blob: 7e3ef3af3db9e35dc175608d83d3e7ab7d3f789a [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));
Vladimir Davydov05eb6e72016-08-02 14:03:01 -0700348 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC | __GFP_ACCOUNT);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100349 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);
Dave Chinner6c60d2b2016-07-26 15:21:50 -0700368 INIT_LIST_HEAD(&inode->i_wb_list);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100369 INIT_LIST_HEAD(&inode->i_lru);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100370 address_space_init_once(&inode->i_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 i_size_ordered_init(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400372#ifdef CONFIG_FSNOTIFY
Eric Parise61ce862009-12-17 21:24:24 -0500373 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
Eric Paris3be25f42009-05-21 17:01:26 -0400374#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376EXPORT_SYMBOL(inode_init_once);
377
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700378static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530380 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
Christoph Lametera35afb82007-05-16 22:10:57 -0700382 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
384
385/*
Dave Chinner250df6e2011-03-22 22:23:36 +1100386 * inode->i_lock must be held
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530388void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389{
Nick Piggin9e38d862010-10-23 06:55:17 -0400390 atomic_inc(&inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
392
Al Viro7de9c6ee2010-10-23 11:11:40 -0400393/*
394 * get additional reference to inode; caller must already hold one.
395 */
396void ihold(struct inode *inode)
397{
398 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
399}
400EXPORT_SYMBOL(ihold);
401
Nick Piggin9e38d862010-10-23 06:55:17 -0400402static void inode_lru_list_add(struct inode *inode)
403{
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000404 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000405 this_cpu_inc(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400406}
407
Jan Kara4eff96d2012-11-26 16:29:51 -0800408/*
409 * Add inode to LRU if needed (inode is unused and clean).
410 *
411 * Needs inode->i_lock held.
412 */
413void inode_add_lru(struct inode *inode)
414{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500415 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
416 I_FREEING | I_WILL_FREE)) &&
Jan Kara4eff96d2012-11-26 16:29:51 -0800417 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
418 inode_lru_list_add(inode);
419}
420
421
Nick Piggin9e38d862010-10-23 06:55:17 -0400422static void inode_lru_list_del(struct inode *inode)
423{
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000424
425 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000426 this_cpu_dec(nr_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427}
428
Christoph Hellwig646ec462010-10-23 07:15:32 -0400429/**
430 * inode_sb_list_add - add inode to the superblock list of inodes
431 * @inode: inode to add
432 */
433void inode_sb_list_add(struct inode *inode)
434{
Dave Chinner74278da2015-03-04 12:37:22 -0500435 spin_lock(&inode->i_sb->s_inode_list_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +1100436 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
Dave Chinner74278da2015-03-04 12:37:22 -0500437 spin_unlock(&inode->i_sb->s_inode_list_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400438}
439EXPORT_SYMBOL_GPL(inode_sb_list_add);
440
Dave Chinner55fa6092011-03-22 22:23:40 +1100441static inline void inode_sb_list_del(struct inode *inode)
Christoph Hellwig646ec462010-10-23 07:15:32 -0400442{
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200443 if (!list_empty(&inode->i_sb_list)) {
Dave Chinner74278da2015-03-04 12:37:22 -0500444 spin_lock(&inode->i_sb->s_inode_list_lock);
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200445 list_del_init(&inode->i_sb_list);
Dave Chinner74278da2015-03-04 12:37:22 -0500446 spin_unlock(&inode->i_sb->s_inode_list_lock);
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200447 }
Christoph Hellwig646ec462010-10-23 07:15:32 -0400448}
449
Dave Chinner4c51acb2010-10-23 06:58:09 -0400450static unsigned long hash(struct super_block *sb, unsigned long hashval)
451{
452 unsigned long tmp;
453
454 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
455 L1_CACHE_BYTES;
Christoph Hellwig4b4563d2011-05-27 09:28:01 -0400456 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
457 return tmp & i_hash_mask;
Dave Chinner4c51acb2010-10-23 06:58:09 -0400458}
459
460/**
461 * __insert_inode_hash - hash an inode
462 * @inode: unhashed inode
463 * @hashval: unsigned long value used to locate this object in the
464 * inode_hashtable.
465 *
466 * Add an inode to the inode hash for this superblock.
467 */
468void __insert_inode_hash(struct inode *inode, unsigned long hashval)
469{
Christoph Hellwig646ec462010-10-23 07:15:32 -0400470 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
471
Dave Chinner67a23c42011-03-22 22:23:42 +1100472 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100473 spin_lock(&inode->i_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400474 hlist_add_head(&inode->i_hash, b);
Dave Chinner250df6e2011-03-22 22:23:36 +1100475 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100476 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400477}
478EXPORT_SYMBOL(__insert_inode_hash);
479
480/**
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200481 * __remove_inode_hash - remove an inode from the hash
Dave Chinner4c51acb2010-10-23 06:58:09 -0400482 * @inode: inode to unhash
483 *
484 * Remove an inode from the superblock.
485 */
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200486void __remove_inode_hash(struct inode *inode)
Dave Chinner4c51acb2010-10-23 06:58:09 -0400487{
Dave Chinner67a23c42011-03-22 22:23:42 +1100488 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100489 spin_lock(&inode->i_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400490 hlist_del_init(&inode->i_hash);
Dave Chinner250df6e2011-03-22 22:23:36 +1100491 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100492 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400493}
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200494EXPORT_SYMBOL(__remove_inode_hash);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400495
Jan Karadbd57682012-05-03 14:48:02 +0200496void clear_inode(struct inode *inode)
Al Virob0683aa2010-06-04 20:55:25 -0400497{
498 might_sleep();
Jan Kara08142572011-06-27 16:18:10 -0700499 /*
500 * We have to cycle tree_lock here because reclaim can be still in the
501 * process of removing the last page (in __delete_from_page_cache())
502 * and we must not free mapping under it.
503 */
504 spin_lock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400505 BUG_ON(inode->i_data.nrpages);
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800506 BUG_ON(inode->i_data.nrexceptional);
Jan Kara08142572011-06-27 16:18:10 -0700507 spin_unlock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400508 BUG_ON(!list_empty(&inode->i_data.private_list));
509 BUG_ON(!(inode->i_state & I_FREEING));
510 BUG_ON(inode->i_state & I_CLEAR);
Dave Chinner6c60d2b2016-07-26 15:21:50 -0700511 BUG_ON(!list_empty(&inode->i_wb_list));
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100512 /* don't need i_lock here, no concurrent mods to i_state */
Al Virob0683aa2010-06-04 20:55:25 -0400513 inode->i_state = I_FREEING | I_CLEAR;
514}
Jan Karadbd57682012-05-03 14:48:02 +0200515EXPORT_SYMBOL(clear_inode);
Al Virob0683aa2010-06-04 20:55:25 -0400516
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100517/*
518 * Free the inode passed in, removing it from the lists it is still connected
519 * to. We remove any pages still attached to the inode and wait for any IO that
520 * is still in progress before finally destroying the inode.
521 *
522 * An inode must already be marked I_FREEING so that we avoid the inode being
523 * moved back onto lists if we race with other code that manipulates the lists
524 * (e.g. writeback_single_inode). The caller is responsible for setting this.
525 *
526 * An inode must already be removed from the LRU list before being evicted from
527 * the cache. This should occur atomically with setting the I_FREEING state
528 * flag, so no inodes here should ever be on the LRU when being evicted.
529 */
Al Viro644da592010-06-07 13:21:05 -0400530static void evict(struct inode *inode)
Al Virob4272d42010-06-04 19:33:20 -0400531{
532 const struct super_operations *op = inode->i_sb->s_op;
533
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100534 BUG_ON(!(inode->i_state & I_FREEING));
535 BUG_ON(!list_empty(&inode->i_lru));
536
Dave Chinnerc7f54082015-03-04 14:07:22 -0500537 if (!list_empty(&inode->i_io_list))
538 inode_io_list_del(inode);
Eric Dumazetb12362b2011-07-28 06:11:47 +0200539
Dave Chinner55fa6092011-03-22 22:23:40 +1100540 inode_sb_list_del(inode);
541
Jan Kara169ebd92012-05-03 14:48:03 +0200542 /*
543 * Wait for flusher thread to be done with the inode so that filesystem
544 * does not start destroying it while writeback is still running. Since
545 * the inode has I_FREEING set, flusher thread won't start new work on
546 * the inode. We just have to wait for running writeback to finish.
547 */
548 inode_wait_for_writeback(inode);
Jan Kara7994e6f2012-05-03 14:48:01 +0200549
Al Virobe7ce412010-06-04 19:40:39 -0400550 if (op->evict_inode) {
551 op->evict_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400552 } else {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700553 truncate_inode_pages_final(&inode->i_data);
Jan Karadbd57682012-05-03 14:48:02 +0200554 clear_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400555 }
Al Viro661074e2010-06-04 20:19:55 -0400556 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
557 bd_forget(inode);
558 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
559 cd_forget(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100560
561 remove_inode_hash(inode);
562
563 spin_lock(&inode->i_lock);
564 wake_up_bit(&inode->i_state, __I_NEW);
565 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
566 spin_unlock(&inode->i_lock);
567
568 destroy_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400569}
570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571/*
572 * dispose_list - dispose of the contents of a local list
573 * @head: the head of the list to free
574 *
575 * Dispose-list gets a local list with local inodes in it, so it doesn't
576 * need to worry about list corruption and SMP locks.
577 */
578static void dispose_list(struct list_head *head)
579{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 while (!list_empty(head)) {
581 struct inode *inode;
582
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100583 inode = list_first_entry(head, struct inode, i_lru);
584 list_del_init(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Al Viro644da592010-06-07 13:21:05 -0400586 evict(inode);
Josef Bacikac05fbb2015-03-04 16:52:52 -0500587 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589}
590
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591/**
Al Viro63997e92010-10-25 20:49:35 -0400592 * evict_inodes - evict all evictable inodes for a superblock
593 * @sb: superblock to operate on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 *
Al Viro63997e92010-10-25 20:49:35 -0400595 * Make sure that no inodes with zero refcount are retained. This is
596 * called by superblock shutdown after having MS_ACTIVE flag removed,
597 * so any inode reaching zero refcount during or after that call will
598 * be immediately evicted.
599 */
600void evict_inodes(struct super_block *sb)
601{
602 struct inode *inode, *next;
603 LIST_HEAD(dispose);
604
Josef Bacikac05fbb2015-03-04 16:52:52 -0500605again:
Dave Chinner74278da2015-03-04 12:37:22 -0500606 spin_lock(&sb->s_inode_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400607 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
608 if (atomic_read(&inode->i_count))
609 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100610
611 spin_lock(&inode->i_lock);
612 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
613 spin_unlock(&inode->i_lock);
Al Viro63997e92010-10-25 20:49:35 -0400614 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100615 }
Al Viro63997e92010-10-25 20:49:35 -0400616
617 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100618 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100619 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100620 list_add(&inode->i_lru, &dispose);
Josef Bacikac05fbb2015-03-04 16:52:52 -0500621
622 /*
623 * We can have a ton of inodes to evict at unmount time given
624 * enough memory, check to see if we need to go to sleep for a
625 * bit so we don't livelock.
626 */
627 if (need_resched()) {
628 spin_unlock(&sb->s_inode_list_lock);
629 cond_resched();
630 dispose_list(&dispose);
631 goto again;
632 }
Al Viro63997e92010-10-25 20:49:35 -0400633 }
Dave Chinner74278da2015-03-04 12:37:22 -0500634 spin_unlock(&sb->s_inode_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400635
636 dispose_list(&dispose);
Al Viro63997e92010-10-25 20:49:35 -0400637}
638
639/**
Christoph Hellwiga0318782010-10-24 19:40:33 +0200640 * invalidate_inodes - attempt to free all inodes on a superblock
641 * @sb: superblock to operate on
NeilBrown93b270f2011-02-24 17:25:47 +1100642 * @kill_dirty: flag to guide handling of dirty inodes
Christoph Hellwiga0318782010-10-24 19:40:33 +0200643 *
644 * Attempts to free all inodes for a given superblock. If there were any
645 * busy inodes return a non-zero value, else zero.
NeilBrown93b270f2011-02-24 17:25:47 +1100646 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
647 * them as busy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 */
NeilBrown93b270f2011-02-24 17:25:47 +1100649int invalidate_inodes(struct super_block *sb, bool kill_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400651 int busy = 0;
Christoph Hellwiga0318782010-10-24 19:40:33 +0200652 struct inode *inode, *next;
653 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
Dave Chinner74278da2015-03-04 12:37:22 -0500655 spin_lock(&sb->s_inode_list_lock);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200656 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100657 spin_lock(&inode->i_lock);
658 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
659 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100661 }
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500662 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100663 spin_unlock(&inode->i_lock);
NeilBrown93b270f2011-02-24 17:25:47 +1100664 busy = 1;
665 continue;
666 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200667 if (atomic_read(&inode->i_count)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100668 spin_unlock(&inode->i_lock);
Christoph Hellwig99a38912010-10-23 19:07:20 +0200669 busy = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 continue;
671 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200672
Christoph Hellwig99a38912010-10-23 19:07:20 +0200673 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100674 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100675 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100676 list_add(&inode->i_lru, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 }
Dave Chinner74278da2015-03-04 12:37:22 -0500678 spin_unlock(&sb->s_inode_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Christoph Hellwiga0318782010-10-24 19:40:33 +0200680 dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
682 return busy;
683}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685/*
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000686 * Isolate the inode from the LRU in preparation for freeing it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 *
688 * Any inodes which are pinned purely because of attached pagecache have their
Nick Piggin9e38d862010-10-23 06:55:17 -0400689 * pagecache removed. If the inode has metadata buffers attached to
690 * mapping->private_list then try to remove them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 *
Nick Piggin9e38d862010-10-23 06:55:17 -0400692 * If the inode has the I_REFERENCED flag set, then it means that it has been
693 * used recently - the flag is set in iput_final(). When we encounter such an
694 * inode, clear the flag and move it to the back of the LRU so it gets another
695 * pass through the LRU before it gets reclaimed. This is necessary because of
696 * the fact we are doing lazy LRU updates to minimise lock contention so the
697 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
698 * with this flag set because they are the inodes that are out of order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800700static enum lru_status inode_lru_isolate(struct list_head *item,
701 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000702{
703 struct list_head *freeable = arg;
704 struct inode *inode = container_of(item, struct inode, i_lru);
705
706 /*
707 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
708 * If we fail to get the lock, just skip it.
709 */
710 if (!spin_trylock(&inode->i_lock))
711 return LRU_SKIP;
712
713 /*
714 * Referenced or dirty inodes are still in use. Give them another pass
715 * through the LRU as we canot reclaim them now.
716 */
717 if (atomic_read(&inode->i_count) ||
718 (inode->i_state & ~I_REFERENCED)) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800719 list_lru_isolate(lru, &inode->i_lru);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000720 spin_unlock(&inode->i_lock);
721 this_cpu_dec(nr_unused);
722 return LRU_REMOVED;
723 }
724
725 /* recently referenced inodes get one more pass */
726 if (inode->i_state & I_REFERENCED) {
727 inode->i_state &= ~I_REFERENCED;
728 spin_unlock(&inode->i_lock);
729 return LRU_ROTATE;
730 }
731
732 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
733 __iget(inode);
734 spin_unlock(&inode->i_lock);
735 spin_unlock(lru_lock);
736 if (remove_inode_buffers(inode)) {
737 unsigned long reap;
738 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
739 if (current_is_kswapd())
740 __count_vm_events(KSWAPD_INODESTEAL, reap);
741 else
742 __count_vm_events(PGINODESTEAL, reap);
743 if (current->reclaim_state)
744 current->reclaim_state->reclaimed_slab += reap;
745 }
746 iput(inode);
747 spin_lock(lru_lock);
748 return LRU_RETRY;
749 }
750
751 WARN_ON(inode->i_state & I_NEW);
752 inode->i_state |= I_FREEING;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800753 list_lru_isolate_move(lru, &inode->i_lru, freeable);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000754 spin_unlock(&inode->i_lock);
755
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000756 this_cpu_dec(nr_unused);
757 return LRU_REMOVED;
758}
759
760/*
761 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
762 * This is called from the superblock shrinker function with a number of inodes
763 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
764 * then are freed outside inode_lock by dispose_list().
765 */
Vladimir Davydov503c3582015-02-12 14:58:47 -0800766long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767{
768 LIST_HEAD(freeable);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000769 long freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Vladimir Davydov503c3582015-02-12 14:58:47 -0800771 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
772 inode_lru_isolate, &freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 dispose_list(&freeable);
Dave Chinner0a234c62013-08-28 10:17:57 +1000774 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775}
776
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777static void __wait_on_freeing_inode(struct inode *inode);
778/*
779 * Called with the inode lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530781static struct inode *find_inode(struct super_block *sb,
782 struct hlist_head *head,
783 int (*test)(struct inode *, void *),
784 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530786 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787
788repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800789 hlist_for_each_entry(inode, head, i_hash) {
Al Viro5a3cd99282013-11-06 09:54:52 -0500790 if (inode->i_sb != sb)
791 continue;
792 if (!test(inode, data))
793 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100794 spin_lock(&inode->i_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -0400795 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 __wait_on_freeing_inode(inode);
797 goto repeat;
798 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400799 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100800 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400801 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400803 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804}
805
806/*
807 * find_inode_fast is the fast path version of find_inode, see the comment at
808 * iget_locked for details.
809 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530810static struct inode *find_inode_fast(struct super_block *sb,
811 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530813 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
815repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800816 hlist_for_each_entry(inode, head, i_hash) {
Al Viro5a3cd99282013-11-06 09:54:52 -0500817 if (inode->i_ino != ino)
818 continue;
819 if (inode->i_sb != sb)
820 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100821 spin_lock(&inode->i_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -0400822 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 __wait_on_freeing_inode(inode);
824 goto repeat;
825 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400826 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100827 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400828 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400830 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831}
832
Eric Dumazetf991bd22010-10-23 11:18:01 -0400833/*
834 * Each cpu owns a range of LAST_INO_BATCH numbers.
835 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
836 * to renew the exhausted range.
David Chinner8290c352008-10-30 17:35:24 +1100837 *
Eric Dumazetf991bd22010-10-23 11:18:01 -0400838 * This does not significantly increase overflow rate because every CPU can
839 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
840 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
841 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
842 * overflow rate by 2x, which does not seem too significant.
843 *
844 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
845 * error if st_ino won't fit in target struct field. Use 32bit counter
846 * here to attempt to avoid that.
David Chinner8290c352008-10-30 17:35:24 +1100847 */
Eric Dumazetf991bd22010-10-23 11:18:01 -0400848#define LAST_INO_BATCH 1024
849static DEFINE_PER_CPU(unsigned int, last_ino);
David Chinner8290c352008-10-30 17:35:24 +1100850
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400851unsigned int get_next_ino(void)
Eric Dumazetf991bd22010-10-23 11:18:01 -0400852{
853 unsigned int *p = &get_cpu_var(last_ino);
854 unsigned int res = *p;
855
856#ifdef CONFIG_SMP
857 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
858 static atomic_t shared_last_ino;
859 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
860
861 res = next - LAST_INO_BATCH;
862 }
863#endif
864
Carlos Maiolino2adc3762015-06-25 12:25:58 -0300865 res++;
866 /* get_next_ino should not provide a 0 inode number */
867 if (unlikely(!res))
868 res++;
869 *p = res;
Eric Dumazetf991bd22010-10-23 11:18:01 -0400870 put_cpu_var(last_ino);
871 return res;
David Chinner8290c352008-10-30 17:35:24 +1100872}
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400873EXPORT_SYMBOL(get_next_ino);
David Chinner8290c352008-10-30 17:35:24 +1100874
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875/**
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200876 * new_inode_pseudo - obtain an inode
877 * @sb: superblock
878 *
879 * Allocates a new inode for given superblock.
880 * Inode wont be chained in superblock s_inodes list
881 * This means :
882 * - fs can't be unmount
883 * - quotas, fsnotify, writeback can't work
884 */
885struct inode *new_inode_pseudo(struct super_block *sb)
886{
887 struct inode *inode = alloc_inode(sb);
888
889 if (inode) {
890 spin_lock(&inode->i_lock);
891 inode->i_state = 0;
892 spin_unlock(&inode->i_lock);
893 INIT_LIST_HEAD(&inode->i_sb_list);
894 }
895 return inode;
896}
897
898/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 * new_inode - obtain an inode
900 * @sb: superblock
901 *
Mel Gorman769848c2007-07-17 04:03:05 -0700902 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800903 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700904 * If HIGHMEM pages are unsuitable or it is known that pages allocated
905 * for the page cache are not reclaimable or migratable,
906 * mapping_set_gfp_mask() must be called with suitable flags on the
907 * newly created inode's mapping
908 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 */
910struct inode *new_inode(struct super_block *sb)
911{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530912 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
Dave Chinner74278da2015-03-04 12:37:22 -0500914 spin_lock_prefetch(&sb->s_inode_list_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530915
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200916 inode = new_inode_pseudo(sb);
917 if (inode)
Dave Chinner55fa6092011-03-22 22:23:40 +1100918 inode_sb_list_add(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 return inode;
920}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921EXPORT_SYMBOL(new_inode);
922
Peter Zijlstra14358e62007-10-14 01:38:33 +0200923#ifdef CONFIG_DEBUG_LOCK_ALLOC
Josh Boyere096d0c2011-08-25 07:48:12 -0400924void lockdep_annotate_inode_mutex_key(struct inode *inode)
925{
Namhyung Kima3314a02010-10-11 22:38:00 +0900926 if (S_ISDIR(inode->i_mode)) {
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200927 struct file_system_type *type = inode->i_sb->s_type;
928
Jan Kara9a7aa122009-06-04 15:26:49 +0200929 /* Set new key only if filesystem hasn't already changed it */
Al Viro9902af72016-04-15 15:08:36 -0400930 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
Jan Kara9a7aa122009-06-04 15:26:49 +0200931 /*
932 * ensure nobody is actually holding i_mutex
933 */
Al Viro9902af72016-04-15 15:08:36 -0400934 // mutex_destroy(&inode->i_mutex);
935 init_rwsem(&inode->i_rwsem);
936 lockdep_set_class(&inode->i_rwsem,
Jan Kara9a7aa122009-06-04 15:26:49 +0200937 &type->i_mutex_dir_key);
938 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200939 }
Josh Boyere096d0c2011-08-25 07:48:12 -0400940}
941EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200942#endif
Josh Boyere096d0c2011-08-25 07:48:12 -0400943
944/**
945 * unlock_new_inode - clear the I_NEW state and wake up any waiters
946 * @inode: new inode to unlock
947 *
948 * Called when the inode is fully initialised to clear the new state of the
949 * inode and wake up anyone waiting for the inode to finish initialisation.
950 */
951void unlock_new_inode(struct inode *inode)
952{
953 lockdep_annotate_inode_mutex_key(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100954 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100955 WARN_ON(!(inode->i_state & I_NEW));
956 inode->i_state &= ~I_NEW;
Al Viro310fa7a2012-03-10 17:07:28 -0500957 smp_mb();
Dave Chinner250df6e2011-03-22 22:23:36 +1100958 wake_up_bit(&inode->i_state, __I_NEW);
959 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961EXPORT_SYMBOL(unlock_new_inode);
962
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400963/**
J. Bruce Fields375e2892012-04-18 15:16:33 -0400964 * lock_two_nondirectories - take two i_mutexes on non-directory objects
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200965 *
966 * Lock any non-NULL argument that is not a directory.
967 * Zero, one or two objects may be locked by this function.
968 *
J. Bruce Fields375e2892012-04-18 15:16:33 -0400969 * @inode1: first inode to lock
970 * @inode2: second inode to lock
971 */
972void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
973{
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200974 if (inode1 > inode2)
975 swap(inode1, inode2);
976
977 if (inode1 && !S_ISDIR(inode1->i_mode))
Al Viro59551022016-01-22 15:40:57 -0500978 inode_lock(inode1);
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200979 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
Al Viro59551022016-01-22 15:40:57 -0500980 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
J. Bruce Fields375e2892012-04-18 15:16:33 -0400981}
982EXPORT_SYMBOL(lock_two_nondirectories);
983
984/**
985 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
986 * @inode1: first inode to unlock
987 * @inode2: second inode to unlock
988 */
989void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
990{
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200991 if (inode1 && !S_ISDIR(inode1->i_mode))
Al Viro59551022016-01-22 15:40:57 -0500992 inode_unlock(inode1);
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200993 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
Al Viro59551022016-01-22 15:40:57 -0500994 inode_unlock(inode2);
J. Bruce Fields375e2892012-04-18 15:16:33 -0400995}
996EXPORT_SYMBOL(unlock_two_nondirectories);
997
998/**
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400999 * iget5_locked - obtain an inode from a mounted file system
1000 * @sb: super block of file system
1001 * @hashval: hash value (usually inode number) to get
1002 * @test: callback used for comparisons between inodes
1003 * @set: callback used to initialize a new struct inode
1004 * @data: opaque data pointer to pass to @test and @set
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001006 * Search for the inode specified by @hashval and @data in the inode cache,
1007 * and if present it is return it with an increased reference count. This is
1008 * a generalized version of iget_locked() for file systems where the inode
1009 * number is not sufficient for unique identification of an inode.
1010 *
1011 * If the inode is not in cache, allocate a new inode and return it locked,
1012 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1013 * before unlocking it via unlock_new_inode().
1014 *
1015 * Note both @test and @set are called with the inode_hash_lock held, so can't
1016 * sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001018struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1019 int (*test)(struct inode *, void *),
1020 int (*set)(struct inode *, void *), void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001022 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301023 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001025 spin_lock(&inode_hash_lock);
1026 inode = find_inode(sb, head, test, data);
1027 spin_unlock(&inode_hash_lock);
1028
1029 if (inode) {
1030 wait_on_inode(inode);
1031 return inode;
1032 }
1033
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 inode = alloc_inode(sb);
1035 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301036 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
Dave Chinner67a23c42011-03-22 22:23:42 +11001038 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 /* We released the lock, so.. */
1040 old = find_inode(sb, head, test, data);
1041 if (!old) {
1042 if (set(inode, data))
1043 goto set_failed;
1044
Dave Chinner250df6e2011-03-22 22:23:36 +11001045 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001046 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001047 hlist_add_head(&inode->i_hash, head);
1048 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001049 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001050 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
1052 /* Return the locked inode with I_NEW set, the
1053 * caller is responsible for filling in the contents
1054 */
1055 return inode;
1056 }
1057
1058 /*
1059 * Uhhuh, somebody else created the same inode under
1060 * us. Use the old inode instead of the one we just
1061 * allocated.
1062 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001063 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 destroy_inode(inode);
1065 inode = old;
1066 wait_on_inode(inode);
1067 }
1068 return inode;
1069
1070set_failed:
Dave Chinner67a23c42011-03-22 22:23:42 +11001071 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 destroy_inode(inode);
1073 return NULL;
1074}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001075EXPORT_SYMBOL(iget5_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001077/**
1078 * iget_locked - obtain an inode from a mounted file system
1079 * @sb: super block of file system
1080 * @ino: inode number to get
1081 *
1082 * Search for the inode specified by @ino in the inode cache and if present
1083 * return it with an increased reference count. This is for file systems
1084 * where the inode number is sufficient for unique identification of an inode.
1085 *
1086 * If the inode is not in cache, allocate a new inode and return it locked,
1087 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1088 * before unlocking it via unlock_new_inode().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001090struct inode *iget_locked(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001092 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301093 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001095 spin_lock(&inode_hash_lock);
1096 inode = find_inode_fast(sb, head, ino);
1097 spin_unlock(&inode_hash_lock);
1098 if (inode) {
1099 wait_on_inode(inode);
1100 return inode;
1101 }
1102
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 inode = alloc_inode(sb);
1104 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301105 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106
Dave Chinner67a23c42011-03-22 22:23:42 +11001107 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 /* We released the lock, so.. */
1109 old = find_inode_fast(sb, head, ino);
1110 if (!old) {
1111 inode->i_ino = ino;
Dave Chinner250df6e2011-03-22 22:23:36 +11001112 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001113 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001114 hlist_add_head(&inode->i_hash, head);
1115 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001116 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001117 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
1119 /* Return the locked inode with I_NEW set, the
1120 * caller is responsible for filling in the contents
1121 */
1122 return inode;
1123 }
1124
1125 /*
1126 * Uhhuh, somebody else created the same inode under
1127 * us. Use the old inode instead of the one we just
1128 * allocated.
1129 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001130 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 destroy_inode(inode);
1132 inode = old;
1133 wait_on_inode(inode);
1134 }
1135 return inode;
1136}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001137EXPORT_SYMBOL(iget_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001139/*
1140 * search the inode cache for a matching inode number.
1141 * If we find one, then the inode number we are trying to
1142 * allocate is not unique and so we should not use it.
1143 *
1144 * Returns 1 if the inode number is unique, 0 if it is not.
1145 */
1146static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1147{
1148 struct hlist_head *b = inode_hashtable + hash(sb, ino);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001149 struct inode *inode;
1150
Dave Chinner67a23c42011-03-22 22:23:42 +11001151 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001152 hlist_for_each_entry(inode, b, i_hash) {
Dave Chinner67a23c42011-03-22 22:23:42 +11001153 if (inode->i_ino == ino && inode->i_sb == sb) {
1154 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001155 return 0;
Dave Chinner67a23c42011-03-22 22:23:42 +11001156 }
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001157 }
Dave Chinner67a23c42011-03-22 22:23:42 +11001158 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001159
1160 return 1;
1161}
1162
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163/**
1164 * iunique - get a unique inode number
1165 * @sb: superblock
1166 * @max_reserved: highest reserved inode number
1167 *
1168 * Obtain an inode number that is unique on the system for a given
1169 * superblock. This is used by file systems that have no natural
1170 * permanent inode numbering system. An inode number is returned that
1171 * is higher than the reserved limit but unique.
1172 *
1173 * BUGS:
1174 * With a large number of inodes live on the file system this function
1175 * currently becomes quite slow.
1176 */
1177ino_t iunique(struct super_block *sb, ino_t max_reserved)
1178{
Jeff Layton866b04f2007-05-08 00:32:29 -07001179 /*
1180 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1181 * error if st_ino won't fit in target struct field. Use 32bit counter
1182 * here to attempt to avoid that.
1183 */
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001184 static DEFINE_SPINLOCK(iunique_lock);
Jeff Layton866b04f2007-05-08 00:32:29 -07001185 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001188 spin_lock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001189 do {
1190 if (counter <= max_reserved)
1191 counter = max_reserved + 1;
1192 res = counter++;
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001193 } while (!test_inode_iunique(sb, res));
1194 spin_unlock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001195
1196 return res;
1197}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198EXPORT_SYMBOL(iunique);
1199
1200struct inode *igrab(struct inode *inode)
1201{
Dave Chinner250df6e2011-03-22 22:23:36 +11001202 spin_lock(&inode->i_lock);
1203 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001205 spin_unlock(&inode->i_lock);
1206 } else {
1207 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 /*
1209 * Handle the case where s_op->clear_inode is not been
1210 * called yet, and somebody is calling igrab
1211 * while the inode is getting freed.
1212 */
1213 inode = NULL;
Dave Chinner250df6e2011-03-22 22:23:36 +11001214 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 return inode;
1216}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217EXPORT_SYMBOL(igrab);
1218
1219/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001220 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 * @sb: super block of file system to search
1222 * @hashval: hash value (usually inode number) to search for
1223 * @test: callback used for comparisons between inodes
1224 * @data: opaque data pointer to pass to @test
1225 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001226 * Search for the inode specified by @hashval and @data in the inode cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 * If the inode is in the cache, the inode is returned with an incremented
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001228 * reference count.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001229 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001230 * Note: I_NEW is not waited upon so you have to be very careful what you do
1231 * with the returned inode. You probably should be using ilookup5() instead.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001232 *
Randy Dunlapb6d0ad62011-03-26 13:27:47 -07001233 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001234 */
1235struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1236 int (*test)(struct inode *, void *), void *data)
1237{
1238 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001239 struct inode *inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001240
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001241 spin_lock(&inode_hash_lock);
1242 inode = find_inode(sb, head, test, data);
1243 spin_unlock(&inode_hash_lock);
1244
1245 return inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001246}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001247EXPORT_SYMBOL(ilookup5_nowait);
1248
1249/**
1250 * ilookup5 - search for an inode in the inode cache
1251 * @sb: super block of file system to search
1252 * @hashval: hash value (usually inode number) to search for
1253 * @test: callback used for comparisons between inodes
1254 * @data: opaque data pointer to pass to @test
1255 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001256 * Search for the inode specified by @hashval and @data in the inode cache,
1257 * and if the inode is in the cache, return the inode with an incremented
1258 * reference count. Waits on I_NEW before returning the inode.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001259 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001261 * This is a generalized version of ilookup() for file systems where the
1262 * inode number is not sufficient for unique identification of an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001264 * Note: @test is called with the inode_hash_lock held, so can't sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 */
1266struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1267 int (*test)(struct inode *, void *), void *data)
1268{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001269 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001271 if (inode)
1272 wait_on_inode(inode);
1273 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275EXPORT_SYMBOL(ilookup5);
1276
1277/**
1278 * ilookup - search for an inode in the inode cache
1279 * @sb: super block of file system to search
1280 * @ino: inode number to search for
1281 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001282 * Search for the inode @ino in the inode cache, and if the inode is in the
1283 * cache, the inode is returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 */
1285struct inode *ilookup(struct super_block *sb, unsigned long ino)
1286{
1287 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001288 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001290 spin_lock(&inode_hash_lock);
1291 inode = find_inode_fast(sb, head, ino);
1292 spin_unlock(&inode_hash_lock);
1293
1294 if (inode)
1295 wait_on_inode(inode);
1296 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298EXPORT_SYMBOL(ilookup);
1299
Theodore Ts'ofe032c42015-02-02 00:37:01 -05001300/**
1301 * find_inode_nowait - find an inode in the inode cache
1302 * @sb: super block of file system to search
1303 * @hashval: hash value (usually inode number) to search for
1304 * @match: callback used for comparisons between inodes
1305 * @data: opaque data pointer to pass to @match
1306 *
1307 * Search for the inode specified by @hashval and @data in the inode
1308 * cache, where the helper function @match will return 0 if the inode
1309 * does not match, 1 if the inode does match, and -1 if the search
1310 * should be stopped. The @match function must be responsible for
1311 * taking the i_lock spin_lock and checking i_state for an inode being
1312 * freed or being initialized, and incrementing the reference count
1313 * before returning 1. It also must not sleep, since it is called with
1314 * the inode_hash_lock spinlock held.
1315 *
1316 * This is a even more generalized version of ilookup5() when the
1317 * function must never block --- find_inode() can block in
1318 * __wait_on_freeing_inode() --- or when the caller can not increment
1319 * the reference count because the resulting iput() might cause an
1320 * inode eviction. The tradeoff is that the @match funtion must be
1321 * very carefully implemented.
1322 */
1323struct inode *find_inode_nowait(struct super_block *sb,
1324 unsigned long hashval,
1325 int (*match)(struct inode *, unsigned long,
1326 void *),
1327 void *data)
1328{
1329 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1330 struct inode *inode, *ret_inode = NULL;
1331 int mval;
1332
1333 spin_lock(&inode_hash_lock);
1334 hlist_for_each_entry(inode, head, i_hash) {
1335 if (inode->i_sb != sb)
1336 continue;
1337 mval = match(inode, hashval, data);
1338 if (mval == 0)
1339 continue;
1340 if (mval == 1)
1341 ret_inode = inode;
1342 goto out;
1343 }
1344out:
1345 spin_unlock(&inode_hash_lock);
1346 return ret_inode;
1347}
1348EXPORT_SYMBOL(find_inode_nowait);
1349
Al Viro261bca82008-12-30 01:48:21 -05001350int insert_inode_locked(struct inode *inode)
1351{
1352 struct super_block *sb = inode->i_sb;
1353 ino_t ino = inode->i_ino;
1354 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001355
Al Viro261bca82008-12-30 01:48:21 -05001356 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001357 struct inode *old = NULL;
Dave Chinner67a23c42011-03-22 22:23:42 +11001358 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001359 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001360 if (old->i_ino != ino)
1361 continue;
1362 if (old->i_sb != sb)
1363 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001364 spin_lock(&old->i_lock);
1365 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1366 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001367 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001368 }
Al Viro72a43d62009-05-13 19:13:40 +01001369 break;
1370 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001371 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001372 spin_lock(&inode->i_lock);
1373 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001374 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001375 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001376 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001377 return 0;
1378 }
1379 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001380 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001381 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001382 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001383 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001384 iput(old);
1385 return -EBUSY;
1386 }
1387 iput(old);
1388 }
1389}
Al Viro261bca82008-12-30 01:48:21 -05001390EXPORT_SYMBOL(insert_inode_locked);
1391
1392int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1393 int (*test)(struct inode *, void *), void *data)
1394{
1395 struct super_block *sb = inode->i_sb;
1396 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001397
Al Viro261bca82008-12-30 01:48:21 -05001398 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001399 struct inode *old = NULL;
1400
Dave Chinner67a23c42011-03-22 22:23:42 +11001401 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001402 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001403 if (old->i_sb != sb)
1404 continue;
1405 if (!test(old, data))
1406 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001407 spin_lock(&old->i_lock);
1408 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1409 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001410 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001411 }
Al Viro72a43d62009-05-13 19:13:40 +01001412 break;
1413 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001414 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001415 spin_lock(&inode->i_lock);
1416 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001417 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001418 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001419 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001420 return 0;
1421 }
1422 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001423 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001424 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001425 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001426 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001427 iput(old);
1428 return -EBUSY;
1429 }
1430 iput(old);
1431 }
1432}
Al Viro261bca82008-12-30 01:48:21 -05001433EXPORT_SYMBOL(insert_inode_locked4);
1434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
Al Viro45321ac2010-06-07 13:43:19 -04001436int generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437{
Al Viro45321ac2010-06-07 13:43:19 -04001438 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440EXPORT_SYMBOL(generic_delete_inode);
1441
Al Viro45321ac2010-06-07 13:43:19 -04001442/*
Al Viro45321ac2010-06-07 13:43:19 -04001443 * Called when we're dropping the last reference
1444 * to an inode.
1445 *
1446 * Call the FS "drop_inode()" function, defaulting to
1447 * the legacy UNIX filesystem behaviour. If it tells
1448 * us to evict inode, do so. Otherwise, retain inode
1449 * in cache if fs is alive, sync and evict if fs is
1450 * shutting down.
1451 */
1452static void iput_final(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453{
1454 struct super_block *sb = inode->i_sb;
Al Viro45321ac2010-06-07 13:43:19 -04001455 const struct super_operations *op = inode->i_sb->s_op;
1456 int drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
Dave Chinner250df6e2011-03-22 22:23:36 +11001458 WARN_ON(inode->i_state & I_NEW);
1459
Al Viroe7f59092011-07-07 15:45:59 -04001460 if (op->drop_inode)
Al Viro45321ac2010-06-07 13:43:19 -04001461 drop = op->drop_inode(inode);
1462 else
1463 drop = generic_drop_inode(inode);
1464
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001465 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1466 inode->i_state |= I_REFERENCED;
Jan Kara4eff96d2012-11-26 16:29:51 -08001467 inode_add_lru(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001468 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001469 return;
1470 }
1471
Al Viro45321ac2010-06-07 13:43:19 -04001472 if (!drop) {
Alexander Viro991114c2005-06-23 00:09:01 -07001473 inode->i_state |= I_WILL_FREE;
Dave Chinner250df6e2011-03-22 22:23:36 +11001474 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 write_inode_now(inode, 1);
Dave Chinner250df6e2011-03-22 22:23:36 +11001476 spin_lock(&inode->i_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001477 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001478 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001480
Alexander Viro991114c2005-06-23 00:09:01 -07001481 inode->i_state |= I_FREEING;
Eric Dumazetc4ae0c62011-07-28 06:55:13 +02001482 if (!list_empty(&inode->i_lru))
1483 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001484 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001485
1486 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487}
1488
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301490 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 * @inode: inode to put
1492 *
1493 * Puts an inode, dropping its usage count. If the inode use count hits
1494 * zero, the inode is then freed and may also be destroyed.
1495 *
1496 * Consequently, iput() can sleep.
1497 */
1498void iput(struct inode *inode)
1499{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001500 if (!inode)
1501 return;
1502 BUG_ON(inode->i_state & I_CLEAR);
1503retry:
1504 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1505 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1506 atomic_inc(&inode->i_count);
1507 inode->i_state &= ~I_DIRTY_TIME;
1508 spin_unlock(&inode->i_lock);
1509 trace_writeback_lazytime_iput(inode);
1510 mark_inode_dirty_sync(inode);
1511 goto retry;
1512 }
1513 iput_final(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 }
1515}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516EXPORT_SYMBOL(iput);
1517
1518/**
1519 * bmap - find a block number in a file
1520 * @inode: inode of file
1521 * @block: block to find
1522 *
1523 * Returns the block number on the device holding the inode that
1524 * is the disk block number for the block of the file requested.
1525 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301526 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 * file.
1528 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301529sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530{
1531 sector_t res = 0;
1532 if (inode->i_mapping->a_ops->bmap)
1533 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1534 return res;
1535}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536EXPORT_SYMBOL(bmap);
1537
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001538/*
1539 * With relative atime, only update atime if the previous atime is
1540 * earlier than either the ctime or mtime or if at least a day has
1541 * passed since the last atime update.
1542 */
1543static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1544 struct timespec now)
1545{
1546
1547 if (!(mnt->mnt_flags & MNT_RELATIME))
1548 return 1;
1549 /*
1550 * Is mtime younger than atime? If yes, update atime:
1551 */
1552 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1553 return 1;
1554 /*
1555 * Is ctime younger than atime? If yes, update atime:
1556 */
1557 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1558 return 1;
1559
1560 /*
1561 * Is the previous atime value older than a day? If yes,
1562 * update atime:
1563 */
1564 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1565 return 1;
1566 /*
1567 * Good, we can skip the atime update:
1568 */
1569 return 0;
1570}
1571
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001572int generic_update_time(struct inode *inode, struct timespec *time, int flags)
Josef Bacikc3b2da32012-03-26 09:59:21 -04001573{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001574 int iflags = I_DIRTY_TIME;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001575
1576 if (flags & S_ATIME)
1577 inode->i_atime = *time;
1578 if (flags & S_VERSION)
1579 inode_inc_iversion(inode);
1580 if (flags & S_CTIME)
1581 inode->i_ctime = *time;
1582 if (flags & S_MTIME)
1583 inode->i_mtime = *time;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001584
1585 if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION))
1586 iflags |= I_DIRTY_SYNC;
1587 __mark_inode_dirty(inode, iflags);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001588 return 0;
1589}
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001590EXPORT_SYMBOL(generic_update_time);
1591
1592/*
1593 * This does the actual work of updating an inodes time or version. Must have
1594 * had called mnt_want_write() before calling this.
1595 */
1596static int update_time(struct inode *inode, struct timespec *time, int flags)
1597{
1598 int (*update_time)(struct inode *, struct timespec *, int);
1599
1600 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1601 generic_update_time;
1602
1603 return update_time(inode, time, flags);
1604}
Josef Bacikc3b2da32012-03-26 09:59:21 -04001605
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001607 * touch_atime - update the access time
Randy Dunlap185553b2012-04-17 17:03:25 -07001608 * @path: the &struct path to update
Randy Dunlap30fdc8e2015-11-09 14:57:58 -08001609 * @inode: inode to update
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 *
1611 * Update the accessed time on an inode and mark it for writeback.
1612 * This function automatically handles read only file systems and media,
1613 * as well as the "noatime" flag and inode specific "noatime" markers.
1614 */
NeilBrown8fa9dd22015-03-23 13:37:40 +11001615bool atime_needs_update(const struct path *path, struct inode *inode)
1616{
1617 struct vfsmount *mnt = path->mnt;
1618 struct timespec now;
1619
1620 if (inode->i_flags & S_NOATIME)
1621 return false;
Eric W. Biederman0bd23d092016-06-29 14:54:46 -05001622
1623 /* Atime updates will likely cause i_uid and i_gid to be written
1624 * back improprely if their true value is unknown to the vfs.
1625 */
1626 if (HAS_UNMAPPED_ID(inode))
1627 return false;
1628
NeilBrown8fa9dd22015-03-23 13:37:40 +11001629 if (IS_NOATIME(inode))
1630 return false;
1631 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1632 return false;
1633
1634 if (mnt->mnt_flags & MNT_NOATIME)
1635 return false;
1636 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1637 return false;
1638
1639 now = current_fs_time(inode->i_sb);
1640
1641 if (!relatime_need_update(mnt, inode, now))
1642 return false;
1643
1644 if (timespec_equal(&inode->i_atime, &now))
1645 return false;
1646
1647 return true;
1648}
1649
Al Virobadcf2b2013-07-16 18:15:46 +04001650void touch_atime(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651{
Al Viro68ac1232012-03-15 08:21:57 -04001652 struct vfsmount *mnt = path->mnt;
David Howellsdf2b1af2015-03-17 22:26:15 +00001653 struct inode *inode = d_inode(path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 struct timespec now;
1655
NeilBrown8fa9dd22015-03-23 13:37:40 +11001656 if (!atime_needs_update(path, inode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001657 return;
1658
Jan Kara5d37e9e2012-06-12 16:20:36 +02001659 if (!sb_start_write_trylock(inode->i_sb))
Andi Kleenb12536c2009-09-18 13:05:47 -07001660 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001661
NeilBrown8fa9dd22015-03-23 13:37:40 +11001662 if (__mnt_want_write(mnt) != 0)
Jan Kara5d37e9e2012-06-12 16:20:36 +02001663 goto skip_update;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001664 /*
1665 * File systems can error out when updating inodes if they need to
1666 * allocate new space to modify an inode (such is the case for
1667 * Btrfs), but since we touch atime while walking down the path we
1668 * really don't care if we failed to update the atime of the file,
1669 * so just ignore the return value.
Alexander Block2bc5565282012-06-15 09:49:33 +02001670 * We may also fail on filesystems that have the ability to make parts
1671 * of the fs read only, e.g. subvolumes in Btrfs.
Josef Bacikc3b2da32012-03-26 09:59:21 -04001672 */
NeilBrown8fa9dd22015-03-23 13:37:40 +11001673 now = current_fs_time(inode->i_sb);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001674 update_time(inode, &now, S_ATIME);
Jan Kara5d37e9e2012-06-12 16:20:36 +02001675 __mnt_drop_write(mnt);
1676skip_update:
1677 sb_end_write(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001679EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
Cong Wang3ed37642012-05-15 14:57:33 +08001681/*
1682 * The logic we want is
1683 *
1684 * if suid or (sgid and xgrp)
1685 * remove privs
1686 */
1687int should_remove_suid(struct dentry *dentry)
1688{
David Howellsdf2b1af2015-03-17 22:26:15 +00001689 umode_t mode = d_inode(dentry)->i_mode;
Cong Wang3ed37642012-05-15 14:57:33 +08001690 int kill = 0;
1691
1692 /* suid always must be killed */
1693 if (unlikely(mode & S_ISUID))
1694 kill = ATTR_KILL_SUID;
1695
1696 /*
1697 * sgid without any exec bits is just a mandatory locking mark; leave
1698 * it alone. If some exec bits are set, it's a real sgid; kill it.
1699 */
1700 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1701 kill |= ATTR_KILL_SGID;
1702
1703 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1704 return kill;
1705
1706 return 0;
1707}
1708EXPORT_SYMBOL(should_remove_suid);
1709
Jan Karadbfae0c2015-05-21 16:05:54 +02001710/*
1711 * Return mask of changes for notify_change() that need to be done as a
1712 * response to write or truncate. Return 0 if nothing has to be changed.
1713 * Negative value on error (change should be denied).
1714 */
Jan Kara45f147a2015-05-21 16:05:55 +02001715int dentry_needs_remove_privs(struct dentry *dentry)
Jan Karadbfae0c2015-05-21 16:05:54 +02001716{
Jan Karadbfae0c2015-05-21 16:05:54 +02001717 struct inode *inode = d_inode(dentry);
1718 int mask = 0;
1719 int ret;
1720
1721 if (IS_NOSEC(inode))
1722 return 0;
1723
1724 mask = should_remove_suid(dentry);
1725 ret = security_inode_need_killpriv(dentry);
1726 if (ret < 0)
1727 return ret;
1728 if (ret)
1729 mask |= ATTR_KILL_PRIV;
1730 return mask;
1731}
Jan Karadbfae0c2015-05-21 16:05:54 +02001732
1733static int __remove_privs(struct dentry *dentry, int kill)
Cong Wang3ed37642012-05-15 14:57:33 +08001734{
1735 struct iattr newattrs;
1736
1737 newattrs.ia_valid = ATTR_FORCE | kill;
J. Bruce Fields27ac0ff2011-09-20 17:19:26 -04001738 /*
1739 * Note we call this on write, so notify_change will not
1740 * encounter any conflicting delegations:
1741 */
1742 return notify_change(dentry, &newattrs, NULL);
Cong Wang3ed37642012-05-15 14:57:33 +08001743}
1744
Jan Kara5fa8e0a2015-05-21 16:05:53 +02001745/*
1746 * Remove special file priviledges (suid, capabilities) when file is written
1747 * to or truncated.
1748 */
1749int file_remove_privs(struct file *file)
Cong Wang3ed37642012-05-15 14:57:33 +08001750{
Miklos Szeredic1892c32016-08-03 13:44:27 +02001751 struct dentry *dentry = file_dentry(file);
1752 struct inode *inode = file_inode(file);
Jan Karadbfae0c2015-05-21 16:05:54 +02001753 int kill;
Cong Wang3ed37642012-05-15 14:57:33 +08001754 int error = 0;
1755
1756 /* Fast path for nothing security related */
1757 if (IS_NOSEC(inode))
1758 return 0;
1759
Miklos Szeredic1892c32016-08-03 13:44:27 +02001760 kill = dentry_needs_remove_privs(dentry);
Jan Karadbfae0c2015-05-21 16:05:54 +02001761 if (kill < 0)
1762 return kill;
1763 if (kill)
1764 error = __remove_privs(dentry, kill);
Jan Kara2426f392015-05-21 16:05:52 +02001765 if (!error)
1766 inode_has_no_xattr(inode);
Cong Wang3ed37642012-05-15 14:57:33 +08001767
1768 return error;
1769}
Jan Kara5fa8e0a2015-05-21 16:05:53 +02001770EXPORT_SYMBOL(file_remove_privs);
Cong Wang3ed37642012-05-15 14:57:33 +08001771
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001773 * file_update_time - update mtime and ctime time
1774 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001776 * Update the mtime and ctime members of an inode and mark the inode
1777 * for writeback. Note that this function is meant exclusively for
1778 * usage in the file write path of filesystems, and filesystems may
1779 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001780 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Josef Bacikc3b2da32012-03-26 09:59:21 -04001781 * timestamps are handled by the server. This can return an error for
1782 * file systems who need to allocate space in order to update an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 */
1784
Josef Bacikc3b2da32012-03-26 09:59:21 -04001785int file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786{
Al Viro496ad9a2013-01-23 17:07:38 -05001787 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 struct timespec now;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001789 int sync_it = 0;
1790 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
Andi Kleence06e0b2009-09-18 13:05:48 -07001792 /* First try to exhaust all avenues to not sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 if (IS_NOCMTIME(inode))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001794 return 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001795
Andi Kleence06e0b2009-09-18 13:05:48 -07001796 now = current_fs_time(inode->i_sb);
1797 if (!timespec_equal(&inode->i_mtime, &now))
1798 sync_it = S_MTIME;
1799
1800 if (!timespec_equal(&inode->i_ctime, &now))
1801 sync_it |= S_CTIME;
1802
1803 if (IS_I_VERSION(inode))
1804 sync_it |= S_VERSION;
1805
1806 if (!sync_it)
Josef Bacikc3b2da32012-03-26 09:59:21 -04001807 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
Andi Kleence06e0b2009-09-18 13:05:48 -07001809 /* Finally allowed to write? Takes lock. */
Jan Karaeb04c282012-06-12 16:20:35 +02001810 if (__mnt_want_write_file(file))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001811 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
Josef Bacikc3b2da32012-03-26 09:59:21 -04001813 ret = update_time(inode, &now, sync_it);
Jan Karaeb04c282012-06-12 16:20:35 +02001814 __mnt_drop_write_file(file);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001815
1816 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001818EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819
1820int inode_needs_sync(struct inode *inode)
1821{
1822 if (IS_SYNC(inode))
1823 return 1;
1824 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1825 return 1;
1826 return 0;
1827}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828EXPORT_SYMBOL(inode_needs_sync);
1829
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001831 * If we try to find an inode in the inode hash while it is being
1832 * deleted, we have to wait until the filesystem completes its
1833 * deletion before reporting that it isn't found. This function waits
1834 * until the deletion _might_ have completed. Callers are responsible
1835 * to recheck inode state.
1836 *
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001837 * It doesn't matter if I_NEW is not set initially, a call to
Dave Chinner250df6e2011-03-22 22:23:36 +11001838 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1839 * will DTRT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 */
1841static void __wait_on_freeing_inode(struct inode *inode)
1842{
1843 wait_queue_head_t *wq;
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001844 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1845 wq = bit_waitqueue(&inode->i_state, __I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +11001847 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001848 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 schedule();
1850 finish_wait(wq, &wait.wait);
Dave Chinner67a23c42011-03-22 22:23:42 +11001851 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852}
1853
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854static __initdata unsigned long ihash_entries;
1855static int __init set_ihash_entries(char *str)
1856{
1857 if (!str)
1858 return 0;
1859 ihash_entries = simple_strtoul(str, &str, 0);
1860 return 1;
1861}
1862__setup("ihash_entries=", set_ihash_entries);
1863
1864/*
1865 * Initialize the waitqueues and inode hash table.
1866 */
1867void __init inode_init_early(void)
1868{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001869 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
1871 /* If hashes are distributed across NUMA nodes, defer
1872 * hash allocation until vmalloc space is available.
1873 */
1874 if (hashdist)
1875 return;
1876
1877 inode_hashtable =
1878 alloc_large_system_hash("Inode-cache",
1879 sizeof(struct hlist_head),
1880 ihash_entries,
1881 14,
1882 HASH_EARLY,
1883 &i_hash_shift,
1884 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001885 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 0);
1887
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001888 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1890}
1891
Denis Cheng74bf17c2007-10-16 23:26:30 -07001892void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001894 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
1896 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001897 inode_cachep = kmem_cache_create("inode_cache",
1898 sizeof(struct inode),
1899 0,
1900 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
Vladimir Davydov5d097052016-01-14 15:18:21 -08001901 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
Paul Mundt20c2df82007-07-20 10:11:58 +09001902 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
1904 /* Hash may have been set up in inode_init_early */
1905 if (!hashdist)
1906 return;
1907
1908 inode_hashtable =
1909 alloc_large_system_hash("Inode-cache",
1910 sizeof(struct hlist_head),
1911 ihash_entries,
1912 14,
1913 0,
1914 &i_hash_shift,
1915 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001916 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 0);
1918
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001919 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1921}
1922
1923void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1924{
1925 inode->i_mode = mode;
1926 if (S_ISCHR(mode)) {
1927 inode->i_fop = &def_chr_fops;
1928 inode->i_rdev = rdev;
1929 } else if (S_ISBLK(mode)) {
1930 inode->i_fop = &def_blk_fops;
1931 inode->i_rdev = rdev;
1932 } else if (S_ISFIFO(mode))
Al Viro599a0ac2013-03-12 09:58:10 -04001933 inode->i_fop = &pipefifo_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 else if (S_ISSOCK(mode))
Al Virobd9b51e2014-11-18 23:38:21 -05001935 ; /* leave it no_open_fops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 else
Manish Katiyaraf0d9ae2009-09-18 13:05:43 -07001937 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1938 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1939 inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940}
1941EXPORT_SYMBOL(init_special_inode);
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001942
1943/**
Ben Hutchingseaae668d2011-02-15 12:48:09 +00001944 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001945 * @inode: New inode
1946 * @dir: Directory inode
1947 * @mode: mode of the new inode
1948 */
1949void inode_init_owner(struct inode *inode, const struct inode *dir,
Al Viro62bb1092011-07-24 23:20:18 -04001950 umode_t mode)
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001951{
1952 inode->i_uid = current_fsuid();
1953 if (dir && dir->i_mode & S_ISGID) {
1954 inode->i_gid = dir->i_gid;
1955 if (S_ISDIR(mode))
1956 mode |= S_ISGID;
1957 } else
1958 inode->i_gid = current_fsgid();
1959 inode->i_mode = mode;
1960}
1961EXPORT_SYMBOL(inode_init_owner);
Serge E. Hallyne795b712011-03-23 16:43:25 -07001962
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001963/**
1964 * inode_owner_or_capable - check current task permissions to inode
1965 * @inode: inode being checked
1966 *
Andy Lutomirski23adbe12014-06-10 12:45:42 -07001967 * Return true if current either has CAP_FOWNER in a namespace with the
1968 * inode owner uid mapped, or owns the file.
Serge E. Hallyne795b712011-03-23 16:43:25 -07001969 */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001970bool inode_owner_or_capable(const struct inode *inode)
Serge E. Hallyne795b712011-03-23 16:43:25 -07001971{
Andy Lutomirski23adbe12014-06-10 12:45:42 -07001972 struct user_namespace *ns;
1973
Eric W. Biederman92361632012-02-08 07:07:50 -08001974 if (uid_eq(current_fsuid(), inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001975 return true;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07001976
1977 ns = current_user_ns();
1978 if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001979 return true;
1980 return false;
1981}
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001982EXPORT_SYMBOL(inode_owner_or_capable);
Trond Myklebust1d59d612012-05-31 12:22:33 -04001983
1984/*
1985 * Direct i/o helper functions
1986 */
1987static void __inode_dio_wait(struct inode *inode)
1988{
1989 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1990 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1991
1992 do {
1993 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1994 if (atomic_read(&inode->i_dio_count))
1995 schedule();
1996 } while (atomic_read(&inode->i_dio_count));
1997 finish_wait(wq, &q.wait);
1998}
1999
2000/**
2001 * inode_dio_wait - wait for outstanding DIO requests to finish
2002 * @inode: inode to wait for
2003 *
2004 * Waits for all pending direct I/O requests to finish so that we can
2005 * proceed with a truncate or equivalent operation.
2006 *
2007 * Must be called under a lock that serializes taking new references
2008 * to i_dio_count, usually by inode->i_mutex.
2009 */
2010void inode_dio_wait(struct inode *inode)
2011{
2012 if (atomic_read(&inode->i_dio_count))
2013 __inode_dio_wait(inode);
2014}
2015EXPORT_SYMBOL(inode_dio_wait);
2016
2017/*
Theodore Ts'o5f16f322014-03-24 14:43:12 -04002018 * inode_set_flags - atomically set some inode flags
2019 *
2020 * Note: the caller should be holding i_mutex, or else be sure that
2021 * they have exclusive access to the inode structure (i.e., while the
2022 * inode is being instantiated). The reason for the cmpxchg() loop
2023 * --- which wouldn't be necessary if all code paths which modify
2024 * i_flags actually followed this rule, is that there is at least one
Jan Kara5fa8e0a2015-05-21 16:05:53 +02002025 * code path which doesn't today so we use cmpxchg() out of an abundance
2026 * of caution.
Theodore Ts'o5f16f322014-03-24 14:43:12 -04002027 *
2028 * In the long run, i_mutex is overkill, and we should probably look
2029 * at using the i_lock spinlock to protect i_flags, and then make sure
2030 * it is so documented in include/linux/fs.h and that all code follows
2031 * the locking convention!!
2032 */
2033void inode_set_flags(struct inode *inode, unsigned int flags,
2034 unsigned int mask)
2035{
2036 unsigned int old_flags, new_flags;
2037
2038 WARN_ON_ONCE(flags & ~mask);
2039 do {
2040 old_flags = ACCESS_ONCE(inode->i_flags);
2041 new_flags = (old_flags & ~mask) | flags;
2042 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2043 new_flags) != old_flags));
2044}
2045EXPORT_SYMBOL(inode_set_flags);
Al Viro21fc61c2015-11-17 01:07:57 -05002046
2047void inode_nohighmem(struct inode *inode)
2048{
2049 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2050}
2051EXPORT_SYMBOL(inode_nohighmem);