blob: 4ccbc21b30ce6fa791c42df70986d41e490cef10 [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;
Eric W. Biederman92361632012-02-08 07:07:50 -0800143 i_uid_write(inode, 0);
144 i_gid_write(inode, 0);
David Chinner2cb15992008-10-30 17:32:23 +1100145 atomic_set(&inode->i_writecount, 0);
146 inode->i_size = 0;
147 inode->i_blocks = 0;
148 inode->i_bytes = 0;
149 inode->i_generation = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100150 inode->i_pipe = NULL;
151 inode->i_bdev = NULL;
152 inode->i_cdev = NULL;
Al Viro61ba64f2015-05-02 09:54:06 -0400153 inode->i_link = NULL;
Al Viro84e710d2016-04-15 00:58:55 -0400154 inode->i_dir_seq = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100155 inode->i_rdev = 0;
156 inode->dirtied_when = 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500157
Tahsin Erdogan3d65ae42016-02-16 13:34:39 -0800158#ifdef CONFIG_CGROUP_WRITEBACK
159 inode->i_wb_frn_winner = 0;
160 inode->i_wb_frn_avg_time = 0;
161 inode->i_wb_frn_history = 0;
162#endif
163
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500164 if (security_inode_alloc(inode))
Christoph Hellwig54e34622009-08-07 14:38:25 -0300165 goto out;
David Chinner2cb15992008-10-30 17:32:23 +1100166 spin_lock_init(&inode->i_lock);
167 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
168
Al Viro9902af72016-04-15 15:08:36 -0400169 init_rwsem(&inode->i_rwsem);
170 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
David Chinner2cb15992008-10-30 17:32:23 +1100171
Christoph Hellwigbd5fe6c2011-06-24 14:29:43 -0400172 atomic_set(&inode->i_dio_count, 0);
David Chinner2cb15992008-10-30 17:32:23 +1100173
174 mapping->a_ops = &empty_aops;
175 mapping->host = inode;
176 mapping->flags = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700177 atomic_set(&mapping->i_mmap_writable, 0);
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800178 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800179 mapping->private_data = NULL;
David Chinner2cb15992008-10-30 17:32:23 +1100180 mapping->writeback_index = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100181 inode->i_private = NULL;
182 inode->i_mapping = mapping;
Al Virob3d9b7a2012-06-09 13:51:19 -0400183 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
Al Virof19d4a82009-06-08 19:50:45 -0400184#ifdef CONFIG_FS_POSIX_ACL
185 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
186#endif
David Chinner2cb15992008-10-30 17:32:23 +1100187
Eric Paris3be25f42009-05-21 17:01:26 -0400188#ifdef CONFIG_FSNOTIFY
189 inode->i_fsnotify_mask = 0;
190#endif
Jeff Layton4a075e32015-01-16 15:05:54 -0500191 inode->i_flctx = NULL;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100192 this_cpu_inc(nr_inodes);
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400193
Christoph Hellwig54e34622009-08-07 14:38:25 -0300194 return 0;
Christoph Hellwig54e34622009-08-07 14:38:25 -0300195out:
196 return -ENOMEM;
David Chinner2cb15992008-10-30 17:32:23 +1100197}
198EXPORT_SYMBOL(inode_init_always);
199
200static struct inode *alloc_inode(struct super_block *sb)
201{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 struct inode *inode;
203
204 if (sb->s_op->alloc_inode)
205 inode = sb->s_op->alloc_inode(sb);
206 else
David Chinner2cb15992008-10-30 17:32:23 +1100207 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
Christoph Hellwig54e34622009-08-07 14:38:25 -0300209 if (!inode)
210 return NULL;
211
212 if (unlikely(inode_init_always(sb, inode))) {
213 if (inode->i_sb->s_op->destroy_inode)
214 inode->i_sb->s_op->destroy_inode(inode);
215 else
216 kmem_cache_free(inode_cachep, inode);
217 return NULL;
218 }
219
220 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221}
222
Nick Pigginff0c7d12011-01-07 17:49:50 +1100223void free_inode_nonrcu(struct inode *inode)
224{
225 kmem_cache_free(inode_cachep, inode);
226}
227EXPORT_SYMBOL(free_inode_nonrcu);
228
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300229void __destroy_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200231 BUG_ON(inode_has_buffers(inode));
Tejun Heo52ebea72015-05-22 17:13:37 -0400232 inode_detach_wb(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 security_inode_free(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400234 fsnotify_inode_delete(inode);
Jeff Laytonf27a0fe2016-01-07 15:08:51 -0500235 locks_free_lock_context(inode);
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100236 if (!inode->i_nlink) {
237 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
238 atomic_long_dec(&inode->i_sb->s_remove_count);
239 }
240
Al Virof19d4a82009-06-08 19:50:45 -0400241#ifdef CONFIG_FS_POSIX_ACL
Andreas Gruenbacherb8a7a3a2016-03-24 14:38:37 +0100242 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
Al Virof19d4a82009-06-08 19:50:45 -0400243 posix_acl_release(inode->i_acl);
Andreas Gruenbacherb8a7a3a2016-03-24 14:38:37 +0100244 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
Al Virof19d4a82009-06-08 19:50:45 -0400245 posix_acl_release(inode->i_default_acl);
246#endif
Nick Piggin3e880fb2011-01-07 17:49:19 +1100247 this_cpu_dec(nr_inodes);
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300248}
249EXPORT_SYMBOL(__destroy_inode);
250
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100251static void i_callback(struct rcu_head *head)
252{
253 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100254 kmem_cache_free(inode_cachep, inode);
255}
256
Christoph Hellwig56b0dac2010-10-06 10:48:55 +0200257static void destroy_inode(struct inode *inode)
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300258{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100259 BUG_ON(!list_empty(&inode->i_lru));
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300260 __destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 if (inode->i_sb->s_op->destroy_inode)
262 inode->i_sb->s_op->destroy_inode(inode);
263 else
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100264 call_rcu(&inode->i_rcu, i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100267/**
268 * drop_nlink - directly drop an inode's link count
269 * @inode: inode
270 *
271 * This is a low-level filesystem helper to replace any
272 * direct filesystem manipulation of i_nlink. In cases
273 * where we are attempting to track writes to the
274 * filesystem, a decrement to zero means an imminent
275 * write when the file is truncated and actually unlinked
276 * on the filesystem.
277 */
278void drop_nlink(struct inode *inode)
279{
280 WARN_ON(inode->i_nlink == 0);
281 inode->__i_nlink--;
282 if (!inode->i_nlink)
283 atomic_long_inc(&inode->i_sb->s_remove_count);
284}
285EXPORT_SYMBOL(drop_nlink);
286
287/**
288 * clear_nlink - directly zero an inode's link count
289 * @inode: inode
290 *
291 * This is a low-level filesystem helper to replace any
292 * direct filesystem manipulation of i_nlink. See
293 * drop_nlink() for why we care about i_nlink hitting zero.
294 */
295void clear_nlink(struct inode *inode)
296{
297 if (inode->i_nlink) {
298 inode->__i_nlink = 0;
299 atomic_long_inc(&inode->i_sb->s_remove_count);
300 }
301}
302EXPORT_SYMBOL(clear_nlink);
303
304/**
305 * set_nlink - directly set an inode's link count
306 * @inode: inode
307 * @nlink: new nlink (should be non-zero)
308 *
309 * This is a low-level filesystem helper to replace any
310 * direct filesystem manipulation of i_nlink.
311 */
312void set_nlink(struct inode *inode, unsigned int nlink)
313{
314 if (!nlink) {
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100315 clear_nlink(inode);
316 } else {
317 /* Yes, some filesystems do change nlink from zero to one */
318 if (inode->i_nlink == 0)
319 atomic_long_dec(&inode->i_sb->s_remove_count);
320
321 inode->__i_nlink = nlink;
322 }
323}
324EXPORT_SYMBOL(set_nlink);
325
326/**
327 * inc_nlink - directly increment an inode's link count
328 * @inode: inode
329 *
330 * This is a low-level filesystem helper to replace any
331 * direct filesystem manipulation of i_nlink. Currently,
332 * it is only here for parity with dec_nlink().
333 */
334void inc_nlink(struct inode *inode)
335{
Al Virof4e0c302013-06-11 08:34:36 +0400336 if (unlikely(inode->i_nlink == 0)) {
337 WARN_ON(!(inode->i_state & I_LINKABLE));
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100338 atomic_long_dec(&inode->i_sb->s_remove_count);
Al Virof4e0c302013-06-11 08:34:36 +0400339 }
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100340
341 inode->__i_nlink++;
342}
343EXPORT_SYMBOL(inc_nlink);
344
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100345void address_space_init_once(struct address_space *mapping)
346{
347 memset(mapping, 0, sizeof(*mapping));
348 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
349 spin_lock_init(&mapping->tree_lock);
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800350 init_rwsem(&mapping->i_mmap_rwsem);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100351 INIT_LIST_HEAD(&mapping->private_list);
352 spin_lock_init(&mapping->private_lock);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700353 mapping->i_mmap = RB_ROOT;
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100354}
355EXPORT_SYMBOL(address_space_init_once);
356
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357/*
358 * These are initializations that only need to be done
359 * once, because the fields are idempotent across use
360 * of the inode, so let the slab aware of that.
361 */
362void inode_init_once(struct inode *inode)
363{
364 memset(inode, 0, sizeof(*inode));
365 INIT_HLIST_NODE(&inode->i_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 INIT_LIST_HEAD(&inode->i_devices);
Dave Chinnerc7f54082015-03-04 14:07:22 -0500367 INIT_LIST_HEAD(&inode->i_io_list);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100368 INIT_LIST_HEAD(&inode->i_lru);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100369 address_space_init_once(&inode->i_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 i_size_ordered_init(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400371#ifdef CONFIG_FSNOTIFY
Eric Parise61ce862009-12-17 21:24:24 -0500372 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
Eric Paris3be25f42009-05-21 17:01:26 -0400373#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375EXPORT_SYMBOL(inode_init_once);
376
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700377static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530379 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
Christoph Lametera35afb82007-05-16 22:10:57 -0700381 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382}
383
384/*
Dave Chinner250df6e2011-03-22 22:23:36 +1100385 * inode->i_lock must be held
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530387void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388{
Nick Piggin9e38d862010-10-23 06:55:17 -0400389 atomic_inc(&inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390}
391
Al Viro7de9c6ee2010-10-23 11:11:40 -0400392/*
393 * get additional reference to inode; caller must already hold one.
394 */
395void ihold(struct inode *inode)
396{
397 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
398}
399EXPORT_SYMBOL(ihold);
400
Nick Piggin9e38d862010-10-23 06:55:17 -0400401static void inode_lru_list_add(struct inode *inode)
402{
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000403 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000404 this_cpu_inc(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400405}
406
Jan Kara4eff96d2012-11-26 16:29:51 -0800407/*
408 * Add inode to LRU if needed (inode is unused and clean).
409 *
410 * Needs inode->i_lock held.
411 */
412void inode_add_lru(struct inode *inode)
413{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500414 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
415 I_FREEING | I_WILL_FREE)) &&
Jan Kara4eff96d2012-11-26 16:29:51 -0800416 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
417 inode_lru_list_add(inode);
418}
419
420
Nick Piggin9e38d862010-10-23 06:55:17 -0400421static void inode_lru_list_del(struct inode *inode)
422{
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000423
424 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000425 this_cpu_dec(nr_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426}
427
Christoph Hellwig646ec462010-10-23 07:15:32 -0400428/**
429 * inode_sb_list_add - add inode to the superblock list of inodes
430 * @inode: inode to add
431 */
432void inode_sb_list_add(struct inode *inode)
433{
Dave Chinner74278da2015-03-04 12:37:22 -0500434 spin_lock(&inode->i_sb->s_inode_list_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +1100435 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
Dave Chinner74278da2015-03-04 12:37:22 -0500436 spin_unlock(&inode->i_sb->s_inode_list_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400437}
438EXPORT_SYMBOL_GPL(inode_sb_list_add);
439
Dave Chinner55fa6092011-03-22 22:23:40 +1100440static inline void inode_sb_list_del(struct inode *inode)
Christoph Hellwig646ec462010-10-23 07:15:32 -0400441{
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200442 if (!list_empty(&inode->i_sb_list)) {
Dave Chinner74278da2015-03-04 12:37:22 -0500443 spin_lock(&inode->i_sb->s_inode_list_lock);
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200444 list_del_init(&inode->i_sb_list);
Dave Chinner74278da2015-03-04 12:37:22 -0500445 spin_unlock(&inode->i_sb->s_inode_list_lock);
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200446 }
Christoph Hellwig646ec462010-10-23 07:15:32 -0400447}
448
Dave Chinner4c51acb2010-10-23 06:58:09 -0400449static unsigned long hash(struct super_block *sb, unsigned long hashval)
450{
451 unsigned long tmp;
452
453 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
454 L1_CACHE_BYTES;
Christoph Hellwig4b4563d2011-05-27 09:28:01 -0400455 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
456 return tmp & i_hash_mask;
Dave Chinner4c51acb2010-10-23 06:58:09 -0400457}
458
459/**
460 * __insert_inode_hash - hash an inode
461 * @inode: unhashed inode
462 * @hashval: unsigned long value used to locate this object in the
463 * inode_hashtable.
464 *
465 * Add an inode to the inode hash for this superblock.
466 */
467void __insert_inode_hash(struct inode *inode, unsigned long hashval)
468{
Christoph Hellwig646ec462010-10-23 07:15:32 -0400469 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
470
Dave Chinner67a23c42011-03-22 22:23:42 +1100471 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100472 spin_lock(&inode->i_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400473 hlist_add_head(&inode->i_hash, b);
Dave Chinner250df6e2011-03-22 22:23:36 +1100474 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100475 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400476}
477EXPORT_SYMBOL(__insert_inode_hash);
478
479/**
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200480 * __remove_inode_hash - remove an inode from the hash
Dave Chinner4c51acb2010-10-23 06:58:09 -0400481 * @inode: inode to unhash
482 *
483 * Remove an inode from the superblock.
484 */
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200485void __remove_inode_hash(struct inode *inode)
Dave Chinner4c51acb2010-10-23 06:58:09 -0400486{
Dave Chinner67a23c42011-03-22 22:23:42 +1100487 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100488 spin_lock(&inode->i_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400489 hlist_del_init(&inode->i_hash);
Dave Chinner250df6e2011-03-22 22:23:36 +1100490 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100491 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400492}
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200493EXPORT_SYMBOL(__remove_inode_hash);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400494
Jan Karadbd57682012-05-03 14:48:02 +0200495void clear_inode(struct inode *inode)
Al Virob0683aa2010-06-04 20:55:25 -0400496{
497 might_sleep();
Jan Kara08142572011-06-27 16:18:10 -0700498 /*
499 * We have to cycle tree_lock here because reclaim can be still in the
500 * process of removing the last page (in __delete_from_page_cache())
501 * and we must not free mapping under it.
502 */
503 spin_lock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400504 BUG_ON(inode->i_data.nrpages);
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800505 BUG_ON(inode->i_data.nrexceptional);
Jan Kara08142572011-06-27 16:18:10 -0700506 spin_unlock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400507 BUG_ON(!list_empty(&inode->i_data.private_list));
508 BUG_ON(!(inode->i_state & I_FREEING));
509 BUG_ON(inode->i_state & I_CLEAR);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100510 /* don't need i_lock here, no concurrent mods to i_state */
Al Virob0683aa2010-06-04 20:55:25 -0400511 inode->i_state = I_FREEING | I_CLEAR;
512}
Jan Karadbd57682012-05-03 14:48:02 +0200513EXPORT_SYMBOL(clear_inode);
Al Virob0683aa2010-06-04 20:55:25 -0400514
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100515/*
516 * Free the inode passed in, removing it from the lists it is still connected
517 * to. We remove any pages still attached to the inode and wait for any IO that
518 * is still in progress before finally destroying the inode.
519 *
520 * An inode must already be marked I_FREEING so that we avoid the inode being
521 * moved back onto lists if we race with other code that manipulates the lists
522 * (e.g. writeback_single_inode). The caller is responsible for setting this.
523 *
524 * An inode must already be removed from the LRU list before being evicted from
525 * the cache. This should occur atomically with setting the I_FREEING state
526 * flag, so no inodes here should ever be on the LRU when being evicted.
527 */
Al Viro644da592010-06-07 13:21:05 -0400528static void evict(struct inode *inode)
Al Virob4272d42010-06-04 19:33:20 -0400529{
530 const struct super_operations *op = inode->i_sb->s_op;
531
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100532 BUG_ON(!(inode->i_state & I_FREEING));
533 BUG_ON(!list_empty(&inode->i_lru));
534
Dave Chinnerc7f54082015-03-04 14:07:22 -0500535 if (!list_empty(&inode->i_io_list))
536 inode_io_list_del(inode);
Eric Dumazetb12362b2011-07-28 06:11:47 +0200537
Dave Chinner55fa6092011-03-22 22:23:40 +1100538 inode_sb_list_del(inode);
539
Jan Kara169ebd92012-05-03 14:48:03 +0200540 /*
541 * Wait for flusher thread to be done with the inode so that filesystem
542 * does not start destroying it while writeback is still running. Since
543 * the inode has I_FREEING set, flusher thread won't start new work on
544 * the inode. We just have to wait for running writeback to finish.
545 */
546 inode_wait_for_writeback(inode);
Jan Kara7994e6f2012-05-03 14:48:01 +0200547
Al Virobe7ce412010-06-04 19:40:39 -0400548 if (op->evict_inode) {
549 op->evict_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400550 } else {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700551 truncate_inode_pages_final(&inode->i_data);
Jan Karadbd57682012-05-03 14:48:02 +0200552 clear_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400553 }
Al Viro661074e2010-06-04 20:19:55 -0400554 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
555 bd_forget(inode);
556 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
557 cd_forget(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100558
559 remove_inode_hash(inode);
560
561 spin_lock(&inode->i_lock);
562 wake_up_bit(&inode->i_state, __I_NEW);
563 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
564 spin_unlock(&inode->i_lock);
565
566 destroy_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400567}
568
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569/*
570 * dispose_list - dispose of the contents of a local list
571 * @head: the head of the list to free
572 *
573 * Dispose-list gets a local list with local inodes in it, so it doesn't
574 * need to worry about list corruption and SMP locks.
575 */
576static void dispose_list(struct list_head *head)
577{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 while (!list_empty(head)) {
579 struct inode *inode;
580
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100581 inode = list_first_entry(head, struct inode, i_lru);
582 list_del_init(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
Al Viro644da592010-06-07 13:21:05 -0400584 evict(inode);
Josef Bacikac05fbb2015-03-04 16:52:52 -0500585 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587}
588
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589/**
Al Viro63997e92010-10-25 20:49:35 -0400590 * evict_inodes - evict all evictable inodes for a superblock
591 * @sb: superblock to operate on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 *
Al Viro63997e92010-10-25 20:49:35 -0400593 * Make sure that no inodes with zero refcount are retained. This is
594 * called by superblock shutdown after having MS_ACTIVE flag removed,
595 * so any inode reaching zero refcount during or after that call will
596 * be immediately evicted.
597 */
598void evict_inodes(struct super_block *sb)
599{
600 struct inode *inode, *next;
601 LIST_HEAD(dispose);
602
Josef Bacikac05fbb2015-03-04 16:52:52 -0500603again:
Dave Chinner74278da2015-03-04 12:37:22 -0500604 spin_lock(&sb->s_inode_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400605 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
606 if (atomic_read(&inode->i_count))
607 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100608
609 spin_lock(&inode->i_lock);
610 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
611 spin_unlock(&inode->i_lock);
Al Viro63997e92010-10-25 20:49:35 -0400612 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100613 }
Al Viro63997e92010-10-25 20:49:35 -0400614
615 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100616 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100617 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100618 list_add(&inode->i_lru, &dispose);
Josef Bacikac05fbb2015-03-04 16:52:52 -0500619
620 /*
621 * We can have a ton of inodes to evict at unmount time given
622 * enough memory, check to see if we need to go to sleep for a
623 * bit so we don't livelock.
624 */
625 if (need_resched()) {
626 spin_unlock(&sb->s_inode_list_lock);
627 cond_resched();
628 dispose_list(&dispose);
629 goto again;
630 }
Al Viro63997e92010-10-25 20:49:35 -0400631 }
Dave Chinner74278da2015-03-04 12:37:22 -0500632 spin_unlock(&sb->s_inode_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400633
634 dispose_list(&dispose);
Al Viro63997e92010-10-25 20:49:35 -0400635}
636
637/**
Christoph Hellwiga0318782010-10-24 19:40:33 +0200638 * invalidate_inodes - attempt to free all inodes on a superblock
639 * @sb: superblock to operate on
NeilBrown93b270f2011-02-24 17:25:47 +1100640 * @kill_dirty: flag to guide handling of dirty inodes
Christoph Hellwiga0318782010-10-24 19:40:33 +0200641 *
642 * Attempts to free all inodes for a given superblock. If there were any
643 * busy inodes return a non-zero value, else zero.
NeilBrown93b270f2011-02-24 17:25:47 +1100644 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
645 * them as busy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 */
NeilBrown93b270f2011-02-24 17:25:47 +1100647int invalidate_inodes(struct super_block *sb, bool kill_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400649 int busy = 0;
Christoph Hellwiga0318782010-10-24 19:40:33 +0200650 struct inode *inode, *next;
651 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Dave Chinner74278da2015-03-04 12:37:22 -0500653 spin_lock(&sb->s_inode_list_lock);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200654 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100655 spin_lock(&inode->i_lock);
656 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
657 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100659 }
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500660 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100661 spin_unlock(&inode->i_lock);
NeilBrown93b270f2011-02-24 17:25:47 +1100662 busy = 1;
663 continue;
664 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200665 if (atomic_read(&inode->i_count)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100666 spin_unlock(&inode->i_lock);
Christoph Hellwig99a38912010-10-23 19:07:20 +0200667 busy = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 continue;
669 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200670
Christoph Hellwig99a38912010-10-23 19:07:20 +0200671 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100672 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100673 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100674 list_add(&inode->i_lru, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 }
Dave Chinner74278da2015-03-04 12:37:22 -0500676 spin_unlock(&sb->s_inode_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
Christoph Hellwiga0318782010-10-24 19:40:33 +0200678 dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
680 return busy;
681}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683/*
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000684 * Isolate the inode from the LRU in preparation for freeing it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 *
686 * Any inodes which are pinned purely because of attached pagecache have their
Nick Piggin9e38d862010-10-23 06:55:17 -0400687 * pagecache removed. If the inode has metadata buffers attached to
688 * mapping->private_list then try to remove them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 *
Nick Piggin9e38d862010-10-23 06:55:17 -0400690 * If the inode has the I_REFERENCED flag set, then it means that it has been
691 * used recently - the flag is set in iput_final(). When we encounter such an
692 * inode, clear the flag and move it to the back of the LRU so it gets another
693 * pass through the LRU before it gets reclaimed. This is necessary because of
694 * the fact we are doing lazy LRU updates to minimise lock contention so the
695 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
696 * with this flag set because they are the inodes that are out of order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800698static enum lru_status inode_lru_isolate(struct list_head *item,
699 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000700{
701 struct list_head *freeable = arg;
702 struct inode *inode = container_of(item, struct inode, i_lru);
703
704 /*
705 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
706 * If we fail to get the lock, just skip it.
707 */
708 if (!spin_trylock(&inode->i_lock))
709 return LRU_SKIP;
710
711 /*
712 * Referenced or dirty inodes are still in use. Give them another pass
713 * through the LRU as we canot reclaim them now.
714 */
715 if (atomic_read(&inode->i_count) ||
716 (inode->i_state & ~I_REFERENCED)) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800717 list_lru_isolate(lru, &inode->i_lru);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000718 spin_unlock(&inode->i_lock);
719 this_cpu_dec(nr_unused);
720 return LRU_REMOVED;
721 }
722
723 /* recently referenced inodes get one more pass */
724 if (inode->i_state & I_REFERENCED) {
725 inode->i_state &= ~I_REFERENCED;
726 spin_unlock(&inode->i_lock);
727 return LRU_ROTATE;
728 }
729
730 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
731 __iget(inode);
732 spin_unlock(&inode->i_lock);
733 spin_unlock(lru_lock);
734 if (remove_inode_buffers(inode)) {
735 unsigned long reap;
736 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
737 if (current_is_kswapd())
738 __count_vm_events(KSWAPD_INODESTEAL, reap);
739 else
740 __count_vm_events(PGINODESTEAL, reap);
741 if (current->reclaim_state)
742 current->reclaim_state->reclaimed_slab += reap;
743 }
744 iput(inode);
745 spin_lock(lru_lock);
746 return LRU_RETRY;
747 }
748
749 WARN_ON(inode->i_state & I_NEW);
750 inode->i_state |= I_FREEING;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800751 list_lru_isolate_move(lru, &inode->i_lru, freeable);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000752 spin_unlock(&inode->i_lock);
753
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000754 this_cpu_dec(nr_unused);
755 return LRU_REMOVED;
756}
757
758/*
759 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
760 * This is called from the superblock shrinker function with a number of inodes
761 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
762 * then are freed outside inode_lock by dispose_list().
763 */
Vladimir Davydov503c3582015-02-12 14:58:47 -0800764long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765{
766 LIST_HEAD(freeable);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000767 long freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
Vladimir Davydov503c3582015-02-12 14:58:47 -0800769 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
770 inode_lru_isolate, &freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 dispose_list(&freeable);
Dave Chinner0a234c62013-08-28 10:17:57 +1000772 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773}
774
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775static void __wait_on_freeing_inode(struct inode *inode);
776/*
777 * Called with the inode lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530779static struct inode *find_inode(struct super_block *sb,
780 struct hlist_head *head,
781 int (*test)(struct inode *, void *),
782 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530784 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
786repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800787 hlist_for_each_entry(inode, head, i_hash) {
Al Viro5a3cd99282013-11-06 09:54:52 -0500788 if (inode->i_sb != sb)
789 continue;
790 if (!test(inode, data))
791 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100792 spin_lock(&inode->i_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -0400793 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 __wait_on_freeing_inode(inode);
795 goto repeat;
796 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400797 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100798 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400799 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400801 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802}
803
804/*
805 * find_inode_fast is the fast path version of find_inode, see the comment at
806 * iget_locked for details.
807 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530808static struct inode *find_inode_fast(struct super_block *sb,
809 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530811 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
813repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800814 hlist_for_each_entry(inode, head, i_hash) {
Al Viro5a3cd99282013-11-06 09:54:52 -0500815 if (inode->i_ino != ino)
816 continue;
817 if (inode->i_sb != sb)
818 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100819 spin_lock(&inode->i_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -0400820 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 __wait_on_freeing_inode(inode);
822 goto repeat;
823 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400824 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100825 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400826 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400828 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829}
830
Eric Dumazetf991bd22010-10-23 11:18:01 -0400831/*
832 * Each cpu owns a range of LAST_INO_BATCH numbers.
833 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
834 * to renew the exhausted range.
David Chinner8290c352008-10-30 17:35:24 +1100835 *
Eric Dumazetf991bd22010-10-23 11:18:01 -0400836 * This does not significantly increase overflow rate because every CPU can
837 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
838 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
839 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
840 * overflow rate by 2x, which does not seem too significant.
841 *
842 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
843 * error if st_ino won't fit in target struct field. Use 32bit counter
844 * here to attempt to avoid that.
David Chinner8290c352008-10-30 17:35:24 +1100845 */
Eric Dumazetf991bd22010-10-23 11:18:01 -0400846#define LAST_INO_BATCH 1024
847static DEFINE_PER_CPU(unsigned int, last_ino);
David Chinner8290c352008-10-30 17:35:24 +1100848
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400849unsigned int get_next_ino(void)
Eric Dumazetf991bd22010-10-23 11:18:01 -0400850{
851 unsigned int *p = &get_cpu_var(last_ino);
852 unsigned int res = *p;
853
854#ifdef CONFIG_SMP
855 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
856 static atomic_t shared_last_ino;
857 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
858
859 res = next - LAST_INO_BATCH;
860 }
861#endif
862
Carlos Maiolino2adc3762015-06-25 12:25:58 -0300863 res++;
864 /* get_next_ino should not provide a 0 inode number */
865 if (unlikely(!res))
866 res++;
867 *p = res;
Eric Dumazetf991bd22010-10-23 11:18:01 -0400868 put_cpu_var(last_ino);
869 return res;
David Chinner8290c352008-10-30 17:35:24 +1100870}
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400871EXPORT_SYMBOL(get_next_ino);
David Chinner8290c352008-10-30 17:35:24 +1100872
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873/**
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200874 * new_inode_pseudo - obtain an inode
875 * @sb: superblock
876 *
877 * Allocates a new inode for given superblock.
878 * Inode wont be chained in superblock s_inodes list
879 * This means :
880 * - fs can't be unmount
881 * - quotas, fsnotify, writeback can't work
882 */
883struct inode *new_inode_pseudo(struct super_block *sb)
884{
885 struct inode *inode = alloc_inode(sb);
886
887 if (inode) {
888 spin_lock(&inode->i_lock);
889 inode->i_state = 0;
890 spin_unlock(&inode->i_lock);
891 INIT_LIST_HEAD(&inode->i_sb_list);
892 }
893 return inode;
894}
895
896/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 * new_inode - obtain an inode
898 * @sb: superblock
899 *
Mel Gorman769848c2007-07-17 04:03:05 -0700900 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800901 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700902 * If HIGHMEM pages are unsuitable or it is known that pages allocated
903 * for the page cache are not reclaimable or migratable,
904 * mapping_set_gfp_mask() must be called with suitable flags on the
905 * newly created inode's mapping
906 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 */
908struct inode *new_inode(struct super_block *sb)
909{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530910 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
Dave Chinner74278da2015-03-04 12:37:22 -0500912 spin_lock_prefetch(&sb->s_inode_list_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530913
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200914 inode = new_inode_pseudo(sb);
915 if (inode)
Dave Chinner55fa6092011-03-22 22:23:40 +1100916 inode_sb_list_add(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 return inode;
918}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919EXPORT_SYMBOL(new_inode);
920
Peter Zijlstra14358e62007-10-14 01:38:33 +0200921#ifdef CONFIG_DEBUG_LOCK_ALLOC
Josh Boyere096d0c2011-08-25 07:48:12 -0400922void lockdep_annotate_inode_mutex_key(struct inode *inode)
923{
Namhyung Kima3314a02010-10-11 22:38:00 +0900924 if (S_ISDIR(inode->i_mode)) {
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200925 struct file_system_type *type = inode->i_sb->s_type;
926
Jan Kara9a7aa122009-06-04 15:26:49 +0200927 /* Set new key only if filesystem hasn't already changed it */
Al Viro9902af72016-04-15 15:08:36 -0400928 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
Jan Kara9a7aa122009-06-04 15:26:49 +0200929 /*
930 * ensure nobody is actually holding i_mutex
931 */
Al Viro9902af72016-04-15 15:08:36 -0400932 // mutex_destroy(&inode->i_mutex);
933 init_rwsem(&inode->i_rwsem);
934 lockdep_set_class(&inode->i_rwsem,
Jan Kara9a7aa122009-06-04 15:26:49 +0200935 &type->i_mutex_dir_key);
936 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200937 }
Josh Boyere096d0c2011-08-25 07:48:12 -0400938}
939EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200940#endif
Josh Boyere096d0c2011-08-25 07:48:12 -0400941
942/**
943 * unlock_new_inode - clear the I_NEW state and wake up any waiters
944 * @inode: new inode to unlock
945 *
946 * Called when the inode is fully initialised to clear the new state of the
947 * inode and wake up anyone waiting for the inode to finish initialisation.
948 */
949void unlock_new_inode(struct inode *inode)
950{
951 lockdep_annotate_inode_mutex_key(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100952 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100953 WARN_ON(!(inode->i_state & I_NEW));
954 inode->i_state &= ~I_NEW;
Al Viro310fa7a2012-03-10 17:07:28 -0500955 smp_mb();
Dave Chinner250df6e2011-03-22 22:23:36 +1100956 wake_up_bit(&inode->i_state, __I_NEW);
957 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959EXPORT_SYMBOL(unlock_new_inode);
960
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400961/**
J. Bruce Fields375e2892012-04-18 15:16:33 -0400962 * lock_two_nondirectories - take two i_mutexes on non-directory objects
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200963 *
964 * Lock any non-NULL argument that is not a directory.
965 * Zero, one or two objects may be locked by this function.
966 *
J. Bruce Fields375e2892012-04-18 15:16:33 -0400967 * @inode1: first inode to lock
968 * @inode2: second inode to lock
969 */
970void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
971{
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200972 if (inode1 > inode2)
973 swap(inode1, inode2);
974
975 if (inode1 && !S_ISDIR(inode1->i_mode))
Al Viro59551022016-01-22 15:40:57 -0500976 inode_lock(inode1);
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200977 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
Al Viro59551022016-01-22 15:40:57 -0500978 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
J. Bruce Fields375e2892012-04-18 15:16:33 -0400979}
980EXPORT_SYMBOL(lock_two_nondirectories);
981
982/**
983 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
984 * @inode1: first inode to unlock
985 * @inode2: second inode to unlock
986 */
987void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
988{
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200989 if (inode1 && !S_ISDIR(inode1->i_mode))
Al Viro59551022016-01-22 15:40:57 -0500990 inode_unlock(inode1);
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200991 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
Al Viro59551022016-01-22 15:40:57 -0500992 inode_unlock(inode2);
J. Bruce Fields375e2892012-04-18 15:16:33 -0400993}
994EXPORT_SYMBOL(unlock_two_nondirectories);
995
996/**
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400997 * iget5_locked - obtain an inode from a mounted file system
998 * @sb: super block of file system
999 * @hashval: hash value (usually inode number) to get
1000 * @test: callback used for comparisons between inodes
1001 * @set: callback used to initialize a new struct inode
1002 * @data: opaque data pointer to pass to @test and @set
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001004 * Search for the inode specified by @hashval and @data in the inode cache,
1005 * and if present it is return it with an increased reference count. This is
1006 * a generalized version of iget_locked() for file systems where the inode
1007 * number is not sufficient for unique identification of an inode.
1008 *
1009 * If the inode is not in cache, allocate a new inode and return it locked,
1010 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1011 * before unlocking it via unlock_new_inode().
1012 *
1013 * Note both @test and @set are called with the inode_hash_lock held, so can't
1014 * sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001016struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1017 int (*test)(struct inode *, void *),
1018 int (*set)(struct inode *, void *), void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001020 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301021 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001023 spin_lock(&inode_hash_lock);
1024 inode = find_inode(sb, head, test, data);
1025 spin_unlock(&inode_hash_lock);
1026
1027 if (inode) {
1028 wait_on_inode(inode);
1029 return inode;
1030 }
1031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 inode = alloc_inode(sb);
1033 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301034 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Dave Chinner67a23c42011-03-22 22:23:42 +11001036 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 /* We released the lock, so.. */
1038 old = find_inode(sb, head, test, data);
1039 if (!old) {
1040 if (set(inode, data))
1041 goto set_failed;
1042
Dave Chinner250df6e2011-03-22 22:23:36 +11001043 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001044 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001045 hlist_add_head(&inode->i_hash, head);
1046 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001047 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001048 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
1050 /* Return the locked inode with I_NEW set, the
1051 * caller is responsible for filling in the contents
1052 */
1053 return inode;
1054 }
1055
1056 /*
1057 * Uhhuh, somebody else created the same inode under
1058 * us. Use the old inode instead of the one we just
1059 * allocated.
1060 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001061 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 destroy_inode(inode);
1063 inode = old;
1064 wait_on_inode(inode);
1065 }
1066 return inode;
1067
1068set_failed:
Dave Chinner67a23c42011-03-22 22:23:42 +11001069 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 destroy_inode(inode);
1071 return NULL;
1072}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001073EXPORT_SYMBOL(iget5_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001075/**
1076 * iget_locked - obtain an inode from a mounted file system
1077 * @sb: super block of file system
1078 * @ino: inode number to get
1079 *
1080 * Search for the inode specified by @ino in the inode cache and if present
1081 * return it with an increased reference count. This is for file systems
1082 * where the inode number is sufficient for unique identification of an inode.
1083 *
1084 * If the inode is not in cache, allocate a new inode and return it locked,
1085 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1086 * before unlocking it via unlock_new_inode().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001088struct inode *iget_locked(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001090 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301091 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001093 spin_lock(&inode_hash_lock);
1094 inode = find_inode_fast(sb, head, ino);
1095 spin_unlock(&inode_hash_lock);
1096 if (inode) {
1097 wait_on_inode(inode);
1098 return inode;
1099 }
1100
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 inode = alloc_inode(sb);
1102 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301103 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104
Dave Chinner67a23c42011-03-22 22:23:42 +11001105 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 /* We released the lock, so.. */
1107 old = find_inode_fast(sb, head, ino);
1108 if (!old) {
1109 inode->i_ino = ino;
Dave Chinner250df6e2011-03-22 22:23:36 +11001110 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001111 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001112 hlist_add_head(&inode->i_hash, head);
1113 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001114 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001115 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
1117 /* Return the locked inode with I_NEW set, the
1118 * caller is responsible for filling in the contents
1119 */
1120 return inode;
1121 }
1122
1123 /*
1124 * Uhhuh, somebody else created the same inode under
1125 * us. Use the old inode instead of the one we just
1126 * allocated.
1127 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001128 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 destroy_inode(inode);
1130 inode = old;
1131 wait_on_inode(inode);
1132 }
1133 return inode;
1134}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001135EXPORT_SYMBOL(iget_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001137/*
1138 * search the inode cache for a matching inode number.
1139 * If we find one, then the inode number we are trying to
1140 * allocate is not unique and so we should not use it.
1141 *
1142 * Returns 1 if the inode number is unique, 0 if it is not.
1143 */
1144static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1145{
1146 struct hlist_head *b = inode_hashtable + hash(sb, ino);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001147 struct inode *inode;
1148
Dave Chinner67a23c42011-03-22 22:23:42 +11001149 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001150 hlist_for_each_entry(inode, b, i_hash) {
Dave Chinner67a23c42011-03-22 22:23:42 +11001151 if (inode->i_ino == ino && inode->i_sb == sb) {
1152 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001153 return 0;
Dave Chinner67a23c42011-03-22 22:23:42 +11001154 }
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001155 }
Dave Chinner67a23c42011-03-22 22:23:42 +11001156 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001157
1158 return 1;
1159}
1160
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161/**
1162 * iunique - get a unique inode number
1163 * @sb: superblock
1164 * @max_reserved: highest reserved inode number
1165 *
1166 * Obtain an inode number that is unique on the system for a given
1167 * superblock. This is used by file systems that have no natural
1168 * permanent inode numbering system. An inode number is returned that
1169 * is higher than the reserved limit but unique.
1170 *
1171 * BUGS:
1172 * With a large number of inodes live on the file system this function
1173 * currently becomes quite slow.
1174 */
1175ino_t iunique(struct super_block *sb, ino_t max_reserved)
1176{
Jeff Layton866b04f2007-05-08 00:32:29 -07001177 /*
1178 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1179 * error if st_ino won't fit in target struct field. Use 32bit counter
1180 * here to attempt to avoid that.
1181 */
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001182 static DEFINE_SPINLOCK(iunique_lock);
Jeff Layton866b04f2007-05-08 00:32:29 -07001183 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001186 spin_lock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001187 do {
1188 if (counter <= max_reserved)
1189 counter = max_reserved + 1;
1190 res = counter++;
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001191 } while (!test_inode_iunique(sb, res));
1192 spin_unlock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001193
1194 return res;
1195}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196EXPORT_SYMBOL(iunique);
1197
1198struct inode *igrab(struct inode *inode)
1199{
Dave Chinner250df6e2011-03-22 22:23:36 +11001200 spin_lock(&inode->i_lock);
1201 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001203 spin_unlock(&inode->i_lock);
1204 } else {
1205 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 /*
1207 * Handle the case where s_op->clear_inode is not been
1208 * called yet, and somebody is calling igrab
1209 * while the inode is getting freed.
1210 */
1211 inode = NULL;
Dave Chinner250df6e2011-03-22 22:23:36 +11001212 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 return inode;
1214}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215EXPORT_SYMBOL(igrab);
1216
1217/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001218 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 * @sb: super block of file system to search
1220 * @hashval: hash value (usually inode number) to search for
1221 * @test: callback used for comparisons between inodes
1222 * @data: opaque data pointer to pass to @test
1223 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001224 * Search for the inode specified by @hashval and @data in the inode cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 * If the inode is in the cache, the inode is returned with an incremented
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001226 * reference count.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001227 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001228 * Note: I_NEW is not waited upon so you have to be very careful what you do
1229 * with the returned inode. You probably should be using ilookup5() instead.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001230 *
Randy Dunlapb6d0ad62011-03-26 13:27:47 -07001231 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001232 */
1233struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1234 int (*test)(struct inode *, void *), void *data)
1235{
1236 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001237 struct inode *inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001238
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001239 spin_lock(&inode_hash_lock);
1240 inode = find_inode(sb, head, test, data);
1241 spin_unlock(&inode_hash_lock);
1242
1243 return inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001244}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001245EXPORT_SYMBOL(ilookup5_nowait);
1246
1247/**
1248 * ilookup5 - search for an inode in the inode cache
1249 * @sb: super block of file system to search
1250 * @hashval: hash value (usually inode number) to search for
1251 * @test: callback used for comparisons between inodes
1252 * @data: opaque data pointer to pass to @test
1253 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001254 * Search for the inode specified by @hashval and @data in the inode cache,
1255 * and if the inode is in the cache, return the inode with an incremented
1256 * reference count. Waits on I_NEW before returning the inode.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001257 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001259 * This is a generalized version of ilookup() for file systems where the
1260 * inode number is not sufficient for unique identification of an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001262 * Note: @test is called with the inode_hash_lock held, so can't sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 */
1264struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1265 int (*test)(struct inode *, void *), void *data)
1266{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001267 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001269 if (inode)
1270 wait_on_inode(inode);
1271 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273EXPORT_SYMBOL(ilookup5);
1274
1275/**
1276 * ilookup - search for an inode in the inode cache
1277 * @sb: super block of file system to search
1278 * @ino: inode number to search for
1279 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001280 * Search for the inode @ino in the inode cache, and if the inode is in the
1281 * cache, the inode is returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 */
1283struct inode *ilookup(struct super_block *sb, unsigned long ino)
1284{
1285 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001286 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001288 spin_lock(&inode_hash_lock);
1289 inode = find_inode_fast(sb, head, ino);
1290 spin_unlock(&inode_hash_lock);
1291
1292 if (inode)
1293 wait_on_inode(inode);
1294 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296EXPORT_SYMBOL(ilookup);
1297
Theodore Ts'ofe032c42015-02-02 00:37:01 -05001298/**
1299 * find_inode_nowait - find an inode in the inode cache
1300 * @sb: super block of file system to search
1301 * @hashval: hash value (usually inode number) to search for
1302 * @match: callback used for comparisons between inodes
1303 * @data: opaque data pointer to pass to @match
1304 *
1305 * Search for the inode specified by @hashval and @data in the inode
1306 * cache, where the helper function @match will return 0 if the inode
1307 * does not match, 1 if the inode does match, and -1 if the search
1308 * should be stopped. The @match function must be responsible for
1309 * taking the i_lock spin_lock and checking i_state for an inode being
1310 * freed or being initialized, and incrementing the reference count
1311 * before returning 1. It also must not sleep, since it is called with
1312 * the inode_hash_lock spinlock held.
1313 *
1314 * This is a even more generalized version of ilookup5() when the
1315 * function must never block --- find_inode() can block in
1316 * __wait_on_freeing_inode() --- or when the caller can not increment
1317 * the reference count because the resulting iput() might cause an
1318 * inode eviction. The tradeoff is that the @match funtion must be
1319 * very carefully implemented.
1320 */
1321struct inode *find_inode_nowait(struct super_block *sb,
1322 unsigned long hashval,
1323 int (*match)(struct inode *, unsigned long,
1324 void *),
1325 void *data)
1326{
1327 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1328 struct inode *inode, *ret_inode = NULL;
1329 int mval;
1330
1331 spin_lock(&inode_hash_lock);
1332 hlist_for_each_entry(inode, head, i_hash) {
1333 if (inode->i_sb != sb)
1334 continue;
1335 mval = match(inode, hashval, data);
1336 if (mval == 0)
1337 continue;
1338 if (mval == 1)
1339 ret_inode = inode;
1340 goto out;
1341 }
1342out:
1343 spin_unlock(&inode_hash_lock);
1344 return ret_inode;
1345}
1346EXPORT_SYMBOL(find_inode_nowait);
1347
Al Viro261bca82008-12-30 01:48:21 -05001348int insert_inode_locked(struct inode *inode)
1349{
1350 struct super_block *sb = inode->i_sb;
1351 ino_t ino = inode->i_ino;
1352 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001353
Al Viro261bca82008-12-30 01:48:21 -05001354 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001355 struct inode *old = NULL;
Dave Chinner67a23c42011-03-22 22:23:42 +11001356 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001357 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001358 if (old->i_ino != ino)
1359 continue;
1360 if (old->i_sb != sb)
1361 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001362 spin_lock(&old->i_lock);
1363 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1364 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001365 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001366 }
Al Viro72a43d62009-05-13 19:13:40 +01001367 break;
1368 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001369 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001370 spin_lock(&inode->i_lock);
1371 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001372 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001373 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001374 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001375 return 0;
1376 }
1377 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001378 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001379 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001380 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001381 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001382 iput(old);
1383 return -EBUSY;
1384 }
1385 iput(old);
1386 }
1387}
Al Viro261bca82008-12-30 01:48:21 -05001388EXPORT_SYMBOL(insert_inode_locked);
1389
1390int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1391 int (*test)(struct inode *, void *), void *data)
1392{
1393 struct super_block *sb = inode->i_sb;
1394 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001395
Al Viro261bca82008-12-30 01:48:21 -05001396 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001397 struct inode *old = NULL;
1398
Dave Chinner67a23c42011-03-22 22:23:42 +11001399 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001400 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001401 if (old->i_sb != sb)
1402 continue;
1403 if (!test(old, data))
1404 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001405 spin_lock(&old->i_lock);
1406 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1407 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001408 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001409 }
Al Viro72a43d62009-05-13 19:13:40 +01001410 break;
1411 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001412 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001413 spin_lock(&inode->i_lock);
1414 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001415 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001416 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001417 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001418 return 0;
1419 }
1420 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001421 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001422 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001423 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001424 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001425 iput(old);
1426 return -EBUSY;
1427 }
1428 iput(old);
1429 }
1430}
Al Viro261bca82008-12-30 01:48:21 -05001431EXPORT_SYMBOL(insert_inode_locked4);
1432
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
Al Viro45321ac2010-06-07 13:43:19 -04001434int generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435{
Al Viro45321ac2010-06-07 13:43:19 -04001436 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438EXPORT_SYMBOL(generic_delete_inode);
1439
Al Viro45321ac2010-06-07 13:43:19 -04001440/*
Al Viro45321ac2010-06-07 13:43:19 -04001441 * Called when we're dropping the last reference
1442 * to an inode.
1443 *
1444 * Call the FS "drop_inode()" function, defaulting to
1445 * the legacy UNIX filesystem behaviour. If it tells
1446 * us to evict inode, do so. Otherwise, retain inode
1447 * in cache if fs is alive, sync and evict if fs is
1448 * shutting down.
1449 */
1450static void iput_final(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451{
1452 struct super_block *sb = inode->i_sb;
Al Viro45321ac2010-06-07 13:43:19 -04001453 const struct super_operations *op = inode->i_sb->s_op;
1454 int drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Dave Chinner250df6e2011-03-22 22:23:36 +11001456 WARN_ON(inode->i_state & I_NEW);
1457
Al Viroe7f59092011-07-07 15:45:59 -04001458 if (op->drop_inode)
Al Viro45321ac2010-06-07 13:43:19 -04001459 drop = op->drop_inode(inode);
1460 else
1461 drop = generic_drop_inode(inode);
1462
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001463 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1464 inode->i_state |= I_REFERENCED;
Jan Kara4eff96d2012-11-26 16:29:51 -08001465 inode_add_lru(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001466 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001467 return;
1468 }
1469
Al Viro45321ac2010-06-07 13:43:19 -04001470 if (!drop) {
Alexander Viro991114c2005-06-23 00:09:01 -07001471 inode->i_state |= I_WILL_FREE;
Dave Chinner250df6e2011-03-22 22:23:36 +11001472 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 write_inode_now(inode, 1);
Dave Chinner250df6e2011-03-22 22:23:36 +11001474 spin_lock(&inode->i_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001475 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001476 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001478
Alexander Viro991114c2005-06-23 00:09:01 -07001479 inode->i_state |= I_FREEING;
Eric Dumazetc4ae0c62011-07-28 06:55:13 +02001480 if (!list_empty(&inode->i_lru))
1481 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001482 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001483
1484 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485}
1486
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301488 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 * @inode: inode to put
1490 *
1491 * Puts an inode, dropping its usage count. If the inode use count hits
1492 * zero, the inode is then freed and may also be destroyed.
1493 *
1494 * Consequently, iput() can sleep.
1495 */
1496void iput(struct inode *inode)
1497{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001498 if (!inode)
1499 return;
1500 BUG_ON(inode->i_state & I_CLEAR);
1501retry:
1502 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1503 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1504 atomic_inc(&inode->i_count);
1505 inode->i_state &= ~I_DIRTY_TIME;
1506 spin_unlock(&inode->i_lock);
1507 trace_writeback_lazytime_iput(inode);
1508 mark_inode_dirty_sync(inode);
1509 goto retry;
1510 }
1511 iput_final(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 }
1513}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514EXPORT_SYMBOL(iput);
1515
1516/**
1517 * bmap - find a block number in a file
1518 * @inode: inode of file
1519 * @block: block to find
1520 *
1521 * Returns the block number on the device holding the inode that
1522 * is the disk block number for the block of the file requested.
1523 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301524 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 * file.
1526 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301527sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528{
1529 sector_t res = 0;
1530 if (inode->i_mapping->a_ops->bmap)
1531 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1532 return res;
1533}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534EXPORT_SYMBOL(bmap);
1535
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001536/*
1537 * With relative atime, only update atime if the previous atime is
1538 * earlier than either the ctime or mtime or if at least a day has
1539 * passed since the last atime update.
1540 */
1541static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1542 struct timespec now)
1543{
1544
1545 if (!(mnt->mnt_flags & MNT_RELATIME))
1546 return 1;
1547 /*
1548 * Is mtime younger than atime? If yes, update atime:
1549 */
1550 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1551 return 1;
1552 /*
1553 * Is ctime younger than atime? If yes, update atime:
1554 */
1555 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1556 return 1;
1557
1558 /*
1559 * Is the previous atime value older than a day? If yes,
1560 * update atime:
1561 */
1562 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1563 return 1;
1564 /*
1565 * Good, we can skip the atime update:
1566 */
1567 return 0;
1568}
1569
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001570int generic_update_time(struct inode *inode, struct timespec *time, int flags)
Josef Bacikc3b2da32012-03-26 09:59:21 -04001571{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001572 int iflags = I_DIRTY_TIME;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001573
1574 if (flags & S_ATIME)
1575 inode->i_atime = *time;
1576 if (flags & S_VERSION)
1577 inode_inc_iversion(inode);
1578 if (flags & S_CTIME)
1579 inode->i_ctime = *time;
1580 if (flags & S_MTIME)
1581 inode->i_mtime = *time;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001582
1583 if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION))
1584 iflags |= I_DIRTY_SYNC;
1585 __mark_inode_dirty(inode, iflags);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001586 return 0;
1587}
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001588EXPORT_SYMBOL(generic_update_time);
1589
1590/*
1591 * This does the actual work of updating an inodes time or version. Must have
1592 * had called mnt_want_write() before calling this.
1593 */
1594static int update_time(struct inode *inode, struct timespec *time, int flags)
1595{
1596 int (*update_time)(struct inode *, struct timespec *, int);
1597
1598 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1599 generic_update_time;
1600
1601 return update_time(inode, time, flags);
1602}
Josef Bacikc3b2da32012-03-26 09:59:21 -04001603
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001605 * touch_atime - update the access time
Randy Dunlap185553b2012-04-17 17:03:25 -07001606 * @path: the &struct path to update
Randy Dunlap30fdc8e2015-11-09 14:57:58 -08001607 * @inode: inode to update
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 *
1609 * Update the accessed time on an inode and mark it for writeback.
1610 * This function automatically handles read only file systems and media,
1611 * as well as the "noatime" flag and inode specific "noatime" markers.
1612 */
NeilBrown8fa9dd22015-03-23 13:37:40 +11001613bool atime_needs_update(const struct path *path, struct inode *inode)
1614{
1615 struct vfsmount *mnt = path->mnt;
1616 struct timespec now;
1617
1618 if (inode->i_flags & S_NOATIME)
1619 return false;
1620 if (IS_NOATIME(inode))
1621 return false;
1622 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1623 return false;
1624
1625 if (mnt->mnt_flags & MNT_NOATIME)
1626 return false;
1627 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1628 return false;
1629
1630 now = current_fs_time(inode->i_sb);
1631
1632 if (!relatime_need_update(mnt, inode, now))
1633 return false;
1634
1635 if (timespec_equal(&inode->i_atime, &now))
1636 return false;
1637
1638 return true;
1639}
1640
Al Virobadcf2b2013-07-16 18:15:46 +04001641void touch_atime(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642{
Al Viro68ac1232012-03-15 08:21:57 -04001643 struct vfsmount *mnt = path->mnt;
David Howellsdf2b1af2015-03-17 22:26:15 +00001644 struct inode *inode = d_inode(path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 struct timespec now;
1646
NeilBrown8fa9dd22015-03-23 13:37:40 +11001647 if (!atime_needs_update(path, inode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001648 return;
1649
Jan Kara5d37e9e2012-06-12 16:20:36 +02001650 if (!sb_start_write_trylock(inode->i_sb))
Andi Kleenb12536c2009-09-18 13:05:47 -07001651 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001652
NeilBrown8fa9dd22015-03-23 13:37:40 +11001653 if (__mnt_want_write(mnt) != 0)
Jan Kara5d37e9e2012-06-12 16:20:36 +02001654 goto skip_update;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001655 /*
1656 * File systems can error out when updating inodes if they need to
1657 * allocate new space to modify an inode (such is the case for
1658 * Btrfs), but since we touch atime while walking down the path we
1659 * really don't care if we failed to update the atime of the file,
1660 * so just ignore the return value.
Alexander Block2bc5565282012-06-15 09:49:33 +02001661 * We may also fail on filesystems that have the ability to make parts
1662 * of the fs read only, e.g. subvolumes in Btrfs.
Josef Bacikc3b2da32012-03-26 09:59:21 -04001663 */
NeilBrown8fa9dd22015-03-23 13:37:40 +11001664 now = current_fs_time(inode->i_sb);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001665 update_time(inode, &now, S_ATIME);
Jan Kara5d37e9e2012-06-12 16:20:36 +02001666 __mnt_drop_write(mnt);
1667skip_update:
1668 sb_end_write(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001670EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
Cong Wang3ed37642012-05-15 14:57:33 +08001672/*
1673 * The logic we want is
1674 *
1675 * if suid or (sgid and xgrp)
1676 * remove privs
1677 */
1678int should_remove_suid(struct dentry *dentry)
1679{
David Howellsdf2b1af2015-03-17 22:26:15 +00001680 umode_t mode = d_inode(dentry)->i_mode;
Cong Wang3ed37642012-05-15 14:57:33 +08001681 int kill = 0;
1682
1683 /* suid always must be killed */
1684 if (unlikely(mode & S_ISUID))
1685 kill = ATTR_KILL_SUID;
1686
1687 /*
1688 * sgid without any exec bits is just a mandatory locking mark; leave
1689 * it alone. If some exec bits are set, it's a real sgid; kill it.
1690 */
1691 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1692 kill |= ATTR_KILL_SGID;
1693
1694 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1695 return kill;
1696
1697 return 0;
1698}
1699EXPORT_SYMBOL(should_remove_suid);
1700
Jan Karadbfae0c2015-05-21 16:05:54 +02001701/*
1702 * Return mask of changes for notify_change() that need to be done as a
1703 * response to write or truncate. Return 0 if nothing has to be changed.
1704 * Negative value on error (change should be denied).
1705 */
Jan Kara45f147a2015-05-21 16:05:55 +02001706int dentry_needs_remove_privs(struct dentry *dentry)
Jan Karadbfae0c2015-05-21 16:05:54 +02001707{
Jan Karadbfae0c2015-05-21 16:05:54 +02001708 struct inode *inode = d_inode(dentry);
1709 int mask = 0;
1710 int ret;
1711
1712 if (IS_NOSEC(inode))
1713 return 0;
1714
1715 mask = should_remove_suid(dentry);
1716 ret = security_inode_need_killpriv(dentry);
1717 if (ret < 0)
1718 return ret;
1719 if (ret)
1720 mask |= ATTR_KILL_PRIV;
1721 return mask;
1722}
Jan Kara45f147a2015-05-21 16:05:55 +02001723EXPORT_SYMBOL(dentry_needs_remove_privs);
Jan Karadbfae0c2015-05-21 16:05:54 +02001724
1725static int __remove_privs(struct dentry *dentry, int kill)
Cong Wang3ed37642012-05-15 14:57:33 +08001726{
1727 struct iattr newattrs;
1728
1729 newattrs.ia_valid = ATTR_FORCE | kill;
J. Bruce Fields27ac0ff2011-09-20 17:19:26 -04001730 /*
1731 * Note we call this on write, so notify_change will not
1732 * encounter any conflicting delegations:
1733 */
1734 return notify_change(dentry, &newattrs, NULL);
Cong Wang3ed37642012-05-15 14:57:33 +08001735}
1736
Jan Kara5fa8e0a2015-05-21 16:05:53 +02001737/*
1738 * Remove special file priviledges (suid, capabilities) when file is written
1739 * to or truncated.
1740 */
1741int file_remove_privs(struct file *file)
Cong Wang3ed37642012-05-15 14:57:33 +08001742{
1743 struct dentry *dentry = file->f_path.dentry;
David Howellsdf2b1af2015-03-17 22:26:15 +00001744 struct inode *inode = d_inode(dentry);
Jan Karadbfae0c2015-05-21 16:05:54 +02001745 int kill;
Cong Wang3ed37642012-05-15 14:57:33 +08001746 int error = 0;
1747
1748 /* Fast path for nothing security related */
1749 if (IS_NOSEC(inode))
1750 return 0;
1751
Jan Karadbfae0c2015-05-21 16:05:54 +02001752 kill = file_needs_remove_privs(file);
1753 if (kill < 0)
1754 return kill;
1755 if (kill)
1756 error = __remove_privs(dentry, kill);
Jan Kara2426f392015-05-21 16:05:52 +02001757 if (!error)
1758 inode_has_no_xattr(inode);
Cong Wang3ed37642012-05-15 14:57:33 +08001759
1760 return error;
1761}
Jan Kara5fa8e0a2015-05-21 16:05:53 +02001762EXPORT_SYMBOL(file_remove_privs);
Cong Wang3ed37642012-05-15 14:57:33 +08001763
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001765 * file_update_time - update mtime and ctime time
1766 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001768 * Update the mtime and ctime members of an inode and mark the inode
1769 * for writeback. Note that this function is meant exclusively for
1770 * usage in the file write path of filesystems, and filesystems may
1771 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001772 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Josef Bacikc3b2da32012-03-26 09:59:21 -04001773 * timestamps are handled by the server. This can return an error for
1774 * file systems who need to allocate space in order to update an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 */
1776
Josef Bacikc3b2da32012-03-26 09:59:21 -04001777int file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778{
Al Viro496ad9a2013-01-23 17:07:38 -05001779 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 struct timespec now;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001781 int sync_it = 0;
1782 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783
Andi Kleence06e0b2009-09-18 13:05:48 -07001784 /* First try to exhaust all avenues to not sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 if (IS_NOCMTIME(inode))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001786 return 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001787
Andi Kleence06e0b2009-09-18 13:05:48 -07001788 now = current_fs_time(inode->i_sb);
1789 if (!timespec_equal(&inode->i_mtime, &now))
1790 sync_it = S_MTIME;
1791
1792 if (!timespec_equal(&inode->i_ctime, &now))
1793 sync_it |= S_CTIME;
1794
1795 if (IS_I_VERSION(inode))
1796 sync_it |= S_VERSION;
1797
1798 if (!sync_it)
Josef Bacikc3b2da32012-03-26 09:59:21 -04001799 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
Andi Kleence06e0b2009-09-18 13:05:48 -07001801 /* Finally allowed to write? Takes lock. */
Jan Karaeb04c282012-06-12 16:20:35 +02001802 if (__mnt_want_write_file(file))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001803 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
Josef Bacikc3b2da32012-03-26 09:59:21 -04001805 ret = update_time(inode, &now, sync_it);
Jan Karaeb04c282012-06-12 16:20:35 +02001806 __mnt_drop_write_file(file);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001807
1808 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001810EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
1812int inode_needs_sync(struct inode *inode)
1813{
1814 if (IS_SYNC(inode))
1815 return 1;
1816 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1817 return 1;
1818 return 0;
1819}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820EXPORT_SYMBOL(inode_needs_sync);
1821
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001823 * If we try to find an inode in the inode hash while it is being
1824 * deleted, we have to wait until the filesystem completes its
1825 * deletion before reporting that it isn't found. This function waits
1826 * until the deletion _might_ have completed. Callers are responsible
1827 * to recheck inode state.
1828 *
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001829 * It doesn't matter if I_NEW is not set initially, a call to
Dave Chinner250df6e2011-03-22 22:23:36 +11001830 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1831 * will DTRT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 */
1833static void __wait_on_freeing_inode(struct inode *inode)
1834{
1835 wait_queue_head_t *wq;
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001836 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1837 wq = bit_waitqueue(&inode->i_state, __I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +11001839 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001840 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 schedule();
1842 finish_wait(wq, &wait.wait);
Dave Chinner67a23c42011-03-22 22:23:42 +11001843 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844}
1845
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846static __initdata unsigned long ihash_entries;
1847static int __init set_ihash_entries(char *str)
1848{
1849 if (!str)
1850 return 0;
1851 ihash_entries = simple_strtoul(str, &str, 0);
1852 return 1;
1853}
1854__setup("ihash_entries=", set_ihash_entries);
1855
1856/*
1857 * Initialize the waitqueues and inode hash table.
1858 */
1859void __init inode_init_early(void)
1860{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001861 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
1863 /* If hashes are distributed across NUMA nodes, defer
1864 * hash allocation until vmalloc space is available.
1865 */
1866 if (hashdist)
1867 return;
1868
1869 inode_hashtable =
1870 alloc_large_system_hash("Inode-cache",
1871 sizeof(struct hlist_head),
1872 ihash_entries,
1873 14,
1874 HASH_EARLY,
1875 &i_hash_shift,
1876 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001877 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 0);
1879
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001880 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1882}
1883
Denis Cheng74bf17c2007-10-16 23:26:30 -07001884void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001886 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887
1888 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001889 inode_cachep = kmem_cache_create("inode_cache",
1890 sizeof(struct inode),
1891 0,
1892 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
Vladimir Davydov5d097052016-01-14 15:18:21 -08001893 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
Paul Mundt20c2df82007-07-20 10:11:58 +09001894 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
1896 /* Hash may have been set up in inode_init_early */
1897 if (!hashdist)
1898 return;
1899
1900 inode_hashtable =
1901 alloc_large_system_hash("Inode-cache",
1902 sizeof(struct hlist_head),
1903 ihash_entries,
1904 14,
1905 0,
1906 &i_hash_shift,
1907 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001908 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 0);
1910
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001911 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1913}
1914
1915void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1916{
1917 inode->i_mode = mode;
1918 if (S_ISCHR(mode)) {
1919 inode->i_fop = &def_chr_fops;
1920 inode->i_rdev = rdev;
1921 } else if (S_ISBLK(mode)) {
1922 inode->i_fop = &def_blk_fops;
1923 inode->i_rdev = rdev;
1924 } else if (S_ISFIFO(mode))
Al Viro599a0ac2013-03-12 09:58:10 -04001925 inode->i_fop = &pipefifo_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 else if (S_ISSOCK(mode))
Al Virobd9b51e2014-11-18 23:38:21 -05001927 ; /* leave it no_open_fops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 else
Manish Katiyaraf0d9ae2009-09-18 13:05:43 -07001929 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1930 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1931 inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932}
1933EXPORT_SYMBOL(init_special_inode);
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001934
1935/**
Ben Hutchingseaae668d2011-02-15 12:48:09 +00001936 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001937 * @inode: New inode
1938 * @dir: Directory inode
1939 * @mode: mode of the new inode
1940 */
1941void inode_init_owner(struct inode *inode, const struct inode *dir,
Al Viro62bb1092011-07-24 23:20:18 -04001942 umode_t mode)
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001943{
1944 inode->i_uid = current_fsuid();
1945 if (dir && dir->i_mode & S_ISGID) {
1946 inode->i_gid = dir->i_gid;
1947 if (S_ISDIR(mode))
1948 mode |= S_ISGID;
1949 } else
1950 inode->i_gid = current_fsgid();
1951 inode->i_mode = mode;
1952}
1953EXPORT_SYMBOL(inode_init_owner);
Serge E. Hallyne795b712011-03-23 16:43:25 -07001954
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001955/**
1956 * inode_owner_or_capable - check current task permissions to inode
1957 * @inode: inode being checked
1958 *
Andy Lutomirski23adbe12014-06-10 12:45:42 -07001959 * Return true if current either has CAP_FOWNER in a namespace with the
1960 * inode owner uid mapped, or owns the file.
Serge E. Hallyne795b712011-03-23 16:43:25 -07001961 */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001962bool inode_owner_or_capable(const struct inode *inode)
Serge E. Hallyne795b712011-03-23 16:43:25 -07001963{
Andy Lutomirski23adbe12014-06-10 12:45:42 -07001964 struct user_namespace *ns;
1965
Eric W. Biederman92361632012-02-08 07:07:50 -08001966 if (uid_eq(current_fsuid(), inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001967 return true;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07001968
1969 ns = current_user_ns();
1970 if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001971 return true;
1972 return false;
1973}
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001974EXPORT_SYMBOL(inode_owner_or_capable);
Trond Myklebust1d59d612012-05-31 12:22:33 -04001975
1976/*
1977 * Direct i/o helper functions
1978 */
1979static void __inode_dio_wait(struct inode *inode)
1980{
1981 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1982 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1983
1984 do {
1985 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1986 if (atomic_read(&inode->i_dio_count))
1987 schedule();
1988 } while (atomic_read(&inode->i_dio_count));
1989 finish_wait(wq, &q.wait);
1990}
1991
1992/**
1993 * inode_dio_wait - wait for outstanding DIO requests to finish
1994 * @inode: inode to wait for
1995 *
1996 * Waits for all pending direct I/O requests to finish so that we can
1997 * proceed with a truncate or equivalent operation.
1998 *
1999 * Must be called under a lock that serializes taking new references
2000 * to i_dio_count, usually by inode->i_mutex.
2001 */
2002void inode_dio_wait(struct inode *inode)
2003{
2004 if (atomic_read(&inode->i_dio_count))
2005 __inode_dio_wait(inode);
2006}
2007EXPORT_SYMBOL(inode_dio_wait);
2008
2009/*
Theodore Ts'o5f16f322014-03-24 14:43:12 -04002010 * inode_set_flags - atomically set some inode flags
2011 *
2012 * Note: the caller should be holding i_mutex, or else be sure that
2013 * they have exclusive access to the inode structure (i.e., while the
2014 * inode is being instantiated). The reason for the cmpxchg() loop
2015 * --- which wouldn't be necessary if all code paths which modify
2016 * i_flags actually followed this rule, is that there is at least one
Jan Kara5fa8e0a2015-05-21 16:05:53 +02002017 * code path which doesn't today so we use cmpxchg() out of an abundance
2018 * of caution.
Theodore Ts'o5f16f322014-03-24 14:43:12 -04002019 *
2020 * In the long run, i_mutex is overkill, and we should probably look
2021 * at using the i_lock spinlock to protect i_flags, and then make sure
2022 * it is so documented in include/linux/fs.h and that all code follows
2023 * the locking convention!!
2024 */
2025void inode_set_flags(struct inode *inode, unsigned int flags,
2026 unsigned int mask)
2027{
2028 unsigned int old_flags, new_flags;
2029
2030 WARN_ON_ONCE(flags & ~mask);
2031 do {
2032 old_flags = ACCESS_ONCE(inode->i_flags);
2033 new_flags = (old_flags & ~mask) | flags;
2034 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2035 new_flags) != old_flags));
2036}
2037EXPORT_SYMBOL(inode_set_flags);
Al Viro21fc61c2015-11-17 01:07:57 -05002038
2039void inode_nohighmem(struct inode *inode)
2040{
2041 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2042}
2043EXPORT_SYMBOL(inode_nohighmem);