blob: ea37cd17b53f0c98b47e2e626b9ff2a9b6e5699b [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 Chinner55fa6092011-03-22 22:23:40 +110031 * inode_sb_list_lock protects:
32 * sb->s_inodes, inode->i_sb_list
Christoph Hellwigf758eea2011-04-21 18:19:44 -060033 * bdi->wb.list_lock protects:
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050034 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_wb_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 *
40 * inode_sb_list_lock
41 * 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
48 * inode_sb_list_lock
49 * 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
Dave Chinner55fa6092011-03-22 22:23:40 +110060__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
61
Linus Torvalds1da177e2005-04-16 15:20:36 -070062/*
Jens Axboe7dcda1c2011-04-05 23:51:48 +020063 * Empty aops. Can be used for the cases where the user does not
64 * define any of the address_space operations.
65 */
66const struct address_space_operations empty_aops = {
67};
68EXPORT_SYMBOL(empty_aops);
69
70/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 * Statistics gathering..
72 */
73struct inodes_stat_t inodes_stat;
74
Glauber Costa3942c072013-08-28 10:17:53 +100075static DEFINE_PER_CPU(unsigned long, nr_inodes);
76static DEFINE_PER_CPU(unsigned long, nr_unused);
Dave Chinnercffbc8a2010-10-23 05:03:02 -040077
Manish Katiyar6b3304b2009-03-31 19:35:54 +053078static struct kmem_cache *inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
Glauber Costa3942c072013-08-28 10:17:53 +100080static long get_nr_inodes(void)
Dave Chinnercffbc8a2010-10-23 05:03:02 -040081{
Nick Piggin3e880fb2011-01-07 17:49:19 +110082 int i;
Glauber Costa3942c072013-08-28 10:17:53 +100083 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +110084 for_each_possible_cpu(i)
85 sum += per_cpu(nr_inodes, i);
86 return sum < 0 ? 0 : sum;
Dave Chinnercffbc8a2010-10-23 05:03:02 -040087}
88
Glauber Costa3942c072013-08-28 10:17:53 +100089static inline long get_nr_inodes_unused(void)
Dave Chinnercffbc8a2010-10-23 05:03:02 -040090{
Dave Chinnerfcb94f72011-07-08 14:14:38 +100091 int i;
Glauber Costa3942c072013-08-28 10:17:53 +100092 long sum = 0;
Dave Chinnerfcb94f72011-07-08 14:14:38 +100093 for_each_possible_cpu(i)
94 sum += per_cpu(nr_unused, i);
95 return sum < 0 ? 0 : sum;
Dave Chinnercffbc8a2010-10-23 05:03:02 -040096}
97
Glauber Costa3942c072013-08-28 10:17:53 +100098long get_nr_dirty_inodes(void)
Dave Chinnercffbc8a2010-10-23 05:03:02 -040099{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100100 /* not actually dirty inodes, but a wild approximation */
Glauber Costa3942c072013-08-28 10:17:53 +1000101 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400102 return nr_dirty > 0 ? nr_dirty : 0;
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400103}
104
105/*
106 * Handle nr_inode sysctl
107 */
108#ifdef CONFIG_SYSCTL
Joe Perches1f7e0612014-06-06 14:38:05 -0700109int proc_nr_inodes(struct ctl_table *table, int write,
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400110 void __user *buffer, size_t *lenp, loff_t *ppos)
111{
112 inodes_stat.nr_inodes = get_nr_inodes();
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000113 inodes_stat.nr_unused = get_nr_inodes_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000114 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400115}
116#endif
117
Al Virobd9b51e2014-11-18 23:38:21 -0500118static int no_open(struct inode *inode, struct file *file)
119{
120 return -ENXIO;
121}
122
David Chinner2cb15992008-10-30 17:32:23 +1100123/**
124 * inode_init_always - perform inode structure intialisation
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800125 * @sb: superblock inode belongs to
126 * @inode: inode to initialise
David Chinner2cb15992008-10-30 17:32:23 +1100127 *
128 * These are initializations that need to be done on every inode
129 * allocation as the fields are not initialised by slab allocation.
130 */
Christoph Hellwig54e34622009-08-07 14:38:25 -0300131int inode_init_always(struct super_block *sb, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132{
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700133 static const struct inode_operations empty_iops;
Al Virobd9b51e2014-11-18 23:38:21 -0500134 static const struct file_operations no_open_fops = {.open = no_open};
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530135 struct address_space *const mapping = &inode->i_data;
David Chinner2cb15992008-10-30 17:32:23 +1100136
137 inode->i_sb = sb;
138 inode->i_blkbits = sb->s_blocksize_bits;
139 inode->i_flags = 0;
140 atomic_set(&inode->i_count, 1);
141 inode->i_op = &empty_iops;
Al Virobd9b51e2014-11-18 23:38:21 -0500142 inode->i_fop = &no_open_fops;
Miklos Szeredia78ef702011-10-28 14:13:30 +0200143 inode->__i_nlink = 1;
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700144 inode->i_opflags = 0;
Eric W. Biederman92361632012-02-08 07:07:50 -0800145 i_uid_write(inode, 0);
146 i_gid_write(inode, 0);
David Chinner2cb15992008-10-30 17:32:23 +1100147 atomic_set(&inode->i_writecount, 0);
148 inode->i_size = 0;
149 inode->i_blocks = 0;
150 inode->i_bytes = 0;
151 inode->i_generation = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100152 inode->i_pipe = NULL;
153 inode->i_bdev = NULL;
154 inode->i_cdev = NULL;
155 inode->i_rdev = 0;
156 inode->dirtied_when = 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500157
158 if (security_inode_alloc(inode))
Christoph Hellwig54e34622009-08-07 14:38:25 -0300159 goto out;
David Chinner2cb15992008-10-30 17:32:23 +1100160 spin_lock_init(&inode->i_lock);
161 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
162
163 mutex_init(&inode->i_mutex);
164 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
165
Christoph Hellwigbd5fe6c2011-06-24 14:29:43 -0400166 atomic_set(&inode->i_dio_count, 0);
David Chinner2cb15992008-10-30 17:32:23 +1100167
168 mapping->a_ops = &empty_aops;
169 mapping->host = inode;
170 mapping->flags = 0;
David Herrmann4bb5f5d2014-08-08 14:25:25 -0700171 atomic_set(&mapping->i_mmap_writable, 0);
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800172 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800173 mapping->private_data = NULL;
David Chinner2cb15992008-10-30 17:32:23 +1100174 mapping->writeback_index = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100175 inode->i_private = NULL;
176 inode->i_mapping = mapping;
Al Virob3d9b7a2012-06-09 13:51:19 -0400177 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
Al Virof19d4a82009-06-08 19:50:45 -0400178#ifdef CONFIG_FS_POSIX_ACL
179 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
180#endif
David Chinner2cb15992008-10-30 17:32:23 +1100181
Eric Paris3be25f42009-05-21 17:01:26 -0400182#ifdef CONFIG_FSNOTIFY
183 inode->i_fsnotify_mask = 0;
184#endif
Jeff Layton4a075e32015-01-16 15:05:54 -0500185 inode->i_flctx = NULL;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100186 this_cpu_inc(nr_inodes);
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400187
Christoph Hellwig54e34622009-08-07 14:38:25 -0300188 return 0;
Christoph Hellwig54e34622009-08-07 14:38:25 -0300189out:
190 return -ENOMEM;
David Chinner2cb15992008-10-30 17:32:23 +1100191}
192EXPORT_SYMBOL(inode_init_always);
193
194static struct inode *alloc_inode(struct super_block *sb)
195{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 struct inode *inode;
197
198 if (sb->s_op->alloc_inode)
199 inode = sb->s_op->alloc_inode(sb);
200 else
David Chinner2cb15992008-10-30 17:32:23 +1100201 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
Christoph Hellwig54e34622009-08-07 14:38:25 -0300203 if (!inode)
204 return NULL;
205
206 if (unlikely(inode_init_always(sb, inode))) {
207 if (inode->i_sb->s_op->destroy_inode)
208 inode->i_sb->s_op->destroy_inode(inode);
209 else
210 kmem_cache_free(inode_cachep, inode);
211 return NULL;
212 }
213
214 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215}
216
Nick Pigginff0c7d12011-01-07 17:49:50 +1100217void free_inode_nonrcu(struct inode *inode)
218{
219 kmem_cache_free(inode_cachep, inode);
220}
221EXPORT_SYMBOL(free_inode_nonrcu);
222
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300223void __destroy_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200225 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 security_inode_free(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400227 fsnotify_inode_delete(inode);
Jeff Layton4a075e32015-01-16 15:05:54 -0500228 locks_free_lock_context(inode->i_flctx);
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100229 if (!inode->i_nlink) {
230 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
231 atomic_long_dec(&inode->i_sb->s_remove_count);
232 }
233
Al Virof19d4a82009-06-08 19:50:45 -0400234#ifdef CONFIG_FS_POSIX_ACL
235 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
236 posix_acl_release(inode->i_acl);
237 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
238 posix_acl_release(inode->i_default_acl);
239#endif
Nick Piggin3e880fb2011-01-07 17:49:19 +1100240 this_cpu_dec(nr_inodes);
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300241}
242EXPORT_SYMBOL(__destroy_inode);
243
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100244static void i_callback(struct rcu_head *head)
245{
246 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100247 kmem_cache_free(inode_cachep, inode);
248}
249
Christoph Hellwig56b0dac2010-10-06 10:48:55 +0200250static void destroy_inode(struct inode *inode)
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300251{
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100252 BUG_ON(!list_empty(&inode->i_lru));
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300253 __destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 if (inode->i_sb->s_op->destroy_inode)
255 inode->i_sb->s_op->destroy_inode(inode);
256 else
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100257 call_rcu(&inode->i_rcu, i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100260/**
261 * drop_nlink - directly drop an inode's link count
262 * @inode: inode
263 *
264 * This is a low-level filesystem helper to replace any
265 * direct filesystem manipulation of i_nlink. In cases
266 * where we are attempting to track writes to the
267 * filesystem, a decrement to zero means an imminent
268 * write when the file is truncated and actually unlinked
269 * on the filesystem.
270 */
271void drop_nlink(struct inode *inode)
272{
273 WARN_ON(inode->i_nlink == 0);
274 inode->__i_nlink--;
275 if (!inode->i_nlink)
276 atomic_long_inc(&inode->i_sb->s_remove_count);
277}
278EXPORT_SYMBOL(drop_nlink);
279
280/**
281 * clear_nlink - directly zero an inode's link count
282 * @inode: inode
283 *
284 * This is a low-level filesystem helper to replace any
285 * direct filesystem manipulation of i_nlink. See
286 * drop_nlink() for why we care about i_nlink hitting zero.
287 */
288void clear_nlink(struct inode *inode)
289{
290 if (inode->i_nlink) {
291 inode->__i_nlink = 0;
292 atomic_long_inc(&inode->i_sb->s_remove_count);
293 }
294}
295EXPORT_SYMBOL(clear_nlink);
296
297/**
298 * set_nlink - directly set an inode's link count
299 * @inode: inode
300 * @nlink: new nlink (should be non-zero)
301 *
302 * This is a low-level filesystem helper to replace any
303 * direct filesystem manipulation of i_nlink.
304 */
305void set_nlink(struct inode *inode, unsigned int nlink)
306{
307 if (!nlink) {
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100308 clear_nlink(inode);
309 } else {
310 /* Yes, some filesystems do change nlink from zero to one */
311 if (inode->i_nlink == 0)
312 atomic_long_dec(&inode->i_sb->s_remove_count);
313
314 inode->__i_nlink = nlink;
315 }
316}
317EXPORT_SYMBOL(set_nlink);
318
319/**
320 * inc_nlink - directly increment an inode's link count
321 * @inode: inode
322 *
323 * This is a low-level filesystem helper to replace any
324 * direct filesystem manipulation of i_nlink. Currently,
325 * it is only here for parity with dec_nlink().
326 */
327void inc_nlink(struct inode *inode)
328{
Al Virof4e0c302013-06-11 08:34:36 +0400329 if (unlikely(inode->i_nlink == 0)) {
330 WARN_ON(!(inode->i_state & I_LINKABLE));
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100331 atomic_long_dec(&inode->i_sb->s_remove_count);
Al Virof4e0c302013-06-11 08:34:36 +0400332 }
Miklos Szeredi7ada4db2011-11-21 12:11:32 +0100333
334 inode->__i_nlink++;
335}
336EXPORT_SYMBOL(inc_nlink);
337
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100338void address_space_init_once(struct address_space *mapping)
339{
340 memset(mapping, 0, sizeof(*mapping));
341 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
342 spin_lock_init(&mapping->tree_lock);
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800343 init_rwsem(&mapping->i_mmap_rwsem);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100344 INIT_LIST_HEAD(&mapping->private_list);
345 spin_lock_init(&mapping->private_lock);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700346 mapping->i_mmap = RB_ROOT;
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100347}
348EXPORT_SYMBOL(address_space_init_once);
349
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350/*
351 * These are initializations that only need to be done
352 * once, because the fields are idempotent across use
353 * of the inode, so let the slab aware of that.
354 */
355void inode_init_once(struct inode *inode)
356{
357 memset(inode, 0, sizeof(*inode));
358 INIT_HLIST_NODE(&inode->i_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 INIT_LIST_HEAD(&inode->i_devices);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100360 INIT_LIST_HEAD(&inode->i_wb_list);
361 INIT_LIST_HEAD(&inode->i_lru);
Miklos Szeredi2aa15892011-02-23 13:49:47 +0100362 address_space_init_once(&inode->i_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 i_size_ordered_init(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400364#ifdef CONFIG_FSNOTIFY
Eric Parise61ce862009-12-17 21:24:24 -0500365 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
Eric Paris3be25f42009-05-21 17:01:26 -0400366#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368EXPORT_SYMBOL(inode_init_once);
369
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700370static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530372 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
Christoph Lametera35afb82007-05-16 22:10:57 -0700374 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375}
376
377/*
Dave Chinner250df6e2011-03-22 22:23:36 +1100378 * inode->i_lock must be held
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530380void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381{
Nick Piggin9e38d862010-10-23 06:55:17 -0400382 atomic_inc(&inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
384
Al Viro7de9c6ee2010-10-23 11:11:40 -0400385/*
386 * get additional reference to inode; caller must already hold one.
387 */
388void ihold(struct inode *inode)
389{
390 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
391}
392EXPORT_SYMBOL(ihold);
393
Nick Piggin9e38d862010-10-23 06:55:17 -0400394static void inode_lru_list_add(struct inode *inode)
395{
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000396 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000397 this_cpu_inc(nr_unused);
Nick Piggin9e38d862010-10-23 06:55:17 -0400398}
399
Jan Kara4eff96d2012-11-26 16:29:51 -0800400/*
401 * Add inode to LRU if needed (inode is unused and clean).
402 *
403 * Needs inode->i_lock held.
404 */
405void inode_add_lru(struct inode *inode)
406{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500407 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
408 I_FREEING | I_WILL_FREE)) &&
Jan Kara4eff96d2012-11-26 16:29:51 -0800409 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
410 inode_lru_list_add(inode);
411}
412
413
Nick Piggin9e38d862010-10-23 06:55:17 -0400414static void inode_lru_list_del(struct inode *inode)
415{
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000416
417 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
Dave Chinnerfcb94f72011-07-08 14:14:38 +1000418 this_cpu_dec(nr_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419}
420
Christoph Hellwig646ec462010-10-23 07:15:32 -0400421/**
422 * inode_sb_list_add - add inode to the superblock list of inodes
423 * @inode: inode to add
424 */
425void inode_sb_list_add(struct inode *inode)
426{
Dave Chinner55fa6092011-03-22 22:23:40 +1100427 spin_lock(&inode_sb_list_lock);
428 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
429 spin_unlock(&inode_sb_list_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400430}
431EXPORT_SYMBOL_GPL(inode_sb_list_add);
432
Dave Chinner55fa6092011-03-22 22:23:40 +1100433static inline void inode_sb_list_del(struct inode *inode)
Christoph Hellwig646ec462010-10-23 07:15:32 -0400434{
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200435 if (!list_empty(&inode->i_sb_list)) {
436 spin_lock(&inode_sb_list_lock);
437 list_del_init(&inode->i_sb_list);
438 spin_unlock(&inode_sb_list_lock);
439 }
Christoph Hellwig646ec462010-10-23 07:15:32 -0400440}
441
Dave Chinner4c51acb2010-10-23 06:58:09 -0400442static unsigned long hash(struct super_block *sb, unsigned long hashval)
443{
444 unsigned long tmp;
445
446 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
447 L1_CACHE_BYTES;
Christoph Hellwig4b4563d2011-05-27 09:28:01 -0400448 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
449 return tmp & i_hash_mask;
Dave Chinner4c51acb2010-10-23 06:58:09 -0400450}
451
452/**
453 * __insert_inode_hash - hash an inode
454 * @inode: unhashed inode
455 * @hashval: unsigned long value used to locate this object in the
456 * inode_hashtable.
457 *
458 * Add an inode to the inode hash for this superblock.
459 */
460void __insert_inode_hash(struct inode *inode, unsigned long hashval)
461{
Christoph Hellwig646ec462010-10-23 07:15:32 -0400462 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
463
Dave Chinner67a23c42011-03-22 22:23:42 +1100464 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100465 spin_lock(&inode->i_lock);
Christoph Hellwig646ec462010-10-23 07:15:32 -0400466 hlist_add_head(&inode->i_hash, b);
Dave Chinner250df6e2011-03-22 22:23:36 +1100467 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100468 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400469}
470EXPORT_SYMBOL(__insert_inode_hash);
471
472/**
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200473 * __remove_inode_hash - remove an inode from the hash
Dave Chinner4c51acb2010-10-23 06:58:09 -0400474 * @inode: inode to unhash
475 *
476 * Remove an inode from the superblock.
477 */
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200478void __remove_inode_hash(struct inode *inode)
Dave Chinner4c51acb2010-10-23 06:58:09 -0400479{
Dave Chinner67a23c42011-03-22 22:23:42 +1100480 spin_lock(&inode_hash_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100481 spin_lock(&inode->i_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400482 hlist_del_init(&inode->i_hash);
Dave Chinner250df6e2011-03-22 22:23:36 +1100483 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +1100484 spin_unlock(&inode_hash_lock);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400485}
Eric Dumazetf2ee7ab2011-07-28 06:41:09 +0200486EXPORT_SYMBOL(__remove_inode_hash);
Dave Chinner4c51acb2010-10-23 06:58:09 -0400487
Jan Karadbd57682012-05-03 14:48:02 +0200488void clear_inode(struct inode *inode)
Al Virob0683aa2010-06-04 20:55:25 -0400489{
490 might_sleep();
Jan Kara08142572011-06-27 16:18:10 -0700491 /*
492 * We have to cycle tree_lock here because reclaim can be still in the
493 * process of removing the last page (in __delete_from_page_cache())
494 * and we must not free mapping under it.
495 */
496 spin_lock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400497 BUG_ON(inode->i_data.nrpages);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700498 BUG_ON(inode->i_data.nrshadows);
Jan Kara08142572011-06-27 16:18:10 -0700499 spin_unlock_irq(&inode->i_data.tree_lock);
Al Virob0683aa2010-06-04 20:55:25 -0400500 BUG_ON(!list_empty(&inode->i_data.private_list));
501 BUG_ON(!(inode->i_state & I_FREEING));
502 BUG_ON(inode->i_state & I_CLEAR);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100503 /* don't need i_lock here, no concurrent mods to i_state */
Al Virob0683aa2010-06-04 20:55:25 -0400504 inode->i_state = I_FREEING | I_CLEAR;
505}
Jan Karadbd57682012-05-03 14:48:02 +0200506EXPORT_SYMBOL(clear_inode);
Al Virob0683aa2010-06-04 20:55:25 -0400507
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100508/*
509 * Free the inode passed in, removing it from the lists it is still connected
510 * to. We remove any pages still attached to the inode and wait for any IO that
511 * is still in progress before finally destroying the inode.
512 *
513 * An inode must already be marked I_FREEING so that we avoid the inode being
514 * moved back onto lists if we race with other code that manipulates the lists
515 * (e.g. writeback_single_inode). The caller is responsible for setting this.
516 *
517 * An inode must already be removed from the LRU list before being evicted from
518 * the cache. This should occur atomically with setting the I_FREEING state
519 * flag, so no inodes here should ever be on the LRU when being evicted.
520 */
Al Viro644da592010-06-07 13:21:05 -0400521static void evict(struct inode *inode)
Al Virob4272d42010-06-04 19:33:20 -0400522{
523 const struct super_operations *op = inode->i_sb->s_op;
524
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100525 BUG_ON(!(inode->i_state & I_FREEING));
526 BUG_ON(!list_empty(&inode->i_lru));
527
Eric Dumazetb12362b2011-07-28 06:11:47 +0200528 if (!list_empty(&inode->i_wb_list))
529 inode_wb_list_del(inode);
530
Dave Chinner55fa6092011-03-22 22:23:40 +1100531 inode_sb_list_del(inode);
532
Jan Kara169ebd92012-05-03 14:48:03 +0200533 /*
534 * Wait for flusher thread to be done with the inode so that filesystem
535 * does not start destroying it while writeback is still running. Since
536 * the inode has I_FREEING set, flusher thread won't start new work on
537 * the inode. We just have to wait for running writeback to finish.
538 */
539 inode_wait_for_writeback(inode);
Jan Kara7994e6f2012-05-03 14:48:01 +0200540
Al Virobe7ce412010-06-04 19:40:39 -0400541 if (op->evict_inode) {
542 op->evict_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400543 } else {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700544 truncate_inode_pages_final(&inode->i_data);
Jan Karadbd57682012-05-03 14:48:02 +0200545 clear_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400546 }
Al Viro661074e2010-06-04 20:19:55 -0400547 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
548 bd_forget(inode);
549 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
550 cd_forget(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +1100551
552 remove_inode_hash(inode);
553
554 spin_lock(&inode->i_lock);
555 wake_up_bit(&inode->i_state, __I_NEW);
556 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
557 spin_unlock(&inode->i_lock);
558
559 destroy_inode(inode);
Al Virob4272d42010-06-04 19:33:20 -0400560}
561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562/*
563 * dispose_list - dispose of the contents of a local list
564 * @head: the head of the list to free
565 *
566 * Dispose-list gets a local list with local inodes in it, so it doesn't
567 * need to worry about list corruption and SMP locks.
568 */
569static void dispose_list(struct list_head *head)
570{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 while (!list_empty(head)) {
572 struct inode *inode;
573
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100574 inode = list_first_entry(head, struct inode, i_lru);
575 list_del_init(&inode->i_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
Al Viro644da592010-06-07 13:21:05 -0400577 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579}
580
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581/**
Al Viro63997e92010-10-25 20:49:35 -0400582 * evict_inodes - evict all evictable inodes for a superblock
583 * @sb: superblock to operate on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 *
Al Viro63997e92010-10-25 20:49:35 -0400585 * Make sure that no inodes with zero refcount are retained. This is
586 * called by superblock shutdown after having MS_ACTIVE flag removed,
587 * so any inode reaching zero refcount during or after that call will
588 * be immediately evicted.
589 */
590void evict_inodes(struct super_block *sb)
591{
592 struct inode *inode, *next;
593 LIST_HEAD(dispose);
594
Dave Chinner55fa6092011-03-22 22:23:40 +1100595 spin_lock(&inode_sb_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400596 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
597 if (atomic_read(&inode->i_count))
598 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100599
600 spin_lock(&inode->i_lock);
601 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
602 spin_unlock(&inode->i_lock);
Al Viro63997e92010-10-25 20:49:35 -0400603 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100604 }
Al Viro63997e92010-10-25 20:49:35 -0400605
606 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100607 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100608 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100609 list_add(&inode->i_lru, &dispose);
Al Viro63997e92010-10-25 20:49:35 -0400610 }
Dave Chinner55fa6092011-03-22 22:23:40 +1100611 spin_unlock(&inode_sb_list_lock);
Al Viro63997e92010-10-25 20:49:35 -0400612
613 dispose_list(&dispose);
Al Viro63997e92010-10-25 20:49:35 -0400614}
615
616/**
Christoph Hellwiga0318782010-10-24 19:40:33 +0200617 * invalidate_inodes - attempt to free all inodes on a superblock
618 * @sb: superblock to operate on
NeilBrown93b270f2011-02-24 17:25:47 +1100619 * @kill_dirty: flag to guide handling of dirty inodes
Christoph Hellwiga0318782010-10-24 19:40:33 +0200620 *
621 * Attempts to free all inodes for a given superblock. If there were any
622 * busy inodes return a non-zero value, else zero.
NeilBrown93b270f2011-02-24 17:25:47 +1100623 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
624 * them as busy.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 */
NeilBrown93b270f2011-02-24 17:25:47 +1100626int invalidate_inodes(struct super_block *sb, bool kill_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627{
Dave Chinnercffbc8a2010-10-23 05:03:02 -0400628 int busy = 0;
Christoph Hellwiga0318782010-10-24 19:40:33 +0200629 struct inode *inode, *next;
630 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
Dave Chinner55fa6092011-03-22 22:23:40 +1100632 spin_lock(&inode_sb_list_lock);
Christoph Hellwiga0318782010-10-24 19:40:33 +0200633 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100634 spin_lock(&inode->i_lock);
635 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
636 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100638 }
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500639 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100640 spin_unlock(&inode->i_lock);
NeilBrown93b270f2011-02-24 17:25:47 +1100641 busy = 1;
642 continue;
643 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200644 if (atomic_read(&inode->i_count)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100645 spin_unlock(&inode->i_lock);
Christoph Hellwig99a38912010-10-23 19:07:20 +0200646 busy = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 continue;
648 }
Christoph Hellwig99a38912010-10-23 19:07:20 +0200649
Christoph Hellwig99a38912010-10-23 19:07:20 +0200650 inode->i_state |= I_FREEING;
Dave Chinner02afc412011-03-22 22:23:38 +1100651 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100652 spin_unlock(&inode->i_lock);
Dave Chinner02afc412011-03-22 22:23:38 +1100653 list_add(&inode->i_lru, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 }
Dave Chinner55fa6092011-03-22 22:23:40 +1100655 spin_unlock(&inode_sb_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Christoph Hellwiga0318782010-10-24 19:40:33 +0200657 dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
659 return busy;
660}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662/*
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000663 * Isolate the inode from the LRU in preparation for freeing it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 *
665 * Any inodes which are pinned purely because of attached pagecache have their
Nick Piggin9e38d862010-10-23 06:55:17 -0400666 * pagecache removed. If the inode has metadata buffers attached to
667 * mapping->private_list then try to remove them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 *
Nick Piggin9e38d862010-10-23 06:55:17 -0400669 * If the inode has the I_REFERENCED flag set, then it means that it has been
670 * used recently - the flag is set in iput_final(). When we encounter such an
671 * inode, clear the flag and move it to the back of the LRU so it gets another
672 * pass through the LRU before it gets reclaimed. This is necessary because of
673 * the fact we are doing lazy LRU updates to minimise lock contention so the
674 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
675 * with this flag set because they are the inodes that are out of order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800677static enum lru_status inode_lru_isolate(struct list_head *item,
678 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000679{
680 struct list_head *freeable = arg;
681 struct inode *inode = container_of(item, struct inode, i_lru);
682
683 /*
684 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
685 * If we fail to get the lock, just skip it.
686 */
687 if (!spin_trylock(&inode->i_lock))
688 return LRU_SKIP;
689
690 /*
691 * Referenced or dirty inodes are still in use. Give them another pass
692 * through the LRU as we canot reclaim them now.
693 */
694 if (atomic_read(&inode->i_count) ||
695 (inode->i_state & ~I_REFERENCED)) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800696 list_lru_isolate(lru, &inode->i_lru);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000697 spin_unlock(&inode->i_lock);
698 this_cpu_dec(nr_unused);
699 return LRU_REMOVED;
700 }
701
702 /* recently referenced inodes get one more pass */
703 if (inode->i_state & I_REFERENCED) {
704 inode->i_state &= ~I_REFERENCED;
705 spin_unlock(&inode->i_lock);
706 return LRU_ROTATE;
707 }
708
709 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
710 __iget(inode);
711 spin_unlock(&inode->i_lock);
712 spin_unlock(lru_lock);
713 if (remove_inode_buffers(inode)) {
714 unsigned long reap;
715 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
716 if (current_is_kswapd())
717 __count_vm_events(KSWAPD_INODESTEAL, reap);
718 else
719 __count_vm_events(PGINODESTEAL, reap);
720 if (current->reclaim_state)
721 current->reclaim_state->reclaimed_slab += reap;
722 }
723 iput(inode);
724 spin_lock(lru_lock);
725 return LRU_RETRY;
726 }
727
728 WARN_ON(inode->i_state & I_NEW);
729 inode->i_state |= I_FREEING;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800730 list_lru_isolate_move(lru, &inode->i_lru, freeable);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000731 spin_unlock(&inode->i_lock);
732
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000733 this_cpu_dec(nr_unused);
734 return LRU_REMOVED;
735}
736
737/*
738 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
739 * This is called from the superblock shrinker function with a number of inodes
740 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
741 * then are freed outside inode_lock by dispose_list().
742 */
Vladimir Davydov503c3582015-02-12 14:58:47 -0800743long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
745 LIST_HEAD(freeable);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +1000746 long freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Vladimir Davydov503c3582015-02-12 14:58:47 -0800748 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
749 inode_lru_isolate, &freeable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 dispose_list(&freeable);
Dave Chinner0a234c62013-08-28 10:17:57 +1000751 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752}
753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754static void __wait_on_freeing_inode(struct inode *inode);
755/*
756 * Called with the inode lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530758static struct inode *find_inode(struct super_block *sb,
759 struct hlist_head *head,
760 int (*test)(struct inode *, void *),
761 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530763 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
765repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800766 hlist_for_each_entry(inode, head, i_hash) {
Al Viro5a3cd99282013-11-06 09:54:52 -0500767 if (inode->i_sb != sb)
768 continue;
769 if (!test(inode, data))
770 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100771 spin_lock(&inode->i_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -0400772 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 __wait_on_freeing_inode(inode);
774 goto repeat;
775 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400776 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100777 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400778 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400780 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781}
782
783/*
784 * find_inode_fast is the fast path version of find_inode, see the comment at
785 * iget_locked for details.
786 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530787static struct inode *find_inode_fast(struct super_block *sb,
788 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530790 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
792repeat:
Sasha Levinb67bfe02013-02-27 17:06:00 -0800793 hlist_for_each_entry(inode, head, i_hash) {
Al Viro5a3cd99282013-11-06 09:54:52 -0500794 if (inode->i_ino != ino)
795 continue;
796 if (inode->i_sb != sb)
797 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +1100798 spin_lock(&inode->i_lock);
Al Viroa4ffdde2010-06-02 17:38:30 -0400799 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 __wait_on_freeing_inode(inode);
801 goto repeat;
802 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400803 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100804 spin_unlock(&inode->i_lock);
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400805 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 }
Christoph Hellwigf7899bd2010-10-23 07:09:06 -0400807 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808}
809
Eric Dumazetf991bd22010-10-23 11:18:01 -0400810/*
811 * Each cpu owns a range of LAST_INO_BATCH numbers.
812 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
813 * to renew the exhausted range.
David Chinner8290c352008-10-30 17:35:24 +1100814 *
Eric Dumazetf991bd22010-10-23 11:18:01 -0400815 * This does not significantly increase overflow rate because every CPU can
816 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
817 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
818 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
819 * overflow rate by 2x, which does not seem too significant.
820 *
821 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
822 * error if st_ino won't fit in target struct field. Use 32bit counter
823 * here to attempt to avoid that.
David Chinner8290c352008-10-30 17:35:24 +1100824 */
Eric Dumazetf991bd22010-10-23 11:18:01 -0400825#define LAST_INO_BATCH 1024
826static DEFINE_PER_CPU(unsigned int, last_ino);
David Chinner8290c352008-10-30 17:35:24 +1100827
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400828unsigned int get_next_ino(void)
Eric Dumazetf991bd22010-10-23 11:18:01 -0400829{
830 unsigned int *p = &get_cpu_var(last_ino);
831 unsigned int res = *p;
832
833#ifdef CONFIG_SMP
834 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
835 static atomic_t shared_last_ino;
836 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
837
838 res = next - LAST_INO_BATCH;
839 }
840#endif
841
842 *p = ++res;
843 put_cpu_var(last_ino);
844 return res;
David Chinner8290c352008-10-30 17:35:24 +1100845}
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400846EXPORT_SYMBOL(get_next_ino);
David Chinner8290c352008-10-30 17:35:24 +1100847
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848/**
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200849 * new_inode_pseudo - obtain an inode
850 * @sb: superblock
851 *
852 * Allocates a new inode for given superblock.
853 * Inode wont be chained in superblock s_inodes list
854 * This means :
855 * - fs can't be unmount
856 * - quotas, fsnotify, writeback can't work
857 */
858struct inode *new_inode_pseudo(struct super_block *sb)
859{
860 struct inode *inode = alloc_inode(sb);
861
862 if (inode) {
863 spin_lock(&inode->i_lock);
864 inode->i_state = 0;
865 spin_unlock(&inode->i_lock);
866 INIT_LIST_HEAD(&inode->i_sb_list);
867 }
868 return inode;
869}
870
871/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 * new_inode - obtain an inode
873 * @sb: superblock
874 *
Mel Gorman769848c2007-07-17 04:03:05 -0700875 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800876 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700877 * If HIGHMEM pages are unsuitable or it is known that pages allocated
878 * for the page cache are not reclaimable or migratable,
879 * mapping_set_gfp_mask() must be called with suitable flags on the
880 * newly created inode's mapping
881 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 */
883struct inode *new_inode(struct super_block *sb)
884{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530885 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
Dave Chinner55fa6092011-03-22 22:23:40 +1100887 spin_lock_prefetch(&inode_sb_list_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530888
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200889 inode = new_inode_pseudo(sb);
890 if (inode)
Dave Chinner55fa6092011-03-22 22:23:40 +1100891 inode_sb_list_add(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 return inode;
893}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894EXPORT_SYMBOL(new_inode);
895
Peter Zijlstra14358e62007-10-14 01:38:33 +0200896#ifdef CONFIG_DEBUG_LOCK_ALLOC
Josh Boyere096d0c2011-08-25 07:48:12 -0400897void lockdep_annotate_inode_mutex_key(struct inode *inode)
898{
Namhyung Kima3314a02010-10-11 22:38:00 +0900899 if (S_ISDIR(inode->i_mode)) {
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200900 struct file_system_type *type = inode->i_sb->s_type;
901
Jan Kara9a7aa122009-06-04 15:26:49 +0200902 /* Set new key only if filesystem hasn't already changed it */
Tyler Hicks978d6d82011-12-12 10:02:30 -0600903 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
Jan Kara9a7aa122009-06-04 15:26:49 +0200904 /*
905 * ensure nobody is actually holding i_mutex
906 */
907 mutex_destroy(&inode->i_mutex);
908 mutex_init(&inode->i_mutex);
909 lockdep_set_class(&inode->i_mutex,
910 &type->i_mutex_dir_key);
911 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200912 }
Josh Boyere096d0c2011-08-25 07:48:12 -0400913}
914EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200915#endif
Josh Boyere096d0c2011-08-25 07:48:12 -0400916
917/**
918 * unlock_new_inode - clear the I_NEW state and wake up any waiters
919 * @inode: new inode to unlock
920 *
921 * Called when the inode is fully initialised to clear the new state of the
922 * inode and wake up anyone waiting for the inode to finish initialisation.
923 */
924void unlock_new_inode(struct inode *inode)
925{
926 lockdep_annotate_inode_mutex_key(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100927 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +0100928 WARN_ON(!(inode->i_state & I_NEW));
929 inode->i_state &= ~I_NEW;
Al Viro310fa7a2012-03-10 17:07:28 -0500930 smp_mb();
Dave Chinner250df6e2011-03-22 22:23:36 +1100931 wake_up_bit(&inode->i_state, __I_NEW);
932 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934EXPORT_SYMBOL(unlock_new_inode);
935
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400936/**
J. Bruce Fields375e2892012-04-18 15:16:33 -0400937 * lock_two_nondirectories - take two i_mutexes on non-directory objects
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200938 *
939 * Lock any non-NULL argument that is not a directory.
940 * Zero, one or two objects may be locked by this function.
941 *
J. Bruce Fields375e2892012-04-18 15:16:33 -0400942 * @inode1: first inode to lock
943 * @inode2: second inode to lock
944 */
945void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
946{
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200947 if (inode1 > inode2)
948 swap(inode1, inode2);
949
950 if (inode1 && !S_ISDIR(inode1->i_mode))
J. Bruce Fields275555162012-04-25 07:19:52 -0400951 mutex_lock(&inode1->i_mutex);
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200952 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
J. Bruce Fields40bd22c2012-04-18 15:21:34 -0400953 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_NONDIR2);
J. Bruce Fields375e2892012-04-18 15:16:33 -0400954}
955EXPORT_SYMBOL(lock_two_nondirectories);
956
957/**
958 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
959 * @inode1: first inode to unlock
960 * @inode2: second inode to unlock
961 */
962void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
963{
J. Bruce Fields4fd699a2014-04-01 17:08:43 +0200964 if (inode1 && !S_ISDIR(inode1->i_mode))
965 mutex_unlock(&inode1->i_mutex);
966 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
J. Bruce Fields375e2892012-04-18 15:16:33 -0400967 mutex_unlock(&inode2->i_mutex);
968}
969EXPORT_SYMBOL(unlock_two_nondirectories);
970
971/**
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400972 * iget5_locked - obtain an inode from a mounted file system
973 * @sb: super block of file system
974 * @hashval: hash value (usually inode number) to get
975 * @test: callback used for comparisons between inodes
976 * @set: callback used to initialize a new struct inode
977 * @data: opaque data pointer to pass to @test and @set
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400979 * Search for the inode specified by @hashval and @data in the inode cache,
980 * and if present it is return it with an increased reference count. This is
981 * a generalized version of iget_locked() for file systems where the inode
982 * number is not sufficient for unique identification of an inode.
983 *
984 * If the inode is not in cache, allocate a new inode and return it locked,
985 * hashed, and with the I_NEW flag set. The file system gets to fill it in
986 * before unlocking it via unlock_new_inode().
987 *
988 * Note both @test and @set are called with the inode_hash_lock held, so can't
989 * sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400991struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
992 int (*test)(struct inode *, void *),
993 int (*set)(struct inode *, void *), void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400995 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530996 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Christoph Hellwig0b2d0722011-03-23 15:03:28 -0400998 spin_lock(&inode_hash_lock);
999 inode = find_inode(sb, head, test, data);
1000 spin_unlock(&inode_hash_lock);
1001
1002 if (inode) {
1003 wait_on_inode(inode);
1004 return inode;
1005 }
1006
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 inode = alloc_inode(sb);
1008 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301009 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
Dave Chinner67a23c42011-03-22 22:23:42 +11001011 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 /* We released the lock, so.. */
1013 old = find_inode(sb, head, test, data);
1014 if (!old) {
1015 if (set(inode, data))
1016 goto set_failed;
1017
Dave Chinner250df6e2011-03-22 22:23:36 +11001018 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001019 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001020 hlist_add_head(&inode->i_hash, head);
1021 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001022 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001023 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
1025 /* Return the locked inode with I_NEW set, the
1026 * caller is responsible for filling in the contents
1027 */
1028 return inode;
1029 }
1030
1031 /*
1032 * Uhhuh, somebody else created the same inode under
1033 * us. Use the old inode instead of the one we just
1034 * allocated.
1035 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001036 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 destroy_inode(inode);
1038 inode = old;
1039 wait_on_inode(inode);
1040 }
1041 return inode;
1042
1043set_failed:
Dave Chinner67a23c42011-03-22 22:23:42 +11001044 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 destroy_inode(inode);
1046 return NULL;
1047}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001048EXPORT_SYMBOL(iget5_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001050/**
1051 * iget_locked - obtain an inode from a mounted file system
1052 * @sb: super block of file system
1053 * @ino: inode number to get
1054 *
1055 * Search for the inode specified by @ino in the inode cache and if present
1056 * return it with an increased reference count. This is for file systems
1057 * where the inode number is sufficient for unique identification of an inode.
1058 *
1059 * If the inode is not in cache, allocate a new inode and return it locked,
1060 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1061 * before unlocking it via unlock_new_inode().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 */
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001063struct inode *iget_locked(struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001065 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301066 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001068 spin_lock(&inode_hash_lock);
1069 inode = find_inode_fast(sb, head, ino);
1070 spin_unlock(&inode_hash_lock);
1071 if (inode) {
1072 wait_on_inode(inode);
1073 return inode;
1074 }
1075
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 inode = alloc_inode(sb);
1077 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301078 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
Dave Chinner67a23c42011-03-22 22:23:42 +11001080 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 /* We released the lock, so.. */
1082 old = find_inode_fast(sb, head, ino);
1083 if (!old) {
1084 inode->i_ino = ino;
Dave Chinner250df6e2011-03-22 22:23:36 +11001085 spin_lock(&inode->i_lock);
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001086 inode->i_state = I_NEW;
Dave Chinner250df6e2011-03-22 22:23:36 +11001087 hlist_add_head(&inode->i_hash, head);
1088 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001089 inode_sb_list_add(inode);
Dave Chinner67a23c42011-03-22 22:23:42 +11001090 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
1092 /* Return the locked inode with I_NEW set, the
1093 * caller is responsible for filling in the contents
1094 */
1095 return inode;
1096 }
1097
1098 /*
1099 * Uhhuh, somebody else created the same inode under
1100 * us. Use the old inode instead of the one we just
1101 * allocated.
1102 */
Dave Chinner67a23c42011-03-22 22:23:42 +11001103 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 destroy_inode(inode);
1105 inode = old;
1106 wait_on_inode(inode);
1107 }
1108 return inode;
1109}
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001110EXPORT_SYMBOL(iget_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001112/*
1113 * search the inode cache for a matching inode number.
1114 * If we find one, then the inode number we are trying to
1115 * allocate is not unique and so we should not use it.
1116 *
1117 * Returns 1 if the inode number is unique, 0 if it is not.
1118 */
1119static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1120{
1121 struct hlist_head *b = inode_hashtable + hash(sb, ino);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001122 struct inode *inode;
1123
Dave Chinner67a23c42011-03-22 22:23:42 +11001124 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001125 hlist_for_each_entry(inode, b, i_hash) {
Dave Chinner67a23c42011-03-22 22:23:42 +11001126 if (inode->i_ino == ino && inode->i_sb == sb) {
1127 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001128 return 0;
Dave Chinner67a23c42011-03-22 22:23:42 +11001129 }
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001130 }
Dave Chinner67a23c42011-03-22 22:23:42 +11001131 spin_unlock(&inode_hash_lock);
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001132
1133 return 1;
1134}
1135
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136/**
1137 * iunique - get a unique inode number
1138 * @sb: superblock
1139 * @max_reserved: highest reserved inode number
1140 *
1141 * Obtain an inode number that is unique on the system for a given
1142 * superblock. This is used by file systems that have no natural
1143 * permanent inode numbering system. An inode number is returned that
1144 * is higher than the reserved limit but unique.
1145 *
1146 * BUGS:
1147 * With a large number of inodes live on the file system this function
1148 * currently becomes quite slow.
1149 */
1150ino_t iunique(struct super_block *sb, ino_t max_reserved)
1151{
Jeff Layton866b04f2007-05-08 00:32:29 -07001152 /*
1153 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1154 * error if st_ino won't fit in target struct field. Use 32bit counter
1155 * here to attempt to avoid that.
1156 */
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001157 static DEFINE_SPINLOCK(iunique_lock);
Jeff Layton866b04f2007-05-08 00:32:29 -07001158 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001161 spin_lock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001162 do {
1163 if (counter <= max_reserved)
1164 counter = max_reserved + 1;
1165 res = counter++;
Christoph Hellwigad5e1952010-10-23 07:00:16 -04001166 } while (!test_inode_iunique(sb, res));
1167 spin_unlock(&iunique_lock);
Jeffrey Layton3361c7b2007-05-08 00:29:48 -07001168
1169 return res;
1170}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171EXPORT_SYMBOL(iunique);
1172
1173struct inode *igrab(struct inode *inode)
1174{
Dave Chinner250df6e2011-03-22 22:23:36 +11001175 spin_lock(&inode->i_lock);
1176 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001178 spin_unlock(&inode->i_lock);
1179 } else {
1180 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 /*
1182 * Handle the case where s_op->clear_inode is not been
1183 * called yet, and somebody is calling igrab
1184 * while the inode is getting freed.
1185 */
1186 inode = NULL;
Dave Chinner250df6e2011-03-22 22:23:36 +11001187 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 return inode;
1189}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190EXPORT_SYMBOL(igrab);
1191
1192/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001193 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 * @sb: super block of file system to search
1195 * @hashval: hash value (usually inode number) to search for
1196 * @test: callback used for comparisons between inodes
1197 * @data: opaque data pointer to pass to @test
1198 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001199 * Search for the inode specified by @hashval and @data in the inode cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 * If the inode is in the cache, the inode is returned with an incremented
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001201 * reference count.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001202 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001203 * Note: I_NEW is not waited upon so you have to be very careful what you do
1204 * with the returned inode. You probably should be using ilookup5() instead.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001205 *
Randy Dunlapb6d0ad62011-03-26 13:27:47 -07001206 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001207 */
1208struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1209 int (*test)(struct inode *, void *), void *data)
1210{
1211 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001212 struct inode *inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001213
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001214 spin_lock(&inode_hash_lock);
1215 inode = find_inode(sb, head, test, data);
1216 spin_unlock(&inode_hash_lock);
1217
1218 return inode;
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001219}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001220EXPORT_SYMBOL(ilookup5_nowait);
1221
1222/**
1223 * ilookup5 - search for an inode in the inode cache
1224 * @sb: super block of file system to search
1225 * @hashval: hash value (usually inode number) to search for
1226 * @test: callback used for comparisons between inodes
1227 * @data: opaque data pointer to pass to @test
1228 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001229 * Search for the inode specified by @hashval and @data in the inode cache,
1230 * and if the inode is in the cache, return the inode with an incremented
1231 * reference count. Waits on I_NEW before returning the inode.
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001232 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001234 * This is a generalized version of ilookup() for file systems where the
1235 * inode number is not sufficient for unique identification of an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001237 * Note: @test is called with the inode_hash_lock held, so can't sleep.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 */
1239struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1240 int (*test)(struct inode *, void *), void *data)
1241{
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001242 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001244 if (inode)
1245 wait_on_inode(inode);
1246 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248EXPORT_SYMBOL(ilookup5);
1249
1250/**
1251 * ilookup - search for an inode in the inode cache
1252 * @sb: super block of file system to search
1253 * @ino: inode number to search for
1254 *
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001255 * Search for the inode @ino in the inode cache, and if the inode is in the
1256 * cache, the inode is returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 */
1258struct inode *ilookup(struct super_block *sb, unsigned long ino)
1259{
1260 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001261 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262
Christoph Hellwig0b2d0722011-03-23 15:03:28 -04001263 spin_lock(&inode_hash_lock);
1264 inode = find_inode_fast(sb, head, ino);
1265 spin_unlock(&inode_hash_lock);
1266
1267 if (inode)
1268 wait_on_inode(inode);
1269 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271EXPORT_SYMBOL(ilookup);
1272
Theodore Ts'ofe032c42015-02-02 00:37:01 -05001273/**
1274 * find_inode_nowait - find an inode in the inode cache
1275 * @sb: super block of file system to search
1276 * @hashval: hash value (usually inode number) to search for
1277 * @match: callback used for comparisons between inodes
1278 * @data: opaque data pointer to pass to @match
1279 *
1280 * Search for the inode specified by @hashval and @data in the inode
1281 * cache, where the helper function @match will return 0 if the inode
1282 * does not match, 1 if the inode does match, and -1 if the search
1283 * should be stopped. The @match function must be responsible for
1284 * taking the i_lock spin_lock and checking i_state for an inode being
1285 * freed or being initialized, and incrementing the reference count
1286 * before returning 1. It also must not sleep, since it is called with
1287 * the inode_hash_lock spinlock held.
1288 *
1289 * This is a even more generalized version of ilookup5() when the
1290 * function must never block --- find_inode() can block in
1291 * __wait_on_freeing_inode() --- or when the caller can not increment
1292 * the reference count because the resulting iput() might cause an
1293 * inode eviction. The tradeoff is that the @match funtion must be
1294 * very carefully implemented.
1295 */
1296struct inode *find_inode_nowait(struct super_block *sb,
1297 unsigned long hashval,
1298 int (*match)(struct inode *, unsigned long,
1299 void *),
1300 void *data)
1301{
1302 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1303 struct inode *inode, *ret_inode = NULL;
1304 int mval;
1305
1306 spin_lock(&inode_hash_lock);
1307 hlist_for_each_entry(inode, head, i_hash) {
1308 if (inode->i_sb != sb)
1309 continue;
1310 mval = match(inode, hashval, data);
1311 if (mval == 0)
1312 continue;
1313 if (mval == 1)
1314 ret_inode = inode;
1315 goto out;
1316 }
1317out:
1318 spin_unlock(&inode_hash_lock);
1319 return ret_inode;
1320}
1321EXPORT_SYMBOL(find_inode_nowait);
1322
Al Viro261bca82008-12-30 01:48:21 -05001323int insert_inode_locked(struct inode *inode)
1324{
1325 struct super_block *sb = inode->i_sb;
1326 ino_t ino = inode->i_ino;
1327 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001328
Al Viro261bca82008-12-30 01:48:21 -05001329 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001330 struct inode *old = NULL;
Dave Chinner67a23c42011-03-22 22:23:42 +11001331 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001332 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001333 if (old->i_ino != ino)
1334 continue;
1335 if (old->i_sb != sb)
1336 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001337 spin_lock(&old->i_lock);
1338 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1339 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001340 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001341 }
Al Viro72a43d62009-05-13 19:13:40 +01001342 break;
1343 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001344 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001345 spin_lock(&inode->i_lock);
1346 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001347 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001348 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001349 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001350 return 0;
1351 }
1352 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001353 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001354 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001355 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001356 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001357 iput(old);
1358 return -EBUSY;
1359 }
1360 iput(old);
1361 }
1362}
Al Viro261bca82008-12-30 01:48:21 -05001363EXPORT_SYMBOL(insert_inode_locked);
1364
1365int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1366 int (*test)(struct inode *, void *), void *data)
1367{
1368 struct super_block *sb = inode->i_sb;
1369 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001370
Al Viro261bca82008-12-30 01:48:21 -05001371 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001372 struct inode *old = NULL;
1373
Dave Chinner67a23c42011-03-22 22:23:42 +11001374 spin_lock(&inode_hash_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -08001375 hlist_for_each_entry(old, head, i_hash) {
Al Viro72a43d62009-05-13 19:13:40 +01001376 if (old->i_sb != sb)
1377 continue;
1378 if (!test(old, data))
1379 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001380 spin_lock(&old->i_lock);
1381 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1382 spin_unlock(&old->i_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001383 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001384 }
Al Viro72a43d62009-05-13 19:13:40 +01001385 break;
1386 }
Sasha Levinb67bfe02013-02-27 17:06:00 -08001387 if (likely(!old)) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001388 spin_lock(&inode->i_lock);
1389 inode->i_state |= I_NEW;
Al Viro261bca82008-12-30 01:48:21 -05001390 hlist_add_head(&inode->i_hash, head);
Dave Chinner250df6e2011-03-22 22:23:36 +11001391 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001392 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001393 return 0;
1394 }
1395 __iget(old);
Dave Chinner250df6e2011-03-22 22:23:36 +11001396 spin_unlock(&old->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001397 spin_unlock(&inode_hash_lock);
Al Viro261bca82008-12-30 01:48:21 -05001398 wait_on_inode(old);
Al Viro1d3382cb2010-10-23 15:19:20 -04001399 if (unlikely(!inode_unhashed(old))) {
Al Viro261bca82008-12-30 01:48:21 -05001400 iput(old);
1401 return -EBUSY;
1402 }
1403 iput(old);
1404 }
1405}
Al Viro261bca82008-12-30 01:48:21 -05001406EXPORT_SYMBOL(insert_inode_locked4);
1407
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
Al Viro45321ac2010-06-07 13:43:19 -04001409int generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410{
Al Viro45321ac2010-06-07 13:43:19 -04001411 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413EXPORT_SYMBOL(generic_delete_inode);
1414
Al Viro45321ac2010-06-07 13:43:19 -04001415/*
Al Viro45321ac2010-06-07 13:43:19 -04001416 * Called when we're dropping the last reference
1417 * to an inode.
1418 *
1419 * Call the FS "drop_inode()" function, defaulting to
1420 * the legacy UNIX filesystem behaviour. If it tells
1421 * us to evict inode, do so. Otherwise, retain inode
1422 * in cache if fs is alive, sync and evict if fs is
1423 * shutting down.
1424 */
1425static void iput_final(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426{
1427 struct super_block *sb = inode->i_sb;
Al Viro45321ac2010-06-07 13:43:19 -04001428 const struct super_operations *op = inode->i_sb->s_op;
1429 int drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
Dave Chinner250df6e2011-03-22 22:23:36 +11001431 WARN_ON(inode->i_state & I_NEW);
1432
Al Viroe7f59092011-07-07 15:45:59 -04001433 if (op->drop_inode)
Al Viro45321ac2010-06-07 13:43:19 -04001434 drop = op->drop_inode(inode);
1435 else
1436 drop = generic_drop_inode(inode);
1437
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001438 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1439 inode->i_state |= I_REFERENCED;
Jan Kara4eff96d2012-11-26 16:29:51 -08001440 inode_add_lru(inode);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001441 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001442 return;
1443 }
1444
Al Viro45321ac2010-06-07 13:43:19 -04001445 if (!drop) {
Alexander Viro991114c2005-06-23 00:09:01 -07001446 inode->i_state |= I_WILL_FREE;
Dave Chinner250df6e2011-03-22 22:23:36 +11001447 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 write_inode_now(inode, 1);
Dave Chinner250df6e2011-03-22 22:23:36 +11001449 spin_lock(&inode->i_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001450 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001451 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001453
Alexander Viro991114c2005-06-23 00:09:01 -07001454 inode->i_state |= I_FREEING;
Eric Dumazetc4ae0c62011-07-28 06:55:13 +02001455 if (!list_empty(&inode->i_lru))
1456 inode_lru_list_del(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001457 spin_unlock(&inode->i_lock);
Dave Chinnerb2b2af82011-03-22 22:23:37 +11001458
1459 evict(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460}
1461
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301463 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 * @inode: inode to put
1465 *
1466 * Puts an inode, dropping its usage count. If the inode use count hits
1467 * zero, the inode is then freed and may also be destroyed.
1468 *
1469 * Consequently, iput() can sleep.
1470 */
1471void iput(struct inode *inode)
1472{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001473 if (!inode)
1474 return;
1475 BUG_ON(inode->i_state & I_CLEAR);
1476retry:
1477 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1478 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1479 atomic_inc(&inode->i_count);
1480 inode->i_state &= ~I_DIRTY_TIME;
1481 spin_unlock(&inode->i_lock);
1482 trace_writeback_lazytime_iput(inode);
1483 mark_inode_dirty_sync(inode);
1484 goto retry;
1485 }
1486 iput_final(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 }
1488}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489EXPORT_SYMBOL(iput);
1490
1491/**
1492 * bmap - find a block number in a file
1493 * @inode: inode of file
1494 * @block: block to find
1495 *
1496 * Returns the block number on the device holding the inode that
1497 * is the disk block number for the block of the file requested.
1498 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301499 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 * file.
1501 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301502sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503{
1504 sector_t res = 0;
1505 if (inode->i_mapping->a_ops->bmap)
1506 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1507 return res;
1508}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509EXPORT_SYMBOL(bmap);
1510
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001511/*
1512 * With relative atime, only update atime if the previous atime is
1513 * earlier than either the ctime or mtime or if at least a day has
1514 * passed since the last atime update.
1515 */
1516static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1517 struct timespec now)
1518{
1519
1520 if (!(mnt->mnt_flags & MNT_RELATIME))
1521 return 1;
1522 /*
1523 * Is mtime younger than atime? If yes, update atime:
1524 */
1525 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1526 return 1;
1527 /*
1528 * Is ctime younger than atime? If yes, update atime:
1529 */
1530 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1531 return 1;
1532
1533 /*
1534 * Is the previous atime value older than a day? If yes,
1535 * update atime:
1536 */
1537 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1538 return 1;
1539 /*
1540 * Good, we can skip the atime update:
1541 */
1542 return 0;
1543}
1544
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001545int generic_update_time(struct inode *inode, struct timespec *time, int flags)
Josef Bacikc3b2da32012-03-26 09:59:21 -04001546{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001547 int iflags = I_DIRTY_TIME;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001548
1549 if (flags & S_ATIME)
1550 inode->i_atime = *time;
1551 if (flags & S_VERSION)
1552 inode_inc_iversion(inode);
1553 if (flags & S_CTIME)
1554 inode->i_ctime = *time;
1555 if (flags & S_MTIME)
1556 inode->i_mtime = *time;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001557
1558 if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION))
1559 iflags |= I_DIRTY_SYNC;
1560 __mark_inode_dirty(inode, iflags);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001561 return 0;
1562}
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001563EXPORT_SYMBOL(generic_update_time);
1564
1565/*
1566 * This does the actual work of updating an inodes time or version. Must have
1567 * had called mnt_want_write() before calling this.
1568 */
1569static int update_time(struct inode *inode, struct timespec *time, int flags)
1570{
1571 int (*update_time)(struct inode *, struct timespec *, int);
1572
1573 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1574 generic_update_time;
1575
1576 return update_time(inode, time, flags);
1577}
Josef Bacikc3b2da32012-03-26 09:59:21 -04001578
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001580 * touch_atime - update the access time
Randy Dunlap185553b2012-04-17 17:03:25 -07001581 * @path: the &struct path to update
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 *
1583 * Update the accessed time on an inode and mark it for writeback.
1584 * This function automatically handles read only file systems and media,
1585 * as well as the "noatime" flag and inode specific "noatime" markers.
1586 */
Al Virobadcf2b2013-07-16 18:15:46 +04001587void touch_atime(const struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588{
Al Viro68ac1232012-03-15 08:21:57 -04001589 struct vfsmount *mnt = path->mnt;
David Howellsdf2b1af2015-03-17 22:26:15 +00001590 struct inode *inode = d_inode(path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 struct timespec now;
1592
Andrew Mortonb2276132006-12-13 00:34:33 -08001593 if (inode->i_flags & S_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001594 return;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001595 if (IS_NOATIME(inode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001596 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001597 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001598 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001599
Dave Hansencdb70f32008-02-15 14:37:41 -08001600 if (mnt->mnt_flags & MNT_NOATIME)
Andi Kleenb12536c2009-09-18 13:05:47 -07001601 return;
Dave Hansencdb70f32008-02-15 14:37:41 -08001602 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
Andi Kleenb12536c2009-09-18 13:05:47 -07001603 return;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001604
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 now = current_fs_time(inode->i_sb);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001606
1607 if (!relatime_need_update(mnt, inode, now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001608 return;
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001609
Valerie Henson47ae32d2006-12-13 00:34:34 -08001610 if (timespec_equal(&inode->i_atime, &now))
Andi Kleenb12536c2009-09-18 13:05:47 -07001611 return;
1612
Jan Kara5d37e9e2012-06-12 16:20:36 +02001613 if (!sb_start_write_trylock(inode->i_sb))
Andi Kleenb12536c2009-09-18 13:05:47 -07001614 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001615
Jan Kara5d37e9e2012-06-12 16:20:36 +02001616 if (__mnt_want_write(mnt))
1617 goto skip_update;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001618 /*
1619 * File systems can error out when updating inodes if they need to
1620 * allocate new space to modify an inode (such is the case for
1621 * Btrfs), but since we touch atime while walking down the path we
1622 * really don't care if we failed to update the atime of the file,
1623 * so just ignore the return value.
Alexander Block2bc5565282012-06-15 09:49:33 +02001624 * We may also fail on filesystems that have the ability to make parts
1625 * of the fs read only, e.g. subvolumes in Btrfs.
Josef Bacikc3b2da32012-03-26 09:59:21 -04001626 */
1627 update_time(inode, &now, S_ATIME);
Jan Kara5d37e9e2012-06-12 16:20:36 +02001628 __mnt_drop_write(mnt);
1629skip_update:
1630 sb_end_write(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001632EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633
Cong Wang3ed37642012-05-15 14:57:33 +08001634/*
1635 * The logic we want is
1636 *
1637 * if suid or (sgid and xgrp)
1638 * remove privs
1639 */
1640int should_remove_suid(struct dentry *dentry)
1641{
David Howellsdf2b1af2015-03-17 22:26:15 +00001642 umode_t mode = d_inode(dentry)->i_mode;
Cong Wang3ed37642012-05-15 14:57:33 +08001643 int kill = 0;
1644
1645 /* suid always must be killed */
1646 if (unlikely(mode & S_ISUID))
1647 kill = ATTR_KILL_SUID;
1648
1649 /*
1650 * sgid without any exec bits is just a mandatory locking mark; leave
1651 * it alone. If some exec bits are set, it's a real sgid; kill it.
1652 */
1653 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1654 kill |= ATTR_KILL_SGID;
1655
1656 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1657 return kill;
1658
1659 return 0;
1660}
1661EXPORT_SYMBOL(should_remove_suid);
1662
1663static int __remove_suid(struct dentry *dentry, int kill)
1664{
1665 struct iattr newattrs;
1666
1667 newattrs.ia_valid = ATTR_FORCE | kill;
J. Bruce Fields27ac0ff2011-09-20 17:19:26 -04001668 /*
1669 * Note we call this on write, so notify_change will not
1670 * encounter any conflicting delegations:
1671 */
1672 return notify_change(dentry, &newattrs, NULL);
Cong Wang3ed37642012-05-15 14:57:33 +08001673}
1674
1675int file_remove_suid(struct file *file)
1676{
1677 struct dentry *dentry = file->f_path.dentry;
David Howellsdf2b1af2015-03-17 22:26:15 +00001678 struct inode *inode = d_inode(dentry);
Cong Wang3ed37642012-05-15 14:57:33 +08001679 int killsuid;
1680 int killpriv;
1681 int error = 0;
1682
1683 /* Fast path for nothing security related */
1684 if (IS_NOSEC(inode))
1685 return 0;
1686
1687 killsuid = should_remove_suid(dentry);
1688 killpriv = security_inode_need_killpriv(dentry);
1689
1690 if (killpriv < 0)
1691 return killpriv;
1692 if (killpriv)
1693 error = security_inode_killpriv(dentry);
1694 if (!error && killsuid)
1695 error = __remove_suid(dentry, killsuid);
1696 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1697 inode->i_flags |= S_NOSEC;
1698
1699 return error;
1700}
1701EXPORT_SYMBOL(file_remove_suid);
1702
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001704 * file_update_time - update mtime and ctime time
1705 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001707 * Update the mtime and ctime members of an inode and mark the inode
1708 * for writeback. Note that this function is meant exclusively for
1709 * usage in the file write path of filesystems, and filesystems may
1710 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001711 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Josef Bacikc3b2da32012-03-26 09:59:21 -04001712 * timestamps are handled by the server. This can return an error for
1713 * file systems who need to allocate space in order to update an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 */
1715
Josef Bacikc3b2da32012-03-26 09:59:21 -04001716int file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717{
Al Viro496ad9a2013-01-23 17:07:38 -05001718 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 struct timespec now;
Josef Bacikc3b2da32012-03-26 09:59:21 -04001720 int sync_it = 0;
1721 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722
Andi Kleence06e0b2009-09-18 13:05:48 -07001723 /* First try to exhaust all avenues to not sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 if (IS_NOCMTIME(inode))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001725 return 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001726
Andi Kleence06e0b2009-09-18 13:05:48 -07001727 now = current_fs_time(inode->i_sb);
1728 if (!timespec_equal(&inode->i_mtime, &now))
1729 sync_it = S_MTIME;
1730
1731 if (!timespec_equal(&inode->i_ctime, &now))
1732 sync_it |= S_CTIME;
1733
1734 if (IS_I_VERSION(inode))
1735 sync_it |= S_VERSION;
1736
1737 if (!sync_it)
Josef Bacikc3b2da32012-03-26 09:59:21 -04001738 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
Andi Kleence06e0b2009-09-18 13:05:48 -07001740 /* Finally allowed to write? Takes lock. */
Jan Karaeb04c282012-06-12 16:20:35 +02001741 if (__mnt_want_write_file(file))
Josef Bacikc3b2da32012-03-26 09:59:21 -04001742 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
Josef Bacikc3b2da32012-03-26 09:59:21 -04001744 ret = update_time(inode, &now, sync_it);
Jan Karaeb04c282012-06-12 16:20:35 +02001745 __mnt_drop_write_file(file);
Josef Bacikc3b2da32012-03-26 09:59:21 -04001746
1747 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001749EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750
1751int inode_needs_sync(struct inode *inode)
1752{
1753 if (IS_SYNC(inode))
1754 return 1;
1755 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1756 return 1;
1757 return 0;
1758}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759EXPORT_SYMBOL(inode_needs_sync);
1760
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001762 * If we try to find an inode in the inode hash while it is being
1763 * deleted, we have to wait until the filesystem completes its
1764 * deletion before reporting that it isn't found. This function waits
1765 * until the deletion _might_ have completed. Callers are responsible
1766 * to recheck inode state.
1767 *
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001768 * It doesn't matter if I_NEW is not set initially, a call to
Dave Chinner250df6e2011-03-22 22:23:36 +11001769 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1770 * will DTRT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 */
1772static void __wait_on_freeing_inode(struct inode *inode)
1773{
1774 wait_queue_head_t *wq;
Christoph Hellwigeaff8072009-12-17 14:25:01 +01001775 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1776 wq = bit_waitqueue(&inode->i_state, __I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +11001778 spin_unlock(&inode->i_lock);
Dave Chinner67a23c42011-03-22 22:23:42 +11001779 spin_unlock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 schedule();
1781 finish_wait(wq, &wait.wait);
Dave Chinner67a23c42011-03-22 22:23:42 +11001782 spin_lock(&inode_hash_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783}
1784
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785static __initdata unsigned long ihash_entries;
1786static int __init set_ihash_entries(char *str)
1787{
1788 if (!str)
1789 return 0;
1790 ihash_entries = simple_strtoul(str, &str, 0);
1791 return 1;
1792}
1793__setup("ihash_entries=", set_ihash_entries);
1794
1795/*
1796 * Initialize the waitqueues and inode hash table.
1797 */
1798void __init inode_init_early(void)
1799{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001800 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801
1802 /* If hashes are distributed across NUMA nodes, defer
1803 * hash allocation until vmalloc space is available.
1804 */
1805 if (hashdist)
1806 return;
1807
1808 inode_hashtable =
1809 alloc_large_system_hash("Inode-cache",
1810 sizeof(struct hlist_head),
1811 ihash_entries,
1812 14,
1813 HASH_EARLY,
1814 &i_hash_shift,
1815 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001816 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 0);
1818
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001819 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1821}
1822
Denis Cheng74bf17c2007-10-16 23:26:30 -07001823void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001825 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
1827 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001828 inode_cachep = kmem_cache_create("inode_cache",
1829 sizeof(struct inode),
1830 0,
1831 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1832 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001833 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834
1835 /* Hash may have been set up in inode_init_early */
1836 if (!hashdist)
1837 return;
1838
1839 inode_hashtable =
1840 alloc_large_system_hash("Inode-cache",
1841 sizeof(struct hlist_head),
1842 ihash_entries,
1843 14,
1844 0,
1845 &i_hash_shift,
1846 &i_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00001847 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 0);
1849
Dimitri Sivanich074b8512012-02-08 12:39:07 -08001850 for (loop = 0; loop < (1U << i_hash_shift); loop++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1852}
1853
1854void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1855{
1856 inode->i_mode = mode;
1857 if (S_ISCHR(mode)) {
1858 inode->i_fop = &def_chr_fops;
1859 inode->i_rdev = rdev;
1860 } else if (S_ISBLK(mode)) {
1861 inode->i_fop = &def_blk_fops;
1862 inode->i_rdev = rdev;
1863 } else if (S_ISFIFO(mode))
Al Viro599a0ac2013-03-12 09:58:10 -04001864 inode->i_fop = &pipefifo_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 else if (S_ISSOCK(mode))
Al Virobd9b51e2014-11-18 23:38:21 -05001866 ; /* leave it no_open_fops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 else
Manish Katiyaraf0d9ae2009-09-18 13:05:43 -07001868 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1869 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1870 inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871}
1872EXPORT_SYMBOL(init_special_inode);
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001873
1874/**
Ben Hutchingseaae668d2011-02-15 12:48:09 +00001875 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001876 * @inode: New inode
1877 * @dir: Directory inode
1878 * @mode: mode of the new inode
1879 */
1880void inode_init_owner(struct inode *inode, const struct inode *dir,
Al Viro62bb1092011-07-24 23:20:18 -04001881 umode_t mode)
Dmitry Monakhova1bd1202010-03-04 17:29:14 +03001882{
1883 inode->i_uid = current_fsuid();
1884 if (dir && dir->i_mode & S_ISGID) {
1885 inode->i_gid = dir->i_gid;
1886 if (S_ISDIR(mode))
1887 mode |= S_ISGID;
1888 } else
1889 inode->i_gid = current_fsgid();
1890 inode->i_mode = mode;
1891}
1892EXPORT_SYMBOL(inode_init_owner);
Serge E. Hallyne795b712011-03-23 16:43:25 -07001893
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001894/**
1895 * inode_owner_or_capable - check current task permissions to inode
1896 * @inode: inode being checked
1897 *
Andy Lutomirski23adbe12014-06-10 12:45:42 -07001898 * Return true if current either has CAP_FOWNER in a namespace with the
1899 * inode owner uid mapped, or owns the file.
Serge E. Hallyne795b712011-03-23 16:43:25 -07001900 */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001901bool inode_owner_or_capable(const struct inode *inode)
Serge E. Hallyne795b712011-03-23 16:43:25 -07001902{
Andy Lutomirski23adbe12014-06-10 12:45:42 -07001903 struct user_namespace *ns;
1904
Eric W. Biederman92361632012-02-08 07:07:50 -08001905 if (uid_eq(current_fsuid(), inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001906 return true;
Andy Lutomirski23adbe12014-06-10 12:45:42 -07001907
1908 ns = current_user_ns();
1909 if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid))
Serge E. Hallyne795b712011-03-23 16:43:25 -07001910 return true;
1911 return false;
1912}
Serge E. Hallyn2e149672011-03-23 16:43:26 -07001913EXPORT_SYMBOL(inode_owner_or_capable);
Trond Myklebust1d59d612012-05-31 12:22:33 -04001914
1915/*
1916 * Direct i/o helper functions
1917 */
1918static void __inode_dio_wait(struct inode *inode)
1919{
1920 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1921 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1922
1923 do {
1924 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1925 if (atomic_read(&inode->i_dio_count))
1926 schedule();
1927 } while (atomic_read(&inode->i_dio_count));
1928 finish_wait(wq, &q.wait);
1929}
1930
1931/**
1932 * inode_dio_wait - wait for outstanding DIO requests to finish
1933 * @inode: inode to wait for
1934 *
1935 * Waits for all pending direct I/O requests to finish so that we can
1936 * proceed with a truncate or equivalent operation.
1937 *
1938 * Must be called under a lock that serializes taking new references
1939 * to i_dio_count, usually by inode->i_mutex.
1940 */
1941void inode_dio_wait(struct inode *inode)
1942{
1943 if (atomic_read(&inode->i_dio_count))
1944 __inode_dio_wait(inode);
1945}
1946EXPORT_SYMBOL(inode_dio_wait);
1947
1948/*
Theodore Ts'o5f16f322014-03-24 14:43:12 -04001949 * inode_set_flags - atomically set some inode flags
1950 *
1951 * Note: the caller should be holding i_mutex, or else be sure that
1952 * they have exclusive access to the inode structure (i.e., while the
1953 * inode is being instantiated). The reason for the cmpxchg() loop
1954 * --- which wouldn't be necessary if all code paths which modify
1955 * i_flags actually followed this rule, is that there is at least one
1956 * code path which doesn't today --- for example,
1957 * __generic_file_aio_write() calls file_remove_suid() without holding
1958 * i_mutex --- so we use cmpxchg() out of an abundance of caution.
1959 *
1960 * In the long run, i_mutex is overkill, and we should probably look
1961 * at using the i_lock spinlock to protect i_flags, and then make sure
1962 * it is so documented in include/linux/fs.h and that all code follows
1963 * the locking convention!!
1964 */
1965void inode_set_flags(struct inode *inode, unsigned int flags,
1966 unsigned int mask)
1967{
1968 unsigned int old_flags, new_flags;
1969
1970 WARN_ON_ONCE(flags & ~mask);
1971 do {
1972 old_flags = ACCESS_ONCE(inode->i_flags);
1973 new_flags = (old_flags & ~mask) | flags;
1974 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
1975 new_flags) != old_flags));
1976}
1977EXPORT_SYMBOL(inode_set_flags);