]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/inode.c
mm/memory-failure.c: fix spinlock vs mutex order
[mirror_ubuntu-artful-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 */
1da177e4
LT
5#include <linux/fs.h>
6#include <linux/mm.h>
7#include <linux/dcache.h>
8#include <linux/init.h>
1da177e4
LT
9#include <linux/slab.h>
10#include <linux/writeback.h>
11#include <linux/module.h>
12#include <linux/backing-dev.h>
13#include <linux/wait.h>
88e0fbc4 14#include <linux/rwsem.h>
1da177e4
LT
15#include <linux/hash.h>
16#include <linux/swap.h>
17#include <linux/security.h>
18#include <linux/pagemap.h>
19#include <linux/cdev.h>
20#include <linux/bootmem.h>
3be25f49 21#include <linux/fsnotify.h>
fc33a7bb 22#include <linux/mount.h>
efaee192 23#include <linux/async.h>
f19d4a8f 24#include <linux/posix_acl.h>
9ce6e0be 25#include <linux/prefetch.h>
a178d202 26#include <linux/ima.h>
e795b717 27#include <linux/cred.h>
4b4563dc 28#include <linux/buffer_head.h> /* for inode_has_buffers */
a66979ab 29#include "internal.h"
1da177e4 30
250df6ed 31/*
4b4563dc 32 * Inode locking rules:
250df6ed
DC
33 *
34 * inode->i_lock protects:
35 * inode->i_state, inode->i_hash, __iget()
02afc410
DC
36 * inode_lru_lock protects:
37 * inode_lru, inode->i_lru
55fa6091
DC
38 * inode_sb_list_lock protects:
39 * sb->s_inodes, inode->i_sb_list
a66979ab
DC
40 * inode_wb_list_lock protects:
41 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
67a23c49
DC
42 * inode_hash_lock protects:
43 * inode_hashtable, inode->i_hash
250df6ed
DC
44 *
45 * Lock ordering:
55fa6091
DC
46 *
47 * inode_sb_list_lock
48 * inode->i_lock
02afc410 49 * inode_lru_lock
a66979ab
DC
50 *
51 * inode_wb_list_lock
52 * inode->i_lock
67a23c49
DC
53 *
54 * inode_hash_lock
55 * inode_sb_list_lock
56 * inode->i_lock
57 *
58 * iunique_lock
59 * inode_hash_lock
250df6ed
DC
60 */
61
fa3536cc
ED
62static unsigned int i_hash_mask __read_mostly;
63static unsigned int i_hash_shift __read_mostly;
67a23c49
DC
64static struct hlist_head *inode_hashtable __read_mostly;
65static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
1da177e4 66
7ccf19a8 67static LIST_HEAD(inode_lru);
02afc410 68static DEFINE_SPINLOCK(inode_lru_lock);
1da177e4 69
55fa6091 70__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
a66979ab 71__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_wb_list_lock);
55fa6091 72
1da177e4 73/*
bab1d944
CH
74 * iprune_sem provides exclusion between the icache shrinking and the
75 * umount path.
88e0fbc4 76 *
bab1d944
CH
77 * We don't actually need it to protect anything in the umount path,
78 * but only need to cycle through it to make sure any inode that
79 * prune_icache took off the LRU list has been fully torn down by the
80 * time we are past evict_inodes.
1da177e4 81 */
88e0fbc4 82static DECLARE_RWSEM(iprune_sem);
1da177e4 83
7dcda1c9
JA
84/*
85 * Empty aops. Can be used for the cases where the user does not
86 * define any of the address_space operations.
87 */
88const struct address_space_operations empty_aops = {
89};
90EXPORT_SYMBOL(empty_aops);
91
1da177e4
LT
92/*
93 * Statistics gathering..
94 */
95struct inodes_stat_t inodes_stat;
96
3e880fb5 97static DEFINE_PER_CPU(unsigned int, nr_inodes);
cffbc8aa 98
6b3304b5 99static struct kmem_cache *inode_cachep __read_mostly;
1da177e4 100
3e880fb5 101static int get_nr_inodes(void)
cffbc8aa 102{
3e880fb5
NP
103 int i;
104 int sum = 0;
105 for_each_possible_cpu(i)
106 sum += per_cpu(nr_inodes, i);
107 return sum < 0 ? 0 : sum;
cffbc8aa
DC
108}
109
110static inline int get_nr_inodes_unused(void)
111{
86c8749e 112 return inodes_stat.nr_unused;
cffbc8aa
DC
113}
114
115int get_nr_dirty_inodes(void)
116{
3e880fb5 117 /* not actually dirty inodes, but a wild approximation */
cffbc8aa
DC
118 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
119 return nr_dirty > 0 ? nr_dirty : 0;
cffbc8aa
DC
120}
121
122/*
123 * Handle nr_inode sysctl
124 */
125#ifdef CONFIG_SYSCTL
126int proc_nr_inodes(ctl_table *table, int write,
127 void __user *buffer, size_t *lenp, loff_t *ppos)
128{
129 inodes_stat.nr_inodes = get_nr_inodes();
cffbc8aa
DC
130 return proc_dointvec(table, write, buffer, lenp, ppos);
131}
132#endif
133
2cb1599f
DC
134/**
135 * inode_init_always - perform inode structure intialisation
0bc02f3f
RD
136 * @sb: superblock inode belongs to
137 * @inode: inode to initialise
2cb1599f
DC
138 *
139 * These are initializations that need to be done on every inode
140 * allocation as the fields are not initialised by slab allocation.
141 */
54e34621 142int inode_init_always(struct super_block *sb, struct inode *inode)
1da177e4 143{
6e1d5dcc 144 static const struct inode_operations empty_iops;
99ac48f5 145 static const struct file_operations empty_fops;
6b3304b5 146 struct address_space *const mapping = &inode->i_data;
2cb1599f
DC
147
148 inode->i_sb = sb;
149 inode->i_blkbits = sb->s_blocksize_bits;
150 inode->i_flags = 0;
151 atomic_set(&inode->i_count, 1);
152 inode->i_op = &empty_iops;
153 inode->i_fop = &empty_fops;
154 inode->i_nlink = 1;
56ff5efa
AV
155 inode->i_uid = 0;
156 inode->i_gid = 0;
2cb1599f
DC
157 atomic_set(&inode->i_writecount, 0);
158 inode->i_size = 0;
159 inode->i_blocks = 0;
160 inode->i_bytes = 0;
161 inode->i_generation = 0;
1da177e4 162#ifdef CONFIG_QUOTA
2cb1599f 163 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
1da177e4 164#endif
2cb1599f
DC
165 inode->i_pipe = NULL;
166 inode->i_bdev = NULL;
167 inode->i_cdev = NULL;
168 inode->i_rdev = 0;
169 inode->dirtied_when = 0;
6146f0d5
MZ
170
171 if (security_inode_alloc(inode))
54e34621 172 goto out;
2cb1599f
DC
173 spin_lock_init(&inode->i_lock);
174 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
175
176 mutex_init(&inode->i_mutex);
177 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
178
179 init_rwsem(&inode->i_alloc_sem);
180 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
181
182 mapping->a_ops = &empty_aops;
183 mapping->host = inode;
184 mapping->flags = 0;
3c1d4378 185 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
2cb1599f
DC
186 mapping->assoc_mapping = NULL;
187 mapping->backing_dev_info = &default_backing_dev_info;
188 mapping->writeback_index = 0;
189
190 /*
191 * If the block_device provides a backing_dev_info for client
192 * inodes then use that. Otherwise the inode share the bdev's
193 * backing_dev_info.
194 */
195 if (sb->s_bdev) {
196 struct backing_dev_info *bdi;
197
2c96ce9f 198 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
2cb1599f
DC
199 mapping->backing_dev_info = bdi;
200 }
201 inode->i_private = NULL;
202 inode->i_mapping = mapping;
f19d4a8f
AV
203#ifdef CONFIG_FS_POSIX_ACL
204 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
205#endif
2cb1599f 206
3be25f49
EP
207#ifdef CONFIG_FSNOTIFY
208 inode->i_fsnotify_mask = 0;
209#endif
210
3e880fb5 211 this_cpu_inc(nr_inodes);
cffbc8aa 212
54e34621 213 return 0;
54e34621
CH
214out:
215 return -ENOMEM;
1da177e4 216}
2cb1599f
DC
217EXPORT_SYMBOL(inode_init_always);
218
219static struct inode *alloc_inode(struct super_block *sb)
220{
221 struct inode *inode;
222
223 if (sb->s_op->alloc_inode)
224 inode = sb->s_op->alloc_inode(sb);
225 else
226 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
227
54e34621
CH
228 if (!inode)
229 return NULL;
230
231 if (unlikely(inode_init_always(sb, inode))) {
232 if (inode->i_sb->s_op->destroy_inode)
233 inode->i_sb->s_op->destroy_inode(inode);
234 else
235 kmem_cache_free(inode_cachep, inode);
236 return NULL;
237 }
238
239 return inode;
2cb1599f 240}
1da177e4 241
ff0c7d15
NP
242void free_inode_nonrcu(struct inode *inode)
243{
244 kmem_cache_free(inode_cachep, inode);
245}
246EXPORT_SYMBOL(free_inode_nonrcu);
247
2e00c97e 248void __destroy_inode(struct inode *inode)
1da177e4 249{
b7542f8c 250 BUG_ON(inode_has_buffers(inode));
1da177e4 251 security_inode_free(inode);
3be25f49 252 fsnotify_inode_delete(inode);
f19d4a8f
AV
253#ifdef CONFIG_FS_POSIX_ACL
254 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
255 posix_acl_release(inode->i_acl);
256 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
257 posix_acl_release(inode->i_default_acl);
258#endif
3e880fb5 259 this_cpu_dec(nr_inodes);
2e00c97e
CH
260}
261EXPORT_SYMBOL(__destroy_inode);
262
fa0d7e3d
NP
263static void i_callback(struct rcu_head *head)
264{
265 struct inode *inode = container_of(head, struct inode, i_rcu);
266 INIT_LIST_HEAD(&inode->i_dentry);
267 kmem_cache_free(inode_cachep, inode);
268}
269
56b0dacf 270static void destroy_inode(struct inode *inode)
2e00c97e 271{
7ccf19a8 272 BUG_ON(!list_empty(&inode->i_lru));
2e00c97e 273 __destroy_inode(inode);
1da177e4
LT
274 if (inode->i_sb->s_op->destroy_inode)
275 inode->i_sb->s_op->destroy_inode(inode);
276 else
fa0d7e3d 277 call_rcu(&inode->i_rcu, i_callback);
1da177e4 278}
1da177e4 279
2aa15890
MS
280void address_space_init_once(struct address_space *mapping)
281{
282 memset(mapping, 0, sizeof(*mapping));
283 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
284 spin_lock_init(&mapping->tree_lock);
3d48ae45 285 mutex_init(&mapping->i_mmap_mutex);
2aa15890
MS
286 INIT_LIST_HEAD(&mapping->private_list);
287 spin_lock_init(&mapping->private_lock);
288 INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
289 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
2aa15890
MS
290}
291EXPORT_SYMBOL(address_space_init_once);
292
1da177e4
LT
293/*
294 * These are initializations that only need to be done
295 * once, because the fields are idempotent across use
296 * of the inode, so let the slab aware of that.
297 */
298void inode_init_once(struct inode *inode)
299{
300 memset(inode, 0, sizeof(*inode));
301 INIT_HLIST_NODE(&inode->i_hash);
302 INIT_LIST_HEAD(&inode->i_dentry);
303 INIT_LIST_HEAD(&inode->i_devices);
7ccf19a8
NP
304 INIT_LIST_HEAD(&inode->i_wb_list);
305 INIT_LIST_HEAD(&inode->i_lru);
2aa15890 306 address_space_init_once(&inode->i_data);
1da177e4 307 i_size_ordered_init(inode);
3be25f49 308#ifdef CONFIG_FSNOTIFY
e61ce867 309 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
3be25f49 310#endif
1da177e4 311}
1da177e4
LT
312EXPORT_SYMBOL(inode_init_once);
313
51cc5068 314static void init_once(void *foo)
1da177e4 315{
6b3304b5 316 struct inode *inode = (struct inode *) foo;
1da177e4 317
a35afb83 318 inode_init_once(inode);
1da177e4
LT
319}
320
321/*
250df6ed 322 * inode->i_lock must be held
1da177e4 323 */
6b3304b5 324void __iget(struct inode *inode)
1da177e4 325{
9e38d86f
NP
326 atomic_inc(&inode->i_count);
327}
2e147f1e 328
7de9c6ee
AV
329/*
330 * get additional reference to inode; caller must already hold one.
331 */
332void ihold(struct inode *inode)
333{
334 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
335}
336EXPORT_SYMBOL(ihold);
337
9e38d86f
NP
338static void inode_lru_list_add(struct inode *inode)
339{
02afc410 340 spin_lock(&inode_lru_lock);
7ccf19a8
NP
341 if (list_empty(&inode->i_lru)) {
342 list_add(&inode->i_lru, &inode_lru);
86c8749e 343 inodes_stat.nr_unused++;
9e38d86f 344 }
02afc410 345 spin_unlock(&inode_lru_lock);
9e38d86f 346}
2e147f1e 347
9e38d86f
NP
348static void inode_lru_list_del(struct inode *inode)
349{
02afc410 350 spin_lock(&inode_lru_lock);
7ccf19a8
NP
351 if (!list_empty(&inode->i_lru)) {
352 list_del_init(&inode->i_lru);
86c8749e 353 inodes_stat.nr_unused--;
9e38d86f 354 }
02afc410 355 spin_unlock(&inode_lru_lock);
1da177e4
LT
356}
357
646ec461
CH
358/**
359 * inode_sb_list_add - add inode to the superblock list of inodes
360 * @inode: inode to add
361 */
362void inode_sb_list_add(struct inode *inode)
363{
55fa6091
DC
364 spin_lock(&inode_sb_list_lock);
365 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
366 spin_unlock(&inode_sb_list_lock);
646ec461
CH
367}
368EXPORT_SYMBOL_GPL(inode_sb_list_add);
369
55fa6091 370static inline void inode_sb_list_del(struct inode *inode)
646ec461 371{
55fa6091 372 spin_lock(&inode_sb_list_lock);
646ec461 373 list_del_init(&inode->i_sb_list);
55fa6091 374 spin_unlock(&inode_sb_list_lock);
646ec461
CH
375}
376
4c51acbc
DC
377static unsigned long hash(struct super_block *sb, unsigned long hashval)
378{
379 unsigned long tmp;
380
381 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
382 L1_CACHE_BYTES;
4b4563dc
CH
383 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
384 return tmp & i_hash_mask;
4c51acbc
DC
385}
386
387/**
388 * __insert_inode_hash - hash an inode
389 * @inode: unhashed inode
390 * @hashval: unsigned long value used to locate this object in the
391 * inode_hashtable.
392 *
393 * Add an inode to the inode hash for this superblock.
394 */
395void __insert_inode_hash(struct inode *inode, unsigned long hashval)
396{
646ec461
CH
397 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
398
67a23c49 399 spin_lock(&inode_hash_lock);
250df6ed 400 spin_lock(&inode->i_lock);
646ec461 401 hlist_add_head(&inode->i_hash, b);
250df6ed 402 spin_unlock(&inode->i_lock);
67a23c49 403 spin_unlock(&inode_hash_lock);
4c51acbc
DC
404}
405EXPORT_SYMBOL(__insert_inode_hash);
406
4c51acbc
DC
407/**
408 * remove_inode_hash - remove an inode from the hash
409 * @inode: inode to unhash
410 *
411 * Remove an inode from the superblock.
412 */
413void remove_inode_hash(struct inode *inode)
414{
67a23c49 415 spin_lock(&inode_hash_lock);
250df6ed 416 spin_lock(&inode->i_lock);
4c51acbc 417 hlist_del_init(&inode->i_hash);
250df6ed 418 spin_unlock(&inode->i_lock);
67a23c49 419 spin_unlock(&inode_hash_lock);
4c51acbc
DC
420}
421EXPORT_SYMBOL(remove_inode_hash);
422
b0683aa6
AV
423void end_writeback(struct inode *inode)
424{
425 might_sleep();
426 BUG_ON(inode->i_data.nrpages);
427 BUG_ON(!list_empty(&inode->i_data.private_list));
428 BUG_ON(!(inode->i_state & I_FREEING));
429 BUG_ON(inode->i_state & I_CLEAR);
430 inode_sync_wait(inode);
fa0d7e3d 431 /* don't need i_lock here, no concurrent mods to i_state */
b0683aa6
AV
432 inode->i_state = I_FREEING | I_CLEAR;
433}
434EXPORT_SYMBOL(end_writeback);
435
b2b2af8e
DC
436/*
437 * Free the inode passed in, removing it from the lists it is still connected
438 * to. We remove any pages still attached to the inode and wait for any IO that
439 * is still in progress before finally destroying the inode.
440 *
441 * An inode must already be marked I_FREEING so that we avoid the inode being
442 * moved back onto lists if we race with other code that manipulates the lists
443 * (e.g. writeback_single_inode). The caller is responsible for setting this.
444 *
445 * An inode must already be removed from the LRU list before being evicted from
446 * the cache. This should occur atomically with setting the I_FREEING state
447 * flag, so no inodes here should ever be on the LRU when being evicted.
448 */
644da596 449static void evict(struct inode *inode)
b4272d4c
AV
450{
451 const struct super_operations *op = inode->i_sb->s_op;
452
b2b2af8e
DC
453 BUG_ON(!(inode->i_state & I_FREEING));
454 BUG_ON(!list_empty(&inode->i_lru));
455
a66979ab 456 inode_wb_list_del(inode);
55fa6091
DC
457 inode_sb_list_del(inode);
458
be7ce416
AV
459 if (op->evict_inode) {
460 op->evict_inode(inode);
b4272d4c
AV
461 } else {
462 if (inode->i_data.nrpages)
463 truncate_inode_pages(&inode->i_data, 0);
30140837 464 end_writeback(inode);
b4272d4c 465 }
661074e9
AV
466 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
467 bd_forget(inode);
468 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
469 cd_forget(inode);
b2b2af8e
DC
470
471 remove_inode_hash(inode);
472
473 spin_lock(&inode->i_lock);
474 wake_up_bit(&inode->i_state, __I_NEW);
475 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
476 spin_unlock(&inode->i_lock);
477
478 destroy_inode(inode);
b4272d4c
AV
479}
480
1da177e4
LT
481/*
482 * dispose_list - dispose of the contents of a local list
483 * @head: the head of the list to free
484 *
485 * Dispose-list gets a local list with local inodes in it, so it doesn't
486 * need to worry about list corruption and SMP locks.
487 */
488static void dispose_list(struct list_head *head)
489{
1da177e4
LT
490 while (!list_empty(head)) {
491 struct inode *inode;
492
7ccf19a8
NP
493 inode = list_first_entry(head, struct inode, i_lru);
494 list_del_init(&inode->i_lru);
1da177e4 495
644da596 496 evict(inode);
1da177e4 497 }
1da177e4
LT
498}
499
63997e98
AV
500/**
501 * evict_inodes - evict all evictable inodes for a superblock
502 * @sb: superblock to operate on
503 *
504 * Make sure that no inodes with zero refcount are retained. This is
505 * called by superblock shutdown after having MS_ACTIVE flag removed,
506 * so any inode reaching zero refcount during or after that call will
507 * be immediately evicted.
1da177e4 508 */
63997e98 509void evict_inodes(struct super_block *sb)
1da177e4 510{
63997e98
AV
511 struct inode *inode, *next;
512 LIST_HEAD(dispose);
1da177e4 513
55fa6091 514 spin_lock(&inode_sb_list_lock);
63997e98
AV
515 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
516 if (atomic_read(&inode->i_count))
aabb8fdb 517 continue;
250df6ed
DC
518
519 spin_lock(&inode->i_lock);
520 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
521 spin_unlock(&inode->i_lock);
1da177e4 522 continue;
250df6ed 523 }
63997e98
AV
524
525 inode->i_state |= I_FREEING;
02afc410 526 inode_lru_list_del(inode);
250df6ed 527 spin_unlock(&inode->i_lock);
02afc410 528 list_add(&inode->i_lru, &dispose);
1da177e4 529 }
55fa6091 530 spin_unlock(&inode_sb_list_lock);
63997e98
AV
531
532 dispose_list(&dispose);
bab1d944
CH
533
534 /*
535 * Cycle through iprune_sem to make sure any inode that prune_icache
536 * moved off the list before we took the lock has been fully torn
537 * down.
538 */
539 down_write(&iprune_sem);
63997e98 540 up_write(&iprune_sem);
1da177e4
LT
541}
542
1da177e4 543/**
a0318786
CH
544 * invalidate_inodes - attempt to free all inodes on a superblock
545 * @sb: superblock to operate on
93b270f7 546 * @kill_dirty: flag to guide handling of dirty inodes
1da177e4 547 *
a0318786
CH
548 * Attempts to free all inodes for a given superblock. If there were any
549 * busy inodes return a non-zero value, else zero.
93b270f7
N
550 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
551 * them as busy.
1da177e4 552 */
93b270f7 553int invalidate_inodes(struct super_block *sb, bool kill_dirty)
1da177e4 554{
cffbc8aa 555 int busy = 0;
a0318786
CH
556 struct inode *inode, *next;
557 LIST_HEAD(dispose);
1da177e4 558
55fa6091 559 spin_lock(&inode_sb_list_lock);
a0318786 560 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
250df6ed
DC
561 spin_lock(&inode->i_lock);
562 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
563 spin_unlock(&inode->i_lock);
aabb8fdb 564 continue;
250df6ed 565 }
93b270f7 566 if (inode->i_state & I_DIRTY && !kill_dirty) {
250df6ed 567 spin_unlock(&inode->i_lock);
93b270f7
N
568 busy = 1;
569 continue;
570 }
99a38919 571 if (atomic_read(&inode->i_count)) {
250df6ed 572 spin_unlock(&inode->i_lock);
99a38919 573 busy = 1;
1da177e4
LT
574 continue;
575 }
99a38919 576
99a38919 577 inode->i_state |= I_FREEING;
02afc410 578 inode_lru_list_del(inode);
250df6ed 579 spin_unlock(&inode->i_lock);
02afc410 580 list_add(&inode->i_lru, &dispose);
1da177e4 581 }
55fa6091 582 spin_unlock(&inode_sb_list_lock);
1da177e4 583
a0318786 584 dispose_list(&dispose);
1da177e4
LT
585
586 return busy;
587}
1da177e4
LT
588
589static int can_unuse(struct inode *inode)
590{
9e38d86f 591 if (inode->i_state & ~I_REFERENCED)
1da177e4
LT
592 return 0;
593 if (inode_has_buffers(inode))
594 return 0;
595 if (atomic_read(&inode->i_count))
596 return 0;
597 if (inode->i_data.nrpages)
598 return 0;
599 return 1;
600}
601
602/*
9e38d86f 603 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
02afc410 604 * temporary list and then are freed outside inode_lru_lock by dispose_list().
1da177e4
LT
605 *
606 * Any inodes which are pinned purely because of attached pagecache have their
9e38d86f
NP
607 * pagecache removed. If the inode has metadata buffers attached to
608 * mapping->private_list then try to remove them.
1da177e4 609 *
9e38d86f
NP
610 * If the inode has the I_REFERENCED flag set, then it means that it has been
611 * used recently - the flag is set in iput_final(). When we encounter such an
612 * inode, clear the flag and move it to the back of the LRU so it gets another
613 * pass through the LRU before it gets reclaimed. This is necessary because of
614 * the fact we are doing lazy LRU updates to minimise lock contention so the
615 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
616 * with this flag set because they are the inodes that are out of order.
1da177e4
LT
617 */
618static void prune_icache(int nr_to_scan)
619{
620 LIST_HEAD(freeable);
1da177e4
LT
621 int nr_scanned;
622 unsigned long reap = 0;
623
88e0fbc4 624 down_read(&iprune_sem);
02afc410 625 spin_lock(&inode_lru_lock);
1da177e4
LT
626 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
627 struct inode *inode;
628
7ccf19a8 629 if (list_empty(&inode_lru))
1da177e4
LT
630 break;
631
7ccf19a8 632 inode = list_entry(inode_lru.prev, struct inode, i_lru);
1da177e4 633
02afc410
DC
634 /*
635 * we are inverting the inode_lru_lock/inode->i_lock here,
636 * so use a trylock. If we fail to get the lock, just move the
637 * inode to the back of the list so we don't spin on it.
638 */
639 if (!spin_trylock(&inode->i_lock)) {
640 list_move(&inode->i_lru, &inode_lru);
641 continue;
642 }
643
9e38d86f
NP
644 /*
645 * Referenced or dirty inodes are still in use. Give them
646 * another pass through the LRU as we canot reclaim them now.
647 */
648 if (atomic_read(&inode->i_count) ||
649 (inode->i_state & ~I_REFERENCED)) {
7ccf19a8 650 list_del_init(&inode->i_lru);
f283c86a 651 spin_unlock(&inode->i_lock);
86c8749e 652 inodes_stat.nr_unused--;
9e38d86f
NP
653 continue;
654 }
655
656 /* recently referenced inodes get one more pass */
657 if (inode->i_state & I_REFERENCED) {
9e38d86f 658 inode->i_state &= ~I_REFERENCED;
250df6ed 659 list_move(&inode->i_lru, &inode_lru);
f283c86a 660 spin_unlock(&inode->i_lock);
1da177e4
LT
661 continue;
662 }
663 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
664 __iget(inode);
250df6ed 665 spin_unlock(&inode->i_lock);
02afc410 666 spin_unlock(&inode_lru_lock);
1da177e4 667 if (remove_inode_buffers(inode))
fc0ecff6
AM
668 reap += invalidate_mapping_pages(&inode->i_data,
669 0, -1);
1da177e4 670 iput(inode);
02afc410 671 spin_lock(&inode_lru_lock);
1da177e4 672
7ccf19a8
NP
673 if (inode != list_entry(inode_lru.next,
674 struct inode, i_lru))
1da177e4 675 continue; /* wrong inode or list_empty */
02afc410
DC
676 /* avoid lock inversions with trylock */
677 if (!spin_trylock(&inode->i_lock))
678 continue;
250df6ed
DC
679 if (!can_unuse(inode)) {
680 spin_unlock(&inode->i_lock);
1da177e4 681 continue;
250df6ed 682 }
1da177e4 683 }
7ef0d737 684 WARN_ON(inode->i_state & I_NEW);
1da177e4 685 inode->i_state |= I_FREEING;
250df6ed 686 spin_unlock(&inode->i_lock);
7ccf19a8 687
7ccf19a8 688 list_move(&inode->i_lru, &freeable);
86c8749e 689 inodes_stat.nr_unused--;
1da177e4 690 }
f8891e5e
CL
691 if (current_is_kswapd())
692 __count_vm_events(KSWAPD_INODESTEAL, reap);
693 else
694 __count_vm_events(PGINODESTEAL, reap);
02afc410 695 spin_unlock(&inode_lru_lock);
1da177e4
LT
696
697 dispose_list(&freeable);
88e0fbc4 698 up_read(&iprune_sem);
1da177e4
LT
699}
700
701/*
702 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
703 * "unused" means that no dentries are referring to the inodes: the files are
704 * not open and the dcache references to those inodes have already been
705 * reclaimed.
706 *
707 * This function is passed the number of inodes to scan, and it returns the
708 * total number of remaining possibly-reclaimable inodes.
709 */
1495f230
YH
710static int shrink_icache_memory(struct shrinker *shrink,
711 struct shrink_control *sc)
1da177e4 712{
1495f230
YH
713 int nr = sc->nr_to_scan;
714 gfp_t gfp_mask = sc->gfp_mask;
715
1da177e4
LT
716 if (nr) {
717 /*
718 * Nasty deadlock avoidance. We may hold various FS locks,
719 * and we don't want to recurse into the FS that called us
720 * in clear_inode() and friends..
6b3304b5 721 */
1da177e4
LT
722 if (!(gfp_mask & __GFP_FS))
723 return -1;
724 prune_icache(nr);
725 }
cffbc8aa 726 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
1da177e4
LT
727}
728
8e1f936b
RR
729static struct shrinker icache_shrinker = {
730 .shrink = shrink_icache_memory,
731 .seeks = DEFAULT_SEEKS,
732};
733
1da177e4
LT
734static void __wait_on_freeing_inode(struct inode *inode);
735/*
736 * Called with the inode lock held.
1da177e4 737 */
6b3304b5
MK
738static struct inode *find_inode(struct super_block *sb,
739 struct hlist_head *head,
740 int (*test)(struct inode *, void *),
741 void *data)
1da177e4
LT
742{
743 struct hlist_node *node;
6b3304b5 744 struct inode *inode = NULL;
1da177e4
LT
745
746repeat:
c5c8be3c 747 hlist_for_each_entry(inode, node, head, i_hash) {
67a23c49
DC
748 spin_lock(&inode->i_lock);
749 if (inode->i_sb != sb) {
750 spin_unlock(&inode->i_lock);
1da177e4 751 continue;
67a23c49
DC
752 }
753 if (!test(inode, data)) {
754 spin_unlock(&inode->i_lock);
1da177e4 755 continue;
67a23c49 756 }
a4ffdde6 757 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1da177e4
LT
758 __wait_on_freeing_inode(inode);
759 goto repeat;
760 }
f7899bd5 761 __iget(inode);
250df6ed 762 spin_unlock(&inode->i_lock);
f7899bd5 763 return inode;
1da177e4 764 }
f7899bd5 765 return NULL;
1da177e4
LT
766}
767
768/*
769 * find_inode_fast is the fast path version of find_inode, see the comment at
770 * iget_locked for details.
771 */
6b3304b5
MK
772static struct inode *find_inode_fast(struct super_block *sb,
773 struct hlist_head *head, unsigned long ino)
1da177e4
LT
774{
775 struct hlist_node *node;
6b3304b5 776 struct inode *inode = NULL;
1da177e4
LT
777
778repeat:
c5c8be3c 779 hlist_for_each_entry(inode, node, head, i_hash) {
67a23c49
DC
780 spin_lock(&inode->i_lock);
781 if (inode->i_ino != ino) {
782 spin_unlock(&inode->i_lock);
1da177e4 783 continue;
67a23c49
DC
784 }
785 if (inode->i_sb != sb) {
786 spin_unlock(&inode->i_lock);
1da177e4 787 continue;
67a23c49 788 }
a4ffdde6 789 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1da177e4
LT
790 __wait_on_freeing_inode(inode);
791 goto repeat;
792 }
f7899bd5 793 __iget(inode);
250df6ed 794 spin_unlock(&inode->i_lock);
f7899bd5 795 return inode;
1da177e4 796 }
f7899bd5 797 return NULL;
8290c35f
DC
798}
799
f991bd2e
ED
800/*
801 * Each cpu owns a range of LAST_INO_BATCH numbers.
802 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
803 * to renew the exhausted range.
8290c35f 804 *
f991bd2e
ED
805 * This does not significantly increase overflow rate because every CPU can
806 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
807 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
808 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
809 * overflow rate by 2x, which does not seem too significant.
810 *
811 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
812 * error if st_ino won't fit in target struct field. Use 32bit counter
813 * here to attempt to avoid that.
8290c35f 814 */
f991bd2e
ED
815#define LAST_INO_BATCH 1024
816static DEFINE_PER_CPU(unsigned int, last_ino);
817
85fe4025 818unsigned int get_next_ino(void)
8290c35f 819{
f991bd2e
ED
820 unsigned int *p = &get_cpu_var(last_ino);
821 unsigned int res = *p;
8290c35f 822
f991bd2e
ED
823#ifdef CONFIG_SMP
824 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
825 static atomic_t shared_last_ino;
826 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
827
828 res = next - LAST_INO_BATCH;
829 }
830#endif
831
832 *p = ++res;
833 put_cpu_var(last_ino);
834 return res;
8290c35f 835}
85fe4025 836EXPORT_SYMBOL(get_next_ino);
8290c35f 837
1da177e4
LT
838/**
839 * new_inode - obtain an inode
840 * @sb: superblock
841 *
769848c0 842 * Allocates a new inode for given superblock. The default gfp_mask
3c1d4378 843 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
769848c0
MG
844 * If HIGHMEM pages are unsuitable or it is known that pages allocated
845 * for the page cache are not reclaimable or migratable,
846 * mapping_set_gfp_mask() must be called with suitable flags on the
847 * newly created inode's mapping
848 *
1da177e4
LT
849 */
850struct inode *new_inode(struct super_block *sb)
851{
6b3304b5 852 struct inode *inode;
1da177e4 853
55fa6091 854 spin_lock_prefetch(&inode_sb_list_lock);
6b3304b5 855
1da177e4
LT
856 inode = alloc_inode(sb);
857 if (inode) {
250df6ed 858 spin_lock(&inode->i_lock);
1da177e4 859 inode->i_state = 0;
250df6ed 860 spin_unlock(&inode->i_lock);
55fa6091 861 inode_sb_list_add(inode);
1da177e4
LT
862 }
863 return inode;
864}
1da177e4
LT
865EXPORT_SYMBOL(new_inode);
866
250df6ed
DC
867/**
868 * unlock_new_inode - clear the I_NEW state and wake up any waiters
869 * @inode: new inode to unlock
870 *
871 * Called when the inode is fully initialised to clear the new state of the
872 * inode and wake up anyone waiting for the inode to finish initialisation.
873 */
1da177e4
LT
874void unlock_new_inode(struct inode *inode)
875{
14358e6d 876#ifdef CONFIG_DEBUG_LOCK_ALLOC
a3314a0e 877 if (S_ISDIR(inode->i_mode)) {
1e89a5e1
PZ
878 struct file_system_type *type = inode->i_sb->s_type;
879
9a7aa12f
JK
880 /* Set new key only if filesystem hasn't already changed it */
881 if (!lockdep_match_class(&inode->i_mutex,
882 &type->i_mutex_key)) {
883 /*
884 * ensure nobody is actually holding i_mutex
885 */
886 mutex_destroy(&inode->i_mutex);
887 mutex_init(&inode->i_mutex);
888 lockdep_set_class(&inode->i_mutex,
889 &type->i_mutex_dir_key);
890 }
1e89a5e1 891 }
14358e6d 892#endif
250df6ed 893 spin_lock(&inode->i_lock);
eaff8079
CH
894 WARN_ON(!(inode->i_state & I_NEW));
895 inode->i_state &= ~I_NEW;
250df6ed
DC
896 wake_up_bit(&inode->i_state, __I_NEW);
897 spin_unlock(&inode->i_lock);
1da177e4 898}
1da177e4
LT
899EXPORT_SYMBOL(unlock_new_inode);
900
0b2d0724
CH
901/**
902 * iget5_locked - obtain an inode from a mounted file system
903 * @sb: super block of file system
904 * @hashval: hash value (usually inode number) to get
905 * @test: callback used for comparisons between inodes
906 * @set: callback used to initialize a new struct inode
907 * @data: opaque data pointer to pass to @test and @set
908 *
909 * Search for the inode specified by @hashval and @data in the inode cache,
910 * and if present it is return it with an increased reference count. This is
911 * a generalized version of iget_locked() for file systems where the inode
912 * number is not sufficient for unique identification of an inode.
913 *
914 * If the inode is not in cache, allocate a new inode and return it locked,
915 * hashed, and with the I_NEW flag set. The file system gets to fill it in
916 * before unlocking it via unlock_new_inode().
1da177e4 917 *
0b2d0724
CH
918 * Note both @test and @set are called with the inode_hash_lock held, so can't
919 * sleep.
1da177e4 920 */
0b2d0724
CH
921struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
922 int (*test)(struct inode *, void *),
923 int (*set)(struct inode *, void *), void *data)
1da177e4 924{
0b2d0724 925 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
6b3304b5 926 struct inode *inode;
1da177e4 927
0b2d0724
CH
928 spin_lock(&inode_hash_lock);
929 inode = find_inode(sb, head, test, data);
930 spin_unlock(&inode_hash_lock);
931
932 if (inode) {
933 wait_on_inode(inode);
934 return inode;
935 }
936
1da177e4
LT
937 inode = alloc_inode(sb);
938 if (inode) {
6b3304b5 939 struct inode *old;
1da177e4 940
67a23c49 941 spin_lock(&inode_hash_lock);
1da177e4
LT
942 /* We released the lock, so.. */
943 old = find_inode(sb, head, test, data);
944 if (!old) {
945 if (set(inode, data))
946 goto set_failed;
947
250df6ed
DC
948 spin_lock(&inode->i_lock);
949 inode->i_state = I_NEW;
646ec461 950 hlist_add_head(&inode->i_hash, head);
250df6ed 951 spin_unlock(&inode->i_lock);
55fa6091 952 inode_sb_list_add(inode);
67a23c49 953 spin_unlock(&inode_hash_lock);
1da177e4
LT
954
955 /* Return the locked inode with I_NEW set, the
956 * caller is responsible for filling in the contents
957 */
958 return inode;
959 }
960
961 /*
962 * Uhhuh, somebody else created the same inode under
963 * us. Use the old inode instead of the one we just
964 * allocated.
965 */
67a23c49 966 spin_unlock(&inode_hash_lock);
1da177e4
LT
967 destroy_inode(inode);
968 inode = old;
969 wait_on_inode(inode);
970 }
971 return inode;
972
973set_failed:
67a23c49 974 spin_unlock(&inode_hash_lock);
1da177e4
LT
975 destroy_inode(inode);
976 return NULL;
977}
0b2d0724 978EXPORT_SYMBOL(iget5_locked);
1da177e4 979
0b2d0724
CH
980/**
981 * iget_locked - obtain an inode from a mounted file system
982 * @sb: super block of file system
983 * @ino: inode number to get
984 *
985 * Search for the inode specified by @ino in the inode cache and if present
986 * return it with an increased reference count. This is for file systems
987 * where the inode number is sufficient for unique identification of an inode.
988 *
989 * If the inode is not in cache, allocate a new inode and return it locked,
990 * hashed, and with the I_NEW flag set. The file system gets to fill it in
991 * before unlocking it via unlock_new_inode().
1da177e4 992 */
0b2d0724 993struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1da177e4 994{
0b2d0724 995 struct hlist_head *head = inode_hashtable + hash(sb, ino);
6b3304b5 996 struct inode *inode;
1da177e4 997
0b2d0724
CH
998 spin_lock(&inode_hash_lock);
999 inode = find_inode_fast(sb, head, ino);
1000 spin_unlock(&inode_hash_lock);
1001 if (inode) {
1002 wait_on_inode(inode);
1003 return inode;
1004 }
1005
1da177e4
LT
1006 inode = alloc_inode(sb);
1007 if (inode) {
6b3304b5 1008 struct inode *old;
1da177e4 1009
67a23c49 1010 spin_lock(&inode_hash_lock);
1da177e4
LT
1011 /* We released the lock, so.. */
1012 old = find_inode_fast(sb, head, ino);
1013 if (!old) {
1014 inode->i_ino = ino;
250df6ed
DC
1015 spin_lock(&inode->i_lock);
1016 inode->i_state = I_NEW;
646ec461 1017 hlist_add_head(&inode->i_hash, head);
250df6ed 1018 spin_unlock(&inode->i_lock);
55fa6091 1019 inode_sb_list_add(inode);
67a23c49 1020 spin_unlock(&inode_hash_lock);
1da177e4
LT
1021
1022 /* Return the locked inode with I_NEW set, the
1023 * caller is responsible for filling in the contents
1024 */
1025 return inode;
1026 }
1027
1028 /*
1029 * Uhhuh, somebody else created the same inode under
1030 * us. Use the old inode instead of the one we just
1031 * allocated.
1032 */
67a23c49 1033 spin_unlock(&inode_hash_lock);
1da177e4
LT
1034 destroy_inode(inode);
1035 inode = old;
1036 wait_on_inode(inode);
1037 }
1038 return inode;
1039}
0b2d0724 1040EXPORT_SYMBOL(iget_locked);
1da177e4 1041
ad5e195a
CH
1042/*
1043 * search the inode cache for a matching inode number.
1044 * If we find one, then the inode number we are trying to
1045 * allocate is not unique and so we should not use it.
1046 *
1047 * Returns 1 if the inode number is unique, 0 if it is not.
1048 */
1049static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1050{
1051 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1052 struct hlist_node *node;
1053 struct inode *inode;
1054
67a23c49 1055 spin_lock(&inode_hash_lock);
ad5e195a 1056 hlist_for_each_entry(inode, node, b, i_hash) {
67a23c49
DC
1057 if (inode->i_ino == ino && inode->i_sb == sb) {
1058 spin_unlock(&inode_hash_lock);
ad5e195a 1059 return 0;
67a23c49 1060 }
ad5e195a 1061 }
67a23c49 1062 spin_unlock(&inode_hash_lock);
ad5e195a
CH
1063
1064 return 1;
1065}
1066
1da177e4
LT
1067/**
1068 * iunique - get a unique inode number
1069 * @sb: superblock
1070 * @max_reserved: highest reserved inode number
1071 *
1072 * Obtain an inode number that is unique on the system for a given
1073 * superblock. This is used by file systems that have no natural
1074 * permanent inode numbering system. An inode number is returned that
1075 * is higher than the reserved limit but unique.
1076 *
1077 * BUGS:
1078 * With a large number of inodes live on the file system this function
1079 * currently becomes quite slow.
1080 */
1081ino_t iunique(struct super_block *sb, ino_t max_reserved)
1082{
866b04fc
JL
1083 /*
1084 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1085 * error if st_ino won't fit in target struct field. Use 32bit counter
1086 * here to attempt to avoid that.
1087 */
ad5e195a 1088 static DEFINE_SPINLOCK(iunique_lock);
866b04fc 1089 static unsigned int counter;
1da177e4 1090 ino_t res;
3361c7be 1091
ad5e195a 1092 spin_lock(&iunique_lock);
3361c7be
JL
1093 do {
1094 if (counter <= max_reserved)
1095 counter = max_reserved + 1;
1da177e4 1096 res = counter++;
ad5e195a
CH
1097 } while (!test_inode_iunique(sb, res));
1098 spin_unlock(&iunique_lock);
1da177e4 1099
3361c7be
JL
1100 return res;
1101}
1da177e4
LT
1102EXPORT_SYMBOL(iunique);
1103
1104struct inode *igrab(struct inode *inode)
1105{
250df6ed
DC
1106 spin_lock(&inode->i_lock);
1107 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1da177e4 1108 __iget(inode);
250df6ed
DC
1109 spin_unlock(&inode->i_lock);
1110 } else {
1111 spin_unlock(&inode->i_lock);
1da177e4
LT
1112 /*
1113 * Handle the case where s_op->clear_inode is not been
1114 * called yet, and somebody is calling igrab
1115 * while the inode is getting freed.
1116 */
1117 inode = NULL;
250df6ed 1118 }
1da177e4
LT
1119 return inode;
1120}
1da177e4
LT
1121EXPORT_SYMBOL(igrab);
1122
1123/**
0b2d0724 1124 * ilookup5_nowait - search for an inode in the inode cache
1da177e4 1125 * @sb: super block of file system to search
0b2d0724 1126 * @hashval: hash value (usually inode number) to search for
1da177e4
LT
1127 * @test: callback used for comparisons between inodes
1128 * @data: opaque data pointer to pass to @test
1da177e4 1129 *
0b2d0724 1130 * Search for the inode specified by @hashval and @data in the inode cache.
1da177e4
LT
1131 * If the inode is in the cache, the inode is returned with an incremented
1132 * reference count.
1133 *
0b2d0724
CH
1134 * Note: I_NEW is not waited upon so you have to be very careful what you do
1135 * with the returned inode. You probably should be using ilookup5() instead.
1da177e4 1136 *
b6d0ad68 1137 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1da177e4 1138 */
0b2d0724
CH
1139struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1140 int (*test)(struct inode *, void *), void *data)
1da177e4 1141{
0b2d0724 1142 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1da177e4
LT
1143 struct inode *inode;
1144
67a23c49 1145 spin_lock(&inode_hash_lock);
1da177e4 1146 inode = find_inode(sb, head, test, data);
67a23c49 1147 spin_unlock(&inode_hash_lock);
88bd5121 1148
0b2d0724 1149 return inode;
88bd5121 1150}
88bd5121
AA
1151EXPORT_SYMBOL(ilookup5_nowait);
1152
1153/**
1154 * ilookup5 - search for an inode in the inode cache
1155 * @sb: super block of file system to search
1156 * @hashval: hash value (usually inode number) to search for
1157 * @test: callback used for comparisons between inodes
1158 * @data: opaque data pointer to pass to @test
1159 *
0b2d0724
CH
1160 * Search for the inode specified by @hashval and @data in the inode cache,
1161 * and if the inode is in the cache, return the inode with an incremented
1162 * reference count. Waits on I_NEW before returning the inode.
88bd5121 1163 * returned with an incremented reference count.
1da177e4 1164 *
0b2d0724
CH
1165 * This is a generalized version of ilookup() for file systems where the
1166 * inode number is not sufficient for unique identification of an inode.
1da177e4 1167 *
0b2d0724 1168 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1da177e4
LT
1169 */
1170struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1171 int (*test)(struct inode *, void *), void *data)
1172{
0b2d0724 1173 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1da177e4 1174
0b2d0724
CH
1175 if (inode)
1176 wait_on_inode(inode);
1177 return inode;
1da177e4 1178}
1da177e4
LT
1179EXPORT_SYMBOL(ilookup5);
1180
1181/**
1182 * ilookup - search for an inode in the inode cache
1183 * @sb: super block of file system to search
1184 * @ino: inode number to search for
1185 *
0b2d0724
CH
1186 * Search for the inode @ino in the inode cache, and if the inode is in the
1187 * cache, the inode is returned with an incremented reference count.
1da177e4
LT
1188 */
1189struct inode *ilookup(struct super_block *sb, unsigned long ino)
1190{
1191 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1da177e4
LT
1192 struct inode *inode;
1193
0b2d0724
CH
1194 spin_lock(&inode_hash_lock);
1195 inode = find_inode_fast(sb, head, ino);
1196 spin_unlock(&inode_hash_lock);
1da177e4 1197
1da177e4 1198 if (inode)
0b2d0724
CH
1199 wait_on_inode(inode);
1200 return inode;
1da177e4 1201}
0b2d0724 1202EXPORT_SYMBOL(ilookup);
1da177e4 1203
261bca86
AV
1204int insert_inode_locked(struct inode *inode)
1205{
1206 struct super_block *sb = inode->i_sb;
1207 ino_t ino = inode->i_ino;
1208 struct hlist_head *head = inode_hashtable + hash(sb, ino);
261bca86 1209
261bca86 1210 while (1) {
72a43d63
AV
1211 struct hlist_node *node;
1212 struct inode *old = NULL;
67a23c49 1213 spin_lock(&inode_hash_lock);
72a43d63
AV
1214 hlist_for_each_entry(old, node, head, i_hash) {
1215 if (old->i_ino != ino)
1216 continue;
1217 if (old->i_sb != sb)
1218 continue;
250df6ed
DC
1219 spin_lock(&old->i_lock);
1220 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1221 spin_unlock(&old->i_lock);
72a43d63 1222 continue;
250df6ed 1223 }
72a43d63
AV
1224 break;
1225 }
1226 if (likely(!node)) {
250df6ed
DC
1227 spin_lock(&inode->i_lock);
1228 inode->i_state |= I_NEW;
261bca86 1229 hlist_add_head(&inode->i_hash, head);
250df6ed 1230 spin_unlock(&inode->i_lock);
67a23c49 1231 spin_unlock(&inode_hash_lock);
261bca86
AV
1232 return 0;
1233 }
1234 __iget(old);
250df6ed 1235 spin_unlock(&old->i_lock);
67a23c49 1236 spin_unlock(&inode_hash_lock);
261bca86 1237 wait_on_inode(old);
1d3382cb 1238 if (unlikely(!inode_unhashed(old))) {
261bca86
AV
1239 iput(old);
1240 return -EBUSY;
1241 }
1242 iput(old);
1243 }
1244}
261bca86
AV
1245EXPORT_SYMBOL(insert_inode_locked);
1246
1247int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1248 int (*test)(struct inode *, void *), void *data)
1249{
1250 struct super_block *sb = inode->i_sb;
1251 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
261bca86 1252
261bca86 1253 while (1) {
72a43d63
AV
1254 struct hlist_node *node;
1255 struct inode *old = NULL;
1256
67a23c49 1257 spin_lock(&inode_hash_lock);
72a43d63
AV
1258 hlist_for_each_entry(old, node, head, i_hash) {
1259 if (old->i_sb != sb)
1260 continue;
1261 if (!test(old, data))
1262 continue;
250df6ed
DC
1263 spin_lock(&old->i_lock);
1264 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1265 spin_unlock(&old->i_lock);
72a43d63 1266 continue;
250df6ed 1267 }
72a43d63
AV
1268 break;
1269 }
1270 if (likely(!node)) {
250df6ed
DC
1271 spin_lock(&inode->i_lock);
1272 inode->i_state |= I_NEW;
261bca86 1273 hlist_add_head(&inode->i_hash, head);
250df6ed 1274 spin_unlock(&inode->i_lock);
67a23c49 1275 spin_unlock(&inode_hash_lock);
261bca86
AV
1276 return 0;
1277 }
1278 __iget(old);
250df6ed 1279 spin_unlock(&old->i_lock);
67a23c49 1280 spin_unlock(&inode_hash_lock);
261bca86 1281 wait_on_inode(old);
1d3382cb 1282 if (unlikely(!inode_unhashed(old))) {
261bca86
AV
1283 iput(old);
1284 return -EBUSY;
1285 }
1286 iput(old);
1287 }
1288}
261bca86
AV
1289EXPORT_SYMBOL(insert_inode_locked4);
1290
1da177e4 1291
45321ac5
AV
1292int generic_delete_inode(struct inode *inode)
1293{
1294 return 1;
1295}
1296EXPORT_SYMBOL(generic_delete_inode);
1297
1da177e4 1298/*
45321ac5
AV
1299 * Normal UNIX filesystem behaviour: delete the
1300 * inode when the usage count drops to zero, and
1301 * i_nlink is zero.
1da177e4 1302 */
45321ac5 1303int generic_drop_inode(struct inode *inode)
1da177e4 1304{
1d3382cb 1305 return !inode->i_nlink || inode_unhashed(inode);
1da177e4 1306}
45321ac5 1307EXPORT_SYMBOL_GPL(generic_drop_inode);
1da177e4 1308
45321ac5
AV
1309/*
1310 * Called when we're dropping the last reference
1311 * to an inode.
22fe4042 1312 *
45321ac5
AV
1313 * Call the FS "drop_inode()" function, defaulting to
1314 * the legacy UNIX filesystem behaviour. If it tells
1315 * us to evict inode, do so. Otherwise, retain inode
1316 * in cache if fs is alive, sync and evict if fs is
1317 * shutting down.
22fe4042 1318 */
45321ac5 1319static void iput_final(struct inode *inode)
1da177e4
LT
1320{
1321 struct super_block *sb = inode->i_sb;
45321ac5
AV
1322 const struct super_operations *op = inode->i_sb->s_op;
1323 int drop;
1324
250df6ed
DC
1325 WARN_ON(inode->i_state & I_NEW);
1326
45321ac5
AV
1327 if (op && op->drop_inode)
1328 drop = op->drop_inode(inode);
1329 else
1330 drop = generic_drop_inode(inode);
1da177e4 1331
b2b2af8e
DC
1332 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1333 inode->i_state |= I_REFERENCED;
1334 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1335 inode_lru_list_add(inode);
1336 spin_unlock(&inode->i_lock);
b2b2af8e
DC
1337 return;
1338 }
1339
45321ac5 1340 if (!drop) {
991114c6 1341 inode->i_state |= I_WILL_FREE;
250df6ed 1342 spin_unlock(&inode->i_lock);
1da177e4 1343 write_inode_now(inode, 1);
250df6ed 1344 spin_lock(&inode->i_lock);
7ef0d737 1345 WARN_ON(inode->i_state & I_NEW);
991114c6 1346 inode->i_state &= ~I_WILL_FREE;
1da177e4 1347 }
7ccf19a8 1348
991114c6 1349 inode->i_state |= I_FREEING;
9e38d86f 1350 inode_lru_list_del(inode);
b2b2af8e 1351 spin_unlock(&inode->i_lock);
b2b2af8e 1352
644da596 1353 evict(inode);
1da177e4
LT
1354}
1355
1da177e4 1356/**
6b3304b5 1357 * iput - put an inode
1da177e4
LT
1358 * @inode: inode to put
1359 *
1360 * Puts an inode, dropping its usage count. If the inode use count hits
1361 * zero, the inode is then freed and may also be destroyed.
1362 *
1363 * Consequently, iput() can sleep.
1364 */
1365void iput(struct inode *inode)
1366{
1367 if (inode) {
a4ffdde6 1368 BUG_ON(inode->i_state & I_CLEAR);
1da177e4 1369
f283c86a 1370 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1da177e4
LT
1371 iput_final(inode);
1372 }
1373}
1da177e4
LT
1374EXPORT_SYMBOL(iput);
1375
1376/**
1377 * bmap - find a block number in a file
1378 * @inode: inode of file
1379 * @block: block to find
1380 *
1381 * Returns the block number on the device holding the inode that
1382 * is the disk block number for the block of the file requested.
1383 * That is, asked for block 4 of inode 1 the function will return the
6b3304b5 1384 * disk block relative to the disk start that holds that block of the
1da177e4
LT
1385 * file.
1386 */
6b3304b5 1387sector_t bmap(struct inode *inode, sector_t block)
1da177e4
LT
1388{
1389 sector_t res = 0;
1390 if (inode->i_mapping->a_ops->bmap)
1391 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1392 return res;
1393}
1da177e4
LT
1394EXPORT_SYMBOL(bmap);
1395
11ff6f05
MG
1396/*
1397 * With relative atime, only update atime if the previous atime is
1398 * earlier than either the ctime or mtime or if at least a day has
1399 * passed since the last atime update.
1400 */
1401static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1402 struct timespec now)
1403{
1404
1405 if (!(mnt->mnt_flags & MNT_RELATIME))
1406 return 1;
1407 /*
1408 * Is mtime younger than atime? If yes, update atime:
1409 */
1410 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1411 return 1;
1412 /*
1413 * Is ctime younger than atime? If yes, update atime:
1414 */
1415 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1416 return 1;
1417
1418 /*
1419 * Is the previous atime value older than a day? If yes,
1420 * update atime:
1421 */
1422 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1423 return 1;
1424 /*
1425 * Good, we can skip the atime update:
1426 */
1427 return 0;
1428}
1429
1da177e4 1430/**
869243a0
CH
1431 * touch_atime - update the access time
1432 * @mnt: mount the inode is accessed on
7045f37b 1433 * @dentry: dentry accessed
1da177e4
LT
1434 *
1435 * Update the accessed time on an inode and mark it for writeback.
1436 * This function automatically handles read only file systems and media,
1437 * as well as the "noatime" flag and inode specific "noatime" markers.
1438 */
869243a0 1439void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1da177e4 1440{
869243a0 1441 struct inode *inode = dentry->d_inode;
1da177e4
LT
1442 struct timespec now;
1443
cdb70f3f 1444 if (inode->i_flags & S_NOATIME)
b12536c2 1445 return;
37756ced 1446 if (IS_NOATIME(inode))
b12536c2 1447 return;
b2276138 1448 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
b12536c2 1449 return;
47ae32d6 1450
cdb70f3f 1451 if (mnt->mnt_flags & MNT_NOATIME)
b12536c2 1452 return;
cdb70f3f 1453 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
b12536c2 1454 return;
1da177e4
LT
1455
1456 now = current_fs_time(inode->i_sb);
11ff6f05
MG
1457
1458 if (!relatime_need_update(mnt, inode, now))
b12536c2 1459 return;
11ff6f05 1460
47ae32d6 1461 if (timespec_equal(&inode->i_atime, &now))
b12536c2
AK
1462 return;
1463
1464 if (mnt_want_write(mnt))
1465 return;
47ae32d6
VH
1466
1467 inode->i_atime = now;
1468 mark_inode_dirty_sync(inode);
cdb70f3f 1469 mnt_drop_write(mnt);
1da177e4 1470}
869243a0 1471EXPORT_SYMBOL(touch_atime);
1da177e4
LT
1472
1473/**
870f4817
CH
1474 * file_update_time - update mtime and ctime time
1475 * @file: file accessed
1da177e4 1476 *
870f4817
CH
1477 * Update the mtime and ctime members of an inode and mark the inode
1478 * for writeback. Note that this function is meant exclusively for
1479 * usage in the file write path of filesystems, and filesystems may
1480 * choose to explicitly ignore update via this function with the
2eadfc0e 1481 * S_NOCMTIME inode flag, e.g. for network filesystem where these
870f4817 1482 * timestamps are handled by the server.
1da177e4
LT
1483 */
1484
870f4817 1485void file_update_time(struct file *file)
1da177e4 1486{
0f7fc9e4 1487 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4 1488 struct timespec now;
ce06e0b2 1489 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1da177e4 1490
ce06e0b2 1491 /* First try to exhaust all avenues to not sync */
1da177e4
LT
1492 if (IS_NOCMTIME(inode))
1493 return;
20ddee2c 1494
1da177e4 1495 now = current_fs_time(inode->i_sb);
ce06e0b2
AK
1496 if (!timespec_equal(&inode->i_mtime, &now))
1497 sync_it = S_MTIME;
1da177e4 1498
ce06e0b2
AK
1499 if (!timespec_equal(&inode->i_ctime, &now))
1500 sync_it |= S_CTIME;
870f4817 1501
ce06e0b2
AK
1502 if (IS_I_VERSION(inode))
1503 sync_it |= S_VERSION;
7a224228 1504
ce06e0b2
AK
1505 if (!sync_it)
1506 return;
1507
1508 /* Finally allowed to write? Takes lock. */
1509 if (mnt_want_write_file(file))
1510 return;
1511
1512 /* Only change inode inside the lock region */
1513 if (sync_it & S_VERSION)
1514 inode_inc_iversion(inode);
1515 if (sync_it & S_CTIME)
1516 inode->i_ctime = now;
1517 if (sync_it & S_MTIME)
1518 inode->i_mtime = now;
1519 mark_inode_dirty_sync(inode);
20ddee2c 1520 mnt_drop_write(file->f_path.mnt);
1da177e4 1521}
870f4817 1522EXPORT_SYMBOL(file_update_time);
1da177e4
LT
1523
1524int inode_needs_sync(struct inode *inode)
1525{
1526 if (IS_SYNC(inode))
1527 return 1;
1528 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1529 return 1;
1530 return 0;
1531}
1da177e4
LT
1532EXPORT_SYMBOL(inode_needs_sync);
1533
1da177e4
LT
1534int inode_wait(void *word)
1535{
1536 schedule();
1537 return 0;
1538}
d44dab8d 1539EXPORT_SYMBOL(inode_wait);
1da177e4
LT
1540
1541/*
168a9fd6
MS
1542 * If we try to find an inode in the inode hash while it is being
1543 * deleted, we have to wait until the filesystem completes its
1544 * deletion before reporting that it isn't found. This function waits
1545 * until the deletion _might_ have completed. Callers are responsible
1546 * to recheck inode state.
1547 *
eaff8079 1548 * It doesn't matter if I_NEW is not set initially, a call to
250df6ed
DC
1549 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1550 * will DTRT.
1da177e4
LT
1551 */
1552static void __wait_on_freeing_inode(struct inode *inode)
1553{
1554 wait_queue_head_t *wq;
eaff8079
CH
1555 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1556 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1da177e4 1557 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
250df6ed 1558 spin_unlock(&inode->i_lock);
67a23c49 1559 spin_unlock(&inode_hash_lock);
1da177e4
LT
1560 schedule();
1561 finish_wait(wq, &wait.wait);
67a23c49 1562 spin_lock(&inode_hash_lock);
1da177e4
LT
1563}
1564
1da177e4
LT
1565static __initdata unsigned long ihash_entries;
1566static int __init set_ihash_entries(char *str)
1567{
1568 if (!str)
1569 return 0;
1570 ihash_entries = simple_strtoul(str, &str, 0);
1571 return 1;
1572}
1573__setup("ihash_entries=", set_ihash_entries);
1574
1575/*
1576 * Initialize the waitqueues and inode hash table.
1577 */
1578void __init inode_init_early(void)
1579{
1580 int loop;
1581
1582 /* If hashes are distributed across NUMA nodes, defer
1583 * hash allocation until vmalloc space is available.
1584 */
1585 if (hashdist)
1586 return;
1587
1588 inode_hashtable =
1589 alloc_large_system_hash("Inode-cache",
1590 sizeof(struct hlist_head),
1591 ihash_entries,
1592 14,
1593 HASH_EARLY,
1594 &i_hash_shift,
1595 &i_hash_mask,
1596 0);
1597
1598 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1599 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1600}
1601
74bf17cf 1602void __init inode_init(void)
1da177e4
LT
1603{
1604 int loop;
1605
1606 /* inode slab cache */
b0196009
PJ
1607 inode_cachep = kmem_cache_create("inode_cache",
1608 sizeof(struct inode),
1609 0,
1610 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1611 SLAB_MEM_SPREAD),
20c2df83 1612 init_once);
8e1f936b 1613 register_shrinker(&icache_shrinker);
1da177e4
LT
1614
1615 /* Hash may have been set up in inode_init_early */
1616 if (!hashdist)
1617 return;
1618
1619 inode_hashtable =
1620 alloc_large_system_hash("Inode-cache",
1621 sizeof(struct hlist_head),
1622 ihash_entries,
1623 14,
1624 0,
1625 &i_hash_shift,
1626 &i_hash_mask,
1627 0);
1628
1629 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1630 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1631}
1632
1633void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1634{
1635 inode->i_mode = mode;
1636 if (S_ISCHR(mode)) {
1637 inode->i_fop = &def_chr_fops;
1638 inode->i_rdev = rdev;
1639 } else if (S_ISBLK(mode)) {
1640 inode->i_fop = &def_blk_fops;
1641 inode->i_rdev = rdev;
1642 } else if (S_ISFIFO(mode))
1643 inode->i_fop = &def_fifo_fops;
1644 else if (S_ISSOCK(mode))
1645 inode->i_fop = &bad_sock_fops;
1646 else
af0d9ae8
MK
1647 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1648 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1649 inode->i_ino);
1da177e4
LT
1650}
1651EXPORT_SYMBOL(init_special_inode);
a1bd120d
DM
1652
1653/**
eaae668d 1654 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
a1bd120d
DM
1655 * @inode: New inode
1656 * @dir: Directory inode
1657 * @mode: mode of the new inode
1658 */
1659void inode_init_owner(struct inode *inode, const struct inode *dir,
1660 mode_t mode)
1661{
1662 inode->i_uid = current_fsuid();
1663 if (dir && dir->i_mode & S_ISGID) {
1664 inode->i_gid = dir->i_gid;
1665 if (S_ISDIR(mode))
1666 mode |= S_ISGID;
1667 } else
1668 inode->i_gid = current_fsgid();
1669 inode->i_mode = mode;
1670}
1671EXPORT_SYMBOL(inode_init_owner);
e795b717 1672
2e149670
SH
1673/**
1674 * inode_owner_or_capable - check current task permissions to inode
1675 * @inode: inode being checked
1676 *
1677 * Return true if current either has CAP_FOWNER to the inode, or
1678 * owns the file.
e795b717 1679 */
2e149670 1680bool inode_owner_or_capable(const struct inode *inode)
e795b717
SH
1681{
1682 struct user_namespace *ns = inode_userns(inode);
1683
1684 if (current_user_ns() == ns && current_fsuid() == inode->i_uid)
1685 return true;
1686 if (ns_capable(ns, CAP_FOWNER))
1687 return true;
1688 return false;
1689}
2e149670 1690EXPORT_SYMBOL(inode_owner_or_capable);