]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/inode.c
fs: switch bdev inode bdi's correctly
[net-next-2.6.git] / fs / inode.c
CommitLineData
1da177e4
LT
1/*
2 * linux/fs/inode.c
3 *
4 * (C) 1997 Linus Torvalds
5 */
6
1da177e4
LT
7#include <linux/fs.h>
8#include <linux/mm.h>
9#include <linux/dcache.h>
10#include <linux/init.h>
1da177e4
LT
11#include <linux/slab.h>
12#include <linux/writeback.h>
13#include <linux/module.h>
14#include <linux/backing-dev.h>
15#include <linux/wait.h>
88e0fbc4 16#include <linux/rwsem.h>
1da177e4
LT
17#include <linux/hash.h>
18#include <linux/swap.h>
19#include <linux/security.h>
20#include <linux/pagemap.h>
21#include <linux/cdev.h>
22#include <linux/bootmem.h>
3be25f49 23#include <linux/fsnotify.h>
fc33a7bb 24#include <linux/mount.h>
efaee192 25#include <linux/async.h>
f19d4a8f 26#include <linux/posix_acl.h>
1da177e4
LT
27
28/*
29 * This is needed for the following functions:
30 * - inode_has_buffers
1da177e4
LT
31 * - invalidate_bdev
32 *
33 * FIXME: remove all knowledge of the buffer layer from this file
34 */
35#include <linux/buffer_head.h>
36
37/*
38 * New inode.c implementation.
39 *
40 * This implementation has the basic premise of trying
41 * to be extremely low-overhead and SMP-safe, yet be
42 * simple enough to be "obviously correct".
43 *
44 * Famous last words.
45 */
46
47/* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
48
49/* #define INODE_PARANOIA 1 */
50/* #define INODE_DEBUG 1 */
51
52/*
53 * Inode lookup is no longer as critical as it used to be:
54 * most of the lookups are going to be through the dcache.
55 */
56#define I_HASHBITS i_hash_shift
57#define I_HASHMASK i_hash_mask
58
fa3536cc
ED
59static unsigned int i_hash_mask __read_mostly;
60static unsigned int i_hash_shift __read_mostly;
1da177e4
LT
61
62/*
63 * Each inode can be on two separate lists. One is
64 * the hash list of the inode, used for lookups. The
65 * other linked list is the "type" list:
66 * "in_use" - valid inode, i_count > 0, i_nlink > 0
67 * "dirty" - as "in_use" but also dirty
68 * "unused" - valid inode, i_count = 0
69 *
70 * A "dirty" list is maintained for each super block,
71 * allowing for low-overhead inode sync() operations.
72 */
73
9e38d86f 74static LIST_HEAD(inode_unused);
fa3536cc 75static struct hlist_head *inode_hashtable __read_mostly;
1da177e4
LT
76
77/*
78 * A simple spinlock to protect the list manipulations.
79 *
80 * NOTE! You also have to own the lock if you change
81 * the i_state of an inode while it is in use..
82 */
83DEFINE_SPINLOCK(inode_lock);
84
85/*
88e0fbc4 86 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
1da177e4
LT
87 * icache shrinking path, and the umount path. Without this exclusion,
88 * by the time prune_icache calls iput for the inode whose pages it has
89 * been invalidating, or by the time it calls clear_inode & destroy_inode
90 * from its final dispose_list, the struct super_block they refer to
91 * (for inode->i_sb->s_op) may already have been freed and reused.
88e0fbc4
NP
92 *
93 * We make this an rwsem because the fastpath is icache shrinking. In
94 * some cases a filesystem may be doing a significant amount of work in
95 * its inode reclaim code, so this should improve parallelism.
1da177e4 96 */
88e0fbc4 97static DECLARE_RWSEM(iprune_sem);
1da177e4
LT
98
99/*
100 * Statistics gathering..
101 */
102struct inodes_stat_t inodes_stat;
103
cffbc8aa
DC
104static struct percpu_counter nr_inodes __cacheline_aligned_in_smp;
105static struct percpu_counter nr_inodes_unused __cacheline_aligned_in_smp;
106
6b3304b5 107static struct kmem_cache *inode_cachep __read_mostly;
1da177e4 108
cffbc8aa
DC
109static inline int get_nr_inodes(void)
110{
111 return percpu_counter_sum_positive(&nr_inodes);
112}
113
114static inline int get_nr_inodes_unused(void)
115{
116 return percpu_counter_sum_positive(&nr_inodes_unused);
117}
118
119int get_nr_dirty_inodes(void)
120{
121 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
122 return nr_dirty > 0 ? nr_dirty : 0;
123
124}
125
126/*
127 * Handle nr_inode sysctl
128 */
129#ifdef CONFIG_SYSCTL
130int proc_nr_inodes(ctl_table *table, int write,
131 void __user *buffer, size_t *lenp, loff_t *ppos)
132{
133 inodes_stat.nr_inodes = get_nr_inodes();
134 inodes_stat.nr_unused = get_nr_inodes_unused();
135 return proc_dointvec(table, write, buffer, lenp, ppos);
136}
137#endif
138
1c0eeaf5
JE
139static void wake_up_inode(struct inode *inode)
140{
141 /*
142 * Prevent speculative execution through spin_unlock(&inode_lock);
143 */
144 smp_mb();
eaff8079 145 wake_up_bit(&inode->i_state, __I_NEW);
1c0eeaf5
JE
146}
147
2cb1599f
DC
148/**
149 * inode_init_always - perform inode structure intialisation
0bc02f3f
RD
150 * @sb: superblock inode belongs to
151 * @inode: inode to initialise
2cb1599f
DC
152 *
153 * These are initializations that need to be done on every inode
154 * allocation as the fields are not initialised by slab allocation.
155 */
54e34621 156int inode_init_always(struct super_block *sb, struct inode *inode)
1da177e4 157{
f5e54d6e 158 static const struct address_space_operations empty_aops;
6e1d5dcc 159 static const struct inode_operations empty_iops;
99ac48f5 160 static const struct file_operations empty_fops;
6b3304b5 161 struct address_space *const mapping = &inode->i_data;
2cb1599f
DC
162
163 inode->i_sb = sb;
164 inode->i_blkbits = sb->s_blocksize_bits;
165 inode->i_flags = 0;
166 atomic_set(&inode->i_count, 1);
167 inode->i_op = &empty_iops;
168 inode->i_fop = &empty_fops;
169 inode->i_nlink = 1;
56ff5efa
AV
170 inode->i_uid = 0;
171 inode->i_gid = 0;
2cb1599f
DC
172 atomic_set(&inode->i_writecount, 0);
173 inode->i_size = 0;
174 inode->i_blocks = 0;
175 inode->i_bytes = 0;
176 inode->i_generation = 0;
1da177e4 177#ifdef CONFIG_QUOTA
2cb1599f 178 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
1da177e4 179#endif
2cb1599f
DC
180 inode->i_pipe = NULL;
181 inode->i_bdev = NULL;
182 inode->i_cdev = NULL;
183 inode->i_rdev = 0;
184 inode->dirtied_when = 0;
6146f0d5
MZ
185
186 if (security_inode_alloc(inode))
54e34621 187 goto out;
2cb1599f
DC
188 spin_lock_init(&inode->i_lock);
189 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
190
191 mutex_init(&inode->i_mutex);
192 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
193
194 init_rwsem(&inode->i_alloc_sem);
195 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
196
197 mapping->a_ops = &empty_aops;
198 mapping->host = inode;
199 mapping->flags = 0;
3c1d4378 200 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
2cb1599f
DC
201 mapping->assoc_mapping = NULL;
202 mapping->backing_dev_info = &default_backing_dev_info;
203 mapping->writeback_index = 0;
204
205 /*
206 * If the block_device provides a backing_dev_info for client
207 * inodes then use that. Otherwise the inode share the bdev's
208 * backing_dev_info.
209 */
210 if (sb->s_bdev) {
211 struct backing_dev_info *bdi;
212
2c96ce9f 213 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
2cb1599f
DC
214 mapping->backing_dev_info = bdi;
215 }
216 inode->i_private = NULL;
217 inode->i_mapping = mapping;
f19d4a8f
AV
218#ifdef CONFIG_FS_POSIX_ACL
219 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
220#endif
2cb1599f 221
3be25f49
EP
222#ifdef CONFIG_FSNOTIFY
223 inode->i_fsnotify_mask = 0;
224#endif
225
cffbc8aa
DC
226 percpu_counter_inc(&nr_inodes);
227
54e34621 228 return 0;
54e34621
CH
229out:
230 return -ENOMEM;
1da177e4 231}
2cb1599f
DC
232EXPORT_SYMBOL(inode_init_always);
233
234static struct inode *alloc_inode(struct super_block *sb)
235{
236 struct inode *inode;
237
238 if (sb->s_op->alloc_inode)
239 inode = sb->s_op->alloc_inode(sb);
240 else
241 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
242
54e34621
CH
243 if (!inode)
244 return NULL;
245
246 if (unlikely(inode_init_always(sb, inode))) {
247 if (inode->i_sb->s_op->destroy_inode)
248 inode->i_sb->s_op->destroy_inode(inode);
249 else
250 kmem_cache_free(inode_cachep, inode);
251 return NULL;
252 }
253
254 return inode;
2cb1599f 255}
1da177e4 256
2e00c97e 257void __destroy_inode(struct inode *inode)
1da177e4 258{
b7542f8c 259 BUG_ON(inode_has_buffers(inode));
1da177e4 260 security_inode_free(inode);
3be25f49 261 fsnotify_inode_delete(inode);
f19d4a8f
AV
262#ifdef CONFIG_FS_POSIX_ACL
263 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
264 posix_acl_release(inode->i_acl);
265 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
266 posix_acl_release(inode->i_default_acl);
267#endif
cffbc8aa 268 percpu_counter_dec(&nr_inodes);
2e00c97e
CH
269}
270EXPORT_SYMBOL(__destroy_inode);
271
56b0dacf 272static void destroy_inode(struct inode *inode)
2e00c97e
CH
273{
274 __destroy_inode(inode);
1da177e4
LT
275 if (inode->i_sb->s_op->destroy_inode)
276 inode->i_sb->s_op->destroy_inode(inode);
277 else
278 kmem_cache_free(inode_cachep, (inode));
279}
1da177e4
LT
280
281/*
282 * These are initializations that only need to be done
283 * once, because the fields are idempotent across use
284 * of the inode, so let the slab aware of that.
285 */
286void inode_init_once(struct inode *inode)
287{
288 memset(inode, 0, sizeof(*inode));
289 INIT_HLIST_NODE(&inode->i_hash);
290 INIT_LIST_HEAD(&inode->i_dentry);
291 INIT_LIST_HEAD(&inode->i_devices);
9e38d86f 292 INIT_LIST_HEAD(&inode->i_list);
1da177e4 293 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
19fd6231 294 spin_lock_init(&inode->i_data.tree_lock);
1da177e4
LT
295 spin_lock_init(&inode->i_data.i_mmap_lock);
296 INIT_LIST_HEAD(&inode->i_data.private_list);
297 spin_lock_init(&inode->i_data.private_lock);
298 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
299 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
1da177e4 300 i_size_ordered_init(inode);
3be25f49 301#ifdef CONFIG_FSNOTIFY
e61ce867 302 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
3be25f49 303#endif
1da177e4 304}
1da177e4
LT
305EXPORT_SYMBOL(inode_init_once);
306
51cc5068 307static void init_once(void *foo)
1da177e4 308{
6b3304b5 309 struct inode *inode = (struct inode *) foo;
1da177e4 310
a35afb83 311 inode_init_once(inode);
1da177e4
LT
312}
313
314/*
315 * inode_lock must be held
316 */
6b3304b5 317void __iget(struct inode *inode)
1da177e4 318{
9e38d86f
NP
319 atomic_inc(&inode->i_count);
320}
2e147f1e 321
7de9c6ee
AV
322/*
323 * get additional reference to inode; caller must already hold one.
324 */
325void ihold(struct inode *inode)
326{
327 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
328}
329EXPORT_SYMBOL(ihold);
330
9e38d86f
NP
331static void inode_lru_list_add(struct inode *inode)
332{
333 if (list_empty(&inode->i_list)) {
334 list_add(&inode->i_list, &inode_unused);
335 percpu_counter_inc(&nr_inodes_unused);
336 }
337}
338
339static void inode_lru_list_del(struct inode *inode)
340{
341 if (!list_empty(&inode->i_list)) {
342 list_del_init(&inode->i_list);
343 percpu_counter_dec(&nr_inodes_unused);
344 }
1da177e4
LT
345}
346
646ec461
CH
347static inline void __inode_sb_list_add(struct inode *inode)
348{
349 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
350}
351
352/**
353 * inode_sb_list_add - add inode to the superblock list of inodes
354 * @inode: inode to add
355 */
356void inode_sb_list_add(struct inode *inode)
357{
358 spin_lock(&inode_lock);
359 __inode_sb_list_add(inode);
360 spin_unlock(&inode_lock);
361}
362EXPORT_SYMBOL_GPL(inode_sb_list_add);
363
364static inline void __inode_sb_list_del(struct inode *inode)
365{
366 list_del_init(&inode->i_sb_list);
367}
368
4c51acbc
DC
369static unsigned long hash(struct super_block *sb, unsigned long hashval)
370{
371 unsigned long tmp;
372
373 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
374 L1_CACHE_BYTES;
375 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
376 return tmp & I_HASHMASK;
377}
378
379/**
380 * __insert_inode_hash - hash an inode
381 * @inode: unhashed inode
382 * @hashval: unsigned long value used to locate this object in the
383 * inode_hashtable.
384 *
385 * Add an inode to the inode hash for this superblock.
386 */
387void __insert_inode_hash(struct inode *inode, unsigned long hashval)
388{
646ec461
CH
389 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
390
4c51acbc 391 spin_lock(&inode_lock);
646ec461 392 hlist_add_head(&inode->i_hash, b);
4c51acbc
DC
393 spin_unlock(&inode_lock);
394}
395EXPORT_SYMBOL(__insert_inode_hash);
396
397/**
398 * __remove_inode_hash - remove an inode from the hash
399 * @inode: inode to unhash
400 *
401 * Remove an inode from the superblock.
402 */
403static void __remove_inode_hash(struct inode *inode)
404{
405 hlist_del_init(&inode->i_hash);
406}
407
408/**
409 * remove_inode_hash - remove an inode from the hash
410 * @inode: inode to unhash
411 *
412 * Remove an inode from the superblock.
413 */
414void remove_inode_hash(struct inode *inode)
415{
416 spin_lock(&inode_lock);
417 hlist_del_init(&inode->i_hash);
418 spin_unlock(&inode_lock);
419}
420EXPORT_SYMBOL(remove_inode_hash);
421
b0683aa6
AV
422void end_writeback(struct inode *inode)
423{
424 might_sleep();
425 BUG_ON(inode->i_data.nrpages);
426 BUG_ON(!list_empty(&inode->i_data.private_list));
427 BUG_ON(!(inode->i_state & I_FREEING));
428 BUG_ON(inode->i_state & I_CLEAR);
429 inode_sync_wait(inode);
430 inode->i_state = I_FREEING | I_CLEAR;
431}
432EXPORT_SYMBOL(end_writeback);
433
644da596 434static void evict(struct inode *inode)
b4272d4c
AV
435{
436 const struct super_operations *op = inode->i_sb->s_op;
437
be7ce416
AV
438 if (op->evict_inode) {
439 op->evict_inode(inode);
b4272d4c
AV
440 } else {
441 if (inode->i_data.nrpages)
442 truncate_inode_pages(&inode->i_data, 0);
30140837 443 end_writeback(inode);
b4272d4c 444 }
661074e9
AV
445 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
446 bd_forget(inode);
447 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
448 cd_forget(inode);
b4272d4c
AV
449}
450
1da177e4
LT
451/*
452 * dispose_list - dispose of the contents of a local list
453 * @head: the head of the list to free
454 *
455 * Dispose-list gets a local list with local inodes in it, so it doesn't
456 * need to worry about list corruption and SMP locks.
457 */
458static void dispose_list(struct list_head *head)
459{
1da177e4
LT
460 while (!list_empty(head)) {
461 struct inode *inode;
462
b5e61818 463 inode = list_first_entry(head, struct inode, i_list);
9e38d86f 464 list_del_init(&inode->i_list);
1da177e4 465
644da596 466 evict(inode);
4120db47
AB
467
468 spin_lock(&inode_lock);
4c51acbc 469 __remove_inode_hash(inode);
646ec461 470 __inode_sb_list_del(inode);
4120db47
AB
471 spin_unlock(&inode_lock);
472
473 wake_up_inode(inode);
1da177e4 474 destroy_inode(inode);
1da177e4 475 }
1da177e4
LT
476}
477
478/*
479 * Invalidate all inodes for a device.
480 */
481static int invalidate_list(struct list_head *head, struct list_head *dispose)
482{
483 struct list_head *next;
cffbc8aa 484 int busy = 0;
1da177e4
LT
485
486 next = head->next;
487 for (;;) {
6b3304b5
MK
488 struct list_head *tmp = next;
489 struct inode *inode;
1da177e4
LT
490
491 /*
492 * We can reschedule here without worrying about the list's
493 * consistency because the per-sb list of inodes must not
88e0fbc4 494 * change during umount anymore, and because iprune_sem keeps
1da177e4
LT
495 * shrink_icache_memory() away.
496 */
497 cond_resched_lock(&inode_lock);
498
499 next = next->next;
500 if (tmp == head)
501 break;
502 inode = list_entry(tmp, struct inode, i_sb_list);
aabb8fdb
NP
503 if (inode->i_state & I_NEW)
504 continue;
99a38919
CH
505 if (atomic_read(&inode->i_count)) {
506 busy = 1;
1da177e4
LT
507 continue;
508 }
99a38919
CH
509
510 list_move(&inode->i_list, dispose);
511 inode->i_state |= I_FREEING;
512 if (!(inode->i_state & (I_DIRTY | I_SYNC)))
513 percpu_counter_dec(&nr_inodes_unused);
1da177e4 514 }
1da177e4
LT
515 return busy;
516}
517
1da177e4
LT
518/**
519 * invalidate_inodes - discard the inodes on a device
520 * @sb: superblock
521 *
522 * Discard all of the inodes for a given superblock. If the discard
523 * fails because there are busy inodes then a non zero value is returned.
524 * If the discard is successful all the inodes have been discarded.
525 */
6b3304b5 526int invalidate_inodes(struct super_block *sb)
1da177e4
LT
527{
528 int busy;
529 LIST_HEAD(throw_away);
530
88e0fbc4 531 down_write(&iprune_sem);
1da177e4 532 spin_lock(&inode_lock);
164bc619 533 fsnotify_unmount_inodes(&sb->s_inodes);
1da177e4
LT
534 busy = invalidate_list(&sb->s_inodes, &throw_away);
535 spin_unlock(&inode_lock);
536
537 dispose_list(&throw_away);
88e0fbc4 538 up_write(&iprune_sem);
1da177e4
LT
539
540 return busy;
541}
1da177e4
LT
542
543static int can_unuse(struct inode *inode)
544{
9e38d86f 545 if (inode->i_state & ~I_REFERENCED)
1da177e4
LT
546 return 0;
547 if (inode_has_buffers(inode))
548 return 0;
549 if (atomic_read(&inode->i_count))
550 return 0;
551 if (inode->i_data.nrpages)
552 return 0;
553 return 1;
554}
555
556/*
9e38d86f
NP
557 * Scan `goal' inodes on the unused list for freeable ones. They are moved to a
558 * temporary list and then are freed outside inode_lock by dispose_list().
1da177e4
LT
559 *
560 * Any inodes which are pinned purely because of attached pagecache have their
9e38d86f
NP
561 * pagecache removed. If the inode has metadata buffers attached to
562 * mapping->private_list then try to remove them.
1da177e4 563 *
9e38d86f
NP
564 * If the inode has the I_REFERENCED flag set, then it means that it has been
565 * used recently - the flag is set in iput_final(). When we encounter such an
566 * inode, clear the flag and move it to the back of the LRU so it gets another
567 * pass through the LRU before it gets reclaimed. This is necessary because of
568 * the fact we are doing lazy LRU updates to minimise lock contention so the
569 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
570 * with this flag set because they are the inodes that are out of order.
1da177e4
LT
571 */
572static void prune_icache(int nr_to_scan)
573{
574 LIST_HEAD(freeable);
1da177e4
LT
575 int nr_scanned;
576 unsigned long reap = 0;
577
88e0fbc4 578 down_read(&iprune_sem);
1da177e4
LT
579 spin_lock(&inode_lock);
580 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
581 struct inode *inode;
582
583 if (list_empty(&inode_unused))
584 break;
585
586 inode = list_entry(inode_unused.prev, struct inode, i_list);
587
9e38d86f
NP
588 /*
589 * Referenced or dirty inodes are still in use. Give them
590 * another pass through the LRU as we canot reclaim them now.
591 */
592 if (atomic_read(&inode->i_count) ||
593 (inode->i_state & ~I_REFERENCED)) {
594 list_del_init(&inode->i_list);
595 percpu_counter_dec(&nr_inodes_unused);
596 continue;
597 }
598
599 /* recently referenced inodes get one more pass */
600 if (inode->i_state & I_REFERENCED) {
1da177e4 601 list_move(&inode->i_list, &inode_unused);
9e38d86f 602 inode->i_state &= ~I_REFERENCED;
1da177e4
LT
603 continue;
604 }
605 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
606 __iget(inode);
607 spin_unlock(&inode_lock);
608 if (remove_inode_buffers(inode))
fc0ecff6
AM
609 reap += invalidate_mapping_pages(&inode->i_data,
610 0, -1);
1da177e4
LT
611 iput(inode);
612 spin_lock(&inode_lock);
613
614 if (inode != list_entry(inode_unused.next,
615 struct inode, i_list))
616 continue; /* wrong inode or list_empty */
617 if (!can_unuse(inode))
618 continue;
619 }
1da177e4 620 list_move(&inode->i_list, &freeable);
7ef0d737 621 WARN_ON(inode->i_state & I_NEW);
1da177e4 622 inode->i_state |= I_FREEING;
cffbc8aa 623 percpu_counter_dec(&nr_inodes_unused);
1da177e4 624 }
f8891e5e
CL
625 if (current_is_kswapd())
626 __count_vm_events(KSWAPD_INODESTEAL, reap);
627 else
628 __count_vm_events(PGINODESTEAL, reap);
1da177e4
LT
629 spin_unlock(&inode_lock);
630
631 dispose_list(&freeable);
88e0fbc4 632 up_read(&iprune_sem);
1da177e4
LT
633}
634
635/*
636 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
637 * "unused" means that no dentries are referring to the inodes: the files are
638 * not open and the dcache references to those inodes have already been
639 * reclaimed.
640 *
641 * This function is passed the number of inodes to scan, and it returns the
642 * total number of remaining possibly-reclaimable inodes.
643 */
7f8275d0 644static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
1da177e4
LT
645{
646 if (nr) {
647 /*
648 * Nasty deadlock avoidance. We may hold various FS locks,
649 * and we don't want to recurse into the FS that called us
650 * in clear_inode() and friends..
6b3304b5 651 */
1da177e4
LT
652 if (!(gfp_mask & __GFP_FS))
653 return -1;
654 prune_icache(nr);
655 }
cffbc8aa 656 return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure;
1da177e4
LT
657}
658
8e1f936b
RR
659static struct shrinker icache_shrinker = {
660 .shrink = shrink_icache_memory,
661 .seeks = DEFAULT_SEEKS,
662};
663
1da177e4
LT
664static void __wait_on_freeing_inode(struct inode *inode);
665/*
666 * Called with the inode lock held.
1da177e4 667 */
6b3304b5
MK
668static struct inode *find_inode(struct super_block *sb,
669 struct hlist_head *head,
670 int (*test)(struct inode *, void *),
671 void *data)
1da177e4
LT
672{
673 struct hlist_node *node;
6b3304b5 674 struct inode *inode = NULL;
1da177e4
LT
675
676repeat:
c5c8be3c 677 hlist_for_each_entry(inode, node, head, i_hash) {
1da177e4
LT
678 if (inode->i_sb != sb)
679 continue;
680 if (!test(inode, data))
681 continue;
a4ffdde6 682 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1da177e4
LT
683 __wait_on_freeing_inode(inode);
684 goto repeat;
685 }
f7899bd5
CH
686 __iget(inode);
687 return inode;
1da177e4 688 }
f7899bd5 689 return NULL;
1da177e4
LT
690}
691
692/*
693 * find_inode_fast is the fast path version of find_inode, see the comment at
694 * iget_locked for details.
695 */
6b3304b5
MK
696static struct inode *find_inode_fast(struct super_block *sb,
697 struct hlist_head *head, unsigned long ino)
1da177e4
LT
698{
699 struct hlist_node *node;
6b3304b5 700 struct inode *inode = NULL;
1da177e4
LT
701
702repeat:
c5c8be3c 703 hlist_for_each_entry(inode, node, head, i_hash) {
1da177e4
LT
704 if (inode->i_ino != ino)
705 continue;
706 if (inode->i_sb != sb)
707 continue;
a4ffdde6 708 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1da177e4
LT
709 __wait_on_freeing_inode(inode);
710 goto repeat;
711 }
f7899bd5
CH
712 __iget(inode);
713 return inode;
1da177e4 714 }
f7899bd5 715 return NULL;
1da177e4
LT
716}
717
f991bd2e
ED
718/*
719 * Each cpu owns a range of LAST_INO_BATCH numbers.
720 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
721 * to renew the exhausted range.
722 *
723 * This does not significantly increase overflow rate because every CPU can
724 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
725 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
726 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
727 * overflow rate by 2x, which does not seem too significant.
728 *
729 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
730 * error if st_ino won't fit in target struct field. Use 32bit counter
731 * here to attempt to avoid that.
732 */
733#define LAST_INO_BATCH 1024
734static DEFINE_PER_CPU(unsigned int, last_ino);
735
85fe4025 736unsigned int get_next_ino(void)
f991bd2e
ED
737{
738 unsigned int *p = &get_cpu_var(last_ino);
739 unsigned int res = *p;
740
741#ifdef CONFIG_SMP
742 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
743 static atomic_t shared_last_ino;
744 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
745
746 res = next - LAST_INO_BATCH;
747 }
748#endif
749
750 *p = ++res;
751 put_cpu_var(last_ino);
752 return res;
753}
85fe4025 754EXPORT_SYMBOL(get_next_ino);
f991bd2e 755
1da177e4
LT
756/**
757 * new_inode - obtain an inode
758 * @sb: superblock
759 *
769848c0 760 * Allocates a new inode for given superblock. The default gfp_mask
3c1d4378 761 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
769848c0
MG
762 * If HIGHMEM pages are unsuitable or it is known that pages allocated
763 * for the page cache are not reclaimable or migratable,
764 * mapping_set_gfp_mask() must be called with suitable flags on the
765 * newly created inode's mapping
766 *
1da177e4
LT
767 */
768struct inode *new_inode(struct super_block *sb)
769{
6b3304b5 770 struct inode *inode;
1da177e4
LT
771
772 spin_lock_prefetch(&inode_lock);
6b3304b5 773
1da177e4
LT
774 inode = alloc_inode(sb);
775 if (inode) {
776 spin_lock(&inode_lock);
646ec461 777 __inode_sb_list_add(inode);
1da177e4
LT
778 inode->i_state = 0;
779 spin_unlock(&inode_lock);
780 }
781 return inode;
782}
1da177e4
LT
783EXPORT_SYMBOL(new_inode);
784
785void unlock_new_inode(struct inode *inode)
786{
14358e6d 787#ifdef CONFIG_DEBUG_LOCK_ALLOC
a3314a0e 788 if (S_ISDIR(inode->i_mode)) {
1e89a5e1
PZ
789 struct file_system_type *type = inode->i_sb->s_type;
790
9a7aa12f
JK
791 /* Set new key only if filesystem hasn't already changed it */
792 if (!lockdep_match_class(&inode->i_mutex,
793 &type->i_mutex_key)) {
794 /*
795 * ensure nobody is actually holding i_mutex
796 */
797 mutex_destroy(&inode->i_mutex);
798 mutex_init(&inode->i_mutex);
799 lockdep_set_class(&inode->i_mutex,
800 &type->i_mutex_dir_key);
801 }
1e89a5e1 802 }
14358e6d 803#endif
1da177e4 804 /*
eaff8079 805 * This is special! We do not need the spinlock when clearing I_NEW,
580be083
JK
806 * because we're guaranteed that nobody else tries to do anything about
807 * the state of the inode when it is locked, as we just created it (so
eaff8079 808 * there can be no old holders that haven't tested I_NEW).
580be083 809 * However we must emit the memory barrier so that other CPUs reliably
eaff8079 810 * see the clearing of I_NEW after the other inode initialisation has
580be083 811 * completed.
1da177e4 812 */
580be083 813 smp_mb();
eaff8079
CH
814 WARN_ON(!(inode->i_state & I_NEW));
815 inode->i_state &= ~I_NEW;
1da177e4
LT
816 wake_up_inode(inode);
817}
1da177e4
LT
818EXPORT_SYMBOL(unlock_new_inode);
819
820/*
821 * This is called without the inode lock held.. Be careful.
822 *
823 * We no longer cache the sb_flags in i_flags - see fs.h
824 * -- rmk@arm.uk.linux.org
825 */
6b3304b5
MK
826static struct inode *get_new_inode(struct super_block *sb,
827 struct hlist_head *head,
828 int (*test)(struct inode *, void *),
829 int (*set)(struct inode *, void *),
830 void *data)
1da177e4 831{
6b3304b5 832 struct inode *inode;
1da177e4
LT
833
834 inode = alloc_inode(sb);
835 if (inode) {
6b3304b5 836 struct inode *old;
1da177e4
LT
837
838 spin_lock(&inode_lock);
839 /* We released the lock, so.. */
840 old = find_inode(sb, head, test, data);
841 if (!old) {
842 if (set(inode, data))
843 goto set_failed;
844
646ec461
CH
845 hlist_add_head(&inode->i_hash, head);
846 __inode_sb_list_add(inode);
eaff8079 847 inode->i_state = I_NEW;
1da177e4
LT
848 spin_unlock(&inode_lock);
849
850 /* Return the locked inode with I_NEW set, the
851 * caller is responsible for filling in the contents
852 */
853 return inode;
854 }
855
856 /*
857 * Uhhuh, somebody else created the same inode under
858 * us. Use the old inode instead of the one we just
859 * allocated.
860 */
1da177e4
LT
861 spin_unlock(&inode_lock);
862 destroy_inode(inode);
863 inode = old;
864 wait_on_inode(inode);
865 }
866 return inode;
867
868set_failed:
869 spin_unlock(&inode_lock);
870 destroy_inode(inode);
871 return NULL;
872}
873
874/*
875 * get_new_inode_fast is the fast path version of get_new_inode, see the
876 * comment at iget_locked for details.
877 */
6b3304b5
MK
878static struct inode *get_new_inode_fast(struct super_block *sb,
879 struct hlist_head *head, unsigned long ino)
1da177e4 880{
6b3304b5 881 struct inode *inode;
1da177e4
LT
882
883 inode = alloc_inode(sb);
884 if (inode) {
6b3304b5 885 struct inode *old;
1da177e4
LT
886
887 spin_lock(&inode_lock);
888 /* We released the lock, so.. */
889 old = find_inode_fast(sb, head, ino);
890 if (!old) {
891 inode->i_ino = ino;
646ec461
CH
892 hlist_add_head(&inode->i_hash, head);
893 __inode_sb_list_add(inode);
eaff8079 894 inode->i_state = I_NEW;
1da177e4
LT
895 spin_unlock(&inode_lock);
896
897 /* Return the locked inode with I_NEW set, the
898 * caller is responsible for filling in the contents
899 */
900 return inode;
901 }
902
903 /*
904 * Uhhuh, somebody else created the same inode under
905 * us. Use the old inode instead of the one we just
906 * allocated.
907 */
1da177e4
LT
908 spin_unlock(&inode_lock);
909 destroy_inode(inode);
910 inode = old;
911 wait_on_inode(inode);
912 }
913 return inode;
914}
915
ad5e195a
CH
916/*
917 * search the inode cache for a matching inode number.
918 * If we find one, then the inode number we are trying to
919 * allocate is not unique and so we should not use it.
920 *
921 * Returns 1 if the inode number is unique, 0 if it is not.
922 */
923static int test_inode_iunique(struct super_block *sb, unsigned long ino)
924{
925 struct hlist_head *b = inode_hashtable + hash(sb, ino);
926 struct hlist_node *node;
927 struct inode *inode;
928
929 hlist_for_each_entry(inode, node, b, i_hash) {
930 if (inode->i_ino == ino && inode->i_sb == sb)
931 return 0;
932 }
933
934 return 1;
935}
936
1da177e4
LT
937/**
938 * iunique - get a unique inode number
939 * @sb: superblock
940 * @max_reserved: highest reserved inode number
941 *
942 * Obtain an inode number that is unique on the system for a given
943 * superblock. This is used by file systems that have no natural
944 * permanent inode numbering system. An inode number is returned that
945 * is higher than the reserved limit but unique.
946 *
947 * BUGS:
948 * With a large number of inodes live on the file system this function
949 * currently becomes quite slow.
950 */
951ino_t iunique(struct super_block *sb, ino_t max_reserved)
952{
866b04fc
JL
953 /*
954 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
955 * error if st_ino won't fit in target struct field. Use 32bit counter
956 * here to attempt to avoid that.
957 */
ad5e195a 958 static DEFINE_SPINLOCK(iunique_lock);
866b04fc 959 static unsigned int counter;
1da177e4 960 ino_t res;
3361c7be 961
1da177e4 962 spin_lock(&inode_lock);
ad5e195a 963 spin_lock(&iunique_lock);
3361c7be
JL
964 do {
965 if (counter <= max_reserved)
966 counter = max_reserved + 1;
1da177e4 967 res = counter++;
ad5e195a
CH
968 } while (!test_inode_iunique(sb, res));
969 spin_unlock(&iunique_lock);
3361c7be 970 spin_unlock(&inode_lock);
1da177e4 971
3361c7be
JL
972 return res;
973}
1da177e4
LT
974EXPORT_SYMBOL(iunique);
975
976struct inode *igrab(struct inode *inode)
977{
978 spin_lock(&inode_lock);
a4ffdde6 979 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
1da177e4
LT
980 __iget(inode);
981 else
982 /*
983 * Handle the case where s_op->clear_inode is not been
984 * called yet, and somebody is calling igrab
985 * while the inode is getting freed.
986 */
987 inode = NULL;
988 spin_unlock(&inode_lock);
989 return inode;
990}
1da177e4
LT
991EXPORT_SYMBOL(igrab);
992
993/**
994 * ifind - internal function, you want ilookup5() or iget5().
995 * @sb: super block of file system to search
996 * @head: the head of the list to search
997 * @test: callback used for comparisons between inodes
998 * @data: opaque data pointer to pass to @test
88bd5121 999 * @wait: if true wait for the inode to be unlocked, if false do not
1da177e4
LT
1000 *
1001 * ifind() searches for the inode specified by @data in the inode
1002 * cache. This is a generalized version of ifind_fast() for file systems where
1003 * the inode number is not sufficient for unique identification of an inode.
1004 *
1005 * If the inode is in the cache, the inode is returned with an incremented
1006 * reference count.
1007 *
1008 * Otherwise NULL is returned.
1009 *
1010 * Note, @test is called with the inode_lock held, so can't sleep.
1011 */
5d2bea45 1012static struct inode *ifind(struct super_block *sb,
1da177e4 1013 struct hlist_head *head, int (*test)(struct inode *, void *),
88bd5121 1014 void *data, const int wait)
1da177e4
LT
1015{
1016 struct inode *inode;
1017
1018 spin_lock(&inode_lock);
1019 inode = find_inode(sb, head, test, data);
1020 if (inode) {
1da177e4 1021 spin_unlock(&inode_lock);
88bd5121
AA
1022 if (likely(wait))
1023 wait_on_inode(inode);
1da177e4
LT
1024 return inode;
1025 }
1026 spin_unlock(&inode_lock);
1027 return NULL;
1028}
1029
1030/**
1031 * ifind_fast - internal function, you want ilookup() or iget().
1032 * @sb: super block of file system to search
1033 * @head: head of the list to search
1034 * @ino: inode number to search for
1035 *
1036 * ifind_fast() searches for the inode @ino in the inode cache. This is for
1037 * file systems where the inode number is sufficient for unique identification
1038 * of an inode.
1039 *
1040 * If the inode is in the cache, the inode is returned with an incremented
1041 * reference count.
1042 *
1043 * Otherwise NULL is returned.
1044 */
5d2bea45 1045static struct inode *ifind_fast(struct super_block *sb,
1da177e4
LT
1046 struct hlist_head *head, unsigned long ino)
1047{
1048 struct inode *inode;
1049
1050 spin_lock(&inode_lock);
1051 inode = find_inode_fast(sb, head, ino);
1052 if (inode) {
1da177e4
LT
1053 spin_unlock(&inode_lock);
1054 wait_on_inode(inode);
1055 return inode;
1056 }
1057 spin_unlock(&inode_lock);
1058 return NULL;
1059}
1060
1061/**
88bd5121 1062 * ilookup5_nowait - search for an inode in the inode cache
1da177e4
LT
1063 * @sb: super block of file system to search
1064 * @hashval: hash value (usually inode number) to search for
1065 * @test: callback used for comparisons between inodes
1066 * @data: opaque data pointer to pass to @test
1067 *
1068 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1069 * @data in the inode cache. This is a generalized version of ilookup() for
1070 * file systems where the inode number is not sufficient for unique
1071 * identification of an inode.
1072 *
1073 * If the inode is in the cache, the inode is returned with an incremented
88bd5121
AA
1074 * reference count. Note, the inode lock is not waited upon so you have to be
1075 * very careful what you do with the returned inode. You probably should be
1076 * using ilookup5() instead.
1077 *
1078 * Otherwise NULL is returned.
1079 *
1080 * Note, @test is called with the inode_lock held, so can't sleep.
1081 */
1082struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1083 int (*test)(struct inode *, void *), void *data)
1084{
1085 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1086
1087 return ifind(sb, head, test, data, 0);
1088}
88bd5121
AA
1089EXPORT_SYMBOL(ilookup5_nowait);
1090
1091/**
1092 * ilookup5 - search for an inode in the inode cache
1093 * @sb: super block of file system to search
1094 * @hashval: hash value (usually inode number) to search for
1095 * @test: callback used for comparisons between inodes
1096 * @data: opaque data pointer to pass to @test
1097 *
1098 * ilookup5() uses ifind() to search for the inode specified by @hashval and
1099 * @data in the inode cache. This is a generalized version of ilookup() for
1100 * file systems where the inode number is not sufficient for unique
1101 * identification of an inode.
1102 *
1103 * If the inode is in the cache, the inode lock is waited upon and the inode is
1104 * returned with an incremented reference count.
1da177e4
LT
1105 *
1106 * Otherwise NULL is returned.
1107 *
1108 * Note, @test is called with the inode_lock held, so can't sleep.
1109 */
1110struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1111 int (*test)(struct inode *, void *), void *data)
1112{
1113 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1114
88bd5121 1115 return ifind(sb, head, test, data, 1);
1da177e4 1116}
1da177e4
LT
1117EXPORT_SYMBOL(ilookup5);
1118
1119/**
1120 * ilookup - search for an inode in the inode cache
1121 * @sb: super block of file system to search
1122 * @ino: inode number to search for
1123 *
1124 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1125 * This is for file systems where the inode number is sufficient for unique
1126 * identification of an inode.
1127 *
1128 * If the inode is in the cache, the inode is returned with an incremented
1129 * reference count.
1130 *
1131 * Otherwise NULL is returned.
1132 */
1133struct inode *ilookup(struct super_block *sb, unsigned long ino)
1134{
1135 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1136
1137 return ifind_fast(sb, head, ino);
1138}
1da177e4
LT
1139EXPORT_SYMBOL(ilookup);
1140
1141/**
1142 * iget5_locked - obtain an inode from a mounted file system
1143 * @sb: super block of file system
1144 * @hashval: hash value (usually inode number) to get
1145 * @test: callback used for comparisons between inodes
1146 * @set: callback used to initialize a new struct inode
1147 * @data: opaque data pointer to pass to @test and @set
1148 *
1da177e4
LT
1149 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1150 * and @data in the inode cache and if present it is returned with an increased
1151 * reference count. This is a generalized version of iget_locked() for file
1152 * systems where the inode number is not sufficient for unique identification
1153 * of an inode.
1154 *
1155 * If the inode is not in cache, get_new_inode() is called to allocate a new
1156 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1157 * file system gets to fill it in before unlocking it via unlock_new_inode().
1158 *
1159 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1160 */
1161struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1162 int (*test)(struct inode *, void *),
1163 int (*set)(struct inode *, void *), void *data)
1164{
1165 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1166 struct inode *inode;
1167
88bd5121 1168 inode = ifind(sb, head, test, data, 1);
1da177e4
LT
1169 if (inode)
1170 return inode;
1171 /*
1172 * get_new_inode() will do the right thing, re-trying the search
1173 * in case it had to block at any point.
1174 */
1175 return get_new_inode(sb, head, test, set, data);
1176}
1da177e4
LT
1177EXPORT_SYMBOL(iget5_locked);
1178
1179/**
1180 * iget_locked - obtain an inode from a mounted file system
1181 * @sb: super block of file system
1182 * @ino: inode number to get
1183 *
1da177e4
LT
1184 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1185 * the inode cache and if present it is returned with an increased reference
1186 * count. This is for file systems where the inode number is sufficient for
1187 * unique identification of an inode.
1188 *
1189 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1190 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1191 * The file system gets to fill it in before unlocking it via
1192 * unlock_new_inode().
1193 */
1194struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1195{
1196 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1197 struct inode *inode;
1198
1199 inode = ifind_fast(sb, head, ino);
1200 if (inode)
1201 return inode;
1202 /*
1203 * get_new_inode_fast() will do the right thing, re-trying the search
1204 * in case it had to block at any point.
1205 */
1206 return get_new_inode_fast(sb, head, ino);
1207}
1da177e4
LT
1208EXPORT_SYMBOL(iget_locked);
1209
261bca86
AV
1210int insert_inode_locked(struct inode *inode)
1211{
1212 struct super_block *sb = inode->i_sb;
1213 ino_t ino = inode->i_ino;
1214 struct hlist_head *head = inode_hashtable + hash(sb, ino);
261bca86 1215
eaff8079 1216 inode->i_state |= I_NEW;
261bca86 1217 while (1) {
72a43d63
AV
1218 struct hlist_node *node;
1219 struct inode *old = NULL;
261bca86 1220 spin_lock(&inode_lock);
72a43d63
AV
1221 hlist_for_each_entry(old, node, head, i_hash) {
1222 if (old->i_ino != ino)
1223 continue;
1224 if (old->i_sb != sb)
1225 continue;
a4ffdde6 1226 if (old->i_state & (I_FREEING|I_WILL_FREE))
72a43d63
AV
1227 continue;
1228 break;
1229 }
1230 if (likely(!node)) {
261bca86
AV
1231 hlist_add_head(&inode->i_hash, head);
1232 spin_unlock(&inode_lock);
1233 return 0;
1234 }
1235 __iget(old);
1236 spin_unlock(&inode_lock);
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
eaff8079 1253 inode->i_state |= I_NEW;
261bca86
AV
1254
1255 while (1) {
72a43d63
AV
1256 struct hlist_node *node;
1257 struct inode *old = NULL;
1258
261bca86 1259 spin_lock(&inode_lock);
72a43d63
AV
1260 hlist_for_each_entry(old, node, head, i_hash) {
1261 if (old->i_sb != sb)
1262 continue;
1263 if (!test(old, data))
1264 continue;
a4ffdde6 1265 if (old->i_state & (I_FREEING|I_WILL_FREE))
72a43d63
AV
1266 continue;
1267 break;
1268 }
1269 if (likely(!node)) {
261bca86
AV
1270 hlist_add_head(&inode->i_hash, head);
1271 spin_unlock(&inode_lock);
1272 return 0;
1273 }
1274 __iget(old);
1275 spin_unlock(&inode_lock);
1276 wait_on_inode(old);
1d3382cb 1277 if (unlikely(!inode_unhashed(old))) {
261bca86
AV
1278 iput(old);
1279 return -EBUSY;
1280 }
1281 iput(old);
1282 }
1283}
261bca86
AV
1284EXPORT_SYMBOL(insert_inode_locked4);
1285
1da177e4 1286
45321ac5
AV
1287int generic_delete_inode(struct inode *inode)
1288{
1289 return 1;
1290}
1291EXPORT_SYMBOL(generic_delete_inode);
1292
1da177e4 1293/*
45321ac5
AV
1294 * Normal UNIX filesystem behaviour: delete the
1295 * inode when the usage count drops to zero, and
1296 * i_nlink is zero.
1da177e4 1297 */
45321ac5 1298int generic_drop_inode(struct inode *inode)
1da177e4 1299{
1d3382cb 1300 return !inode->i_nlink || inode_unhashed(inode);
1da177e4 1301}
45321ac5 1302EXPORT_SYMBOL_GPL(generic_drop_inode);
1da177e4 1303
45321ac5
AV
1304/*
1305 * Called when we're dropping the last reference
1306 * to an inode.
22fe4042 1307 *
45321ac5
AV
1308 * Call the FS "drop_inode()" function, defaulting to
1309 * the legacy UNIX filesystem behaviour. If it tells
1310 * us to evict inode, do so. Otherwise, retain inode
1311 * in cache if fs is alive, sync and evict if fs is
1312 * shutting down.
22fe4042 1313 */
45321ac5 1314static void iput_final(struct inode *inode)
1da177e4
LT
1315{
1316 struct super_block *sb = inode->i_sb;
45321ac5
AV
1317 const struct super_operations *op = inode->i_sb->s_op;
1318 int drop;
1319
1320 if (op && op->drop_inode)
1321 drop = op->drop_inode(inode);
1322 else
1323 drop = generic_drop_inode(inode);
1da177e4 1324
45321ac5 1325 if (!drop) {
acb0c854 1326 if (sb->s_flags & MS_ACTIVE) {
9e38d86f
NP
1327 inode->i_state |= I_REFERENCED;
1328 if (!(inode->i_state & (I_DIRTY|I_SYNC))) {
1329 inode_lru_list_add(inode);
1330 }
991114c6 1331 spin_unlock(&inode_lock);
45321ac5 1332 return;
991114c6 1333 }
7ef0d737 1334 WARN_ON(inode->i_state & I_NEW);
991114c6
AV
1335 inode->i_state |= I_WILL_FREE;
1336 spin_unlock(&inode_lock);
1da177e4
LT
1337 write_inode_now(inode, 1);
1338 spin_lock(&inode_lock);
7ef0d737 1339 WARN_ON(inode->i_state & I_NEW);
991114c6 1340 inode->i_state &= ~I_WILL_FREE;
4c51acbc 1341 __remove_inode_hash(inode);
1da177e4 1342 }
7ef0d737 1343 WARN_ON(inode->i_state & I_NEW);
991114c6 1344 inode->i_state |= I_FREEING;
9e38d86f
NP
1345
1346 /*
1347 * After we delete the inode from the LRU here, we avoid moving dirty
1348 * inodes back onto the LRU now because I_FREEING is set and hence
1349 * writeback_single_inode() won't move the inode around.
1350 */
1351 inode_lru_list_del(inode);
1352
646ec461 1353 __inode_sb_list_del(inode);
1da177e4 1354 spin_unlock(&inode_lock);
644da596 1355 evict(inode);
4c51acbc 1356 remove_inode_hash(inode);
7f04c26d 1357 wake_up_inode(inode);
45321ac5 1358 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
1da177e4
LT
1359 destroy_inode(inode);
1360}
1361
1da177e4 1362/**
6b3304b5 1363 * iput - put an inode
1da177e4
LT
1364 * @inode: inode to put
1365 *
1366 * Puts an inode, dropping its usage count. If the inode use count hits
1367 * zero, the inode is then freed and may also be destroyed.
1368 *
1369 * Consequently, iput() can sleep.
1370 */
1371void iput(struct inode *inode)
1372{
1373 if (inode) {
a4ffdde6 1374 BUG_ON(inode->i_state & I_CLEAR);
1da177e4 1375
1da177e4
LT
1376 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1377 iput_final(inode);
1378 }
1379}
1da177e4
LT
1380EXPORT_SYMBOL(iput);
1381
1382/**
1383 * bmap - find a block number in a file
1384 * @inode: inode of file
1385 * @block: block to find
1386 *
1387 * Returns the block number on the device holding the inode that
1388 * is the disk block number for the block of the file requested.
1389 * That is, asked for block 4 of inode 1 the function will return the
6b3304b5 1390 * disk block relative to the disk start that holds that block of the
1da177e4
LT
1391 * file.
1392 */
6b3304b5 1393sector_t bmap(struct inode *inode, sector_t block)
1da177e4
LT
1394{
1395 sector_t res = 0;
1396 if (inode->i_mapping->a_ops->bmap)
1397 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1398 return res;
1399}
1da177e4
LT
1400EXPORT_SYMBOL(bmap);
1401
11ff6f05
MG
1402/*
1403 * With relative atime, only update atime if the previous atime is
1404 * earlier than either the ctime or mtime or if at least a day has
1405 * passed since the last atime update.
1406 */
1407static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1408 struct timespec now)
1409{
1410
1411 if (!(mnt->mnt_flags & MNT_RELATIME))
1412 return 1;
1413 /*
1414 * Is mtime younger than atime? If yes, update atime:
1415 */
1416 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1417 return 1;
1418 /*
1419 * Is ctime younger than atime? If yes, update atime:
1420 */
1421 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1422 return 1;
1423
1424 /*
1425 * Is the previous atime value older than a day? If yes,
1426 * update atime:
1427 */
1428 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1429 return 1;
1430 /*
1431 * Good, we can skip the atime update:
1432 */
1433 return 0;
1434}
1435
1da177e4 1436/**
869243a0
CH
1437 * touch_atime - update the access time
1438 * @mnt: mount the inode is accessed on
7045f37b 1439 * @dentry: dentry accessed
1da177e4
LT
1440 *
1441 * Update the accessed time on an inode and mark it for writeback.
1442 * This function automatically handles read only file systems and media,
1443 * as well as the "noatime" flag and inode specific "noatime" markers.
1444 */
869243a0 1445void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1da177e4 1446{
869243a0 1447 struct inode *inode = dentry->d_inode;
1da177e4
LT
1448 struct timespec now;
1449
cdb70f3f 1450 if (inode->i_flags & S_NOATIME)
b12536c2 1451 return;
37756ced 1452 if (IS_NOATIME(inode))
b12536c2 1453 return;
b2276138 1454 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
b12536c2 1455 return;
47ae32d6 1456
cdb70f3f 1457 if (mnt->mnt_flags & MNT_NOATIME)
b12536c2 1458 return;
cdb70f3f 1459 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
b12536c2 1460 return;
1da177e4
LT
1461
1462 now = current_fs_time(inode->i_sb);
11ff6f05
MG
1463
1464 if (!relatime_need_update(mnt, inode, now))
b12536c2 1465 return;
11ff6f05 1466
47ae32d6 1467 if (timespec_equal(&inode->i_atime, &now))
b12536c2
AK
1468 return;
1469
1470 if (mnt_want_write(mnt))
1471 return;
47ae32d6
VH
1472
1473 inode->i_atime = now;
1474 mark_inode_dirty_sync(inode);
cdb70f3f 1475 mnt_drop_write(mnt);
1da177e4 1476}
869243a0 1477EXPORT_SYMBOL(touch_atime);
1da177e4
LT
1478
1479/**
870f4817
CH
1480 * file_update_time - update mtime and ctime time
1481 * @file: file accessed
1da177e4 1482 *
870f4817
CH
1483 * Update the mtime and ctime members of an inode and mark the inode
1484 * for writeback. Note that this function is meant exclusively for
1485 * usage in the file write path of filesystems, and filesystems may
1486 * choose to explicitly ignore update via this function with the
2eadfc0e 1487 * S_NOCMTIME inode flag, e.g. for network filesystem where these
870f4817 1488 * timestamps are handled by the server.
1da177e4
LT
1489 */
1490
870f4817 1491void file_update_time(struct file *file)
1da177e4 1492{
0f7fc9e4 1493 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4 1494 struct timespec now;
ce06e0b2 1495 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1da177e4 1496
ce06e0b2 1497 /* First try to exhaust all avenues to not sync */
1da177e4
LT
1498 if (IS_NOCMTIME(inode))
1499 return;
20ddee2c 1500
1da177e4 1501 now = current_fs_time(inode->i_sb);
ce06e0b2
AK
1502 if (!timespec_equal(&inode->i_mtime, &now))
1503 sync_it = S_MTIME;
1da177e4 1504
ce06e0b2
AK
1505 if (!timespec_equal(&inode->i_ctime, &now))
1506 sync_it |= S_CTIME;
870f4817 1507
ce06e0b2
AK
1508 if (IS_I_VERSION(inode))
1509 sync_it |= S_VERSION;
7a224228 1510
ce06e0b2
AK
1511 if (!sync_it)
1512 return;
1513
1514 /* Finally allowed to write? Takes lock. */
1515 if (mnt_want_write_file(file))
1516 return;
1517
1518 /* Only change inode inside the lock region */
1519 if (sync_it & S_VERSION)
1520 inode_inc_iversion(inode);
1521 if (sync_it & S_CTIME)
1522 inode->i_ctime = now;
1523 if (sync_it & S_MTIME)
1524 inode->i_mtime = now;
1525 mark_inode_dirty_sync(inode);
20ddee2c 1526 mnt_drop_write(file->f_path.mnt);
1da177e4 1527}
870f4817 1528EXPORT_SYMBOL(file_update_time);
1da177e4
LT
1529
1530int inode_needs_sync(struct inode *inode)
1531{
1532 if (IS_SYNC(inode))
1533 return 1;
1534 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1535 return 1;
1536 return 0;
1537}
1da177e4
LT
1538EXPORT_SYMBOL(inode_needs_sync);
1539
1da177e4
LT
1540int inode_wait(void *word)
1541{
1542 schedule();
1543 return 0;
1544}
d44dab8d 1545EXPORT_SYMBOL(inode_wait);
1da177e4
LT
1546
1547/*
168a9fd6
MS
1548 * If we try to find an inode in the inode hash while it is being
1549 * deleted, we have to wait until the filesystem completes its
1550 * deletion before reporting that it isn't found. This function waits
1551 * until the deletion _might_ have completed. Callers are responsible
1552 * to recheck inode state.
1553 *
eaff8079 1554 * It doesn't matter if I_NEW is not set initially, a call to
168a9fd6
MS
1555 * wake_up_inode() after removing from the hash list will DTRT.
1556 *
1da177e4
LT
1557 * This is called with inode_lock held.
1558 */
1559static void __wait_on_freeing_inode(struct inode *inode)
1560{
1561 wait_queue_head_t *wq;
eaff8079
CH
1562 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1563 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1da177e4
LT
1564 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1565 spin_unlock(&inode_lock);
1566 schedule();
1567 finish_wait(wq, &wait.wait);
1568 spin_lock(&inode_lock);
1569}
1570
1da177e4
LT
1571static __initdata unsigned long ihash_entries;
1572static int __init set_ihash_entries(char *str)
1573{
1574 if (!str)
1575 return 0;
1576 ihash_entries = simple_strtoul(str, &str, 0);
1577 return 1;
1578}
1579__setup("ihash_entries=", set_ihash_entries);
1580
1581/*
1582 * Initialize the waitqueues and inode hash table.
1583 */
1584void __init inode_init_early(void)
1585{
1586 int loop;
1587
1588 /* If hashes are distributed across NUMA nodes, defer
1589 * hash allocation until vmalloc space is available.
1590 */
1591 if (hashdist)
1592 return;
1593
1594 inode_hashtable =
1595 alloc_large_system_hash("Inode-cache",
1596 sizeof(struct hlist_head),
1597 ihash_entries,
1598 14,
1599 HASH_EARLY,
1600 &i_hash_shift,
1601 &i_hash_mask,
1602 0);
1603
1604 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1605 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1606}
1607
74bf17cf 1608void __init inode_init(void)
1da177e4
LT
1609{
1610 int loop;
1611
1612 /* inode slab cache */
b0196009
PJ
1613 inode_cachep = kmem_cache_create("inode_cache",
1614 sizeof(struct inode),
1615 0,
1616 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1617 SLAB_MEM_SPREAD),
20c2df83 1618 init_once);
8e1f936b 1619 register_shrinker(&icache_shrinker);
cffbc8aa
DC
1620 percpu_counter_init(&nr_inodes, 0);
1621 percpu_counter_init(&nr_inodes_unused, 0);
1da177e4
LT
1622
1623 /* Hash may have been set up in inode_init_early */
1624 if (!hashdist)
1625 return;
1626
1627 inode_hashtable =
1628 alloc_large_system_hash("Inode-cache",
1629 sizeof(struct hlist_head),
1630 ihash_entries,
1631 14,
1632 0,
1633 &i_hash_shift,
1634 &i_hash_mask,
1635 0);
1636
1637 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1638 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1639}
1640
1641void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1642{
1643 inode->i_mode = mode;
1644 if (S_ISCHR(mode)) {
1645 inode->i_fop = &def_chr_fops;
1646 inode->i_rdev = rdev;
1647 } else if (S_ISBLK(mode)) {
1648 inode->i_fop = &def_blk_fops;
1649 inode->i_rdev = rdev;
1650 } else if (S_ISFIFO(mode))
1651 inode->i_fop = &def_fifo_fops;
1652 else if (S_ISSOCK(mode))
1653 inode->i_fop = &bad_sock_fops;
1654 else
af0d9ae8
MK
1655 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1656 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1657 inode->i_ino);
1da177e4
LT
1658}
1659EXPORT_SYMBOL(init_special_inode);
a1bd120d
DM
1660
1661/**
1662 * Init uid,gid,mode for new inode according to posix standards
1663 * @inode: New inode
1664 * @dir: Directory inode
1665 * @mode: mode of the new inode
1666 */
1667void inode_init_owner(struct inode *inode, const struct inode *dir,
1668 mode_t mode)
1669{
1670 inode->i_uid = current_fsuid();
1671 if (dir && dir->i_mode & S_ISGID) {
1672 inode->i_gid = dir->i_gid;
1673 if (S_ISDIR(mode))
1674 mode |= S_ISGID;
1675 } else
1676 inode->i_gid = current_fsgid();
1677 inode->i_mode = mode;
1678}
1679EXPORT_SYMBOL(inode_init_owner);