blob: 6eecb7ff0b9aa2dae0ebee49a1c73ddb3f2db398 [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 Chinnerbc3b14c2013-08-28 10:17:58 +100020#include <linux/list_lru.h>
Dave Chinnera66979a2011-03-22 22:23:41 +110021#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
23/*
Christoph Hellwig4b4563d2011-05-27 09:28:01 -040024 * Inode locking rules:
Dave Chinner250df6e2011-03-22 22:23:36 +110025 *
26 * inode->i_lock protects:
27 * inode->i_state, inode->i_hash, __iget()
Dave Chinnerbc3b14c2013-08-28 10:17:58 +100028 * Inode LRU list locks protect:
Dave Chinner98b745c2011-07-08 14:14:39 +100029 * inode->i_sb->s_inode_lru, inode->i_lru
Dave Chinner55fa6092011-03-22 22:23:40 +110030 * inode_sb_list_lock protects:
31 * sb->s_inodes, inode->i_sb_list
Christoph Hellwigf758eea2011-04-21 18:19:44 -060032 * bdi->wb.list_lock protects:
Dave Chinnera66979a2011-03-22 22:23:41 +110033 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
Dave Chinner67a23c42011-03-22 22:23:42 +110034 * inode_hash_lock protects:
35 * inode_hashtable, inode->i_hash
Dave Chinner250df6e2011-03-22 22:23:36 +110036 *
37 * Lock ordering:
Dave Chinner55fa6092011-03-22 22:23:40 +110038 *
39 * inode_sb_list_lock
40 * inode->i_lock
Dave Chinnerbc3b14c2013-08-28 10:17:58 +100041 * Inode LRU list locks
Dave Chinnera66979a2011-03-22 22:23:41 +110042 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060043 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110044 * inode->i_lock
Dave Chinner67a23c42011-03-22 22:23:42 +110045 *
46 * inode_hash_lock
47 * inode_sb_list_lock
48 * inode->i_lock
49 *
50 * iunique_lock
51 * inode_hash_lock
Dave Chinner250df6e2011-03-22 22:23:36 +110052 */
53
Eric Dumazetfa3536c2006-03-26 01:37:24 -080054static unsigned int i_hash_mask __read_mostly;
55static unsigned int i_hash_shift __read_mostly;
Dave Chinner67a23c42011-03-22 22:23:42 +110056static struct hlist_head *inode_hashtable __read_mostly;
57static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Dave Chinner55fa6092011-03-22 22:23:40 +110059__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
60
Linus Torvalds1da177e2005-04-16 15:20:36 -070061/*
Jens Axboe7dcda1c2011-04-05 23:51:48 +020062 * Empty aops. Can be used for the cases where the user does not
63 * define any of the address_space operations.
64 */
65const struct address_space_operations empty_aops = {
66};
67EXPORT_SYMBOL(empty_aops);
68
69/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 * Statistics gathering..
71 */
72struct inodes_stat_t inodes_stat;
73
Glauber Costa3942c072013-08-28 10:17:53 +100074static DEFINE_PER_CPU(unsigned long, nr_inodes);
75static DEFINE_PER_CPU(unsigned long, nr_unused);
Dave Chinnercffbc8a2010-10-23 05:03:02 -040076
Manish Katiyar6b3304b2009-03-31 19:35:54 +053077static struct kmem_cache *inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
Glauber Costa3942c072013-08-28 10:17:53 +100079static long get_nr_inodes(void)
Dave Chinnercffbc8a2010-10-23 05:03:02 -040080{
Nick Piggin3e880fb2011-01-07 17:49:19 +110081 int i;
Glauber Costa3942c072013-08-28 10:17:53 +100082 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +110083 for_each_possible_cpu(i)
84 sum += per_cpu(nr_inodes, i);
85 return sum < 0 ? 0 : sum;
Dave Chinnercffbc8a2010-10-23 05:03:02 -040086}
87
Glauber Costa3942c072013-08-28 10:17:53 +100088static inline long get_nr_inodes_unused(void)
Dave Chinnercffbc8a2010-10-23 05:03:02 -040089{
Dave Chinnerfcb94f72011-07-08 14:14:38 +100090 int i;
Glauber Costa3942c072013-08-28 10:17:53 +100091 long sum = 0;
Dave Chinnerfcb94f72011-07-08 14:14:38 +100092 for_each_possible_cpu(i)
93 sum += per_cpu(nr_unused, i);
94 return sum < 0 ? 0 : sum;
Dave Chinnercffbc8a2010-10-23 05:03:02 -040095}
96
Glauber Costa3942c072013-08-28 10:17:53 +100097long get_nr_dirty_inodes(void)
Dave Chinnercffbc8a2010-10-23 05:03:02 -040098{
Nick Piggin3e880fb2011-01-07 17:49:19 +110099 /* not actually dirty inodes, but a wild approximation */
Glauber Costa3942c072013-08-28 10:17:53 +1000100 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400101 return nr_dirty > 0 ? nr_dirty : 0;
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400102}
103
104/*
105 * Handle nr_inode sysctl
106 */
107#ifdef CONFIG_SYSCTL
Joe Perches1f7e0612014-06-06 14:38:05 -0700108int proc_nr_inodes(struct ctl_table *table, int write,
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400109 void __user *buffer, size_t *lenp, loff_t *ppos)
110{
111 inodes_stat.nr_inodes = get_nr_inodes();
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000112 inodes_stat.nr_unused = get_nr_inodes_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000113 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400114}
115#endif
116
David Chinner2cb15992008-10-30 17:32:23 +1100117/**
118 * inode_init_always - perform inode structure intialisation
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800119 * @sb: superblock inode belongs to
120 * @inode: inode to initialise
David Chinner2cb15992008-10-30 17:32:23 +1100121 *
122 * These are initializations that need to be done on every inode
123 * allocation as the fields are not initialised by slab allocation.
124 */
Christoph Hellwig54e34622009-08-07 14:38:25 -0300125int inode_init_always(struct super_block *sb, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126{
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700127 static const struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800128 static const struct file_operations empty_fops;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530129 struct address_space *const mapping = &inode->i_data;
David Chinner2cb15992008-10-30 17:32:23 +1100130
131 inode->i_sb = sb;
132 inode->i_blkbits = sb->s_blocksize_bits;
133 inode->i_flags = 0;
134 atomic_set(&inode->i_count, 1);
135 inode->i_op = &empty_iops;
136 inode->i_fop = &empty_fops;
Miklos Szeredia78ef702011-10-28 14:13:30 +0200137 inode->__i_nlink = 1;
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700138 inode->i_opflags = 0;
Eric W. Biederman92361632012-02-08 07:07:50 -0800139 i_uid_write(inode, 0);
140 i_gid_write(inode, 0);
David Chinner2cb15992008-10-30 17:32:23 +1100141 atomic_set(&inode->i_writecount, 0);
142 inode->i_size = 0;
143 inode->i_blocks = 0;
144 inode->i_bytes = 0;
145 inode->i_generation = 0;
146#ifdef CONFIG_QUOTA
147 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
148#endif
149 inode->i_pipe = NULL;
150 inode->i_bdev = NULL;
151 inode->i_cdev = NULL;
152 inode->i_rdev = 0;
153 inode->dirtied_when = 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500154
155 if (security_inode_alloc(inode))
Christoph Hellwig54e34622009-08-07 14:38:25 -0300156 goto out;
David Chinner2cb15992008-10-30 17:32:23 +1100157 spin_lock_init(&inode->i_lock);
158 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
159
160 mutex_init(&inode->i_mutex);
161 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
162
Christoph Hellwigbd5fe6c2011-06-24 14:29:43 -0400163 atomic_set(&inode->i_dio_count, 0);
David Chinner2cb15992008-10-30 17:32:23 +1100164
165 mapping->a_ops = &empty_aops;
166 mapping->host = inode;
167 mapping->flags = 0;
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800168 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800169 mapping->private_data = NULL;
David Chinner2cb15992008-10-30 17:32:23 +1100170 mapping->backing_dev_info = &default_backing_dev_info;
171 mapping->writeback_index = 0;
172
173 /*
174 * If the block_device provides a backing_dev_info for client
175 * inodes then use that. Otherwise the inode share the bdev's
176 * backing_dev_info.
177 */
178 if (sb->s_bdev) {
179 struct backing_dev_info *bdi;
180
Jens Axboe2c96ce92009-09-15 09:43:56 +0200181 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
David Chinner2cb15992008-10-30 17:32:23 +1100182 mapping->backing_dev_info = bdi;
183 }
184 inode->i_private = NULL;
185 inode->i_mapping = mapping;
Al Virob3d9b7a2012-06-09 13:51:19 -0400186 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
Al Virof19d4a82009-06-08 19:50:45 -0400187#ifdef CONFIG_FS_POSIX_ACL
188 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
189#endif
David Chinner2cb15992008-10-30 17:32:23 +1100190
Eric Paris3be25f42009-05-21 17:01:26 -0400191#ifdef CONFIG_FSNOTIFY
192 inode->i_fsnotify_mask = 0;
193#endif
194
Nick Piggin3e880fb2011-01-07 17:49:19 +1100195 this_cpu_inc(nr_inodes);
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400196
Christoph Hellwig54e34622009-08-07 14:38:25 -0300197 return 0;
Christoph Hellwig54e34622009-08-07 14:38:25 -0300198out:
199 return -ENOMEM;
David Chinner2cb15992008-10-30 17:32:23 +1100200}
201EXPORT_SYMBOL(inode_init_always);
202
203static struct inode *alloc_inode(struct super_block *sb)
204{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 struct inode *inode;
206
207 if (sb->s_op->alloc_inode)
208 inode = sb->s_op->alloc_inode(sb);
209 else
David Chinner2cb15992008-10-30 17:32:23 +1100210 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
Christoph Hellwig54e34622009-08-07 14:38:25 -0300212 if (!inode)
213 return NULL;
214
215 if (unlikely(inode_init_always(sb, inode))) {
216 if (inode->i_sb->s_op->destroy_inode)
217 inode->i_sb->s_op->destroy_inode(inode);
218 else
219 kmem_cache_free(inode_cachep, inode);
220 return NULL;
221 }
222
223 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225
Nick Pigginff0c7d12011-01-07 17:49:50 +1100226void free_inode_nonrcu(struct inode *inode)
227{
228 kmem_cache_free(inode_cachep, inode);
229}
230EXPORT_SYMBOL(free_inode_nonrcu);
231
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300232void __destroy_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200234 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 security_inode_free(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400236 fsnotify_inode_delete(inode);
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100237 if (!inode->i_nlink) {
238 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
239 atomic_long_dec(&inode->i_sb->s_remove_count);
240 }
241
Al Virof19d4a82009-06-08 19:50:45 -0400242#ifdef CONFIG_FS_POSIX_ACL
243 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
244 posix_acl_release(inode->i_acl);
245 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
246 posix_acl_release(inode->i_default_acl);
247#endif
Nick Piggin3e880fb2011-01-07 17:49:19 +1100248 this_cpu_dec(nr_inodes);
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300249}
250EXPORT_SYMBOL(__destroy_inode);
251
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100252static void i_callback(struct rcu_head *head)
253{
254 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100255 kmem_cache_free(inode_cachep, inode);
256}
257
Christoph Hellwig56b0dac2010-10-06 10:48:55 +0200258static void destroy_inode(struct inode *inode)
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300259{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100260 BUG_ON(!list_empty(&inode->i_lru));
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300261 __destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 if (inode->i_sb->s_op->destroy_inode)
263 inode->i_sb->s_op->destroy_inode(inode);
264 else
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100265 call_rcu(&inode->i_rcu, i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100268/**
269 * drop_nlink - directly drop an inode's link count
270 * @inode: inode
271 *
272 * This is a low-level filesystem helper to replace any
273 * direct filesystem manipulation of i_nlink. In cases
274 * where we are attempting to track writes to the
275 * filesystem, a decrement to zero means an imminent
276 * write when the file is truncated and actually unlinked
277 * on the filesystem.
278 */
279void drop_nlink(struct inode *inode)
280{
281 WARN_ON(inode->i_nlink == 0);
282 inode->__i_nlink--;
283 if (!inode->i_nlink)
284 atomic_long_inc(&inode->i_sb->s_remove_count);
285}
286EXPORT_SYMBOL(drop_nlink);
287
288/**
289 * clear_nlink - directly zero an inode's link count
290 * @inode: inode
291 *
292 * This is a low-level filesystem helper to replace any
293 * direct filesystem manipulation of i_nlink. See
294 * drop_nlink() for why we care about i_nlink hitting zero.
295 */
296void clear_nlink(struct inode *inode)
297{
298 if (inode->i_nlink) {
299 inode->__i_nlink = 0;
300 atomic_long_inc(&inode->i_sb->s_remove_count);
301 }
302}
303EXPORT_SYMBOL(clear_nlink);
304
305/**
306 * set_nlink - directly set an inode's link count
307 * @inode: inode
308 * @nlink: new nlink (should be non-zero)
309 *
310 * This is a low-level filesystem helper to replace any
311 * direct filesystem manipulation of i_nlink.
312 */
313void set_nlink(struct inode *inode, unsigned int nlink)
314{
315 if (!nlink) {
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100316 clear_nlink(inode);
317 } else {
318 /* Yes, some filesystems do change nlink from zero to one */
319 if (inode->i_nlink == 0)
320 atomic_long_dec(&inode->i_sb->s_remove_count);
321
322 inode->__i_nlink = nlink;
323 }
324}
325EXPORT_SYMBOL(set_nlink);
326
327/**
328 * inc_nlink - directly increment an inode's link count
329 * @inode: inode
330 *
331 * This is a low-level filesystem helper to replace any
332 * direct filesystem manipulation of i_nlink. Currently,
333 * it is only here for parity with dec_nlink().
334 */
335void inc_nlink(struct inode *inode)
336{
Al Virof4e0c302013-06-11 08:34:36 +0400337 if (unlikely(inode->i_nlink == 0)) {
338 WARN_ON(!(inode->i_state & I_LINKABLE));
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100339 atomic_long_dec(&inode->i_sb->s_remove_count);
Al Virof4e0c302013-06-11 08:34:36 +0400340 }
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100341
342 inode->__i_nlink++;
343}
344EXPORT_SYMBOL(inc_nlink);
345
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100346void address_space_init_once(struct address_space *mapping)
347{
348 memset(mapping, 0, sizeof(*mapping));
349 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
350 spin_lock_init(&mapping->tree_lock);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700351 mutex_init(&mapping->i_mmap_mutex);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100352 INIT_LIST_HEAD(&mapping->private_list);
353 spin_lock_init(&mapping->private_lock);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700354 mapping->i_mmap = RB_ROOT;
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100355 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100356}
357EXPORT_SYMBOL(address_space_init_once);
358
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359/*
360 * These are initializations that only need to be done
361 * once, because the fields are idempotent across use
362 * of the inode, so let the slab aware of that.
363 */
364void inode_init_once(struct inode *inode)
365{
366 memset(inode, 0, sizeof(*inode));
367 INIT_HLIST_NODE(&inode->i_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 INIT_LIST_HEAD(&inode->i_devices);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100369 INIT_LIST_HEAD(&inode->i_wb_list);
370 INIT_LIST_HEAD(&inode->i_lru);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100371 address_space_init_once(&inode->i_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 i_size_ordered_init(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400373#ifdef CONFIG_FSNOTIFY
Eric Parise61ce862009-12-17 21:24:24 -0500374 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
Eric Paris3be25f42009-05-21 17:01:26 -0400375#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377EXPORT_SYMBOL(inode_init_once);
378
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700379static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530381 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
Christoph Lametera35afb82007-05-16 22:10:57 -0700383 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384}
385
386/*
Dave Chinner250df6e2011-03-22 22:23:36 +1100387 * inode->i_lock must be held
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530389void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390{
Nick Piggin9e38d862010-10-23 06:55:17 -0400391 atomic_inc(&inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392}
393
Al Viro7de9c6ee2010-10-23 11:11:40 -0400394/*
395 * get additional reference to inode; caller must already hold one.
396 */
397void ihold(struct inode *inode)
398{
399 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
400}
401EXPORT_SYMBOL(ihold);
402
Nick Piggin9e38d862010-10-23 06:55:17 -0400403static void inode_lru_list_add(struct inode *inode)
404{
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000405 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000406 this_cpu_inc(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400407}
408
Jan Kara4eff96d2012-11-26 16:29:51 -0800409/*
410 * Add inode to LRU if needed (inode is unused and clean).
411 *
412 * Needs inode->i_lock held.
413 */
414void inode_add_lru(struct inode *inode)
415{
416 if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) &&
417 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
418 inode_lru_list_add(inode);
419}
420
421
Nick Piggin9e38d862010-10-23 06:55:17 -0400422static void inode_lru_list_del(struct inode *inode)
423{
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000424
425 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000426 this_cpu_dec(nr_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427}
428
Christoph Hellwig646ec462010-10-23 07:15:32 -0400429/**
430 * inode_sb_list_add - add inode to the superblock list of inodes
431 * @inode: inode to add
432 */
433void inode_sb_list_add(struct inode *inode)
434{
Dave Chinner55fa6092011-03-22 22:23:40 +1100435 spin_lock(&inode_sb_list_lock);
436 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
437 spin_unlock(&inode_sb_list_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400438}
439EXPORT_SYMBOL_GPL(inode_sb_list_add);
440
Dave Chinner55fa6092011-03-22 22:23:40 +1100441static inline void inode_sb_list_del(struct inode *inode)
Christoph Hellwig646ec462010-10-23 07:15:32 -0400442{
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200443 if (!list_empty(&inode->i_sb_list)) {
444 spin_lock(&inode_sb_list_lock);
445 list_del_init(&inode->i_sb_list);
446 spin_unlock(&inode_sb_list_lock);
447 }
Christoph Hellwig646ec462010-10-23 07:15:32 -0400448}
449
Dave Chinner4c51acb2010-10-23 06:58:09 -0400450static unsigned long hash(struct super_block *sb, unsigned long hashval)
451{
452 unsigned long tmp;
453
454 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
455 L1_CACHE_BYTES;
Christoph Hellwig4b4563d2011-05-27 09:28:01 -0400456 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
457 return tmp & i_hash_mask;
Dave Chinner4c51acb2010-10-23 06:58:09 -0400458}
459
460/**
461 * __insert_inode_hash - hash an inode
462 * @inode: unhashed inode
463 * @hashval: unsigned long value used to locate this object in the
464 * inode_hashtable.
465 *
466 * Add an inode to the inode hash for this superblock.
467 */
468void __insert_inode_hash(struct inode *inode, unsigned long hashval)
469{
Christoph Hellwig646ec462010-10-23 07:15:32 -0400470 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
471
Dave Chinner67a23c42011-03-22 22:23:42 +1100472 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100473 spin_lock(&inode->i_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400474 hlist_add_head(&inode->i_hash, b);
Dave Chinner250df6e2011-03-22 22:23:36 +1100475 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100476 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400477}
478EXPORT_SYMBOL(__insert_inode_hash);
479
480/**
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200481 * __remove_inode_hash - remove an inode from the hash
Dave Chinner4c51acb2010-10-23 06:58:09 -0400482 * @inode: inode to unhash
483 *
484 * Remove an inode from the superblock.
485 */
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200486void __remove_inode_hash(struct inode *inode)
Dave Chinner4c51acb2010-10-23 06:58:09 -0400487{
Dave Chinner67a23c42011-03-22 22:23:42 +1100488 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100489 spin_lock(&inode->i_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400490 hlist_del_init(&inode->i_hash);
Dave Chinner250df6e2011-03-22 22:23:36 +1100491 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100492 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400493}
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200494EXPORT_SYMBOL(__remove_inode_hash);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400495
Jan Karadbd57682012-05-03 14:48:02 +0200496void clear_inode(struct inode *inode)
Al Virob0683aa2010-06-04 20:55:25 -0400497{
498 might_sleep();
Jan Kara08142572011-06-27 16:18:10 -0700499 /*
500 * We have to cycle tree_lock here because reclaim can be still in the
501 * process of removing the last page (in __delete_from_page_cache())
502 * and we must not free mapping under it.
503 */
504 spin_lock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400505 BUG_ON(inode->i_data.nrpages);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700506 BUG_ON(inode->i_data.nrshadows);
Jan Kara08142572011-06-27 16:18:10 -0700507 spin_unlock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400508 BUG_ON(!list_empty(&inode->i_data.private_list));
509 BUG_ON(!(inode->i_state & I_FREEING));
510 BUG_ON(inode->i_state & I_CLEAR);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100511 /* don't need i_lock here, no concurrent mods to i_state */
Al Virob0683aa2010-06-04 20:55:25 -0400512 inode->i_state = I_FREEING | I_CLEAR;
513}
Jan Karadbd57682012-05-03 14:48:02 +0200514EXPORT_SYMBOL(clear_inode);
Al Virob0683aa2010-06-04 20:55:25 -0400515
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100516/*
517 * Free the inode passed in, removing it from the lists it is still connected
518 * to. We remove any pages still attached to the inode and wait for any IO that
519 * is still in progress before finally destroying the inode.
520 *
521 * An inode must already be marked I_FREEING so that we avoid the inode being
522 * moved back onto lists if we race with other code that manipulates the lists
523 * (e.g. writeback_single_inode). The caller is responsible for setting this.
524 *
525 * An inode must already be removed from the LRU list before being evicted from
526 * the cache. This should occur atomically with setting the I_FREEING state
527 * flag, so no inodes here should ever be on the LRU when being evicted.
528 */
Al Viro644da592010-06-07 13:21:05 -0400529static void evict(struct inode *inode)
Al Virob4272d42010-06-04 19:33:20 -0400530{
531 const struct super_operations *op = inode->i_sb->s_op;
532
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100533 BUG_ON(!(inode->i_state & I_FREEING));
534 BUG_ON(!list_empty(&inode->i_lru));
535
Eric Dumazetb12362b2011-07-28 06:11:47 +0200536 if (!list_empty(&inode->i_wb_list))
537 inode_wb_list_del(inode);
538
Dave Chinner55fa6092011-03-22 22:23:40 +1100539 inode_sb_list_del(inode);
540
Jan Kara169ebd92012-05-03 14:48:03 +0200541 /*
542 * Wait for flusher thread to be done with the inode so that filesystem
543 * does not start destroying it while writeback is still running. Since
544 * the inode has I_FREEING set, flusher thread won't start new work on
545 * the inode. We just have to wait for running writeback to finish.
546 */
547 inode_wait_for_writeback(inode);
Jan Kara7994e6f2012-05-03 14:48:01 +0200548
Al Virobe7ce412010-06-04 19:40:39 -0400549 if (op->evict_inode) {
550 op->evict_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400551 } else {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700552 truncate_inode_pages_final(&inode->i_data);
Jan Karadbd57682012-05-03 14:48:02 +0200553 clear_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400554 }
Al Viro661074e2010-06-04 20:19:55 -0400555 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
556 bd_forget(inode);
557 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
558 cd_forget(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100559
560 remove_inode_hash(inode);
561
562 spin_lock(&inode->i_lock);
563 wake_up_bit(&inode->i_state, __I_NEW);
564 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
565 spin_unlock(&inode->i_lock);
566
567 destroy_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400568}
569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570/*
571 * dispose_list - dispose of the contents of a local list
572 * @head: the head of the list to free
573 *
574 * Dispose-list gets a local list with local inodes in it, so it doesn't
575 * need to worry about list corruption and SMP locks.
576 */
577static void dispose_list(struct list_head *head)
578{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 while (!list_empty(head)) {
580 struct inode *inode;
581
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100582 inode = list_first_entry(head, struct inode, i_lru);
583 list_del_init(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Al Viro644da592010-06-07 13:21:05 -0400585 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587}
588
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589/**
Al Viro63997e92010-10-25 20:49:35 -0400590 * evict_inodes - evict all evictable inodes for a superblock
591 * @sb: superblock to operate on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 *
Al Viro63997e92010-10-25 20:49:35 -0400593 * Make sure that no inodes with zero refcount are retained. This is
594 * called by superblock shutdown after having MS_ACTIVE flag removed,
595 * so any inode reaching zero refcount during or after that call will
596 * be immediately evicted.
597 */
598void evict_inodes(struct super_block *sb)
599{
600 struct inode *inode, *next;
601 LIST_HEAD(dispose);
602
Dave Chinner55fa6092011-03-22 22:23:40 +1100603 spin_lock(&inode_sb_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400604 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
605 if (atomic_read(&inode->i_count))
606 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100607
608 spin_lock(&inode->i_lock);
609 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
610 spin_unlock(&inode->i_lock);
Al Viro63997e92010-10-25 20:49:35 -0400611 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100612 }
Al Viro63997e92010-10-25 20:49:35 -0400613
614 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100615 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100616 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100617 list_add(&inode->i_lru, &dispose);
Al Viro63997e92010-10-25 20:49:35 -0400618 }
Dave Chinner55fa6092011-03-22 22:23:40 +1100619 spin_unlock(&inode_sb_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400620
621 dispose_list(&dispose);
Al Viro63997e92010-10-25 20:49:35 -0400622}
623
624/**
Christoph Hellwiga0318782010-10-24 19:40:33 +0200625 * invalidate_inodes - attempt to free all inodes on a superblock
626 * @sb: superblock to operate on
NeilBrown93b270f2011-02-24 17:25:47 +1100627 * @kill_dirty: flag to guide handling of dirty inodes
Christoph Hellwiga0318782010-10-24 19:40:33 +0200628 *
629 * Attempts to free all inodes for a given superblock. If there were any
630 * busy inodes return a non-zero value, else zero.
NeilBrown93b270f2011-02-24 17:25:47 +1100631 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
632 * them as busy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 */
NeilBrown93b270f2011-02-24 17:25:47 +1100634int invalidate_inodes(struct super_block *sb, bool kill_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635{
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400636 int busy = 0;
Christoph Hellwiga0318782010-10-24 19:40:33 +0200637 struct inode *inode, *next;
638 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
Dave Chinner55fa6092011-03-22 22:23:40 +1100640 spin_lock(&inode_sb_list_lock);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200641 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100642 spin_lock(&inode->i_lock);
643 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
644 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100646 }
NeilBrown93b270f2011-02-24 17:25:47 +1100647 if (inode->i_state & I_DIRTY && !kill_dirty) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100648 spin_unlock(&inode->i_lock);
NeilBrown93b270f2011-02-24 17:25:47 +1100649 busy = 1;
650 continue;
651 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200652 if (atomic_read(&inode->i_count)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100653 spin_unlock(&inode->i_lock);
Christoph Hellwig99a38912010-10-23 19:07:20 +0200654 busy = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 continue;
656 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200657
Christoph Hellwig99a38912010-10-23 19:07:20 +0200658 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100659 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100660 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100661 list_add(&inode->i_lru, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 }
Dave Chinner55fa6092011-03-22 22:23:40 +1100663 spin_unlock(&inode_sb_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
Christoph Hellwiga0318782010-10-24 19:40:33 +0200665 dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666
667 return busy;
668}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670/*
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000671 * Isolate the inode from the LRU in preparation for freeing it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 *
673 * Any inodes which are pinned purely because of attached pagecache have their
Nick Piggin9e38d862010-10-23 06:55:17 -0400674 * pagecache removed. If the inode has metadata buffers attached to
675 * mapping->private_list then try to remove them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 *
Nick Piggin9e38d862010-10-23 06:55:17 -0400677 * If the inode has the I_REFERENCED flag set, then it means that it has been
678 * used recently - the flag is set in iput_final(). When we encounter such an
679 * inode, clear the flag and move it to the back of the LRU so it gets another
680 * pass through the LRU before it gets reclaimed. This is necessary because of
681 * the fact we are doing lazy LRU updates to minimise lock contention so the
682 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
683 * with this flag set because they are the inodes that are out of order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 */
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000685static enum lru_status
686inode_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
687{
688 struct list_head *freeable = arg;
689 struct inode *inode = container_of(item, struct inode, i_lru);
690
691 /*
692 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
693 * If we fail to get the lock, just skip it.
694 */
695 if (!spin_trylock(&inode->i_lock))
696 return LRU_SKIP;
697
698 /*
699 * Referenced or dirty inodes are still in use. Give them another pass
700 * through the LRU as we canot reclaim them now.
701 */
702 if (atomic_read(&inode->i_count) ||
703 (inode->i_state & ~I_REFERENCED)) {
704 list_del_init(&inode->i_lru);
705 spin_unlock(&inode->i_lock);
706 this_cpu_dec(nr_unused);
707 return LRU_REMOVED;
708 }
709
710 /* recently referenced inodes get one more pass */
711 if (inode->i_state & I_REFERENCED) {
712 inode->i_state &= ~I_REFERENCED;
713 spin_unlock(&inode->i_lock);
714 return LRU_ROTATE;
715 }
716
717 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
718 __iget(inode);
719 spin_unlock(&inode->i_lock);
720 spin_unlock(lru_lock);
721 if (remove_inode_buffers(inode)) {
722 unsigned long reap;
723 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
724 if (current_is_kswapd())
725 __count_vm_events(KSWAPD_INODESTEAL, reap);
726 else
727 __count_vm_events(PGINODESTEAL, reap);
728 if (current->reclaim_state)
729 current->reclaim_state->reclaimed_slab += reap;
730 }
731 iput(inode);
732 spin_lock(lru_lock);
733 return LRU_RETRY;
734 }
735
736 WARN_ON(inode->i_state & I_NEW);
737 inode->i_state |= I_FREEING;
Glauber Costad38fa692013-08-28 10:17:59 +1000738 list_move(&inode->i_lru, freeable);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000739 spin_unlock(&inode->i_lock);
740
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000741 this_cpu_dec(nr_unused);
742 return LRU_REMOVED;
743}
744
745/*
746 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
747 * This is called from the superblock shrinker function with a number of inodes
748 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
749 * then are freed outside inode_lock by dispose_list().
750 */
Dave Chinner9b17c622013-08-28 10:18:05 +1000751long prune_icache_sb(struct super_block *sb, unsigned long nr_to_scan,
752 int nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753{
754 LIST_HEAD(freeable);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000755 long freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
Dave Chinner9b17c622013-08-28 10:18:05 +1000757 freed = list_lru_walk_node(&sb->s_inode_lru, nid, inode_lru_isolate,
758 &freeable, &nr_to_scan);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 dispose_list(&freeable);
Dave Chinner0a234c62013-08-28 10:17:57 +1000760 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761}
762
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763static void __wait_on_freeing_inode(struct inode *inode);
764/*
765 * Called with the inode lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530767static struct inode *find_inode(struct super_block *sb,
768 struct hlist_head *head,
769 int (*test)(struct inode *, void *),
770 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530772 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
774repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800775 hlist_for_each_entry(inode, head, i_hash) {
Al Viro5a3cd99282013-11-06 09:54:52 -0500776 if (inode->i_sb != sb)
777 continue;
778 if (!test(inode, data))
779 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100780 spin_lock(&inode->i_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -0400781 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 __wait_on_freeing_inode(inode);
783 goto repeat;
784 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400785 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100786 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400787 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400789 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790}
791
792/*
793 * find_inode_fast is the fast path version of find_inode, see the comment at
794 * iget_locked for details.
795 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530796static struct inode *find_inode_fast(struct super_block *sb,
797 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530799 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
801repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800802 hlist_for_each_entry(inode, head, i_hash) {
Al Viro5a3cd99282013-11-06 09:54:52 -0500803 if (inode->i_ino != ino)
804 continue;
805 if (inode->i_sb != sb)
806 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100807 spin_lock(&inode->i_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -0400808 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 __wait_on_freeing_inode(inode);
810 goto repeat;
811 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400812 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100813 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400814 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400816 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817}
818
Eric Dumazetf991bd22010-10-23 11:18:01 -0400819/*
820 * Each cpu owns a range of LAST_INO_BATCH numbers.
821 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
822 * to renew the exhausted range.
David Chinner8290c352008-10-30 17:35:24 +1100823 *
Eric Dumazetf991bd22010-10-23 11:18:01 -0400824 * This does not significantly increase overflow rate because every CPU can
825 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
826 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
827 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
828 * overflow rate by 2x, which does not seem too significant.
829 *
830 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
831 * error if st_ino won't fit in target struct field. Use 32bit counter
832 * here to attempt to avoid that.
David Chinner8290c352008-10-30 17:35:24 +1100833 */
Eric Dumazetf991bd22010-10-23 11:18:01 -0400834#define LAST_INO_BATCH 1024
835static DEFINE_PER_CPU(unsigned int, last_ino);
David Chinner8290c352008-10-30 17:35:24 +1100836
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400837unsigned int get_next_ino(void)
Eric Dumazetf991bd22010-10-23 11:18:01 -0400838{
839 unsigned int *p = &get_cpu_var(last_ino);
840 unsigned int res = *p;
841
842#ifdef CONFIG_SMP
843 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
844 static atomic_t shared_last_ino;
845 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
846
847 res = next - LAST_INO_BATCH;
848 }
849#endif
850
851 *p = ++res;
852 put_cpu_var(last_ino);
853 return res;
David Chinner8290c352008-10-30 17:35:24 +1100854}
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400855EXPORT_SYMBOL(get_next_ino);
David Chinner8290c352008-10-30 17:35:24 +1100856
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857/**
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200858 * new_inode_pseudo - obtain an inode
859 * @sb: superblock
860 *
861 * Allocates a new inode for given superblock.
862 * Inode wont be chained in superblock s_inodes list
863 * This means :
864 * - fs can't be unmount
865 * - quotas, fsnotify, writeback can't work
866 */
867struct inode *new_inode_pseudo(struct super_block *sb)
868{
869 struct inode *inode = alloc_inode(sb);
870
871 if (inode) {
872 spin_lock(&inode->i_lock);
873 inode->i_state = 0;
874 spin_unlock(&inode->i_lock);
875 INIT_LIST_HEAD(&inode->i_sb_list);
876 }
877 return inode;
878}
879
880/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 * new_inode - obtain an inode
882 * @sb: superblock
883 *
Mel Gorman769848c2007-07-17 04:03:05 -0700884 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800885 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700886 * If HIGHMEM pages are unsuitable or it is known that pages allocated
887 * for the page cache are not reclaimable or migratable,
888 * mapping_set_gfp_mask() must be called with suitable flags on the
889 * newly created inode's mapping
890 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 */
892struct inode *new_inode(struct super_block *sb)
893{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530894 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895
Dave Chinner55fa6092011-03-22 22:23:40 +1100896 spin_lock_prefetch(&inode_sb_list_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530897
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200898 inode = new_inode_pseudo(sb);
899 if (inode)
Dave Chinner55fa6092011-03-22 22:23:40 +1100900 inode_sb_list_add(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 return inode;
902}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903EXPORT_SYMBOL(new_inode);
904
Peter Zijlstra14358e62007-10-14 01:38:33 +0200905#ifdef CONFIG_DEBUG_LOCK_ALLOC
Josh Boyere096d0c2011-08-25 07:48:12 -0400906void lockdep_annotate_inode_mutex_key(struct inode *inode)
907{
Namhyung Kima3314a02010-10-11 22:38:00 +0900908 if (S_ISDIR(inode->i_mode)) {
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200909 struct file_system_type *type = inode->i_sb->s_type;
910
Jan Kara9a7aa122009-06-04 15:26:49 +0200911 /* Set new key only if filesystem hasn't already changed it */
Tyler Hicks978d6d82011-12-12 10:02:30 -0600912 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
Jan Kara9a7aa122009-06-04 15:26:49 +0200913 /*
914 * ensure nobody is actually holding i_mutex
915 */
916 mutex_destroy(&inode->i_mutex);
917 mutex_init(&inode->i_mutex);
918 lockdep_set_class(&inode->i_mutex,
919 &type->i_mutex_dir_key);
920 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200921 }
Josh Boyere096d0c2011-08-25 07:48:12 -0400922}
923EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200924#endif
Josh Boyere096d0c2011-08-25 07:48:12 -0400925
926/**
927 * unlock_new_inode - clear the I_NEW state and wake up any waiters
928 * @inode: new inode to unlock
929 *
930 * Called when the inode is fully initialised to clear the new state of the
931 * inode and wake up anyone waiting for the inode to finish initialisation.
932 */
933void unlock_new_inode(struct inode *inode)
934{
935 lockdep_annotate_inode_mutex_key(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100936 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100937 WARN_ON(!(inode->i_state & I_NEW));
938 inode->i_state &= ~I_NEW;
Al Viro310fa7a2012-03-10 17:07:28 -0500939 smp_mb();
Dave Chinner250df6e2011-03-22 22:23:36 +1100940 wake_up_bit(&inode->i_state, __I_NEW);
941 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943EXPORT_SYMBOL(unlock_new_inode);
944
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400945/**
J. Bruce Fields375e2892012-04-18 15:16:33 -0400946 * lock_two_nondirectories - take two i_mutexes on non-directory objects
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200947 *
948 * Lock any non-NULL argument that is not a directory.
949 * Zero, one or two objects may be locked by this function.
950 *
J. Bruce Fields375e2892012-04-18 15:16:33 -0400951 * @inode1: first inode to lock
952 * @inode2: second inode to lock
953 */
954void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
955{
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200956 if (inode1 > inode2)
957 swap(inode1, inode2);
958
959 if (inode1 && !S_ISDIR(inode1->i_mode))
J. Bruce Fields275555162012-04-25 07:19:52 -0400960 mutex_lock(&inode1->i_mutex);
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200961 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
J. Bruce Fields40bd22c2012-04-18 15:21:34 -0400962 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_NONDIR2);
J. Bruce Fields375e2892012-04-18 15:16:33 -0400963}
964EXPORT_SYMBOL(lock_two_nondirectories);
965
966/**
967 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
968 * @inode1: first inode to unlock
969 * @inode2: second inode to unlock
970 */
971void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
972{
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200973 if (inode1 && !S_ISDIR(inode1->i_mode))
974 mutex_unlock(&inode1->i_mutex);
975 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
J. Bruce Fields375e2892012-04-18 15:16:33 -0400976 mutex_unlock(&inode2->i_mutex);
977}
978EXPORT_SYMBOL(unlock_two_nondirectories);
979
980/**
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400981 * iget5_locked - obtain an inode from a mounted file system
982 * @sb: super block of file system
983 * @hashval: hash value (usually inode number) to get
984 * @test: callback used for comparisons between inodes
985 * @set: callback used to initialize a new struct inode
986 * @data: opaque data pointer to pass to @test and @set
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400988 * Search for the inode specified by @hashval and @data in the inode cache,
989 * and if present it is return it with an increased reference count. This is
990 * a generalized version of iget_locked() for file systems where the inode
991 * number is not sufficient for unique identification of an inode.
992 *
993 * If the inode is not in cache, allocate a new inode and return it locked,
994 * hashed, and with the I_NEW flag set. The file system gets to fill it in
995 * before unlocking it via unlock_new_inode().
996 *
997 * Note both @test and @set are called with the inode_hash_lock held, so can't
998 * sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001000struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1001 int (*test)(struct inode *, void *),
1002 int (*set)(struct inode *, void *), void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001004 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301005 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001007 spin_lock(&inode_hash_lock);
1008 inode = find_inode(sb, head, test, data);
1009 spin_unlock(&inode_hash_lock);
1010
1011 if (inode) {
1012 wait_on_inode(inode);
1013 return inode;
1014 }
1015
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 inode = alloc_inode(sb);
1017 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301018 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
Dave Chinner67a23c42011-03-22 22:23:42 +11001020 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 /* We released the lock, so.. */
1022 old = find_inode(sb, head, test, data);
1023 if (!old) {
1024 if (set(inode, data))
1025 goto set_failed;
1026
Dave Chinner250df6e2011-03-22 22:23:36 +11001027 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001028 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001029 hlist_add_head(&inode->i_hash, head);
1030 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001031 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001032 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
1034 /* Return the locked inode with I_NEW set, the
1035 * caller is responsible for filling in the contents
1036 */
1037 return inode;
1038 }
1039
1040 /*
1041 * Uhhuh, somebody else created the same inode under
1042 * us. Use the old inode instead of the one we just
1043 * allocated.
1044 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001045 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 destroy_inode(inode);
1047 inode = old;
1048 wait_on_inode(inode);
1049 }
1050 return inode;
1051
1052set_failed:
Dave Chinner67a23c42011-03-22 22:23:42 +11001053 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 destroy_inode(inode);
1055 return NULL;
1056}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001057EXPORT_SYMBOL(iget5_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001059/**
1060 * iget_locked - obtain an inode from a mounted file system
1061 * @sb: super block of file system
1062 * @ino: inode number to get
1063 *
1064 * Search for the inode specified by @ino in the inode cache and if present
1065 * return it with an increased reference count. This is for file systems
1066 * where the inode number is sufficient for unique identification of an inode.
1067 *
1068 * If the inode is not in cache, allocate a new inode and return it locked,
1069 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1070 * before unlocking it via unlock_new_inode().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001072struct inode *iget_locked(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001074 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301075 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001077 spin_lock(&inode_hash_lock);
1078 inode = find_inode_fast(sb, head, ino);
1079 spin_unlock(&inode_hash_lock);
1080 if (inode) {
1081 wait_on_inode(inode);
1082 return inode;
1083 }
1084
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 inode = alloc_inode(sb);
1086 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301087 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
Dave Chinner67a23c42011-03-22 22:23:42 +11001089 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 /* We released the lock, so.. */
1091 old = find_inode_fast(sb, head, ino);
1092 if (!old) {
1093 inode->i_ino = ino;
Dave Chinner250df6e2011-03-22 22:23:36 +11001094 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001095 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001096 hlist_add_head(&inode->i_hash, head);
1097 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001098 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001099 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
1101 /* Return the locked inode with I_NEW set, the
1102 * caller is responsible for filling in the contents
1103 */
1104 return inode;
1105 }
1106
1107 /*
1108 * Uhhuh, somebody else created the same inode under
1109 * us. Use the old inode instead of the one we just
1110 * allocated.
1111 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001112 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 destroy_inode(inode);
1114 inode = old;
1115 wait_on_inode(inode);
1116 }
1117 return inode;
1118}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001119EXPORT_SYMBOL(iget_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001121/*
1122 * search the inode cache for a matching inode number.
1123 * If we find one, then the inode number we are trying to
1124 * allocate is not unique and so we should not use it.
1125 *
1126 * Returns 1 if the inode number is unique, 0 if it is not.
1127 */
1128static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1129{
1130 struct hlist_head *b = inode_hashtable + hash(sb, ino);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001131 struct inode *inode;
1132
Dave Chinner67a23c42011-03-22 22:23:42 +11001133 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001134 hlist_for_each_entry(inode, b, i_hash) {
Dave Chinner67a23c42011-03-22 22:23:42 +11001135 if (inode->i_ino == ino && inode->i_sb == sb) {
1136 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001137 return 0;
Dave Chinner67a23c42011-03-22 22:23:42 +11001138 }
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001139 }
Dave Chinner67a23c42011-03-22 22:23:42 +11001140 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001141
1142 return 1;
1143}
1144
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145/**
1146 * iunique - get a unique inode number
1147 * @sb: superblock
1148 * @max_reserved: highest reserved inode number
1149 *
1150 * Obtain an inode number that is unique on the system for a given
1151 * superblock. This is used by file systems that have no natural
1152 * permanent inode numbering system. An inode number is returned that
1153 * is higher than the reserved limit but unique.
1154 *
1155 * BUGS:
1156 * With a large number of inodes live on the file system this function
1157 * currently becomes quite slow.
1158 */
1159ino_t iunique(struct super_block *sb, ino_t max_reserved)
1160{
Jeff Layton866b04f2007-05-08 00:32:29 -07001161 /*
1162 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1163 * error if st_ino won't fit in target struct field. Use 32bit counter
1164 * here to attempt to avoid that.
1165 */
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001166 static DEFINE_SPINLOCK(iunique_lock);
Jeff Layton866b04f2007-05-08 00:32:29 -07001167 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001170 spin_lock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001171 do {
1172 if (counter <= max_reserved)
1173 counter = max_reserved + 1;
1174 res = counter++;
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001175 } while (!test_inode_iunique(sb, res));
1176 spin_unlock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001177
1178 return res;
1179}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180EXPORT_SYMBOL(iunique);
1181
1182struct inode *igrab(struct inode *inode)
1183{
Dave Chinner250df6e2011-03-22 22:23:36 +11001184 spin_lock(&inode->i_lock);
1185 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001187 spin_unlock(&inode->i_lock);
1188 } else {
1189 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 /*
1191 * Handle the case where s_op->clear_inode is not been
1192 * called yet, and somebody is calling igrab
1193 * while the inode is getting freed.
1194 */
1195 inode = NULL;
Dave Chinner250df6e2011-03-22 22:23:36 +11001196 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 return inode;
1198}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199EXPORT_SYMBOL(igrab);
1200
1201/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001202 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 * @sb: super block of file system to search
1204 * @hashval: hash value (usually inode number) to search for
1205 * @test: callback used for comparisons between inodes
1206 * @data: opaque data pointer to pass to @test
1207 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001208 * Search for the inode specified by @hashval and @data in the inode cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 * If the inode is in the cache, the inode is returned with an incremented
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001210 * reference count.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001211 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001212 * Note: I_NEW is not waited upon so you have to be very careful what you do
1213 * with the returned inode. You probably should be using ilookup5() instead.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001214 *
Randy Dunlapb6d0ad62011-03-26 13:27:47 -07001215 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001216 */
1217struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1218 int (*test)(struct inode *, void *), void *data)
1219{
1220 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001221 struct inode *inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001222
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001223 spin_lock(&inode_hash_lock);
1224 inode = find_inode(sb, head, test, data);
1225 spin_unlock(&inode_hash_lock);
1226
1227 return inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001228}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001229EXPORT_SYMBOL(ilookup5_nowait);
1230
1231/**
1232 * ilookup5 - search for an inode in the inode cache
1233 * @sb: super block of file system to search
1234 * @hashval: hash value (usually inode number) to search for
1235 * @test: callback used for comparisons between inodes
1236 * @data: opaque data pointer to pass to @test
1237 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001238 * Search for the inode specified by @hashval and @data in the inode cache,
1239 * and if the inode is in the cache, return the inode with an incremented
1240 * reference count. Waits on I_NEW before returning the inode.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001241 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001243 * This is a generalized version of ilookup() for file systems where the
1244 * inode number is not sufficient for unique identification of an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001246 * Note: @test is called with the inode_hash_lock held, so can't sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 */
1248struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1249 int (*test)(struct inode *, void *), void *data)
1250{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001251 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001253 if (inode)
1254 wait_on_inode(inode);
1255 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257EXPORT_SYMBOL(ilookup5);
1258
1259/**
1260 * ilookup - search for an inode in the inode cache
1261 * @sb: super block of file system to search
1262 * @ino: inode number to search for
1263 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001264 * Search for the inode @ino in the inode cache, and if the inode is in the
1265 * cache, the inode is returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 */
1267struct inode *ilookup(struct super_block *sb, unsigned long ino)
1268{
1269 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001270 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001272 spin_lock(&inode_hash_lock);
1273 inode = find_inode_fast(sb, head, ino);
1274 spin_unlock(&inode_hash_lock);
1275
1276 if (inode)
1277 wait_on_inode(inode);
1278 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280EXPORT_SYMBOL(ilookup);
1281
Al Viro261bca82008-12-30 01:48:21 -05001282int insert_inode_locked(struct inode *inode)
1283{
1284 struct super_block *sb = inode->i_sb;
1285 ino_t ino = inode->i_ino;
1286 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001287
Al Viro261bca82008-12-30 01:48:21 -05001288 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001289 struct inode *old = NULL;
Dave Chinner67a23c42011-03-22 22:23:42 +11001290 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001291 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001292 if (old->i_ino != ino)
1293 continue;
1294 if (old->i_sb != sb)
1295 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001296 spin_lock(&old->i_lock);
1297 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1298 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001299 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001300 }
Al Viro72a43d62009-05-13 19:13:40 +01001301 break;
1302 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001303 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001304 spin_lock(&inode->i_lock);
1305 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001306 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001307 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001308 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001309 return 0;
1310 }
1311 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001312 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001313 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001314 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001315 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001316 iput(old);
1317 return -EBUSY;
1318 }
1319 iput(old);
1320 }
1321}
Al Viro261bca82008-12-30 01:48:21 -05001322EXPORT_SYMBOL(insert_inode_locked);
1323
1324int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1325 int (*test)(struct inode *, void *), void *data)
1326{
1327 struct super_block *sb = inode->i_sb;
1328 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001329
Al Viro261bca82008-12-30 01:48:21 -05001330 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001331 struct inode *old = NULL;
1332
Dave Chinner67a23c42011-03-22 22:23:42 +11001333 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001334 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001335 if (old->i_sb != sb)
1336 continue;
1337 if (!test(old, data))
1338 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001339 spin_lock(&old->i_lock);
1340 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1341 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001342 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001343 }
Al Viro72a43d62009-05-13 19:13:40 +01001344 break;
1345 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001346 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001347 spin_lock(&inode->i_lock);
1348 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001349 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001350 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001351 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001352 return 0;
1353 }
1354 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001355 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001356 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001357 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001358 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001359 iput(old);
1360 return -EBUSY;
1361 }
1362 iput(old);
1363 }
1364}
Al Viro261bca82008-12-30 01:48:21 -05001365EXPORT_SYMBOL(insert_inode_locked4);
1366
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Al Viro45321ac2010-06-07 13:43:19 -04001368int generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369{
Al Viro45321ac2010-06-07 13:43:19 -04001370 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372EXPORT_SYMBOL(generic_delete_inode);
1373
Al Viro45321ac2010-06-07 13:43:19 -04001374/*
Al Viro45321ac2010-06-07 13:43:19 -04001375 * Called when we're dropping the last reference
1376 * to an inode.
1377 *
1378 * Call the FS "drop_inode()" function, defaulting to
1379 * the legacy UNIX filesystem behaviour. If it tells
1380 * us to evict inode, do so. Otherwise, retain inode
1381 * in cache if fs is alive, sync and evict if fs is
1382 * shutting down.
1383 */
1384static void iput_final(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385{
1386 struct super_block *sb = inode->i_sb;
Al Viro45321ac2010-06-07 13:43:19 -04001387 const struct super_operations *op = inode->i_sb->s_op;
1388 int drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
Dave Chinner250df6e2011-03-22 22:23:36 +11001390 WARN_ON(inode->i_state & I_NEW);
1391
Al Viroe7f59092011-07-07 15:45:59 -04001392 if (op->drop_inode)
Al Viro45321ac2010-06-07 13:43:19 -04001393 drop = op->drop_inode(inode);
1394 else
1395 drop = generic_drop_inode(inode);
1396
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001397 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1398 inode->i_state |= I_REFERENCED;
Jan Kara4eff96d2012-11-26 16:29:51 -08001399 inode_add_lru(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001400 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001401 return;
1402 }
1403
Al Viro45321ac2010-06-07 13:43:19 -04001404 if (!drop) {
Alexander Viro991114c2005-06-23 00:09:01 -07001405 inode->i_state |= I_WILL_FREE;
Dave Chinner250df6e2011-03-22 22:23:36 +11001406 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 write_inode_now(inode, 1);
Dave Chinner250df6e2011-03-22 22:23:36 +11001408 spin_lock(&inode->i_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001409 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001410 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001412
Alexander Viro991114c2005-06-23 00:09:01 -07001413 inode->i_state |= I_FREEING;
Eric Dumazetc4ae0c62011-07-28 06:55:13 +02001414 if (!list_empty(&inode->i_lru))
1415 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001416 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001417
1418 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419}
1420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301422 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 * @inode: inode to put
1424 *
1425 * Puts an inode, dropping its usage count. If the inode use count hits
1426 * zero, the inode is then freed and may also be destroyed.
1427 *
1428 * Consequently, iput() can sleep.
1429 */
1430void iput(struct inode *inode)
1431{
1432 if (inode) {
Al Viroa4ffdde2010-06-02 17:38:30 -04001433 BUG_ON(inode->i_state & I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434
Dave Chinnerf283c862011-03-22 22:23:39 +11001435 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 iput_final(inode);
1437 }
1438}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439EXPORT_SYMBOL(iput);
1440
1441/**
1442 * bmap - find a block number in a file
1443 * @inode: inode of file
1444 * @block: block to find
1445 *
1446 * Returns the block number on the device holding the inode that
1447 * is the disk block number for the block of the file requested.
1448 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301449 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 * file.
1451 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301452sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453{
1454 sector_t res = 0;
1455 if (inode->i_mapping->a_ops->bmap)
1456 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1457 return res;
1458}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459EXPORT_SYMBOL(bmap);
1460
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001461/*
1462 * With relative atime, only update atime if the previous atime is
1463 * earlier than either the ctime or mtime or if at least a day has
1464 * passed since the last atime update.
1465 */
1466static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1467 struct timespec now)
1468{
1469
1470 if (!(mnt->mnt_flags & MNT_RELATIME))
1471 return 1;
1472 /*
1473 * Is mtime younger than atime? If yes, update atime:
1474 */
1475 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1476 return 1;
1477 /*
1478 * Is ctime younger than atime? If yes, update atime:
1479 */
1480 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1481 return 1;
1482
1483 /*
1484 * Is the previous atime value older than a day? If yes,
1485 * update atime:
1486 */
1487 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1488 return 1;
1489 /*
1490 * Good, we can skip the atime update:
1491 */
1492 return 0;
1493}
1494
Josef Bacikc3b2da32012-03-26 09:59:21 -04001495/*
1496 * This does the actual work of updating an inodes time or version. Must have
1497 * had called mnt_want_write() before calling this.
1498 */
1499static int update_time(struct inode *inode, struct timespec *time, int flags)
1500{
1501 if (inode->i_op->update_time)
1502 return inode->i_op->update_time(inode, time, flags);
1503
1504 if (flags & S_ATIME)
1505 inode->i_atime = *time;
1506 if (flags & S_VERSION)
1507 inode_inc_iversion(inode);
1508 if (flags & S_CTIME)
1509 inode->i_ctime = *time;
1510 if (flags & S_MTIME)
1511 inode->i_mtime = *time;
1512 mark_inode_dirty_sync(inode);
1513 return 0;
1514}
1515
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001517 * touch_atime - update the access time
Randy Dunlap185553b2012-04-17 17:03:25 -07001518 * @path: the &struct path to update
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 *
1520 * Update the accessed time on an inode and mark it for writeback.
1521 * This function automatically handles read only file systems and media,
1522 * as well as the "noatime" flag and inode specific "noatime" markers.
1523 */
Al Virobadcf2b2013-07-16 18:15:46 +04001524void touch_atime(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525{
Al Viro68ac1232012-03-15 08:21:57 -04001526 struct vfsmount *mnt = path->mnt;
1527 struct inode *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 struct timespec now;
1529
Andrew Mortonb2276132006-12-13 00:34:33 -08001530 if (inode->i_flags & S_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001531 return;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001532 if (IS_NOATIME(inode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001533 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001534 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001535 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001536
Dave Hansencdb70f32008-02-15 14:37:41 -08001537 if (mnt->mnt_flags & MNT_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001538 return;
Dave Hansencdb70f32008-02-15 14:37:41 -08001539 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001540 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 now = current_fs_time(inode->i_sb);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001543
1544 if (!relatime_need_update(mnt, inode, now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001545 return;
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001546
Valerie Henson47ae32d2006-12-13 00:34:34 -08001547 if (timespec_equal(&inode->i_atime, &now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001548 return;
1549
Jan Kara5d37e9e2012-06-12 16:20:36 +02001550 if (!sb_start_write_trylock(inode->i_sb))
Andi Kleenb12536c2009-09-18 13:05:47 -07001551 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001552
Jan Kara5d37e9e2012-06-12 16:20:36 +02001553 if (__mnt_want_write(mnt))
1554 goto skip_update;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001555 /*
1556 * File systems can error out when updating inodes if they need to
1557 * allocate new space to modify an inode (such is the case for
1558 * Btrfs), but since we touch atime while walking down the path we
1559 * really don't care if we failed to update the atime of the file,
1560 * so just ignore the return value.
Alexander Block2bc5565282012-06-15 09:49:33 +02001561 * We may also fail on filesystems that have the ability to make parts
1562 * of the fs read only, e.g. subvolumes in Btrfs.
Josef Bacikc3b2da32012-03-26 09:59:21 -04001563 */
1564 update_time(inode, &now, S_ATIME);
Jan Kara5d37e9e2012-06-12 16:20:36 +02001565 __mnt_drop_write(mnt);
1566skip_update:
1567 sb_end_write(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001569EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570
Cong Wang3ed37642012-05-15 14:57:33 +08001571/*
1572 * The logic we want is
1573 *
1574 * if suid or (sgid and xgrp)
1575 * remove privs
1576 */
1577int should_remove_suid(struct dentry *dentry)
1578{
1579 umode_t mode = dentry->d_inode->i_mode;
1580 int kill = 0;
1581
1582 /* suid always must be killed */
1583 if (unlikely(mode & S_ISUID))
1584 kill = ATTR_KILL_SUID;
1585
1586 /*
1587 * sgid without any exec bits is just a mandatory locking mark; leave
1588 * it alone. If some exec bits are set, it's a real sgid; kill it.
1589 */
1590 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1591 kill |= ATTR_KILL_SGID;
1592
1593 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1594 return kill;
1595
1596 return 0;
1597}
1598EXPORT_SYMBOL(should_remove_suid);
1599
1600static int __remove_suid(struct dentry *dentry, int kill)
1601{
1602 struct iattr newattrs;
1603
1604 newattrs.ia_valid = ATTR_FORCE | kill;
J. Bruce Fields27ac0ff2011-09-20 17:19:26 -04001605 /*
1606 * Note we call this on write, so notify_change will not
1607 * encounter any conflicting delegations:
1608 */
1609 return notify_change(dentry, &newattrs, NULL);
Cong Wang3ed37642012-05-15 14:57:33 +08001610}
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 *
Andy Lutomirski23adbe12014-06-10 12:45:42 -07001842 * Return true if current either has CAP_FOWNER in a namespace with the
1843 * inode owner uid mapped, or 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{
Andy Lutomirski23adbe12014-06-10 12:45:42 -07001847 struct user_namespace *ns;
1848
Eric W. Biederman92361632012-02-08 07:07:50 -08001849 if (uid_eq(current_fsuid(), inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001850 return true;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07001851
1852 ns = current_user_ns();
1853 if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001854 return true;
1855 return false;
1856}
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001857EXPORT_SYMBOL(inode_owner_or_capable);
Trond Myklebust1d59d612012-05-31 12:22:33 -04001858
1859/*
1860 * Direct i/o helper functions
1861 */
1862static void __inode_dio_wait(struct inode *inode)
1863{
1864 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1865 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1866
1867 do {
1868 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1869 if (atomic_read(&inode->i_dio_count))
1870 schedule();
1871 } while (atomic_read(&inode->i_dio_count));
1872 finish_wait(wq, &q.wait);
1873}
1874
1875/**
1876 * inode_dio_wait - wait for outstanding DIO requests to finish
1877 * @inode: inode to wait for
1878 *
1879 * Waits for all pending direct I/O requests to finish so that we can
1880 * proceed with a truncate or equivalent operation.
1881 *
1882 * Must be called under a lock that serializes taking new references
1883 * to i_dio_count, usually by inode->i_mutex.
1884 */
1885void inode_dio_wait(struct inode *inode)
1886{
1887 if (atomic_read(&inode->i_dio_count))
1888 __inode_dio_wait(inode);
1889}
1890EXPORT_SYMBOL(inode_dio_wait);
1891
1892/*
1893 * inode_dio_done - signal finish of a direct I/O requests
1894 * @inode: inode the direct I/O happens on
1895 *
1896 * This is called once we've finished processing a direct I/O request,
1897 * and is used to wake up callers waiting for direct I/O to be quiesced.
1898 */
1899void inode_dio_done(struct inode *inode)
1900{
1901 if (atomic_dec_and_test(&inode->i_dio_count))
1902 wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1903}
1904EXPORT_SYMBOL(inode_dio_done);
Theodore Ts'o5f16f322014-03-24 14:43:12 -04001905
1906/*
1907 * inode_set_flags - atomically set some inode flags
1908 *
1909 * Note: the caller should be holding i_mutex, or else be sure that
1910 * they have exclusive access to the inode structure (i.e., while the
1911 * inode is being instantiated). The reason for the cmpxchg() loop
1912 * --- which wouldn't be necessary if all code paths which modify
1913 * i_flags actually followed this rule, is that there is at least one
1914 * code path which doesn't today --- for example,
1915 * __generic_file_aio_write() calls file_remove_suid() without holding
1916 * i_mutex --- so we use cmpxchg() out of an abundance of caution.
1917 *
1918 * In the long run, i_mutex is overkill, and we should probably look
1919 * at using the i_lock spinlock to protect i_flags, and then make sure
1920 * it is so documented in include/linux/fs.h and that all code follows
1921 * the locking convention!!
1922 */
1923void inode_set_flags(struct inode *inode, unsigned int flags,
1924 unsigned int mask)
1925{
1926 unsigned int old_flags, new_flags;
1927
1928 WARN_ON_ONCE(flags & ~mask);
1929 do {
1930 old_flags = ACCESS_ONCE(inode->i_flags);
1931 new_flags = (old_flags & ~mask) | flags;
1932 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
1933 new_flags) != old_flags));
1934}
1935EXPORT_SYMBOL(inode_set_flags);