]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/inode.c
Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
[mirror_ubuntu-zesty-kernel.git] / fs / inode.c
CommitLineData
1da177e4 1/*
1da177e4 2 * (C) 1997 Linus Torvalds
4b4563dc 3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
1da177e4 4 */
e59cc473 5#include <linux/export.h>
1da177e4
LT
6#include <linux/fs.h>
7#include <linux/mm.h>
1da177e4 8#include <linux/backing-dev.h>
1da177e4
LT
9#include <linux/hash.h>
10#include <linux/swap.h>
11#include <linux/security.h>
1da177e4
LT
12#include <linux/cdev.h>
13#include <linux/bootmem.h>
3be25f49 14#include <linux/fsnotify.h>
fc33a7bb 15#include <linux/mount.h>
f19d4a8f 16#include <linux/posix_acl.h>
9ce6e0be 17#include <linux/prefetch.h>
4b4563dc 18#include <linux/buffer_head.h> /* for inode_has_buffers */
7ada4db8 19#include <linux/ratelimit.h>
bc3b14cb 20#include <linux/list_lru.h>
0ae45f63 21#include <trace/events/writeback.h>
a66979ab 22#include "internal.h"
1da177e4 23
250df6ed 24/*
4b4563dc 25 * Inode locking rules:
250df6ed
DC
26 *
27 * inode->i_lock protects:
28 * inode->i_state, inode->i_hash, __iget()
bc3b14cb 29 * Inode LRU list locks protect:
98b745c6 30 * inode->i_sb->s_inode_lru, inode->i_lru
55fa6091
DC
31 * inode_sb_list_lock protects:
32 * sb->s_inodes, inode->i_sb_list
f758eeab 33 * bdi->wb.list_lock protects:
0ae45f63 34 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_wb_list
67a23c49
DC
35 * inode_hash_lock protects:
36 * inode_hashtable, inode->i_hash
250df6ed
DC
37 *
38 * Lock ordering:
55fa6091
DC
39 *
40 * inode_sb_list_lock
41 * inode->i_lock
bc3b14cb 42 * Inode LRU list locks
a66979ab 43 *
f758eeab 44 * bdi->wb.list_lock
a66979ab 45 * inode->i_lock
67a23c49
DC
46 *
47 * inode_hash_lock
48 * inode_sb_list_lock
49 * inode->i_lock
50 *
51 * iunique_lock
52 * inode_hash_lock
250df6ed
DC
53 */
54
fa3536cc
ED
55static unsigned int i_hash_mask __read_mostly;
56static unsigned int i_hash_shift __read_mostly;
67a23c49
DC
57static struct hlist_head *inode_hashtable __read_mostly;
58static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
1da177e4 59
55fa6091
DC
60__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
61
7dcda1c9
JA
62/*
63 * 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
1da177e4
LT
70/*
71 * Statistics gathering..
72 */
73struct inodes_stat_t inodes_stat;
74
3942c07c
GC
75static DEFINE_PER_CPU(unsigned long, nr_inodes);
76static DEFINE_PER_CPU(unsigned long, nr_unused);
cffbc8aa 77
6b3304b5 78static struct kmem_cache *inode_cachep __read_mostly;
1da177e4 79
3942c07c 80static long get_nr_inodes(void)
cffbc8aa 81{
3e880fb5 82 int i;
3942c07c 83 long sum = 0;
3e880fb5
NP
84 for_each_possible_cpu(i)
85 sum += per_cpu(nr_inodes, i);
86 return sum < 0 ? 0 : sum;
cffbc8aa
DC
87}
88
3942c07c 89static inline long get_nr_inodes_unused(void)
cffbc8aa 90{
fcb94f72 91 int i;
3942c07c 92 long sum = 0;
fcb94f72
DC
93 for_each_possible_cpu(i)
94 sum += per_cpu(nr_unused, i);
95 return sum < 0 ? 0 : sum;
cffbc8aa
DC
96}
97
3942c07c 98long get_nr_dirty_inodes(void)
cffbc8aa 99{
3e880fb5 100 /* not actually dirty inodes, but a wild approximation */
3942c07c 101 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
cffbc8aa 102 return nr_dirty > 0 ? nr_dirty : 0;
cffbc8aa
DC
103}
104
105/*
106 * Handle nr_inode sysctl
107 */
108#ifdef CONFIG_SYSCTL
1f7e0616 109int proc_nr_inodes(struct ctl_table *table, int write,
cffbc8aa
DC
110 void __user *buffer, size_t *lenp, loff_t *ppos)
111{
112 inodes_stat.nr_inodes = get_nr_inodes();
fcb94f72 113 inodes_stat.nr_unused = get_nr_inodes_unused();
3942c07c 114 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
cffbc8aa
DC
115}
116#endif
117
bd9b51e7
AV
118static int no_open(struct inode *inode, struct file *file)
119{
120 return -ENXIO;
121}
122
2cb1599f
DC
123/**
124 * inode_init_always - perform inode structure intialisation
0bc02f3f
RD
125 * @sb: superblock inode belongs to
126 * @inode: inode to initialise
2cb1599f
DC
127 *
128 * These are initializations that need to be done on every inode
129 * allocation as the fields are not initialised by slab allocation.
130 */
54e34621 131int inode_init_always(struct super_block *sb, struct inode *inode)
1da177e4 132{
6e1d5dcc 133 static const struct inode_operations empty_iops;
bd9b51e7 134 static const struct file_operations no_open_fops = {.open = no_open};
6b3304b5 135 struct address_space *const mapping = &inode->i_data;
2cb1599f
DC
136
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;
bd9b51e7 142 inode->i_fop = &no_open_fops;
a78ef704 143 inode->__i_nlink = 1;
3ddcd056 144 inode->i_opflags = 0;
92361636
EB
145 i_uid_write(inode, 0);
146 i_gid_write(inode, 0);
2cb1599f
DC
147 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;
2cb1599f
DC
152 inode->i_pipe = NULL;
153 inode->i_bdev = NULL;
154 inode->i_cdev = NULL;
155 inode->i_rdev = 0;
156 inode->dirtied_when = 0;
6146f0d5
MZ
157
158 if (security_inode_alloc(inode))
54e34621 159 goto out;
2cb1599f
DC
160 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
bd5fe6c5 166 atomic_set(&inode->i_dio_count, 0);
2cb1599f
DC
167
168 mapping->a_ops = &empty_aops;
169 mapping->host = inode;
170 mapping->flags = 0;
4bb5f5d9 171 atomic_set(&mapping->i_mmap_writable, 0);
3c1d4378 172 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
252aa6f5 173 mapping->private_data = NULL;
2cb1599f 174 mapping->writeback_index = 0;
2cb1599f
DC
175 inode->i_private = NULL;
176 inode->i_mapping = mapping;
b3d9b7a3 177 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
f19d4a8f
AV
178#ifdef CONFIG_FS_POSIX_ACL
179 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
180#endif
2cb1599f 181
3be25f49
EP
182#ifdef CONFIG_FSNOTIFY
183 inode->i_fsnotify_mask = 0;
184#endif
4a075e39 185 inode->i_flctx = NULL;
3e880fb5 186 this_cpu_inc(nr_inodes);
cffbc8aa 187
54e34621 188 return 0;
54e34621
CH
189out:
190 return -ENOMEM;
1da177e4 191}
2cb1599f
DC
192EXPORT_SYMBOL(inode_init_always);
193
194static struct inode *alloc_inode(struct super_block *sb)
195{
196 struct inode *inode;
197
198 if (sb->s_op->alloc_inode)
199 inode = sb->s_op->alloc_inode(sb);
200 else
201 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
202
54e34621
CH
203 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;
2cb1599f 215}
1da177e4 216
ff0c7d15
NP
217void free_inode_nonrcu(struct inode *inode)
218{
219 kmem_cache_free(inode_cachep, inode);
220}
221EXPORT_SYMBOL(free_inode_nonrcu);
222
2e00c97e 223void __destroy_inode(struct inode *inode)
1da177e4 224{
b7542f8c 225 BUG_ON(inode_has_buffers(inode));
1da177e4 226 security_inode_free(inode);
3be25f49 227 fsnotify_inode_delete(inode);
4a075e39 228 locks_free_lock_context(inode->i_flctx);
7ada4db8
MS
229 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
f19d4a8f
AV
234#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
3e880fb5 240 this_cpu_dec(nr_inodes);
2e00c97e
CH
241}
242EXPORT_SYMBOL(__destroy_inode);
243
fa0d7e3d
NP
244static void i_callback(struct rcu_head *head)
245{
246 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP
247 kmem_cache_free(inode_cachep, inode);
248}
249
56b0dacf 250static void destroy_inode(struct inode *inode)
2e00c97e 251{
7ccf19a8 252 BUG_ON(!list_empty(&inode->i_lru));
2e00c97e 253 __destroy_inode(inode);
1da177e4
LT
254 if (inode->i_sb->s_op->destroy_inode)
255 inode->i_sb->s_op->destroy_inode(inode);
256 else
fa0d7e3d 257 call_rcu(&inode->i_rcu, i_callback);
1da177e4 258}
1da177e4 259
7ada4db8
MS
260/**
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) {
7ada4db8
MS
308 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{
f4e0c30c
AV
329 if (unlikely(inode->i_nlink == 0)) {
330 WARN_ON(!(inode->i_state & I_LINKABLE));
7ada4db8 331 atomic_long_dec(&inode->i_sb->s_remove_count);
f4e0c30c 332 }
7ada4db8
MS
333
334 inode->__i_nlink++;
335}
336EXPORT_SYMBOL(inc_nlink);
337
2aa15890
MS
338void 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);
c8c06efa 343 init_rwsem(&mapping->i_mmap_rwsem);
2aa15890
MS
344 INIT_LIST_HEAD(&mapping->private_list);
345 spin_lock_init(&mapping->private_lock);
6b2dbba8 346 mapping->i_mmap = RB_ROOT;
2aa15890
MS
347}
348EXPORT_SYMBOL(address_space_init_once);
349
1da177e4
LT
350/*
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);
1da177e4 359 INIT_LIST_HEAD(&inode->i_devices);
7ccf19a8
NP
360 INIT_LIST_HEAD(&inode->i_wb_list);
361 INIT_LIST_HEAD(&inode->i_lru);
2aa15890 362 address_space_init_once(&inode->i_data);
1da177e4 363 i_size_ordered_init(inode);
3be25f49 364#ifdef CONFIG_FSNOTIFY
e61ce867 365 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
3be25f49 366#endif
1da177e4 367}
1da177e4
LT
368EXPORT_SYMBOL(inode_init_once);
369
51cc5068 370static void init_once(void *foo)
1da177e4 371{
6b3304b5 372 struct inode *inode = (struct inode *) foo;
1da177e4 373
a35afb83 374 inode_init_once(inode);
1da177e4
LT
375}
376
377/*
250df6ed 378 * inode->i_lock must be held
1da177e4 379 */
6b3304b5 380void __iget(struct inode *inode)
1da177e4 381{
9e38d86f
NP
382 atomic_inc(&inode->i_count);
383}
2e147f1e 384
7de9c6ee
AV
385/*
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
9e38d86f
NP
394static void inode_lru_list_add(struct inode *inode)
395{
bc3b14cb 396 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
fcb94f72 397 this_cpu_inc(nr_unused);
9e38d86f 398}
2e147f1e 399
4eff96dd
JK
400/*
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{
0ae45f63
TT
407 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
408 I_FREEING | I_WILL_FREE)) &&
4eff96dd
JK
409 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
410 inode_lru_list_add(inode);
411}
412
413
9e38d86f
NP
414static void inode_lru_list_del(struct inode *inode)
415{
bc3b14cb
DC
416
417 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
fcb94f72 418 this_cpu_dec(nr_unused);
1da177e4
LT
419}
420
646ec461
CH
421/**
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{
55fa6091
DC
427 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);
646ec461
CH
430}
431EXPORT_SYMBOL_GPL(inode_sb_list_add);
432
55fa6091 433static inline void inode_sb_list_del(struct inode *inode)
646ec461 434{
a209dfc7
ED
435 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 }
646ec461
CH
440}
441
4c51acbc
DC
442static 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;
4b4563dc
CH
448 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
449 return tmp & i_hash_mask;
4c51acbc
DC
450}
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{
646ec461
CH
462 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
463
67a23c49 464 spin_lock(&inode_hash_lock);
250df6ed 465 spin_lock(&inode->i_lock);
646ec461 466 hlist_add_head(&inode->i_hash, b);
250df6ed 467 spin_unlock(&inode->i_lock);
67a23c49 468 spin_unlock(&inode_hash_lock);
4c51acbc
DC
469}
470EXPORT_SYMBOL(__insert_inode_hash);
471
4c51acbc 472/**
f2ee7abf 473 * __remove_inode_hash - remove an inode from the hash
4c51acbc
DC
474 * @inode: inode to unhash
475 *
476 * Remove an inode from the superblock.
477 */
f2ee7abf 478void __remove_inode_hash(struct inode *inode)
4c51acbc 479{
67a23c49 480 spin_lock(&inode_hash_lock);
250df6ed 481 spin_lock(&inode->i_lock);
4c51acbc 482 hlist_del_init(&inode->i_hash);
250df6ed 483 spin_unlock(&inode->i_lock);
67a23c49 484 spin_unlock(&inode_hash_lock);
4c51acbc 485}
f2ee7abf 486EXPORT_SYMBOL(__remove_inode_hash);
4c51acbc 487
dbd5768f 488void clear_inode(struct inode *inode)
b0683aa6
AV
489{
490 might_sleep();
08142579
JK
491 /*
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);
b0683aa6 497 BUG_ON(inode->i_data.nrpages);
91b0abe3 498 BUG_ON(inode->i_data.nrshadows);
08142579 499 spin_unlock_irq(&inode->i_data.tree_lock);
b0683aa6
AV
500 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);
fa0d7e3d 503 /* don't need i_lock here, no concurrent mods to i_state */
b0683aa6
AV
504 inode->i_state = I_FREEING | I_CLEAR;
505}
dbd5768f 506EXPORT_SYMBOL(clear_inode);
b0683aa6 507
b2b2af8e
DC
508/*
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 */
644da596 521static void evict(struct inode *inode)
b4272d4c
AV
522{
523 const struct super_operations *op = inode->i_sb->s_op;
524
b2b2af8e
DC
525 BUG_ON(!(inode->i_state & I_FREEING));
526 BUG_ON(!list_empty(&inode->i_lru));
527
b12362bd
ED
528 if (!list_empty(&inode->i_wb_list))
529 inode_wb_list_del(inode);
530
55fa6091
DC
531 inode_sb_list_del(inode);
532
169ebd90
JK
533 /*
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);
7994e6f7 540
be7ce416
AV
541 if (op->evict_inode) {
542 op->evict_inode(inode);
b4272d4c 543 } else {
91b0abe3 544 truncate_inode_pages_final(&inode->i_data);
dbd5768f 545 clear_inode(inode);
b4272d4c 546 }
661074e9
AV
547 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);
b2b2af8e
DC
551
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);
b4272d4c
AV
560}
561
1da177e4
LT
562/*
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{
1da177e4
LT
571 while (!list_empty(head)) {
572 struct inode *inode;
573
7ccf19a8
NP
574 inode = list_first_entry(head, struct inode, i_lru);
575 list_del_init(&inode->i_lru);
1da177e4 576
644da596 577 evict(inode);
1da177e4 578 }
1da177e4
LT
579}
580
63997e98
AV
581/**
582 * evict_inodes - evict all evictable inodes for a superblock
583 * @sb: superblock to operate on
584 *
585 * 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.
1da177e4 589 */
63997e98 590void evict_inodes(struct super_block *sb)
1da177e4 591{
63997e98
AV
592 struct inode *inode, *next;
593 LIST_HEAD(dispose);
1da177e4 594
55fa6091 595 spin_lock(&inode_sb_list_lock);
63997e98
AV
596 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
597 if (atomic_read(&inode->i_count))
aabb8fdb 598 continue;
250df6ed
DC
599
600 spin_lock(&inode->i_lock);
601 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
602 spin_unlock(&inode->i_lock);
1da177e4 603 continue;
250df6ed 604 }
63997e98
AV
605
606 inode->i_state |= I_FREEING;
02afc410 607 inode_lru_list_del(inode);
250df6ed 608 spin_unlock(&inode->i_lock);
02afc410 609 list_add(&inode->i_lru, &dispose);
1da177e4 610 }
55fa6091 611 spin_unlock(&inode_sb_list_lock);
63997e98
AV
612
613 dispose_list(&dispose);
1da177e4
LT
614}
615
1da177e4 616/**
a0318786
CH
617 * invalidate_inodes - attempt to free all inodes on a superblock
618 * @sb: superblock to operate on
93b270f7 619 * @kill_dirty: flag to guide handling of dirty inodes
1da177e4 620 *
a0318786
CH
621 * Attempts to free all inodes for a given superblock. If there were any
622 * busy inodes return a non-zero value, else zero.
93b270f7
N
623 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
624 * them as busy.
1da177e4 625 */
93b270f7 626int invalidate_inodes(struct super_block *sb, bool kill_dirty)
1da177e4 627{
cffbc8aa 628 int busy = 0;
a0318786
CH
629 struct inode *inode, *next;
630 LIST_HEAD(dispose);
1da177e4 631
55fa6091 632 spin_lock(&inode_sb_list_lock);
a0318786 633 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
250df6ed
DC
634 spin_lock(&inode->i_lock);
635 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
636 spin_unlock(&inode->i_lock);
aabb8fdb 637 continue;
250df6ed 638 }
0ae45f63 639 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
250df6ed 640 spin_unlock(&inode->i_lock);
93b270f7
N
641 busy = 1;
642 continue;
643 }
99a38919 644 if (atomic_read(&inode->i_count)) {
250df6ed 645 spin_unlock(&inode->i_lock);
99a38919 646 busy = 1;
1da177e4
LT
647 continue;
648 }
99a38919 649
99a38919 650 inode->i_state |= I_FREEING;
02afc410 651 inode_lru_list_del(inode);
250df6ed 652 spin_unlock(&inode->i_lock);
02afc410 653 list_add(&inode->i_lru, &dispose);
1da177e4 654 }
55fa6091 655 spin_unlock(&inode_sb_list_lock);
1da177e4 656
a0318786 657 dispose_list(&dispose);
1da177e4
LT
658
659 return busy;
660}
1da177e4 661
1da177e4 662/*
bc3b14cb 663 * Isolate the inode from the LRU in preparation for freeing it.
1da177e4
LT
664 *
665 * Any inodes which are pinned purely because of attached pagecache have their
9e38d86f
NP
666 * pagecache removed. If the inode has metadata buffers attached to
667 * mapping->private_list then try to remove them.
1da177e4 668 *
9e38d86f
NP
669 * 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.
1da177e4 676 */
3f97b163
VD
677static enum lru_status inode_lru_isolate(struct list_head *item,
678 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1da177e4 679{
bc3b14cb
DC
680 struct list_head *freeable = arg;
681 struct inode *inode = container_of(item, struct inode, i_lru);
1da177e4 682
bc3b14cb
DC
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;
1da177e4 689
bc3b14cb
DC
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)) {
3f97b163 696 list_lru_isolate(lru, &inode->i_lru);
bc3b14cb
DC
697 spin_unlock(&inode->i_lock);
698 this_cpu_dec(nr_unused);
699 return LRU_REMOVED;
700 }
1da177e4 701
bc3b14cb
DC
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 }
1da177e4 708
bc3b14cb
DC
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;
02afc410 722 }
bc3b14cb
DC
723 iput(inode);
724 spin_lock(lru_lock);
725 return LRU_RETRY;
726 }
02afc410 727
bc3b14cb
DC
728 WARN_ON(inode->i_state & I_NEW);
729 inode->i_state |= I_FREEING;
3f97b163 730 list_lru_isolate_move(lru, &inode->i_lru, freeable);
bc3b14cb 731 spin_unlock(&inode->i_lock);
9e38d86f 732
bc3b14cb
DC
733 this_cpu_dec(nr_unused);
734 return LRU_REMOVED;
735}
7ccf19a8 736
bc3b14cb
DC
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 */
503c358c 743long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
bc3b14cb
DC
744{
745 LIST_HEAD(freeable);
746 long freed;
1da177e4 747
503c358c
VD
748 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
749 inode_lru_isolate, &freeable);
1da177e4 750 dispose_list(&freeable);
0a234c6d 751 return freed;
1da177e4
LT
752}
753
1da177e4
LT
754static void __wait_on_freeing_inode(struct inode *inode);
755/*
756 * Called with the inode lock held.
1da177e4 757 */
6b3304b5
MK
758static struct inode *find_inode(struct super_block *sb,
759 struct hlist_head *head,
760 int (*test)(struct inode *, void *),
761 void *data)
1da177e4 762{
6b3304b5 763 struct inode *inode = NULL;
1da177e4
LT
764
765repeat:
b67bfe0d 766 hlist_for_each_entry(inode, head, i_hash) {
5a3cd992 767 if (inode->i_sb != sb)
1da177e4 768 continue;
5a3cd992 769 if (!test(inode, data))
1da177e4 770 continue;
5a3cd992 771 spin_lock(&inode->i_lock);
a4ffdde6 772 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1da177e4
LT
773 __wait_on_freeing_inode(inode);
774 goto repeat;
775 }
f7899bd5 776 __iget(inode);
250df6ed 777 spin_unlock(&inode->i_lock);
f7899bd5 778 return inode;
1da177e4 779 }
f7899bd5 780 return NULL;
1da177e4
LT
781}
782
783/*
784 * find_inode_fast is the fast path version of find_inode, see the comment at
785 * iget_locked for details.
786 */
6b3304b5
MK
787static struct inode *find_inode_fast(struct super_block *sb,
788 struct hlist_head *head, unsigned long ino)
1da177e4 789{
6b3304b5 790 struct inode *inode = NULL;
1da177e4
LT
791
792repeat:
b67bfe0d 793 hlist_for_each_entry(inode, head, i_hash) {
5a3cd992 794 if (inode->i_ino != ino)
1da177e4 795 continue;
5a3cd992 796 if (inode->i_sb != sb)
1da177e4 797 continue;
5a3cd992 798 spin_lock(&inode->i_lock);
a4ffdde6 799 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1da177e4
LT
800 __wait_on_freeing_inode(inode);
801 goto repeat;
802 }
f7899bd5 803 __iget(inode);
250df6ed 804 spin_unlock(&inode->i_lock);
f7899bd5 805 return inode;
1da177e4 806 }
f7899bd5 807 return NULL;
8290c35f
DC
808}
809
f991bd2e
ED
810/*
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.
8290c35f 814 *
f991bd2e
ED
815 * 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.
8290c35f 824 */
f991bd2e
ED
825#define LAST_INO_BATCH 1024
826static DEFINE_PER_CPU(unsigned int, last_ino);
827
85fe4025 828unsigned int get_next_ino(void)
8290c35f 829{
f991bd2e
ED
830 unsigned int *p = &get_cpu_var(last_ino);
831 unsigned int res = *p;
8290c35f 832
f991bd2e
ED
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;
8290c35f 845}
85fe4025 846EXPORT_SYMBOL(get_next_ino);
8290c35f 847
a209dfc7
ED
848/**
849 * 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
1da177e4
LT
871/**
872 * new_inode - obtain an inode
873 * @sb: superblock
874 *
769848c0 875 * Allocates a new inode for given superblock. The default gfp_mask
3c1d4378 876 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
769848c0
MG
877 * 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 *
1da177e4
LT
882 */
883struct inode *new_inode(struct super_block *sb)
884{
6b3304b5 885 struct inode *inode;
1da177e4 886
55fa6091 887 spin_lock_prefetch(&inode_sb_list_lock);
6b3304b5 888
a209dfc7
ED
889 inode = new_inode_pseudo(sb);
890 if (inode)
55fa6091 891 inode_sb_list_add(inode);
1da177e4
LT
892 return inode;
893}
1da177e4
LT
894EXPORT_SYMBOL(new_inode);
895
14358e6d 896#ifdef CONFIG_DEBUG_LOCK_ALLOC
e096d0c7
JB
897void lockdep_annotate_inode_mutex_key(struct inode *inode)
898{
a3314a0e 899 if (S_ISDIR(inode->i_mode)) {
1e89a5e1
PZ
900 struct file_system_type *type = inode->i_sb->s_type;
901
9a7aa12f 902 /* Set new key only if filesystem hasn't already changed it */
978d6d8c 903 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
9a7aa12f
JK
904 /*
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 }
1e89a5e1 912 }
e096d0c7
JB
913}
914EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
14358e6d 915#endif
e096d0c7
JB
916
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);
250df6ed 927 spin_lock(&inode->i_lock);
eaff8079
CH
928 WARN_ON(!(inode->i_state & I_NEW));
929 inode->i_state &= ~I_NEW;
310fa7a3 930 smp_mb();
250df6ed
DC
931 wake_up_bit(&inode->i_state, __I_NEW);
932 spin_unlock(&inode->i_lock);
1da177e4 933}
1da177e4
LT
934EXPORT_SYMBOL(unlock_new_inode);
935
375e289e
BF
936/**
937 * lock_two_nondirectories - take two i_mutexes on non-directory objects
4fd699ae
BF
938 *
939 * Lock any non-NULL argument that is not a directory.
940 * Zero, one or two objects may be locked by this function.
941 *
375e289e
BF
942 * @inode1: first inode to lock
943 * @inode2: second inode to lock
944 */
945void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
946{
4fd699ae
BF
947 if (inode1 > inode2)
948 swap(inode1, inode2);
949
950 if (inode1 && !S_ISDIR(inode1->i_mode))
27555516 951 mutex_lock(&inode1->i_mutex);
4fd699ae 952 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
40bd22c9 953 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_NONDIR2);
375e289e
BF
954}
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{
4fd699ae
BF
964 if (inode1 && !S_ISDIR(inode1->i_mode))
965 mutex_unlock(&inode1->i_mutex);
966 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
375e289e
BF
967 mutex_unlock(&inode2->i_mutex);
968}
969EXPORT_SYMBOL(unlock_two_nondirectories);
970
0b2d0724
CH
971/**
972 * 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
978 *
979 * 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().
1da177e4 987 *
0b2d0724
CH
988 * Note both @test and @set are called with the inode_hash_lock held, so can't
989 * sleep.
1da177e4 990 */
0b2d0724
CH
991struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
992 int (*test)(struct inode *, void *),
993 int (*set)(struct inode *, void *), void *data)
1da177e4 994{
0b2d0724 995 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
6b3304b5 996 struct inode *inode;
1da177e4 997
0b2d0724
CH
998 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
1da177e4
LT
1007 inode = alloc_inode(sb);
1008 if (inode) {
6b3304b5 1009 struct inode *old;
1da177e4 1010
67a23c49 1011 spin_lock(&inode_hash_lock);
1da177e4
LT
1012 /* 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
250df6ed
DC
1018 spin_lock(&inode->i_lock);
1019 inode->i_state = I_NEW;
646ec461 1020 hlist_add_head(&inode->i_hash, head);
250df6ed 1021 spin_unlock(&inode->i_lock);
55fa6091 1022 inode_sb_list_add(inode);
67a23c49 1023 spin_unlock(&inode_hash_lock);
1da177e4
LT
1024
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 */
67a23c49 1036 spin_unlock(&inode_hash_lock);
1da177e4
LT
1037 destroy_inode(inode);
1038 inode = old;
1039 wait_on_inode(inode);
1040 }
1041 return inode;
1042
1043set_failed:
67a23c49 1044 spin_unlock(&inode_hash_lock);
1da177e4
LT
1045 destroy_inode(inode);
1046 return NULL;
1047}
0b2d0724 1048EXPORT_SYMBOL(iget5_locked);
1da177e4 1049
0b2d0724
CH
1050/**
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().
1da177e4 1062 */
0b2d0724 1063struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1da177e4 1064{
0b2d0724 1065 struct hlist_head *head = inode_hashtable + hash(sb, ino);
6b3304b5 1066 struct inode *inode;
1da177e4 1067
0b2d0724
CH
1068 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
1da177e4
LT
1076 inode = alloc_inode(sb);
1077 if (inode) {
6b3304b5 1078 struct inode *old;
1da177e4 1079
67a23c49 1080 spin_lock(&inode_hash_lock);
1da177e4
LT
1081 /* We released the lock, so.. */
1082 old = find_inode_fast(sb, head, ino);
1083 if (!old) {
1084 inode->i_ino = ino;
250df6ed
DC
1085 spin_lock(&inode->i_lock);
1086 inode->i_state = I_NEW;
646ec461 1087 hlist_add_head(&inode->i_hash, head);
250df6ed 1088 spin_unlock(&inode->i_lock);
55fa6091 1089 inode_sb_list_add(inode);
67a23c49 1090 spin_unlock(&inode_hash_lock);
1da177e4
LT
1091
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 */
67a23c49 1103 spin_unlock(&inode_hash_lock);
1da177e4
LT
1104 destroy_inode(inode);
1105 inode = old;
1106 wait_on_inode(inode);
1107 }
1108 return inode;
1109}
0b2d0724 1110EXPORT_SYMBOL(iget_locked);
1da177e4 1111
ad5e195a
CH
1112/*
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);
ad5e195a
CH
1122 struct inode *inode;
1123
67a23c49 1124 spin_lock(&inode_hash_lock);
b67bfe0d 1125 hlist_for_each_entry(inode, b, i_hash) {
67a23c49
DC
1126 if (inode->i_ino == ino && inode->i_sb == sb) {
1127 spin_unlock(&inode_hash_lock);
ad5e195a 1128 return 0;
67a23c49 1129 }
ad5e195a 1130 }
67a23c49 1131 spin_unlock(&inode_hash_lock);
ad5e195a
CH
1132
1133 return 1;
1134}
1135
1da177e4
LT
1136/**
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{
866b04fc
JL
1152 /*
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 */
ad5e195a 1157 static DEFINE_SPINLOCK(iunique_lock);
866b04fc 1158 static unsigned int counter;
1da177e4 1159 ino_t res;
3361c7be 1160
ad5e195a 1161 spin_lock(&iunique_lock);
3361c7be
JL
1162 do {
1163 if (counter <= max_reserved)
1164 counter = max_reserved + 1;
1da177e4 1165 res = counter++;
ad5e195a
CH
1166 } while (!test_inode_iunique(sb, res));
1167 spin_unlock(&iunique_lock);
1da177e4 1168
3361c7be
JL
1169 return res;
1170}
1da177e4
LT
1171EXPORT_SYMBOL(iunique);
1172
1173struct inode *igrab(struct inode *inode)
1174{
250df6ed
DC
1175 spin_lock(&inode->i_lock);
1176 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1da177e4 1177 __iget(inode);
250df6ed
DC
1178 spin_unlock(&inode->i_lock);
1179 } else {
1180 spin_unlock(&inode->i_lock);
1da177e4
LT
1181 /*
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;
250df6ed 1187 }
1da177e4
LT
1188 return inode;
1189}
1da177e4
LT
1190EXPORT_SYMBOL(igrab);
1191
1192/**
0b2d0724 1193 * ilookup5_nowait - search for an inode in the inode cache
1da177e4 1194 * @sb: super block of file system to search
0b2d0724 1195 * @hashval: hash value (usually inode number) to search for
1da177e4
LT
1196 * @test: callback used for comparisons between inodes
1197 * @data: opaque data pointer to pass to @test
1da177e4 1198 *
0b2d0724 1199 * Search for the inode specified by @hashval and @data in the inode cache.
1da177e4
LT
1200 * If the inode is in the cache, the inode is returned with an incremented
1201 * reference count.
1202 *
0b2d0724
CH
1203 * 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.
1da177e4 1205 *
b6d0ad68 1206 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1da177e4 1207 */
0b2d0724
CH
1208struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1209 int (*test)(struct inode *, void *), void *data)
1da177e4 1210{
0b2d0724 1211 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1da177e4
LT
1212 struct inode *inode;
1213
67a23c49 1214 spin_lock(&inode_hash_lock);
1da177e4 1215 inode = find_inode(sb, head, test, data);
67a23c49 1216 spin_unlock(&inode_hash_lock);
88bd5121 1217
0b2d0724 1218 return inode;
88bd5121 1219}
88bd5121
AA
1220EXPORT_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 *
0b2d0724
CH
1229 * 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.
88bd5121 1232 * returned with an incremented reference count.
1da177e4 1233 *
0b2d0724
CH
1234 * This is a generalized version of ilookup() for file systems where the
1235 * inode number is not sufficient for unique identification of an inode.
1da177e4 1236 *
0b2d0724 1237 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1da177e4
LT
1238 */
1239struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1240 int (*test)(struct inode *, void *), void *data)
1241{
0b2d0724 1242 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1da177e4 1243
0b2d0724
CH
1244 if (inode)
1245 wait_on_inode(inode);
1246 return inode;
1da177e4 1247}
1da177e4
LT
1248EXPORT_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 *
0b2d0724
CH
1255 * 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.
1da177e4
LT
1257 */
1258struct inode *ilookup(struct super_block *sb, unsigned long ino)
1259{
1260 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1da177e4
LT
1261 struct inode *inode;
1262
0b2d0724
CH
1263 spin_lock(&inode_hash_lock);
1264 inode = find_inode_fast(sb, head, ino);
1265 spin_unlock(&inode_hash_lock);
1da177e4 1266
1da177e4 1267 if (inode)
0b2d0724
CH
1268 wait_on_inode(inode);
1269 return inode;
1da177e4 1270}
0b2d0724 1271EXPORT_SYMBOL(ilookup);
1da177e4 1272
fe032c42
TT
1273/**
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
261bca86
AV
1323int 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);
261bca86 1328
261bca86 1329 while (1) {
72a43d63 1330 struct inode *old = NULL;
67a23c49 1331 spin_lock(&inode_hash_lock);
b67bfe0d 1332 hlist_for_each_entry(old, head, i_hash) {
72a43d63
AV
1333 if (old->i_ino != ino)
1334 continue;
1335 if (old->i_sb != sb)
1336 continue;
250df6ed
DC
1337 spin_lock(&old->i_lock);
1338 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1339 spin_unlock(&old->i_lock);
72a43d63 1340 continue;
250df6ed 1341 }
72a43d63
AV
1342 break;
1343 }
b67bfe0d 1344 if (likely(!old)) {
250df6ed
DC
1345 spin_lock(&inode->i_lock);
1346 inode->i_state |= I_NEW;
261bca86 1347 hlist_add_head(&inode->i_hash, head);
250df6ed 1348 spin_unlock(&inode->i_lock);
67a23c49 1349 spin_unlock(&inode_hash_lock);
261bca86
AV
1350 return 0;
1351 }
1352 __iget(old);
250df6ed 1353 spin_unlock(&old->i_lock);
67a23c49 1354 spin_unlock(&inode_hash_lock);
261bca86 1355 wait_on_inode(old);
1d3382cb 1356 if (unlikely(!inode_unhashed(old))) {
261bca86
AV
1357 iput(old);
1358 return -EBUSY;
1359 }
1360 iput(old);
1361 }
1362}
261bca86
AV
1363EXPORT_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);
261bca86 1370
261bca86 1371 while (1) {
72a43d63
AV
1372 struct inode *old = NULL;
1373
67a23c49 1374 spin_lock(&inode_hash_lock);
b67bfe0d 1375 hlist_for_each_entry(old, head, i_hash) {
72a43d63
AV
1376 if (old->i_sb != sb)
1377 continue;
1378 if (!test(old, data))
1379 continue;
250df6ed
DC
1380 spin_lock(&old->i_lock);
1381 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1382 spin_unlock(&old->i_lock);
72a43d63 1383 continue;
250df6ed 1384 }
72a43d63
AV
1385 break;
1386 }
b67bfe0d 1387 if (likely(!old)) {
250df6ed
DC
1388 spin_lock(&inode->i_lock);
1389 inode->i_state |= I_NEW;
261bca86 1390 hlist_add_head(&inode->i_hash, head);
250df6ed 1391 spin_unlock(&inode->i_lock);
67a23c49 1392 spin_unlock(&inode_hash_lock);
261bca86
AV
1393 return 0;
1394 }
1395 __iget(old);
250df6ed 1396 spin_unlock(&old->i_lock);
67a23c49 1397 spin_unlock(&inode_hash_lock);
261bca86 1398 wait_on_inode(old);
1d3382cb 1399 if (unlikely(!inode_unhashed(old))) {
261bca86
AV
1400 iput(old);
1401 return -EBUSY;
1402 }
1403 iput(old);
1404 }
1405}
261bca86
AV
1406EXPORT_SYMBOL(insert_inode_locked4);
1407
1da177e4 1408
45321ac5
AV
1409int generic_delete_inode(struct inode *inode)
1410{
1411 return 1;
1412}
1413EXPORT_SYMBOL(generic_delete_inode);
1414
45321ac5
AV
1415/*
1416 * Called when we're dropping the last reference
1417 * to an inode.
22fe4042 1418 *
45321ac5
AV
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.
22fe4042 1424 */
45321ac5 1425static void iput_final(struct inode *inode)
1da177e4
LT
1426{
1427 struct super_block *sb = inode->i_sb;
45321ac5
AV
1428 const struct super_operations *op = inode->i_sb->s_op;
1429 int drop;
1430
250df6ed
DC
1431 WARN_ON(inode->i_state & I_NEW);
1432
e7f59097 1433 if (op->drop_inode)
45321ac5
AV
1434 drop = op->drop_inode(inode);
1435 else
1436 drop = generic_drop_inode(inode);
1da177e4 1437
b2b2af8e
DC
1438 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1439 inode->i_state |= I_REFERENCED;
4eff96dd 1440 inode_add_lru(inode);
b2b2af8e 1441 spin_unlock(&inode->i_lock);
b2b2af8e
DC
1442 return;
1443 }
1444
45321ac5 1445 if (!drop) {
991114c6 1446 inode->i_state |= I_WILL_FREE;
250df6ed 1447 spin_unlock(&inode->i_lock);
1da177e4 1448 write_inode_now(inode, 1);
250df6ed 1449 spin_lock(&inode->i_lock);
7ef0d737 1450 WARN_ON(inode->i_state & I_NEW);
991114c6 1451 inode->i_state &= ~I_WILL_FREE;
1da177e4 1452 }
7ccf19a8 1453
991114c6 1454 inode->i_state |= I_FREEING;
c4ae0c65
ED
1455 if (!list_empty(&inode->i_lru))
1456 inode_lru_list_del(inode);
b2b2af8e 1457 spin_unlock(&inode->i_lock);
b2b2af8e 1458
644da596 1459 evict(inode);
1da177e4
LT
1460}
1461
1da177e4 1462/**
6b3304b5 1463 * iput - put an inode
1da177e4
LT
1464 * @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{
0ae45f63
TT
1473 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);
1da177e4
LT
1487 }
1488}
1da177e4
LT
1489EXPORT_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
6b3304b5 1499 * disk block relative to the disk start that holds that block of the
1da177e4
LT
1500 * file.
1501 */
6b3304b5 1502sector_t bmap(struct inode *inode, sector_t block)
1da177e4
LT
1503{
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}
1da177e4
LT
1509EXPORT_SYMBOL(bmap);
1510
11ff6f05
MG
1511/*
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
0ae45f63 1545int generic_update_time(struct inode *inode, struct timespec *time, int flags)
c3b2da31 1546{
0ae45f63 1547 int iflags = I_DIRTY_TIME;
c3b2da31
JB
1548
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;
0ae45f63
TT
1557
1558 if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION))
1559 iflags |= I_DIRTY_SYNC;
1560 __mark_inode_dirty(inode, iflags);
c3b2da31
JB
1561 return 0;
1562}
0ae45f63
TT
1563EXPORT_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}
c3b2da31 1578
1da177e4 1579/**
869243a0 1580 * touch_atime - update the access time
185553b2 1581 * @path: the &struct path to update
1da177e4
LT
1582 *
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 */
badcf2b7 1587void touch_atime(const struct path *path)
1da177e4 1588{
68ac1234
AV
1589 struct vfsmount *mnt = path->mnt;
1590 struct inode *inode = path->dentry->d_inode;
1da177e4
LT
1591 struct timespec now;
1592
cdb70f3f 1593 if (inode->i_flags & S_NOATIME)
b12536c2 1594 return;
37756ced 1595 if (IS_NOATIME(inode))
b12536c2 1596 return;
b2276138 1597 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
b12536c2 1598 return;
47ae32d6 1599
cdb70f3f 1600 if (mnt->mnt_flags & MNT_NOATIME)
b12536c2 1601 return;
cdb70f3f 1602 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
b12536c2 1603 return;
1da177e4
LT
1604
1605 now = current_fs_time(inode->i_sb);
11ff6f05
MG
1606
1607 if (!relatime_need_update(mnt, inode, now))
b12536c2 1608 return;
11ff6f05 1609
47ae32d6 1610 if (timespec_equal(&inode->i_atime, &now))
b12536c2
AK
1611 return;
1612
5d37e9e6 1613 if (!sb_start_write_trylock(inode->i_sb))
b12536c2 1614 return;
47ae32d6 1615
5d37e9e6
JK
1616 if (__mnt_want_write(mnt))
1617 goto skip_update;
c3b2da31
JB
1618 /*
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.
2bc55652
AB
1624 * We may also fail on filesystems that have the ability to make parts
1625 * of the fs read only, e.g. subvolumes in Btrfs.
c3b2da31
JB
1626 */
1627 update_time(inode, &now, S_ATIME);
5d37e9e6
JK
1628 __mnt_drop_write(mnt);
1629skip_update:
1630 sb_end_write(inode->i_sb);
1da177e4 1631}
869243a0 1632EXPORT_SYMBOL(touch_atime);
1da177e4 1633
3ed37648
CW
1634/*
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{
1642 umode_t mode = dentry->d_inode->i_mode;
1643 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;
27ac0ffe
BF
1668 /*
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);
3ed37648
CW
1673}
1674
1675int file_remove_suid(struct file *file)
1676{
1677 struct dentry *dentry = file->f_path.dentry;
1678 struct inode *inode = dentry->d_inode;
1679 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
1da177e4 1703/**
870f4817
CH
1704 * file_update_time - update mtime and ctime time
1705 * @file: file accessed
1da177e4 1706 *
870f4817
CH
1707 * 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
2eadfc0e 1711 * S_NOCMTIME inode flag, e.g. for network filesystem where these
c3b2da31
JB
1712 * 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.
1da177e4
LT
1714 */
1715
c3b2da31 1716int file_update_time(struct file *file)
1da177e4 1717{
496ad9aa 1718 struct inode *inode = file_inode(file);
1da177e4 1719 struct timespec now;
c3b2da31
JB
1720 int sync_it = 0;
1721 int ret;
1da177e4 1722
ce06e0b2 1723 /* First try to exhaust all avenues to not sync */
1da177e4 1724 if (IS_NOCMTIME(inode))
c3b2da31 1725 return 0;
20ddee2c 1726
1da177e4 1727 now = current_fs_time(inode->i_sb);
ce06e0b2
AK
1728 if (!timespec_equal(&inode->i_mtime, &now))
1729 sync_it = S_MTIME;
1da177e4 1730
ce06e0b2
AK
1731 if (!timespec_equal(&inode->i_ctime, &now))
1732 sync_it |= S_CTIME;
870f4817 1733
ce06e0b2
AK
1734 if (IS_I_VERSION(inode))
1735 sync_it |= S_VERSION;
7a224228 1736
ce06e0b2 1737 if (!sync_it)
c3b2da31 1738 return 0;
ce06e0b2
AK
1739
1740 /* Finally allowed to write? Takes lock. */
eb04c282 1741 if (__mnt_want_write_file(file))
c3b2da31 1742 return 0;
ce06e0b2 1743
c3b2da31 1744 ret = update_time(inode, &now, sync_it);
eb04c282 1745 __mnt_drop_write_file(file);
c3b2da31
JB
1746
1747 return ret;
1da177e4 1748}
870f4817 1749EXPORT_SYMBOL(file_update_time);
1da177e4
LT
1750
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}
1da177e4
LT
1759EXPORT_SYMBOL(inode_needs_sync);
1760
1da177e4 1761/*
168a9fd6
MS
1762 * 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 *
eaff8079 1768 * It doesn't matter if I_NEW is not set initially, a call to
250df6ed
DC
1769 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1770 * will DTRT.
1da177e4
LT
1771 */
1772static void __wait_on_freeing_inode(struct inode *inode)
1773{
1774 wait_queue_head_t *wq;
eaff8079
CH
1775 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1776 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1da177e4 1777 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
250df6ed 1778 spin_unlock(&inode->i_lock);
67a23c49 1779 spin_unlock(&inode_hash_lock);
1da177e4
LT
1780 schedule();
1781 finish_wait(wq, &wait.wait);
67a23c49 1782 spin_lock(&inode_hash_lock);
1da177e4
LT
1783}
1784
1da177e4
LT
1785static __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{
074b8517 1800 unsigned int loop;
1da177e4
LT
1801
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,
31fe62b9 1816 0,
1da177e4
LT
1817 0);
1818
074b8517 1819 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1da177e4
LT
1820 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1821}
1822
74bf17cf 1823void __init inode_init(void)
1da177e4 1824{
074b8517 1825 unsigned int loop;
1da177e4
LT
1826
1827 /* inode slab cache */
b0196009
PJ
1828 inode_cachep = kmem_cache_create("inode_cache",
1829 sizeof(struct inode),
1830 0,
1831 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1832 SLAB_MEM_SPREAD),
20c2df83 1833 init_once);
1da177e4
LT
1834
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,
31fe62b9 1847 0,
1da177e4
LT
1848 0);
1849
074b8517 1850 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1da177e4
LT
1851 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))
599a0ac1 1864 inode->i_fop = &pipefifo_fops;
1da177e4 1865 else if (S_ISSOCK(mode))
bd9b51e7 1866 ; /* leave it no_open_fops */
1da177e4 1867 else
af0d9ae8
MK
1868 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);
1da177e4
LT
1871}
1872EXPORT_SYMBOL(init_special_inode);
a1bd120d
DM
1873
1874/**
eaae668d 1875 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
a1bd120d
DM
1876 * @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,
62bb1091 1881 umode_t mode)
a1bd120d
DM
1882{
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);
e795b717 1893
2e149670
SH
1894/**
1895 * inode_owner_or_capable - check current task permissions to inode
1896 * @inode: inode being checked
1897 *
23adbe12
AL
1898 * Return true if current either has CAP_FOWNER in a namespace with the
1899 * inode owner uid mapped, or owns the file.
e795b717 1900 */
2e149670 1901bool inode_owner_or_capable(const struct inode *inode)
e795b717 1902{
23adbe12
AL
1903 struct user_namespace *ns;
1904
92361636 1905 if (uid_eq(current_fsuid(), inode->i_uid))
e795b717 1906 return true;
23adbe12
AL
1907
1908 ns = current_user_ns();
1909 if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid))
e795b717
SH
1910 return true;
1911 return false;
1912}
2e149670 1913EXPORT_SYMBOL(inode_owner_or_capable);
1d59d61f
TM
1914
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/*
1949 * inode_dio_done - signal finish of a direct I/O requests
1950 * @inode: inode the direct I/O happens on
1951 *
1952 * This is called once we've finished processing a direct I/O request,
1953 * and is used to wake up callers waiting for direct I/O to be quiesced.
1954 */
1955void inode_dio_done(struct inode *inode)
1956{
1957 if (atomic_dec_and_test(&inode->i_dio_count))
1958 wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1959}
1960EXPORT_SYMBOL(inode_dio_done);
5f16f322
TT
1961
1962/*
1963 * inode_set_flags - atomically set some inode flags
1964 *
1965 * Note: the caller should be holding i_mutex, or else be sure that
1966 * they have exclusive access to the inode structure (i.e., while the
1967 * inode is being instantiated). The reason for the cmpxchg() loop
1968 * --- which wouldn't be necessary if all code paths which modify
1969 * i_flags actually followed this rule, is that there is at least one
1970 * code path which doesn't today --- for example,
1971 * __generic_file_aio_write() calls file_remove_suid() without holding
1972 * i_mutex --- so we use cmpxchg() out of an abundance of caution.
1973 *
1974 * In the long run, i_mutex is overkill, and we should probably look
1975 * at using the i_lock spinlock to protect i_flags, and then make sure
1976 * it is so documented in include/linux/fs.h and that all code follows
1977 * the locking convention!!
1978 */
1979void inode_set_flags(struct inode *inode, unsigned int flags,
1980 unsigned int mask)
1981{
1982 unsigned int old_flags, new_flags;
1983
1984 WARN_ON_ONCE(flags & ~mask);
1985 do {
1986 old_flags = ACCESS_ONCE(inode->i_flags);
1987 new_flags = (old_flags & ~mask) | flags;
1988 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
1989 new_flags) != old_flags));
1990}
1991EXPORT_SYMBOL(inode_set_flags);