blob: 88110fd0b282e49246dc9cd93a1d6e173d951d7b [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>
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050021#include <trace/events/writeback.h>
Dave Chinnera66979a2011-03-22 22:23:41 +110022#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
24/*
Christoph Hellwig4b4563d2011-05-27 09:28:01 -040025 * Inode locking rules:
Dave Chinner250df6e2011-03-22 22:23:36 +110026 *
27 * inode->i_lock protects:
28 * inode->i_state, inode->i_hash, __iget()
Dave Chinnerbc3b14c2013-08-28 10:17:58 +100029 * Inode LRU list locks protect:
Dave Chinner98b745c2011-07-08 14:14:39 +100030 * inode->i_sb->s_inode_lru, inode->i_lru
Dave Chinner74278da2015-03-04 12:37:22 -050031 * inode->i_sb->s_inode_list_lock protects:
32 * inode->i_sb->s_inodes, inode->i_sb_list
Christoph Hellwigf758eea2011-04-21 18:19:44 -060033 * bdi->wb.list_lock protects:
Dave Chinnerc7f54082015-03-04 14:07:22 -050034 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
Dave Chinner67a23c42011-03-22 22:23:42 +110035 * inode_hash_lock protects:
36 * inode_hashtable, inode->i_hash
Dave Chinner250df6e2011-03-22 22:23:36 +110037 *
38 * Lock ordering:
Dave Chinner55fa6092011-03-22 22:23:40 +110039 *
Dave Chinner74278da2015-03-04 12:37:22 -050040 * inode->i_sb->s_inode_list_lock
Dave Chinner55fa6092011-03-22 22:23:40 +110041 * inode->i_lock
Dave Chinnerbc3b14c2013-08-28 10:17:58 +100042 * Inode LRU list locks
Dave Chinnera66979a2011-03-22 22:23:41 +110043 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060044 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110045 * inode->i_lock
Dave Chinner67a23c42011-03-22 22:23:42 +110046 *
47 * inode_hash_lock
Dave Chinner74278da2015-03-04 12:37:22 -050048 * inode->i_sb->s_inode_list_lock
Dave Chinner67a23c42011-03-22 22:23:42 +110049 * inode->i_lock
50 *
51 * iunique_lock
52 * inode_hash_lock
Dave Chinner250df6e2011-03-22 22:23:36 +110053 */
54
Eric Dumazetfa3536c2006-03-26 01:37:24 -080055static unsigned int i_hash_mask __read_mostly;
56static unsigned int i_hash_shift __read_mostly;
Dave Chinner67a23c42011-03-22 22:23:42 +110057static struct hlist_head *inode_hashtable __read_mostly;
58static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
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
Glauber Costa3942c072013-08-28 10:17:53 +100073static DEFINE_PER_CPU(unsigned long, nr_inodes);
74static DEFINE_PER_CPU(unsigned long, 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
Glauber Costa3942c072013-08-28 10:17:53 +100078static long get_nr_inodes(void)
Dave Chinnercffbc8a2010-10-23 05:03:02 -040079{
Nick Piggin3e880fb2011-01-07 17:49:19 +110080 int i;
Glauber Costa3942c072013-08-28 10:17:53 +100081 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +110082 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
Glauber Costa3942c072013-08-28 10:17:53 +100087static inline long get_nr_inodes_unused(void)
Dave Chinnercffbc8a2010-10-23 05:03:02 -040088{
Dave Chinnerfcb94f72011-07-08 14:14:38 +100089 int i;
Glauber Costa3942c072013-08-28 10:17:53 +100090 long sum = 0;
Dave Chinnerfcb94f72011-07-08 14:14:38 +100091 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
Glauber Costa3942c072013-08-28 10:17:53 +100096long get_nr_dirty_inodes(void)
Dave Chinnercffbc8a2010-10-23 05:03:02 -040097{
Nick Piggin3e880fb2011-01-07 17:49:19 +110098 /* not actually dirty inodes, but a wild approximation */
Glauber Costa3942c072013-08-28 10:17:53 +100099 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400100 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
Joe Perches1f7e0612014-06-06 14:38:05 -0700107int proc_nr_inodes(struct ctl_table *table, int write,
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400108 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();
Glauber Costa3942c072013-08-28 10:17:53 +1000112 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400113}
114#endif
115
Al Virobd9b51e2014-11-18 23:38:21 -0500116static int no_open(struct inode *inode, struct file *file)
117{
118 return -ENXIO;
119}
120
David Chinner2cb15992008-10-30 17:32:23 +1100121/**
122 * inode_init_always - perform inode structure intialisation
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800123 * @sb: superblock inode belongs to
124 * @inode: inode to initialise
David Chinner2cb15992008-10-30 17:32:23 +1100125 *
126 * These are initializations that need to be done on every inode
127 * allocation as the fields are not initialised by slab allocation.
128 */
Christoph Hellwig54e34622009-08-07 14:38:25 -0300129int inode_init_always(struct super_block *sb, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130{
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700131 static const struct inode_operations empty_iops;
Al Virobd9b51e2014-11-18 23:38:21 -0500132 static const struct file_operations no_open_fops = {.open = no_open};
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530133 struct address_space *const mapping = &inode->i_data;
David Chinner2cb15992008-10-30 17:32:23 +1100134
135 inode->i_sb = sb;
136 inode->i_blkbits = sb->s_blocksize_bits;
137 inode->i_flags = 0;
138 atomic_set(&inode->i_count, 1);
139 inode->i_op = &empty_iops;
Al Virobd9b51e2014-11-18 23:38:21 -0500140 inode->i_fop = &no_open_fops;
Miklos Szeredia78ef702011-10-28 14:13:30 +0200141 inode->__i_nlink = 1;
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700142 inode->i_opflags = 0;
Andreas Gruenbacherd0a5b992016-09-29 17:48:39 +0200143 if (sb->s_xattr)
144 inode->i_opflags |= IOP_XATTR;
Eric W. Biederman92361632012-02-08 07:07:50 -0800145 i_uid_write(inode, 0);
146 i_gid_write(inode, 0);
David Chinner2cb15992008-10-30 17:32:23 +1100147 atomic_set(&inode->i_writecount, 0);
148 inode->i_size = 0;
149 inode->i_blocks = 0;
150 inode->i_bytes = 0;
151 inode->i_generation = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100152 inode->i_pipe = NULL;
153 inode->i_bdev = NULL;
154 inode->i_cdev = NULL;
Al Viro61ba64f2015-05-02 09:54:06 -0400155 inode->i_link = NULL;
Al Viro84e710d2016-04-15 00:58:55 -0400156 inode->i_dir_seq = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100157 inode->i_rdev = 0;
158 inode->dirtied_when = 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500159
Tahsin Erdogan3d65ae42016-02-16 13:34:39 -0800160#ifdef CONFIG_CGROUP_WRITEBACK
161 inode->i_wb_frn_winner = 0;
162 inode->i_wb_frn_avg_time = 0;
163 inode->i_wb_frn_history = 0;
164#endif
165
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500166 if (security_inode_alloc(inode))
Christoph Hellwig54e34622009-08-07 14:38:25 -0300167 goto out;
David Chinner2cb15992008-10-30 17:32:23 +1100168 spin_lock_init(&inode->i_lock);
169 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
170
Al Viro9902af72016-04-15 15:08:36 -0400171 init_rwsem(&inode->i_rwsem);
172 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
David Chinner2cb15992008-10-30 17:32:23 +1100173
Christoph Hellwigbd5fe6c2011-06-24 14:29:43 -0400174 atomic_set(&inode->i_dio_count, 0);
David Chinner2cb15992008-10-30 17:32:23 +1100175
176 mapping->a_ops = &empty_aops;
177 mapping->host = inode;
178 mapping->flags = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700179 atomic_set(&mapping->i_mmap_writable, 0);
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800180 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800181 mapping->private_data = NULL;
David Chinner2cb15992008-10-30 17:32:23 +1100182 mapping->writeback_index = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100183 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
Jeff Layton4a075e32015-01-16 15:05:54 -0500193 inode->i_flctx = NULL;
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));
Tejun Heo52ebea72015-05-22 17:13:37 -0400234 inode_detach_wb(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);
Jeff Laytonf27a0fe2016-01-07 15:08:51 -0500237 locks_free_lock_context(inode);
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100238 if (!inode->i_nlink) {
239 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
240 atomic_long_dec(&inode->i_sb->s_remove_count);
241 }
242
Al Virof19d4a82009-06-08 19:50:45 -0400243#ifdef CONFIG_FS_POSIX_ACL
Andreas Gruenbacherb8a7a3a2016-03-24 14:38:37 +0100244 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
Al Virof19d4a82009-06-08 19:50:45 -0400245 posix_acl_release(inode->i_acl);
Andreas Gruenbacherb8a7a3a2016-03-24 14:38:37 +0100246 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
Al Virof19d4a82009-06-08 19:50:45 -0400247 posix_acl_release(inode->i_default_acl);
248#endif
Nick Piggin3e880fb2011-01-07 17:49:19 +1100249 this_cpu_dec(nr_inodes);
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300250}
251EXPORT_SYMBOL(__destroy_inode);
252
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100253static void i_callback(struct rcu_head *head)
254{
255 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100256 kmem_cache_free(inode_cachep, inode);
257}
258
Christoph Hellwig56b0dac2010-10-06 10:48:55 +0200259static void destroy_inode(struct inode *inode)
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300260{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100261 BUG_ON(!list_empty(&inode->i_lru));
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300262 __destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 if (inode->i_sb->s_op->destroy_inode)
264 inode->i_sb->s_op->destroy_inode(inode);
265 else
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100266 call_rcu(&inode->i_rcu, i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100269/**
270 * drop_nlink - directly drop an inode's link count
271 * @inode: inode
272 *
273 * This is a low-level filesystem helper to replace any
274 * direct filesystem manipulation of i_nlink. In cases
275 * where we are attempting to track writes to the
276 * filesystem, a decrement to zero means an imminent
277 * write when the file is truncated and actually unlinked
278 * on the filesystem.
279 */
280void drop_nlink(struct inode *inode)
281{
282 WARN_ON(inode->i_nlink == 0);
283 inode->__i_nlink--;
284 if (!inode->i_nlink)
285 atomic_long_inc(&inode->i_sb->s_remove_count);
286}
287EXPORT_SYMBOL(drop_nlink);
288
289/**
290 * clear_nlink - directly zero an inode's link count
291 * @inode: inode
292 *
293 * This is a low-level filesystem helper to replace any
294 * direct filesystem manipulation of i_nlink. See
295 * drop_nlink() for why we care about i_nlink hitting zero.
296 */
297void clear_nlink(struct inode *inode)
298{
299 if (inode->i_nlink) {
300 inode->__i_nlink = 0;
301 atomic_long_inc(&inode->i_sb->s_remove_count);
302 }
303}
304EXPORT_SYMBOL(clear_nlink);
305
306/**
307 * set_nlink - directly set an inode's link count
308 * @inode: inode
309 * @nlink: new nlink (should be non-zero)
310 *
311 * This is a low-level filesystem helper to replace any
312 * direct filesystem manipulation of i_nlink.
313 */
314void set_nlink(struct inode *inode, unsigned int nlink)
315{
316 if (!nlink) {
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100317 clear_nlink(inode);
318 } else {
319 /* Yes, some filesystems do change nlink from zero to one */
320 if (inode->i_nlink == 0)
321 atomic_long_dec(&inode->i_sb->s_remove_count);
322
323 inode->__i_nlink = nlink;
324 }
325}
326EXPORT_SYMBOL(set_nlink);
327
328/**
329 * inc_nlink - directly increment an inode's link count
330 * @inode: inode
331 *
332 * This is a low-level filesystem helper to replace any
333 * direct filesystem manipulation of i_nlink. Currently,
334 * it is only here for parity with dec_nlink().
335 */
336void inc_nlink(struct inode *inode)
337{
Al Virof4e0c302013-06-11 08:34:36 +0400338 if (unlikely(inode->i_nlink == 0)) {
339 WARN_ON(!(inode->i_state & I_LINKABLE));
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100340 atomic_long_dec(&inode->i_sb->s_remove_count);
Al Virof4e0c302013-06-11 08:34:36 +0400341 }
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100342
343 inode->__i_nlink++;
344}
345EXPORT_SYMBOL(inc_nlink);
346
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100347void address_space_init_once(struct address_space *mapping)
348{
349 memset(mapping, 0, sizeof(*mapping));
Vladimir Davydov05eb6e72016-08-02 14:03:01 -0700350 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC | __GFP_ACCOUNT);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100351 spin_lock_init(&mapping->tree_lock);
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800352 init_rwsem(&mapping->i_mmap_rwsem);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100353 INIT_LIST_HEAD(&mapping->private_list);
354 spin_lock_init(&mapping->private_lock);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700355 mapping->i_mmap = RB_ROOT;
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);
Dave Chinnerc7f54082015-03-04 14:07:22 -0500369 INIT_LIST_HEAD(&inode->i_io_list);
Dave Chinner6c60d2b2016-07-26 15:21:50 -0700370 INIT_LIST_HEAD(&inode->i_wb_list);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100371 INIT_LIST_HEAD(&inode->i_lru);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100372 address_space_init_once(&inode->i_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 i_size_ordered_init(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400374#ifdef CONFIG_FSNOTIFY
Eric Parise61ce862009-12-17 21:24:24 -0500375 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
Eric Paris3be25f42009-05-21 17:01:26 -0400376#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378EXPORT_SYMBOL(inode_init_once);
379
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700380static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530382 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383
Christoph Lametera35afb82007-05-16 22:10:57 -0700384 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385}
386
387/*
Dave Chinner250df6e2011-03-22 22:23:36 +1100388 * inode->i_lock must be held
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530390void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391{
Nick Piggin9e38d862010-10-23 06:55:17 -0400392 atomic_inc(&inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393}
394
Al Viro7de9c6ee2010-10-23 11:11:40 -0400395/*
396 * get additional reference to inode; caller must already hold one.
397 */
398void ihold(struct inode *inode)
399{
400 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
401}
402EXPORT_SYMBOL(ihold);
403
Nick Piggin9e38d862010-10-23 06:55:17 -0400404static void inode_lru_list_add(struct inode *inode)
405{
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000406 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000407 this_cpu_inc(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400408}
409
Jan Kara4eff96d2012-11-26 16:29:51 -0800410/*
411 * Add inode to LRU if needed (inode is unused and clean).
412 *
413 * Needs inode->i_lock held.
414 */
415void inode_add_lru(struct inode *inode)
416{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500417 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
418 I_FREEING | I_WILL_FREE)) &&
Jan Kara4eff96d2012-11-26 16:29:51 -0800419 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
420 inode_lru_list_add(inode);
421}
422
423
Nick Piggin9e38d862010-10-23 06:55:17 -0400424static void inode_lru_list_del(struct inode *inode)
425{
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000426
427 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000428 this_cpu_dec(nr_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429}
430
Christoph Hellwig646ec462010-10-23 07:15:32 -0400431/**
432 * inode_sb_list_add - add inode to the superblock list of inodes
433 * @inode: inode to add
434 */
435void inode_sb_list_add(struct inode *inode)
436{
Dave Chinner74278da2015-03-04 12:37:22 -0500437 spin_lock(&inode->i_sb->s_inode_list_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +1100438 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
Dave Chinner74278da2015-03-04 12:37:22 -0500439 spin_unlock(&inode->i_sb->s_inode_list_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400440}
441EXPORT_SYMBOL_GPL(inode_sb_list_add);
442
Dave Chinner55fa6092011-03-22 22:23:40 +1100443static inline void inode_sb_list_del(struct inode *inode)
Christoph Hellwig646ec462010-10-23 07:15:32 -0400444{
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200445 if (!list_empty(&inode->i_sb_list)) {
Dave Chinner74278da2015-03-04 12:37:22 -0500446 spin_lock(&inode->i_sb->s_inode_list_lock);
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200447 list_del_init(&inode->i_sb_list);
Dave Chinner74278da2015-03-04 12:37:22 -0500448 spin_unlock(&inode->i_sb->s_inode_list_lock);
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200449 }
Christoph Hellwig646ec462010-10-23 07:15:32 -0400450}
451
Dave Chinner4c51acb2010-10-23 06:58:09 -0400452static unsigned long hash(struct super_block *sb, unsigned long hashval)
453{
454 unsigned long tmp;
455
456 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
457 L1_CACHE_BYTES;
Christoph Hellwig4b4563d2011-05-27 09:28:01 -0400458 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
459 return tmp & i_hash_mask;
Dave Chinner4c51acb2010-10-23 06:58:09 -0400460}
461
462/**
463 * __insert_inode_hash - hash an inode
464 * @inode: unhashed inode
465 * @hashval: unsigned long value used to locate this object in the
466 * inode_hashtable.
467 *
468 * Add an inode to the inode hash for this superblock.
469 */
470void __insert_inode_hash(struct inode *inode, unsigned long hashval)
471{
Christoph Hellwig646ec462010-10-23 07:15:32 -0400472 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
473
Dave Chinner67a23c42011-03-22 22:23:42 +1100474 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100475 spin_lock(&inode->i_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400476 hlist_add_head(&inode->i_hash, b);
Dave Chinner250df6e2011-03-22 22:23:36 +1100477 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100478 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400479}
480EXPORT_SYMBOL(__insert_inode_hash);
481
482/**
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200483 * __remove_inode_hash - remove an inode from the hash
Dave Chinner4c51acb2010-10-23 06:58:09 -0400484 * @inode: inode to unhash
485 *
486 * Remove an inode from the superblock.
487 */
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200488void __remove_inode_hash(struct inode *inode)
Dave Chinner4c51acb2010-10-23 06:58:09 -0400489{
Dave Chinner67a23c42011-03-22 22:23:42 +1100490 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100491 spin_lock(&inode->i_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400492 hlist_del_init(&inode->i_hash);
Dave Chinner250df6e2011-03-22 22:23:36 +1100493 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100494 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400495}
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200496EXPORT_SYMBOL(__remove_inode_hash);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400497
Jan Karadbd57682012-05-03 14:48:02 +0200498void clear_inode(struct inode *inode)
Al Virob0683aa2010-06-04 20:55:25 -0400499{
500 might_sleep();
Jan Kara08142572011-06-27 16:18:10 -0700501 /*
502 * We have to cycle tree_lock here because reclaim can be still in the
503 * process of removing the last page (in __delete_from_page_cache())
504 * and we must not free mapping under it.
505 */
506 spin_lock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400507 BUG_ON(inode->i_data.nrpages);
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800508 BUG_ON(inode->i_data.nrexceptional);
Jan Kara08142572011-06-27 16:18:10 -0700509 spin_unlock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400510 BUG_ON(!list_empty(&inode->i_data.private_list));
511 BUG_ON(!(inode->i_state & I_FREEING));
512 BUG_ON(inode->i_state & I_CLEAR);
Dave Chinner6c60d2b2016-07-26 15:21:50 -0700513 BUG_ON(!list_empty(&inode->i_wb_list));
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100514 /* don't need i_lock here, no concurrent mods to i_state */
Al Virob0683aa2010-06-04 20:55:25 -0400515 inode->i_state = I_FREEING | I_CLEAR;
516}
Jan Karadbd57682012-05-03 14:48:02 +0200517EXPORT_SYMBOL(clear_inode);
Al Virob0683aa2010-06-04 20:55:25 -0400518
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100519/*
520 * Free the inode passed in, removing it from the lists it is still connected
521 * to. We remove any pages still attached to the inode and wait for any IO that
522 * is still in progress before finally destroying the inode.
523 *
524 * An inode must already be marked I_FREEING so that we avoid the inode being
525 * moved back onto lists if we race with other code that manipulates the lists
526 * (e.g. writeback_single_inode). The caller is responsible for setting this.
527 *
528 * An inode must already be removed from the LRU list before being evicted from
529 * the cache. This should occur atomically with setting the I_FREEING state
530 * flag, so no inodes here should ever be on the LRU when being evicted.
531 */
Al Viro644da592010-06-07 13:21:05 -0400532static void evict(struct inode *inode)
Al Virob4272d42010-06-04 19:33:20 -0400533{
534 const struct super_operations *op = inode->i_sb->s_op;
535
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100536 BUG_ON(!(inode->i_state & I_FREEING));
537 BUG_ON(!list_empty(&inode->i_lru));
538
Dave Chinnerc7f54082015-03-04 14:07:22 -0500539 if (!list_empty(&inode->i_io_list))
540 inode_io_list_del(inode);
Eric Dumazetb12362b2011-07-28 06:11:47 +0200541
Dave Chinner55fa6092011-03-22 22:23:40 +1100542 inode_sb_list_del(inode);
543
Jan Kara169ebd92012-05-03 14:48:03 +0200544 /*
545 * Wait for flusher thread to be done with the inode so that filesystem
546 * does not start destroying it while writeback is still running. Since
547 * the inode has I_FREEING set, flusher thread won't start new work on
548 * the inode. We just have to wait for running writeback to finish.
549 */
550 inode_wait_for_writeback(inode);
Jan Kara7994e6f2012-05-03 14:48:01 +0200551
Al Virobe7ce412010-06-04 19:40:39 -0400552 if (op->evict_inode) {
553 op->evict_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400554 } else {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700555 truncate_inode_pages_final(&inode->i_data);
Jan Karadbd57682012-05-03 14:48:02 +0200556 clear_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400557 }
Al Viro661074e2010-06-04 20:19:55 -0400558 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
559 bd_forget(inode);
560 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
561 cd_forget(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100562
563 remove_inode_hash(inode);
564
565 spin_lock(&inode->i_lock);
566 wake_up_bit(&inode->i_state, __I_NEW);
567 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
568 spin_unlock(&inode->i_lock);
569
570 destroy_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400571}
572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573/*
574 * dispose_list - dispose of the contents of a local list
575 * @head: the head of the list to free
576 *
577 * Dispose-list gets a local list with local inodes in it, so it doesn't
578 * need to worry about list corruption and SMP locks.
579 */
580static void dispose_list(struct list_head *head)
581{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 while (!list_empty(head)) {
583 struct inode *inode;
584
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100585 inode = list_first_entry(head, struct inode, i_lru);
586 list_del_init(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Al Viro644da592010-06-07 13:21:05 -0400588 evict(inode);
Josef Bacikac05fbb2015-03-04 16:52:52 -0500589 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591}
592
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593/**
Al Viro63997e92010-10-25 20:49:35 -0400594 * evict_inodes - evict all evictable inodes for a superblock
595 * @sb: superblock to operate on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 *
Al Viro63997e92010-10-25 20:49:35 -0400597 * Make sure that no inodes with zero refcount are retained. This is
598 * called by superblock shutdown after having MS_ACTIVE flag removed,
599 * so any inode reaching zero refcount during or after that call will
600 * be immediately evicted.
601 */
602void evict_inodes(struct super_block *sb)
603{
604 struct inode *inode, *next;
605 LIST_HEAD(dispose);
606
Josef Bacikac05fbb2015-03-04 16:52:52 -0500607again:
Dave Chinner74278da2015-03-04 12:37:22 -0500608 spin_lock(&sb->s_inode_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400609 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
610 if (atomic_read(&inode->i_count))
611 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100612
613 spin_lock(&inode->i_lock);
614 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
615 spin_unlock(&inode->i_lock);
Al Viro63997e92010-10-25 20:49:35 -0400616 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100617 }
Al Viro63997e92010-10-25 20:49:35 -0400618
619 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100620 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100621 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100622 list_add(&inode->i_lru, &dispose);
Josef Bacikac05fbb2015-03-04 16:52:52 -0500623
624 /*
625 * We can have a ton of inodes to evict at unmount time given
626 * enough memory, check to see if we need to go to sleep for a
627 * bit so we don't livelock.
628 */
629 if (need_resched()) {
630 spin_unlock(&sb->s_inode_list_lock);
631 cond_resched();
632 dispose_list(&dispose);
633 goto again;
634 }
Al Viro63997e92010-10-25 20:49:35 -0400635 }
Dave Chinner74278da2015-03-04 12:37:22 -0500636 spin_unlock(&sb->s_inode_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400637
638 dispose_list(&dispose);
Al Viro63997e92010-10-25 20:49:35 -0400639}
640
641/**
Christoph Hellwiga0318782010-10-24 19:40:33 +0200642 * invalidate_inodes - attempt to free all inodes on a superblock
643 * @sb: superblock to operate on
NeilBrown93b270f2011-02-24 17:25:47 +1100644 * @kill_dirty: flag to guide handling of dirty inodes
Christoph Hellwiga0318782010-10-24 19:40:33 +0200645 *
646 * Attempts to free all inodes for a given superblock. If there were any
647 * busy inodes return a non-zero value, else zero.
NeilBrown93b270f2011-02-24 17:25:47 +1100648 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
649 * them as busy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 */
NeilBrown93b270f2011-02-24 17:25:47 +1100651int invalidate_inodes(struct super_block *sb, bool kill_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400653 int busy = 0;
Christoph Hellwiga0318782010-10-24 19:40:33 +0200654 struct inode *inode, *next;
655 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Dave Chinner74278da2015-03-04 12:37:22 -0500657 spin_lock(&sb->s_inode_list_lock);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200658 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100659 spin_lock(&inode->i_lock);
660 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
661 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100663 }
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500664 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100665 spin_unlock(&inode->i_lock);
NeilBrown93b270f2011-02-24 17:25:47 +1100666 busy = 1;
667 continue;
668 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200669 if (atomic_read(&inode->i_count)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100670 spin_unlock(&inode->i_lock);
Christoph Hellwig99a38912010-10-23 19:07:20 +0200671 busy = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 continue;
673 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200674
Christoph Hellwig99a38912010-10-23 19:07:20 +0200675 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100676 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100677 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100678 list_add(&inode->i_lru, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 }
Dave Chinner74278da2015-03-04 12:37:22 -0500680 spin_unlock(&sb->s_inode_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
Christoph Hellwiga0318782010-10-24 19:40:33 +0200682 dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
684 return busy;
685}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687/*
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000688 * Isolate the inode from the LRU in preparation for freeing it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 *
690 * Any inodes which are pinned purely because of attached pagecache have their
Nick Piggin9e38d862010-10-23 06:55:17 -0400691 * pagecache removed. If the inode has metadata buffers attached to
692 * mapping->private_list then try to remove them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 *
Nick Piggin9e38d862010-10-23 06:55:17 -0400694 * If the inode has the I_REFERENCED flag set, then it means that it has been
695 * used recently - the flag is set in iput_final(). When we encounter such an
696 * inode, clear the flag and move it to the back of the LRU so it gets another
697 * pass through the LRU before it gets reclaimed. This is necessary because of
698 * the fact we are doing lazy LRU updates to minimise lock contention so the
699 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
700 * with this flag set because they are the inodes that are out of order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800702static enum lru_status inode_lru_isolate(struct list_head *item,
703 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000704{
705 struct list_head *freeable = arg;
706 struct inode *inode = container_of(item, struct inode, i_lru);
707
708 /*
709 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
710 * If we fail to get the lock, just skip it.
711 */
712 if (!spin_trylock(&inode->i_lock))
713 return LRU_SKIP;
714
715 /*
716 * Referenced or dirty inodes are still in use. Give them another pass
717 * through the LRU as we canot reclaim them now.
718 */
719 if (atomic_read(&inode->i_count) ||
720 (inode->i_state & ~I_REFERENCED)) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800721 list_lru_isolate(lru, &inode->i_lru);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000722 spin_unlock(&inode->i_lock);
723 this_cpu_dec(nr_unused);
724 return LRU_REMOVED;
725 }
726
727 /* recently referenced inodes get one more pass */
728 if (inode->i_state & I_REFERENCED) {
729 inode->i_state &= ~I_REFERENCED;
730 spin_unlock(&inode->i_lock);
731 return LRU_ROTATE;
732 }
733
734 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
735 __iget(inode);
736 spin_unlock(&inode->i_lock);
737 spin_unlock(lru_lock);
738 if (remove_inode_buffers(inode)) {
739 unsigned long reap;
740 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
741 if (current_is_kswapd())
742 __count_vm_events(KSWAPD_INODESTEAL, reap);
743 else
744 __count_vm_events(PGINODESTEAL, reap);
745 if (current->reclaim_state)
746 current->reclaim_state->reclaimed_slab += reap;
747 }
748 iput(inode);
749 spin_lock(lru_lock);
750 return LRU_RETRY;
751 }
752
753 WARN_ON(inode->i_state & I_NEW);
754 inode->i_state |= I_FREEING;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800755 list_lru_isolate_move(lru, &inode->i_lru, freeable);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000756 spin_unlock(&inode->i_lock);
757
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000758 this_cpu_dec(nr_unused);
759 return LRU_REMOVED;
760}
761
762/*
763 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
764 * This is called from the superblock shrinker function with a number of inodes
765 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
766 * then are freed outside inode_lock by dispose_list().
767 */
Vladimir Davydov503c3582015-02-12 14:58:47 -0800768long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769{
770 LIST_HEAD(freeable);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000771 long freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
Vladimir Davydov503c3582015-02-12 14:58:47 -0800773 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
774 inode_lru_isolate, &freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 dispose_list(&freeable);
Dave Chinner0a234c62013-08-28 10:17:57 +1000776 return freed;
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{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530788 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
790repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800791 hlist_for_each_entry(inode, head, i_hash) {
Al Viro5a3cd99282013-11-06 09:54:52 -0500792 if (inode->i_sb != sb)
793 continue;
794 if (!test(inode, data))
795 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100796 spin_lock(&inode->i_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -0400797 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 __wait_on_freeing_inode(inode);
799 goto repeat;
800 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400801 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100802 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400803 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400805 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806}
807
808/*
809 * find_inode_fast is the fast path version of find_inode, see the comment at
810 * iget_locked for details.
811 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530812static struct inode *find_inode_fast(struct super_block *sb,
813 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530815 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
817repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800818 hlist_for_each_entry(inode, head, i_hash) {
Al Viro5a3cd99282013-11-06 09:54:52 -0500819 if (inode->i_ino != ino)
820 continue;
821 if (inode->i_sb != sb)
822 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100823 spin_lock(&inode->i_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -0400824 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 __wait_on_freeing_inode(inode);
826 goto repeat;
827 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400828 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100829 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400830 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400832 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833}
834
Eric Dumazetf991bd22010-10-23 11:18:01 -0400835/*
836 * Each cpu owns a range of LAST_INO_BATCH numbers.
837 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
838 * to renew the exhausted range.
David Chinner8290c352008-10-30 17:35:24 +1100839 *
Eric Dumazetf991bd22010-10-23 11:18:01 -0400840 * This does not significantly increase overflow rate because every CPU can
841 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
842 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
843 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
844 * overflow rate by 2x, which does not seem too significant.
845 *
846 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
847 * error if st_ino won't fit in target struct field. Use 32bit counter
848 * here to attempt to avoid that.
David Chinner8290c352008-10-30 17:35:24 +1100849 */
Eric Dumazetf991bd22010-10-23 11:18:01 -0400850#define LAST_INO_BATCH 1024
851static DEFINE_PER_CPU(unsigned int, last_ino);
David Chinner8290c352008-10-30 17:35:24 +1100852
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400853unsigned int get_next_ino(void)
Eric Dumazetf991bd22010-10-23 11:18:01 -0400854{
855 unsigned int *p = &get_cpu_var(last_ino);
856 unsigned int res = *p;
857
858#ifdef CONFIG_SMP
859 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
860 static atomic_t shared_last_ino;
861 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
862
863 res = next - LAST_INO_BATCH;
864 }
865#endif
866
Carlos Maiolino2adc3762015-06-25 12:25:58 -0300867 res++;
868 /* get_next_ino should not provide a 0 inode number */
869 if (unlikely(!res))
870 res++;
871 *p = res;
Eric Dumazetf991bd22010-10-23 11:18:01 -0400872 put_cpu_var(last_ino);
873 return res;
David Chinner8290c352008-10-30 17:35:24 +1100874}
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400875EXPORT_SYMBOL(get_next_ino);
David Chinner8290c352008-10-30 17:35:24 +1100876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877/**
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200878 * new_inode_pseudo - obtain an inode
879 * @sb: superblock
880 *
881 * Allocates a new inode for given superblock.
882 * Inode wont be chained in superblock s_inodes list
883 * This means :
884 * - fs can't be unmount
885 * - quotas, fsnotify, writeback can't work
886 */
887struct inode *new_inode_pseudo(struct super_block *sb)
888{
889 struct inode *inode = alloc_inode(sb);
890
891 if (inode) {
892 spin_lock(&inode->i_lock);
893 inode->i_state = 0;
894 spin_unlock(&inode->i_lock);
895 INIT_LIST_HEAD(&inode->i_sb_list);
896 }
897 return inode;
898}
899
900/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 * new_inode - obtain an inode
902 * @sb: superblock
903 *
Mel Gorman769848c2007-07-17 04:03:05 -0700904 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800905 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700906 * If HIGHMEM pages are unsuitable or it is known that pages allocated
907 * for the page cache are not reclaimable or migratable,
908 * mapping_set_gfp_mask() must be called with suitable flags on the
909 * newly created inode's mapping
910 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 */
912struct inode *new_inode(struct super_block *sb)
913{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530914 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
Dave Chinner74278da2015-03-04 12:37:22 -0500916 spin_lock_prefetch(&sb->s_inode_list_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530917
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200918 inode = new_inode_pseudo(sb);
919 if (inode)
Dave Chinner55fa6092011-03-22 22:23:40 +1100920 inode_sb_list_add(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 return inode;
922}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923EXPORT_SYMBOL(new_inode);
924
Peter Zijlstra14358e62007-10-14 01:38:33 +0200925#ifdef CONFIG_DEBUG_LOCK_ALLOC
Josh Boyere096d0c2011-08-25 07:48:12 -0400926void lockdep_annotate_inode_mutex_key(struct inode *inode)
927{
Namhyung Kima3314a02010-10-11 22:38:00 +0900928 if (S_ISDIR(inode->i_mode)) {
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200929 struct file_system_type *type = inode->i_sb->s_type;
930
Jan Kara9a7aa122009-06-04 15:26:49 +0200931 /* Set new key only if filesystem hasn't already changed it */
Al Viro9902af72016-04-15 15:08:36 -0400932 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
Jan Kara9a7aa122009-06-04 15:26:49 +0200933 /*
934 * ensure nobody is actually holding i_mutex
935 */
Al Viro9902af72016-04-15 15:08:36 -0400936 // mutex_destroy(&inode->i_mutex);
937 init_rwsem(&inode->i_rwsem);
938 lockdep_set_class(&inode->i_rwsem,
Jan Kara9a7aa122009-06-04 15:26:49 +0200939 &type->i_mutex_dir_key);
940 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200941 }
Josh Boyere096d0c2011-08-25 07:48:12 -0400942}
943EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200944#endif
Josh Boyere096d0c2011-08-25 07:48:12 -0400945
946/**
947 * unlock_new_inode - clear the I_NEW state and wake up any waiters
948 * @inode: new inode to unlock
949 *
950 * Called when the inode is fully initialised to clear the new state of the
951 * inode and wake up anyone waiting for the inode to finish initialisation.
952 */
953void unlock_new_inode(struct inode *inode)
954{
955 lockdep_annotate_inode_mutex_key(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100956 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100957 WARN_ON(!(inode->i_state & I_NEW));
958 inode->i_state &= ~I_NEW;
Al Viro310fa7a2012-03-10 17:07:28 -0500959 smp_mb();
Dave Chinner250df6e2011-03-22 22:23:36 +1100960 wake_up_bit(&inode->i_state, __I_NEW);
961 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963EXPORT_SYMBOL(unlock_new_inode);
964
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400965/**
J. Bruce Fields375e2892012-04-18 15:16:33 -0400966 * lock_two_nondirectories - take two i_mutexes on non-directory objects
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200967 *
968 * Lock any non-NULL argument that is not a directory.
969 * Zero, one or two objects may be locked by this function.
970 *
J. Bruce Fields375e2892012-04-18 15:16:33 -0400971 * @inode1: first inode to lock
972 * @inode2: second inode to lock
973 */
974void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
975{
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200976 if (inode1 > inode2)
977 swap(inode1, inode2);
978
979 if (inode1 && !S_ISDIR(inode1->i_mode))
Al Viro59551022016-01-22 15:40:57 -0500980 inode_lock(inode1);
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200981 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
Al Viro59551022016-01-22 15:40:57 -0500982 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
J. Bruce Fields375e2892012-04-18 15:16:33 -0400983}
984EXPORT_SYMBOL(lock_two_nondirectories);
985
986/**
987 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
988 * @inode1: first inode to unlock
989 * @inode2: second inode to unlock
990 */
991void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
992{
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200993 if (inode1 && !S_ISDIR(inode1->i_mode))
Al Viro59551022016-01-22 15:40:57 -0500994 inode_unlock(inode1);
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200995 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
Al Viro59551022016-01-22 15:40:57 -0500996 inode_unlock(inode2);
J. Bruce Fields375e2892012-04-18 15:16:33 -0400997}
998EXPORT_SYMBOL(unlock_two_nondirectories);
999
1000/**
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001001 * iget5_locked - obtain an inode from a mounted file system
1002 * @sb: super block of file system
1003 * @hashval: hash value (usually inode number) to get
1004 * @test: callback used for comparisons between inodes
1005 * @set: callback used to initialize a new struct inode
1006 * @data: opaque data pointer to pass to @test and @set
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001008 * Search for the inode specified by @hashval and @data in the inode cache,
1009 * and if present it is return it with an increased reference count. This is
1010 * a generalized version of iget_locked() for file systems where the inode
1011 * number is not sufficient for unique identification of an inode.
1012 *
1013 * If the inode is not in cache, allocate a new inode and return it locked,
1014 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1015 * before unlocking it via unlock_new_inode().
1016 *
1017 * Note both @test and @set are called with the inode_hash_lock held, so can't
1018 * sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001020struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1021 int (*test)(struct inode *, void *),
1022 int (*set)(struct inode *, void *), void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001024 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301025 struct inode *inode;
Al Viro2864f302016-07-03 23:15:21 -04001026again:
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001027 spin_lock(&inode_hash_lock);
1028 inode = find_inode(sb, head, test, data);
1029 spin_unlock(&inode_hash_lock);
1030
1031 if (inode) {
1032 wait_on_inode(inode);
Al Viro2864f302016-07-03 23:15:21 -04001033 if (unlikely(inode_unhashed(inode))) {
1034 iput(inode);
1035 goto again;
1036 }
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001037 return inode;
1038 }
1039
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 inode = alloc_inode(sb);
1041 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301042 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Dave Chinner67a23c42011-03-22 22:23:42 +11001044 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 /* We released the lock, so.. */
1046 old = find_inode(sb, head, test, data);
1047 if (!old) {
1048 if (set(inode, data))
1049 goto set_failed;
1050
Dave Chinner250df6e2011-03-22 22:23:36 +11001051 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001052 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001053 hlist_add_head(&inode->i_hash, head);
1054 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001055 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001056 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
1058 /* Return the locked inode with I_NEW set, the
1059 * caller is responsible for filling in the contents
1060 */
1061 return inode;
1062 }
1063
1064 /*
1065 * Uhhuh, somebody else created the same inode under
1066 * us. Use the old inode instead of the one we just
1067 * allocated.
1068 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001069 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 destroy_inode(inode);
1071 inode = old;
1072 wait_on_inode(inode);
Al Viro2864f302016-07-03 23:15:21 -04001073 if (unlikely(inode_unhashed(inode))) {
1074 iput(inode);
1075 goto again;
1076 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 }
1078 return inode;
1079
1080set_failed:
Dave Chinner67a23c42011-03-22 22:23:42 +11001081 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 destroy_inode(inode);
1083 return NULL;
1084}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001085EXPORT_SYMBOL(iget5_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001087/**
1088 * iget_locked - obtain an inode from a mounted file system
1089 * @sb: super block of file system
1090 * @ino: inode number to get
1091 *
1092 * Search for the inode specified by @ino in the inode cache and if present
1093 * return it with an increased reference count. This is for file systems
1094 * where the inode number is sufficient for unique identification of an inode.
1095 *
1096 * If the inode is not in cache, allocate a new inode and return it locked,
1097 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1098 * before unlocking it via unlock_new_inode().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001100struct inode *iget_locked(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001102 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301103 struct inode *inode;
Al Viro2864f302016-07-03 23:15:21 -04001104again:
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001105 spin_lock(&inode_hash_lock);
1106 inode = find_inode_fast(sb, head, ino);
1107 spin_unlock(&inode_hash_lock);
1108 if (inode) {
1109 wait_on_inode(inode);
Al Viro2864f302016-07-03 23:15:21 -04001110 if (unlikely(inode_unhashed(inode))) {
1111 iput(inode);
1112 goto again;
1113 }
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001114 return inode;
1115 }
1116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 inode = alloc_inode(sb);
1118 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301119 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
Dave Chinner67a23c42011-03-22 22:23:42 +11001121 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 /* We released the lock, so.. */
1123 old = find_inode_fast(sb, head, ino);
1124 if (!old) {
1125 inode->i_ino = ino;
Dave Chinner250df6e2011-03-22 22:23:36 +11001126 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001127 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001128 hlist_add_head(&inode->i_hash, head);
1129 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001130 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001131 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
1133 /* Return the locked inode with I_NEW set, the
1134 * caller is responsible for filling in the contents
1135 */
1136 return inode;
1137 }
1138
1139 /*
1140 * Uhhuh, somebody else created the same inode under
1141 * us. Use the old inode instead of the one we just
1142 * allocated.
1143 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001144 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 destroy_inode(inode);
1146 inode = old;
1147 wait_on_inode(inode);
Al Viro2864f302016-07-03 23:15:21 -04001148 if (unlikely(inode_unhashed(inode))) {
1149 iput(inode);
1150 goto again;
1151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 }
1153 return inode;
1154}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001155EXPORT_SYMBOL(iget_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001157/*
1158 * search the inode cache for a matching inode number.
1159 * If we find one, then the inode number we are trying to
1160 * allocate is not unique and so we should not use it.
1161 *
1162 * Returns 1 if the inode number is unique, 0 if it is not.
1163 */
1164static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1165{
1166 struct hlist_head *b = inode_hashtable + hash(sb, ino);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001167 struct inode *inode;
1168
Dave Chinner67a23c42011-03-22 22:23:42 +11001169 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001170 hlist_for_each_entry(inode, b, i_hash) {
Dave Chinner67a23c42011-03-22 22:23:42 +11001171 if (inode->i_ino == ino && inode->i_sb == sb) {
1172 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001173 return 0;
Dave Chinner67a23c42011-03-22 22:23:42 +11001174 }
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001175 }
Dave Chinner67a23c42011-03-22 22:23:42 +11001176 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001177
1178 return 1;
1179}
1180
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181/**
1182 * iunique - get a unique inode number
1183 * @sb: superblock
1184 * @max_reserved: highest reserved inode number
1185 *
1186 * Obtain an inode number that is unique on the system for a given
1187 * superblock. This is used by file systems that have no natural
1188 * permanent inode numbering system. An inode number is returned that
1189 * is higher than the reserved limit but unique.
1190 *
1191 * BUGS:
1192 * With a large number of inodes live on the file system this function
1193 * currently becomes quite slow.
1194 */
1195ino_t iunique(struct super_block *sb, ino_t max_reserved)
1196{
Jeff Layton866b04f2007-05-08 00:32:29 -07001197 /*
1198 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1199 * error if st_ino won't fit in target struct field. Use 32bit counter
1200 * here to attempt to avoid that.
1201 */
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001202 static DEFINE_SPINLOCK(iunique_lock);
Jeff Layton866b04f2007-05-08 00:32:29 -07001203 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001206 spin_lock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001207 do {
1208 if (counter <= max_reserved)
1209 counter = max_reserved + 1;
1210 res = counter++;
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001211 } while (!test_inode_iunique(sb, res));
1212 spin_unlock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001213
1214 return res;
1215}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216EXPORT_SYMBOL(iunique);
1217
1218struct inode *igrab(struct inode *inode)
1219{
Dave Chinner250df6e2011-03-22 22:23:36 +11001220 spin_lock(&inode->i_lock);
1221 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001223 spin_unlock(&inode->i_lock);
1224 } else {
1225 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 /*
1227 * Handle the case where s_op->clear_inode is not been
1228 * called yet, and somebody is calling igrab
1229 * while the inode is getting freed.
1230 */
1231 inode = NULL;
Dave Chinner250df6e2011-03-22 22:23:36 +11001232 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 return inode;
1234}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235EXPORT_SYMBOL(igrab);
1236
1237/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001238 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 * @sb: super block of file system to search
1240 * @hashval: hash value (usually inode number) to search for
1241 * @test: callback used for comparisons between inodes
1242 * @data: opaque data pointer to pass to @test
1243 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001244 * Search for the inode specified by @hashval and @data in the inode cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 * If the inode is in the cache, the inode is returned with an incremented
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001246 * reference count.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001247 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001248 * Note: I_NEW is not waited upon so you have to be very careful what you do
1249 * with the returned inode. You probably should be using ilookup5() instead.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001250 *
Randy Dunlapb6d0ad62011-03-26 13:27:47 -07001251 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001252 */
1253struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1254 int (*test)(struct inode *, void *), void *data)
1255{
1256 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001257 struct inode *inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001258
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001259 spin_lock(&inode_hash_lock);
1260 inode = find_inode(sb, head, test, data);
1261 spin_unlock(&inode_hash_lock);
1262
1263 return inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001264}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001265EXPORT_SYMBOL(ilookup5_nowait);
1266
1267/**
1268 * ilookup5 - search for an inode in the inode cache
1269 * @sb: super block of file system to search
1270 * @hashval: hash value (usually inode number) to search for
1271 * @test: callback used for comparisons between inodes
1272 * @data: opaque data pointer to pass to @test
1273 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001274 * Search for the inode specified by @hashval and @data in the inode cache,
1275 * and if the inode is in the cache, return the inode with an incremented
1276 * reference count. Waits on I_NEW before returning the inode.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001277 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001279 * This is a generalized version of ilookup() for file systems where the
1280 * inode number is not sufficient for unique identification of an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001282 * Note: @test is called with the inode_hash_lock held, so can't sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 */
1284struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1285 int (*test)(struct inode *, void *), void *data)
1286{
Al Viro2864f302016-07-03 23:15:21 -04001287 struct inode *inode;
1288again:
1289 inode = ilookup5_nowait(sb, hashval, test, data);
1290 if (inode) {
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001291 wait_on_inode(inode);
Al Viro2864f302016-07-03 23:15:21 -04001292 if (unlikely(inode_unhashed(inode))) {
1293 iput(inode);
1294 goto again;
1295 }
1296 }
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001297 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299EXPORT_SYMBOL(ilookup5);
1300
1301/**
1302 * ilookup - search for an inode in the inode cache
1303 * @sb: super block of file system to search
1304 * @ino: inode number to search for
1305 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001306 * Search for the inode @ino in the inode cache, and if the inode is in the
1307 * cache, the inode is returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 */
1309struct inode *ilookup(struct super_block *sb, unsigned long ino)
1310{
1311 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001312 struct inode *inode;
Al Viro2864f302016-07-03 23:15:21 -04001313again:
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001314 spin_lock(&inode_hash_lock);
1315 inode = find_inode_fast(sb, head, ino);
1316 spin_unlock(&inode_hash_lock);
1317
Al Viro2864f302016-07-03 23:15:21 -04001318 if (inode) {
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001319 wait_on_inode(inode);
Al Viro2864f302016-07-03 23:15:21 -04001320 if (unlikely(inode_unhashed(inode))) {
1321 iput(inode);
1322 goto again;
1323 }
1324 }
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001325 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327EXPORT_SYMBOL(ilookup);
1328
Theodore Ts'ofe032c42015-02-02 00:37:01 -05001329/**
1330 * find_inode_nowait - find an inode in the inode cache
1331 * @sb: super block of file system to search
1332 * @hashval: hash value (usually inode number) to search for
1333 * @match: callback used for comparisons between inodes
1334 * @data: opaque data pointer to pass to @match
1335 *
1336 * Search for the inode specified by @hashval and @data in the inode
1337 * cache, where the helper function @match will return 0 if the inode
1338 * does not match, 1 if the inode does match, and -1 if the search
1339 * should be stopped. The @match function must be responsible for
1340 * taking the i_lock spin_lock and checking i_state for an inode being
1341 * freed or being initialized, and incrementing the reference count
1342 * before returning 1. It also must not sleep, since it is called with
1343 * the inode_hash_lock spinlock held.
1344 *
1345 * This is a even more generalized version of ilookup5() when the
1346 * function must never block --- find_inode() can block in
1347 * __wait_on_freeing_inode() --- or when the caller can not increment
1348 * the reference count because the resulting iput() might cause an
1349 * inode eviction. The tradeoff is that the @match funtion must be
1350 * very carefully implemented.
1351 */
1352struct inode *find_inode_nowait(struct super_block *sb,
1353 unsigned long hashval,
1354 int (*match)(struct inode *, unsigned long,
1355 void *),
1356 void *data)
1357{
1358 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1359 struct inode *inode, *ret_inode = NULL;
1360 int mval;
1361
1362 spin_lock(&inode_hash_lock);
1363 hlist_for_each_entry(inode, head, i_hash) {
1364 if (inode->i_sb != sb)
1365 continue;
1366 mval = match(inode, hashval, data);
1367 if (mval == 0)
1368 continue;
1369 if (mval == 1)
1370 ret_inode = inode;
1371 goto out;
1372 }
1373out:
1374 spin_unlock(&inode_hash_lock);
1375 return ret_inode;
1376}
1377EXPORT_SYMBOL(find_inode_nowait);
1378
Al Viro261bca82008-12-30 01:48:21 -05001379int insert_inode_locked(struct inode *inode)
1380{
1381 struct super_block *sb = inode->i_sb;
1382 ino_t ino = inode->i_ino;
1383 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001384
Al Viro261bca82008-12-30 01:48:21 -05001385 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001386 struct inode *old = NULL;
Dave Chinner67a23c42011-03-22 22:23:42 +11001387 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001388 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001389 if (old->i_ino != ino)
1390 continue;
1391 if (old->i_sb != sb)
1392 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001393 spin_lock(&old->i_lock);
1394 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1395 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001396 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001397 }
Al Viro72a43d62009-05-13 19:13:40 +01001398 break;
1399 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001400 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001401 spin_lock(&inode->i_lock);
1402 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001403 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001404 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001405 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001406 return 0;
1407 }
1408 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001409 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001410 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001411 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001412 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001413 iput(old);
1414 return -EBUSY;
1415 }
1416 iput(old);
1417 }
1418}
Al Viro261bca82008-12-30 01:48:21 -05001419EXPORT_SYMBOL(insert_inode_locked);
1420
1421int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1422 int (*test)(struct inode *, void *), void *data)
1423{
1424 struct super_block *sb = inode->i_sb;
1425 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001426
Al Viro261bca82008-12-30 01:48:21 -05001427 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001428 struct inode *old = NULL;
1429
Dave Chinner67a23c42011-03-22 22:23:42 +11001430 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001431 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001432 if (old->i_sb != sb)
1433 continue;
1434 if (!test(old, data))
1435 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001436 spin_lock(&old->i_lock);
1437 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1438 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001439 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001440 }
Al Viro72a43d62009-05-13 19:13:40 +01001441 break;
1442 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001443 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001444 spin_lock(&inode->i_lock);
1445 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001446 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001447 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001448 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001449 return 0;
1450 }
1451 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001452 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001453 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001454 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001455 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001456 iput(old);
1457 return -EBUSY;
1458 }
1459 iput(old);
1460 }
1461}
Al Viro261bca82008-12-30 01:48:21 -05001462EXPORT_SYMBOL(insert_inode_locked4);
1463
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
Al Viro45321ac2010-06-07 13:43:19 -04001465int generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466{
Al Viro45321ac2010-06-07 13:43:19 -04001467 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469EXPORT_SYMBOL(generic_delete_inode);
1470
Al Viro45321ac2010-06-07 13:43:19 -04001471/*
Al Viro45321ac2010-06-07 13:43:19 -04001472 * Called when we're dropping the last reference
1473 * to an inode.
1474 *
1475 * Call the FS "drop_inode()" function, defaulting to
1476 * the legacy UNIX filesystem behaviour. If it tells
1477 * us to evict inode, do so. Otherwise, retain inode
1478 * in cache if fs is alive, sync and evict if fs is
1479 * shutting down.
1480 */
1481static void iput_final(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482{
1483 struct super_block *sb = inode->i_sb;
Al Viro45321ac2010-06-07 13:43:19 -04001484 const struct super_operations *op = inode->i_sb->s_op;
1485 int drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486
Dave Chinner250df6e2011-03-22 22:23:36 +11001487 WARN_ON(inode->i_state & I_NEW);
1488
Al Viroe7f59092011-07-07 15:45:59 -04001489 if (op->drop_inode)
Al Viro45321ac2010-06-07 13:43:19 -04001490 drop = op->drop_inode(inode);
1491 else
1492 drop = generic_drop_inode(inode);
1493
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001494 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1495 inode->i_state |= I_REFERENCED;
Jan Kara4eff96d2012-11-26 16:29:51 -08001496 inode_add_lru(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001497 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001498 return;
1499 }
1500
Al Viro45321ac2010-06-07 13:43:19 -04001501 if (!drop) {
Alexander Viro991114c2005-06-23 00:09:01 -07001502 inode->i_state |= I_WILL_FREE;
Dave Chinner250df6e2011-03-22 22:23:36 +11001503 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 write_inode_now(inode, 1);
Dave Chinner250df6e2011-03-22 22:23:36 +11001505 spin_lock(&inode->i_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001506 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001507 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001509
Alexander Viro991114c2005-06-23 00:09:01 -07001510 inode->i_state |= I_FREEING;
Eric Dumazetc4ae0c62011-07-28 06:55:13 +02001511 if (!list_empty(&inode->i_lru))
1512 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001513 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001514
1515 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516}
1517
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301519 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 * @inode: inode to put
1521 *
1522 * Puts an inode, dropping its usage count. If the inode use count hits
1523 * zero, the inode is then freed and may also be destroyed.
1524 *
1525 * Consequently, iput() can sleep.
1526 */
1527void iput(struct inode *inode)
1528{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001529 if (!inode)
1530 return;
1531 BUG_ON(inode->i_state & I_CLEAR);
1532retry:
1533 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1534 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1535 atomic_inc(&inode->i_count);
1536 inode->i_state &= ~I_DIRTY_TIME;
1537 spin_unlock(&inode->i_lock);
1538 trace_writeback_lazytime_iput(inode);
1539 mark_inode_dirty_sync(inode);
1540 goto retry;
1541 }
1542 iput_final(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 }
1544}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545EXPORT_SYMBOL(iput);
1546
1547/**
1548 * bmap - find a block number in a file
1549 * @inode: inode of file
1550 * @block: block to find
1551 *
1552 * Returns the block number on the device holding the inode that
1553 * is the disk block number for the block of the file requested.
1554 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301555 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 * file.
1557 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301558sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559{
1560 sector_t res = 0;
1561 if (inode->i_mapping->a_ops->bmap)
1562 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1563 return res;
1564}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565EXPORT_SYMBOL(bmap);
1566
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001567/*
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001568 * Update times in overlayed inode from underlying real inode
1569 */
1570static void update_ovl_inode_times(struct dentry *dentry, struct inode *inode,
1571 bool rcu)
1572{
1573 if (!rcu) {
1574 struct inode *realinode = d_real_inode(dentry);
1575
1576 if (unlikely(inode != realinode) &&
1577 (!timespec_equal(&inode->i_mtime, &realinode->i_mtime) ||
1578 !timespec_equal(&inode->i_ctime, &realinode->i_ctime))) {
1579 inode->i_mtime = realinode->i_mtime;
1580 inode->i_ctime = realinode->i_ctime;
1581 }
1582 }
1583}
1584
1585/*
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001586 * With relative atime, only update atime if the previous atime is
1587 * earlier than either the ctime or mtime or if at least a day has
1588 * passed since the last atime update.
1589 */
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001590static int relatime_need_update(const struct path *path, struct inode *inode,
1591 struct timespec now, bool rcu)
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001592{
1593
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001594 if (!(path->mnt->mnt_flags & MNT_RELATIME))
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001595 return 1;
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001596
1597 update_ovl_inode_times(path->dentry, inode, rcu);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001598 /*
1599 * Is mtime younger than atime? If yes, update atime:
1600 */
1601 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1602 return 1;
1603 /*
1604 * Is ctime younger than atime? If yes, update atime:
1605 */
1606 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1607 return 1;
1608
1609 /*
1610 * Is the previous atime value older than a day? If yes,
1611 * update atime:
1612 */
1613 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1614 return 1;
1615 /*
1616 * Good, we can skip the atime update:
1617 */
1618 return 0;
1619}
1620
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001621int generic_update_time(struct inode *inode, struct timespec *time, int flags)
Josef Bacikc3b2da32012-03-26 09:59:21 -04001622{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001623 int iflags = I_DIRTY_TIME;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001624
1625 if (flags & S_ATIME)
1626 inode->i_atime = *time;
1627 if (flags & S_VERSION)
1628 inode_inc_iversion(inode);
1629 if (flags & S_CTIME)
1630 inode->i_ctime = *time;
1631 if (flags & S_MTIME)
1632 inode->i_mtime = *time;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001633
1634 if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION))
1635 iflags |= I_DIRTY_SYNC;
1636 __mark_inode_dirty(inode, iflags);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001637 return 0;
1638}
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001639EXPORT_SYMBOL(generic_update_time);
1640
1641/*
1642 * This does the actual work of updating an inodes time or version. Must have
1643 * had called mnt_want_write() before calling this.
1644 */
1645static int update_time(struct inode *inode, struct timespec *time, int flags)
1646{
1647 int (*update_time)(struct inode *, struct timespec *, int);
1648
1649 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1650 generic_update_time;
1651
1652 return update_time(inode, time, flags);
1653}
Josef Bacikc3b2da32012-03-26 09:59:21 -04001654
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001656 * touch_atime - update the access time
Randy Dunlap185553b2012-04-17 17:03:25 -07001657 * @path: the &struct path to update
Randy Dunlap30fdc8e2015-11-09 14:57:58 -08001658 * @inode: inode to update
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 *
1660 * Update the accessed time on an inode and mark it for writeback.
1661 * This function automatically handles read only file systems and media,
1662 * as well as the "noatime" flag and inode specific "noatime" markers.
1663 */
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001664bool __atime_needs_update(const struct path *path, struct inode *inode,
1665 bool rcu)
NeilBrown8fa9dd22015-03-23 13:37:40 +11001666{
1667 struct vfsmount *mnt = path->mnt;
1668 struct timespec now;
1669
1670 if (inode->i_flags & S_NOATIME)
1671 return false;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05001672
1673 /* Atime updates will likely cause i_uid and i_gid to be written
1674 * back improprely if their true value is unknown to the vfs.
1675 */
1676 if (HAS_UNMAPPED_ID(inode))
1677 return false;
1678
NeilBrown8fa9dd22015-03-23 13:37:40 +11001679 if (IS_NOATIME(inode))
1680 return false;
1681 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1682 return false;
1683
1684 if (mnt->mnt_flags & MNT_NOATIME)
1685 return false;
1686 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1687 return false;
1688
Deepa Dinamanic2050a42016-09-14 07:48:06 -07001689 now = current_time(inode);
NeilBrown8fa9dd22015-03-23 13:37:40 +11001690
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001691 if (!relatime_need_update(path, inode, now, rcu))
NeilBrown8fa9dd22015-03-23 13:37:40 +11001692 return false;
1693
1694 if (timespec_equal(&inode->i_atime, &now))
1695 return false;
1696
1697 return true;
1698}
1699
Al Virobadcf2b2013-07-16 18:15:46 +04001700void touch_atime(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701{
Al Viro68ac1232012-03-15 08:21:57 -04001702 struct vfsmount *mnt = path->mnt;
David Howellsdf2b1af2015-03-17 22:26:15 +00001703 struct inode *inode = d_inode(path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 struct timespec now;
1705
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001706 if (!__atime_needs_update(path, inode, false))
Andi Kleenb12536c2009-09-18 13:05:47 -07001707 return;
1708
Jan Kara5d37e9e2012-06-12 16:20:36 +02001709 if (!sb_start_write_trylock(inode->i_sb))
Andi Kleenb12536c2009-09-18 13:05:47 -07001710 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001711
NeilBrown8fa9dd22015-03-23 13:37:40 +11001712 if (__mnt_want_write(mnt) != 0)
Jan Kara5d37e9e2012-06-12 16:20:36 +02001713 goto skip_update;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001714 /*
1715 * File systems can error out when updating inodes if they need to
1716 * allocate new space to modify an inode (such is the case for
1717 * Btrfs), but since we touch atime while walking down the path we
1718 * really don't care if we failed to update the atime of the file,
1719 * so just ignore the return value.
Alexander Block2bc5565282012-06-15 09:49:33 +02001720 * We may also fail on filesystems that have the ability to make parts
1721 * of the fs read only, e.g. subvolumes in Btrfs.
Josef Bacikc3b2da32012-03-26 09:59:21 -04001722 */
Deepa Dinamanic2050a42016-09-14 07:48:06 -07001723 now = current_time(inode);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001724 update_time(inode, &now, S_ATIME);
Jan Kara5d37e9e2012-06-12 16:20:36 +02001725 __mnt_drop_write(mnt);
1726skip_update:
1727 sb_end_write(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001729EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730
Cong Wang3ed37642012-05-15 14:57:33 +08001731/*
1732 * The logic we want is
1733 *
1734 * if suid or (sgid and xgrp)
1735 * remove privs
1736 */
1737int should_remove_suid(struct dentry *dentry)
1738{
David Howellsdf2b1af2015-03-17 22:26:15 +00001739 umode_t mode = d_inode(dentry)->i_mode;
Cong Wang3ed37642012-05-15 14:57:33 +08001740 int kill = 0;
1741
1742 /* suid always must be killed */
1743 if (unlikely(mode & S_ISUID))
1744 kill = ATTR_KILL_SUID;
1745
1746 /*
1747 * sgid without any exec bits is just a mandatory locking mark; leave
1748 * it alone. If some exec bits are set, it's a real sgid; kill it.
1749 */
1750 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1751 kill |= ATTR_KILL_SGID;
1752
1753 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1754 return kill;
1755
1756 return 0;
1757}
1758EXPORT_SYMBOL(should_remove_suid);
1759
Jan Karadbfae0c2015-05-21 16:05:54 +02001760/*
1761 * Return mask of changes for notify_change() that need to be done as a
1762 * response to write or truncate. Return 0 if nothing has to be changed.
1763 * Negative value on error (change should be denied).
1764 */
Jan Kara45f147a2015-05-21 16:05:55 +02001765int dentry_needs_remove_privs(struct dentry *dentry)
Jan Karadbfae0c2015-05-21 16:05:54 +02001766{
Jan Karadbfae0c2015-05-21 16:05:54 +02001767 struct inode *inode = d_inode(dentry);
1768 int mask = 0;
1769 int ret;
1770
1771 if (IS_NOSEC(inode))
1772 return 0;
1773
1774 mask = should_remove_suid(dentry);
1775 ret = security_inode_need_killpriv(dentry);
1776 if (ret < 0)
1777 return ret;
1778 if (ret)
1779 mask |= ATTR_KILL_PRIV;
1780 return mask;
1781}
Jan Karadbfae0c2015-05-21 16:05:54 +02001782
1783static int __remove_privs(struct dentry *dentry, int kill)
Cong Wang3ed37642012-05-15 14:57:33 +08001784{
1785 struct iattr newattrs;
1786
1787 newattrs.ia_valid = ATTR_FORCE | kill;
J. Bruce Fields27ac0ff2011-09-20 17:19:26 -04001788 /*
1789 * Note we call this on write, so notify_change will not
1790 * encounter any conflicting delegations:
1791 */
1792 return notify_change(dentry, &newattrs, NULL);
Cong Wang3ed37642012-05-15 14:57:33 +08001793}
1794
Jan Kara5fa8e0a2015-05-21 16:05:53 +02001795/*
1796 * Remove special file priviledges (suid, capabilities) when file is written
1797 * to or truncated.
1798 */
1799int file_remove_privs(struct file *file)
Cong Wang3ed37642012-05-15 14:57:33 +08001800{
Miklos Szeredic1892c32016-08-03 13:44:27 +02001801 struct dentry *dentry = file_dentry(file);
1802 struct inode *inode = file_inode(file);
Jan Karadbfae0c2015-05-21 16:05:54 +02001803 int kill;
Cong Wang3ed37642012-05-15 14:57:33 +08001804 int error = 0;
1805
1806 /* Fast path for nothing security related */
1807 if (IS_NOSEC(inode))
1808 return 0;
1809
Miklos Szeredic1892c32016-08-03 13:44:27 +02001810 kill = dentry_needs_remove_privs(dentry);
Jan Karadbfae0c2015-05-21 16:05:54 +02001811 if (kill < 0)
1812 return kill;
1813 if (kill)
1814 error = __remove_privs(dentry, kill);
Jan Kara2426f392015-05-21 16:05:52 +02001815 if (!error)
1816 inode_has_no_xattr(inode);
Cong Wang3ed37642012-05-15 14:57:33 +08001817
1818 return error;
1819}
Jan Kara5fa8e0a2015-05-21 16:05:53 +02001820EXPORT_SYMBOL(file_remove_privs);
Cong Wang3ed37642012-05-15 14:57:33 +08001821
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001823 * file_update_time - update mtime and ctime time
1824 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001826 * Update the mtime and ctime members of an inode and mark the inode
1827 * for writeback. Note that this function is meant exclusively for
1828 * usage in the file write path of filesystems, and filesystems may
1829 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001830 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Josef Bacikc3b2da32012-03-26 09:59:21 -04001831 * timestamps are handled by the server. This can return an error for
1832 * file systems who need to allocate space in order to update an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 */
1834
Josef Bacikc3b2da32012-03-26 09:59:21 -04001835int file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836{
Al Viro496ad9a2013-01-23 17:07:38 -05001837 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 struct timespec now;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001839 int sync_it = 0;
1840 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841
Andi Kleence06e0b2009-09-18 13:05:48 -07001842 /* First try to exhaust all avenues to not sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 if (IS_NOCMTIME(inode))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001844 return 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001845
Deepa Dinamanic2050a42016-09-14 07:48:06 -07001846 now = current_time(inode);
Andi Kleence06e0b2009-09-18 13:05:48 -07001847 if (!timespec_equal(&inode->i_mtime, &now))
1848 sync_it = S_MTIME;
1849
1850 if (!timespec_equal(&inode->i_ctime, &now))
1851 sync_it |= S_CTIME;
1852
1853 if (IS_I_VERSION(inode))
1854 sync_it |= S_VERSION;
1855
1856 if (!sync_it)
Josef Bacikc3b2da32012-03-26 09:59:21 -04001857 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
Andi Kleence06e0b2009-09-18 13:05:48 -07001859 /* Finally allowed to write? Takes lock. */
Jan Karaeb04c282012-06-12 16:20:35 +02001860 if (__mnt_want_write_file(file))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001861 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
Josef Bacikc3b2da32012-03-26 09:59:21 -04001863 ret = update_time(inode, &now, sync_it);
Jan Karaeb04c282012-06-12 16:20:35 +02001864 __mnt_drop_write_file(file);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001865
1866 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001868EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869
1870int inode_needs_sync(struct inode *inode)
1871{
1872 if (IS_SYNC(inode))
1873 return 1;
1874 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1875 return 1;
1876 return 0;
1877}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878EXPORT_SYMBOL(inode_needs_sync);
1879
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001881 * If we try to find an inode in the inode hash while it is being
1882 * deleted, we have to wait until the filesystem completes its
1883 * deletion before reporting that it isn't found. This function waits
1884 * until the deletion _might_ have completed. Callers are responsible
1885 * to recheck inode state.
1886 *
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001887 * It doesn't matter if I_NEW is not set initially, a call to
Dave Chinner250df6e2011-03-22 22:23:36 +11001888 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1889 * will DTRT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 */
1891static void __wait_on_freeing_inode(struct inode *inode)
1892{
1893 wait_queue_head_t *wq;
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001894 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1895 wq = bit_waitqueue(&inode->i_state, __I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +11001897 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001898 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 schedule();
1900 finish_wait(wq, &wait.wait);
Dave Chinner67a23c42011-03-22 22:23:42 +11001901 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902}
1903
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904static __initdata unsigned long ihash_entries;
1905static int __init set_ihash_entries(char *str)
1906{
1907 if (!str)
1908 return 0;
1909 ihash_entries = simple_strtoul(str, &str, 0);
1910 return 1;
1911}
1912__setup("ihash_entries=", set_ihash_entries);
1913
1914/*
1915 * Initialize the waitqueues and inode hash table.
1916 */
1917void __init inode_init_early(void)
1918{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001919 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
1921 /* If hashes are distributed across NUMA nodes, defer
1922 * hash allocation until vmalloc space is available.
1923 */
1924 if (hashdist)
1925 return;
1926
1927 inode_hashtable =
1928 alloc_large_system_hash("Inode-cache",
1929 sizeof(struct hlist_head),
1930 ihash_entries,
1931 14,
1932 HASH_EARLY,
1933 &i_hash_shift,
1934 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001935 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 0);
1937
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001938 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1940}
1941
Denis Cheng74bf17c2007-10-16 23:26:30 -07001942void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001944 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
1946 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001947 inode_cachep = kmem_cache_create("inode_cache",
1948 sizeof(struct inode),
1949 0,
1950 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
Vladimir Davydov5d097052016-01-14 15:18:21 -08001951 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
Paul Mundt20c2df82007-07-20 10:11:58 +09001952 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
1954 /* Hash may have been set up in inode_init_early */
1955 if (!hashdist)
1956 return;
1957
1958 inode_hashtable =
1959 alloc_large_system_hash("Inode-cache",
1960 sizeof(struct hlist_head),
1961 ihash_entries,
1962 14,
1963 0,
1964 &i_hash_shift,
1965 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001966 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 0);
1968
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001969 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1971}
1972
1973void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1974{
1975 inode->i_mode = mode;
1976 if (S_ISCHR(mode)) {
1977 inode->i_fop = &def_chr_fops;
1978 inode->i_rdev = rdev;
1979 } else if (S_ISBLK(mode)) {
1980 inode->i_fop = &def_blk_fops;
1981 inode->i_rdev = rdev;
1982 } else if (S_ISFIFO(mode))
Al Viro599a0ac2013-03-12 09:58:10 -04001983 inode->i_fop = &pipefifo_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 else if (S_ISSOCK(mode))
Al Virobd9b51e2014-11-18 23:38:21 -05001985 ; /* leave it no_open_fops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 else
Manish Katiyaraf0d9ae2009-09-18 13:05:43 -07001987 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1988 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1989 inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990}
1991EXPORT_SYMBOL(init_special_inode);
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001992
1993/**
Ben Hutchingseaae668d2011-02-15 12:48:09 +00001994 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001995 * @inode: New inode
1996 * @dir: Directory inode
1997 * @mode: mode of the new inode
1998 */
1999void inode_init_owner(struct inode *inode, const struct inode *dir,
Al Viro62bb1092011-07-24 23:20:18 -04002000 umode_t mode)
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03002001{
2002 inode->i_uid = current_fsuid();
2003 if (dir && dir->i_mode & S_ISGID) {
2004 inode->i_gid = dir->i_gid;
2005 if (S_ISDIR(mode))
2006 mode |= S_ISGID;
2007 } else
2008 inode->i_gid = current_fsgid();
2009 inode->i_mode = mode;
2010}
2011EXPORT_SYMBOL(inode_init_owner);
Serge E. Hallyne795b712011-03-23 16:43:25 -07002012
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002013/**
2014 * inode_owner_or_capable - check current task permissions to inode
2015 * @inode: inode being checked
2016 *
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002017 * Return true if current either has CAP_FOWNER in a namespace with the
2018 * inode owner uid mapped, or owns the file.
Serge E. Hallyne795b712011-03-23 16:43:25 -07002019 */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002020bool inode_owner_or_capable(const struct inode *inode)
Serge E. Hallyne795b712011-03-23 16:43:25 -07002021{
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002022 struct user_namespace *ns;
2023
Eric W. Biederman92361632012-02-08 07:07:50 -08002024 if (uid_eq(current_fsuid(), inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07002025 return true;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002026
2027 ns = current_user_ns();
2028 if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07002029 return true;
2030 return false;
2031}
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002032EXPORT_SYMBOL(inode_owner_or_capable);
Trond Myklebust1d59d612012-05-31 12:22:33 -04002033
2034/*
2035 * Direct i/o helper functions
2036 */
2037static void __inode_dio_wait(struct inode *inode)
2038{
2039 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2040 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2041
2042 do {
2043 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
2044 if (atomic_read(&inode->i_dio_count))
2045 schedule();
2046 } while (atomic_read(&inode->i_dio_count));
2047 finish_wait(wq, &q.wait);
2048}
2049
2050/**
2051 * inode_dio_wait - wait for outstanding DIO requests to finish
2052 * @inode: inode to wait for
2053 *
2054 * Waits for all pending direct I/O requests to finish so that we can
2055 * proceed with a truncate or equivalent operation.
2056 *
2057 * Must be called under a lock that serializes taking new references
2058 * to i_dio_count, usually by inode->i_mutex.
2059 */
2060void inode_dio_wait(struct inode *inode)
2061{
2062 if (atomic_read(&inode->i_dio_count))
2063 __inode_dio_wait(inode);
2064}
2065EXPORT_SYMBOL(inode_dio_wait);
2066
2067/*
Theodore Ts'o5f16f322014-03-24 14:43:12 -04002068 * inode_set_flags - atomically set some inode flags
2069 *
2070 * Note: the caller should be holding i_mutex, or else be sure that
2071 * they have exclusive access to the inode structure (i.e., while the
2072 * inode is being instantiated). The reason for the cmpxchg() loop
2073 * --- which wouldn't be necessary if all code paths which modify
2074 * i_flags actually followed this rule, is that there is at least one
Jan Kara5fa8e0a2015-05-21 16:05:53 +02002075 * code path which doesn't today so we use cmpxchg() out of an abundance
2076 * of caution.
Theodore Ts'o5f16f322014-03-24 14:43:12 -04002077 *
2078 * In the long run, i_mutex is overkill, and we should probably look
2079 * at using the i_lock spinlock to protect i_flags, and then make sure
2080 * it is so documented in include/linux/fs.h and that all code follows
2081 * the locking convention!!
2082 */
2083void inode_set_flags(struct inode *inode, unsigned int flags,
2084 unsigned int mask)
2085{
2086 unsigned int old_flags, new_flags;
2087
2088 WARN_ON_ONCE(flags & ~mask);
2089 do {
2090 old_flags = ACCESS_ONCE(inode->i_flags);
2091 new_flags = (old_flags & ~mask) | flags;
2092 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2093 new_flags) != old_flags));
2094}
2095EXPORT_SYMBOL(inode_set_flags);
Al Viro21fc61c2015-11-17 01:07:57 -05002096
2097void inode_nohighmem(struct inode *inode)
2098{
2099 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2100}
2101EXPORT_SYMBOL(inode_nohighmem);
Deepa Dinamani3cd88662016-09-14 07:48:02 -07002102
2103/**
2104 * current_time - Return FS time
2105 * @inode: inode.
2106 *
2107 * Return the current time truncated to the time granularity supported by
2108 * the fs.
2109 *
2110 * Note that inode and inode->sb cannot be NULL.
2111 * Otherwise, the function warns and returns time without truncation.
2112 */
2113struct timespec current_time(struct inode *inode)
2114{
2115 struct timespec now = current_kernel_time();
2116
2117 if (unlikely(!inode->i_sb)) {
2118 WARN(1, "current_time() called with uninitialized super_block in the inode");
2119 return now;
2120 }
2121
2122 return timespec_trunc(now, inode->i_sb->s_time_gran);
2123}
2124EXPORT_SYMBOL(current_time);