Loading...
1/*
2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
4 */
5#include <linux/fs.h>
6#include <linux/mm.h>
7#include <linux/dcache.h>
8#include <linux/init.h>
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>
14#include <linux/rwsem.h>
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>
21#include <linux/fsnotify.h>
22#include <linux/mount.h>
23#include <linux/async.h>
24#include <linux/posix_acl.h>
25#include <linux/prefetch.h>
26#include <linux/ima.h>
27#include <linux/cred.h>
28#include <linux/buffer_head.h> /* for inode_has_buffers */
29#include "internal.h"
30
31/*
32 * Inode locking rules:
33 *
34 * inode->i_lock protects:
35 * inode->i_state, inode->i_hash, __iget()
36 * inode->i_sb->s_inode_lru_lock protects:
37 * inode->i_sb->s_inode_lru, inode->i_lru
38 * inode_sb_list_lock protects:
39 * sb->s_inodes, inode->i_sb_list
40 * bdi->wb.list_lock protects:
41 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
42 * inode_hash_lock protects:
43 * inode_hashtable, inode->i_hash
44 *
45 * Lock ordering:
46 *
47 * inode_sb_list_lock
48 * inode->i_lock
49 * inode->i_sb->s_inode_lru_lock
50 *
51 * bdi->wb.list_lock
52 * inode->i_lock
53 *
54 * inode_hash_lock
55 * inode_sb_list_lock
56 * inode->i_lock
57 *
58 * iunique_lock
59 * inode_hash_lock
60 */
61
62static unsigned int i_hash_mask __read_mostly;
63static unsigned int i_hash_shift __read_mostly;
64static struct hlist_head *inode_hashtable __read_mostly;
65static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
66
67__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
68
69/*
70 * Empty aops. Can be used for the cases where the user does not
71 * define any of the address_space operations.
72 */
73const struct address_space_operations empty_aops = {
74};
75EXPORT_SYMBOL(empty_aops);
76
77/*
78 * Statistics gathering..
79 */
80struct inodes_stat_t inodes_stat;
81
82static DEFINE_PER_CPU(unsigned int, nr_inodes);
83static DEFINE_PER_CPU(unsigned int, nr_unused);
84
85static struct kmem_cache *inode_cachep __read_mostly;
86
87static int get_nr_inodes(void)
88{
89 int i;
90 int sum = 0;
91 for_each_possible_cpu(i)
92 sum += per_cpu(nr_inodes, i);
93 return sum < 0 ? 0 : sum;
94}
95
96static inline int get_nr_inodes_unused(void)
97{
98 int i;
99 int sum = 0;
100 for_each_possible_cpu(i)
101 sum += per_cpu(nr_unused, i);
102 return sum < 0 ? 0 : sum;
103}
104
105int get_nr_dirty_inodes(void)
106{
107 /* not actually dirty inodes, but a wild approximation */
108 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
109 return nr_dirty > 0 ? nr_dirty : 0;
110}
111
112/*
113 * Handle nr_inode sysctl
114 */
115#ifdef CONFIG_SYSCTL
116int proc_nr_inodes(ctl_table *table, int write,
117 void __user *buffer, size_t *lenp, loff_t *ppos)
118{
119 inodes_stat.nr_inodes = get_nr_inodes();
120 inodes_stat.nr_unused = get_nr_inodes_unused();
121 return proc_dointvec(table, write, buffer, lenp, ppos);
122}
123#endif
124
125/**
126 * inode_init_always - perform inode structure intialisation
127 * @sb: superblock inode belongs to
128 * @inode: inode to initialise
129 *
130 * These are initializations that need to be done on every inode
131 * allocation as the fields are not initialised by slab allocation.
132 */
133int inode_init_always(struct super_block *sb, struct inode *inode)
134{
135 static const struct inode_operations empty_iops;
136 static const struct file_operations empty_fops;
137 struct address_space *const mapping = &inode->i_data;
138
139 inode->i_sb = sb;
140 inode->i_blkbits = sb->s_blocksize_bits;
141 inode->i_flags = 0;
142 atomic_set(&inode->i_count, 1);
143 inode->i_op = &empty_iops;
144 inode->i_fop = &empty_fops;
145 inode->i_nlink = 1;
146 inode->i_opflags = 0;
147 inode->i_uid = 0;
148 inode->i_gid = 0;
149 atomic_set(&inode->i_writecount, 0);
150 inode->i_size = 0;
151 inode->i_blocks = 0;
152 inode->i_bytes = 0;
153 inode->i_generation = 0;
154#ifdef CONFIG_QUOTA
155 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
156#endif
157 inode->i_pipe = NULL;
158 inode->i_bdev = NULL;
159 inode->i_cdev = NULL;
160 inode->i_rdev = 0;
161 inode->dirtied_when = 0;
162
163 if (security_inode_alloc(inode))
164 goto out;
165 spin_lock_init(&inode->i_lock);
166 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
167
168 mutex_init(&inode->i_mutex);
169 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
170
171 atomic_set(&inode->i_dio_count, 0);
172
173 mapping->a_ops = &empty_aops;
174 mapping->host = inode;
175 mapping->flags = 0;
176 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
177 mapping->assoc_mapping = NULL;
178 mapping->backing_dev_info = &default_backing_dev_info;
179 mapping->writeback_index = 0;
180
181 /*
182 * If the block_device provides a backing_dev_info for client
183 * inodes then use that. Otherwise the inode share the bdev's
184 * backing_dev_info.
185 */
186 if (sb->s_bdev) {
187 struct backing_dev_info *bdi;
188
189 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
190 mapping->backing_dev_info = bdi;
191 }
192 inode->i_private = NULL;
193 inode->i_mapping = mapping;
194#ifdef CONFIG_FS_POSIX_ACL
195 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
196#endif
197
198#ifdef CONFIG_FSNOTIFY
199 inode->i_fsnotify_mask = 0;
200#endif
201
202 this_cpu_inc(nr_inodes);
203
204 return 0;
205out:
206 return -ENOMEM;
207}
208EXPORT_SYMBOL(inode_init_always);
209
210static struct inode *alloc_inode(struct super_block *sb)
211{
212 struct inode *inode;
213
214 if (sb->s_op->alloc_inode)
215 inode = sb->s_op->alloc_inode(sb);
216 else
217 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
218
219 if (!inode)
220 return NULL;
221
222 if (unlikely(inode_init_always(sb, inode))) {
223 if (inode->i_sb->s_op->destroy_inode)
224 inode->i_sb->s_op->destroy_inode(inode);
225 else
226 kmem_cache_free(inode_cachep, inode);
227 return NULL;
228 }
229
230 return inode;
231}
232
233void free_inode_nonrcu(struct inode *inode)
234{
235 kmem_cache_free(inode_cachep, inode);
236}
237EXPORT_SYMBOL(free_inode_nonrcu);
238
239void __destroy_inode(struct inode *inode)
240{
241 BUG_ON(inode_has_buffers(inode));
242 security_inode_free(inode);
243 fsnotify_inode_delete(inode);
244#ifdef CONFIG_FS_POSIX_ACL
245 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
246 posix_acl_release(inode->i_acl);
247 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
248 posix_acl_release(inode->i_default_acl);
249#endif
250 this_cpu_dec(nr_inodes);
251}
252EXPORT_SYMBOL(__destroy_inode);
253
254static void i_callback(struct rcu_head *head)
255{
256 struct inode *inode = container_of(head, struct inode, i_rcu);
257 INIT_LIST_HEAD(&inode->i_dentry);
258 kmem_cache_free(inode_cachep, inode);
259}
260
261static void destroy_inode(struct inode *inode)
262{
263 BUG_ON(!list_empty(&inode->i_lru));
264 __destroy_inode(inode);
265 if (inode->i_sb->s_op->destroy_inode)
266 inode->i_sb->s_op->destroy_inode(inode);
267 else
268 call_rcu(&inode->i_rcu, i_callback);
269}
270
271void address_space_init_once(struct address_space *mapping)
272{
273 memset(mapping, 0, sizeof(*mapping));
274 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
275 spin_lock_init(&mapping->tree_lock);
276 mutex_init(&mapping->i_mmap_mutex);
277 INIT_LIST_HEAD(&mapping->private_list);
278 spin_lock_init(&mapping->private_lock);
279 INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
280 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
281}
282EXPORT_SYMBOL(address_space_init_once);
283
284/*
285 * These are initializations that only need to be done
286 * once, because the fields are idempotent across use
287 * of the inode, so let the slab aware of that.
288 */
289void inode_init_once(struct inode *inode)
290{
291 memset(inode, 0, sizeof(*inode));
292 INIT_HLIST_NODE(&inode->i_hash);
293 INIT_LIST_HEAD(&inode->i_dentry);
294 INIT_LIST_HEAD(&inode->i_devices);
295 INIT_LIST_HEAD(&inode->i_wb_list);
296 INIT_LIST_HEAD(&inode->i_lru);
297 address_space_init_once(&inode->i_data);
298 i_size_ordered_init(inode);
299#ifdef CONFIG_FSNOTIFY
300 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
301#endif
302}
303EXPORT_SYMBOL(inode_init_once);
304
305static void init_once(void *foo)
306{
307 struct inode *inode = (struct inode *) foo;
308
309 inode_init_once(inode);
310}
311
312/*
313 * inode->i_lock must be held
314 */
315void __iget(struct inode *inode)
316{
317 atomic_inc(&inode->i_count);
318}
319
320/*
321 * get additional reference to inode; caller must already hold one.
322 */
323void ihold(struct inode *inode)
324{
325 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
326}
327EXPORT_SYMBOL(ihold);
328
329static void inode_lru_list_add(struct inode *inode)
330{
331 spin_lock(&inode->i_sb->s_inode_lru_lock);
332 if (list_empty(&inode->i_lru)) {
333 list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
334 inode->i_sb->s_nr_inodes_unused++;
335 this_cpu_inc(nr_unused);
336 }
337 spin_unlock(&inode->i_sb->s_inode_lru_lock);
338}
339
340static void inode_lru_list_del(struct inode *inode)
341{
342 spin_lock(&inode->i_sb->s_inode_lru_lock);
343 if (!list_empty(&inode->i_lru)) {
344 list_del_init(&inode->i_lru);
345 inode->i_sb->s_nr_inodes_unused--;
346 this_cpu_dec(nr_unused);
347 }
348 spin_unlock(&inode->i_sb->s_inode_lru_lock);
349}
350
351/**
352 * inode_sb_list_add - add inode to the superblock list of inodes
353 * @inode: inode to add
354 */
355void inode_sb_list_add(struct inode *inode)
356{
357 spin_lock(&inode_sb_list_lock);
358 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
359 spin_unlock(&inode_sb_list_lock);
360}
361EXPORT_SYMBOL_GPL(inode_sb_list_add);
362
363static inline void inode_sb_list_del(struct inode *inode)
364{
365 if (!list_empty(&inode->i_sb_list)) {
366 spin_lock(&inode_sb_list_lock);
367 list_del_init(&inode->i_sb_list);
368 spin_unlock(&inode_sb_list_lock);
369 }
370}
371
372static unsigned long hash(struct super_block *sb, unsigned long hashval)
373{
374 unsigned long tmp;
375
376 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
377 L1_CACHE_BYTES;
378 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
379 return tmp & i_hash_mask;
380}
381
382/**
383 * __insert_inode_hash - hash an inode
384 * @inode: unhashed inode
385 * @hashval: unsigned long value used to locate this object in the
386 * inode_hashtable.
387 *
388 * Add an inode to the inode hash for this superblock.
389 */
390void __insert_inode_hash(struct inode *inode, unsigned long hashval)
391{
392 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
393
394 spin_lock(&inode_hash_lock);
395 spin_lock(&inode->i_lock);
396 hlist_add_head(&inode->i_hash, b);
397 spin_unlock(&inode->i_lock);
398 spin_unlock(&inode_hash_lock);
399}
400EXPORT_SYMBOL(__insert_inode_hash);
401
402/**
403 * __remove_inode_hash - remove an inode from the hash
404 * @inode: inode to unhash
405 *
406 * Remove an inode from the superblock.
407 */
408void __remove_inode_hash(struct inode *inode)
409{
410 spin_lock(&inode_hash_lock);
411 spin_lock(&inode->i_lock);
412 hlist_del_init(&inode->i_hash);
413 spin_unlock(&inode->i_lock);
414 spin_unlock(&inode_hash_lock);
415}
416EXPORT_SYMBOL(__remove_inode_hash);
417
418void end_writeback(struct inode *inode)
419{
420 might_sleep();
421 /*
422 * We have to cycle tree_lock here because reclaim can be still in the
423 * process of removing the last page (in __delete_from_page_cache())
424 * and we must not free mapping under it.
425 */
426 spin_lock_irq(&inode->i_data.tree_lock);
427 BUG_ON(inode->i_data.nrpages);
428 spin_unlock_irq(&inode->i_data.tree_lock);
429 BUG_ON(!list_empty(&inode->i_data.private_list));
430 BUG_ON(!(inode->i_state & I_FREEING));
431 BUG_ON(inode->i_state & I_CLEAR);
432 inode_sync_wait(inode);
433 /* don't need i_lock here, no concurrent mods to i_state */
434 inode->i_state = I_FREEING | I_CLEAR;
435}
436EXPORT_SYMBOL(end_writeback);
437
438/*
439 * Free the inode passed in, removing it from the lists it is still connected
440 * to. We remove any pages still attached to the inode and wait for any IO that
441 * is still in progress before finally destroying the inode.
442 *
443 * An inode must already be marked I_FREEING so that we avoid the inode being
444 * moved back onto lists if we race with other code that manipulates the lists
445 * (e.g. writeback_single_inode). The caller is responsible for setting this.
446 *
447 * An inode must already be removed from the LRU list before being evicted from
448 * the cache. This should occur atomically with setting the I_FREEING state
449 * flag, so no inodes here should ever be on the LRU when being evicted.
450 */
451static void evict(struct inode *inode)
452{
453 const struct super_operations *op = inode->i_sb->s_op;
454
455 BUG_ON(!(inode->i_state & I_FREEING));
456 BUG_ON(!list_empty(&inode->i_lru));
457
458 if (!list_empty(&inode->i_wb_list))
459 inode_wb_list_del(inode);
460
461 inode_sb_list_del(inode);
462
463 if (op->evict_inode) {
464 op->evict_inode(inode);
465 } else {
466 if (inode->i_data.nrpages)
467 truncate_inode_pages(&inode->i_data, 0);
468 end_writeback(inode);
469 }
470 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
471 bd_forget(inode);
472 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
473 cd_forget(inode);
474
475 remove_inode_hash(inode);
476
477 spin_lock(&inode->i_lock);
478 wake_up_bit(&inode->i_state, __I_NEW);
479 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
480 spin_unlock(&inode->i_lock);
481
482 destroy_inode(inode);
483}
484
485/*
486 * dispose_list - dispose of the contents of a local list
487 * @head: the head of the list to free
488 *
489 * Dispose-list gets a local list with local inodes in it, so it doesn't
490 * need to worry about list corruption and SMP locks.
491 */
492static void dispose_list(struct list_head *head)
493{
494 while (!list_empty(head)) {
495 struct inode *inode;
496
497 inode = list_first_entry(head, struct inode, i_lru);
498 list_del_init(&inode->i_lru);
499
500 evict(inode);
501 }
502}
503
504/**
505 * evict_inodes - evict all evictable inodes for a superblock
506 * @sb: superblock to operate on
507 *
508 * Make sure that no inodes with zero refcount are retained. This is
509 * called by superblock shutdown after having MS_ACTIVE flag removed,
510 * so any inode reaching zero refcount during or after that call will
511 * be immediately evicted.
512 */
513void evict_inodes(struct super_block *sb)
514{
515 struct inode *inode, *next;
516 LIST_HEAD(dispose);
517
518 spin_lock(&inode_sb_list_lock);
519 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
520 if (atomic_read(&inode->i_count))
521 continue;
522
523 spin_lock(&inode->i_lock);
524 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
525 spin_unlock(&inode->i_lock);
526 continue;
527 }
528
529 inode->i_state |= I_FREEING;
530 inode_lru_list_del(inode);
531 spin_unlock(&inode->i_lock);
532 list_add(&inode->i_lru, &dispose);
533 }
534 spin_unlock(&inode_sb_list_lock);
535
536 dispose_list(&dispose);
537}
538
539/**
540 * invalidate_inodes - attempt to free all inodes on a superblock
541 * @sb: superblock to operate on
542 * @kill_dirty: flag to guide handling of dirty inodes
543 *
544 * Attempts to free all inodes for a given superblock. If there were any
545 * busy inodes return a non-zero value, else zero.
546 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
547 * them as busy.
548 */
549int invalidate_inodes(struct super_block *sb, bool kill_dirty)
550{
551 int busy = 0;
552 struct inode *inode, *next;
553 LIST_HEAD(dispose);
554
555 spin_lock(&inode_sb_list_lock);
556 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
557 spin_lock(&inode->i_lock);
558 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
559 spin_unlock(&inode->i_lock);
560 continue;
561 }
562 if (inode->i_state & I_DIRTY && !kill_dirty) {
563 spin_unlock(&inode->i_lock);
564 busy = 1;
565 continue;
566 }
567 if (atomic_read(&inode->i_count)) {
568 spin_unlock(&inode->i_lock);
569 busy = 1;
570 continue;
571 }
572
573 inode->i_state |= I_FREEING;
574 inode_lru_list_del(inode);
575 spin_unlock(&inode->i_lock);
576 list_add(&inode->i_lru, &dispose);
577 }
578 spin_unlock(&inode_sb_list_lock);
579
580 dispose_list(&dispose);
581
582 return busy;
583}
584
585static int can_unuse(struct inode *inode)
586{
587 if (inode->i_state & ~I_REFERENCED)
588 return 0;
589 if (inode_has_buffers(inode))
590 return 0;
591 if (atomic_read(&inode->i_count))
592 return 0;
593 if (inode->i_data.nrpages)
594 return 0;
595 return 1;
596}
597
598/*
599 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
600 * This is called from the superblock shrinker function with a number of inodes
601 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
602 * then are freed outside inode_lock by dispose_list().
603 *
604 * Any inodes which are pinned purely because of attached pagecache have their
605 * pagecache removed. If the inode has metadata buffers attached to
606 * mapping->private_list then try to remove them.
607 *
608 * If the inode has the I_REFERENCED flag set, then it means that it has been
609 * used recently - the flag is set in iput_final(). When we encounter such an
610 * inode, clear the flag and move it to the back of the LRU so it gets another
611 * pass through the LRU before it gets reclaimed. This is necessary because of
612 * the fact we are doing lazy LRU updates to minimise lock contention so the
613 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
614 * with this flag set because they are the inodes that are out of order.
615 */
616void prune_icache_sb(struct super_block *sb, int nr_to_scan)
617{
618 LIST_HEAD(freeable);
619 int nr_scanned;
620 unsigned long reap = 0;
621
622 spin_lock(&sb->s_inode_lru_lock);
623 for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
624 struct inode *inode;
625
626 if (list_empty(&sb->s_inode_lru))
627 break;
628
629 inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
630
631 /*
632 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
633 * so use a trylock. If we fail to get the lock, just move the
634 * inode to the back of the list so we don't spin on it.
635 */
636 if (!spin_trylock(&inode->i_lock)) {
637 list_move(&inode->i_lru, &sb->s_inode_lru);
638 continue;
639 }
640
641 /*
642 * Referenced or dirty inodes are still in use. Give them
643 * another pass through the LRU as we canot reclaim them now.
644 */
645 if (atomic_read(&inode->i_count) ||
646 (inode->i_state & ~I_REFERENCED)) {
647 list_del_init(&inode->i_lru);
648 spin_unlock(&inode->i_lock);
649 sb->s_nr_inodes_unused--;
650 this_cpu_dec(nr_unused);
651 continue;
652 }
653
654 /* recently referenced inodes get one more pass */
655 if (inode->i_state & I_REFERENCED) {
656 inode->i_state &= ~I_REFERENCED;
657 list_move(&inode->i_lru, &sb->s_inode_lru);
658 spin_unlock(&inode->i_lock);
659 continue;
660 }
661 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
662 __iget(inode);
663 spin_unlock(&inode->i_lock);
664 spin_unlock(&sb->s_inode_lru_lock);
665 if (remove_inode_buffers(inode))
666 reap += invalidate_mapping_pages(&inode->i_data,
667 0, -1);
668 iput(inode);
669 spin_lock(&sb->s_inode_lru_lock);
670
671 if (inode != list_entry(sb->s_inode_lru.next,
672 struct inode, i_lru))
673 continue; /* wrong inode or list_empty */
674 /* avoid lock inversions with trylock */
675 if (!spin_trylock(&inode->i_lock))
676 continue;
677 if (!can_unuse(inode)) {
678 spin_unlock(&inode->i_lock);
679 continue;
680 }
681 }
682 WARN_ON(inode->i_state & I_NEW);
683 inode->i_state |= I_FREEING;
684 spin_unlock(&inode->i_lock);
685
686 list_move(&inode->i_lru, &freeable);
687 sb->s_nr_inodes_unused--;
688 this_cpu_dec(nr_unused);
689 }
690 if (current_is_kswapd())
691 __count_vm_events(KSWAPD_INODESTEAL, reap);
692 else
693 __count_vm_events(PGINODESTEAL, reap);
694 spin_unlock(&sb->s_inode_lru_lock);
695
696 dispose_list(&freeable);
697}
698
699static void __wait_on_freeing_inode(struct inode *inode);
700/*
701 * Called with the inode lock held.
702 */
703static struct inode *find_inode(struct super_block *sb,
704 struct hlist_head *head,
705 int (*test)(struct inode *, void *),
706 void *data)
707{
708 struct hlist_node *node;
709 struct inode *inode = NULL;
710
711repeat:
712 hlist_for_each_entry(inode, node, head, i_hash) {
713 spin_lock(&inode->i_lock);
714 if (inode->i_sb != sb) {
715 spin_unlock(&inode->i_lock);
716 continue;
717 }
718 if (!test(inode, data)) {
719 spin_unlock(&inode->i_lock);
720 continue;
721 }
722 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
723 __wait_on_freeing_inode(inode);
724 goto repeat;
725 }
726 __iget(inode);
727 spin_unlock(&inode->i_lock);
728 return inode;
729 }
730 return NULL;
731}
732
733/*
734 * find_inode_fast is the fast path version of find_inode, see the comment at
735 * iget_locked for details.
736 */
737static struct inode *find_inode_fast(struct super_block *sb,
738 struct hlist_head *head, unsigned long ino)
739{
740 struct hlist_node *node;
741 struct inode *inode = NULL;
742
743repeat:
744 hlist_for_each_entry(inode, node, head, i_hash) {
745 spin_lock(&inode->i_lock);
746 if (inode->i_ino != ino) {
747 spin_unlock(&inode->i_lock);
748 continue;
749 }
750 if (inode->i_sb != sb) {
751 spin_unlock(&inode->i_lock);
752 continue;
753 }
754 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
755 __wait_on_freeing_inode(inode);
756 goto repeat;
757 }
758 __iget(inode);
759 spin_unlock(&inode->i_lock);
760 return inode;
761 }
762 return NULL;
763}
764
765/*
766 * Each cpu owns a range of LAST_INO_BATCH numbers.
767 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
768 * to renew the exhausted range.
769 *
770 * This does not significantly increase overflow rate because every CPU can
771 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
772 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
773 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
774 * overflow rate by 2x, which does not seem too significant.
775 *
776 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
777 * error if st_ino won't fit in target struct field. Use 32bit counter
778 * here to attempt to avoid that.
779 */
780#define LAST_INO_BATCH 1024
781static DEFINE_PER_CPU(unsigned int, last_ino);
782
783unsigned int get_next_ino(void)
784{
785 unsigned int *p = &get_cpu_var(last_ino);
786 unsigned int res = *p;
787
788#ifdef CONFIG_SMP
789 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
790 static atomic_t shared_last_ino;
791 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
792
793 res = next - LAST_INO_BATCH;
794 }
795#endif
796
797 *p = ++res;
798 put_cpu_var(last_ino);
799 return res;
800}
801EXPORT_SYMBOL(get_next_ino);
802
803/**
804 * new_inode_pseudo - obtain an inode
805 * @sb: superblock
806 *
807 * Allocates a new inode for given superblock.
808 * Inode wont be chained in superblock s_inodes list
809 * This means :
810 * - fs can't be unmount
811 * - quotas, fsnotify, writeback can't work
812 */
813struct inode *new_inode_pseudo(struct super_block *sb)
814{
815 struct inode *inode = alloc_inode(sb);
816
817 if (inode) {
818 spin_lock(&inode->i_lock);
819 inode->i_state = 0;
820 spin_unlock(&inode->i_lock);
821 INIT_LIST_HEAD(&inode->i_sb_list);
822 }
823 return inode;
824}
825
826/**
827 * new_inode - obtain an inode
828 * @sb: superblock
829 *
830 * Allocates a new inode for given superblock. The default gfp_mask
831 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
832 * If HIGHMEM pages are unsuitable or it is known that pages allocated
833 * for the page cache are not reclaimable or migratable,
834 * mapping_set_gfp_mask() must be called with suitable flags on the
835 * newly created inode's mapping
836 *
837 */
838struct inode *new_inode(struct super_block *sb)
839{
840 struct inode *inode;
841
842 spin_lock_prefetch(&inode_sb_list_lock);
843
844 inode = new_inode_pseudo(sb);
845 if (inode)
846 inode_sb_list_add(inode);
847 return inode;
848}
849EXPORT_SYMBOL(new_inode);
850
851#ifdef CONFIG_DEBUG_LOCK_ALLOC
852void lockdep_annotate_inode_mutex_key(struct inode *inode)
853{
854 if (S_ISDIR(inode->i_mode)) {
855 struct file_system_type *type = inode->i_sb->s_type;
856
857 /* Set new key only if filesystem hasn't already changed it */
858 if (!lockdep_match_class(&inode->i_mutex,
859 &type->i_mutex_key)) {
860 /*
861 * ensure nobody is actually holding i_mutex
862 */
863 mutex_destroy(&inode->i_mutex);
864 mutex_init(&inode->i_mutex);
865 lockdep_set_class(&inode->i_mutex,
866 &type->i_mutex_dir_key);
867 }
868 }
869}
870EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
871#endif
872
873/**
874 * unlock_new_inode - clear the I_NEW state and wake up any waiters
875 * @inode: new inode to unlock
876 *
877 * Called when the inode is fully initialised to clear the new state of the
878 * inode and wake up anyone waiting for the inode to finish initialisation.
879 */
880void unlock_new_inode(struct inode *inode)
881{
882 lockdep_annotate_inode_mutex_key(inode);
883 spin_lock(&inode->i_lock);
884 WARN_ON(!(inode->i_state & I_NEW));
885 inode->i_state &= ~I_NEW;
886 wake_up_bit(&inode->i_state, __I_NEW);
887 spin_unlock(&inode->i_lock);
888}
889EXPORT_SYMBOL(unlock_new_inode);
890
891/**
892 * iget5_locked - obtain an inode from a mounted file system
893 * @sb: super block of file system
894 * @hashval: hash value (usually inode number) to get
895 * @test: callback used for comparisons between inodes
896 * @set: callback used to initialize a new struct inode
897 * @data: opaque data pointer to pass to @test and @set
898 *
899 * Search for the inode specified by @hashval and @data in the inode cache,
900 * and if present it is return it with an increased reference count. This is
901 * a generalized version of iget_locked() for file systems where the inode
902 * number is not sufficient for unique identification of an inode.
903 *
904 * If the inode is not in cache, allocate a new inode and return it locked,
905 * hashed, and with the I_NEW flag set. The file system gets to fill it in
906 * before unlocking it via unlock_new_inode().
907 *
908 * Note both @test and @set are called with the inode_hash_lock held, so can't
909 * sleep.
910 */
911struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
912 int (*test)(struct inode *, void *),
913 int (*set)(struct inode *, void *), void *data)
914{
915 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
916 struct inode *inode;
917
918 spin_lock(&inode_hash_lock);
919 inode = find_inode(sb, head, test, data);
920 spin_unlock(&inode_hash_lock);
921
922 if (inode) {
923 wait_on_inode(inode);
924 return inode;
925 }
926
927 inode = alloc_inode(sb);
928 if (inode) {
929 struct inode *old;
930
931 spin_lock(&inode_hash_lock);
932 /* We released the lock, so.. */
933 old = find_inode(sb, head, test, data);
934 if (!old) {
935 if (set(inode, data))
936 goto set_failed;
937
938 spin_lock(&inode->i_lock);
939 inode->i_state = I_NEW;
940 hlist_add_head(&inode->i_hash, head);
941 spin_unlock(&inode->i_lock);
942 inode_sb_list_add(inode);
943 spin_unlock(&inode_hash_lock);
944
945 /* Return the locked inode with I_NEW set, the
946 * caller is responsible for filling in the contents
947 */
948 return inode;
949 }
950
951 /*
952 * Uhhuh, somebody else created the same inode under
953 * us. Use the old inode instead of the one we just
954 * allocated.
955 */
956 spin_unlock(&inode_hash_lock);
957 destroy_inode(inode);
958 inode = old;
959 wait_on_inode(inode);
960 }
961 return inode;
962
963set_failed:
964 spin_unlock(&inode_hash_lock);
965 destroy_inode(inode);
966 return NULL;
967}
968EXPORT_SYMBOL(iget5_locked);
969
970/**
971 * iget_locked - obtain an inode from a mounted file system
972 * @sb: super block of file system
973 * @ino: inode number to get
974 *
975 * Search for the inode specified by @ino in the inode cache and if present
976 * return it with an increased reference count. This is for file systems
977 * where the inode number is sufficient for unique identification of an inode.
978 *
979 * If the inode is not in cache, allocate a new inode and return it locked,
980 * hashed, and with the I_NEW flag set. The file system gets to fill it in
981 * before unlocking it via unlock_new_inode().
982 */
983struct inode *iget_locked(struct super_block *sb, unsigned long ino)
984{
985 struct hlist_head *head = inode_hashtable + hash(sb, ino);
986 struct inode *inode;
987
988 spin_lock(&inode_hash_lock);
989 inode = find_inode_fast(sb, head, ino);
990 spin_unlock(&inode_hash_lock);
991 if (inode) {
992 wait_on_inode(inode);
993 return inode;
994 }
995
996 inode = alloc_inode(sb);
997 if (inode) {
998 struct inode *old;
999
1000 spin_lock(&inode_hash_lock);
1001 /* We released the lock, so.. */
1002 old = find_inode_fast(sb, head, ino);
1003 if (!old) {
1004 inode->i_ino = ino;
1005 spin_lock(&inode->i_lock);
1006 inode->i_state = I_NEW;
1007 hlist_add_head(&inode->i_hash, head);
1008 spin_unlock(&inode->i_lock);
1009 inode_sb_list_add(inode);
1010 spin_unlock(&inode_hash_lock);
1011
1012 /* Return the locked inode with I_NEW set, the
1013 * caller is responsible for filling in the contents
1014 */
1015 return inode;
1016 }
1017
1018 /*
1019 * Uhhuh, somebody else created the same inode under
1020 * us. Use the old inode instead of the one we just
1021 * allocated.
1022 */
1023 spin_unlock(&inode_hash_lock);
1024 destroy_inode(inode);
1025 inode = old;
1026 wait_on_inode(inode);
1027 }
1028 return inode;
1029}
1030EXPORT_SYMBOL(iget_locked);
1031
1032/*
1033 * search the inode cache for a matching inode number.
1034 * If we find one, then the inode number we are trying to
1035 * allocate is not unique and so we should not use it.
1036 *
1037 * Returns 1 if the inode number is unique, 0 if it is not.
1038 */
1039static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1040{
1041 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1042 struct hlist_node *node;
1043 struct inode *inode;
1044
1045 spin_lock(&inode_hash_lock);
1046 hlist_for_each_entry(inode, node, b, i_hash) {
1047 if (inode->i_ino == ino && inode->i_sb == sb) {
1048 spin_unlock(&inode_hash_lock);
1049 return 0;
1050 }
1051 }
1052 spin_unlock(&inode_hash_lock);
1053
1054 return 1;
1055}
1056
1057/**
1058 * iunique - get a unique inode number
1059 * @sb: superblock
1060 * @max_reserved: highest reserved inode number
1061 *
1062 * Obtain an inode number that is unique on the system for a given
1063 * superblock. This is used by file systems that have no natural
1064 * permanent inode numbering system. An inode number is returned that
1065 * is higher than the reserved limit but unique.
1066 *
1067 * BUGS:
1068 * With a large number of inodes live on the file system this function
1069 * currently becomes quite slow.
1070 */
1071ino_t iunique(struct super_block *sb, ino_t max_reserved)
1072{
1073 /*
1074 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1075 * error if st_ino won't fit in target struct field. Use 32bit counter
1076 * here to attempt to avoid that.
1077 */
1078 static DEFINE_SPINLOCK(iunique_lock);
1079 static unsigned int counter;
1080 ino_t res;
1081
1082 spin_lock(&iunique_lock);
1083 do {
1084 if (counter <= max_reserved)
1085 counter = max_reserved + 1;
1086 res = counter++;
1087 } while (!test_inode_iunique(sb, res));
1088 spin_unlock(&iunique_lock);
1089
1090 return res;
1091}
1092EXPORT_SYMBOL(iunique);
1093
1094struct inode *igrab(struct inode *inode)
1095{
1096 spin_lock(&inode->i_lock);
1097 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1098 __iget(inode);
1099 spin_unlock(&inode->i_lock);
1100 } else {
1101 spin_unlock(&inode->i_lock);
1102 /*
1103 * Handle the case where s_op->clear_inode is not been
1104 * called yet, and somebody is calling igrab
1105 * while the inode is getting freed.
1106 */
1107 inode = NULL;
1108 }
1109 return inode;
1110}
1111EXPORT_SYMBOL(igrab);
1112
1113/**
1114 * ilookup5_nowait - search for an inode in the inode cache
1115 * @sb: super block of file system to search
1116 * @hashval: hash value (usually inode number) to search for
1117 * @test: callback used for comparisons between inodes
1118 * @data: opaque data pointer to pass to @test
1119 *
1120 * Search for the inode specified by @hashval and @data in the inode cache.
1121 * If the inode is in the cache, the inode is returned with an incremented
1122 * reference count.
1123 *
1124 * Note: I_NEW is not waited upon so you have to be very careful what you do
1125 * with the returned inode. You probably should be using ilookup5() instead.
1126 *
1127 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1128 */
1129struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1130 int (*test)(struct inode *, void *), void *data)
1131{
1132 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1133 struct inode *inode;
1134
1135 spin_lock(&inode_hash_lock);
1136 inode = find_inode(sb, head, test, data);
1137 spin_unlock(&inode_hash_lock);
1138
1139 return inode;
1140}
1141EXPORT_SYMBOL(ilookup5_nowait);
1142
1143/**
1144 * ilookup5 - search for an inode in the inode cache
1145 * @sb: super block of file system to search
1146 * @hashval: hash value (usually inode number) to search for
1147 * @test: callback used for comparisons between inodes
1148 * @data: opaque data pointer to pass to @test
1149 *
1150 * Search for the inode specified by @hashval and @data in the inode cache,
1151 * and if the inode is in the cache, return the inode with an incremented
1152 * reference count. Waits on I_NEW before returning the inode.
1153 * returned with an incremented reference count.
1154 *
1155 * This is a generalized version of ilookup() for file systems where the
1156 * inode number is not sufficient for unique identification of an inode.
1157 *
1158 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1159 */
1160struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1161 int (*test)(struct inode *, void *), void *data)
1162{
1163 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1164
1165 if (inode)
1166 wait_on_inode(inode);
1167 return inode;
1168}
1169EXPORT_SYMBOL(ilookup5);
1170
1171/**
1172 * ilookup - search for an inode in the inode cache
1173 * @sb: super block of file system to search
1174 * @ino: inode number to search for
1175 *
1176 * Search for the inode @ino in the inode cache, and if the inode is in the
1177 * cache, the inode is returned with an incremented reference count.
1178 */
1179struct inode *ilookup(struct super_block *sb, unsigned long ino)
1180{
1181 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1182 struct inode *inode;
1183
1184 spin_lock(&inode_hash_lock);
1185 inode = find_inode_fast(sb, head, ino);
1186 spin_unlock(&inode_hash_lock);
1187
1188 if (inode)
1189 wait_on_inode(inode);
1190 return inode;
1191}
1192EXPORT_SYMBOL(ilookup);
1193
1194int insert_inode_locked(struct inode *inode)
1195{
1196 struct super_block *sb = inode->i_sb;
1197 ino_t ino = inode->i_ino;
1198 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1199
1200 while (1) {
1201 struct hlist_node *node;
1202 struct inode *old = NULL;
1203 spin_lock(&inode_hash_lock);
1204 hlist_for_each_entry(old, node, head, i_hash) {
1205 if (old->i_ino != ino)
1206 continue;
1207 if (old->i_sb != sb)
1208 continue;
1209 spin_lock(&old->i_lock);
1210 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1211 spin_unlock(&old->i_lock);
1212 continue;
1213 }
1214 break;
1215 }
1216 if (likely(!node)) {
1217 spin_lock(&inode->i_lock);
1218 inode->i_state |= I_NEW;
1219 hlist_add_head(&inode->i_hash, head);
1220 spin_unlock(&inode->i_lock);
1221 spin_unlock(&inode_hash_lock);
1222 return 0;
1223 }
1224 __iget(old);
1225 spin_unlock(&old->i_lock);
1226 spin_unlock(&inode_hash_lock);
1227 wait_on_inode(old);
1228 if (unlikely(!inode_unhashed(old))) {
1229 iput(old);
1230 return -EBUSY;
1231 }
1232 iput(old);
1233 }
1234}
1235EXPORT_SYMBOL(insert_inode_locked);
1236
1237int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1238 int (*test)(struct inode *, void *), void *data)
1239{
1240 struct super_block *sb = inode->i_sb;
1241 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1242
1243 while (1) {
1244 struct hlist_node *node;
1245 struct inode *old = NULL;
1246
1247 spin_lock(&inode_hash_lock);
1248 hlist_for_each_entry(old, node, head, i_hash) {
1249 if (old->i_sb != sb)
1250 continue;
1251 if (!test(old, data))
1252 continue;
1253 spin_lock(&old->i_lock);
1254 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1255 spin_unlock(&old->i_lock);
1256 continue;
1257 }
1258 break;
1259 }
1260 if (likely(!node)) {
1261 spin_lock(&inode->i_lock);
1262 inode->i_state |= I_NEW;
1263 hlist_add_head(&inode->i_hash, head);
1264 spin_unlock(&inode->i_lock);
1265 spin_unlock(&inode_hash_lock);
1266 return 0;
1267 }
1268 __iget(old);
1269 spin_unlock(&old->i_lock);
1270 spin_unlock(&inode_hash_lock);
1271 wait_on_inode(old);
1272 if (unlikely(!inode_unhashed(old))) {
1273 iput(old);
1274 return -EBUSY;
1275 }
1276 iput(old);
1277 }
1278}
1279EXPORT_SYMBOL(insert_inode_locked4);
1280
1281
1282int generic_delete_inode(struct inode *inode)
1283{
1284 return 1;
1285}
1286EXPORT_SYMBOL(generic_delete_inode);
1287
1288/*
1289 * Normal UNIX filesystem behaviour: delete the
1290 * inode when the usage count drops to zero, and
1291 * i_nlink is zero.
1292 */
1293int generic_drop_inode(struct inode *inode)
1294{
1295 return !inode->i_nlink || inode_unhashed(inode);
1296}
1297EXPORT_SYMBOL_GPL(generic_drop_inode);
1298
1299/*
1300 * Called when we're dropping the last reference
1301 * to an inode.
1302 *
1303 * Call the FS "drop_inode()" function, defaulting to
1304 * the legacy UNIX filesystem behaviour. If it tells
1305 * us to evict inode, do so. Otherwise, retain inode
1306 * in cache if fs is alive, sync and evict if fs is
1307 * shutting down.
1308 */
1309static void iput_final(struct inode *inode)
1310{
1311 struct super_block *sb = inode->i_sb;
1312 const struct super_operations *op = inode->i_sb->s_op;
1313 int drop;
1314
1315 WARN_ON(inode->i_state & I_NEW);
1316
1317 if (op->drop_inode)
1318 drop = op->drop_inode(inode);
1319 else
1320 drop = generic_drop_inode(inode);
1321
1322 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1323 inode->i_state |= I_REFERENCED;
1324 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1325 inode_lru_list_add(inode);
1326 spin_unlock(&inode->i_lock);
1327 return;
1328 }
1329
1330 if (!drop) {
1331 inode->i_state |= I_WILL_FREE;
1332 spin_unlock(&inode->i_lock);
1333 write_inode_now(inode, 1);
1334 spin_lock(&inode->i_lock);
1335 WARN_ON(inode->i_state & I_NEW);
1336 inode->i_state &= ~I_WILL_FREE;
1337 }
1338
1339 inode->i_state |= I_FREEING;
1340 if (!list_empty(&inode->i_lru))
1341 inode_lru_list_del(inode);
1342 spin_unlock(&inode->i_lock);
1343
1344 evict(inode);
1345}
1346
1347/**
1348 * iput - put an inode
1349 * @inode: inode to put
1350 *
1351 * Puts an inode, dropping its usage count. If the inode use count hits
1352 * zero, the inode is then freed and may also be destroyed.
1353 *
1354 * Consequently, iput() can sleep.
1355 */
1356void iput(struct inode *inode)
1357{
1358 if (inode) {
1359 BUG_ON(inode->i_state & I_CLEAR);
1360
1361 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1362 iput_final(inode);
1363 }
1364}
1365EXPORT_SYMBOL(iput);
1366
1367/**
1368 * bmap - find a block number in a file
1369 * @inode: inode of file
1370 * @block: block to find
1371 *
1372 * Returns the block number on the device holding the inode that
1373 * is the disk block number for the block of the file requested.
1374 * That is, asked for block 4 of inode 1 the function will return the
1375 * disk block relative to the disk start that holds that block of the
1376 * file.
1377 */
1378sector_t bmap(struct inode *inode, sector_t block)
1379{
1380 sector_t res = 0;
1381 if (inode->i_mapping->a_ops->bmap)
1382 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1383 return res;
1384}
1385EXPORT_SYMBOL(bmap);
1386
1387/*
1388 * With relative atime, only update atime if the previous atime is
1389 * earlier than either the ctime or mtime or if at least a day has
1390 * passed since the last atime update.
1391 */
1392static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1393 struct timespec now)
1394{
1395
1396 if (!(mnt->mnt_flags & MNT_RELATIME))
1397 return 1;
1398 /*
1399 * Is mtime younger than atime? If yes, update atime:
1400 */
1401 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1402 return 1;
1403 /*
1404 * Is ctime younger than atime? If yes, update atime:
1405 */
1406 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1407 return 1;
1408
1409 /*
1410 * Is the previous atime value older than a day? If yes,
1411 * update atime:
1412 */
1413 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1414 return 1;
1415 /*
1416 * Good, we can skip the atime update:
1417 */
1418 return 0;
1419}
1420
1421/**
1422 * touch_atime - update the access time
1423 * @mnt: mount the inode is accessed on
1424 * @dentry: dentry accessed
1425 *
1426 * Update the accessed time on an inode and mark it for writeback.
1427 * This function automatically handles read only file systems and media,
1428 * as well as the "noatime" flag and inode specific "noatime" markers.
1429 */
1430void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1431{
1432 struct inode *inode = dentry->d_inode;
1433 struct timespec now;
1434
1435 if (inode->i_flags & S_NOATIME)
1436 return;
1437 if (IS_NOATIME(inode))
1438 return;
1439 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1440 return;
1441
1442 if (mnt->mnt_flags & MNT_NOATIME)
1443 return;
1444 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1445 return;
1446
1447 now = current_fs_time(inode->i_sb);
1448
1449 if (!relatime_need_update(mnt, inode, now))
1450 return;
1451
1452 if (timespec_equal(&inode->i_atime, &now))
1453 return;
1454
1455 if (mnt_want_write(mnt))
1456 return;
1457
1458 inode->i_atime = now;
1459 mark_inode_dirty_sync(inode);
1460 mnt_drop_write(mnt);
1461}
1462EXPORT_SYMBOL(touch_atime);
1463
1464/**
1465 * file_update_time - update mtime and ctime time
1466 * @file: file accessed
1467 *
1468 * Update the mtime and ctime members of an inode and mark the inode
1469 * for writeback. Note that this function is meant exclusively for
1470 * usage in the file write path of filesystems, and filesystems may
1471 * choose to explicitly ignore update via this function with the
1472 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1473 * timestamps are handled by the server.
1474 */
1475
1476void file_update_time(struct file *file)
1477{
1478 struct inode *inode = file->f_path.dentry->d_inode;
1479 struct timespec now;
1480 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1481
1482 /* First try to exhaust all avenues to not sync */
1483 if (IS_NOCMTIME(inode))
1484 return;
1485
1486 now = current_fs_time(inode->i_sb);
1487 if (!timespec_equal(&inode->i_mtime, &now))
1488 sync_it = S_MTIME;
1489
1490 if (!timespec_equal(&inode->i_ctime, &now))
1491 sync_it |= S_CTIME;
1492
1493 if (IS_I_VERSION(inode))
1494 sync_it |= S_VERSION;
1495
1496 if (!sync_it)
1497 return;
1498
1499 /* Finally allowed to write? Takes lock. */
1500 if (mnt_want_write_file(file))
1501 return;
1502
1503 /* Only change inode inside the lock region */
1504 if (sync_it & S_VERSION)
1505 inode_inc_iversion(inode);
1506 if (sync_it & S_CTIME)
1507 inode->i_ctime = now;
1508 if (sync_it & S_MTIME)
1509 inode->i_mtime = now;
1510 mark_inode_dirty_sync(inode);
1511 mnt_drop_write(file->f_path.mnt);
1512}
1513EXPORT_SYMBOL(file_update_time);
1514
1515int inode_needs_sync(struct inode *inode)
1516{
1517 if (IS_SYNC(inode))
1518 return 1;
1519 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1520 return 1;
1521 return 0;
1522}
1523EXPORT_SYMBOL(inode_needs_sync);
1524
1525int inode_wait(void *word)
1526{
1527 schedule();
1528 return 0;
1529}
1530EXPORT_SYMBOL(inode_wait);
1531
1532/*
1533 * If we try to find an inode in the inode hash while it is being
1534 * deleted, we have to wait until the filesystem completes its
1535 * deletion before reporting that it isn't found. This function waits
1536 * until the deletion _might_ have completed. Callers are responsible
1537 * to recheck inode state.
1538 *
1539 * It doesn't matter if I_NEW is not set initially, a call to
1540 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1541 * will DTRT.
1542 */
1543static void __wait_on_freeing_inode(struct inode *inode)
1544{
1545 wait_queue_head_t *wq;
1546 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1547 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1548 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1549 spin_unlock(&inode->i_lock);
1550 spin_unlock(&inode_hash_lock);
1551 schedule();
1552 finish_wait(wq, &wait.wait);
1553 spin_lock(&inode_hash_lock);
1554}
1555
1556static __initdata unsigned long ihash_entries;
1557static int __init set_ihash_entries(char *str)
1558{
1559 if (!str)
1560 return 0;
1561 ihash_entries = simple_strtoul(str, &str, 0);
1562 return 1;
1563}
1564__setup("ihash_entries=", set_ihash_entries);
1565
1566/*
1567 * Initialize the waitqueues and inode hash table.
1568 */
1569void __init inode_init_early(void)
1570{
1571 int loop;
1572
1573 /* If hashes are distributed across NUMA nodes, defer
1574 * hash allocation until vmalloc space is available.
1575 */
1576 if (hashdist)
1577 return;
1578
1579 inode_hashtable =
1580 alloc_large_system_hash("Inode-cache",
1581 sizeof(struct hlist_head),
1582 ihash_entries,
1583 14,
1584 HASH_EARLY,
1585 &i_hash_shift,
1586 &i_hash_mask,
1587 0);
1588
1589 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1590 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1591}
1592
1593void __init inode_init(void)
1594{
1595 int loop;
1596
1597 /* inode slab cache */
1598 inode_cachep = kmem_cache_create("inode_cache",
1599 sizeof(struct inode),
1600 0,
1601 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1602 SLAB_MEM_SPREAD),
1603 init_once);
1604
1605 /* Hash may have been set up in inode_init_early */
1606 if (!hashdist)
1607 return;
1608
1609 inode_hashtable =
1610 alloc_large_system_hash("Inode-cache",
1611 sizeof(struct hlist_head),
1612 ihash_entries,
1613 14,
1614 0,
1615 &i_hash_shift,
1616 &i_hash_mask,
1617 0);
1618
1619 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1620 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1621}
1622
1623void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1624{
1625 inode->i_mode = mode;
1626 if (S_ISCHR(mode)) {
1627 inode->i_fop = &def_chr_fops;
1628 inode->i_rdev = rdev;
1629 } else if (S_ISBLK(mode)) {
1630 inode->i_fop = &def_blk_fops;
1631 inode->i_rdev = rdev;
1632 } else if (S_ISFIFO(mode))
1633 inode->i_fop = &def_fifo_fops;
1634 else if (S_ISSOCK(mode))
1635 inode->i_fop = &bad_sock_fops;
1636 else
1637 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1638 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1639 inode->i_ino);
1640}
1641EXPORT_SYMBOL(init_special_inode);
1642
1643/**
1644 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1645 * @inode: New inode
1646 * @dir: Directory inode
1647 * @mode: mode of the new inode
1648 */
1649void inode_init_owner(struct inode *inode, const struct inode *dir,
1650 mode_t mode)
1651{
1652 inode->i_uid = current_fsuid();
1653 if (dir && dir->i_mode & S_ISGID) {
1654 inode->i_gid = dir->i_gid;
1655 if (S_ISDIR(mode))
1656 mode |= S_ISGID;
1657 } else
1658 inode->i_gid = current_fsgid();
1659 inode->i_mode = mode;
1660}
1661EXPORT_SYMBOL(inode_init_owner);
1662
1663/**
1664 * inode_owner_or_capable - check current task permissions to inode
1665 * @inode: inode being checked
1666 *
1667 * Return true if current either has CAP_FOWNER to the inode, or
1668 * owns the file.
1669 */
1670bool inode_owner_or_capable(const struct inode *inode)
1671{
1672 struct user_namespace *ns = inode_userns(inode);
1673
1674 if (current_user_ns() == ns && current_fsuid() == inode->i_uid)
1675 return true;
1676 if (ns_capable(ns, CAP_FOWNER))
1677 return true;
1678 return false;
1679}
1680EXPORT_SYMBOL(inode_owner_or_capable);
1/*
2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
4 */
5#include <linux/export.h>
6#include <linux/fs.h>
7#include <linux/mm.h>
8#include <linux/backing-dev.h>
9#include <linux/hash.h>
10#include <linux/swap.h>
11#include <linux/security.h>
12#include <linux/cdev.h>
13#include <linux/bootmem.h>
14#include <linux/fsnotify.h>
15#include <linux/mount.h>
16#include <linux/posix_acl.h>
17#include <linux/prefetch.h>
18#include <linux/buffer_head.h> /* for inode_has_buffers */
19#include <linux/ratelimit.h>
20#include <linux/list_lru.h>
21#include "internal.h"
22
23/*
24 * Inode locking rules:
25 *
26 * inode->i_lock protects:
27 * inode->i_state, inode->i_hash, __iget()
28 * Inode LRU list locks protect:
29 * inode->i_sb->s_inode_lru, inode->i_lru
30 * inode_sb_list_lock protects:
31 * sb->s_inodes, inode->i_sb_list
32 * bdi->wb.list_lock protects:
33 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
34 * inode_hash_lock protects:
35 * inode_hashtable, inode->i_hash
36 *
37 * Lock ordering:
38 *
39 * inode_sb_list_lock
40 * inode->i_lock
41 * Inode LRU list locks
42 *
43 * bdi->wb.list_lock
44 * inode->i_lock
45 *
46 * inode_hash_lock
47 * inode_sb_list_lock
48 * inode->i_lock
49 *
50 * iunique_lock
51 * inode_hash_lock
52 */
53
54static unsigned int i_hash_mask __read_mostly;
55static unsigned int i_hash_shift __read_mostly;
56static struct hlist_head *inode_hashtable __read_mostly;
57static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
58
59__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
60
61/*
62 * Empty aops. Can be used for the cases where the user does not
63 * define any of the address_space operations.
64 */
65const struct address_space_operations empty_aops = {
66};
67EXPORT_SYMBOL(empty_aops);
68
69/*
70 * Statistics gathering..
71 */
72struct inodes_stat_t inodes_stat;
73
74static DEFINE_PER_CPU(unsigned long, nr_inodes);
75static DEFINE_PER_CPU(unsigned long, nr_unused);
76
77static struct kmem_cache *inode_cachep __read_mostly;
78
79static long get_nr_inodes(void)
80{
81 int i;
82 long sum = 0;
83 for_each_possible_cpu(i)
84 sum += per_cpu(nr_inodes, i);
85 return sum < 0 ? 0 : sum;
86}
87
88static inline long get_nr_inodes_unused(void)
89{
90 int i;
91 long sum = 0;
92 for_each_possible_cpu(i)
93 sum += per_cpu(nr_unused, i);
94 return sum < 0 ? 0 : sum;
95}
96
97long get_nr_dirty_inodes(void)
98{
99 /* not actually dirty inodes, but a wild approximation */
100 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
101 return nr_dirty > 0 ? nr_dirty : 0;
102}
103
104/*
105 * Handle nr_inode sysctl
106 */
107#ifdef CONFIG_SYSCTL
108int proc_nr_inodes(ctl_table *table, int write,
109 void __user *buffer, size_t *lenp, loff_t *ppos)
110{
111 inodes_stat.nr_inodes = get_nr_inodes();
112 inodes_stat.nr_unused = get_nr_inodes_unused();
113 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
114}
115#endif
116
117/**
118 * inode_init_always - perform inode structure intialisation
119 * @sb: superblock inode belongs to
120 * @inode: inode to initialise
121 *
122 * These are initializations that need to be done on every inode
123 * allocation as the fields are not initialised by slab allocation.
124 */
125int inode_init_always(struct super_block *sb, struct inode *inode)
126{
127 static const struct inode_operations empty_iops;
128 static const struct file_operations empty_fops;
129 struct address_space *const mapping = &inode->i_data;
130
131 inode->i_sb = sb;
132 inode->i_blkbits = sb->s_blocksize_bits;
133 inode->i_flags = 0;
134 atomic_set(&inode->i_count, 1);
135 inode->i_op = &empty_iops;
136 inode->i_fop = &empty_fops;
137 inode->__i_nlink = 1;
138 inode->i_opflags = 0;
139 i_uid_write(inode, 0);
140 i_gid_write(inode, 0);
141 atomic_set(&inode->i_writecount, 0);
142 inode->i_size = 0;
143 inode->i_blocks = 0;
144 inode->i_bytes = 0;
145 inode->i_generation = 0;
146#ifdef CONFIG_QUOTA
147 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
148#endif
149 inode->i_pipe = NULL;
150 inode->i_bdev = NULL;
151 inode->i_cdev = NULL;
152 inode->i_rdev = 0;
153 inode->dirtied_when = 0;
154
155 if (security_inode_alloc(inode))
156 goto out;
157 spin_lock_init(&inode->i_lock);
158 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
159
160 mutex_init(&inode->i_mutex);
161 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
162
163 atomic_set(&inode->i_dio_count, 0);
164
165 mapping->a_ops = &empty_aops;
166 mapping->host = inode;
167 mapping->flags = 0;
168 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
169 mapping->private_data = NULL;
170 mapping->backing_dev_info = &default_backing_dev_info;
171 mapping->writeback_index = 0;
172
173 /*
174 * If the block_device provides a backing_dev_info for client
175 * inodes then use that. Otherwise the inode share the bdev's
176 * backing_dev_info.
177 */
178 if (sb->s_bdev) {
179 struct backing_dev_info *bdi;
180
181 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
182 mapping->backing_dev_info = bdi;
183 }
184 inode->i_private = NULL;
185 inode->i_mapping = mapping;
186 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
187#ifdef CONFIG_FS_POSIX_ACL
188 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
189#endif
190
191#ifdef CONFIG_FSNOTIFY
192 inode->i_fsnotify_mask = 0;
193#endif
194
195 this_cpu_inc(nr_inodes);
196
197 return 0;
198out:
199 return -ENOMEM;
200}
201EXPORT_SYMBOL(inode_init_always);
202
203static struct inode *alloc_inode(struct super_block *sb)
204{
205 struct inode *inode;
206
207 if (sb->s_op->alloc_inode)
208 inode = sb->s_op->alloc_inode(sb);
209 else
210 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
211
212 if (!inode)
213 return NULL;
214
215 if (unlikely(inode_init_always(sb, inode))) {
216 if (inode->i_sb->s_op->destroy_inode)
217 inode->i_sb->s_op->destroy_inode(inode);
218 else
219 kmem_cache_free(inode_cachep, inode);
220 return NULL;
221 }
222
223 return inode;
224}
225
226void free_inode_nonrcu(struct inode *inode)
227{
228 kmem_cache_free(inode_cachep, inode);
229}
230EXPORT_SYMBOL(free_inode_nonrcu);
231
232void __destroy_inode(struct inode *inode)
233{
234 BUG_ON(inode_has_buffers(inode));
235 security_inode_free(inode);
236 fsnotify_inode_delete(inode);
237 if (!inode->i_nlink) {
238 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
239 atomic_long_dec(&inode->i_sb->s_remove_count);
240 }
241
242#ifdef CONFIG_FS_POSIX_ACL
243 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
244 posix_acl_release(inode->i_acl);
245 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
246 posix_acl_release(inode->i_default_acl);
247#endif
248 this_cpu_dec(nr_inodes);
249}
250EXPORT_SYMBOL(__destroy_inode);
251
252static void i_callback(struct rcu_head *head)
253{
254 struct inode *inode = container_of(head, struct inode, i_rcu);
255 kmem_cache_free(inode_cachep, inode);
256}
257
258static void destroy_inode(struct inode *inode)
259{
260 BUG_ON(!list_empty(&inode->i_lru));
261 __destroy_inode(inode);
262 if (inode->i_sb->s_op->destroy_inode)
263 inode->i_sb->s_op->destroy_inode(inode);
264 else
265 call_rcu(&inode->i_rcu, i_callback);
266}
267
268/**
269 * drop_nlink - directly drop an inode's link count
270 * @inode: inode
271 *
272 * This is a low-level filesystem helper to replace any
273 * direct filesystem manipulation of i_nlink. In cases
274 * where we are attempting to track writes to the
275 * filesystem, a decrement to zero means an imminent
276 * write when the file is truncated and actually unlinked
277 * on the filesystem.
278 */
279void drop_nlink(struct inode *inode)
280{
281 WARN_ON(inode->i_nlink == 0);
282 inode->__i_nlink--;
283 if (!inode->i_nlink)
284 atomic_long_inc(&inode->i_sb->s_remove_count);
285}
286EXPORT_SYMBOL(drop_nlink);
287
288/**
289 * clear_nlink - directly zero an inode's link count
290 * @inode: inode
291 *
292 * This is a low-level filesystem helper to replace any
293 * direct filesystem manipulation of i_nlink. See
294 * drop_nlink() for why we care about i_nlink hitting zero.
295 */
296void clear_nlink(struct inode *inode)
297{
298 if (inode->i_nlink) {
299 inode->__i_nlink = 0;
300 atomic_long_inc(&inode->i_sb->s_remove_count);
301 }
302}
303EXPORT_SYMBOL(clear_nlink);
304
305/**
306 * set_nlink - directly set an inode's link count
307 * @inode: inode
308 * @nlink: new nlink (should be non-zero)
309 *
310 * This is a low-level filesystem helper to replace any
311 * direct filesystem manipulation of i_nlink.
312 */
313void set_nlink(struct inode *inode, unsigned int nlink)
314{
315 if (!nlink) {
316 clear_nlink(inode);
317 } else {
318 /* Yes, some filesystems do change nlink from zero to one */
319 if (inode->i_nlink == 0)
320 atomic_long_dec(&inode->i_sb->s_remove_count);
321
322 inode->__i_nlink = nlink;
323 }
324}
325EXPORT_SYMBOL(set_nlink);
326
327/**
328 * inc_nlink - directly increment an inode's link count
329 * @inode: inode
330 *
331 * This is a low-level filesystem helper to replace any
332 * direct filesystem manipulation of i_nlink. Currently,
333 * it is only here for parity with dec_nlink().
334 */
335void inc_nlink(struct inode *inode)
336{
337 if (unlikely(inode->i_nlink == 0)) {
338 WARN_ON(!(inode->i_state & I_LINKABLE));
339 atomic_long_dec(&inode->i_sb->s_remove_count);
340 }
341
342 inode->__i_nlink++;
343}
344EXPORT_SYMBOL(inc_nlink);
345
346void address_space_init_once(struct address_space *mapping)
347{
348 memset(mapping, 0, sizeof(*mapping));
349 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
350 spin_lock_init(&mapping->tree_lock);
351 mutex_init(&mapping->i_mmap_mutex);
352 INIT_LIST_HEAD(&mapping->private_list);
353 spin_lock_init(&mapping->private_lock);
354 mapping->i_mmap = RB_ROOT;
355 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
356}
357EXPORT_SYMBOL(address_space_init_once);
358
359/*
360 * These are initializations that only need to be done
361 * once, because the fields are idempotent across use
362 * of the inode, so let the slab aware of that.
363 */
364void inode_init_once(struct inode *inode)
365{
366 memset(inode, 0, sizeof(*inode));
367 INIT_HLIST_NODE(&inode->i_hash);
368 INIT_LIST_HEAD(&inode->i_devices);
369 INIT_LIST_HEAD(&inode->i_wb_list);
370 INIT_LIST_HEAD(&inode->i_lru);
371 address_space_init_once(&inode->i_data);
372 i_size_ordered_init(inode);
373#ifdef CONFIG_FSNOTIFY
374 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
375#endif
376}
377EXPORT_SYMBOL(inode_init_once);
378
379static void init_once(void *foo)
380{
381 struct inode *inode = (struct inode *) foo;
382
383 inode_init_once(inode);
384}
385
386/*
387 * inode->i_lock must be held
388 */
389void __iget(struct inode *inode)
390{
391 atomic_inc(&inode->i_count);
392}
393
394/*
395 * get additional reference to inode; caller must already hold one.
396 */
397void ihold(struct inode *inode)
398{
399 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
400}
401EXPORT_SYMBOL(ihold);
402
403static void inode_lru_list_add(struct inode *inode)
404{
405 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
406 this_cpu_inc(nr_unused);
407}
408
409/*
410 * Add inode to LRU if needed (inode is unused and clean).
411 *
412 * Needs inode->i_lock held.
413 */
414void inode_add_lru(struct inode *inode)
415{
416 if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) &&
417 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
418 inode_lru_list_add(inode);
419}
420
421
422static void inode_lru_list_del(struct inode *inode)
423{
424
425 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
426 this_cpu_dec(nr_unused);
427}
428
429/**
430 * inode_sb_list_add - add inode to the superblock list of inodes
431 * @inode: inode to add
432 */
433void inode_sb_list_add(struct inode *inode)
434{
435 spin_lock(&inode_sb_list_lock);
436 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
437 spin_unlock(&inode_sb_list_lock);
438}
439EXPORT_SYMBOL_GPL(inode_sb_list_add);
440
441static inline void inode_sb_list_del(struct inode *inode)
442{
443 if (!list_empty(&inode->i_sb_list)) {
444 spin_lock(&inode_sb_list_lock);
445 list_del_init(&inode->i_sb_list);
446 spin_unlock(&inode_sb_list_lock);
447 }
448}
449
450static unsigned long hash(struct super_block *sb, unsigned long hashval)
451{
452 unsigned long tmp;
453
454 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
455 L1_CACHE_BYTES;
456 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
457 return tmp & i_hash_mask;
458}
459
460/**
461 * __insert_inode_hash - hash an inode
462 * @inode: unhashed inode
463 * @hashval: unsigned long value used to locate this object in the
464 * inode_hashtable.
465 *
466 * Add an inode to the inode hash for this superblock.
467 */
468void __insert_inode_hash(struct inode *inode, unsigned long hashval)
469{
470 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
471
472 spin_lock(&inode_hash_lock);
473 spin_lock(&inode->i_lock);
474 hlist_add_head(&inode->i_hash, b);
475 spin_unlock(&inode->i_lock);
476 spin_unlock(&inode_hash_lock);
477}
478EXPORT_SYMBOL(__insert_inode_hash);
479
480/**
481 * __remove_inode_hash - remove an inode from the hash
482 * @inode: inode to unhash
483 *
484 * Remove an inode from the superblock.
485 */
486void __remove_inode_hash(struct inode *inode)
487{
488 spin_lock(&inode_hash_lock);
489 spin_lock(&inode->i_lock);
490 hlist_del_init(&inode->i_hash);
491 spin_unlock(&inode->i_lock);
492 spin_unlock(&inode_hash_lock);
493}
494EXPORT_SYMBOL(__remove_inode_hash);
495
496void clear_inode(struct inode *inode)
497{
498 might_sleep();
499 /*
500 * We have to cycle tree_lock here because reclaim can be still in the
501 * process of removing the last page (in __delete_from_page_cache())
502 * and we must not free mapping under it.
503 */
504 spin_lock_irq(&inode->i_data.tree_lock);
505 BUG_ON(inode->i_data.nrpages);
506 BUG_ON(inode->i_data.nrshadows);
507 spin_unlock_irq(&inode->i_data.tree_lock);
508 BUG_ON(!list_empty(&inode->i_data.private_list));
509 BUG_ON(!(inode->i_state & I_FREEING));
510 BUG_ON(inode->i_state & I_CLEAR);
511 /* don't need i_lock here, no concurrent mods to i_state */
512 inode->i_state = I_FREEING | I_CLEAR;
513}
514EXPORT_SYMBOL(clear_inode);
515
516/*
517 * Free the inode passed in, removing it from the lists it is still connected
518 * to. We remove any pages still attached to the inode and wait for any IO that
519 * is still in progress before finally destroying the inode.
520 *
521 * An inode must already be marked I_FREEING so that we avoid the inode being
522 * moved back onto lists if we race with other code that manipulates the lists
523 * (e.g. writeback_single_inode). The caller is responsible for setting this.
524 *
525 * An inode must already be removed from the LRU list before being evicted from
526 * the cache. This should occur atomically with setting the I_FREEING state
527 * flag, so no inodes here should ever be on the LRU when being evicted.
528 */
529static void evict(struct inode *inode)
530{
531 const struct super_operations *op = inode->i_sb->s_op;
532
533 BUG_ON(!(inode->i_state & I_FREEING));
534 BUG_ON(!list_empty(&inode->i_lru));
535
536 if (!list_empty(&inode->i_wb_list))
537 inode_wb_list_del(inode);
538
539 inode_sb_list_del(inode);
540
541 /*
542 * Wait for flusher thread to be done with the inode so that filesystem
543 * does not start destroying it while writeback is still running. Since
544 * the inode has I_FREEING set, flusher thread won't start new work on
545 * the inode. We just have to wait for running writeback to finish.
546 */
547 inode_wait_for_writeback(inode);
548
549 if (op->evict_inode) {
550 op->evict_inode(inode);
551 } else {
552 truncate_inode_pages_final(&inode->i_data);
553 clear_inode(inode);
554 }
555 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
556 bd_forget(inode);
557 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
558 cd_forget(inode);
559
560 remove_inode_hash(inode);
561
562 spin_lock(&inode->i_lock);
563 wake_up_bit(&inode->i_state, __I_NEW);
564 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
565 spin_unlock(&inode->i_lock);
566
567 destroy_inode(inode);
568}
569
570/*
571 * dispose_list - dispose of the contents of a local list
572 * @head: the head of the list to free
573 *
574 * Dispose-list gets a local list with local inodes in it, so it doesn't
575 * need to worry about list corruption and SMP locks.
576 */
577static void dispose_list(struct list_head *head)
578{
579 while (!list_empty(head)) {
580 struct inode *inode;
581
582 inode = list_first_entry(head, struct inode, i_lru);
583 list_del_init(&inode->i_lru);
584
585 evict(inode);
586 }
587}
588
589/**
590 * evict_inodes - evict all evictable inodes for a superblock
591 * @sb: superblock to operate on
592 *
593 * Make sure that no inodes with zero refcount are retained. This is
594 * called by superblock shutdown after having MS_ACTIVE flag removed,
595 * so any inode reaching zero refcount during or after that call will
596 * be immediately evicted.
597 */
598void evict_inodes(struct super_block *sb)
599{
600 struct inode *inode, *next;
601 LIST_HEAD(dispose);
602
603 spin_lock(&inode_sb_list_lock);
604 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
605 if (atomic_read(&inode->i_count))
606 continue;
607
608 spin_lock(&inode->i_lock);
609 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
610 spin_unlock(&inode->i_lock);
611 continue;
612 }
613
614 inode->i_state |= I_FREEING;
615 inode_lru_list_del(inode);
616 spin_unlock(&inode->i_lock);
617 list_add(&inode->i_lru, &dispose);
618 }
619 spin_unlock(&inode_sb_list_lock);
620
621 dispose_list(&dispose);
622}
623
624/**
625 * invalidate_inodes - attempt to free all inodes on a superblock
626 * @sb: superblock to operate on
627 * @kill_dirty: flag to guide handling of dirty inodes
628 *
629 * Attempts to free all inodes for a given superblock. If there were any
630 * busy inodes return a non-zero value, else zero.
631 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
632 * them as busy.
633 */
634int invalidate_inodes(struct super_block *sb, bool kill_dirty)
635{
636 int busy = 0;
637 struct inode *inode, *next;
638 LIST_HEAD(dispose);
639
640 spin_lock(&inode_sb_list_lock);
641 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
642 spin_lock(&inode->i_lock);
643 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
644 spin_unlock(&inode->i_lock);
645 continue;
646 }
647 if (inode->i_state & I_DIRTY && !kill_dirty) {
648 spin_unlock(&inode->i_lock);
649 busy = 1;
650 continue;
651 }
652 if (atomic_read(&inode->i_count)) {
653 spin_unlock(&inode->i_lock);
654 busy = 1;
655 continue;
656 }
657
658 inode->i_state |= I_FREEING;
659 inode_lru_list_del(inode);
660 spin_unlock(&inode->i_lock);
661 list_add(&inode->i_lru, &dispose);
662 }
663 spin_unlock(&inode_sb_list_lock);
664
665 dispose_list(&dispose);
666
667 return busy;
668}
669
670/*
671 * Isolate the inode from the LRU in preparation for freeing it.
672 *
673 * Any inodes which are pinned purely because of attached pagecache have their
674 * pagecache removed. If the inode has metadata buffers attached to
675 * mapping->private_list then try to remove them.
676 *
677 * If the inode has the I_REFERENCED flag set, then it means that it has been
678 * used recently - the flag is set in iput_final(). When we encounter such an
679 * inode, clear the flag and move it to the back of the LRU so it gets another
680 * pass through the LRU before it gets reclaimed. This is necessary because of
681 * the fact we are doing lazy LRU updates to minimise lock contention so the
682 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
683 * with this flag set because they are the inodes that are out of order.
684 */
685static enum lru_status
686inode_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
687{
688 struct list_head *freeable = arg;
689 struct inode *inode = container_of(item, struct inode, i_lru);
690
691 /*
692 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
693 * If we fail to get the lock, just skip it.
694 */
695 if (!spin_trylock(&inode->i_lock))
696 return LRU_SKIP;
697
698 /*
699 * Referenced or dirty inodes are still in use. Give them another pass
700 * through the LRU as we canot reclaim them now.
701 */
702 if (atomic_read(&inode->i_count) ||
703 (inode->i_state & ~I_REFERENCED)) {
704 list_del_init(&inode->i_lru);
705 spin_unlock(&inode->i_lock);
706 this_cpu_dec(nr_unused);
707 return LRU_REMOVED;
708 }
709
710 /* recently referenced inodes get one more pass */
711 if (inode->i_state & I_REFERENCED) {
712 inode->i_state &= ~I_REFERENCED;
713 spin_unlock(&inode->i_lock);
714 return LRU_ROTATE;
715 }
716
717 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
718 __iget(inode);
719 spin_unlock(&inode->i_lock);
720 spin_unlock(lru_lock);
721 if (remove_inode_buffers(inode)) {
722 unsigned long reap;
723 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
724 if (current_is_kswapd())
725 __count_vm_events(KSWAPD_INODESTEAL, reap);
726 else
727 __count_vm_events(PGINODESTEAL, reap);
728 if (current->reclaim_state)
729 current->reclaim_state->reclaimed_slab += reap;
730 }
731 iput(inode);
732 spin_lock(lru_lock);
733 return LRU_RETRY;
734 }
735
736 WARN_ON(inode->i_state & I_NEW);
737 inode->i_state |= I_FREEING;
738 list_move(&inode->i_lru, freeable);
739 spin_unlock(&inode->i_lock);
740
741 this_cpu_dec(nr_unused);
742 return LRU_REMOVED;
743}
744
745/*
746 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
747 * This is called from the superblock shrinker function with a number of inodes
748 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
749 * then are freed outside inode_lock by dispose_list().
750 */
751long prune_icache_sb(struct super_block *sb, unsigned long nr_to_scan,
752 int nid)
753{
754 LIST_HEAD(freeable);
755 long freed;
756
757 freed = list_lru_walk_node(&sb->s_inode_lru, nid, inode_lru_isolate,
758 &freeable, &nr_to_scan);
759 dispose_list(&freeable);
760 return freed;
761}
762
763static void __wait_on_freeing_inode(struct inode *inode);
764/*
765 * Called with the inode lock held.
766 */
767static struct inode *find_inode(struct super_block *sb,
768 struct hlist_head *head,
769 int (*test)(struct inode *, void *),
770 void *data)
771{
772 struct inode *inode = NULL;
773
774repeat:
775 hlist_for_each_entry(inode, head, i_hash) {
776 if (inode->i_sb != sb)
777 continue;
778 if (!test(inode, data))
779 continue;
780 spin_lock(&inode->i_lock);
781 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
782 __wait_on_freeing_inode(inode);
783 goto repeat;
784 }
785 __iget(inode);
786 spin_unlock(&inode->i_lock);
787 return inode;
788 }
789 return NULL;
790}
791
792/*
793 * find_inode_fast is the fast path version of find_inode, see the comment at
794 * iget_locked for details.
795 */
796static struct inode *find_inode_fast(struct super_block *sb,
797 struct hlist_head *head, unsigned long ino)
798{
799 struct inode *inode = NULL;
800
801repeat:
802 hlist_for_each_entry(inode, head, i_hash) {
803 if (inode->i_ino != ino)
804 continue;
805 if (inode->i_sb != sb)
806 continue;
807 spin_lock(&inode->i_lock);
808 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
809 __wait_on_freeing_inode(inode);
810 goto repeat;
811 }
812 __iget(inode);
813 spin_unlock(&inode->i_lock);
814 return inode;
815 }
816 return NULL;
817}
818
819/*
820 * Each cpu owns a range of LAST_INO_BATCH numbers.
821 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
822 * to renew the exhausted range.
823 *
824 * This does not significantly increase overflow rate because every CPU can
825 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
826 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
827 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
828 * overflow rate by 2x, which does not seem too significant.
829 *
830 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
831 * error if st_ino won't fit in target struct field. Use 32bit counter
832 * here to attempt to avoid that.
833 */
834#define LAST_INO_BATCH 1024
835static DEFINE_PER_CPU(unsigned int, last_ino);
836
837unsigned int get_next_ino(void)
838{
839 unsigned int *p = &get_cpu_var(last_ino);
840 unsigned int res = *p;
841
842#ifdef CONFIG_SMP
843 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
844 static atomic_t shared_last_ino;
845 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
846
847 res = next - LAST_INO_BATCH;
848 }
849#endif
850
851 *p = ++res;
852 put_cpu_var(last_ino);
853 return res;
854}
855EXPORT_SYMBOL(get_next_ino);
856
857/**
858 * new_inode_pseudo - obtain an inode
859 * @sb: superblock
860 *
861 * Allocates a new inode for given superblock.
862 * Inode wont be chained in superblock s_inodes list
863 * This means :
864 * - fs can't be unmount
865 * - quotas, fsnotify, writeback can't work
866 */
867struct inode *new_inode_pseudo(struct super_block *sb)
868{
869 struct inode *inode = alloc_inode(sb);
870
871 if (inode) {
872 spin_lock(&inode->i_lock);
873 inode->i_state = 0;
874 spin_unlock(&inode->i_lock);
875 INIT_LIST_HEAD(&inode->i_sb_list);
876 }
877 return inode;
878}
879
880/**
881 * new_inode - obtain an inode
882 * @sb: superblock
883 *
884 * Allocates a new inode for given superblock. The default gfp_mask
885 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
886 * If HIGHMEM pages are unsuitable or it is known that pages allocated
887 * for the page cache are not reclaimable or migratable,
888 * mapping_set_gfp_mask() must be called with suitable flags on the
889 * newly created inode's mapping
890 *
891 */
892struct inode *new_inode(struct super_block *sb)
893{
894 struct inode *inode;
895
896 spin_lock_prefetch(&inode_sb_list_lock);
897
898 inode = new_inode_pseudo(sb);
899 if (inode)
900 inode_sb_list_add(inode);
901 return inode;
902}
903EXPORT_SYMBOL(new_inode);
904
905#ifdef CONFIG_DEBUG_LOCK_ALLOC
906void lockdep_annotate_inode_mutex_key(struct inode *inode)
907{
908 if (S_ISDIR(inode->i_mode)) {
909 struct file_system_type *type = inode->i_sb->s_type;
910
911 /* Set new key only if filesystem hasn't already changed it */
912 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
913 /*
914 * ensure nobody is actually holding i_mutex
915 */
916 mutex_destroy(&inode->i_mutex);
917 mutex_init(&inode->i_mutex);
918 lockdep_set_class(&inode->i_mutex,
919 &type->i_mutex_dir_key);
920 }
921 }
922}
923EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
924#endif
925
926/**
927 * unlock_new_inode - clear the I_NEW state and wake up any waiters
928 * @inode: new inode to unlock
929 *
930 * Called when the inode is fully initialised to clear the new state of the
931 * inode and wake up anyone waiting for the inode to finish initialisation.
932 */
933void unlock_new_inode(struct inode *inode)
934{
935 lockdep_annotate_inode_mutex_key(inode);
936 spin_lock(&inode->i_lock);
937 WARN_ON(!(inode->i_state & I_NEW));
938 inode->i_state &= ~I_NEW;
939 smp_mb();
940 wake_up_bit(&inode->i_state, __I_NEW);
941 spin_unlock(&inode->i_lock);
942}
943EXPORT_SYMBOL(unlock_new_inode);
944
945/**
946 * lock_two_nondirectories - take two i_mutexes on non-directory objects
947 *
948 * Lock any non-NULL argument that is not a directory.
949 * Zero, one or two objects may be locked by this function.
950 *
951 * @inode1: first inode to lock
952 * @inode2: second inode to lock
953 */
954void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
955{
956 if (inode1 > inode2)
957 swap(inode1, inode2);
958
959 if (inode1 && !S_ISDIR(inode1->i_mode))
960 mutex_lock(&inode1->i_mutex);
961 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
962 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_NONDIR2);
963}
964EXPORT_SYMBOL(lock_two_nondirectories);
965
966/**
967 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
968 * @inode1: first inode to unlock
969 * @inode2: second inode to unlock
970 */
971void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
972{
973 if (inode1 && !S_ISDIR(inode1->i_mode))
974 mutex_unlock(&inode1->i_mutex);
975 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
976 mutex_unlock(&inode2->i_mutex);
977}
978EXPORT_SYMBOL(unlock_two_nondirectories);
979
980/**
981 * iget5_locked - obtain an inode from a mounted file system
982 * @sb: super block of file system
983 * @hashval: hash value (usually inode number) to get
984 * @test: callback used for comparisons between inodes
985 * @set: callback used to initialize a new struct inode
986 * @data: opaque data pointer to pass to @test and @set
987 *
988 * Search for the inode specified by @hashval and @data in the inode cache,
989 * and if present it is return it with an increased reference count. This is
990 * a generalized version of iget_locked() for file systems where the inode
991 * number is not sufficient for unique identification of an inode.
992 *
993 * If the inode is not in cache, allocate a new inode and return it locked,
994 * hashed, and with the I_NEW flag set. The file system gets to fill it in
995 * before unlocking it via unlock_new_inode().
996 *
997 * Note both @test and @set are called with the inode_hash_lock held, so can't
998 * sleep.
999 */
1000struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1001 int (*test)(struct inode *, void *),
1002 int (*set)(struct inode *, void *), void *data)
1003{
1004 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1005 struct inode *inode;
1006
1007 spin_lock(&inode_hash_lock);
1008 inode = find_inode(sb, head, test, data);
1009 spin_unlock(&inode_hash_lock);
1010
1011 if (inode) {
1012 wait_on_inode(inode);
1013 return inode;
1014 }
1015
1016 inode = alloc_inode(sb);
1017 if (inode) {
1018 struct inode *old;
1019
1020 spin_lock(&inode_hash_lock);
1021 /* We released the lock, so.. */
1022 old = find_inode(sb, head, test, data);
1023 if (!old) {
1024 if (set(inode, data))
1025 goto set_failed;
1026
1027 spin_lock(&inode->i_lock);
1028 inode->i_state = I_NEW;
1029 hlist_add_head(&inode->i_hash, head);
1030 spin_unlock(&inode->i_lock);
1031 inode_sb_list_add(inode);
1032 spin_unlock(&inode_hash_lock);
1033
1034 /* Return the locked inode with I_NEW set, the
1035 * caller is responsible for filling in the contents
1036 */
1037 return inode;
1038 }
1039
1040 /*
1041 * Uhhuh, somebody else created the same inode under
1042 * us. Use the old inode instead of the one we just
1043 * allocated.
1044 */
1045 spin_unlock(&inode_hash_lock);
1046 destroy_inode(inode);
1047 inode = old;
1048 wait_on_inode(inode);
1049 }
1050 return inode;
1051
1052set_failed:
1053 spin_unlock(&inode_hash_lock);
1054 destroy_inode(inode);
1055 return NULL;
1056}
1057EXPORT_SYMBOL(iget5_locked);
1058
1059/**
1060 * iget_locked - obtain an inode from a mounted file system
1061 * @sb: super block of file system
1062 * @ino: inode number to get
1063 *
1064 * Search for the inode specified by @ino in the inode cache and if present
1065 * return it with an increased reference count. This is for file systems
1066 * where the inode number is sufficient for unique identification of an inode.
1067 *
1068 * If the inode is not in cache, allocate a new inode and return it locked,
1069 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1070 * before unlocking it via unlock_new_inode().
1071 */
1072struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1073{
1074 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1075 struct inode *inode;
1076
1077 spin_lock(&inode_hash_lock);
1078 inode = find_inode_fast(sb, head, ino);
1079 spin_unlock(&inode_hash_lock);
1080 if (inode) {
1081 wait_on_inode(inode);
1082 return inode;
1083 }
1084
1085 inode = alloc_inode(sb);
1086 if (inode) {
1087 struct inode *old;
1088
1089 spin_lock(&inode_hash_lock);
1090 /* We released the lock, so.. */
1091 old = find_inode_fast(sb, head, ino);
1092 if (!old) {
1093 inode->i_ino = ino;
1094 spin_lock(&inode->i_lock);
1095 inode->i_state = I_NEW;
1096 hlist_add_head(&inode->i_hash, head);
1097 spin_unlock(&inode->i_lock);
1098 inode_sb_list_add(inode);
1099 spin_unlock(&inode_hash_lock);
1100
1101 /* Return the locked inode with I_NEW set, the
1102 * caller is responsible for filling in the contents
1103 */
1104 return inode;
1105 }
1106
1107 /*
1108 * Uhhuh, somebody else created the same inode under
1109 * us. Use the old inode instead of the one we just
1110 * allocated.
1111 */
1112 spin_unlock(&inode_hash_lock);
1113 destroy_inode(inode);
1114 inode = old;
1115 wait_on_inode(inode);
1116 }
1117 return inode;
1118}
1119EXPORT_SYMBOL(iget_locked);
1120
1121/*
1122 * search the inode cache for a matching inode number.
1123 * If we find one, then the inode number we are trying to
1124 * allocate is not unique and so we should not use it.
1125 *
1126 * Returns 1 if the inode number is unique, 0 if it is not.
1127 */
1128static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1129{
1130 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1131 struct inode *inode;
1132
1133 spin_lock(&inode_hash_lock);
1134 hlist_for_each_entry(inode, b, i_hash) {
1135 if (inode->i_ino == ino && inode->i_sb == sb) {
1136 spin_unlock(&inode_hash_lock);
1137 return 0;
1138 }
1139 }
1140 spin_unlock(&inode_hash_lock);
1141
1142 return 1;
1143}
1144
1145/**
1146 * iunique - get a unique inode number
1147 * @sb: superblock
1148 * @max_reserved: highest reserved inode number
1149 *
1150 * Obtain an inode number that is unique on the system for a given
1151 * superblock. This is used by file systems that have no natural
1152 * permanent inode numbering system. An inode number is returned that
1153 * is higher than the reserved limit but unique.
1154 *
1155 * BUGS:
1156 * With a large number of inodes live on the file system this function
1157 * currently becomes quite slow.
1158 */
1159ino_t iunique(struct super_block *sb, ino_t max_reserved)
1160{
1161 /*
1162 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1163 * error if st_ino won't fit in target struct field. Use 32bit counter
1164 * here to attempt to avoid that.
1165 */
1166 static DEFINE_SPINLOCK(iunique_lock);
1167 static unsigned int counter;
1168 ino_t res;
1169
1170 spin_lock(&iunique_lock);
1171 do {
1172 if (counter <= max_reserved)
1173 counter = max_reserved + 1;
1174 res = counter++;
1175 } while (!test_inode_iunique(sb, res));
1176 spin_unlock(&iunique_lock);
1177
1178 return res;
1179}
1180EXPORT_SYMBOL(iunique);
1181
1182struct inode *igrab(struct inode *inode)
1183{
1184 spin_lock(&inode->i_lock);
1185 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1186 __iget(inode);
1187 spin_unlock(&inode->i_lock);
1188 } else {
1189 spin_unlock(&inode->i_lock);
1190 /*
1191 * Handle the case where s_op->clear_inode is not been
1192 * called yet, and somebody is calling igrab
1193 * while the inode is getting freed.
1194 */
1195 inode = NULL;
1196 }
1197 return inode;
1198}
1199EXPORT_SYMBOL(igrab);
1200
1201/**
1202 * ilookup5_nowait - search for an inode in the inode cache
1203 * @sb: super block of file system to search
1204 * @hashval: hash value (usually inode number) to search for
1205 * @test: callback used for comparisons between inodes
1206 * @data: opaque data pointer to pass to @test
1207 *
1208 * Search for the inode specified by @hashval and @data in the inode cache.
1209 * If the inode is in the cache, the inode is returned with an incremented
1210 * reference count.
1211 *
1212 * Note: I_NEW is not waited upon so you have to be very careful what you do
1213 * with the returned inode. You probably should be using ilookup5() instead.
1214 *
1215 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1216 */
1217struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1218 int (*test)(struct inode *, void *), void *data)
1219{
1220 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1221 struct inode *inode;
1222
1223 spin_lock(&inode_hash_lock);
1224 inode = find_inode(sb, head, test, data);
1225 spin_unlock(&inode_hash_lock);
1226
1227 return inode;
1228}
1229EXPORT_SYMBOL(ilookup5_nowait);
1230
1231/**
1232 * ilookup5 - search for an inode in the inode cache
1233 * @sb: super block of file system to search
1234 * @hashval: hash value (usually inode number) to search for
1235 * @test: callback used for comparisons between inodes
1236 * @data: opaque data pointer to pass to @test
1237 *
1238 * Search for the inode specified by @hashval and @data in the inode cache,
1239 * and if the inode is in the cache, return the inode with an incremented
1240 * reference count. Waits on I_NEW before returning the inode.
1241 * returned with an incremented reference count.
1242 *
1243 * This is a generalized version of ilookup() for file systems where the
1244 * inode number is not sufficient for unique identification of an inode.
1245 *
1246 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1247 */
1248struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1249 int (*test)(struct inode *, void *), void *data)
1250{
1251 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1252
1253 if (inode)
1254 wait_on_inode(inode);
1255 return inode;
1256}
1257EXPORT_SYMBOL(ilookup5);
1258
1259/**
1260 * ilookup - search for an inode in the inode cache
1261 * @sb: super block of file system to search
1262 * @ino: inode number to search for
1263 *
1264 * Search for the inode @ino in the inode cache, and if the inode is in the
1265 * cache, the inode is returned with an incremented reference count.
1266 */
1267struct inode *ilookup(struct super_block *sb, unsigned long ino)
1268{
1269 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1270 struct inode *inode;
1271
1272 spin_lock(&inode_hash_lock);
1273 inode = find_inode_fast(sb, head, ino);
1274 spin_unlock(&inode_hash_lock);
1275
1276 if (inode)
1277 wait_on_inode(inode);
1278 return inode;
1279}
1280EXPORT_SYMBOL(ilookup);
1281
1282int insert_inode_locked(struct inode *inode)
1283{
1284 struct super_block *sb = inode->i_sb;
1285 ino_t ino = inode->i_ino;
1286 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1287
1288 while (1) {
1289 struct inode *old = NULL;
1290 spin_lock(&inode_hash_lock);
1291 hlist_for_each_entry(old, head, i_hash) {
1292 if (old->i_ino != ino)
1293 continue;
1294 if (old->i_sb != sb)
1295 continue;
1296 spin_lock(&old->i_lock);
1297 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1298 spin_unlock(&old->i_lock);
1299 continue;
1300 }
1301 break;
1302 }
1303 if (likely(!old)) {
1304 spin_lock(&inode->i_lock);
1305 inode->i_state |= I_NEW;
1306 hlist_add_head(&inode->i_hash, head);
1307 spin_unlock(&inode->i_lock);
1308 spin_unlock(&inode_hash_lock);
1309 return 0;
1310 }
1311 __iget(old);
1312 spin_unlock(&old->i_lock);
1313 spin_unlock(&inode_hash_lock);
1314 wait_on_inode(old);
1315 if (unlikely(!inode_unhashed(old))) {
1316 iput(old);
1317 return -EBUSY;
1318 }
1319 iput(old);
1320 }
1321}
1322EXPORT_SYMBOL(insert_inode_locked);
1323
1324int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1325 int (*test)(struct inode *, void *), void *data)
1326{
1327 struct super_block *sb = inode->i_sb;
1328 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1329
1330 while (1) {
1331 struct inode *old = NULL;
1332
1333 spin_lock(&inode_hash_lock);
1334 hlist_for_each_entry(old, head, i_hash) {
1335 if (old->i_sb != sb)
1336 continue;
1337 if (!test(old, data))
1338 continue;
1339 spin_lock(&old->i_lock);
1340 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1341 spin_unlock(&old->i_lock);
1342 continue;
1343 }
1344 break;
1345 }
1346 if (likely(!old)) {
1347 spin_lock(&inode->i_lock);
1348 inode->i_state |= I_NEW;
1349 hlist_add_head(&inode->i_hash, head);
1350 spin_unlock(&inode->i_lock);
1351 spin_unlock(&inode_hash_lock);
1352 return 0;
1353 }
1354 __iget(old);
1355 spin_unlock(&old->i_lock);
1356 spin_unlock(&inode_hash_lock);
1357 wait_on_inode(old);
1358 if (unlikely(!inode_unhashed(old))) {
1359 iput(old);
1360 return -EBUSY;
1361 }
1362 iput(old);
1363 }
1364}
1365EXPORT_SYMBOL(insert_inode_locked4);
1366
1367
1368int generic_delete_inode(struct inode *inode)
1369{
1370 return 1;
1371}
1372EXPORT_SYMBOL(generic_delete_inode);
1373
1374/*
1375 * Called when we're dropping the last reference
1376 * to an inode.
1377 *
1378 * Call the FS "drop_inode()" function, defaulting to
1379 * the legacy UNIX filesystem behaviour. If it tells
1380 * us to evict inode, do so. Otherwise, retain inode
1381 * in cache if fs is alive, sync and evict if fs is
1382 * shutting down.
1383 */
1384static void iput_final(struct inode *inode)
1385{
1386 struct super_block *sb = inode->i_sb;
1387 const struct super_operations *op = inode->i_sb->s_op;
1388 int drop;
1389
1390 WARN_ON(inode->i_state & I_NEW);
1391
1392 if (op->drop_inode)
1393 drop = op->drop_inode(inode);
1394 else
1395 drop = generic_drop_inode(inode);
1396
1397 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1398 inode->i_state |= I_REFERENCED;
1399 inode_add_lru(inode);
1400 spin_unlock(&inode->i_lock);
1401 return;
1402 }
1403
1404 if (!drop) {
1405 inode->i_state |= I_WILL_FREE;
1406 spin_unlock(&inode->i_lock);
1407 write_inode_now(inode, 1);
1408 spin_lock(&inode->i_lock);
1409 WARN_ON(inode->i_state & I_NEW);
1410 inode->i_state &= ~I_WILL_FREE;
1411 }
1412
1413 inode->i_state |= I_FREEING;
1414 if (!list_empty(&inode->i_lru))
1415 inode_lru_list_del(inode);
1416 spin_unlock(&inode->i_lock);
1417
1418 evict(inode);
1419}
1420
1421/**
1422 * iput - put an inode
1423 * @inode: inode to put
1424 *
1425 * Puts an inode, dropping its usage count. If the inode use count hits
1426 * zero, the inode is then freed and may also be destroyed.
1427 *
1428 * Consequently, iput() can sleep.
1429 */
1430void iput(struct inode *inode)
1431{
1432 if (inode) {
1433 BUG_ON(inode->i_state & I_CLEAR);
1434
1435 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1436 iput_final(inode);
1437 }
1438}
1439EXPORT_SYMBOL(iput);
1440
1441/**
1442 * bmap - find a block number in a file
1443 * @inode: inode of file
1444 * @block: block to find
1445 *
1446 * Returns the block number on the device holding the inode that
1447 * is the disk block number for the block of the file requested.
1448 * That is, asked for block 4 of inode 1 the function will return the
1449 * disk block relative to the disk start that holds that block of the
1450 * file.
1451 */
1452sector_t bmap(struct inode *inode, sector_t block)
1453{
1454 sector_t res = 0;
1455 if (inode->i_mapping->a_ops->bmap)
1456 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1457 return res;
1458}
1459EXPORT_SYMBOL(bmap);
1460
1461/*
1462 * With relative atime, only update atime if the previous atime is
1463 * earlier than either the ctime or mtime or if at least a day has
1464 * passed since the last atime update.
1465 */
1466static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1467 struct timespec now)
1468{
1469
1470 if (!(mnt->mnt_flags & MNT_RELATIME))
1471 return 1;
1472 /*
1473 * Is mtime younger than atime? If yes, update atime:
1474 */
1475 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1476 return 1;
1477 /*
1478 * Is ctime younger than atime? If yes, update atime:
1479 */
1480 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1481 return 1;
1482
1483 /*
1484 * Is the previous atime value older than a day? If yes,
1485 * update atime:
1486 */
1487 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1488 return 1;
1489 /*
1490 * Good, we can skip the atime update:
1491 */
1492 return 0;
1493}
1494
1495/*
1496 * This does the actual work of updating an inodes time or version. Must have
1497 * had called mnt_want_write() before calling this.
1498 */
1499static int update_time(struct inode *inode, struct timespec *time, int flags)
1500{
1501 if (inode->i_op->update_time)
1502 return inode->i_op->update_time(inode, time, flags);
1503
1504 if (flags & S_ATIME)
1505 inode->i_atime = *time;
1506 if (flags & S_VERSION)
1507 inode_inc_iversion(inode);
1508 if (flags & S_CTIME)
1509 inode->i_ctime = *time;
1510 if (flags & S_MTIME)
1511 inode->i_mtime = *time;
1512 mark_inode_dirty_sync(inode);
1513 return 0;
1514}
1515
1516/**
1517 * touch_atime - update the access time
1518 * @path: the &struct path to update
1519 *
1520 * Update the accessed time on an inode and mark it for writeback.
1521 * This function automatically handles read only file systems and media,
1522 * as well as the "noatime" flag and inode specific "noatime" markers.
1523 */
1524void touch_atime(const struct path *path)
1525{
1526 struct vfsmount *mnt = path->mnt;
1527 struct inode *inode = path->dentry->d_inode;
1528 struct timespec now;
1529
1530 if (inode->i_flags & S_NOATIME)
1531 return;
1532 if (IS_NOATIME(inode))
1533 return;
1534 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1535 return;
1536
1537 if (mnt->mnt_flags & MNT_NOATIME)
1538 return;
1539 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1540 return;
1541
1542 now = current_fs_time(inode->i_sb);
1543
1544 if (!relatime_need_update(mnt, inode, now))
1545 return;
1546
1547 if (timespec_equal(&inode->i_atime, &now))
1548 return;
1549
1550 if (!sb_start_write_trylock(inode->i_sb))
1551 return;
1552
1553 if (__mnt_want_write(mnt))
1554 goto skip_update;
1555 /*
1556 * File systems can error out when updating inodes if they need to
1557 * allocate new space to modify an inode (such is the case for
1558 * Btrfs), but since we touch atime while walking down the path we
1559 * really don't care if we failed to update the atime of the file,
1560 * so just ignore the return value.
1561 * We may also fail on filesystems that have the ability to make parts
1562 * of the fs read only, e.g. subvolumes in Btrfs.
1563 */
1564 update_time(inode, &now, S_ATIME);
1565 __mnt_drop_write(mnt);
1566skip_update:
1567 sb_end_write(inode->i_sb);
1568}
1569EXPORT_SYMBOL(touch_atime);
1570
1571/*
1572 * The logic we want is
1573 *
1574 * if suid or (sgid and xgrp)
1575 * remove privs
1576 */
1577int should_remove_suid(struct dentry *dentry)
1578{
1579 umode_t mode = dentry->d_inode->i_mode;
1580 int kill = 0;
1581
1582 /* suid always must be killed */
1583 if (unlikely(mode & S_ISUID))
1584 kill = ATTR_KILL_SUID;
1585
1586 /*
1587 * sgid without any exec bits is just a mandatory locking mark; leave
1588 * it alone. If some exec bits are set, it's a real sgid; kill it.
1589 */
1590 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1591 kill |= ATTR_KILL_SGID;
1592
1593 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1594 return kill;
1595
1596 return 0;
1597}
1598EXPORT_SYMBOL(should_remove_suid);
1599
1600static int __remove_suid(struct dentry *dentry, int kill)
1601{
1602 struct iattr newattrs;
1603
1604 newattrs.ia_valid = ATTR_FORCE | kill;
1605 /*
1606 * Note we call this on write, so notify_change will not
1607 * encounter any conflicting delegations:
1608 */
1609 return notify_change(dentry, &newattrs, NULL);
1610}
1611
1612int file_remove_suid(struct file *file)
1613{
1614 struct dentry *dentry = file->f_path.dentry;
1615 struct inode *inode = dentry->d_inode;
1616 int killsuid;
1617 int killpriv;
1618 int error = 0;
1619
1620 /* Fast path for nothing security related */
1621 if (IS_NOSEC(inode))
1622 return 0;
1623
1624 killsuid = should_remove_suid(dentry);
1625 killpriv = security_inode_need_killpriv(dentry);
1626
1627 if (killpriv < 0)
1628 return killpriv;
1629 if (killpriv)
1630 error = security_inode_killpriv(dentry);
1631 if (!error && killsuid)
1632 error = __remove_suid(dentry, killsuid);
1633 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1634 inode->i_flags |= S_NOSEC;
1635
1636 return error;
1637}
1638EXPORT_SYMBOL(file_remove_suid);
1639
1640/**
1641 * file_update_time - update mtime and ctime time
1642 * @file: file accessed
1643 *
1644 * Update the mtime and ctime members of an inode and mark the inode
1645 * for writeback. Note that this function is meant exclusively for
1646 * usage in the file write path of filesystems, and filesystems may
1647 * choose to explicitly ignore update via this function with the
1648 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1649 * timestamps are handled by the server. This can return an error for
1650 * file systems who need to allocate space in order to update an inode.
1651 */
1652
1653int file_update_time(struct file *file)
1654{
1655 struct inode *inode = file_inode(file);
1656 struct timespec now;
1657 int sync_it = 0;
1658 int ret;
1659
1660 /* First try to exhaust all avenues to not sync */
1661 if (IS_NOCMTIME(inode))
1662 return 0;
1663
1664 now = current_fs_time(inode->i_sb);
1665 if (!timespec_equal(&inode->i_mtime, &now))
1666 sync_it = S_MTIME;
1667
1668 if (!timespec_equal(&inode->i_ctime, &now))
1669 sync_it |= S_CTIME;
1670
1671 if (IS_I_VERSION(inode))
1672 sync_it |= S_VERSION;
1673
1674 if (!sync_it)
1675 return 0;
1676
1677 /* Finally allowed to write? Takes lock. */
1678 if (__mnt_want_write_file(file))
1679 return 0;
1680
1681 ret = update_time(inode, &now, sync_it);
1682 __mnt_drop_write_file(file);
1683
1684 return ret;
1685}
1686EXPORT_SYMBOL(file_update_time);
1687
1688int inode_needs_sync(struct inode *inode)
1689{
1690 if (IS_SYNC(inode))
1691 return 1;
1692 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1693 return 1;
1694 return 0;
1695}
1696EXPORT_SYMBOL(inode_needs_sync);
1697
1698int inode_wait(void *word)
1699{
1700 schedule();
1701 return 0;
1702}
1703EXPORT_SYMBOL(inode_wait);
1704
1705/*
1706 * If we try to find an inode in the inode hash while it is being
1707 * deleted, we have to wait until the filesystem completes its
1708 * deletion before reporting that it isn't found. This function waits
1709 * until the deletion _might_ have completed. Callers are responsible
1710 * to recheck inode state.
1711 *
1712 * It doesn't matter if I_NEW is not set initially, a call to
1713 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1714 * will DTRT.
1715 */
1716static void __wait_on_freeing_inode(struct inode *inode)
1717{
1718 wait_queue_head_t *wq;
1719 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1720 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1721 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1722 spin_unlock(&inode->i_lock);
1723 spin_unlock(&inode_hash_lock);
1724 schedule();
1725 finish_wait(wq, &wait.wait);
1726 spin_lock(&inode_hash_lock);
1727}
1728
1729static __initdata unsigned long ihash_entries;
1730static int __init set_ihash_entries(char *str)
1731{
1732 if (!str)
1733 return 0;
1734 ihash_entries = simple_strtoul(str, &str, 0);
1735 return 1;
1736}
1737__setup("ihash_entries=", set_ihash_entries);
1738
1739/*
1740 * Initialize the waitqueues and inode hash table.
1741 */
1742void __init inode_init_early(void)
1743{
1744 unsigned int loop;
1745
1746 /* If hashes are distributed across NUMA nodes, defer
1747 * hash allocation until vmalloc space is available.
1748 */
1749 if (hashdist)
1750 return;
1751
1752 inode_hashtable =
1753 alloc_large_system_hash("Inode-cache",
1754 sizeof(struct hlist_head),
1755 ihash_entries,
1756 14,
1757 HASH_EARLY,
1758 &i_hash_shift,
1759 &i_hash_mask,
1760 0,
1761 0);
1762
1763 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1764 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1765}
1766
1767void __init inode_init(void)
1768{
1769 unsigned int loop;
1770
1771 /* inode slab cache */
1772 inode_cachep = kmem_cache_create("inode_cache",
1773 sizeof(struct inode),
1774 0,
1775 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1776 SLAB_MEM_SPREAD),
1777 init_once);
1778
1779 /* Hash may have been set up in inode_init_early */
1780 if (!hashdist)
1781 return;
1782
1783 inode_hashtable =
1784 alloc_large_system_hash("Inode-cache",
1785 sizeof(struct hlist_head),
1786 ihash_entries,
1787 14,
1788 0,
1789 &i_hash_shift,
1790 &i_hash_mask,
1791 0,
1792 0);
1793
1794 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1795 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1796}
1797
1798void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1799{
1800 inode->i_mode = mode;
1801 if (S_ISCHR(mode)) {
1802 inode->i_fop = &def_chr_fops;
1803 inode->i_rdev = rdev;
1804 } else if (S_ISBLK(mode)) {
1805 inode->i_fop = &def_blk_fops;
1806 inode->i_rdev = rdev;
1807 } else if (S_ISFIFO(mode))
1808 inode->i_fop = &pipefifo_fops;
1809 else if (S_ISSOCK(mode))
1810 inode->i_fop = &bad_sock_fops;
1811 else
1812 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1813 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1814 inode->i_ino);
1815}
1816EXPORT_SYMBOL(init_special_inode);
1817
1818/**
1819 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1820 * @inode: New inode
1821 * @dir: Directory inode
1822 * @mode: mode of the new inode
1823 */
1824void inode_init_owner(struct inode *inode, const struct inode *dir,
1825 umode_t mode)
1826{
1827 inode->i_uid = current_fsuid();
1828 if (dir && dir->i_mode & S_ISGID) {
1829 inode->i_gid = dir->i_gid;
1830 if (S_ISDIR(mode))
1831 mode |= S_ISGID;
1832 } else
1833 inode->i_gid = current_fsgid();
1834 inode->i_mode = mode;
1835}
1836EXPORT_SYMBOL(inode_init_owner);
1837
1838/**
1839 * inode_owner_or_capable - check current task permissions to inode
1840 * @inode: inode being checked
1841 *
1842 * Return true if current either has CAP_FOWNER to the inode, or
1843 * owns the file.
1844 */
1845bool inode_owner_or_capable(const struct inode *inode)
1846{
1847 if (uid_eq(current_fsuid(), inode->i_uid))
1848 return true;
1849 if (inode_capable(inode, CAP_FOWNER))
1850 return true;
1851 return false;
1852}
1853EXPORT_SYMBOL(inode_owner_or_capable);
1854
1855/*
1856 * Direct i/o helper functions
1857 */
1858static void __inode_dio_wait(struct inode *inode)
1859{
1860 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1861 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1862
1863 do {
1864 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1865 if (atomic_read(&inode->i_dio_count))
1866 schedule();
1867 } while (atomic_read(&inode->i_dio_count));
1868 finish_wait(wq, &q.wait);
1869}
1870
1871/**
1872 * inode_dio_wait - wait for outstanding DIO requests to finish
1873 * @inode: inode to wait for
1874 *
1875 * Waits for all pending direct I/O requests to finish so that we can
1876 * proceed with a truncate or equivalent operation.
1877 *
1878 * Must be called under a lock that serializes taking new references
1879 * to i_dio_count, usually by inode->i_mutex.
1880 */
1881void inode_dio_wait(struct inode *inode)
1882{
1883 if (atomic_read(&inode->i_dio_count))
1884 __inode_dio_wait(inode);
1885}
1886EXPORT_SYMBOL(inode_dio_wait);
1887
1888/*
1889 * inode_dio_done - signal finish of a direct I/O requests
1890 * @inode: inode the direct I/O happens on
1891 *
1892 * This is called once we've finished processing a direct I/O request,
1893 * and is used to wake up callers waiting for direct I/O to be quiesced.
1894 */
1895void inode_dio_done(struct inode *inode)
1896{
1897 if (atomic_dec_and_test(&inode->i_dio_count))
1898 wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1899}
1900EXPORT_SYMBOL(inode_dio_done);
1901
1902/*
1903 * inode_set_flags - atomically set some inode flags
1904 *
1905 * Note: the caller should be holding i_mutex, or else be sure that
1906 * they have exclusive access to the inode structure (i.e., while the
1907 * inode is being instantiated). The reason for the cmpxchg() loop
1908 * --- which wouldn't be necessary if all code paths which modify
1909 * i_flags actually followed this rule, is that there is at least one
1910 * code path which doesn't today --- for example,
1911 * __generic_file_aio_write() calls file_remove_suid() without holding
1912 * i_mutex --- so we use cmpxchg() out of an abundance of caution.
1913 *
1914 * In the long run, i_mutex is overkill, and we should probably look
1915 * at using the i_lock spinlock to protect i_flags, and then make sure
1916 * it is so documented in include/linux/fs.h and that all code follows
1917 * the locking convention!!
1918 */
1919void inode_set_flags(struct inode *inode, unsigned int flags,
1920 unsigned int mask)
1921{
1922 unsigned int old_flags, new_flags;
1923
1924 WARN_ON_ONCE(flags & ~mask);
1925 do {
1926 old_flags = ACCESS_ONCE(inode->i_flags);
1927 new_flags = (old_flags & ~mask) | flags;
1928 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
1929 new_flags) != old_flags));
1930}
1931EXPORT_SYMBOL(inode_set_flags);