->i_lru, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 }
Dave Chinner55fa6092011-03-22 22:23:40 +1100656 spin_unlock(&inode_sb_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
Christoph Hellwiga0318782010-10-24 19:40:33 +0200658 dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
660 return busy;
661}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
663static int can_unuse(struct inode *inode)
664{
Nick Piggin9e38d862010-10-23 06:55:17 -0400665 if (inode->i_state & ~I_REFERENCED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 return 0;
667 if (inode_has_buffers(inode))
668 return 0;
669 if (atomic_read(&inode->i_count))
670 return 0;
671 if (inode->i_data.nrpages)
672 return 0;
673 return 1;
674}
675
676/*
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000677 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
678 * This is called from the superblock shrinker function with a number of inodes
679 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
680 * then are freed outside inode_lock by dispose_list().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 *
682 * Any inodes which are pinned purely because of attached pagecache have their
Nick Piggin9e38d862010-10-23 06:55:17 -0400683 * pagecache removed. If the inode has metadata buffers attached to
684 * mapping->private_list then try to remove them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 *
Nick Piggin9e38d862010-10-23 06:55:17 -0400686 * If the inode has the I_REFERENCED flag set, then it means that it has been
687 * used recently - the flag is set in iput_final(). When we encounter such an
688 * inode, clear the flag and move it to the back of the LRU so it gets another
689 * pass through the LRU before it gets reclaimed. This is necessary because of
690 * the fact we are doing lazy LRU updates to minimise lock contention so the
691 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
692 * with this flag set because they are the inodes that are out of order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 */
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000694void prune_icache_sb(struct super_block *sb, int nr_to_scan)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
696 LIST_HEAD(freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 int nr_scanned;
698 unsigned long reap = 0;
699
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000700 spin_lock(&sb->s_inode_lru_lock);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000701 for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 struct inode *inode;
703
Dave Chinner98b745c2011-07-08 14:14:39 +1000704 if (list_empty(&sb->s_inode_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 break;
706
Dave Chinner98b745c2011-07-08 14:14:39 +1000707 inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708
Nick Piggin9e38d862010-10-23 06:55:17 -0400709 /*
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000710 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
Dave Chinner02afc412011-03-22 22:23:38 +1100711 * so use a trylock. If we fail to get the lock, just move the
712 * inode to the back of the list so we don't spin on it.
713 */
714 if (!spin_trylock(&inode->i_lock)) {
Christoph Hellwig62a3dde2011-10-28 10:03:41 +0200715 list_move_tail(&inode->i_lru, &sb->s_inode_lru);
Dave Chinner02afc412011-03-22 22:23:38 +1100716 continue;
717 }
718
719 /*
Nick Piggin9e38d862010-10-23 06:55:17 -0400720 * Referenced or dirty inodes are still in use. Give them
721 * another pass through the LRU as we canot reclaim them now.
722 */
723 if (atomic_read(&inode->i_count) ||
724 (inode->i_state & ~I_REFERENCED)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100725 list_del_init(&inode->i_lru);
Dave Chinnerf283c862011-03-22 22:23:39 +1100726 spin_unlock(&inode->i_lock);
Dave Chinner98b745c2011-07-08 14:14:39 +1000727 sb->s_nr_inodes_unused--;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000728 this_cpu_dec(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400729 continue;
730 }
731
732 /* recently referenced inodes get one more pass */
733 if (inode->i_state & I_REFERENCED) {
Nick Piggin9e38d862010-10-23 06:55:17 -0400734 inode->i_state &= ~I_REFERENCED;
Dave Chinner98b745c2011-07-08 14:14:39 +1000735 list_move(&inode->i_lru, &sb->s_inode_lru);
Dave Chinnerf283c862011-03-22 22:23:39 +1100736 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 continue;
738 }
739 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
740 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100741 spin_unlock(&inode->i_lock);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000742 spin_unlock(&sb->s_inode_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800744 reap += invalidate_mapping_pages(&inode->i_data,
745 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 iput(inode);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000747 spin_lock(&sb->s_inode_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
Dave Chinner98b745c2011-07-08 14:14:39 +1000749 if (inode != list_entry(sb->s_inode_lru.next,
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100750 struct inode, i_lru))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 continue; /* wrong inode or list_empty */
Dave Chinner02afc412011-03-22 22:23:38 +1100752 /* avoid lock inversions with trylock */
753 if (!spin_trylock(&inode->i_lock))
754 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100755 if (!can_unuse(inode)) {
756 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100758 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 }
Nick Piggin7ef0d732009-03-12 14:31:38 -0700760 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 inode->i_state |= I_FREEING;
Dave Chinner250df6e2011-03-22 22:23:36 +1100762 spin_unlock(&inode->i_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100763
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100764 list_move(&inode->i_lru, &freeable);
Dave Chinner98b745c2011-07-08 14:14:39 +1000765 sb->s_nr_inodes_unused--;
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000766 this_cpu_dec(nr_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 }
Christoph Lameterf8891e52006-06-30 01:55:45 -0700768 if (current_is_kswapd())
769 __count_vm_events(KSWAPD_INODESTEAL, reap);
770 else
771 __count_vm_events(PGINODESTEAL, reap);
Dave Chinner09cc9fc2011-07-08 14:14:40 +1000772 spin_unlock(&sb->s_inode_lru_lock);
Konstantin Khlebnikov5f8aefd2012-01-10 15:07:18 -0800773 if (current->reclaim_state)
774 current->reclaim_state->reclaimed_slab += reap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
776 dispose_list(&freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777}
778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779static void __wait_on_freeing_inode(struct inode *inode);
780/*
781 * Called with the inode lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530783static struct inode *find_inode(struct super_block *sb,
784 struct hlist_head *head,
785 int (*test)(struct inode *, void *),
786 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787{
788 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530789 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
791repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700792 hlist_for_each_entry(inode, node, head, i_hash) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100793 spin_lock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100794 if (inode->i_sb != sb) {
795 spin_unlock(&inode->i_lock);
796 continue;
797 }
798 if (!test(inode, data)) {
799 spin_unlock(&inode->i_lock);
800 continue;
801 }
Al Viroa4ffdde2010-06-02 17:38:30 -0400802 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 __wait_on_freeing_inode(inode);
804 goto repeat;
805 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400806 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100807 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400808 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400810 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811}
812
813/*
814 * find_inode_fast is the fast path version of find_inode, see the comment at
815 * iget_locked for details.
816 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530817static struct inode *find_inode_fast(struct super_block *sb,
818 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819{
820 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530821 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
823repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700824 hlist_for_each_entry(inode, node, head, i_hash) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100825 spin_lock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100826 if (inode->i_ino != ino) {
827 spin_unlock(&inode->i_lock);
828 continue;
829 }
830 if (inode->i_sb != sb) {
831 spin_unlock(&inode->i_lock);
832 continue;
833 }
Al Viroa4ffdde2010-06-02 17:38:30 -0400834 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 __wait_on_freeing_inode(inode);
836 goto repeat;
837 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400838 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100839 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400840 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400842 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843}
844
Eric Dumazetf991bd22010-10-23 11:18:01 -0400845/*
846 * Each cpu owns a range of LAST_INO_BATCH numbers.
847 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
848 * to renew the exhausted range.
David Chinner8290c352008-10-30 17:35:24 +1100849 *
Eric Dumazetf991bd22010-10-23 11:18:01 -0400850 * This does not significantly increase overflow rate because every CPU can
851 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
852 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
853 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
854 * overflow rate by 2x, which does not seem too significant.
855 *
856 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
857 * error if st_ino won't fit in target struct field. Use 32bit counter
858 * here to attempt to avoid that.
David Chinner8290c352008-10-30 17:35:24 +1100859 */
Eric Dumazetf991bd22010-10-23 11:18:01 -0400860#define LAST_INO_BATCH 1024
861static DEFINE_PER_CPU(unsigned int, last_ino);
David Chinner8290c352008-10-30 17:35:24 +1100862
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400863unsigned int get_next_ino(void)
Eric Dumazetf991bd22010-10-23 11:18:01 -0400864{
865 unsigned int *p = &get_cpu_var(last_ino);
866 unsigned int res = *p;
867
868#ifdef CONFIG_SMP
869 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
870 static atomic_t shared_last_ino;
871 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
872
873 res = next - LAST_INO_BATCH;
874 }
875#endif
876
877 *p = ++res;
878 put_cpu_var(last_ino);
879 return res;
David Chinner8290c352008-10-30 17:35:24 +1100880}
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400881EXPORT_SYMBOL(get_next_ino);
David Chinner8290c352008-10-30 17:35:24 +1100882
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883/**
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200884 * new_inode_pseudo - obtain an inode
885 * @sb: superblock
886 *
887 * Allocates a new inode for given superblock.
888 * Inode wont be chained in superblock s_inodes list
889 * This means :
890 * - fs can't be unmount
891 * - quotas, fsnotify, writeback can't work
892 */
893struct inode *new_inode_pseudo(struct super_block *sb)
894{
895 struct inode *inode = alloc_inode(sb);
896
897 if (inode) {
898 spin_lock(&inode->i_lock);
899 inode->i_state = 0;
900 spin_unlock(&inode->i_lock);
901 INIT_LIST_HEAD(&inode->i_sb_list);
902 }
903 return inode;
904}
905
906/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 * new_inode - obtain an inode
908 * @sb: superblock
909 *
Mel Gorman769848c2007-07-17 04:03:05 -0700910 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800911 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700912 * If HIGHMEM pages are unsuitable or it is known that pages allocated
913 * for the page cache are not reclaimable or migratable,
914 * mapping_set_gfp_mask() must be called with suitable flags on the
915 * newly created inode's mapping
916 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 */
918struct inode *new_inode(struct super_block *sb)
919{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530920 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
Dave Chinner55fa6092011-03-22 22:23:40 +1100922 spin_lock_prefetch(&inode_sb_list_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530923
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200924 inode = new_inode_pseudo(sb);
925 if (inode)
Dave Chinner55fa6092011-03-22 22:23:40 +1100926 inode_sb_list_add(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 return inode;
928}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929EXPORT_SYMBOL(new_inode);
930
Peter Zijlstra14358e62007-10-14 01:38:33 +0200931#ifdef CONFIG_DEBUG_LOCK_ALLOC
Josh Boyere096d0c2011-08-25 07:48:12 -0400932void lockdep_annotate_inode_mutex_key(struct inode *inode)
933{
Namhyung Kima3314a02010-10-11 22:38:00 +0900934 if (S_ISDIR(inode->i_mode)) {
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200935 struct file_system_type *type = inode->i_sb->s_type;
936
Jan Kara9a7aa122009-06-04 15:26:49 +0200937 /* Set new key only if filesystem hasn't already changed it */
Tyler Hicks978d6d82011-12-12 10:02:30 -0600938 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
Jan Kara9a7aa122009-06-04 15:26:49 +0200939 /*
940 * ensure nobody is actually holding i_mutex
941 */
942 mutex_destroy(&inode->i_mutex);
943 mutex_init(&inode->i_mutex);
944 lockdep_set_class(&inode->i_mutex,
945 &type->i_mutex_dir_key);
946 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200947 }
Josh Boyere096d0c2011-08-25 07:48:12 -0400948}
949EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200950#endif
Josh Boyere096d0c2011-08-25 07:48:12 -0400951
952/**
953 * unlock_new_inode - clear the I_NEW state and wake up any waiters
954 * @inode: new inode to unlock
955 *
956 * Called when the inode is fully initialised to clear the new state of the
957 * inode and wake up anyone waiting for the inode to finish initialisation.
958 */
959void unlock_new_inode(struct inode *inode)
960{
961 lockdep_annotate_inode_mutex_key(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100962 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100963 WARN_ON(!(inode->i_state & I_NEW));
964 inode->i_state &= ~I_NEW;
Al Viro310fa7a2012-03-10 17:07:28 -0500965 smp_mb();
Dave Chinner250df6e2011-03-22 22:23:36 +1100966 wake_up_bit(&inode->i_state, __I_NEW);
967 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969EXPORT_SYMBOL(unlock_new_inode);
970
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400971/**
972 * iget5_locked - obtain an inode from a mounted file system
973 * @sb: super block of file system
974 * @hashval: hash value (usually inode number) to get
975 * @test: callback used for comparisons between inodes
976 * @set: callback used to initialize a new struct inode
977 * @data: opaque data pointer to pass to @test and @set
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400979 * Search for the inode specified by @hashval and @data in the inode cache,
980 * and if present it is return it with an increased reference count. This is
981 * a generalized version of iget_locked() for file systems where the inode
982 * number is not sufficient for unique identification of an inode.
983 *
984 * If the inode is not in cache, allocate a new inode and return it locked,
985 * hashed, and with the I_NEW flag set. The file system gets to fill it in
986 * before unlocking it via unlock_new_inode().
987 *
988 * Note both @test and @set are called with the inode_hash_lock held, so can't
989 * sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400991struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
992 int (*test)(struct inode *, void *),
993 int (*set)(struct inode *, void *), void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400995 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530996 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400998 spin_lock(&inode_hash_lock);
999 inode = find_inode(sb, head, test, data);
1000 spin_unlock(&inode_hash_lock);
1001
1002 if (inode) {
1003 wait_on_inode(inode);
1004 return inode;
1005 }
1006
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 inode = alloc_inode(sb);
1008 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301009 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
Dave Chinner67a23c42011-03-22 22:23:42 +11001011 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 /* We released the lock, so.. */
1013 old = find_inode(sb, head, test, data);
1014 if (!old) {
1015 if (set(inode, data))
1016 goto set_failed;
1017
Dave Chinner250df6e2011-03-22 22:23:36 +11001018 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001019 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001020 hlist_add_head(&inode->i_hash, head);
1021 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001022 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001023 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
1025 /* Return the locked inode with I_NEW set, the
1026 * caller is responsible for filling in the contents
1027 */
1028 return inode;
1029 }
1030
1031 /*
1032 * Uhhuh, somebody else created the same inode under
1033 * us. Use the old inode instead of the one we just
1034 * allocated.
1035 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001036 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 destroy_inode(inode);
1038 inode = old;
1039 wait_on_inode(inode);
1040 }
1041 return inode;
1042
1043set_failed:
Dave Chinner67a23c42011-03-22 22:23:42 +11001044 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 destroy_inode(inode);
1046 return NULL;
1047}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001048EXPORT_SYMBOL(iget5_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001050/**
1051 * iget_locked - obtain an inode from a mounted file system
1052 * @sb: super block of file system
1053 * @ino: inode number to get
1054 *
1055 * Search for the inode specified by @ino in the inode cache and if present
1056 * return it with an increased reference count. This is for file systems
1057 * where the inode number is sufficient for unique identification of an inode.
1058 *
1059 * If the inode is not in cache, allocate a new inode and return it locked,
1060 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1061 * before unlocking it via unlock_new_inode().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001063struct inode *iget_locked(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001065 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301066 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001068 spin_lock(&inode_hash_lock);
1069 inode = find_inode_fast(sb, head, ino);
1070 spin_unlock(&inode_hash_lock);
1071 if (inode) {
1072 wait_on_inode(inode);
1073 return inode;
1074 }
1075
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 inode = alloc_inode(sb);
1077 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301078 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
Dave Chinner67a23c42011-03-22 22:23:42 +11001080 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 /* We released the lock, so.. */
1082 old = find_inode_fast(sb, head, ino);
1083 if (!old) {
1084 inode->i_ino = ino;
Dave Chinner250df6e2011-03-22 22:23:36 +11001085 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001086 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001087 hlist_add_head(&inode->i_hash, head);
1088 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001089 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001090 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
1092 /* Return the locked inode with I_NEW set, the
1093 * caller is responsible for filling in the contents
1094 */
1095 return inode;
1096 }
1097
1098 /*
1099 * Uhhuh, somebody else created the same inode under
1100 * us. Use the old inode instead of the one we just
1101 * allocated.
1102 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001103 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 destroy_inode(inode);
1105 inode = old;
1106 wait_on_inode(inode);
1107 }
1108 return inode;
1109}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001110EXPORT_SYMBOL(iget_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001112/*
1113 * search the inode cache for a matching inode number.
1114 * If we find one, then the inode number we are trying to
1115 * allocate is not unique and so we should not use it.
1116 *
1117 * Returns 1 if the inode number is unique, 0 if it is not.
1118 */
1119static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1120{
1121 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1122 struct hlist_node *node;
1123 struct inode *inode;
1124
Dave Chinner67a23c42011-03-22 22:23:42 +11001125 spin_lock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001126 hlist_for_each_entry(inode, node, b, i_hash) {
Dave Chinner67a23c42011-03-22 22:23:42 +11001127 if (inode->i_ino == ino && inode->i_sb == sb) {
1128 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001129 return 0;
Dave Chinner67a23c42011-03-22 22:23:42 +11001130 }
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001131 }
Dave Chinner67a23c42011-03-22 22:23:42 +11001132 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001133
1134 return 1;
1135}
1136
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137/**
1138 * iunique - get a unique inode number
1139 * @sb: superblock
1140 * @max_reserved: highest reserved inode number
1141 *
1142 * Obtain an inode number that is unique on the system for a given
1143 * superblock. This is used by file systems that have no natural
1144 * permanent inode numbering system. An inode number is returned that
1145 * is higher than the reserved limit but unique.
1146 *
1147 * BUGS:
1148 * With a large number of inodes live on the file system this function
1149 * currently becomes quite slow.
1150 */
1151ino_t iunique(struct super_block *sb, ino_t max_reserved)
1152{
Jeff Layton866b04f2007-05-08 00:32:29 -07001153 /*
1154 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1155 * error if st_ino won't fit in target struct field. Use 32bit counter
1156 * here to attempt to avoid that.
1157 */
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001158 static DEFINE_SPINLOCK(iunique_lock);
Jeff Layton866b04f2007-05-08 00:32:29 -07001159 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001162 spin_lock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001163 do {
1164 if (counter <= max_reserved)
1165 counter = max_reserved + 1;
1166 res = counter++;
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001167 } while (!test_inode_iunique(sb, res));
1168 spin_unlock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001169
1170 return res;
1171}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172EXPORT_SYMBOL(iunique);
1173
1174struct inode *igrab(struct inode *inode)
1175{
Dave Chinner250df6e2011-03-22 22:23:36 +11001176 spin_lock(&inode->i_lock);
1177 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001179 spin_unlock(&inode->i_lock);
1180 } else {
1181 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 /*
1183 * Handle the case where s_op->clear_inode is not been
1184 * called yet, and somebody is calling igrab
1185 * while the inode is getting freed.
1186 */
1187 inode = NULL;
Dave Chinner250df6e2011-03-22 22:23:36 +11001188 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 return inode;
1190}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191EXPORT_SYMBOL(igrab);
1192
1193/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001194 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 * @sb: super block of file system to search
1196 * @hashval: hash value (usually inode number) to search for
1197 * @test: callback used for comparisons between inodes
1198 * @data: opaque data pointer to pass to @test
1199 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001200 * Search for the inode specified by @hashval and @data in the inode cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 * If the inode is in the cache, the inode is returned with an incremented
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001202 * reference count.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001203 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001204 * Note: I_NEW is not waited upon so you have to be very careful what you do
1205 * with the returned inode. You probably should be using ilookup5() instead.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001206 *
Randy Dunlapb6d0ad62011-03-26 13:27:47 -07001207 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001208 */
1209struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1210 int (*test)(struct inode *, void *), void *data)
1211{
1212 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001213 struct inode *inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001214
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001215 spin_lock(&inode_hash_lock);
1216 inode = find_inode(sb, head, test, data);
1217 spin_unlock(&inode_hash_lock);
1218
1219 return inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001220}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001221EXPORT_SYMBOL(ilookup5_nowait);
1222
1223/**
1224 * ilookup5 - search for an inode in the inode cache
1225 * @sb: super block of file system to search
1226 * @hashval: hash value (usually inode number) to search for
1227 * @test: callback used for comparisons between inodes
1228 * @data: opaque data pointer to pass to @test
1229 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001230 * Search for the inode specified by @hashval and @data in the inode cache,
1231 * and if the inode is in the cache, return the inode with an incremented
1232 * reference count. Waits on I_NEW before returning the inode.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001233 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001235 * This is a generalized version of ilookup() for file systems where the
1236 * inode number is not sufficient for unique identification of an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001238 * Note: @test is called with the inode_hash_lock held, so can't sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 */
1240struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1241 int (*test)(struct inode *, void *), void *data)
1242{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001243 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001245 if (inode)
1246 wait_on_inode(inode);
1247 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249EXPORT_SYMBOL(ilookup5);
1250
1251/**
1252 * ilookup - search for an inode in the inode cache
1253 * @sb: super block of file system to search
1254 * @ino: inode number to search for
1255 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001256 * Search for the inode @ino in the inode cache, and if the inode is in the
1257 * cache, the inode is returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 */
1259struct inode *ilookup(struct super_block *sb, unsigned long ino)
1260{
1261 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001262 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001264 spin_lock(&inode_hash_lock);
1265 inode = find_inode_fast(sb, head, ino);
1266 spin_unlock(&inode_hash_lock);
1267
1268 if (inode)
1269 wait_on_inode(inode);
1270 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272EXPORT_SYMBOL(ilookup);
1273
Al Viro261bca82008-12-30 01:48:21 -05001274int insert_inode_locked(struct inode *inode)
1275{
1276 struct super_block *sb = inode->i_sb;
1277 ino_t ino = inode->i_ino;
1278 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001279
Al Viro261bca82008-12-30 01:48:21 -05001280 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001281 struct hlist_node *node;
1282 struct inode *old = NULL;
Dave Chinner67a23c42011-03-22 22:23:42 +11001283 spin_lock(&inode_hash_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001284 hlist_for_each_entry(old, node, head, i_hash) {
1285 if (old->i_ino != ino)
1286 continue;
1287 if (old->i_sb != sb)
1288 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001289 spin_lock(&old->i_lock);
1290 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1291 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001292 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001293 }
Al Viro72a43d62009-05-13 19:13:40 +01001294 break;
1295 }
1296 if (likely(!node)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001297 spin_lock(&inode->i_lock);
1298 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001299 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001300 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001301 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001302 return 0;
1303 }
1304 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001305 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001306 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001307 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001308 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001309 iput(old);
1310 return -EBUSY;
1311 }
1312 iput(old);
1313 }
1314}
Al Viro261bca82008-12-30 01:48:21 -05001315EXPORT_SYMBOL(insert_inode_locked);
1316
1317int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1318 int (*test)(struct inode *, void *), void *data)
1319{
1320 struct super_block *sb = inode->i_sb;
1321 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001322
Al Viro261bca82008-12-30 01:48:21 -05001323 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001324 struct hlist_node *node;
1325 struct inode *old = NULL;
1326
Dave Chinner67a23c42011-03-22 22:23:42 +11001327 spin_lock(&inode_hash_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001328 hlist_for_each_entry(old, node, head, i_hash) {
1329 if (old->i_sb != sb)
1330 continue;
1331 if (!test(old, data))
1332 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001333 spin_lock(&old->i_lock);
1334 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1335 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001336 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001337 }
Al Viro72a43d62009-05-13 19:13:40 +01001338 break;
1339 }
1340 if (likely(!node)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001341 spin_lock(&inode->i_lock);
1342 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001343 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001344 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001345 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001346 return 0;
1347 }
1348 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001349 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001350 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001351 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001352 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001353 iput(old);
1354 return -EBUSY;
1355 }
1356 iput(old);
1357 }
1358}
Al Viro261bca82008-12-30 01:48:21 -05001359EXPORT_SYMBOL(insert_inode_locked4);
1360
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
Al Viro45321ac2010-06-07 13:43:19 -04001362int generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363{
Al Viro45321ac2010-06-07 13:43:19 -04001364 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366EXPORT_SYMBOL(generic_delete_inode);
1367
Al Viro45321ac2010-06-07 13:43:19 -04001368/*
Al Viro45321ac2010-06-07 13:43:19 -04001369 * Called when we're dropping the last reference
1370 * to an inode.
1371 *
1372 * Call the FS "drop_inode()" function, defaulting to
1373 * the legacy UNIX filesystem behaviour. If it tells
1374 * us to evict inode, do so. Otherwise, retain inode
1375 * in cache if fs is alive, sync and evict if fs is
1376 * shutting down.
1377 */
1378static void iput_final(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379{
1380 struct super_block *sb = inode->i_sb;
Al Viro45321ac2010-06-07 13:43:19 -04001381 const struct super_operations *op = inode->i_sb->s_op;
1382 int drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
Dave Chinner250df6e2011-03-22 22:23:36 +11001384 WARN_ON(inode->i_state & I_NEW);
1385
Al Viroe7f59092011-07-07 15:45:59 -04001386 if (op->drop_inode)
Al Viro45321ac2010-06-07 13:43:19 -04001387 drop = op->drop_inode(inode);
1388 else
1389 drop = generic_drop_inode(inode);
1390
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001391 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1392 inode->i_state |= I_REFERENCED;
1393 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1394 inode_lru_list_add(inode);
1395 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001396 return;
1397 }
1398
Al Viro45321ac2010-06-07 13:43:19 -04001399 if (!drop) {
Alexander Viro991114c2005-06-23 00:09:01 -07001400 inode->i_state |= I_WILL_FREE;
Dave Chinner250df6e2011-03-22 22:23:36 +11001401 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 write_inode_now(inode, 1);
Dave Chinner250df6e2011-03-22 22:23:36 +11001403 spin_lock(&inode->i_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001404 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001405 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001407
Alexander Viro991114c2005-06-23 00:09:01 -07001408 inode->i_state |= I_FREEING;
Eric Dumazetc4ae0c62011-07-28 06:55:13 +02001409 if (!list_empty(&inode->i_lru))
1410 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001411 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001412
1413 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414}
1415
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301417 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 * @inode: inode to put
1419 *
1420 * Puts an inode, dropping its usage count. If the inode use count hits
1421 * zero, the inode is then freed and may also be destroyed.
1422 *
1423 * Consequently, iput() can sleep.
1424 */
1425void iput(struct inode *inode)
1426{
1427 if (inode) {
Al Viroa4ffdde2010-06-02 17:38:30 -04001428 BUG_ON(inode->i_state & I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
Dave Chinnerf283c862011-03-22 22:23:39 +11001430 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 iput_final(inode);
1432 }
1433}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434EXPORT_SYMBOL(iput);
1435
1436/**
1437 * bmap - find a block number in a file
1438 * @inode: inode of file
1439 * @block: block to find
1440 *
1441 * Returns the block number on the device holding the inode that
1442 * is the disk block number for the block of the file requested.
1443 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301444 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 * file.
1446 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301447sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448{
1449 sector_t res = 0;
1450 if (inode->i_mapping->a_ops->bmap)
1451 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1452 return res;
1453}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454EXPORT_SYMBOL(bmap);
1455
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001456/*
1457 * With relative atime, only update atime if the previous atime is
1458 * earlier than either the ctime or mtime or if at least a day has
1459 * passed since the last atime update.
1460 */
1461static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1462 struct timespec now)
1463{
1464
1465 if (!(mnt->mnt_flags & MNT_RELATIME))
1466 return 1;
1467 /*
1468 * Is mtime younger than atime? If yes, update atime:
1469 */
1470 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1471 return 1;
1472 /*
1473 * Is ctime younger than atime? If yes, update atime:
1474 */
1475 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1476 return 1;
1477
1478 /*
1479 * Is the previous atime value older than a day? If yes,
1480 * update atime:
1481 */
1482 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1483 return 1;
1484 /*
1485 * Good, we can skip the atime update:
1486 */
1487 return 0;
1488}
1489
Josef Bacikc3b2da32012-03-26 09:59:21 -04001490/*
1491 * This does the actual work of updating an inodes time or version. Must have
1492 * had called mnt_want_write() before calling this.
1493 */
1494static int update_time(struct inode *inode, struct timespec *time, int flags)
1495{
1496 if (inode->i_op->update_time)
1497 return inode->i_op->update_time(inode, time, flags);
1498
1499 if (flags & S_ATIME)
1500 inode->i_atime = *time;
1501 if (flags & S_VERSION)
1502 inode_inc_iversion(inode);
1503 if (flags & S_CTIME)
1504 inode->i_ctime = *time;
1505 if (flags & S_MTIME)
1506 inode->i_mtime = *time;
1507 mark_inode_dirty_sync(inode);
1508 return 0;
1509}
1510
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001512 * touch_atime - update the access time
Randy Dunlap185553b2012-04-17 17:03:25 -07001513 * @path: the &struct path to update
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 *
1515 * Update the accessed time on an inode and mark it for writeback.
1516 * This function automatically handles read only file systems and media,
1517 * as well as the "noatime" flag and inode specific "noatime" markers.
1518 */
Al Viro68ac1232012-03-15 08:21:57 -04001519void touch_atime(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520{
Al Viro68ac1232012-03-15 08:21:57 -04001521 struct vfsmount *mnt = path->mnt;
1522 struct inode *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 struct timespec now;
1524
Andrew Mortonb2276132006-12-13 00:34:33 -08001525 if (inode->i_flags & S_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001526 return;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001527 if (IS_NOATIME(inode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001528 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001529 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001530 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001531
Dave Hansencdb70f32008-02-15 14:37:41 -08001532 if (mnt->mnt_flags & MNT_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001533 return;
Dave Hansencdb70f32008-02-15 14:37:41 -08001534 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001535 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001536
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 now = current_fs_time(inode->i_sb);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001538
1539 if (!relatime_need_update(mnt, inode, now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001540 return;
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001541
Valerie Henson47ae32d2006-12-13 00:34:34 -08001542 if (timespec_equal(&inode->i_atime, &now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001543 return;
1544
Jan Kara5d37e9e2012-06-12 16:20:36 +02001545 if (!sb_start_write_trylock(inode->i_sb))
Andi Kleenb12536c2009-09-18 13:05:47 -07001546 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001547
Jan Kara5d37e9e2012-06-12 16:20:36 +02001548 if (__mnt_want_write(mnt))
1549 goto skip_update;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001550 /*
1551 * File systems can error out when updating inodes if they need to
1552 * allocate new space to modify an inode (such is the case for
1553 * Btrfs), but since we touch atime while walking down the path we
1554 * really don't care if we failed to update the atime of the file,
1555 * so just ignore the return value.
Alexander Block2bc5565282012-06-15 09:49:33 +02001556 * We may also fail on filesystems that have the ability to make parts
1557 * of the fs read only, e.g. subvolumes in Btrfs.
Josef Bacikc3b2da32012-03-26 09:59:21 -04001558 */
1559 update_time(inode, &now, S_ATIME);
Jan Kara5d37e9e2012-06-12 16:20:36 +02001560 __mnt_drop_write(mnt);
1561skip_update:
1562 sb_end_write(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001564EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565
Cong Wang3ed37642012-05-15 14:57:33 +08001566/*
1567 * The logic we want is
1568 *
1569 * if suid or (sgid and xgrp)
1570 * remove privs
1571 */
1572int should_remove_suid(struct dentry *dentry)
1573{
1574 umode_t mode = dentry->d_inode->i_mode;
1575 int kill = 0;
1576
1577 /* suid always must be killed */
1578 if (unlikely(mode & S_ISUID))
1579 kill = ATTR_KILL_SUID;
1580
1581 /*
1582 * sgid without any exec bits is just a mandatory locking mark; leave
1583 * it alone. If some exec bits are set, it's a real sgid; kill it.
1584 */
1585 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1586 kill |= ATTR_KILL_SGID;
1587
1588 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1589 return kill;
1590
1591 return 0;
1592}
1593EXPORT_SYMBOL(should_remove_suid);
1594
1595static int __remove_suid(struct dentry *dentry, int kill)
1596{
1597 struct iattr newattrs;
1598
1599 newattrs.ia_valid = ATTR_FORCE | kill;
1600 return notify_change(dentry, &newattrs);
1601}
1602
1603int file_remove_suid(struct file *file)
1604{
1605 struct dentry *dentry = file->f_path.dentry;
1606 struct inode *inode = dentry->d_inode;
1607 int killsuid;
1608 int killpriv;
1609 int error = 0;
1610
1611 /* Fast path for nothing security related */
1612 if (IS_NOSEC(inode))
1613 return 0;
1614
1615 killsuid = should_remove_suid(dentry);
1616 killpriv = security_inode_need_killpriv(dentry);
1617
1618 if (killpriv < 0)
1619 return killpriv;
1620 if (killpriv)
1621 error = security_inode_killpriv(dentry);
1622 if (!error && killsuid)
1623 error = __remove_suid(dentry, killsuid);
1624 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1625 inode->i_flags |= S_NOSEC;
1626
1627 return error;
1628}
1629EXPORT_SYMBOL(file_remove_suid);
1630
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001632 * file_update_time - update mtime and ctime time
1633 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001635 * Update the mtime and ctime members of an inode and mark the inode
1636 * for writeback. Note that this function is meant exclusively for
1637 * usage in the file write path of filesystems, and filesystems may
1638 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001639 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Josef Bacikc3b2da32012-03-26 09:59:21 -04001640 * timestamps are handled by the server. This can return an error for
1641 * file systems who need to allocate space in order to update an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 */
1643
Josef Bacikc3b2da32012-03-26 09:59:21 -04001644int file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001646 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 struct timespec now;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001648 int sync_it = 0;
1649 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
Andi Kleence06e0b2009-09-18 13:05:48 -07001651 /* First try to exhaust all avenues to not sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 if (IS_NOCMTIME(inode))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001653 return 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001654
Andi Kleence06e0b2009-09-18 13:05:48 -07001655 now = current_fs_time(inode->i_sb);
1656 if (!timespec_equal(&inode->i_mtime, &now))
1657 sync_it = S_MTIME;
1658
1659 if (!timespec_equal(&inode->i_ctime, &now))
1660 sync_it |= S_CTIME;
1661
1662 if (IS_I_VERSION(inode))
1663 sync_it |= S_VERSION;
1664
1665 if (!sync_it)
Josef Bacikc3b2da32012-03-26 09:59:21 -04001666 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
Andi Kleence06e0b2009-09-18 13:05:48 -07001668 /* Finally allowed to write? Takes lock. */
Jan Karaeb04c282012-06-12 16:20:35 +02001669 if (__mnt_want_write_file(file))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001670 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
Josef Bacikc3b2da32012-03-26 09:59:21 -04001672 ret = update_time(inode, &now, sync_it);
Jan Karaeb04c282012-06-12 16:20:35 +02001673 __mnt_drop_write_file(file);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001674
1675 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001677EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678
1679int inode_needs_sync(struct inode *inode)
1680{
1681 if (IS_SYNC(inode))
1682 return 1;
1683 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1684 return 1;
1685 return 0;
1686}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687EXPORT_SYMBOL(inode_needs_sync);
1688
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689int inode_wait(void *word)
1690{
1691 schedule();
1692 return 0;
1693}
Stephen Rothwelld44dab82008-11-10 17:06:05 +11001694EXPORT_SYMBOL(inode_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695
1696/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001697 * If we try to find an inode in the inode hash while it is being
1698 * deleted, we have to wait until the filesystem completes its
1699 * deletion before reporting that it isn't found. This function waits
1700 * until the deletion _might_ have completed. Callers are responsible
1701 * to recheck inode state.
1702 *
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001703 * It doesn't matter if I_NEW is not set initially, a call to
Dave Chinner250df6e2011-03-22 22:23:36 +11001704 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1705 * will DTRT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 */
1707static void __wait_on_freeing_inode(struct inode *inode)
1708{
1709 wait_queue_head_t *wq;
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001710 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1711 wq = bit_waitqueue(&inode->i_state, __I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +11001713 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001714 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 schedule();
1716 finish_wait(wq, &wait.wait);
Dave Chinner67a23c42011-03-22 22:23:42 +11001717 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718}
1719
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720static __initdata unsigned long ihash_entries;
1721static int __init set_ihash_entries(char *str)
1722{
1723 if (!str)
1724 return 0;
1725 ihash_entries = simple_strtoul(str, &str, 0);
1726 return 1;
1727}
1728__setup("ihash_entries=", set_ihash_entries);
1729
1730/*
1731 * Initialize the waitqueues and inode hash table.
1732 */
1733void __init inode_init_early(void)
1734{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001735 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736
1737 /* If hashes are distributed across NUMA nodes, defer
1738 * hash allocation until vmalloc space is available.
1739 */
1740 if (hashdist)
1741 return;
1742
1743 inode_hashtable =
1744 alloc_large_system_hash("Inode-cache",
1745 sizeof(struct hlist_head),
1746 ihash_entries,
1747 14,
1748 HASH_EARLY,
1749 &i_hash_shift,
1750 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001751 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 0);
1753
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001754 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1756}
1757
Denis Cheng74bf17c2007-10-16 23:26:30 -07001758void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001760 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761
1762 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001763 inode_cachep = kmem_cache_create("inode_cache",
1764 sizeof(struct inode),
1765 0,
1766 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1767 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001768 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770 /* Hash may have been set up in inode_init_early */
1771 if (!hashdist)
1772 return;
1773
1774 inode_hashtable =
1775 alloc_large_system_hash("Inode-cache",
1776 sizeof(struct hlist_head),
1777 ihash_entries,
1778 14,
1779 0,
1780 &i_hash_shift,
1781 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001782 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 0);
1784
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001785 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1787}
1788
1789void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1790{
1791 inode->i_mode = mode;
1792 if (S_ISCHR(mode)) {
1793 inode->i_fop = &def_chr_fops;
1794 inode->i_rdev = rdev;
1795 } else if (S_ISBLK(mode)) {
1796 inode->i_fop = &def_blk_fops;
1797 inode->i_rdev = rdev;
1798 } else if (S_ISFIFO(mode))
1799 inode->i_fop = &def_fifo_fops;
1800 else if (S_ISSOCK(mode))
1801 inode->i_fop = &bad_sock_fops;
1802 else
Manish Katiyaraf0d9ae2009-09-18 13:05:43 -07001803 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1804 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1805 inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806}
1807EXPORT_SYMBOL(init_special_inode);
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001808
1809/**
Ben Hutchingseaae668d2011-02-15 12:48:09 +00001810 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001811 * @inode: New inode
1812 * @dir: Directory inode
1813 * @mode: mode of the new inode
1814 */
1815void inode_init_owner(struct inode *inode, const struct inode *dir,
Al Viro62bb1092011-07-24 23:20:18 -04001816 umode_t mode)
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001817{
1818 inode->i_uid = current_fsuid();
1819 if (dir && dir->i_mode & S_ISGID) {
1820 inode->i_gid = dir->i_gid;
1821 if (S_ISDIR(mode))
1822 mode |= S_ISGID;
1823 } else
1824 inode->i_gid = current_fsgid();
1825 inode->i_mode = mode;
1826}
1827EXPORT_SYMBOL(inode_init_owner);
Serge E. Hallyne795b712011-03-23 16:43:25 -07001828
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001829/**
1830 * inode_owner_or_capable - check current task permissions to inode
1831 * @inode: inode being checked
1832 *
1833 * Return true if current either has CAP_FOWNER to the inode, or
1834 * owns the file.
Serge E. Hallyne795b712011-03-23 16:43:25 -07001835 */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001836bool inode_owner_or_capable(const struct inode *inode)
Serge E. Hallyne795b712011-03-23 16:43:25 -07001837{
Eric W. Biederman92361632012-02-08 07:07:50 -08001838 if (uid_eq(current_fsuid(), inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001839 return true;
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -08001840 if (inode_capable(inode, CAP_FOWNER))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001841 return true;
1842 return false;
1843}
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001844EXPORT_SYMBOL(inode_owner_or_capable);
Trond Myklebust1d59d612012-05-31 12:22:33 -04001845
1846/*
1847 * Direct i/o helper functions
1848 */
1849static void __inode_dio_wait(struct inode *inode)
1850{
1851 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1852 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1853
1854 do {
1855 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1856 if (atomic_read(&inode->i_dio_count))
1857 schedule();
1858 } while (atomic_read(&inode->i_dio_count));
1859 finish_wait(wq, &q.wait);
1860}
1861
1862/**
1863 * inode_dio_wait - wait for outstanding DIO requests to finish
1864 * @inode: inode to wait for
1865 *
1866 * Waits for all pending direct I/O requests to finish so that we can
1867 * proceed with a truncate or equivalent operation.
1868 *
1869 * Must be called under a lock that serializes taking new references
1870 * to i_dio_count, usually by inode->i_mutex.
1871 */
1872void inode_dio_wait(struct inode *inode)
1873{
1874 if (atomic_read(&inode->i_dio_count))
1875 __inode_dio_wait(inode);
1876}
1877EXPORT_SYMBOL(inode_dio_wait);
1878
1879/*
1880 * inode_dio_done - signal finish of a direct I/O requests
1881 * @inode: inode the direct I/O happens on
1882 *
1883 * This is called once we've finished processing a direct I/O request,
1884 * and is used to wake up callers waiting for direct I/O to be quiesced.
1885 */
1886void inode_dio_done(struct inode *inode)
1887{
1888 if (atomic_dec_and_test(&inode->i_dio_count))
1889 wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1890}
1891EXPORT_SYMBOL(inode_dio_done);