blob: bbd45312a949cfdd66683e2800f883ca8134e8ca [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}
Darrick J. Wong63d184d2017-09-17 14:06:57 -0700640EXPORT_SYMBOL_GPL(evict_inodes);
Al Viro63997e92010-10-25 20:49:35 -0400641
642/**
Christoph Hellwiga0318782010-10-24 19:40:33 +0200643 * invalidate_inodes - attempt to free all inodes on a superblock
644 * @sb: superblock to operate on
NeilBrown93b270f2011-02-24 17:25:47 +1100645 * @kill_dirty: flag to guide handling of dirty inodes
Christoph Hellwiga0318782010-10-24 19:40:33 +0200646 *
647 * Attempts to free all inodes for a given superblock. If there were any
648 * busy inodes return a non-zero value, else zero.
NeilBrown93b270f2011-02-24 17:25:47 +1100649 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
650 * them as busy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 */
NeilBrown93b270f2011-02-24 17:25:47 +1100652int invalidate_inodes(struct super_block *sb, bool kill_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653{
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400654 int busy = 0;
Christoph Hellwiga0318782010-10-24 19:40:33 +0200655 struct inode *inode, *next;
656 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
Dave Chinner74278da2015-03-04 12:37:22 -0500658 spin_lock(&sb->s_inode_list_lock);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200659 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100660 spin_lock(&inode->i_lock);
661 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
662 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100664 }
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500665 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100666 spin_unlock(&inode->i_lock);
NeilBrown93b270f2011-02-24 17:25:47 +1100667 busy = 1;
668 continue;
669 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200670 if (atomic_read(&inode->i_count)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100671 spin_unlock(&inode->i_lock);
Christoph Hellwig99a38912010-10-23 19:07:20 +0200672 busy = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 continue;
674 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200675
Christoph Hellwig99a38912010-10-23 19:07:20 +0200676 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100677 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100678 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100679 list_add(&inode->i_lru, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 }
Dave Chinner74278da2015-03-04 12:37:22 -0500681 spin_unlock(&sb->s_inode_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682
Christoph Hellwiga0318782010-10-24 19:40:33 +0200683 dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
685 return busy;
686}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688/*
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000689 * Isolate the inode from the LRU in preparation for freeing it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 *
691 * Any inodes which are pinned purely because of attached pagecache have their
Nick Piggin9e38d862010-10-23 06:55:17 -0400692 * pagecache removed. If the inode has metadata buffers attached to
693 * mapping->private_list then try to remove them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 *
Nick Piggin9e38d862010-10-23 06:55:17 -0400695 * If the inode has the I_REFERENCED flag set, then it means that it has been
696 * used recently - the flag is set in iput_final(). When we encounter such an
697 * inode, clear the flag and move it to the back of the LRU so it gets another
698 * pass through the LRU before it gets reclaimed. This is necessary because of
699 * the fact we are doing lazy LRU updates to minimise lock contention so the
700 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
701 * with this flag set because they are the inodes that are out of order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800703static enum lru_status inode_lru_isolate(struct list_head *item,
704 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000705{
706 struct list_head *freeable = arg;
707 struct inode *inode = container_of(item, struct inode, i_lru);
708
709 /*
710 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
711 * If we fail to get the lock, just skip it.
712 */
713 if (!spin_trylock(&inode->i_lock))
714 return LRU_SKIP;
715
716 /*
717 * Referenced or dirty inodes are still in use. Give them another pass
718 * through the LRU as we canot reclaim them now.
719 */
720 if (atomic_read(&inode->i_count) ||
721 (inode->i_state & ~I_REFERENCED)) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800722 list_lru_isolate(lru, &inode->i_lru);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000723 spin_unlock(&inode->i_lock);
724 this_cpu_dec(nr_unused);
725 return LRU_REMOVED;
726 }
727
728 /* recently referenced inodes get one more pass */
729 if (inode->i_state & I_REFERENCED) {
730 inode->i_state &= ~I_REFERENCED;
731 spin_unlock(&inode->i_lock);
732 return LRU_ROTATE;
733 }
734
735 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
736 __iget(inode);
737 spin_unlock(&inode->i_lock);
738 spin_unlock(lru_lock);
739 if (remove_inode_buffers(inode)) {
740 unsigned long reap;
741 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
742 if (current_is_kswapd())
743 __count_vm_events(KSWAPD_INODESTEAL, reap);
744 else
745 __count_vm_events(PGINODESTEAL, reap);
746 if (current->reclaim_state)
747 current->reclaim_state->reclaimed_slab += reap;
748 }
749 iput(inode);
750 spin_lock(lru_lock);
751 return LRU_RETRY;
752 }
753
754 WARN_ON(inode->i_state & I_NEW);
755 inode->i_state |= I_FREEING;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800756 list_lru_isolate_move(lru, &inode->i_lru, freeable);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000757 spin_unlock(&inode->i_lock);
758
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000759 this_cpu_dec(nr_unused);
760 return LRU_REMOVED;
761}
762
763/*
764 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
765 * This is called from the superblock shrinker function with a number of inodes
766 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
767 * then are freed outside inode_lock by dispose_list().
768 */
Vladimir Davydov503c3582015-02-12 14:58:47 -0800769long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770{
771 LIST_HEAD(freeable);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000772 long freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
Vladimir Davydov503c3582015-02-12 14:58:47 -0800774 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
775 inode_lru_isolate, &freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 dispose_list(&freeable);
Dave Chinner0a234c62013-08-28 10:17:57 +1000777 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
779
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780static void __wait_on_freeing_inode(struct inode *inode);
781/*
782 * Called with the inode lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530784static struct inode *find_inode(struct super_block *sb,
785 struct hlist_head *head,
786 int (*test)(struct inode *, void *),
787 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530789 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
791repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800792 hlist_for_each_entry(inode, head, i_hash) {
Al Viro5a3cd99282013-11-06 09:54:52 -0500793 if (inode->i_sb != sb)
794 continue;
795 if (!test(inode, data))
796 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100797 spin_lock(&inode->i_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -0400798 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 __wait_on_freeing_inode(inode);
800 goto repeat;
801 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400802 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100803 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400804 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400806 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807}
808
809/*
810 * find_inode_fast is the fast path version of find_inode, see the comment at
811 * iget_locked for details.
812 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530813static struct inode *find_inode_fast(struct super_block *sb,
814 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530816 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
818repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800819 hlist_for_each_entry(inode, head, i_hash) {
Al Viro5a3cd99282013-11-06 09:54:52 -0500820 if (inode->i_ino != ino)
821 continue;
822 if (inode->i_sb != sb)
823 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100824 spin_lock(&inode->i_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -0400825 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 __wait_on_freeing_inode(inode);
827 goto repeat;
828 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400829 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100830 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400831 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400833 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834}
835
Eric Dumazetf991bd22010-10-23 11:18:01 -0400836/*
837 * Each cpu owns a range of LAST_INO_BATCH numbers.
838 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
839 * to renew the exhausted range.
David Chinner8290c352008-10-30 17:35:24 +1100840 *
Eric Dumazetf991bd22010-10-23 11:18:01 -0400841 * This does not significantly increase overflow rate because every CPU can
842 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
843 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
844 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
845 * overflow rate by 2x, which does not seem too significant.
846 *
847 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
848 * error if st_ino won't fit in target struct field. Use 32bit counter
849 * here to attempt to avoid that.
David Chinner8290c352008-10-30 17:35:24 +1100850 */
Eric Dumazetf991bd22010-10-23 11:18:01 -0400851#define LAST_INO_BATCH 1024
852static DEFINE_PER_CPU(unsigned int, last_ino);
David Chinner8290c352008-10-30 17:35:24 +1100853
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400854unsigned int get_next_ino(void)
Eric Dumazetf991bd22010-10-23 11:18:01 -0400855{
856 unsigned int *p = &get_cpu_var(last_ino);
857 unsigned int res = *p;
858
859#ifdef CONFIG_SMP
860 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
861 static atomic_t shared_last_ino;
862 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
863
864 res = next - LAST_INO_BATCH;
865 }
866#endif
867
Carlos Maiolino2adc3762015-06-25 12:25:58 -0300868 res++;
869 /* get_next_ino should not provide a 0 inode number */
870 if (unlikely(!res))
871 res++;
872 *p = res;
Eric Dumazetf991bd22010-10-23 11:18:01 -0400873 put_cpu_var(last_ino);
874 return res;
David Chinner8290c352008-10-30 17:35:24 +1100875}
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400876EXPORT_SYMBOL(get_next_ino);
David Chinner8290c352008-10-30 17:35:24 +1100877
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878/**
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200879 * new_inode_pseudo - obtain an inode
880 * @sb: superblock
881 *
882 * Allocates a new inode for given superblock.
883 * Inode wont be chained in superblock s_inodes list
884 * This means :
885 * - fs can't be unmount
886 * - quotas, fsnotify, writeback can't work
887 */
888struct inode *new_inode_pseudo(struct super_block *sb)
889{
890 struct inode *inode = alloc_inode(sb);
891
892 if (inode) {
893 spin_lock(&inode->i_lock);
894 inode->i_state = 0;
895 spin_unlock(&inode->i_lock);
896 INIT_LIST_HEAD(&inode->i_sb_list);
897 }
898 return inode;
899}
900
901/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 * new_inode - obtain an inode
903 * @sb: superblock
904 *
Mel Gorman769848c2007-07-17 04:03:05 -0700905 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800906 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700907 * If HIGHMEM pages are unsuitable or it is known that pages allocated
908 * for the page cache are not reclaimable or migratable,
909 * mapping_set_gfp_mask() must be called with suitable flags on the
910 * newly created inode's mapping
911 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 */
913struct inode *new_inode(struct super_block *sb)
914{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530915 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
Dave Chinner74278da2015-03-04 12:37:22 -0500917 spin_lock_prefetch(&sb->s_inode_list_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530918
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200919 inode = new_inode_pseudo(sb);
920 if (inode)
Dave Chinner55fa6092011-03-22 22:23:40 +1100921 inode_sb_list_add(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 return inode;
923}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924EXPORT_SYMBOL(new_inode);
925
Peter Zijlstra14358e62007-10-14 01:38:33 +0200926#ifdef CONFIG_DEBUG_LOCK_ALLOC
Josh Boyere096d0c2011-08-25 07:48:12 -0400927void lockdep_annotate_inode_mutex_key(struct inode *inode)
928{
Namhyung Kima3314a02010-10-11 22:38:00 +0900929 if (S_ISDIR(inode->i_mode)) {
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200930 struct file_system_type *type = inode->i_sb->s_type;
931
Jan Kara9a7aa122009-06-04 15:26:49 +0200932 /* Set new key only if filesystem hasn't already changed it */
Al Viro9902af72016-04-15 15:08:36 -0400933 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
Jan Kara9a7aa122009-06-04 15:26:49 +0200934 /*
935 * ensure nobody is actually holding i_mutex
936 */
Al Viro9902af72016-04-15 15:08:36 -0400937 // mutex_destroy(&inode->i_mutex);
938 init_rwsem(&inode->i_rwsem);
939 lockdep_set_class(&inode->i_rwsem,
Jan Kara9a7aa122009-06-04 15:26:49 +0200940 &type->i_mutex_dir_key);
941 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200942 }
Josh Boyere096d0c2011-08-25 07:48:12 -0400943}
944EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200945#endif
Josh Boyere096d0c2011-08-25 07:48:12 -0400946
947/**
948 * unlock_new_inode - clear the I_NEW state and wake up any waiters
949 * @inode: new inode to unlock
950 *
951 * Called when the inode is fully initialised to clear the new state of the
952 * inode and wake up anyone waiting for the inode to finish initialisation.
953 */
954void unlock_new_inode(struct inode *inode)
955{
956 lockdep_annotate_inode_mutex_key(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100957 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100958 WARN_ON(!(inode->i_state & I_NEW));
959 inode->i_state &= ~I_NEW;
Al Viro310fa7a2012-03-10 17:07:28 -0500960 smp_mb();
Dave Chinner250df6e2011-03-22 22:23:36 +1100961 wake_up_bit(&inode->i_state, __I_NEW);
962 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964EXPORT_SYMBOL(unlock_new_inode);
965
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400966/**
J. Bruce Fields375e2892012-04-18 15:16:33 -0400967 * lock_two_nondirectories - take two i_mutexes on non-directory objects
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200968 *
969 * Lock any non-NULL argument that is not a directory.
970 * Zero, one or two objects may be locked by this function.
971 *
J. Bruce Fields375e2892012-04-18 15:16:33 -0400972 * @inode1: first inode to lock
973 * @inode2: second inode to lock
974 */
975void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
976{
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200977 if (inode1 > inode2)
978 swap(inode1, inode2);
979
980 if (inode1 && !S_ISDIR(inode1->i_mode))
Al Viro59551022016-01-22 15:40:57 -0500981 inode_lock(inode1);
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200982 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
Al Viro59551022016-01-22 15:40:57 -0500983 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
J. Bruce Fields375e2892012-04-18 15:16:33 -0400984}
985EXPORT_SYMBOL(lock_two_nondirectories);
986
987/**
988 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
989 * @inode1: first inode to unlock
990 * @inode2: second inode to unlock
991 */
992void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
993{
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200994 if (inode1 && !S_ISDIR(inode1->i_mode))
Al Viro59551022016-01-22 15:40:57 -0500995 inode_unlock(inode1);
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200996 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
Al Viro59551022016-01-22 15:40:57 -0500997 inode_unlock(inode2);
J. Bruce Fields375e2892012-04-18 15:16:33 -0400998}
999EXPORT_SYMBOL(unlock_two_nondirectories);
1000
1001/**
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001002 * iget5_locked - obtain an inode from a mounted file system
1003 * @sb: super block of file system
1004 * @hashval: hash value (usually inode number) to get
1005 * @test: callback used for comparisons between inodes
1006 * @set: callback used to initialize a new struct inode
1007 * @data: opaque data pointer to pass to @test and @set
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001009 * Search for the inode specified by @hashval and @data in the inode cache,
1010 * and if present it is return it with an increased reference count. This is
1011 * a generalized version of iget_locked() for file systems where the inode
1012 * number is not sufficient for unique identification of an inode.
1013 *
1014 * If the inode is not in cache, allocate a new inode and return it locked,
1015 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1016 * before unlocking it via unlock_new_inode().
1017 *
1018 * Note both @test and @set are called with the inode_hash_lock held, so can't
1019 * sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001021struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1022 int (*test)(struct inode *, void *),
1023 int (*set)(struct inode *, void *), void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001025 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301026 struct inode *inode;
Al Viro2864f302016-07-03 23:15:21 -04001027again:
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001028 spin_lock(&inode_hash_lock);
1029 inode = find_inode(sb, head, test, data);
1030 spin_unlock(&inode_hash_lock);
1031
1032 if (inode) {
1033 wait_on_inode(inode);
Al Viro2864f302016-07-03 23:15:21 -04001034 if (unlikely(inode_unhashed(inode))) {
1035 iput(inode);
1036 goto again;
1037 }
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001038 return inode;
1039 }
1040
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 inode = alloc_inode(sb);
1042 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301043 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
Dave Chinner67a23c42011-03-22 22:23:42 +11001045 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 /* We released the lock, so.. */
1047 old = find_inode(sb, head, test, data);
1048 if (!old) {
1049 if (set(inode, data))
1050 goto set_failed;
1051
Dave Chinner250df6e2011-03-22 22:23:36 +11001052 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001053 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001054 hlist_add_head(&inode->i_hash, head);
1055 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001056 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001057 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
1059 /* Return the locked inode with I_NEW set, the
1060 * caller is responsible for filling in the contents
1061 */
1062 return inode;
1063 }
1064
1065 /*
1066 * Uhhuh, somebody else created the same inode under
1067 * us. Use the old inode instead of the one we just
1068 * allocated.
1069 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001070 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 destroy_inode(inode);
1072 inode = old;
1073 wait_on_inode(inode);
Al Viro2864f302016-07-03 23:15:21 -04001074 if (unlikely(inode_unhashed(inode))) {
1075 iput(inode);
1076 goto again;
1077 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 }
1079 return inode;
1080
1081set_failed:
Dave Chinner67a23c42011-03-22 22:23:42 +11001082 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 destroy_inode(inode);
1084 return NULL;
1085}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001086EXPORT_SYMBOL(iget5_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001088/**
1089 * iget_locked - obtain an inode from a mounted file system
1090 * @sb: super block of file system
1091 * @ino: inode number to get
1092 *
1093 * Search for the inode specified by @ino in the inode cache and if present
1094 * return it with an increased reference count. This is for file systems
1095 * where the inode number is sufficient for unique identification of an inode.
1096 *
1097 * If the inode is not in cache, allocate a new inode and return it locked,
1098 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1099 * before unlocking it via unlock_new_inode().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001101struct inode *iget_locked(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001103 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301104 struct inode *inode;
Al Viro2864f302016-07-03 23:15:21 -04001105again:
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001106 spin_lock(&inode_hash_lock);
1107 inode = find_inode_fast(sb, head, ino);
1108 spin_unlock(&inode_hash_lock);
1109 if (inode) {
1110 wait_on_inode(inode);
Al Viro2864f302016-07-03 23:15:21 -04001111 if (unlikely(inode_unhashed(inode))) {
1112 iput(inode);
1113 goto again;
1114 }
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001115 return inode;
1116 }
1117
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 inode = alloc_inode(sb);
1119 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301120 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
Dave Chinner67a23c42011-03-22 22:23:42 +11001122 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 /* We released the lock, so.. */
1124 old = find_inode_fast(sb, head, ino);
1125 if (!old) {
1126 inode->i_ino = ino;
Dave Chinner250df6e2011-03-22 22:23:36 +11001127 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001128 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001129 hlist_add_head(&inode->i_hash, head);
1130 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001131 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001132 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133
1134 /* Return the locked inode with I_NEW set, the
1135 * caller is responsible for filling in the contents
1136 */
1137 return inode;
1138 }
1139
1140 /*
1141 * Uhhuh, somebody else created the same inode under
1142 * us. Use the old inode instead of the one we just
1143 * allocated.
1144 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001145 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 destroy_inode(inode);
1147 inode = old;
1148 wait_on_inode(inode);
Al Viro2864f302016-07-03 23:15:21 -04001149 if (unlikely(inode_unhashed(inode))) {
1150 iput(inode);
1151 goto again;
1152 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 }
1154 return inode;
1155}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001156EXPORT_SYMBOL(iget_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001158/*
1159 * search the inode cache for a matching inode number.
1160 * If we find one, then the inode number we are trying to
1161 * allocate is not unique and so we should not use it.
1162 *
1163 * Returns 1 if the inode number is unique, 0 if it is not.
1164 */
1165static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1166{
1167 struct hlist_head *b = inode_hashtable + hash(sb, ino);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001168 struct inode *inode;
1169
Dave Chinner67a23c42011-03-22 22:23:42 +11001170 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001171 hlist_for_each_entry(inode, b, i_hash) {
Dave Chinner67a23c42011-03-22 22:23:42 +11001172 if (inode->i_ino == ino && inode->i_sb == sb) {
1173 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001174 return 0;
Dave Chinner67a23c42011-03-22 22:23:42 +11001175 }
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001176 }
Dave Chinner67a23c42011-03-22 22:23:42 +11001177 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001178
1179 return 1;
1180}
1181
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182/**
1183 * iunique - get a unique inode number
1184 * @sb: superblock
1185 * @max_reserved: highest reserved inode number
1186 *
1187 * Obtain an inode number that is unique on the system for a given
1188 * superblock. This is used by file systems that have no natural
1189 * permanent inode numbering system. An inode number is returned that
1190 * is higher than the reserved limit but unique.
1191 *
1192 * BUGS:
1193 * With a large number of inodes live on the file system this function
1194 * currently becomes quite slow.
1195 */
1196ino_t iunique(struct super_block *sb, ino_t max_reserved)
1197{
Jeff Layton866b04f2007-05-08 00:32:29 -07001198 /*
1199 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1200 * error if st_ino won't fit in target struct field. Use 32bit counter
1201 * here to attempt to avoid that.
1202 */
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001203 static DEFINE_SPINLOCK(iunique_lock);
Jeff Layton866b04f2007-05-08 00:32:29 -07001204 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001207 spin_lock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001208 do {
1209 if (counter <= max_reserved)
1210 counter = max_reserved + 1;
1211 res = counter++;
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001212 } while (!test_inode_iunique(sb, res));
1213 spin_unlock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001214
1215 return res;
1216}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217EXPORT_SYMBOL(iunique);
1218
1219struct inode *igrab(struct inode *inode)
1220{
Dave Chinner250df6e2011-03-22 22:23:36 +11001221 spin_lock(&inode->i_lock);
1222 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001224 spin_unlock(&inode->i_lock);
1225 } else {
1226 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 /*
1228 * Handle the case where s_op->clear_inode is not been
1229 * called yet, and somebody is calling igrab
1230 * while the inode is getting freed.
1231 */
1232 inode = NULL;
Dave Chinner250df6e2011-03-22 22:23:36 +11001233 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 return inode;
1235}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236EXPORT_SYMBOL(igrab);
1237
1238/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001239 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 * @sb: super block of file system to search
1241 * @hashval: hash value (usually inode number) to search for
1242 * @test: callback used for comparisons between inodes
1243 * @data: opaque data pointer to pass to @test
1244 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001245 * Search for the inode specified by @hashval and @data in the inode cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 * If the inode is in the cache, the inode is returned with an incremented
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001247 * reference count.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001248 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001249 * Note: I_NEW is not waited upon so you have to be very careful what you do
1250 * with the returned inode. You probably should be using ilookup5() instead.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001251 *
Randy Dunlapb6d0ad62011-03-26 13:27:47 -07001252 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001253 */
1254struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1255 int (*test)(struct inode *, void *), void *data)
1256{
1257 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001258 struct inode *inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001259
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001260 spin_lock(&inode_hash_lock);
1261 inode = find_inode(sb, head, test, data);
1262 spin_unlock(&inode_hash_lock);
1263
1264 return inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001265}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001266EXPORT_SYMBOL(ilookup5_nowait);
1267
1268/**
1269 * ilookup5 - search for an inode in the inode cache
1270 * @sb: super block of file system to search
1271 * @hashval: hash value (usually inode number) to search for
1272 * @test: callback used for comparisons between inodes
1273 * @data: opaque data pointer to pass to @test
1274 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001275 * Search for the inode specified by @hashval and @data in the inode cache,
1276 * and if the inode is in the cache, return the inode with an incremented
1277 * reference count. Waits on I_NEW before returning the inode.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001278 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001280 * This is a generalized version of ilookup() for file systems where the
1281 * inode number is not sufficient for unique identification of an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001283 * Note: @test is called with the inode_hash_lock held, so can't sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 */
1285struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1286 int (*test)(struct inode *, void *), void *data)
1287{
Al Viro2864f302016-07-03 23:15:21 -04001288 struct inode *inode;
1289again:
1290 inode = ilookup5_nowait(sb, hashval, test, data);
1291 if (inode) {
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001292 wait_on_inode(inode);
Al Viro2864f302016-07-03 23:15:21 -04001293 if (unlikely(inode_unhashed(inode))) {
1294 iput(inode);
1295 goto again;
1296 }
1297 }
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001298 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300EXPORT_SYMBOL(ilookup5);
1301
1302/**
1303 * ilookup - search for an inode in the inode cache
1304 * @sb: super block of file system to search
1305 * @ino: inode number to search for
1306 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001307 * Search for the inode @ino in the inode cache, and if the inode is in the
1308 * cache, the inode is returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 */
1310struct inode *ilookup(struct super_block *sb, unsigned long ino)
1311{
1312 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001313 struct inode *inode;
Al Viro2864f302016-07-03 23:15:21 -04001314again:
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001315 spin_lock(&inode_hash_lock);
1316 inode = find_inode_fast(sb, head, ino);
1317 spin_unlock(&inode_hash_lock);
1318
Al Viro2864f302016-07-03 23:15:21 -04001319 if (inode) {
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001320 wait_on_inode(inode);
Al Viro2864f302016-07-03 23:15:21 -04001321 if (unlikely(inode_unhashed(inode))) {
1322 iput(inode);
1323 goto again;
1324 }
1325 }
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001326 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328EXPORT_SYMBOL(ilookup);
1329
Theodore Ts'ofe032c42015-02-02 00:37:01 -05001330/**
1331 * find_inode_nowait - find an inode in the inode cache
1332 * @sb: super block of file system to search
1333 * @hashval: hash value (usually inode number) to search for
1334 * @match: callback used for comparisons between inodes
1335 * @data: opaque data pointer to pass to @match
1336 *
1337 * Search for the inode specified by @hashval and @data in the inode
1338 * cache, where the helper function @match will return 0 if the inode
1339 * does not match, 1 if the inode does match, and -1 if the search
1340 * should be stopped. The @match function must be responsible for
1341 * taking the i_lock spin_lock and checking i_state for an inode being
1342 * freed or being initialized, and incrementing the reference count
1343 * before returning 1. It also must not sleep, since it is called with
1344 * the inode_hash_lock spinlock held.
1345 *
1346 * This is a even more generalized version of ilookup5() when the
1347 * function must never block --- find_inode() can block in
1348 * __wait_on_freeing_inode() --- or when the caller can not increment
1349 * the reference count because the resulting iput() might cause an
1350 * inode eviction. The tradeoff is that the @match funtion must be
1351 * very carefully implemented.
1352 */
1353struct inode *find_inode_nowait(struct super_block *sb,
1354 unsigned long hashval,
1355 int (*match)(struct inode *, unsigned long,
1356 void *),
1357 void *data)
1358{
1359 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1360 struct inode *inode, *ret_inode = NULL;
1361 int mval;
1362
1363 spin_lock(&inode_hash_lock);
1364 hlist_for_each_entry(inode, head, i_hash) {
1365 if (inode->i_sb != sb)
1366 continue;
1367 mval = match(inode, hashval, data);
1368 if (mval == 0)
1369 continue;
1370 if (mval == 1)
1371 ret_inode = inode;
1372 goto out;
1373 }
1374out:
1375 spin_unlock(&inode_hash_lock);
1376 return ret_inode;
1377}
1378EXPORT_SYMBOL(find_inode_nowait);
1379
Al Viro261bca82008-12-30 01:48:21 -05001380int insert_inode_locked(struct inode *inode)
1381{
1382 struct super_block *sb = inode->i_sb;
1383 ino_t ino = inode->i_ino;
1384 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001385
Al Viro261bca82008-12-30 01:48:21 -05001386 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001387 struct inode *old = NULL;
Dave Chinner67a23c42011-03-22 22:23:42 +11001388 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001389 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001390 if (old->i_ino != ino)
1391 continue;
1392 if (old->i_sb != sb)
1393 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001394 spin_lock(&old->i_lock);
1395 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1396 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001397 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001398 }
Al Viro72a43d62009-05-13 19:13:40 +01001399 break;
1400 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001401 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001402 spin_lock(&inode->i_lock);
1403 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001404 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001405 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001406 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001407 return 0;
1408 }
1409 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001410 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001411 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001412 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001413 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001414 iput(old);
1415 return -EBUSY;
1416 }
1417 iput(old);
1418 }
1419}
Al Viro261bca82008-12-30 01:48:21 -05001420EXPORT_SYMBOL(insert_inode_locked);
1421
1422int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1423 int (*test)(struct inode *, void *), void *data)
1424{
1425 struct super_block *sb = inode->i_sb;
1426 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001427
Al Viro261bca82008-12-30 01:48:21 -05001428 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001429 struct inode *old = NULL;
1430
Dave Chinner67a23c42011-03-22 22:23:42 +11001431 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001432 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001433 if (old->i_sb != sb)
1434 continue;
1435 if (!test(old, data))
1436 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001437 spin_lock(&old->i_lock);
1438 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1439 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001440 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001441 }
Al Viro72a43d62009-05-13 19:13:40 +01001442 break;
1443 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001444 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001445 spin_lock(&inode->i_lock);
1446 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001447 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001448 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001449 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001450 return 0;
1451 }
1452 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001453 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001454 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001455 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001456 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001457 iput(old);
1458 return -EBUSY;
1459 }
1460 iput(old);
1461 }
1462}
Al Viro261bca82008-12-30 01:48:21 -05001463EXPORT_SYMBOL(insert_inode_locked4);
1464
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465
Al Viro45321ac2010-06-07 13:43:19 -04001466int generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467{
Al Viro45321ac2010-06-07 13:43:19 -04001468 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470EXPORT_SYMBOL(generic_delete_inode);
1471
Al Viro45321ac2010-06-07 13:43:19 -04001472/*
Al Viro45321ac2010-06-07 13:43:19 -04001473 * Called when we're dropping the last reference
1474 * to an inode.
1475 *
1476 * Call the FS "drop_inode()" function, defaulting to
1477 * the legacy UNIX filesystem behaviour. If it tells
1478 * us to evict inode, do so. Otherwise, retain inode
1479 * in cache if fs is alive, sync and evict if fs is
1480 * shutting down.
1481 */
1482static void iput_final(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483{
1484 struct super_block *sb = inode->i_sb;
Al Viro45321ac2010-06-07 13:43:19 -04001485 const struct super_operations *op = inode->i_sb->s_op;
1486 int drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
Dave Chinner250df6e2011-03-22 22:23:36 +11001488 WARN_ON(inode->i_state & I_NEW);
1489
Al Viroe7f59092011-07-07 15:45:59 -04001490 if (op->drop_inode)
Al Viro45321ac2010-06-07 13:43:19 -04001491 drop = op->drop_inode(inode);
1492 else
1493 drop = generic_drop_inode(inode);
1494
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001495 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1496 inode->i_state |= I_REFERENCED;
Jan Kara4eff96d2012-11-26 16:29:51 -08001497 inode_add_lru(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001498 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001499 return;
1500 }
1501
Al Viro45321ac2010-06-07 13:43:19 -04001502 if (!drop) {
Alexander Viro991114c2005-06-23 00:09:01 -07001503 inode->i_state |= I_WILL_FREE;
Dave Chinner250df6e2011-03-22 22:23:36 +11001504 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 write_inode_now(inode, 1);
Dave Chinner250df6e2011-03-22 22:23:36 +11001506 spin_lock(&inode->i_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001507 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001508 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001510
Alexander Viro991114c2005-06-23 00:09:01 -07001511 inode->i_state |= I_FREEING;
Eric Dumazetc4ae0c62011-07-28 06:55:13 +02001512 if (!list_empty(&inode->i_lru))
1513 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001514 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001515
1516 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517}
1518
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301520 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 * @inode: inode to put
1522 *
1523 * Puts an inode, dropping its usage count. If the inode use count hits
1524 * zero, the inode is then freed and may also be destroyed.
1525 *
1526 * Consequently, iput() can sleep.
1527 */
1528void iput(struct inode *inode)
1529{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001530 if (!inode)
1531 return;
1532 BUG_ON(inode->i_state & I_CLEAR);
1533retry:
1534 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1535 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1536 atomic_inc(&inode->i_count);
1537 inode->i_state &= ~I_DIRTY_TIME;
1538 spin_unlock(&inode->i_lock);
1539 trace_writeback_lazytime_iput(inode);
1540 mark_inode_dirty_sync(inode);
1541 goto retry;
1542 }
1543 iput_final(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 }
1545}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546EXPORT_SYMBOL(iput);
1547
1548/**
1549 * bmap - find a block number in a file
1550 * @inode: inode of file
1551 * @block: block to find
1552 *
1553 * Returns the block number on the device holding the inode that
1554 * is the disk block number for the block of the file requested.
1555 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301556 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 * file.
1558 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301559sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560{
1561 sector_t res = 0;
1562 if (inode->i_mapping->a_ops->bmap)
1563 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1564 return res;
1565}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566EXPORT_SYMBOL(bmap);
1567
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001568/*
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001569 * Update times in overlayed inode from underlying real inode
1570 */
1571static void update_ovl_inode_times(struct dentry *dentry, struct inode *inode,
1572 bool rcu)
1573{
1574 if (!rcu) {
1575 struct inode *realinode = d_real_inode(dentry);
1576
1577 if (unlikely(inode != realinode) &&
1578 (!timespec_equal(&inode->i_mtime, &realinode->i_mtime) ||
1579 !timespec_equal(&inode->i_ctime, &realinode->i_ctime))) {
1580 inode->i_mtime = realinode->i_mtime;
1581 inode->i_ctime = realinode->i_ctime;
1582 }
1583 }
1584}
1585
1586/*
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001587 * With relative atime, only update atime if the previous atime is
1588 * earlier than either the ctime or mtime or if at least a day has
1589 * passed since the last atime update.
1590 */
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001591static int relatime_need_update(const struct path *path, struct inode *inode,
1592 struct timespec now, bool rcu)
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001593{
1594
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001595 if (!(path->mnt->mnt_flags & MNT_RELATIME))
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001596 return 1;
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001597
1598 update_ovl_inode_times(path->dentry, inode, rcu);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001599 /*
1600 * Is mtime younger than atime? If yes, update atime:
1601 */
1602 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1603 return 1;
1604 /*
1605 * Is ctime younger than atime? If yes, update atime:
1606 */
1607 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1608 return 1;
1609
1610 /*
1611 * Is the previous atime value older than a day? If yes,
1612 * update atime:
1613 */
1614 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1615 return 1;
1616 /*
1617 * Good, we can skip the atime update:
1618 */
1619 return 0;
1620}
1621
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001622int generic_update_time(struct inode *inode, struct timespec *time, int flags)
Josef Bacikc3b2da32012-03-26 09:59:21 -04001623{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001624 int iflags = I_DIRTY_TIME;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001625
1626 if (flags & S_ATIME)
1627 inode->i_atime = *time;
1628 if (flags & S_VERSION)
1629 inode_inc_iversion(inode);
1630 if (flags & S_CTIME)
1631 inode->i_ctime = *time;
1632 if (flags & S_MTIME)
1633 inode->i_mtime = *time;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001634
1635 if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION))
1636 iflags |= I_DIRTY_SYNC;
1637 __mark_inode_dirty(inode, iflags);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001638 return 0;
1639}
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001640EXPORT_SYMBOL(generic_update_time);
1641
1642/*
1643 * This does the actual work of updating an inodes time or version. Must have
1644 * had called mnt_want_write() before calling this.
1645 */
1646static int update_time(struct inode *inode, struct timespec *time, int flags)
1647{
1648 int (*update_time)(struct inode *, struct timespec *, int);
1649
1650 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1651 generic_update_time;
1652
1653 return update_time(inode, time, flags);
1654}
Josef Bacikc3b2da32012-03-26 09:59:21 -04001655
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001657 * touch_atime - update the access time
Randy Dunlap185553b2012-04-17 17:03:25 -07001658 * @path: the &struct path to update
Randy Dunlap30fdc8e2015-11-09 14:57:58 -08001659 * @inode: inode to update
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 *
1661 * Update the accessed time on an inode and mark it for writeback.
1662 * This function automatically handles read only file systems and media,
1663 * as well as the "noatime" flag and inode specific "noatime" markers.
1664 */
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001665bool __atime_needs_update(const struct path *path, struct inode *inode,
1666 bool rcu)
NeilBrown8fa9dd22015-03-23 13:37:40 +11001667{
1668 struct vfsmount *mnt = path->mnt;
1669 struct timespec now;
1670
1671 if (inode->i_flags & S_NOATIME)
1672 return false;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05001673
1674 /* Atime updates will likely cause i_uid and i_gid to be written
1675 * back improprely if their true value is unknown to the vfs.
1676 */
1677 if (HAS_UNMAPPED_ID(inode))
1678 return false;
1679
NeilBrown8fa9dd22015-03-23 13:37:40 +11001680 if (IS_NOATIME(inode))
1681 return false;
1682 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1683 return false;
1684
1685 if (mnt->mnt_flags & MNT_NOATIME)
1686 return false;
1687 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1688 return false;
1689
Deepa Dinamanic2050a42016-09-14 07:48:06 -07001690 now = current_time(inode);
NeilBrown8fa9dd22015-03-23 13:37:40 +11001691
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001692 if (!relatime_need_update(path, inode, now, rcu))
NeilBrown8fa9dd22015-03-23 13:37:40 +11001693 return false;
1694
1695 if (timespec_equal(&inode->i_atime, &now))
1696 return false;
1697
1698 return true;
1699}
1700
Al Virobadcf2b2013-07-16 18:15:46 +04001701void touch_atime(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702{
Al Viro68ac1232012-03-15 08:21:57 -04001703 struct vfsmount *mnt = path->mnt;
David Howellsdf2b1af2015-03-17 22:26:15 +00001704 struct inode *inode = d_inode(path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 struct timespec now;
1706
Miklos Szeredi598e3c82016-09-16 12:44:20 +02001707 if (!__atime_needs_update(path, inode, false))
Andi Kleenb12536c2009-09-18 13:05:47 -07001708 return;
1709
Jan Kara5d37e9e2012-06-12 16:20:36 +02001710 if (!sb_start_write_trylock(inode->i_sb))
Andi Kleenb12536c2009-09-18 13:05:47 -07001711 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001712
NeilBrown8fa9dd22015-03-23 13:37:40 +11001713 if (__mnt_want_write(mnt) != 0)
Jan Kara5d37e9e2012-06-12 16:20:36 +02001714 goto skip_update;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001715 /*
1716 * File systems can error out when updating inodes if they need to
1717 * allocate new space to modify an inode (such is the case for
1718 * Btrfs), but since we touch atime while walking down the path we
1719 * really don't care if we failed to update the atime of the file,
1720 * so just ignore the return value.
Alexander Block2bc5565282012-06-15 09:49:33 +02001721 * We may also fail on filesystems that have the ability to make parts
1722 * of the fs read only, e.g. subvolumes in Btrfs.
Josef Bacikc3b2da32012-03-26 09:59:21 -04001723 */
Deepa Dinamanic2050a42016-09-14 07:48:06 -07001724 now = current_time(inode);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001725 update_time(inode, &now, S_ATIME);
Jan Kara5d37e9e2012-06-12 16:20:36 +02001726 __mnt_drop_write(mnt);
1727skip_update:
1728 sb_end_write(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001730EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731
Cong Wang3ed37642012-05-15 14:57:33 +08001732/*
1733 * The logic we want is
1734 *
1735 * if suid or (sgid and xgrp)
1736 * remove privs
1737 */
1738int should_remove_suid(struct dentry *dentry)
1739{
David Howellsdf2b1af2015-03-17 22:26:15 +00001740 umode_t mode = d_inode(dentry)->i_mode;
Cong Wang3ed37642012-05-15 14:57:33 +08001741 int kill = 0;
1742
1743 /* suid always must be killed */
1744 if (unlikely(mode & S_ISUID))
1745 kill = ATTR_KILL_SUID;
1746
1747 /*
1748 * sgid without any exec bits is just a mandatory locking mark; leave
1749 * it alone. If some exec bits are set, it's a real sgid; kill it.
1750 */
1751 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1752 kill |= ATTR_KILL_SGID;
1753
1754 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1755 return kill;
1756
1757 return 0;
1758}
1759EXPORT_SYMBOL(should_remove_suid);
1760
Jan Karadbfae0c2015-05-21 16:05:54 +02001761/*
1762 * Return mask of changes for notify_change() that need to be done as a
1763 * response to write or truncate. Return 0 if nothing has to be changed.
1764 * Negative value on error (change should be denied).
1765 */
Jan Kara45f147a2015-05-21 16:05:55 +02001766int dentry_needs_remove_privs(struct dentry *dentry)
Jan Karadbfae0c2015-05-21 16:05:54 +02001767{
Jan Karadbfae0c2015-05-21 16:05:54 +02001768 struct inode *inode = d_inode(dentry);
1769 int mask = 0;
1770 int ret;
1771
1772 if (IS_NOSEC(inode))
1773 return 0;
1774
1775 mask = should_remove_suid(dentry);
1776 ret = security_inode_need_killpriv(dentry);
1777 if (ret < 0)
1778 return ret;
1779 if (ret)
1780 mask |= ATTR_KILL_PRIV;
1781 return mask;
1782}
Jan Karadbfae0c2015-05-21 16:05:54 +02001783
Daniel Rosenberg2757e9b2016-10-26 16:33:11 -07001784static int __remove_privs(struct vfsmount *mnt, struct dentry *dentry, int kill)
Cong Wang3ed37642012-05-15 14:57:33 +08001785{
1786 struct iattr newattrs;
1787
1788 newattrs.ia_valid = ATTR_FORCE | kill;
J. Bruce Fields27ac0ff2011-09-20 17:19:26 -04001789 /*
1790 * Note we call this on write, so notify_change will not
1791 * encounter any conflicting delegations:
1792 */
Daniel Rosenberg2757e9b2016-10-26 16:33:11 -07001793 return notify_change2(mnt, dentry, &newattrs, NULL);
Cong Wang3ed37642012-05-15 14:57:33 +08001794}
1795
Jan Kara5fa8e0a2015-05-21 16:05:53 +02001796/*
1797 * Remove special file priviledges (suid, capabilities) when file is written
1798 * to or truncated.
1799 */
1800int file_remove_privs(struct file *file)
Cong Wang3ed37642012-05-15 14:57:33 +08001801{
Miklos Szeredic1892c32016-08-03 13:44:27 +02001802 struct dentry *dentry = file_dentry(file);
1803 struct inode *inode = file_inode(file);
Jan Karadbfae0c2015-05-21 16:05:54 +02001804 int kill;
Cong Wang3ed37642012-05-15 14:57:33 +08001805 int error = 0;
1806
Alexander Lochmann77beb272018-12-14 11:55:52 +01001807 /*
1808 * Fast path for nothing security related.
1809 * As well for non-regular files, e.g. blkdev inodes.
1810 * For example, blkdev_write_iter() might get here
1811 * trying to remove privs which it is not allowed to.
1812 */
1813 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
Cong Wang3ed37642012-05-15 14:57:33 +08001814 return 0;
1815
Miklos Szeredic1892c32016-08-03 13:44:27 +02001816 kill = dentry_needs_remove_privs(dentry);
Jan Karadbfae0c2015-05-21 16:05:54 +02001817 if (kill < 0)
1818 return kill;
1819 if (kill)
Daniel Rosenberg2757e9b2016-10-26 16:33:11 -07001820 error = __remove_privs(file->f_path.mnt, dentry, kill);
Jan Kara2426f392015-05-21 16:05:52 +02001821 if (!error)
1822 inode_has_no_xattr(inode);
Cong Wang3ed37642012-05-15 14:57:33 +08001823
1824 return error;
1825}
Jan Kara5fa8e0a2015-05-21 16:05:53 +02001826EXPORT_SYMBOL(file_remove_privs);
Cong Wang3ed37642012-05-15 14:57:33 +08001827
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001829 * file_update_time - update mtime and ctime time
1830 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001832 * Update the mtime and ctime members of an inode and mark the inode
1833 * for writeback. Note that this function is meant exclusively for
1834 * usage in the file write path of filesystems, and filesystems may
1835 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001836 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Josef Bacikc3b2da32012-03-26 09:59:21 -04001837 * timestamps are handled by the server. This can return an error for
1838 * file systems who need to allocate space in order to update an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 */
1840
Josef Bacikc3b2da32012-03-26 09:59:21 -04001841int file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842{
Al Viro496ad9a2013-01-23 17:07:38 -05001843 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 struct timespec now;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001845 int sync_it = 0;
1846 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
Andi Kleence06e0b2009-09-18 13:05:48 -07001848 /* First try to exhaust all avenues to not sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 if (IS_NOCMTIME(inode))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001850 return 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001851
Deepa Dinamanic2050a42016-09-14 07:48:06 -07001852 now = current_time(inode);
Andi Kleence06e0b2009-09-18 13:05:48 -07001853 if (!timespec_equal(&inode->i_mtime, &now))
1854 sync_it = S_MTIME;
1855
1856 if (!timespec_equal(&inode->i_ctime, &now))
1857 sync_it |= S_CTIME;
1858
1859 if (IS_I_VERSION(inode))
1860 sync_it |= S_VERSION;
1861
1862 if (!sync_it)
Josef Bacikc3b2da32012-03-26 09:59:21 -04001863 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
Andi Kleence06e0b2009-09-18 13:05:48 -07001865 /* Finally allowed to write? Takes lock. */
Jan Karaeb04c282012-06-12 16:20:35 +02001866 if (__mnt_want_write_file(file))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001867 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868
Josef Bacikc3b2da32012-03-26 09:59:21 -04001869 ret = update_time(inode, &now, sync_it);
Jan Karaeb04c282012-06-12 16:20:35 +02001870 __mnt_drop_write_file(file);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001871
1872 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001874EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
1876int inode_needs_sync(struct inode *inode)
1877{
1878 if (IS_SYNC(inode))
1879 return 1;
1880 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1881 return 1;
1882 return 0;
1883}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884EXPORT_SYMBOL(inode_needs_sync);
1885
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001887 * If we try to find an inode in the inode hash while it is being
1888 * deleted, we have to wait until the filesystem completes its
1889 * deletion before reporting that it isn't found. This function waits
1890 * until the deletion _might_ have completed. Callers are responsible
1891 * to recheck inode state.
1892 *
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001893 * It doesn't matter if I_NEW is not set initially, a call to
Dave Chinner250df6e2011-03-22 22:23:36 +11001894 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1895 * will DTRT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 */
1897static void __wait_on_freeing_inode(struct inode *inode)
1898{
1899 wait_queue_head_t *wq;
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001900 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1901 wq = bit_waitqueue(&inode->i_state, __I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +11001903 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001904 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 schedule();
1906 finish_wait(wq, &wait.wait);
Dave Chinner67a23c42011-03-22 22:23:42 +11001907 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908}
1909
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910static __initdata unsigned long ihash_entries;
1911static int __init set_ihash_entries(char *str)
1912{
1913 if (!str)
1914 return 0;
1915 ihash_entries = simple_strtoul(str, &str, 0);
1916 return 1;
1917}
1918__setup("ihash_entries=", set_ihash_entries);
1919
1920/*
1921 * Initialize the waitqueues and inode hash table.
1922 */
1923void __init inode_init_early(void)
1924{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001925 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926
1927 /* If hashes are distributed across NUMA nodes, defer
1928 * hash allocation until vmalloc space is available.
1929 */
1930 if (hashdist)
1931 return;
1932
1933 inode_hashtable =
1934 alloc_large_system_hash("Inode-cache",
1935 sizeof(struct hlist_head),
1936 ihash_entries,
1937 14,
1938 HASH_EARLY,
1939 &i_hash_shift,
1940 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001941 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 0);
1943
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001944 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1946}
1947
Denis Cheng74bf17c2007-10-16 23:26:30 -07001948void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001950 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951
1952 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001953 inode_cachep = kmem_cache_create("inode_cache",
1954 sizeof(struct inode),
1955 0,
1956 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
Vladimir Davydov5d097052016-01-14 15:18:21 -08001957 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
Paul Mundt20c2df82007-07-20 10:11:58 +09001958 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
1960 /* Hash may have been set up in inode_init_early */
1961 if (!hashdist)
1962 return;
1963
1964 inode_hashtable =
1965 alloc_large_system_hash("Inode-cache",
1966 sizeof(struct hlist_head),
1967 ihash_entries,
1968 14,
1969 0,
1970 &i_hash_shift,
1971 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001972 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 0);
1974
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001975 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1977}
1978
1979void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1980{
1981 inode->i_mode = mode;
1982 if (S_ISCHR(mode)) {
1983 inode->i_fop = &def_chr_fops;
1984 inode->i_rdev = rdev;
1985 } else if (S_ISBLK(mode)) {
1986 inode->i_fop = &def_blk_fops;
1987 inode->i_rdev = rdev;
1988 } else if (S_ISFIFO(mode))
Al Viro599a0ac2013-03-12 09:58:10 -04001989 inode->i_fop = &pipefifo_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 else if (S_ISSOCK(mode))
Al Virobd9b51e2014-11-18 23:38:21 -05001991 ; /* leave it no_open_fops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 else
Manish Katiyaraf0d9ae2009-09-18 13:05:43 -07001993 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1994 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1995 inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996}
1997EXPORT_SYMBOL(init_special_inode);
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001998
1999/**
Ben Hutchingseaae668d2011-02-15 12:48:09 +00002000 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03002001 * @inode: New inode
2002 * @dir: Directory inode
2003 * @mode: mode of the new inode
2004 */
2005void inode_init_owner(struct inode *inode, const struct inode *dir,
Al Viro62bb1092011-07-24 23:20:18 -04002006 umode_t mode)
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03002007{
2008 inode->i_uid = current_fsuid();
2009 if (dir && dir->i_mode & S_ISGID) {
2010 inode->i_gid = dir->i_gid;
Linus Torvaldsd2c7c522018-07-03 17:10:19 -07002011
2012 /* Directories are special, and always inherit S_ISGID */
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03002013 if (S_ISDIR(mode))
2014 mode |= S_ISGID;
Linus Torvaldsd2c7c522018-07-03 17:10:19 -07002015 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2016 !in_group_p(inode->i_gid) &&
2017 !capable_wrt_inode_uidgid(dir, CAP_FSETID))
2018 mode &= ~S_ISGID;
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03002019 } else
2020 inode->i_gid = current_fsgid();
2021 inode->i_mode = mode;
2022}
2023EXPORT_SYMBOL(inode_init_owner);
Serge E. Hallyne795b712011-03-23 16:43:25 -07002024
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002025/**
2026 * inode_owner_or_capable - check current task permissions to inode
2027 * @inode: inode being checked
2028 *
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002029 * Return true if current either has CAP_FOWNER in a namespace with the
2030 * inode owner uid mapped, or owns the file.
Serge E. Hallyne795b712011-03-23 16:43:25 -07002031 */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002032bool inode_owner_or_capable(const struct inode *inode)
Serge E. Hallyne795b712011-03-23 16:43:25 -07002033{
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002034 struct user_namespace *ns;
2035
Eric W. Biederman92361632012-02-08 07:07:50 -08002036 if (uid_eq(current_fsuid(), inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07002037 return true;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07002038
2039 ns = current_user_ns();
2040 if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07002041 return true;
2042 return false;
2043}
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002044EXPORT_SYMBOL(inode_owner_or_capable);
Trond Myklebust1d59d612012-05-31 12:22:33 -04002045
2046/*
2047 * Direct i/o helper functions
2048 */
2049static void __inode_dio_wait(struct inode *inode)
2050{
2051 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2052 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2053
2054 do {
2055 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
2056 if (atomic_read(&inode->i_dio_count))
2057 schedule();
2058 } while (atomic_read(&inode->i_dio_count));
2059 finish_wait(wq, &q.wait);
2060}
2061
2062/**
2063 * inode_dio_wait - wait for outstanding DIO requests to finish
2064 * @inode: inode to wait for
2065 *
2066 * Waits for all pending direct I/O requests to finish so that we can
2067 * proceed with a truncate or equivalent operation.
2068 *
2069 * Must be called under a lock that serializes taking new references
2070 * to i_dio_count, usually by inode->i_mutex.
2071 */
2072void inode_dio_wait(struct inode *inode)
2073{
2074 if (atomic_read(&inode->i_dio_count))
2075 __inode_dio_wait(inode);
2076}
2077EXPORT_SYMBOL(inode_dio_wait);
2078
2079/*
Theodore Ts'o5f16f322014-03-24 14:43:12 -04002080 * inode_set_flags - atomically set some inode flags
2081 *
2082 * Note: the caller should be holding i_mutex, or else be sure that
2083 * they have exclusive access to the inode structure (i.e., while the
2084 * inode is being instantiated). The reason for the cmpxchg() loop
2085 * --- which wouldn't be necessary if all code paths which modify
2086 * i_flags actually followed this rule, is that there is at least one
Jan Kara5fa8e0a2015-05-21 16:05:53 +02002087 * code path which doesn't today so we use cmpxchg() out of an abundance
2088 * of caution.
Theodore Ts'o5f16f322014-03-24 14:43:12 -04002089 *
2090 * In the long run, i_mutex is overkill, and we should probably look
2091 * at using the i_lock spinlock to protect i_flags, and then make sure
2092 * it is so documented in include/linux/fs.h and that all code follows
2093 * the locking convention!!
2094 */
2095void inode_set_flags(struct inode *inode, unsigned int flags,
2096 unsigned int mask)
2097{
2098 unsigned int old_flags, new_flags;
2099
2100 WARN_ON_ONCE(flags & ~mask);
2101 do {
2102 old_flags = ACCESS_ONCE(inode->i_flags);
2103 new_flags = (old_flags & ~mask) | flags;
2104 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2105 new_flags) != old_flags));
2106}
2107EXPORT_SYMBOL(inode_set_flags);
Al Viro21fc61c2015-11-17 01:07:57 -05002108
2109void inode_nohighmem(struct inode *inode)
2110{
2111 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2112}
2113EXPORT_SYMBOL(inode_nohighmem);
Deepa Dinamani3cd88662016-09-14 07:48:02 -07002114
2115/**
2116 * current_time - Return FS time
2117 * @inode: inode.
2118 *
2119 * Return the current time truncated to the time granularity supported by
2120 * the fs.
2121 *
2122 * Note that inode and inode->sb cannot be NULL.
2123 * Otherwise, the function warns and returns time without truncation.
2124 */
2125struct timespec current_time(struct inode *inode)
2126{
2127 struct timespec now = current_kernel_time();
2128
2129 if (unlikely(!inode->i_sb)) {
2130 WARN(1, "current_time() called with uninitialized super_block in the inode");
2131 return now;
2132 }
2133
2134 return timespec_trunc(now, inode->i_sb->s_time_gran);
2135}
2136EXPORT_SYMBOL(current_time);
Eric Biggers8040d8d2019-07-01 13:26:28 -07002137
2138/*
2139 * Generic function to check FS_IOC_SETFLAGS values and reject any invalid
2140 * configurations.
2141 *
2142 * Note: the caller should be holding i_mutex, or else be sure that they have
2143 * exclusive access to the inode structure.
2144 */
2145int vfs_ioc_setflags_prepare(struct inode *inode, unsigned int oldflags,
2146 unsigned int flags)
2147{
2148 /*
2149 * The IMMUTABLE and APPEND_ONLY flags can only be changed by
2150 * the relevant capability.
2151 *
2152 * This test looks nicer. Thanks to Pauline Middelink
2153 */
2154 if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL) &&
2155 !capable(CAP_LINUX_IMMUTABLE))
2156 return -EPERM;
2157
2158 return 0;
2159}
2160EXPORT_SYMBOL(vfs_ioc_setflags_prepare);