Revert "ARM: sunxi: dt: Convert to the new SID compatibles"
[linux-2.6-block.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>
a66979ab 21#include "internal.h"
1da177e4 22
250df6ed 23/*
4b4563dc 24 * Inode locking rules:
250df6ed
DC
25 *
26 * inode->i_lock protects:
27 * inode->i_state, inode->i_hash, __iget()
bc3b14cb 28 * Inode LRU list locks protect:
98b745c6 29 * inode->i_sb->s_inode_lru, inode->i_lru
55fa6091
DC
30 * inode_sb_list_lock protects:
31 * sb->s_inodes, inode->i_sb_list
f758eeab 32 * bdi->wb.list_lock protects:
a66979ab 33 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
67a23c49
DC
34 * inode_hash_lock protects:
35 * inode_hashtable, inode->i_hash
250df6ed
DC
36 *
37 * Lock ordering:
55fa6091
DC
38 *
39 * inode_sb_list_lock
40 * inode->i_lock
bc3b14cb 41 * Inode LRU list locks
a66979ab 42 *
f758eeab 43 * bdi->wb.list_lock
a66979ab 44 * inode->i_lock
67a23c49
DC
45 *
46 * inode_hash_lock
47 * inode_sb_list_lock
48 * inode->i_lock
49 *
50 * iunique_lock
51 * inode_hash_lock
250df6ed
DC
52 */
53
fa3536cc
ED
54static unsigned int i_hash_mask __read_mostly;
55static unsigned int i_hash_shift __read_mostly;
67a23c49
DC
56static struct hlist_head *inode_hashtable __read_mostly;
57static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
1da177e4 58
55fa6091
DC
59__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
60
7dcda1c9
JA
61/*
62 * Empty aops. Can be used for the cases where the user does not
63 * define any of the address_space operations.
64 */
65const struct address_space_operations empty_aops = {
66};
67EXPORT_SYMBOL(empty_aops);
68
1da177e4
LT
69/*
70 * Statistics gathering..
71 */
72struct inodes_stat_t inodes_stat;
73
3942c07c
GC
74static DEFINE_PER_CPU(unsigned long, nr_inodes);
75static DEFINE_PER_CPU(unsigned long, nr_unused);
cffbc8aa 76
6b3304b5 77static struct kmem_cache *inode_cachep __read_mostly;
1da177e4 78
3942c07c 79static long get_nr_inodes(void)
cffbc8aa 80{
3e880fb5 81 int i;
3942c07c 82 long sum = 0;
3e880fb5
NP
83 for_each_possible_cpu(i)
84 sum += per_cpu(nr_inodes, i);
85 return sum < 0 ? 0 : sum;
cffbc8aa
DC
86}
87
3942c07c 88static inline long get_nr_inodes_unused(void)
cffbc8aa 89{
fcb94f72 90 int i;
3942c07c 91 long sum = 0;
fcb94f72
DC
92 for_each_possible_cpu(i)
93 sum += per_cpu(nr_unused, i);
94 return sum < 0 ? 0 : sum;
cffbc8aa
DC
95}
96
3942c07c 97long get_nr_dirty_inodes(void)
cffbc8aa 98{
3e880fb5 99 /* not actually dirty inodes, but a wild approximation */
3942c07c 100 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
cffbc8aa 101 return nr_dirty > 0 ? nr_dirty : 0;
cffbc8aa
DC
102}
103
104/*
105 * Handle nr_inode sysctl
106 */
107#ifdef CONFIG_SYSCTL
108int proc_nr_inodes(ctl_table *table, int write,
109 void __user *buffer, size_t *lenp, loff_t *ppos)
110{
111 inodes_stat.nr_inodes = get_nr_inodes();
fcb94f72 112 inodes_stat.nr_unused = get_nr_inodes_unused();
3942c07c 113 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
cffbc8aa
DC
114}
115#endif
116
2cb1599f
DC
117/**
118 * inode_init_always - perform inode structure intialisation
0bc02f3f
RD
119 * @sb: superblock inode belongs to
120 * @inode: inode to initialise
2cb1599f
DC
121 *
122 * These are initializations that need to be done on every inode
123 * allocation as the fields are not initialised by slab allocation.
124 */
54e34621 125int inode_init_always(struct super_block *sb, struct inode *inode)
1da177e4 126{
6e1d5dcc 127 static const struct inode_operations empty_iops;
99ac48f5 128 static const struct file_operations empty_fops;
6b3304b5 129 struct address_space *const mapping = &inode->i_data;
2cb1599f
DC
130
131 inode->i_sb = sb;
132 inode->i_blkbits = sb->s_blocksize_bits;
133 inode->i_flags = 0;
134 atomic_set(&inode->i_count, 1);
135 inode->i_op = &empty_iops;
136 inode->i_fop = &empty_fops;
a78ef704 137 inode->__i_nlink = 1;
3ddcd056 138 inode->i_opflags = 0;
92361636
EB
139 i_uid_write(inode, 0);
140 i_gid_write(inode, 0);
2cb1599f
DC
141 atomic_set(&inode->i_writecount, 0);
142 inode->i_size = 0;
143 inode->i_blocks = 0;
144 inode->i_bytes = 0;
145 inode->i_generation = 0;
1da177e4 146#ifdef CONFIG_QUOTA
2cb1599f 147 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
1da177e4 148#endif
2cb1599f
DC
149 inode->i_pipe = NULL;
150 inode->i_bdev = NULL;
151 inode->i_cdev = NULL;
152 inode->i_rdev = 0;
153 inode->dirtied_when = 0;
6146f0d5
MZ
154
155 if (security_inode_alloc(inode))
54e34621 156 goto out;
2cb1599f
DC
157 spin_lock_init(&inode->i_lock);
158 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
159
160 mutex_init(&inode->i_mutex);
161 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
162
bd5fe6c5 163 atomic_set(&inode->i_dio_count, 0);
2cb1599f
DC
164
165 mapping->a_ops = &empty_aops;
166 mapping->host = inode;
167 mapping->flags = 0;
3c1d4378 168 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
252aa6f5 169 mapping->private_data = NULL;
2cb1599f
DC
170 mapping->backing_dev_info = &default_backing_dev_info;
171 mapping->writeback_index = 0;
172
173 /*
174 * If the block_device provides a backing_dev_info for client
175 * inodes then use that. Otherwise the inode share the bdev's
176 * backing_dev_info.
177 */
178 if (sb->s_bdev) {
179 struct backing_dev_info *bdi;
180
2c96ce9f 181 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
2cb1599f
DC
182 mapping->backing_dev_info = bdi;
183 }
184 inode->i_private = NULL;
185 inode->i_mapping = mapping;
b3d9b7a3 186 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
f19d4a8f
AV
187#ifdef CONFIG_FS_POSIX_ACL
188 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
189#endif
2cb1599f 190
3be25f49
EP
191#ifdef CONFIG_FSNOTIFY
192 inode->i_fsnotify_mask = 0;
193#endif
194
3e880fb5 195 this_cpu_inc(nr_inodes);
cffbc8aa 196
54e34621 197 return 0;
54e34621
CH
198out:
199 return -ENOMEM;
1da177e4 200}
2cb1599f
DC
201EXPORT_SYMBOL(inode_init_always);
202
203static struct inode *alloc_inode(struct super_block *sb)
204{
205 struct inode *inode;
206
207 if (sb->s_op->alloc_inode)
208 inode = sb->s_op->alloc_inode(sb);
209 else
210 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
211
54e34621
CH
212 if (!inode)
213 return NULL;
214
215 if (unlikely(inode_init_always(sb, inode))) {
216 if (inode->i_sb->s_op->destroy_inode)
217 inode->i_sb->s_op->destroy_inode(inode);
218 else
219 kmem_cache_free(inode_cachep, inode);
220 return NULL;
221 }
222
223 return inode;
2cb1599f 224}
1da177e4 225
ff0c7d15
NP
226void free_inode_nonrcu(struct inode *inode)
227{
228 kmem_cache_free(inode_cachep, inode);
229}
230EXPORT_SYMBOL(free_inode_nonrcu);
231
2e00c97e 232void __destroy_inode(struct inode *inode)
1da177e4 233{
b7542f8c 234 BUG_ON(inode_has_buffers(inode));
1da177e4 235 security_inode_free(inode);
3be25f49 236 fsnotify_inode_delete(inode);
7ada4db8
MS
237 if (!inode->i_nlink) {
238 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
239 atomic_long_dec(&inode->i_sb->s_remove_count);
240 }
241
f19d4a8f
AV
242#ifdef CONFIG_FS_POSIX_ACL
243 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
244 posix_acl_release(inode->i_acl);
245 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
246 posix_acl_release(inode->i_default_acl);
247#endif
3e880fb5 248 this_cpu_dec(nr_inodes);
2e00c97e
CH
249}
250EXPORT_SYMBOL(__destroy_inode);
251
fa0d7e3d
NP
252static void i_callback(struct rcu_head *head)
253{
254 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP
255 kmem_cache_free(inode_cachep, inode);
256}
257
56b0dacf 258static void destroy_inode(struct inode *inode)
2e00c97e 259{
7ccf19a8 260 BUG_ON(!list_empty(&inode->i_lru));
2e00c97e 261 __destroy_inode(inode);
1da177e4
LT
262 if (inode->i_sb->s_op->destroy_inode)
263 inode->i_sb->s_op->destroy_inode(inode);
264 else
fa0d7e3d 265 call_rcu(&inode->i_rcu, i_callback);
1da177e4 266}
1da177e4 267
7ada4db8
MS
268/**
269 * drop_nlink - directly drop an inode's link count
270 * @inode: inode
271 *
272 * This is a low-level filesystem helper to replace any
273 * direct filesystem manipulation of i_nlink. In cases
274 * where we are attempting to track writes to the
275 * filesystem, a decrement to zero means an imminent
276 * write when the file is truncated and actually unlinked
277 * on the filesystem.
278 */
279void drop_nlink(struct inode *inode)
280{
281 WARN_ON(inode->i_nlink == 0);
282 inode->__i_nlink--;
283 if (!inode->i_nlink)
284 atomic_long_inc(&inode->i_sb->s_remove_count);
285}
286EXPORT_SYMBOL(drop_nlink);
287
288/**
289 * clear_nlink - directly zero an inode's link count
290 * @inode: inode
291 *
292 * This is a low-level filesystem helper to replace any
293 * direct filesystem manipulation of i_nlink. See
294 * drop_nlink() for why we care about i_nlink hitting zero.
295 */
296void clear_nlink(struct inode *inode)
297{
298 if (inode->i_nlink) {
299 inode->__i_nlink = 0;
300 atomic_long_inc(&inode->i_sb->s_remove_count);
301 }
302}
303EXPORT_SYMBOL(clear_nlink);
304
305/**
306 * set_nlink - directly set an inode's link count
307 * @inode: inode
308 * @nlink: new nlink (should be non-zero)
309 *
310 * This is a low-level filesystem helper to replace any
311 * direct filesystem manipulation of i_nlink.
312 */
313void set_nlink(struct inode *inode, unsigned int nlink)
314{
315 if (!nlink) {
7ada4db8
MS
316 clear_nlink(inode);
317 } else {
318 /* Yes, some filesystems do change nlink from zero to one */
319 if (inode->i_nlink == 0)
320 atomic_long_dec(&inode->i_sb->s_remove_count);
321
322 inode->__i_nlink = nlink;
323 }
324}
325EXPORT_SYMBOL(set_nlink);
326
327/**
328 * inc_nlink - directly increment an inode's link count
329 * @inode: inode
330 *
331 * This is a low-level filesystem helper to replace any
332 * direct filesystem manipulation of i_nlink. Currently,
333 * it is only here for parity with dec_nlink().
334 */
335void inc_nlink(struct inode *inode)
336{
f4e0c30c
AV
337 if (unlikely(inode->i_nlink == 0)) {
338 WARN_ON(!(inode->i_state & I_LINKABLE));
7ada4db8 339 atomic_long_dec(&inode->i_sb->s_remove_count);
f4e0c30c 340 }
7ada4db8
MS
341
342 inode->__i_nlink++;
343}
344EXPORT_SYMBOL(inc_nlink);
345
2aa15890
MS
346void address_space_init_once(struct address_space *mapping)
347{
348 memset(mapping, 0, sizeof(*mapping));
349 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
350 spin_lock_init(&mapping->tree_lock);
3d48ae45 351 mutex_init(&mapping->i_mmap_mutex);
2aa15890
MS
352 INIT_LIST_HEAD(&mapping->private_list);
353 spin_lock_init(&mapping->private_lock);
6b2dbba8 354 mapping->i_mmap = RB_ROOT;
2aa15890 355 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
2aa15890
MS
356}
357EXPORT_SYMBOL(address_space_init_once);
358
1da177e4
LT
359/*
360 * These are initializations that only need to be done
361 * once, because the fields are idempotent across use
362 * of the inode, so let the slab aware of that.
363 */
364void inode_init_once(struct inode *inode)
365{
366 memset(inode, 0, sizeof(*inode));
367 INIT_HLIST_NODE(&inode->i_hash);
1da177e4 368 INIT_LIST_HEAD(&inode->i_devices);
7ccf19a8
NP
369 INIT_LIST_HEAD(&inode->i_wb_list);
370 INIT_LIST_HEAD(&inode->i_lru);
2aa15890 371 address_space_init_once(&inode->i_data);
1da177e4 372 i_size_ordered_init(inode);
3be25f49 373#ifdef CONFIG_FSNOTIFY
e61ce867 374 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
3be25f49 375#endif
1da177e4 376}
1da177e4
LT
377EXPORT_SYMBOL(inode_init_once);
378
51cc5068 379static void init_once(void *foo)
1da177e4 380{
6b3304b5 381 struct inode *inode = (struct inode *) foo;
1da177e4 382
a35afb83 383 inode_init_once(inode);
1da177e4
LT
384}
385
386/*
250df6ed 387 * inode->i_lock must be held
1da177e4 388 */
6b3304b5 389void __iget(struct inode *inode)
1da177e4 390{
9e38d86f
NP
391 atomic_inc(&inode->i_count);
392}
2e147f1e 393
7de9c6ee
AV
394/*
395 * get additional reference to inode; caller must already hold one.
396 */
397void ihold(struct inode *inode)
398{
399 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
400}
401EXPORT_SYMBOL(ihold);
402
9e38d86f
NP
403static void inode_lru_list_add(struct inode *inode)
404{
bc3b14cb 405 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
fcb94f72 406 this_cpu_inc(nr_unused);
9e38d86f 407}
2e147f1e 408
4eff96dd
JK
409/*
410 * Add inode to LRU if needed (inode is unused and clean).
411 *
412 * Needs inode->i_lock held.
413 */
414void inode_add_lru(struct inode *inode)
415{
416 if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) &&
417 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
418 inode_lru_list_add(inode);
419}
420
421
9e38d86f
NP
422static void inode_lru_list_del(struct inode *inode)
423{
bc3b14cb
DC
424
425 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
fcb94f72 426 this_cpu_dec(nr_unused);
1da177e4
LT
427}
428
646ec461
CH
429/**
430 * inode_sb_list_add - add inode to the superblock list of inodes
431 * @inode: inode to add
432 */
433void inode_sb_list_add(struct inode *inode)
434{
55fa6091
DC
435 spin_lock(&inode_sb_list_lock);
436 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
437 spin_unlock(&inode_sb_list_lock);
646ec461
CH
438}
439EXPORT_SYMBOL_GPL(inode_sb_list_add);
440
55fa6091 441static inline void inode_sb_list_del(struct inode *inode)
646ec461 442{
a209dfc7
ED
443 if (!list_empty(&inode->i_sb_list)) {
444 spin_lock(&inode_sb_list_lock);
445 list_del_init(&inode->i_sb_list);
446 spin_unlock(&inode_sb_list_lock);
447 }
646ec461
CH
448}
449
4c51acbc
DC
450static unsigned long hash(struct super_block *sb, unsigned long hashval)
451{
452 unsigned long tmp;
453
454 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
455 L1_CACHE_BYTES;
4b4563dc
CH
456 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
457 return tmp & i_hash_mask;
4c51acbc
DC
458}
459
460/**
461 * __insert_inode_hash - hash an inode
462 * @inode: unhashed inode
463 * @hashval: unsigned long value used to locate this object in the
464 * inode_hashtable.
465 *
466 * Add an inode to the inode hash for this superblock.
467 */
468void __insert_inode_hash(struct inode *inode, unsigned long hashval)
469{
646ec461
CH
470 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
471
67a23c49 472 spin_lock(&inode_hash_lock);
250df6ed 473 spin_lock(&inode->i_lock);
646ec461 474 hlist_add_head(&inode->i_hash, b);
250df6ed 475 spin_unlock(&inode->i_lock);
67a23c49 476 spin_unlock(&inode_hash_lock);
4c51acbc
DC
477}
478EXPORT_SYMBOL(__insert_inode_hash);
479
4c51acbc 480/**
f2ee7abf 481 * __remove_inode_hash - remove an inode from the hash
4c51acbc
DC
482 * @inode: inode to unhash
483 *
484 * Remove an inode from the superblock.
485 */
f2ee7abf 486void __remove_inode_hash(struct inode *inode)
4c51acbc 487{
67a23c49 488 spin_lock(&inode_hash_lock);
250df6ed 489 spin_lock(&inode->i_lock);
4c51acbc 490 hlist_del_init(&inode->i_hash);
250df6ed 491 spin_unlock(&inode->i_lock);
67a23c49 492 spin_unlock(&inode_hash_lock);
4c51acbc 493}
f2ee7abf 494EXPORT_SYMBOL(__remove_inode_hash);
4c51acbc 495
dbd5768f 496void clear_inode(struct inode *inode)
b0683aa6
AV
497{
498 might_sleep();
08142579
JK
499 /*
500 * We have to cycle tree_lock here because reclaim can be still in the
501 * process of removing the last page (in __delete_from_page_cache())
502 * and we must not free mapping under it.
503 */
504 spin_lock_irq(&inode->i_data.tree_lock);
b0683aa6 505 BUG_ON(inode->i_data.nrpages);
08142579 506 spin_unlock_irq(&inode->i_data.tree_lock);
b0683aa6
AV
507 BUG_ON(!list_empty(&inode->i_data.private_list));
508 BUG_ON(!(inode->i_state & I_FREEING));
509 BUG_ON(inode->i_state & I_CLEAR);
fa0d7e3d 510 /* don't need i_lock here, no concurrent mods to i_state */
b0683aa6
AV
511 inode->i_state = I_FREEING | I_CLEAR;
512}
dbd5768f 513EXPORT_SYMBOL(clear_inode);
b0683aa6 514
b2b2af8e
DC
515/*
516 * Free the inode passed in, removing it from the lists it is still connected
517 * to. We remove any pages still attached to the inode and wait for any IO that
518 * is still in progress before finally destroying the inode.
519 *
520 * An inode must already be marked I_FREEING so that we avoid the inode being
521 * moved back onto lists if we race with other code that manipulates the lists
522 * (e.g. writeback_single_inode). The caller is responsible for setting this.
523 *
524 * An inode must already be removed from the LRU list before being evicted from
525 * the cache. This should occur atomically with setting the I_FREEING state
526 * flag, so no inodes here should ever be on the LRU when being evicted.
527 */
644da596 528static void evict(struct inode *inode)
b4272d4c
AV
529{
530 const struct super_operations *op = inode->i_sb->s_op;
531
b2b2af8e
DC
532 BUG_ON(!(inode->i_state & I_FREEING));
533 BUG_ON(!list_empty(&inode->i_lru));
534
b12362bd
ED
535 if (!list_empty(&inode->i_wb_list))
536 inode_wb_list_del(inode);
537
55fa6091
DC
538 inode_sb_list_del(inode);
539
169ebd90
JK
540 /*
541 * Wait for flusher thread to be done with the inode so that filesystem
542 * does not start destroying it while writeback is still running. Since
543 * the inode has I_FREEING set, flusher thread won't start new work on
544 * the inode. We just have to wait for running writeback to finish.
545 */
546 inode_wait_for_writeback(inode);
7994e6f7 547
be7ce416
AV
548 if (op->evict_inode) {
549 op->evict_inode(inode);
b4272d4c
AV
550 } else {
551 if (inode->i_data.nrpages)
552 truncate_inode_pages(&inode->i_data, 0);
dbd5768f 553 clear_inode(inode);
b4272d4c 554 }
661074e9
AV
555 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
556 bd_forget(inode);
557 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
558 cd_forget(inode);
b2b2af8e
DC
559
560 remove_inode_hash(inode);
561
562 spin_lock(&inode->i_lock);
563 wake_up_bit(&inode->i_state, __I_NEW);
564 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
565 spin_unlock(&inode->i_lock);
566
567 destroy_inode(inode);
b4272d4c
AV
568}
569
1da177e4
LT
570/*
571 * dispose_list - dispose of the contents of a local list
572 * @head: the head of the list to free
573 *
574 * Dispose-list gets a local list with local inodes in it, so it doesn't
575 * need to worry about list corruption and SMP locks.
576 */
577static void dispose_list(struct list_head *head)
578{
1da177e4
LT
579 while (!list_empty(head)) {
580 struct inode *inode;
581
7ccf19a8
NP
582 inode = list_first_entry(head, struct inode, i_lru);
583 list_del_init(&inode->i_lru);
1da177e4 584
644da596 585 evict(inode);
1da177e4 586 }
1da177e4
LT
587}
588
63997e98
AV
589/**
590 * evict_inodes - evict all evictable inodes for a superblock
591 * @sb: superblock to operate on
592 *
593 * Make sure that no inodes with zero refcount are retained. This is
594 * called by superblock shutdown after having MS_ACTIVE flag removed,
595 * so any inode reaching zero refcount during or after that call will
596 * be immediately evicted.
1da177e4 597 */
63997e98 598void evict_inodes(struct super_block *sb)
1da177e4 599{
63997e98
AV
600 struct inode *inode, *next;
601 LIST_HEAD(dispose);
1da177e4 602
55fa6091 603 spin_lock(&inode_sb_list_lock);
63997e98
AV
604 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
605 if (atomic_read(&inode->i_count))
aabb8fdb 606 continue;
250df6ed
DC
607
608 spin_lock(&inode->i_lock);
609 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
610 spin_unlock(&inode->i_lock);
1da177e4 611 continue;
250df6ed 612 }
63997e98
AV
613
614 inode->i_state |= I_FREEING;
02afc410 615 inode_lru_list_del(inode);
250df6ed 616 spin_unlock(&inode->i_lock);
02afc410 617 list_add(&inode->i_lru, &dispose);
1da177e4 618 }
55fa6091 619 spin_unlock(&inode_sb_list_lock);
63997e98
AV
620
621 dispose_list(&dispose);
1da177e4
LT
622}
623
1da177e4 624/**
a0318786
CH
625 * invalidate_inodes - attempt to free all inodes on a superblock
626 * @sb: superblock to operate on
93b270f7 627 * @kill_dirty: flag to guide handling of dirty inodes
1da177e4 628 *
a0318786
CH
629 * Attempts to free all inodes for a given superblock. If there were any
630 * busy inodes return a non-zero value, else zero.
93b270f7
N
631 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
632 * them as busy.
1da177e4 633 */
93b270f7 634int invalidate_inodes(struct super_block *sb, bool kill_dirty)
1da177e4 635{
cffbc8aa 636 int busy = 0;
a0318786
CH
637 struct inode *inode, *next;
638 LIST_HEAD(dispose);
1da177e4 639
55fa6091 640 spin_lock(&inode_sb_list_lock);
a0318786 641 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
250df6ed
DC
642 spin_lock(&inode->i_lock);
643 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
644 spin_unlock(&inode->i_lock);
aabb8fdb 645 continue;
250df6ed 646 }
93b270f7 647 if (inode->i_state & I_DIRTY && !kill_dirty) {
250df6ed 648 spin_unlock(&inode->i_lock);
93b270f7
N
649 busy = 1;
650 continue;
651 }
99a38919 652 if (atomic_read(&inode->i_count)) {
250df6ed 653 spin_unlock(&inode->i_lock);
99a38919 654 busy = 1;
1da177e4
LT
655 continue;
656 }
99a38919 657
99a38919 658 inode->i_state |= I_FREEING;
02afc410 659 inode_lru_list_del(inode);
250df6ed 660 spin_unlock(&inode->i_lock);
02afc410 661 list_add(&inode->i_lru, &dispose);
1da177e4 662 }
55fa6091 663 spin_unlock(&inode_sb_list_lock);
1da177e4 664
a0318786 665 dispose_list(&dispose);
1da177e4
LT
666
667 return busy;
668}
1da177e4 669
1da177e4 670/*
bc3b14cb 671 * Isolate the inode from the LRU in preparation for freeing it.
1da177e4
LT
672 *
673 * Any inodes which are pinned purely because of attached pagecache have their
9e38d86f
NP
674 * pagecache removed. If the inode has metadata buffers attached to
675 * mapping->private_list then try to remove them.
1da177e4 676 *
9e38d86f
NP
677 * If the inode has the I_REFERENCED flag set, then it means that it has been
678 * used recently - the flag is set in iput_final(). When we encounter such an
679 * inode, clear the flag and move it to the back of the LRU so it gets another
680 * pass through the LRU before it gets reclaimed. This is necessary because of
681 * the fact we are doing lazy LRU updates to minimise lock contention so the
682 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
683 * with this flag set because they are the inodes that are out of order.
1da177e4 684 */
bc3b14cb
DC
685static enum lru_status
686inode_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
1da177e4 687{
bc3b14cb
DC
688 struct list_head *freeable = arg;
689 struct inode *inode = container_of(item, struct inode, i_lru);
1da177e4 690
bc3b14cb
DC
691 /*
692 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
693 * If we fail to get the lock, just skip it.
694 */
695 if (!spin_trylock(&inode->i_lock))
696 return LRU_SKIP;
1da177e4 697
bc3b14cb
DC
698 /*
699 * Referenced or dirty inodes are still in use. Give them another pass
700 * through the LRU as we canot reclaim them now.
701 */
702 if (atomic_read(&inode->i_count) ||
703 (inode->i_state & ~I_REFERENCED)) {
704 list_del_init(&inode->i_lru);
705 spin_unlock(&inode->i_lock);
706 this_cpu_dec(nr_unused);
707 return LRU_REMOVED;
708 }
1da177e4 709
bc3b14cb
DC
710 /* recently referenced inodes get one more pass */
711 if (inode->i_state & I_REFERENCED) {
712 inode->i_state &= ~I_REFERENCED;
713 spin_unlock(&inode->i_lock);
714 return LRU_ROTATE;
715 }
1da177e4 716
bc3b14cb
DC
717 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
718 __iget(inode);
719 spin_unlock(&inode->i_lock);
720 spin_unlock(lru_lock);
721 if (remove_inode_buffers(inode)) {
722 unsigned long reap;
723 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
724 if (current_is_kswapd())
725 __count_vm_events(KSWAPD_INODESTEAL, reap);
726 else
727 __count_vm_events(PGINODESTEAL, reap);
728 if (current->reclaim_state)
729 current->reclaim_state->reclaimed_slab += reap;
02afc410 730 }
bc3b14cb
DC
731 iput(inode);
732 spin_lock(lru_lock);
733 return LRU_RETRY;
734 }
02afc410 735
bc3b14cb
DC
736 WARN_ON(inode->i_state & I_NEW);
737 inode->i_state |= I_FREEING;
d38fa698 738 list_move(&inode->i_lru, freeable);
bc3b14cb 739 spin_unlock(&inode->i_lock);
9e38d86f 740
bc3b14cb
DC
741 this_cpu_dec(nr_unused);
742 return LRU_REMOVED;
743}
7ccf19a8 744
bc3b14cb
DC
745/*
746 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
747 * This is called from the superblock shrinker function with a number of inodes
748 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
749 * then are freed outside inode_lock by dispose_list().
750 */
9b17c623
DC
751long prune_icache_sb(struct super_block *sb, unsigned long nr_to_scan,
752 int nid)
bc3b14cb
DC
753{
754 LIST_HEAD(freeable);
755 long freed;
1da177e4 756
9b17c623
DC
757 freed = list_lru_walk_node(&sb->s_inode_lru, nid, inode_lru_isolate,
758 &freeable, &nr_to_scan);
1da177e4 759 dispose_list(&freeable);
0a234c6d 760 return freed;
1da177e4
LT
761}
762
1da177e4
LT
763static void __wait_on_freeing_inode(struct inode *inode);
764/*
765 * Called with the inode lock held.
1da177e4 766 */
6b3304b5
MK
767static struct inode *find_inode(struct super_block *sb,
768 struct hlist_head *head,
769 int (*test)(struct inode *, void *),
770 void *data)
1da177e4 771{
6b3304b5 772 struct inode *inode = NULL;
1da177e4
LT
773
774repeat:
b67bfe0d 775 hlist_for_each_entry(inode, head, i_hash) {
5a3cd992 776 if (inode->i_sb != sb)
1da177e4 777 continue;
5a3cd992 778 if (!test(inode, data))
1da177e4 779 continue;
5a3cd992 780 spin_lock(&inode->i_lock);
a4ffdde6 781 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1da177e4
LT
782 __wait_on_freeing_inode(inode);
783 goto repeat;
784 }
f7899bd5 785 __iget(inode);
250df6ed 786 spin_unlock(&inode->i_lock);
f7899bd5 787 return inode;
1da177e4 788 }
f7899bd5 789 return NULL;
1da177e4
LT
790}
791
792/*
793 * find_inode_fast is the fast path version of find_inode, see the comment at
794 * iget_locked for details.
795 */
6b3304b5
MK
796static struct inode *find_inode_fast(struct super_block *sb,
797 struct hlist_head *head, unsigned long ino)
1da177e4 798{
6b3304b5 799 struct inode *inode = NULL;
1da177e4
LT
800
801repeat:
b67bfe0d 802 hlist_for_each_entry(inode, head, i_hash) {
5a3cd992 803 if (inode->i_ino != ino)
1da177e4 804 continue;
5a3cd992 805 if (inode->i_sb != sb)
1da177e4 806 continue;
5a3cd992 807 spin_lock(&inode->i_lock);
a4ffdde6 808 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1da177e4
LT
809 __wait_on_freeing_inode(inode);
810 goto repeat;
811 }
f7899bd5 812 __iget(inode);
250df6ed 813 spin_unlock(&inode->i_lock);
f7899bd5 814 return inode;
1da177e4 815 }
f7899bd5 816 return NULL;
8290c35f
DC
817}
818
f991bd2e
ED
819/*
820 * Each cpu owns a range of LAST_INO_BATCH numbers.
821 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
822 * to renew the exhausted range.
8290c35f 823 *
f991bd2e
ED
824 * This does not significantly increase overflow rate because every CPU can
825 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
826 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
827 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
828 * overflow rate by 2x, which does not seem too significant.
829 *
830 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
831 * error if st_ino won't fit in target struct field. Use 32bit counter
832 * here to attempt to avoid that.
8290c35f 833 */
f991bd2e
ED
834#define LAST_INO_BATCH 1024
835static DEFINE_PER_CPU(unsigned int, last_ino);
836
85fe4025 837unsigned int get_next_ino(void)
8290c35f 838{
f991bd2e
ED
839 unsigned int *p = &get_cpu_var(last_ino);
840 unsigned int res = *p;
8290c35f 841
f991bd2e
ED
842#ifdef CONFIG_SMP
843 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
844 static atomic_t shared_last_ino;
845 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
846
847 res = next - LAST_INO_BATCH;
848 }
849#endif
850
851 *p = ++res;
852 put_cpu_var(last_ino);
853 return res;
8290c35f 854}
85fe4025 855EXPORT_SYMBOL(get_next_ino);
8290c35f 856
a209dfc7
ED
857/**
858 * new_inode_pseudo - obtain an inode
859 * @sb: superblock
860 *
861 * Allocates a new inode for given superblock.
862 * Inode wont be chained in superblock s_inodes list
863 * This means :
864 * - fs can't be unmount
865 * - quotas, fsnotify, writeback can't work
866 */
867struct inode *new_inode_pseudo(struct super_block *sb)
868{
869 struct inode *inode = alloc_inode(sb);
870
871 if (inode) {
872 spin_lock(&inode->i_lock);
873 inode->i_state = 0;
874 spin_unlock(&inode->i_lock);
875 INIT_LIST_HEAD(&inode->i_sb_list);
876 }
877 return inode;
878}
879
1da177e4
LT
880/**
881 * new_inode - obtain an inode
882 * @sb: superblock
883 *
769848c0 884 * Allocates a new inode for given superblock. The default gfp_mask
3c1d4378 885 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
769848c0
MG
886 * If HIGHMEM pages are unsuitable or it is known that pages allocated
887 * for the page cache are not reclaimable or migratable,
888 * mapping_set_gfp_mask() must be called with suitable flags on the
889 * newly created inode's mapping
890 *
1da177e4
LT
891 */
892struct inode *new_inode(struct super_block *sb)
893{
6b3304b5 894 struct inode *inode;
1da177e4 895
55fa6091 896 spin_lock_prefetch(&inode_sb_list_lock);
6b3304b5 897
a209dfc7
ED
898 inode = new_inode_pseudo(sb);
899 if (inode)
55fa6091 900 inode_sb_list_add(inode);
1da177e4
LT
901 return inode;
902}
1da177e4
LT
903EXPORT_SYMBOL(new_inode);
904
14358e6d 905#ifdef CONFIG_DEBUG_LOCK_ALLOC
e096d0c7
JB
906void lockdep_annotate_inode_mutex_key(struct inode *inode)
907{
a3314a0e 908 if (S_ISDIR(inode->i_mode)) {
1e89a5e1
PZ
909 struct file_system_type *type = inode->i_sb->s_type;
910
9a7aa12f 911 /* Set new key only if filesystem hasn't already changed it */
978d6d8c 912 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
9a7aa12f
JK
913 /*
914 * ensure nobody is actually holding i_mutex
915 */
916 mutex_destroy(&inode->i_mutex);
917 mutex_init(&inode->i_mutex);
918 lockdep_set_class(&inode->i_mutex,
919 &type->i_mutex_dir_key);
920 }
1e89a5e1 921 }
e096d0c7
JB
922}
923EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
14358e6d 924#endif
e096d0c7
JB
925
926/**
927 * unlock_new_inode - clear the I_NEW state and wake up any waiters
928 * @inode: new inode to unlock
929 *
930 * Called when the inode is fully initialised to clear the new state of the
931 * inode and wake up anyone waiting for the inode to finish initialisation.
932 */
933void unlock_new_inode(struct inode *inode)
934{
935 lockdep_annotate_inode_mutex_key(inode);
250df6ed 936 spin_lock(&inode->i_lock);
eaff8079
CH
937 WARN_ON(!(inode->i_state & I_NEW));
938 inode->i_state &= ~I_NEW;
310fa7a3 939 smp_mb();
250df6ed
DC
940 wake_up_bit(&inode->i_state, __I_NEW);
941 spin_unlock(&inode->i_lock);
1da177e4 942}
1da177e4
LT
943EXPORT_SYMBOL(unlock_new_inode);
944
375e289e
BF
945/**
946 * lock_two_nondirectories - take two i_mutexes on non-directory objects
947 * @inode1: first inode to lock
948 * @inode2: second inode to lock
949 */
950void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
951{
952 WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
953 if (inode1 == inode2 || !inode2) {
27555516 954 mutex_lock(&inode1->i_mutex);
375e289e
BF
955 return;
956 }
957 WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
958 if (inode1 < inode2) {
27555516 959 mutex_lock(&inode1->i_mutex);
40bd22c9 960 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_NONDIR2);
375e289e 961 } else {
27555516 962 mutex_lock(&inode2->i_mutex);
40bd22c9 963 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_NONDIR2);
375e289e
BF
964 }
965}
966EXPORT_SYMBOL(lock_two_nondirectories);
967
968/**
969 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
970 * @inode1: first inode to unlock
971 * @inode2: second inode to unlock
972 */
973void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
974{
975 mutex_unlock(&inode1->i_mutex);
976 if (inode2 && inode2 != inode1)
977 mutex_unlock(&inode2->i_mutex);
978}
979EXPORT_SYMBOL(unlock_two_nondirectories);
980
0b2d0724
CH
981/**
982 * iget5_locked - obtain an inode from a mounted file system
983 * @sb: super block of file system
984 * @hashval: hash value (usually inode number) to get
985 * @test: callback used for comparisons between inodes
986 * @set: callback used to initialize a new struct inode
987 * @data: opaque data pointer to pass to @test and @set
988 *
989 * Search for the inode specified by @hashval and @data in the inode cache,
990 * and if present it is return it with an increased reference count. This is
991 * a generalized version of iget_locked() for file systems where the inode
992 * number is not sufficient for unique identification of an inode.
993 *
994 * If the inode is not in cache, allocate a new inode and return it locked,
995 * hashed, and with the I_NEW flag set. The file system gets to fill it in
996 * before unlocking it via unlock_new_inode().
1da177e4 997 *
0b2d0724
CH
998 * Note both @test and @set are called with the inode_hash_lock held, so can't
999 * sleep.
1da177e4 1000 */
0b2d0724
CH
1001struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1002 int (*test)(struct inode *, void *),
1003 int (*set)(struct inode *, void *), void *data)
1da177e4 1004{
0b2d0724 1005 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
6b3304b5 1006 struct inode *inode;
1da177e4 1007
0b2d0724
CH
1008 spin_lock(&inode_hash_lock);
1009 inode = find_inode(sb, head, test, data);
1010 spin_unlock(&inode_hash_lock);
1011
1012 if (inode) {
1013 wait_on_inode(inode);
1014 return inode;
1015 }
1016
1da177e4
LT
1017 inode = alloc_inode(sb);
1018 if (inode) {
6b3304b5 1019 struct inode *old;
1da177e4 1020
67a23c49 1021 spin_lock(&inode_hash_lock);
1da177e4
LT
1022 /* We released the lock, so.. */
1023 old = find_inode(sb, head, test, data);
1024 if (!old) {
1025 if (set(inode, data))
1026 goto set_failed;
1027
250df6ed
DC
1028 spin_lock(&inode->i_lock);
1029 inode->i_state = I_NEW;
646ec461 1030 hlist_add_head(&inode->i_hash, head);
250df6ed 1031 spin_unlock(&inode->i_lock);
55fa6091 1032 inode_sb_list_add(inode);
67a23c49 1033 spin_unlock(&inode_hash_lock);
1da177e4
LT
1034
1035 /* Return the locked inode with I_NEW set, the
1036 * caller is responsible for filling in the contents
1037 */
1038 return inode;
1039 }
1040
1041 /*
1042 * Uhhuh, somebody else created the same inode under
1043 * us. Use the old inode instead of the one we just
1044 * allocated.
1045 */
67a23c49 1046 spin_unlock(&inode_hash_lock);
1da177e4
LT
1047 destroy_inode(inode);
1048 inode = old;
1049 wait_on_inode(inode);
1050 }
1051 return inode;
1052
1053set_failed:
67a23c49 1054 spin_unlock(&inode_hash_lock);
1da177e4
LT
1055 destroy_inode(inode);
1056 return NULL;
1057}
0b2d0724 1058EXPORT_SYMBOL(iget5_locked);
1da177e4 1059
0b2d0724
CH
1060/**
1061 * iget_locked - obtain an inode from a mounted file system
1062 * @sb: super block of file system
1063 * @ino: inode number to get
1064 *
1065 * Search for the inode specified by @ino in the inode cache and if present
1066 * return it with an increased reference count. This is for file systems
1067 * where the inode number is sufficient for unique identification of an inode.
1068 *
1069 * If the inode is not in cache, allocate a new inode and return it locked,
1070 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1071 * before unlocking it via unlock_new_inode().
1da177e4 1072 */
0b2d0724 1073struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1da177e4 1074{
0b2d0724 1075 struct hlist_head *head = inode_hashtable + hash(sb, ino);
6b3304b5 1076 struct inode *inode;
1da177e4 1077
0b2d0724
CH
1078 spin_lock(&inode_hash_lock);
1079 inode = find_inode_fast(sb, head, ino);
1080 spin_unlock(&inode_hash_lock);
1081 if (inode) {
1082 wait_on_inode(inode);
1083 return inode;
1084 }
1085
1da177e4
LT
1086 inode = alloc_inode(sb);
1087 if (inode) {
6b3304b5 1088 struct inode *old;
1da177e4 1089
67a23c49 1090 spin_lock(&inode_hash_lock);
1da177e4
LT
1091 /* We released the lock, so.. */
1092 old = find_inode_fast(sb, head, ino);
1093 if (!old) {
1094 inode->i_ino = ino;
250df6ed
DC
1095 spin_lock(&inode->i_lock);
1096 inode->i_state = I_NEW;
646ec461 1097 hlist_add_head(&inode->i_hash, head);
250df6ed 1098 spin_unlock(&inode->i_lock);
55fa6091 1099 inode_sb_list_add(inode);
67a23c49 1100 spin_unlock(&inode_hash_lock);
1da177e4
LT
1101
1102 /* Return the locked inode with I_NEW set, the
1103 * caller is responsible for filling in the contents
1104 */
1105 return inode;
1106 }
1107
1108 /*
1109 * Uhhuh, somebody else created the same inode under
1110 * us. Use the old inode instead of the one we just
1111 * allocated.
1112 */
67a23c49 1113 spin_unlock(&inode_hash_lock);
1da177e4
LT
1114 destroy_inode(inode);
1115 inode = old;
1116 wait_on_inode(inode);
1117 }
1118 return inode;
1119}
0b2d0724 1120EXPORT_SYMBOL(iget_locked);
1da177e4 1121
ad5e195a
CH
1122/*
1123 * search the inode cache for a matching inode number.
1124 * If we find one, then the inode number we are trying to
1125 * allocate is not unique and so we should not use it.
1126 *
1127 * Returns 1 if the inode number is unique, 0 if it is not.
1128 */
1129static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1130{
1131 struct hlist_head *b = inode_hashtable + hash(sb, ino);
ad5e195a
CH
1132 struct inode *inode;
1133
67a23c49 1134 spin_lock(&inode_hash_lock);
b67bfe0d 1135 hlist_for_each_entry(inode, b, i_hash) {
67a23c49
DC
1136 if (inode->i_ino == ino && inode->i_sb == sb) {
1137 spin_unlock(&inode_hash_lock);
ad5e195a 1138 return 0;
67a23c49 1139 }
ad5e195a 1140 }
67a23c49 1141 spin_unlock(&inode_hash_lock);
ad5e195a
CH
1142
1143 return 1;
1144}
1145
1da177e4
LT
1146/**
1147 * iunique - get a unique inode number
1148 * @sb: superblock
1149 * @max_reserved: highest reserved inode number
1150 *
1151 * Obtain an inode number that is unique on the system for a given
1152 * superblock. This is used by file systems that have no natural
1153 * permanent inode numbering system. An inode number is returned that
1154 * is higher than the reserved limit but unique.
1155 *
1156 * BUGS:
1157 * With a large number of inodes live on the file system this function
1158 * currently becomes quite slow.
1159 */
1160ino_t iunique(struct super_block *sb, ino_t max_reserved)
1161{
866b04fc
JL
1162 /*
1163 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1164 * error if st_ino won't fit in target struct field. Use 32bit counter
1165 * here to attempt to avoid that.
1166 */
ad5e195a 1167 static DEFINE_SPINLOCK(iunique_lock);
866b04fc 1168 static unsigned int counter;
1da177e4 1169 ino_t res;
3361c7be 1170
ad5e195a 1171 spin_lock(&iunique_lock);
3361c7be
JL
1172 do {
1173 if (counter <= max_reserved)
1174 counter = max_reserved + 1;
1da177e4 1175 res = counter++;
ad5e195a
CH
1176 } while (!test_inode_iunique(sb, res));
1177 spin_unlock(&iunique_lock);
1da177e4 1178
3361c7be
JL
1179 return res;
1180}
1da177e4
LT
1181EXPORT_SYMBOL(iunique);
1182
1183struct inode *igrab(struct inode *inode)
1184{
250df6ed
DC
1185 spin_lock(&inode->i_lock);
1186 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1da177e4 1187 __iget(inode);
250df6ed
DC
1188 spin_unlock(&inode->i_lock);
1189 } else {
1190 spin_unlock(&inode->i_lock);
1da177e4
LT
1191 /*
1192 * Handle the case where s_op->clear_inode is not been
1193 * called yet, and somebody is calling igrab
1194 * while the inode is getting freed.
1195 */
1196 inode = NULL;
250df6ed 1197 }
1da177e4
LT
1198 return inode;
1199}
1da177e4
LT
1200EXPORT_SYMBOL(igrab);
1201
1202/**
0b2d0724 1203 * ilookup5_nowait - search for an inode in the inode cache
1da177e4 1204 * @sb: super block of file system to search
0b2d0724 1205 * @hashval: hash value (usually inode number) to search for
1da177e4
LT
1206 * @test: callback used for comparisons between inodes
1207 * @data: opaque data pointer to pass to @test
1da177e4 1208 *
0b2d0724 1209 * Search for the inode specified by @hashval and @data in the inode cache.
1da177e4
LT
1210 * If the inode is in the cache, the inode is returned with an incremented
1211 * reference count.
1212 *
0b2d0724
CH
1213 * Note: I_NEW is not waited upon so you have to be very careful what you do
1214 * with the returned inode. You probably should be using ilookup5() instead.
1da177e4 1215 *
b6d0ad68 1216 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1da177e4 1217 */
0b2d0724
CH
1218struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1219 int (*test)(struct inode *, void *), void *data)
1da177e4 1220{
0b2d0724 1221 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1da177e4
LT
1222 struct inode *inode;
1223
67a23c49 1224 spin_lock(&inode_hash_lock);
1da177e4 1225 inode = find_inode(sb, head, test, data);
67a23c49 1226 spin_unlock(&inode_hash_lock);
88bd5121 1227
0b2d0724 1228 return inode;
88bd5121 1229}
88bd5121
AA
1230EXPORT_SYMBOL(ilookup5_nowait);
1231
1232/**
1233 * ilookup5 - search for an inode in the inode cache
1234 * @sb: super block of file system to search
1235 * @hashval: hash value (usually inode number) to search for
1236 * @test: callback used for comparisons between inodes
1237 * @data: opaque data pointer to pass to @test
1238 *
0b2d0724
CH
1239 * Search for the inode specified by @hashval and @data in the inode cache,
1240 * and if the inode is in the cache, return the inode with an incremented
1241 * reference count. Waits on I_NEW before returning the inode.
88bd5121 1242 * returned with an incremented reference count.
1da177e4 1243 *
0b2d0724
CH
1244 * This is a generalized version of ilookup() for file systems where the
1245 * inode number is not sufficient for unique identification of an inode.
1da177e4 1246 *
0b2d0724 1247 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1da177e4
LT
1248 */
1249struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1250 int (*test)(struct inode *, void *), void *data)
1251{
0b2d0724 1252 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1da177e4 1253
0b2d0724
CH
1254 if (inode)
1255 wait_on_inode(inode);
1256 return inode;
1da177e4 1257}
1da177e4
LT
1258EXPORT_SYMBOL(ilookup5);
1259
1260/**
1261 * ilookup - search for an inode in the inode cache
1262 * @sb: super block of file system to search
1263 * @ino: inode number to search for
1264 *
0b2d0724
CH
1265 * Search for the inode @ino in the inode cache, and if the inode is in the
1266 * cache, the inode is returned with an incremented reference count.
1da177e4
LT
1267 */
1268struct inode *ilookup(struct super_block *sb, unsigned long ino)
1269{
1270 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1da177e4
LT
1271 struct inode *inode;
1272
0b2d0724
CH
1273 spin_lock(&inode_hash_lock);
1274 inode = find_inode_fast(sb, head, ino);
1275 spin_unlock(&inode_hash_lock);
1da177e4 1276
1da177e4 1277 if (inode)
0b2d0724
CH
1278 wait_on_inode(inode);
1279 return inode;
1da177e4 1280}
0b2d0724 1281EXPORT_SYMBOL(ilookup);
1da177e4 1282
261bca86
AV
1283int insert_inode_locked(struct inode *inode)
1284{
1285 struct super_block *sb = inode->i_sb;
1286 ino_t ino = inode->i_ino;
1287 struct hlist_head *head = inode_hashtable + hash(sb, ino);
261bca86 1288
261bca86 1289 while (1) {
72a43d63 1290 struct inode *old = NULL;
67a23c49 1291 spin_lock(&inode_hash_lock);
b67bfe0d 1292 hlist_for_each_entry(old, head, i_hash) {
72a43d63
AV
1293 if (old->i_ino != ino)
1294 continue;
1295 if (old->i_sb != sb)
1296 continue;
250df6ed
DC
1297 spin_lock(&old->i_lock);
1298 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1299 spin_unlock(&old->i_lock);
72a43d63 1300 continue;
250df6ed 1301 }
72a43d63
AV
1302 break;
1303 }
b67bfe0d 1304 if (likely(!old)) {
250df6ed
DC
1305 spin_lock(&inode->i_lock);
1306 inode->i_state |= I_NEW;
261bca86 1307 hlist_add_head(&inode->i_hash, head);
250df6ed 1308 spin_unlock(&inode->i_lock);
67a23c49 1309 spin_unlock(&inode_hash_lock);
261bca86
AV
1310 return 0;
1311 }
1312 __iget(old);
250df6ed 1313 spin_unlock(&old->i_lock);
67a23c49 1314 spin_unlock(&inode_hash_lock);
261bca86 1315 wait_on_inode(old);
1d3382cb 1316 if (unlikely(!inode_unhashed(old))) {
261bca86
AV
1317 iput(old);
1318 return -EBUSY;
1319 }
1320 iput(old);
1321 }
1322}
261bca86
AV
1323EXPORT_SYMBOL(insert_inode_locked);
1324
1325int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1326 int (*test)(struct inode *, void *), void *data)
1327{
1328 struct super_block *sb = inode->i_sb;
1329 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
261bca86 1330
261bca86 1331 while (1) {
72a43d63
AV
1332 struct inode *old = NULL;
1333
67a23c49 1334 spin_lock(&inode_hash_lock);
b67bfe0d 1335 hlist_for_each_entry(old, head, i_hash) {
72a43d63
AV
1336 if (old->i_sb != sb)
1337 continue;
1338 if (!test(old, data))
1339 continue;
250df6ed
DC
1340 spin_lock(&old->i_lock);
1341 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1342 spin_unlock(&old->i_lock);
72a43d63 1343 continue;
250df6ed 1344 }
72a43d63
AV
1345 break;
1346 }
b67bfe0d 1347 if (likely(!old)) {
250df6ed
DC
1348 spin_lock(&inode->i_lock);
1349 inode->i_state |= I_NEW;
261bca86 1350 hlist_add_head(&inode->i_hash, head);
250df6ed 1351 spin_unlock(&inode->i_lock);
67a23c49 1352 spin_unlock(&inode_hash_lock);
261bca86
AV
1353 return 0;
1354 }
1355 __iget(old);
250df6ed 1356 spin_unlock(&old->i_lock);
67a23c49 1357 spin_unlock(&inode_hash_lock);
261bca86 1358 wait_on_inode(old);
1d3382cb 1359 if (unlikely(!inode_unhashed(old))) {
261bca86
AV
1360 iput(old);
1361 return -EBUSY;
1362 }
1363 iput(old);
1364 }
1365}
261bca86
AV
1366EXPORT_SYMBOL(insert_inode_locked4);
1367
1da177e4 1368
45321ac5
AV
1369int generic_delete_inode(struct inode *inode)
1370{
1371 return 1;
1372}
1373EXPORT_SYMBOL(generic_delete_inode);
1374
45321ac5
AV
1375/*
1376 * Called when we're dropping the last reference
1377 * to an inode.
22fe4042 1378 *
45321ac5
AV
1379 * Call the FS "drop_inode()" function, defaulting to
1380 * the legacy UNIX filesystem behaviour. If it tells
1381 * us to evict inode, do so. Otherwise, retain inode
1382 * in cache if fs is alive, sync and evict if fs is
1383 * shutting down.
22fe4042 1384 */
45321ac5 1385static void iput_final(struct inode *inode)
1da177e4
LT
1386{
1387 struct super_block *sb = inode->i_sb;
45321ac5
AV
1388 const struct super_operations *op = inode->i_sb->s_op;
1389 int drop;
1390
250df6ed
DC
1391 WARN_ON(inode->i_state & I_NEW);
1392
e7f59097 1393 if (op->drop_inode)
45321ac5
AV
1394 drop = op->drop_inode(inode);
1395 else
1396 drop = generic_drop_inode(inode);
1da177e4 1397
b2b2af8e
DC
1398 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1399 inode->i_state |= I_REFERENCED;
4eff96dd 1400 inode_add_lru(inode);
b2b2af8e 1401 spin_unlock(&inode->i_lock);
b2b2af8e
DC
1402 return;
1403 }
1404
45321ac5 1405 if (!drop) {
991114c6 1406 inode->i_state |= I_WILL_FREE;
250df6ed 1407 spin_unlock(&inode->i_lock);
1da177e4 1408 write_inode_now(inode, 1);
250df6ed 1409 spin_lock(&inode->i_lock);
7ef0d737 1410 WARN_ON(inode->i_state & I_NEW);
991114c6 1411 inode->i_state &= ~I_WILL_FREE;
1da177e4 1412 }
7ccf19a8 1413
991114c6 1414 inode->i_state |= I_FREEING;
c4ae0c65
ED
1415 if (!list_empty(&inode->i_lru))
1416 inode_lru_list_del(inode);
b2b2af8e 1417 spin_unlock(&inode->i_lock);
b2b2af8e 1418
644da596 1419 evict(inode);
1da177e4
LT
1420}
1421
1da177e4 1422/**
6b3304b5 1423 * iput - put an inode
1da177e4
LT
1424 * @inode: inode to put
1425 *
1426 * Puts an inode, dropping its usage count. If the inode use count hits
1427 * zero, the inode is then freed and may also be destroyed.
1428 *
1429 * Consequently, iput() can sleep.
1430 */
1431void iput(struct inode *inode)
1432{
1433 if (inode) {
a4ffdde6 1434 BUG_ON(inode->i_state & I_CLEAR);
1da177e4 1435
f283c86a 1436 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1da177e4
LT
1437 iput_final(inode);
1438 }
1439}
1da177e4
LT
1440EXPORT_SYMBOL(iput);
1441
1442/**
1443 * bmap - find a block number in a file
1444 * @inode: inode of file
1445 * @block: block to find
1446 *
1447 * Returns the block number on the device holding the inode that
1448 * is the disk block number for the block of the file requested.
1449 * That is, asked for block 4 of inode 1 the function will return the
6b3304b5 1450 * disk block relative to the disk start that holds that block of the
1da177e4
LT
1451 * file.
1452 */
6b3304b5 1453sector_t bmap(struct inode *inode, sector_t block)
1da177e4
LT
1454{
1455 sector_t res = 0;
1456 if (inode->i_mapping->a_ops->bmap)
1457 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1458 return res;
1459}
1da177e4
LT
1460EXPORT_SYMBOL(bmap);
1461
11ff6f05
MG
1462/*
1463 * With relative atime, only update atime if the previous atime is
1464 * earlier than either the ctime or mtime or if at least a day has
1465 * passed since the last atime update.
1466 */
1467static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1468 struct timespec now)
1469{
1470
1471 if (!(mnt->mnt_flags & MNT_RELATIME))
1472 return 1;
1473 /*
1474 * Is mtime younger than atime? If yes, update atime:
1475 */
1476 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1477 return 1;
1478 /*
1479 * Is ctime younger than atime? If yes, update atime:
1480 */
1481 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1482 return 1;
1483
1484 /*
1485 * Is the previous atime value older than a day? If yes,
1486 * update atime:
1487 */
1488 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1489 return 1;
1490 /*
1491 * Good, we can skip the atime update:
1492 */
1493 return 0;
1494}
1495
c3b2da31
JB
1496/*
1497 * This does the actual work of updating an inodes time or version. Must have
1498 * had called mnt_want_write() before calling this.
1499 */
1500static int update_time(struct inode *inode, struct timespec *time, int flags)
1501{
1502 if (inode->i_op->update_time)
1503 return inode->i_op->update_time(inode, time, flags);
1504
1505 if (flags & S_ATIME)
1506 inode->i_atime = *time;
1507 if (flags & S_VERSION)
1508 inode_inc_iversion(inode);
1509 if (flags & S_CTIME)
1510 inode->i_ctime = *time;
1511 if (flags & S_MTIME)
1512 inode->i_mtime = *time;
1513 mark_inode_dirty_sync(inode);
1514 return 0;
1515}
1516
1da177e4 1517/**
869243a0 1518 * touch_atime - update the access time
185553b2 1519 * @path: the &struct path to update
1da177e4
LT
1520 *
1521 * Update the accessed time on an inode and mark it for writeback.
1522 * This function automatically handles read only file systems and media,
1523 * as well as the "noatime" flag and inode specific "noatime" markers.
1524 */
badcf2b7 1525void touch_atime(const struct path *path)
1da177e4 1526{
68ac1234
AV
1527 struct vfsmount *mnt = path->mnt;
1528 struct inode *inode = path->dentry->d_inode;
1da177e4
LT
1529 struct timespec now;
1530
cdb70f3f 1531 if (inode->i_flags & S_NOATIME)
b12536c2 1532 return;
37756ced 1533 if (IS_NOATIME(inode))
b12536c2 1534 return;
b2276138 1535 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
b12536c2 1536 return;
47ae32d6 1537
cdb70f3f 1538 if (mnt->mnt_flags & MNT_NOATIME)
b12536c2 1539 return;
cdb70f3f 1540 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
b12536c2 1541 return;
1da177e4
LT
1542
1543 now = current_fs_time(inode->i_sb);
11ff6f05
MG
1544
1545 if (!relatime_need_update(mnt, inode, now))
b12536c2 1546 return;
11ff6f05 1547
47ae32d6 1548 if (timespec_equal(&inode->i_atime, &now))
b12536c2
AK
1549 return;
1550
5d37e9e6 1551 if (!sb_start_write_trylock(inode->i_sb))
b12536c2 1552 return;
47ae32d6 1553
5d37e9e6
JK
1554 if (__mnt_want_write(mnt))
1555 goto skip_update;
c3b2da31
JB
1556 /*
1557 * File systems can error out when updating inodes if they need to
1558 * allocate new space to modify an inode (such is the case for
1559 * Btrfs), but since we touch atime while walking down the path we
1560 * really don't care if we failed to update the atime of the file,
1561 * so just ignore the return value.
2bc55652
AB
1562 * We may also fail on filesystems that have the ability to make parts
1563 * of the fs read only, e.g. subvolumes in Btrfs.
c3b2da31
JB
1564 */
1565 update_time(inode, &now, S_ATIME);
5d37e9e6
JK
1566 __mnt_drop_write(mnt);
1567skip_update:
1568 sb_end_write(inode->i_sb);
1da177e4 1569}
869243a0 1570EXPORT_SYMBOL(touch_atime);
1da177e4 1571
3ed37648
CW
1572/*
1573 * The logic we want is
1574 *
1575 * if suid or (sgid and xgrp)
1576 * remove privs
1577 */
1578int should_remove_suid(struct dentry *dentry)
1579{
1580 umode_t mode = dentry->d_inode->i_mode;
1581 int kill = 0;
1582
1583 /* suid always must be killed */
1584 if (unlikely(mode & S_ISUID))
1585 kill = ATTR_KILL_SUID;
1586
1587 /*
1588 * sgid without any exec bits is just a mandatory locking mark; leave
1589 * it alone. If some exec bits are set, it's a real sgid; kill it.
1590 */
1591 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1592 kill |= ATTR_KILL_SGID;
1593
1594 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1595 return kill;
1596
1597 return 0;
1598}
1599EXPORT_SYMBOL(should_remove_suid);
1600
1601static int __remove_suid(struct dentry *dentry, int kill)
1602{
1603 struct iattr newattrs;
1604
1605 newattrs.ia_valid = ATTR_FORCE | kill;
27ac0ffe
BF
1606 /*
1607 * Note we call this on write, so notify_change will not
1608 * encounter any conflicting delegations:
1609 */
1610 return notify_change(dentry, &newattrs, NULL);
3ed37648
CW
1611}
1612
1613int file_remove_suid(struct file *file)
1614{
1615 struct dentry *dentry = file->f_path.dentry;
1616 struct inode *inode = dentry->d_inode;
1617 int killsuid;
1618 int killpriv;
1619 int error = 0;
1620
1621 /* Fast path for nothing security related */
1622 if (IS_NOSEC(inode))
1623 return 0;
1624
1625 killsuid = should_remove_suid(dentry);
1626 killpriv = security_inode_need_killpriv(dentry);
1627
1628 if (killpriv < 0)
1629 return killpriv;
1630 if (killpriv)
1631 error = security_inode_killpriv(dentry);
1632 if (!error && killsuid)
1633 error = __remove_suid(dentry, killsuid);
1634 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1635 inode->i_flags |= S_NOSEC;
1636
1637 return error;
1638}
1639EXPORT_SYMBOL(file_remove_suid);
1640
1da177e4 1641/**
870f4817
CH
1642 * file_update_time - update mtime and ctime time
1643 * @file: file accessed
1da177e4 1644 *
870f4817
CH
1645 * Update the mtime and ctime members of an inode and mark the inode
1646 * for writeback. Note that this function is meant exclusively for
1647 * usage in the file write path of filesystems, and filesystems may
1648 * choose to explicitly ignore update via this function with the
2eadfc0e 1649 * S_NOCMTIME inode flag, e.g. for network filesystem where these
c3b2da31
JB
1650 * timestamps are handled by the server. This can return an error for
1651 * file systems who need to allocate space in order to update an inode.
1da177e4
LT
1652 */
1653
c3b2da31 1654int file_update_time(struct file *file)
1da177e4 1655{
496ad9aa 1656 struct inode *inode = file_inode(file);
1da177e4 1657 struct timespec now;
c3b2da31
JB
1658 int sync_it = 0;
1659 int ret;
1da177e4 1660
ce06e0b2 1661 /* First try to exhaust all avenues to not sync */
1da177e4 1662 if (IS_NOCMTIME(inode))
c3b2da31 1663 return 0;
20ddee2c 1664
1da177e4 1665 now = current_fs_time(inode->i_sb);
ce06e0b2
AK
1666 if (!timespec_equal(&inode->i_mtime, &now))
1667 sync_it = S_MTIME;
1da177e4 1668
ce06e0b2
AK
1669 if (!timespec_equal(&inode->i_ctime, &now))
1670 sync_it |= S_CTIME;
870f4817 1671
ce06e0b2
AK
1672 if (IS_I_VERSION(inode))
1673 sync_it |= S_VERSION;
7a224228 1674
ce06e0b2 1675 if (!sync_it)
c3b2da31 1676 return 0;
ce06e0b2
AK
1677
1678 /* Finally allowed to write? Takes lock. */
eb04c282 1679 if (__mnt_want_write_file(file))
c3b2da31 1680 return 0;
ce06e0b2 1681
c3b2da31 1682 ret = update_time(inode, &now, sync_it);
eb04c282 1683 __mnt_drop_write_file(file);
c3b2da31
JB
1684
1685 return ret;
1da177e4 1686}
870f4817 1687EXPORT_SYMBOL(file_update_time);
1da177e4
LT
1688
1689int inode_needs_sync(struct inode *inode)
1690{
1691 if (IS_SYNC(inode))
1692 return 1;
1693 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1694 return 1;
1695 return 0;
1696}
1da177e4
LT
1697EXPORT_SYMBOL(inode_needs_sync);
1698
1da177e4
LT
1699int inode_wait(void *word)
1700{
1701 schedule();
1702 return 0;
1703}
d44dab8d 1704EXPORT_SYMBOL(inode_wait);
1da177e4
LT
1705
1706/*
168a9fd6
MS
1707 * If we try to find an inode in the inode hash while it is being
1708 * deleted, we have to wait until the filesystem completes its
1709 * deletion before reporting that it isn't found. This function waits
1710 * until the deletion _might_ have completed. Callers are responsible
1711 * to recheck inode state.
1712 *
eaff8079 1713 * It doesn't matter if I_NEW is not set initially, a call to
250df6ed
DC
1714 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1715 * will DTRT.
1da177e4
LT
1716 */
1717static void __wait_on_freeing_inode(struct inode *inode)
1718{
1719 wait_queue_head_t *wq;
eaff8079
CH
1720 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1721 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1da177e4 1722 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
250df6ed 1723 spin_unlock(&inode->i_lock);
67a23c49 1724 spin_unlock(&inode_hash_lock);
1da177e4
LT
1725 schedule();
1726 finish_wait(wq, &wait.wait);
67a23c49 1727 spin_lock(&inode_hash_lock);
1da177e4
LT
1728}
1729
1da177e4
LT
1730static __initdata unsigned long ihash_entries;
1731static int __init set_ihash_entries(char *str)
1732{
1733 if (!str)
1734 return 0;
1735 ihash_entries = simple_strtoul(str, &str, 0);
1736 return 1;
1737}
1738__setup("ihash_entries=", set_ihash_entries);
1739
1740/*
1741 * Initialize the waitqueues and inode hash table.
1742 */
1743void __init inode_init_early(void)
1744{
074b8517 1745 unsigned int loop;
1da177e4
LT
1746
1747 /* If hashes are distributed across NUMA nodes, defer
1748 * hash allocation until vmalloc space is available.
1749 */
1750 if (hashdist)
1751 return;
1752
1753 inode_hashtable =
1754 alloc_large_system_hash("Inode-cache",
1755 sizeof(struct hlist_head),
1756 ihash_entries,
1757 14,
1758 HASH_EARLY,
1759 &i_hash_shift,
1760 &i_hash_mask,
31fe62b9 1761 0,
1da177e4
LT
1762 0);
1763
074b8517 1764 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1da177e4
LT
1765 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1766}
1767
74bf17cf 1768void __init inode_init(void)
1da177e4 1769{
074b8517 1770 unsigned int loop;
1da177e4
LT
1771
1772 /* inode slab cache */
b0196009
PJ
1773 inode_cachep = kmem_cache_create("inode_cache",
1774 sizeof(struct inode),
1775 0,
1776 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1777 SLAB_MEM_SPREAD),
20c2df83 1778 init_once);
1da177e4
LT
1779
1780 /* Hash may have been set up in inode_init_early */
1781 if (!hashdist)
1782 return;
1783
1784 inode_hashtable =
1785 alloc_large_system_hash("Inode-cache",
1786 sizeof(struct hlist_head),
1787 ihash_entries,
1788 14,
1789 0,
1790 &i_hash_shift,
1791 &i_hash_mask,
31fe62b9 1792 0,
1da177e4
LT
1793 0);
1794
074b8517 1795 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1da177e4
LT
1796 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1797}
1798
1799void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1800{
1801 inode->i_mode = mode;
1802 if (S_ISCHR(mode)) {
1803 inode->i_fop = &def_chr_fops;
1804 inode->i_rdev = rdev;
1805 } else if (S_ISBLK(mode)) {
1806 inode->i_fop = &def_blk_fops;
1807 inode->i_rdev = rdev;
1808 } else if (S_ISFIFO(mode))
599a0ac1 1809 inode->i_fop = &pipefifo_fops;
1da177e4
LT
1810 else if (S_ISSOCK(mode))
1811 inode->i_fop = &bad_sock_fops;
1812 else
af0d9ae8
MK
1813 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1814 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1815 inode->i_ino);
1da177e4
LT
1816}
1817EXPORT_SYMBOL(init_special_inode);
a1bd120d
DM
1818
1819/**
eaae668d 1820 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
a1bd120d
DM
1821 * @inode: New inode
1822 * @dir: Directory inode
1823 * @mode: mode of the new inode
1824 */
1825void inode_init_owner(struct inode *inode, const struct inode *dir,
62bb1091 1826 umode_t mode)
a1bd120d
DM
1827{
1828 inode->i_uid = current_fsuid();
1829 if (dir && dir->i_mode & S_ISGID) {
1830 inode->i_gid = dir->i_gid;
1831 if (S_ISDIR(mode))
1832 mode |= S_ISGID;
1833 } else
1834 inode->i_gid = current_fsgid();
1835 inode->i_mode = mode;
1836}
1837EXPORT_SYMBOL(inode_init_owner);
e795b717 1838
2e149670
SH
1839/**
1840 * inode_owner_or_capable - check current task permissions to inode
1841 * @inode: inode being checked
1842 *
1843 * Return true if current either has CAP_FOWNER to the inode, or
1844 * owns the file.
e795b717 1845 */
2e149670 1846bool inode_owner_or_capable(const struct inode *inode)
e795b717 1847{
92361636 1848 if (uid_eq(current_fsuid(), inode->i_uid))
e795b717 1849 return true;
1a48e2ac 1850 if (inode_capable(inode, CAP_FOWNER))
e795b717
SH
1851 return true;
1852 return false;
1853}
2e149670 1854EXPORT_SYMBOL(inode_owner_or_capable);
1d59d61f
TM
1855
1856/*
1857 * Direct i/o helper functions
1858 */
1859static void __inode_dio_wait(struct inode *inode)
1860{
1861 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1862 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1863
1864 do {
1865 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1866 if (atomic_read(&inode->i_dio_count))
1867 schedule();
1868 } while (atomic_read(&inode->i_dio_count));
1869 finish_wait(wq, &q.wait);
1870}
1871
1872/**
1873 * inode_dio_wait - wait for outstanding DIO requests to finish
1874 * @inode: inode to wait for
1875 *
1876 * Waits for all pending direct I/O requests to finish so that we can
1877 * proceed with a truncate or equivalent operation.
1878 *
1879 * Must be called under a lock that serializes taking new references
1880 * to i_dio_count, usually by inode->i_mutex.
1881 */
1882void inode_dio_wait(struct inode *inode)
1883{
1884 if (atomic_read(&inode->i_dio_count))
1885 __inode_dio_wait(inode);
1886}
1887EXPORT_SYMBOL(inode_dio_wait);
1888
1889/*
1890 * inode_dio_done - signal finish of a direct I/O requests
1891 * @inode: inode the direct I/O happens on
1892 *
1893 * This is called once we've finished processing a direct I/O request,
1894 * and is used to wake up callers waiting for direct I/O to be quiesced.
1895 */
1896void inode_dio_done(struct inode *inode)
1897{
1898 if (atomic_dec_and_test(&inode->i_dio_count))
1899 wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1900}
1901EXPORT_SYMBOL(inode_dio_done);