]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/inode.c
arm64: SW PAN: Point saved ttbr0 at the zero page when switching to init_mm
[mirror_ubuntu-artful-kernel.git] / fs / inode.c
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 <trace/events/writeback.h>
22 #include "internal.h"
23
24 /*
25 * Inode locking rules:
26 *
27 * inode->i_lock protects:
28 * inode->i_state, inode->i_hash, __iget()
29 * Inode LRU list locks protect:
30 * inode->i_sb->s_inode_lru, inode->i_lru
31 * inode->i_sb->s_inode_list_lock protects:
32 * inode->i_sb->s_inodes, inode->i_sb_list
33 * bdi->wb.list_lock protects:
34 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
35 * inode_hash_lock protects:
36 * inode_hashtable, inode->i_hash
37 *
38 * Lock ordering:
39 *
40 * inode->i_sb->s_inode_list_lock
41 * inode->i_lock
42 * Inode LRU list locks
43 *
44 * bdi->wb.list_lock
45 * inode->i_lock
46 *
47 * inode_hash_lock
48 * inode->i_sb->s_inode_list_lock
49 * inode->i_lock
50 *
51 * iunique_lock
52 * inode_hash_lock
53 */
54
55 static unsigned int i_hash_mask __read_mostly;
56 static unsigned int i_hash_shift __read_mostly;
57 static struct hlist_head *inode_hashtable __read_mostly;
58 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
59
60 /*
61 * Empty aops. Can be used for the cases where the user does not
62 * define any of the address_space operations.
63 */
64 const struct address_space_operations empty_aops = {
65 };
66 EXPORT_SYMBOL(empty_aops);
67
68 /*
69 * Statistics gathering..
70 */
71 struct inodes_stat_t inodes_stat;
72
73 static DEFINE_PER_CPU(unsigned long, nr_inodes);
74 static DEFINE_PER_CPU(unsigned long, nr_unused);
75
76 static struct kmem_cache *inode_cachep __read_mostly;
77
78 static long get_nr_inodes(void)
79 {
80 int i;
81 long sum = 0;
82 for_each_possible_cpu(i)
83 sum += per_cpu(nr_inodes, i);
84 return sum < 0 ? 0 : sum;
85 }
86
87 static inline long get_nr_inodes_unused(void)
88 {
89 int i;
90 long sum = 0;
91 for_each_possible_cpu(i)
92 sum += per_cpu(nr_unused, i);
93 return sum < 0 ? 0 : sum;
94 }
95
96 long get_nr_dirty_inodes(void)
97 {
98 /* not actually dirty inodes, but a wild approximation */
99 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
100 return nr_dirty > 0 ? nr_dirty : 0;
101 }
102
103 /*
104 * Handle nr_inode sysctl
105 */
106 #ifdef CONFIG_SYSCTL
107 int proc_nr_inodes(struct ctl_table *table, int write,
108 void __user *buffer, size_t *lenp, loff_t *ppos)
109 {
110 inodes_stat.nr_inodes = get_nr_inodes();
111 inodes_stat.nr_unused = get_nr_inodes_unused();
112 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
113 }
114 #endif
115
116 static int no_open(struct inode *inode, struct file *file)
117 {
118 return -ENXIO;
119 }
120
121 /**
122 * inode_init_always - perform inode structure initialisation
123 * @sb: superblock inode belongs to
124 * @inode: inode to initialise
125 *
126 * These are initializations that need to be done on every inode
127 * allocation as the fields are not initialised by slab allocation.
128 */
129 int inode_init_always(struct super_block *sb, struct inode *inode)
130 {
131 static const struct inode_operations empty_iops;
132 static const struct file_operations no_open_fops = {.open = no_open};
133 struct address_space *const mapping = &inode->i_data;
134
135 inode->i_sb = sb;
136 inode->i_blkbits = sb->s_blocksize_bits;
137 inode->i_flags = 0;
138 atomic_set(&inode->i_count, 1);
139 inode->i_op = &empty_iops;
140 inode->i_fop = &no_open_fops;
141 inode->__i_nlink = 1;
142 inode->i_opflags = 0;
143 if (sb->s_xattr)
144 inode->i_opflags |= IOP_XATTR;
145 i_uid_write(inode, 0);
146 i_gid_write(inode, 0);
147 atomic_set(&inode->i_writecount, 0);
148 inode->i_size = 0;
149 inode->i_write_hint = WRITE_LIFE_NOT_SET;
150 inode->i_blocks = 0;
151 inode->i_bytes = 0;
152 inode->i_generation = 0;
153 inode->i_pipe = NULL;
154 inode->i_bdev = NULL;
155 inode->i_cdev = NULL;
156 inode->i_link = NULL;
157 inode->i_dir_seq = 0;
158 inode->i_rdev = 0;
159 inode->dirtied_when = 0;
160
161 #ifdef CONFIG_CGROUP_WRITEBACK
162 inode->i_wb_frn_winner = 0;
163 inode->i_wb_frn_avg_time = 0;
164 inode->i_wb_frn_history = 0;
165 #endif
166
167 if (security_inode_alloc(inode))
168 goto out;
169 spin_lock_init(&inode->i_lock);
170 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
171
172 init_rwsem(&inode->i_rwsem);
173 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
174
175 atomic_set(&inode->i_dio_count, 0);
176
177 mapping->a_ops = &empty_aops;
178 mapping->host = inode;
179 mapping->flags = 0;
180 atomic_set(&mapping->i_mmap_writable, 0);
181 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
182 mapping->private_data = NULL;
183 mapping->writeback_index = 0;
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 inode->i_flctx = NULL;
195 this_cpu_inc(nr_inodes);
196
197 return 0;
198 out:
199 return -ENOMEM;
200 }
201 EXPORT_SYMBOL(inode_init_always);
202
203 static 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
226 void free_inode_nonrcu(struct inode *inode)
227 {
228 kmem_cache_free(inode_cachep, inode);
229 }
230 EXPORT_SYMBOL(free_inode_nonrcu);
231
232 void __destroy_inode(struct inode *inode)
233 {
234 BUG_ON(inode_has_buffers(inode));
235 inode_detach_wb(inode);
236 security_inode_free(inode);
237 fsnotify_inode_delete(inode);
238 locks_free_lock_context(inode);
239 if (!inode->i_nlink) {
240 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
241 atomic_long_dec(&inode->i_sb->s_remove_count);
242 }
243
244 #ifdef CONFIG_FS_POSIX_ACL
245 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
246 posix_acl_release(inode->i_acl);
247 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
248 posix_acl_release(inode->i_default_acl);
249 #endif
250 this_cpu_dec(nr_inodes);
251 }
252 EXPORT_SYMBOL(__destroy_inode);
253
254 static void i_callback(struct rcu_head *head)
255 {
256 struct inode *inode = container_of(head, struct inode, i_rcu);
257 kmem_cache_free(inode_cachep, inode);
258 }
259
260 static void destroy_inode(struct inode *inode)
261 {
262 BUG_ON(!list_empty(&inode->i_lru));
263 __destroy_inode(inode);
264 if (inode->i_sb->s_op->destroy_inode)
265 inode->i_sb->s_op->destroy_inode(inode);
266 else
267 call_rcu(&inode->i_rcu, i_callback);
268 }
269
270 /**
271 * drop_nlink - directly drop an inode's link count
272 * @inode: inode
273 *
274 * This is a low-level filesystem helper to replace any
275 * direct filesystem manipulation of i_nlink. In cases
276 * where we are attempting to track writes to the
277 * filesystem, a decrement to zero means an imminent
278 * write when the file is truncated and actually unlinked
279 * on the filesystem.
280 */
281 void drop_nlink(struct inode *inode)
282 {
283 WARN_ON(inode->i_nlink == 0);
284 inode->__i_nlink--;
285 if (!inode->i_nlink)
286 atomic_long_inc(&inode->i_sb->s_remove_count);
287 }
288 EXPORT_SYMBOL(drop_nlink);
289
290 /**
291 * clear_nlink - directly zero an inode's link count
292 * @inode: inode
293 *
294 * This is a low-level filesystem helper to replace any
295 * direct filesystem manipulation of i_nlink. See
296 * drop_nlink() for why we care about i_nlink hitting zero.
297 */
298 void clear_nlink(struct inode *inode)
299 {
300 if (inode->i_nlink) {
301 inode->__i_nlink = 0;
302 atomic_long_inc(&inode->i_sb->s_remove_count);
303 }
304 }
305 EXPORT_SYMBOL(clear_nlink);
306
307 /**
308 * set_nlink - directly set an inode's link count
309 * @inode: inode
310 * @nlink: new nlink (should be non-zero)
311 *
312 * This is a low-level filesystem helper to replace any
313 * direct filesystem manipulation of i_nlink.
314 */
315 void set_nlink(struct inode *inode, unsigned int nlink)
316 {
317 if (!nlink) {
318 clear_nlink(inode);
319 } else {
320 /* Yes, some filesystems do change nlink from zero to one */
321 if (inode->i_nlink == 0)
322 atomic_long_dec(&inode->i_sb->s_remove_count);
323
324 inode->__i_nlink = nlink;
325 }
326 }
327 EXPORT_SYMBOL(set_nlink);
328
329 /**
330 * inc_nlink - directly increment an inode's link count
331 * @inode: inode
332 *
333 * This is a low-level filesystem helper to replace any
334 * direct filesystem manipulation of i_nlink. Currently,
335 * it is only here for parity with dec_nlink().
336 */
337 void inc_nlink(struct inode *inode)
338 {
339 if (unlikely(inode->i_nlink == 0)) {
340 WARN_ON(!(inode->i_state & I_LINKABLE));
341 atomic_long_dec(&inode->i_sb->s_remove_count);
342 }
343
344 inode->__i_nlink++;
345 }
346 EXPORT_SYMBOL(inc_nlink);
347
348 void address_space_init_once(struct address_space *mapping)
349 {
350 memset(mapping, 0, sizeof(*mapping));
351 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC | __GFP_ACCOUNT);
352 spin_lock_init(&mapping->tree_lock);
353 init_rwsem(&mapping->i_mmap_rwsem);
354 INIT_LIST_HEAD(&mapping->private_list);
355 spin_lock_init(&mapping->private_lock);
356 mapping->i_mmap = RB_ROOT;
357 }
358 EXPORT_SYMBOL(address_space_init_once);
359
360 /*
361 * These are initializations that only need to be done
362 * once, because the fields are idempotent across use
363 * of the inode, so let the slab aware of that.
364 */
365 void inode_init_once(struct inode *inode)
366 {
367 memset(inode, 0, sizeof(*inode));
368 INIT_HLIST_NODE(&inode->i_hash);
369 INIT_LIST_HEAD(&inode->i_devices);
370 INIT_LIST_HEAD(&inode->i_io_list);
371 INIT_LIST_HEAD(&inode->i_wb_list);
372 INIT_LIST_HEAD(&inode->i_lru);
373 address_space_init_once(&inode->i_data);
374 i_size_ordered_init(inode);
375 }
376 EXPORT_SYMBOL(inode_init_once);
377
378 static void init_once(void *foo)
379 {
380 struct inode *inode = (struct inode *) foo;
381
382 inode_init_once(inode);
383 }
384
385 /*
386 * inode->i_lock must be held
387 */
388 void __iget(struct inode *inode)
389 {
390 atomic_inc(&inode->i_count);
391 }
392
393 /*
394 * get additional reference to inode; caller must already hold one.
395 */
396 void ihold(struct inode *inode)
397 {
398 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
399 }
400 EXPORT_SYMBOL(ihold);
401
402 static void inode_lru_list_add(struct inode *inode)
403 {
404 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
405 this_cpu_inc(nr_unused);
406 else
407 inode->i_state |= I_REFERENCED;
408 }
409
410 /*
411 * Add inode to LRU if needed (inode is unused and clean).
412 *
413 * Needs inode->i_lock held.
414 */
415 void inode_add_lru(struct inode *inode)
416 {
417 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
418 I_FREEING | I_WILL_FREE)) &&
419 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
420 inode_lru_list_add(inode);
421 }
422
423
424 static void inode_lru_list_del(struct inode *inode)
425 {
426
427 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
428 this_cpu_dec(nr_unused);
429 }
430
431 /**
432 * inode_sb_list_add - add inode to the superblock list of inodes
433 * @inode: inode to add
434 */
435 void inode_sb_list_add(struct inode *inode)
436 {
437 spin_lock(&inode->i_sb->s_inode_list_lock);
438 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
439 spin_unlock(&inode->i_sb->s_inode_list_lock);
440 }
441 EXPORT_SYMBOL_GPL(inode_sb_list_add);
442
443 static inline void inode_sb_list_del(struct inode *inode)
444 {
445 if (!list_empty(&inode->i_sb_list)) {
446 spin_lock(&inode->i_sb->s_inode_list_lock);
447 list_del_init(&inode->i_sb_list);
448 spin_unlock(&inode->i_sb->s_inode_list_lock);
449 }
450 }
451
452 static unsigned long hash(struct super_block *sb, unsigned long hashval)
453 {
454 unsigned long tmp;
455
456 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
457 L1_CACHE_BYTES;
458 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
459 return tmp & i_hash_mask;
460 }
461
462 /**
463 * __insert_inode_hash - hash an inode
464 * @inode: unhashed inode
465 * @hashval: unsigned long value used to locate this object in the
466 * inode_hashtable.
467 *
468 * Add an inode to the inode hash for this superblock.
469 */
470 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
471 {
472 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
473
474 spin_lock(&inode_hash_lock);
475 spin_lock(&inode->i_lock);
476 hlist_add_head(&inode->i_hash, b);
477 spin_unlock(&inode->i_lock);
478 spin_unlock(&inode_hash_lock);
479 }
480 EXPORT_SYMBOL(__insert_inode_hash);
481
482 /**
483 * __remove_inode_hash - remove an inode from the hash
484 * @inode: inode to unhash
485 *
486 * Remove an inode from the superblock.
487 */
488 void __remove_inode_hash(struct inode *inode)
489 {
490 spin_lock(&inode_hash_lock);
491 spin_lock(&inode->i_lock);
492 hlist_del_init(&inode->i_hash);
493 spin_unlock(&inode->i_lock);
494 spin_unlock(&inode_hash_lock);
495 }
496 EXPORT_SYMBOL(__remove_inode_hash);
497
498 void clear_inode(struct inode *inode)
499 {
500 might_sleep();
501 /*
502 * We have to cycle tree_lock here because reclaim can be still in the
503 * process of removing the last page (in __delete_from_page_cache())
504 * and we must not free mapping under it.
505 */
506 spin_lock_irq(&inode->i_data.tree_lock);
507 BUG_ON(inode->i_data.nrpages);
508 BUG_ON(inode->i_data.nrexceptional);
509 spin_unlock_irq(&inode->i_data.tree_lock);
510 BUG_ON(!list_empty(&inode->i_data.private_list));
511 BUG_ON(!(inode->i_state & I_FREEING));
512 BUG_ON(inode->i_state & I_CLEAR);
513 BUG_ON(!list_empty(&inode->i_wb_list));
514 /* don't need i_lock here, no concurrent mods to i_state */
515 inode->i_state = I_FREEING | I_CLEAR;
516 }
517 EXPORT_SYMBOL(clear_inode);
518
519 /*
520 * Free the inode passed in, removing it from the lists it is still connected
521 * to. We remove any pages still attached to the inode and wait for any IO that
522 * is still in progress before finally destroying the inode.
523 *
524 * An inode must already be marked I_FREEING so that we avoid the inode being
525 * moved back onto lists if we race with other code that manipulates the lists
526 * (e.g. writeback_single_inode). The caller is responsible for setting this.
527 *
528 * An inode must already be removed from the LRU list before being evicted from
529 * the cache. This should occur atomically with setting the I_FREEING state
530 * flag, so no inodes here should ever be on the LRU when being evicted.
531 */
532 static void evict(struct inode *inode)
533 {
534 const struct super_operations *op = inode->i_sb->s_op;
535
536 BUG_ON(!(inode->i_state & I_FREEING));
537 BUG_ON(!list_empty(&inode->i_lru));
538
539 if (!list_empty(&inode->i_io_list))
540 inode_io_list_del(inode);
541
542 inode_sb_list_del(inode);
543
544 /*
545 * Wait for flusher thread to be done with the inode so that filesystem
546 * does not start destroying it while writeback is still running. Since
547 * the inode has I_FREEING set, flusher thread won't start new work on
548 * the inode. We just have to wait for running writeback to finish.
549 */
550 inode_wait_for_writeback(inode);
551
552 if (op->evict_inode) {
553 op->evict_inode(inode);
554 } else {
555 truncate_inode_pages_final(&inode->i_data);
556 clear_inode(inode);
557 }
558 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
559 bd_forget(inode);
560 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
561 cd_forget(inode);
562
563 remove_inode_hash(inode);
564
565 spin_lock(&inode->i_lock);
566 wake_up_bit(&inode->i_state, __I_NEW);
567 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
568 spin_unlock(&inode->i_lock);
569
570 destroy_inode(inode);
571 }
572
573 /*
574 * dispose_list - dispose of the contents of a local list
575 * @head: the head of the list to free
576 *
577 * Dispose-list gets a local list with local inodes in it, so it doesn't
578 * need to worry about list corruption and SMP locks.
579 */
580 static void dispose_list(struct list_head *head)
581 {
582 while (!list_empty(head)) {
583 struct inode *inode;
584
585 inode = list_first_entry(head, struct inode, i_lru);
586 list_del_init(&inode->i_lru);
587
588 evict(inode);
589 cond_resched();
590 }
591 }
592
593 /**
594 * evict_inodes - evict all evictable inodes for a superblock
595 * @sb: superblock to operate on
596 *
597 * Make sure that no inodes with zero refcount are retained. This is
598 * called by superblock shutdown after having MS_ACTIVE flag removed,
599 * so any inode reaching zero refcount during or after that call will
600 * be immediately evicted.
601 */
602 void evict_inodes(struct super_block *sb)
603 {
604 struct inode *inode, *next;
605 LIST_HEAD(dispose);
606
607 again:
608 spin_lock(&sb->s_inode_list_lock);
609 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
610 if (atomic_read(&inode->i_count))
611 continue;
612
613 spin_lock(&inode->i_lock);
614 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
615 spin_unlock(&inode->i_lock);
616 continue;
617 }
618
619 inode->i_state |= I_FREEING;
620 inode_lru_list_del(inode);
621 spin_unlock(&inode->i_lock);
622 list_add(&inode->i_lru, &dispose);
623
624 /*
625 * We can have a ton of inodes to evict at unmount time given
626 * enough memory, check to see if we need to go to sleep for a
627 * bit so we don't livelock.
628 */
629 if (need_resched()) {
630 spin_unlock(&sb->s_inode_list_lock);
631 cond_resched();
632 dispose_list(&dispose);
633 goto again;
634 }
635 }
636 spin_unlock(&sb->s_inode_list_lock);
637
638 dispose_list(&dispose);
639 }
640 EXPORT_SYMBOL_GPL(evict_inodes);
641
642 /**
643 * invalidate_inodes - attempt to free all inodes on a superblock
644 * @sb: superblock to operate on
645 * @kill_dirty: flag to guide handling of dirty inodes
646 *
647 * Attempts to free all inodes for a given superblock. If there were any
648 * busy inodes return a non-zero value, else zero.
649 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
650 * them as busy.
651 */
652 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
653 {
654 int busy = 0;
655 struct inode *inode, *next;
656 LIST_HEAD(dispose);
657
658 spin_lock(&sb->s_inode_list_lock);
659 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
660 spin_lock(&inode->i_lock);
661 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
662 spin_unlock(&inode->i_lock);
663 continue;
664 }
665 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
666 spin_unlock(&inode->i_lock);
667 busy = 1;
668 continue;
669 }
670 if (atomic_read(&inode->i_count)) {
671 spin_unlock(&inode->i_lock);
672 busy = 1;
673 continue;
674 }
675
676 inode->i_state |= I_FREEING;
677 inode_lru_list_del(inode);
678 spin_unlock(&inode->i_lock);
679 list_add(&inode->i_lru, &dispose);
680 }
681 spin_unlock(&sb->s_inode_list_lock);
682
683 dispose_list(&dispose);
684
685 return busy;
686 }
687
688 /*
689 * Isolate the inode from the LRU in preparation for freeing it.
690 *
691 * Any inodes which are pinned purely because of attached pagecache have their
692 * pagecache removed. If the inode has metadata buffers attached to
693 * mapping->private_list then try to remove them.
694 *
695 * If the inode has the I_REFERENCED flag set, then it means that it has been
696 * used recently - the flag is set in iput_final(). When we encounter such an
697 * inode, clear the flag and move it to the back of the LRU so it gets another
698 * pass through the LRU before it gets reclaimed. This is necessary because of
699 * the fact we are doing lazy LRU updates to minimise lock contention so the
700 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
701 * with this flag set because they are the inodes that are out of order.
702 */
703 static enum lru_status inode_lru_isolate(struct list_head *item,
704 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
705 {
706 struct list_head *freeable = arg;
707 struct inode *inode = container_of(item, struct inode, i_lru);
708
709 /*
710 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
711 * If we fail to get the lock, just skip it.
712 */
713 if (!spin_trylock(&inode->i_lock))
714 return LRU_SKIP;
715
716 /*
717 * Referenced or dirty inodes are still in use. Give them another pass
718 * through the LRU as we canot reclaim them now.
719 */
720 if (atomic_read(&inode->i_count) ||
721 (inode->i_state & ~I_REFERENCED)) {
722 list_lru_isolate(lru, &inode->i_lru);
723 spin_unlock(&inode->i_lock);
724 this_cpu_dec(nr_unused);
725 return LRU_REMOVED;
726 }
727
728 /* recently referenced inodes get one more pass */
729 if (inode->i_state & I_REFERENCED) {
730 inode->i_state &= ~I_REFERENCED;
731 spin_unlock(&inode->i_lock);
732 return LRU_ROTATE;
733 }
734
735 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
736 __iget(inode);
737 spin_unlock(&inode->i_lock);
738 spin_unlock(lru_lock);
739 if (remove_inode_buffers(inode)) {
740 unsigned long reap;
741 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
742 if (current_is_kswapd())
743 __count_vm_events(KSWAPD_INODESTEAL, reap);
744 else
745 __count_vm_events(PGINODESTEAL, reap);
746 if (current->reclaim_state)
747 current->reclaim_state->reclaimed_slab += reap;
748 }
749 iput(inode);
750 spin_lock(lru_lock);
751 return LRU_RETRY;
752 }
753
754 WARN_ON(inode->i_state & I_NEW);
755 inode->i_state |= I_FREEING;
756 list_lru_isolate_move(lru, &inode->i_lru, freeable);
757 spin_unlock(&inode->i_lock);
758
759 this_cpu_dec(nr_unused);
760 return LRU_REMOVED;
761 }
762
763 /*
764 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
765 * This is called from the superblock shrinker function with a number of inodes
766 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
767 * then are freed outside inode_lock by dispose_list().
768 */
769 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
770 {
771 LIST_HEAD(freeable);
772 long freed;
773
774 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
775 inode_lru_isolate, &freeable);
776 dispose_list(&freeable);
777 return freed;
778 }
779
780 static void __wait_on_freeing_inode(struct inode *inode);
781 /*
782 * Called with the inode lock held.
783 */
784 static struct inode *find_inode(struct super_block *sb,
785 struct hlist_head *head,
786 int (*test)(struct inode *, void *),
787 void *data)
788 {
789 struct inode *inode = NULL;
790
791 repeat:
792 hlist_for_each_entry(inode, head, i_hash) {
793 if (inode->i_sb != sb)
794 continue;
795 if (!test(inode, data))
796 continue;
797 spin_lock(&inode->i_lock);
798 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
799 __wait_on_freeing_inode(inode);
800 goto repeat;
801 }
802 __iget(inode);
803 spin_unlock(&inode->i_lock);
804 return inode;
805 }
806 return NULL;
807 }
808
809 /*
810 * find_inode_fast is the fast path version of find_inode, see the comment at
811 * iget_locked for details.
812 */
813 static struct inode *find_inode_fast(struct super_block *sb,
814 struct hlist_head *head, unsigned long ino)
815 {
816 struct inode *inode = NULL;
817
818 repeat:
819 hlist_for_each_entry(inode, head, i_hash) {
820 if (inode->i_ino != ino)
821 continue;
822 if (inode->i_sb != sb)
823 continue;
824 spin_lock(&inode->i_lock);
825 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
826 __wait_on_freeing_inode(inode);
827 goto repeat;
828 }
829 __iget(inode);
830 spin_unlock(&inode->i_lock);
831 return inode;
832 }
833 return NULL;
834 }
835
836 /*
837 * Each cpu owns a range of LAST_INO_BATCH numbers.
838 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
839 * to renew the exhausted range.
840 *
841 * This does not significantly increase overflow rate because every CPU can
842 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
843 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
844 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
845 * overflow rate by 2x, which does not seem too significant.
846 *
847 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
848 * error if st_ino won't fit in target struct field. Use 32bit counter
849 * here to attempt to avoid that.
850 */
851 #define LAST_INO_BATCH 1024
852 static DEFINE_PER_CPU(unsigned int, last_ino);
853
854 unsigned int get_next_ino(void)
855 {
856 unsigned int *p = &get_cpu_var(last_ino);
857 unsigned int res = *p;
858
859 #ifdef CONFIG_SMP
860 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
861 static atomic_t shared_last_ino;
862 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
863
864 res = next - LAST_INO_BATCH;
865 }
866 #endif
867
868 res++;
869 /* get_next_ino should not provide a 0 inode number */
870 if (unlikely(!res))
871 res++;
872 *p = res;
873 put_cpu_var(last_ino);
874 return res;
875 }
876 EXPORT_SYMBOL(get_next_ino);
877
878 /**
879 * new_inode_pseudo - obtain an inode
880 * @sb: superblock
881 *
882 * Allocates a new inode for given superblock.
883 * Inode wont be chained in superblock s_inodes list
884 * This means :
885 * - fs can't be unmount
886 * - quotas, fsnotify, writeback can't work
887 */
888 struct inode *new_inode_pseudo(struct super_block *sb)
889 {
890 struct inode *inode = alloc_inode(sb);
891
892 if (inode) {
893 spin_lock(&inode->i_lock);
894 inode->i_state = 0;
895 spin_unlock(&inode->i_lock);
896 INIT_LIST_HEAD(&inode->i_sb_list);
897 }
898 return inode;
899 }
900
901 /**
902 * new_inode - obtain an inode
903 * @sb: superblock
904 *
905 * Allocates a new inode for given superblock. The default gfp_mask
906 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
907 * If HIGHMEM pages are unsuitable or it is known that pages allocated
908 * for the page cache are not reclaimable or migratable,
909 * mapping_set_gfp_mask() must be called with suitable flags on the
910 * newly created inode's mapping
911 *
912 */
913 struct inode *new_inode(struct super_block *sb)
914 {
915 struct inode *inode;
916
917 spin_lock_prefetch(&sb->s_inode_list_lock);
918
919 inode = new_inode_pseudo(sb);
920 if (inode)
921 inode_sb_list_add(inode);
922 return inode;
923 }
924 EXPORT_SYMBOL(new_inode);
925
926 #ifdef CONFIG_DEBUG_LOCK_ALLOC
927 void lockdep_annotate_inode_mutex_key(struct inode *inode)
928 {
929 if (S_ISDIR(inode->i_mode)) {
930 struct file_system_type *type = inode->i_sb->s_type;
931
932 /* Set new key only if filesystem hasn't already changed it */
933 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
934 /*
935 * ensure nobody is actually holding i_mutex
936 */
937 // mutex_destroy(&inode->i_mutex);
938 init_rwsem(&inode->i_rwsem);
939 lockdep_set_class(&inode->i_rwsem,
940 &type->i_mutex_dir_key);
941 }
942 }
943 }
944 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
945 #endif
946
947 /**
948 * unlock_new_inode - clear the I_NEW state and wake up any waiters
949 * @inode: new inode to unlock
950 *
951 * Called when the inode is fully initialised to clear the new state of the
952 * inode and wake up anyone waiting for the inode to finish initialisation.
953 */
954 void unlock_new_inode(struct inode *inode)
955 {
956 lockdep_annotate_inode_mutex_key(inode);
957 spin_lock(&inode->i_lock);
958 WARN_ON(!(inode->i_state & I_NEW));
959 inode->i_state &= ~I_NEW;
960 smp_mb();
961 wake_up_bit(&inode->i_state, __I_NEW);
962 spin_unlock(&inode->i_lock);
963 }
964 EXPORT_SYMBOL(unlock_new_inode);
965
966 /**
967 * lock_two_nondirectories - take two i_mutexes on non-directory objects
968 *
969 * Lock any non-NULL argument that is not a directory.
970 * Zero, one or two objects may be locked by this function.
971 *
972 * @inode1: first inode to lock
973 * @inode2: second inode to lock
974 */
975 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
976 {
977 if (inode1 > inode2)
978 swap(inode1, inode2);
979
980 if (inode1 && !S_ISDIR(inode1->i_mode))
981 inode_lock(inode1);
982 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
983 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
984 }
985 EXPORT_SYMBOL(lock_two_nondirectories);
986
987 /**
988 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
989 * @inode1: first inode to unlock
990 * @inode2: second inode to unlock
991 */
992 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
993 {
994 if (inode1 && !S_ISDIR(inode1->i_mode))
995 inode_unlock(inode1);
996 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
997 inode_unlock(inode2);
998 }
999 EXPORT_SYMBOL(unlock_two_nondirectories);
1000
1001 /**
1002 * iget5_locked - obtain an inode from a mounted file system
1003 * @sb: super block of file system
1004 * @hashval: hash value (usually inode number) to get
1005 * @test: callback used for comparisons between inodes
1006 * @set: callback used to initialize a new struct inode
1007 * @data: opaque data pointer to pass to @test and @set
1008 *
1009 * Search for the inode specified by @hashval and @data in the inode cache,
1010 * and if present it is return it with an increased reference count. This is
1011 * a generalized version of iget_locked() for file systems where the inode
1012 * number is not sufficient for unique identification of an inode.
1013 *
1014 * If the inode is not in cache, allocate a new inode and return it locked,
1015 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1016 * before unlocking it via unlock_new_inode().
1017 *
1018 * Note both @test and @set are called with the inode_hash_lock held, so can't
1019 * sleep.
1020 */
1021 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1022 int (*test)(struct inode *, void *),
1023 int (*set)(struct inode *, void *), void *data)
1024 {
1025 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1026 struct inode *inode;
1027 again:
1028 spin_lock(&inode_hash_lock);
1029 inode = find_inode(sb, head, test, data);
1030 spin_unlock(&inode_hash_lock);
1031
1032 if (inode) {
1033 wait_on_inode(inode);
1034 if (unlikely(inode_unhashed(inode))) {
1035 iput(inode);
1036 goto again;
1037 }
1038 return inode;
1039 }
1040
1041 inode = alloc_inode(sb);
1042 if (inode) {
1043 struct inode *old;
1044
1045 spin_lock(&inode_hash_lock);
1046 /* We released the lock, so.. */
1047 old = find_inode(sb, head, test, data);
1048 if (!old) {
1049 if (set(inode, data))
1050 goto set_failed;
1051
1052 spin_lock(&inode->i_lock);
1053 inode->i_state = I_NEW;
1054 hlist_add_head(&inode->i_hash, head);
1055 spin_unlock(&inode->i_lock);
1056 inode_sb_list_add(inode);
1057 spin_unlock(&inode_hash_lock);
1058
1059 /* Return the locked inode with I_NEW set, the
1060 * caller is responsible for filling in the contents
1061 */
1062 return inode;
1063 }
1064
1065 /*
1066 * Uhhuh, somebody else created the same inode under
1067 * us. Use the old inode instead of the one we just
1068 * allocated.
1069 */
1070 spin_unlock(&inode_hash_lock);
1071 destroy_inode(inode);
1072 inode = old;
1073 wait_on_inode(inode);
1074 if (unlikely(inode_unhashed(inode))) {
1075 iput(inode);
1076 goto again;
1077 }
1078 }
1079 return inode;
1080
1081 set_failed:
1082 spin_unlock(&inode_hash_lock);
1083 destroy_inode(inode);
1084 return NULL;
1085 }
1086 EXPORT_SYMBOL(iget5_locked);
1087
1088 /**
1089 * iget_locked - obtain an inode from a mounted file system
1090 * @sb: super block of file system
1091 * @ino: inode number to get
1092 *
1093 * Search for the inode specified by @ino in the inode cache and if present
1094 * return it with an increased reference count. This is for file systems
1095 * where the inode number is sufficient for unique identification of an inode.
1096 *
1097 * If the inode is not in cache, allocate a new inode and return it locked,
1098 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1099 * before unlocking it via unlock_new_inode().
1100 */
1101 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1102 {
1103 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1104 struct inode *inode;
1105 again:
1106 spin_lock(&inode_hash_lock);
1107 inode = find_inode_fast(sb, head, ino);
1108 spin_unlock(&inode_hash_lock);
1109 if (inode) {
1110 wait_on_inode(inode);
1111 if (unlikely(inode_unhashed(inode))) {
1112 iput(inode);
1113 goto again;
1114 }
1115 return inode;
1116 }
1117
1118 inode = alloc_inode(sb);
1119 if (inode) {
1120 struct inode *old;
1121
1122 spin_lock(&inode_hash_lock);
1123 /* We released the lock, so.. */
1124 old = find_inode_fast(sb, head, ino);
1125 if (!old) {
1126 inode->i_ino = ino;
1127 spin_lock(&inode->i_lock);
1128 inode->i_state = I_NEW;
1129 hlist_add_head(&inode->i_hash, head);
1130 spin_unlock(&inode->i_lock);
1131 inode_sb_list_add(inode);
1132 spin_unlock(&inode_hash_lock);
1133
1134 /* Return the locked inode with I_NEW set, the
1135 * caller is responsible for filling in the contents
1136 */
1137 return inode;
1138 }
1139
1140 /*
1141 * Uhhuh, somebody else created the same inode under
1142 * us. Use the old inode instead of the one we just
1143 * allocated.
1144 */
1145 spin_unlock(&inode_hash_lock);
1146 destroy_inode(inode);
1147 inode = old;
1148 wait_on_inode(inode);
1149 if (unlikely(inode_unhashed(inode))) {
1150 iput(inode);
1151 goto again;
1152 }
1153 }
1154 return inode;
1155 }
1156 EXPORT_SYMBOL(iget_locked);
1157
1158 /*
1159 * search the inode cache for a matching inode number.
1160 * If we find one, then the inode number we are trying to
1161 * allocate is not unique and so we should not use it.
1162 *
1163 * Returns 1 if the inode number is unique, 0 if it is not.
1164 */
1165 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1166 {
1167 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1168 struct inode *inode;
1169
1170 spin_lock(&inode_hash_lock);
1171 hlist_for_each_entry(inode, b, i_hash) {
1172 if (inode->i_ino == ino && inode->i_sb == sb) {
1173 spin_unlock(&inode_hash_lock);
1174 return 0;
1175 }
1176 }
1177 spin_unlock(&inode_hash_lock);
1178
1179 return 1;
1180 }
1181
1182 /**
1183 * iunique - get a unique inode number
1184 * @sb: superblock
1185 * @max_reserved: highest reserved inode number
1186 *
1187 * Obtain an inode number that is unique on the system for a given
1188 * superblock. This is used by file systems that have no natural
1189 * permanent inode numbering system. An inode number is returned that
1190 * is higher than the reserved limit but unique.
1191 *
1192 * BUGS:
1193 * With a large number of inodes live on the file system this function
1194 * currently becomes quite slow.
1195 */
1196 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1197 {
1198 /*
1199 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1200 * error if st_ino won't fit in target struct field. Use 32bit counter
1201 * here to attempt to avoid that.
1202 */
1203 static DEFINE_SPINLOCK(iunique_lock);
1204 static unsigned int counter;
1205 ino_t res;
1206
1207 spin_lock(&iunique_lock);
1208 do {
1209 if (counter <= max_reserved)
1210 counter = max_reserved + 1;
1211 res = counter++;
1212 } while (!test_inode_iunique(sb, res));
1213 spin_unlock(&iunique_lock);
1214
1215 return res;
1216 }
1217 EXPORT_SYMBOL(iunique);
1218
1219 struct inode *igrab(struct inode *inode)
1220 {
1221 spin_lock(&inode->i_lock);
1222 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1223 __iget(inode);
1224 spin_unlock(&inode->i_lock);
1225 } else {
1226 spin_unlock(&inode->i_lock);
1227 /*
1228 * Handle the case where s_op->clear_inode is not been
1229 * called yet, and somebody is calling igrab
1230 * while the inode is getting freed.
1231 */
1232 inode = NULL;
1233 }
1234 return inode;
1235 }
1236 EXPORT_SYMBOL(igrab);
1237
1238 /**
1239 * ilookup5_nowait - search for an inode in the inode cache
1240 * @sb: super block of file system to search
1241 * @hashval: hash value (usually inode number) to search for
1242 * @test: callback used for comparisons between inodes
1243 * @data: opaque data pointer to pass to @test
1244 *
1245 * Search for the inode specified by @hashval and @data in the inode cache.
1246 * If the inode is in the cache, the inode is returned with an incremented
1247 * reference count.
1248 *
1249 * Note: I_NEW is not waited upon so you have to be very careful what you do
1250 * with the returned inode. You probably should be using ilookup5() instead.
1251 *
1252 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1253 */
1254 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1255 int (*test)(struct inode *, void *), void *data)
1256 {
1257 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1258 struct inode *inode;
1259
1260 spin_lock(&inode_hash_lock);
1261 inode = find_inode(sb, head, test, data);
1262 spin_unlock(&inode_hash_lock);
1263
1264 return inode;
1265 }
1266 EXPORT_SYMBOL(ilookup5_nowait);
1267
1268 /**
1269 * ilookup5 - search for an inode in the inode cache
1270 * @sb: super block of file system to search
1271 * @hashval: hash value (usually inode number) to search for
1272 * @test: callback used for comparisons between inodes
1273 * @data: opaque data pointer to pass to @test
1274 *
1275 * Search for the inode specified by @hashval and @data in the inode cache,
1276 * and if the inode is in the cache, return the inode with an incremented
1277 * reference count. Waits on I_NEW before returning the inode.
1278 * returned with an incremented reference count.
1279 *
1280 * This is a generalized version of ilookup() for file systems where the
1281 * inode number is not sufficient for unique identification of an inode.
1282 *
1283 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1284 */
1285 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1286 int (*test)(struct inode *, void *), void *data)
1287 {
1288 struct inode *inode;
1289 again:
1290 inode = ilookup5_nowait(sb, hashval, test, data);
1291 if (inode) {
1292 wait_on_inode(inode);
1293 if (unlikely(inode_unhashed(inode))) {
1294 iput(inode);
1295 goto again;
1296 }
1297 }
1298 return inode;
1299 }
1300 EXPORT_SYMBOL(ilookup5);
1301
1302 /**
1303 * ilookup - search for an inode in the inode cache
1304 * @sb: super block of file system to search
1305 * @ino: inode number to search for
1306 *
1307 * Search for the inode @ino in the inode cache, and if the inode is in the
1308 * cache, the inode is returned with an incremented reference count.
1309 */
1310 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1311 {
1312 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1313 struct inode *inode;
1314 again:
1315 spin_lock(&inode_hash_lock);
1316 inode = find_inode_fast(sb, head, ino);
1317 spin_unlock(&inode_hash_lock);
1318
1319 if (inode) {
1320 wait_on_inode(inode);
1321 if (unlikely(inode_unhashed(inode))) {
1322 iput(inode);
1323 goto again;
1324 }
1325 }
1326 return inode;
1327 }
1328 EXPORT_SYMBOL(ilookup);
1329
1330 /**
1331 * find_inode_nowait - find an inode in the inode cache
1332 * @sb: super block of file system to search
1333 * @hashval: hash value (usually inode number) to search for
1334 * @match: callback used for comparisons between inodes
1335 * @data: opaque data pointer to pass to @match
1336 *
1337 * Search for the inode specified by @hashval and @data in the inode
1338 * cache, where the helper function @match will return 0 if the inode
1339 * does not match, 1 if the inode does match, and -1 if the search
1340 * should be stopped. The @match function must be responsible for
1341 * taking the i_lock spin_lock and checking i_state for an inode being
1342 * freed or being initialized, and incrementing the reference count
1343 * before returning 1. It also must not sleep, since it is called with
1344 * the inode_hash_lock spinlock held.
1345 *
1346 * This is a even more generalized version of ilookup5() when the
1347 * function must never block --- find_inode() can block in
1348 * __wait_on_freeing_inode() --- or when the caller can not increment
1349 * the reference count because the resulting iput() might cause an
1350 * inode eviction. The tradeoff is that the @match funtion must be
1351 * very carefully implemented.
1352 */
1353 struct inode *find_inode_nowait(struct super_block *sb,
1354 unsigned long hashval,
1355 int (*match)(struct inode *, unsigned long,
1356 void *),
1357 void *data)
1358 {
1359 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1360 struct inode *inode, *ret_inode = NULL;
1361 int mval;
1362
1363 spin_lock(&inode_hash_lock);
1364 hlist_for_each_entry(inode, head, i_hash) {
1365 if (inode->i_sb != sb)
1366 continue;
1367 mval = match(inode, hashval, data);
1368 if (mval == 0)
1369 continue;
1370 if (mval == 1)
1371 ret_inode = inode;
1372 goto out;
1373 }
1374 out:
1375 spin_unlock(&inode_hash_lock);
1376 return ret_inode;
1377 }
1378 EXPORT_SYMBOL(find_inode_nowait);
1379
1380 int insert_inode_locked(struct inode *inode)
1381 {
1382 struct super_block *sb = inode->i_sb;
1383 ino_t ino = inode->i_ino;
1384 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1385
1386 while (1) {
1387 struct inode *old = NULL;
1388 spin_lock(&inode_hash_lock);
1389 hlist_for_each_entry(old, head, i_hash) {
1390 if (old->i_ino != ino)
1391 continue;
1392 if (old->i_sb != sb)
1393 continue;
1394 spin_lock(&old->i_lock);
1395 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1396 spin_unlock(&old->i_lock);
1397 continue;
1398 }
1399 break;
1400 }
1401 if (likely(!old)) {
1402 spin_lock(&inode->i_lock);
1403 inode->i_state |= I_NEW;
1404 hlist_add_head(&inode->i_hash, head);
1405 spin_unlock(&inode->i_lock);
1406 spin_unlock(&inode_hash_lock);
1407 return 0;
1408 }
1409 __iget(old);
1410 spin_unlock(&old->i_lock);
1411 spin_unlock(&inode_hash_lock);
1412 wait_on_inode(old);
1413 if (unlikely(!inode_unhashed(old))) {
1414 iput(old);
1415 return -EBUSY;
1416 }
1417 iput(old);
1418 }
1419 }
1420 EXPORT_SYMBOL(insert_inode_locked);
1421
1422 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1423 int (*test)(struct inode *, void *), void *data)
1424 {
1425 struct super_block *sb = inode->i_sb;
1426 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1427
1428 while (1) {
1429 struct inode *old = NULL;
1430
1431 spin_lock(&inode_hash_lock);
1432 hlist_for_each_entry(old, head, i_hash) {
1433 if (old->i_sb != sb)
1434 continue;
1435 if (!test(old, data))
1436 continue;
1437 spin_lock(&old->i_lock);
1438 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1439 spin_unlock(&old->i_lock);
1440 continue;
1441 }
1442 break;
1443 }
1444 if (likely(!old)) {
1445 spin_lock(&inode->i_lock);
1446 inode->i_state |= I_NEW;
1447 hlist_add_head(&inode->i_hash, head);
1448 spin_unlock(&inode->i_lock);
1449 spin_unlock(&inode_hash_lock);
1450 return 0;
1451 }
1452 __iget(old);
1453 spin_unlock(&old->i_lock);
1454 spin_unlock(&inode_hash_lock);
1455 wait_on_inode(old);
1456 if (unlikely(!inode_unhashed(old))) {
1457 iput(old);
1458 return -EBUSY;
1459 }
1460 iput(old);
1461 }
1462 }
1463 EXPORT_SYMBOL(insert_inode_locked4);
1464
1465
1466 int generic_delete_inode(struct inode *inode)
1467 {
1468 return 1;
1469 }
1470 EXPORT_SYMBOL(generic_delete_inode);
1471
1472 /*
1473 * Called when we're dropping the last reference
1474 * to an inode.
1475 *
1476 * Call the FS "drop_inode()" function, defaulting to
1477 * the legacy UNIX filesystem behaviour. If it tells
1478 * us to evict inode, do so. Otherwise, retain inode
1479 * in cache if fs is alive, sync and evict if fs is
1480 * shutting down.
1481 */
1482 static void iput_final(struct inode *inode)
1483 {
1484 struct super_block *sb = inode->i_sb;
1485 const struct super_operations *op = inode->i_sb->s_op;
1486 int drop;
1487
1488 WARN_ON(inode->i_state & I_NEW);
1489
1490 if (op->drop_inode)
1491 drop = op->drop_inode(inode);
1492 else
1493 drop = generic_drop_inode(inode);
1494
1495 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1496 inode_add_lru(inode);
1497 spin_unlock(&inode->i_lock);
1498 return;
1499 }
1500
1501 if (!drop) {
1502 inode->i_state |= I_WILL_FREE;
1503 spin_unlock(&inode->i_lock);
1504 write_inode_now(inode, 1);
1505 spin_lock(&inode->i_lock);
1506 WARN_ON(inode->i_state & I_NEW);
1507 inode->i_state &= ~I_WILL_FREE;
1508 }
1509
1510 inode->i_state |= I_FREEING;
1511 if (!list_empty(&inode->i_lru))
1512 inode_lru_list_del(inode);
1513 spin_unlock(&inode->i_lock);
1514
1515 evict(inode);
1516 }
1517
1518 /**
1519 * iput - put an inode
1520 * @inode: inode to put
1521 *
1522 * Puts an inode, dropping its usage count. If the inode use count hits
1523 * zero, the inode is then freed and may also be destroyed.
1524 *
1525 * Consequently, iput() can sleep.
1526 */
1527 void iput(struct inode *inode)
1528 {
1529 if (!inode)
1530 return;
1531 BUG_ON(inode->i_state & I_CLEAR);
1532 retry:
1533 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1534 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1535 atomic_inc(&inode->i_count);
1536 inode->i_state &= ~I_DIRTY_TIME;
1537 spin_unlock(&inode->i_lock);
1538 trace_writeback_lazytime_iput(inode);
1539 mark_inode_dirty_sync(inode);
1540 goto retry;
1541 }
1542 iput_final(inode);
1543 }
1544 }
1545 EXPORT_SYMBOL(iput);
1546
1547 /**
1548 * bmap - find a block number in a file
1549 * @inode: inode of file
1550 * @block: block to find
1551 *
1552 * Returns the block number on the device holding the inode that
1553 * is the disk block number for the block of the file requested.
1554 * That is, asked for block 4 of inode 1 the function will return the
1555 * disk block relative to the disk start that holds that block of the
1556 * file.
1557 */
1558 sector_t bmap(struct inode *inode, sector_t block)
1559 {
1560 sector_t res = 0;
1561 if (inode->i_mapping->a_ops->bmap)
1562 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1563 return res;
1564 }
1565 EXPORT_SYMBOL(bmap);
1566
1567 /*
1568 * Update times in overlayed inode from underlying real inode
1569 */
1570 static void update_ovl_inode_times(struct dentry *dentry, struct inode *inode,
1571 bool rcu)
1572 {
1573 if (!rcu) {
1574 struct inode *realinode = d_real_inode(dentry);
1575
1576 if (unlikely(inode != realinode) &&
1577 (!timespec_equal(&inode->i_mtime, &realinode->i_mtime) ||
1578 !timespec_equal(&inode->i_ctime, &realinode->i_ctime))) {
1579 inode->i_mtime = realinode->i_mtime;
1580 inode->i_ctime = realinode->i_ctime;
1581 }
1582 }
1583 }
1584
1585 /*
1586 * With relative atime, only update atime if the previous atime is
1587 * earlier than either the ctime or mtime or if at least a day has
1588 * passed since the last atime update.
1589 */
1590 static int relatime_need_update(const struct path *path, struct inode *inode,
1591 struct timespec now, bool rcu)
1592 {
1593
1594 if (!(path->mnt->mnt_flags & MNT_RELATIME))
1595 return 1;
1596
1597 update_ovl_inode_times(path->dentry, inode, rcu);
1598 /*
1599 * Is mtime younger than atime? If yes, update atime:
1600 */
1601 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1602 return 1;
1603 /*
1604 * Is ctime younger than atime? If yes, update atime:
1605 */
1606 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1607 return 1;
1608
1609 /*
1610 * Is the previous atime value older than a day? If yes,
1611 * update atime:
1612 */
1613 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1614 return 1;
1615 /*
1616 * Good, we can skip the atime update:
1617 */
1618 return 0;
1619 }
1620
1621 int generic_update_time(struct inode *inode, struct timespec *time, int flags)
1622 {
1623 int iflags = I_DIRTY_TIME;
1624
1625 if (flags & S_ATIME)
1626 inode->i_atime = *time;
1627 if (flags & S_VERSION)
1628 inode_inc_iversion(inode);
1629 if (flags & S_CTIME)
1630 inode->i_ctime = *time;
1631 if (flags & S_MTIME)
1632 inode->i_mtime = *time;
1633
1634 if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION))
1635 iflags |= I_DIRTY_SYNC;
1636 __mark_inode_dirty(inode, iflags);
1637 return 0;
1638 }
1639 EXPORT_SYMBOL(generic_update_time);
1640
1641 /*
1642 * This does the actual work of updating an inodes time or version. Must have
1643 * had called mnt_want_write() before calling this.
1644 */
1645 int update_time(struct inode *inode, struct timespec *time, int flags)
1646 {
1647 int (*update_time)(struct inode *, struct timespec *, int);
1648
1649 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1650 generic_update_time;
1651
1652 return update_time(inode, time, flags);
1653 }
1654 EXPORT_SYMBOL_GPL(update_time);
1655
1656 /**
1657 * touch_atime - update the access time
1658 * @path: the &struct path to update
1659 * @inode: inode to update
1660 *
1661 * Update the accessed time on an inode and mark it for writeback.
1662 * This function automatically handles read only file systems and media,
1663 * as well as the "noatime" flag and inode specific "noatime" markers.
1664 */
1665 bool __atime_needs_update(const struct path *path, struct inode *inode,
1666 bool rcu)
1667 {
1668 struct vfsmount *mnt = path->mnt;
1669 struct timespec now;
1670
1671 if (inode->i_flags & S_NOATIME)
1672 return false;
1673
1674 /* Atime updates will likely cause i_uid and i_gid to be written
1675 * back improprely if their true value is unknown to the vfs.
1676 */
1677 if (HAS_UNMAPPED_ID(inode))
1678 return false;
1679
1680 if (IS_NOATIME(inode))
1681 return false;
1682 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1683 return false;
1684
1685 if (mnt->mnt_flags & MNT_NOATIME)
1686 return false;
1687 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1688 return false;
1689
1690 now = current_time(inode);
1691
1692 if (!relatime_need_update(path, inode, now, rcu))
1693 return false;
1694
1695 if (timespec_equal(&inode->i_atime, &now))
1696 return false;
1697
1698 return true;
1699 }
1700
1701 void touch_atime(const struct path *path)
1702 {
1703 struct vfsmount *mnt = path->mnt;
1704 struct inode *inode = d_inode(path->dentry);
1705 struct timespec now;
1706
1707 if (!__atime_needs_update(path, inode, false))
1708 return;
1709
1710 if (!sb_start_write_trylock(inode->i_sb))
1711 return;
1712
1713 if (__mnt_want_write(mnt) != 0)
1714 goto skip_update;
1715 /*
1716 * File systems can error out when updating inodes if they need to
1717 * allocate new space to modify an inode (such is the case for
1718 * Btrfs), but since we touch atime while walking down the path we
1719 * really don't care if we failed to update the atime of the file,
1720 * so just ignore the return value.
1721 * We may also fail on filesystems that have the ability to make parts
1722 * of the fs read only, e.g. subvolumes in Btrfs.
1723 */
1724 now = current_time(inode);
1725 update_time(inode, &now, S_ATIME);
1726 __mnt_drop_write(mnt);
1727 skip_update:
1728 sb_end_write(inode->i_sb);
1729 }
1730 EXPORT_SYMBOL(touch_atime);
1731
1732 /*
1733 * The logic we want is
1734 *
1735 * if suid or (sgid and xgrp)
1736 * remove privs
1737 */
1738 int should_remove_suid(struct dentry *dentry)
1739 {
1740 struct inode *inode = d_inode(dentry);
1741 umode_t mode = inode->i_mode;
1742 int kill = 0;
1743
1744 /* suid always must be killed */
1745 if (unlikely(mode & S_ISUID))
1746 kill = ATTR_KILL_SUID;
1747
1748 /*
1749 * sgid without any exec bits is just a mandatory locking mark; leave
1750 * it alone. If some exec bits are set, it's a real sgid; kill it.
1751 */
1752 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1753 kill |= ATTR_KILL_SGID;
1754
1755 if (unlikely(kill && !capable_wrt_inode_uidgid(inode, CAP_FSETID) &&
1756 S_ISREG(mode)))
1757 return kill;
1758
1759 return 0;
1760 }
1761 EXPORT_SYMBOL(should_remove_suid);
1762
1763 /*
1764 * Return mask of changes for notify_change() that need to be done as a
1765 * response to write or truncate. Return 0 if nothing has to be changed.
1766 * Negative value on error (change should be denied).
1767 */
1768 int dentry_needs_remove_privs(struct dentry *dentry)
1769 {
1770 struct inode *inode = d_inode(dentry);
1771 int mask = 0;
1772 int ret;
1773
1774 if (IS_NOSEC(inode))
1775 return 0;
1776
1777 mask = should_remove_suid(dentry);
1778 ret = security_inode_need_killpriv(dentry);
1779 if (ret < 0)
1780 return ret;
1781 if (ret)
1782 mask |= ATTR_KILL_PRIV;
1783 return mask;
1784 }
1785
1786 static int __remove_privs(struct dentry *dentry, int kill)
1787 {
1788 struct iattr newattrs;
1789
1790 newattrs.ia_valid = ATTR_FORCE | kill;
1791 /*
1792 * Note we call this on write, so notify_change will not
1793 * encounter any conflicting delegations:
1794 */
1795 return notify_change(dentry, &newattrs, NULL);
1796 }
1797
1798 /*
1799 * Remove special file priviledges (suid, capabilities) when file is written
1800 * to or truncated.
1801 */
1802 int file_remove_privs(struct file *file)
1803 {
1804 struct dentry *dentry = file_dentry(file);
1805 struct inode *inode = file_inode(file);
1806 int kill;
1807 int error = 0;
1808
1809 /* Fast path for nothing security related */
1810 if (IS_NOSEC(inode))
1811 return 0;
1812
1813 kill = dentry_needs_remove_privs(dentry);
1814 if (kill < 0)
1815 return kill;
1816 if (kill)
1817 error = __remove_privs(dentry, kill);
1818 if (!error)
1819 inode_has_no_xattr(inode);
1820
1821 return error;
1822 }
1823 EXPORT_SYMBOL(file_remove_privs);
1824
1825 /**
1826 * file_update_time - update mtime and ctime time
1827 * @file: file accessed
1828 *
1829 * Update the mtime and ctime members of an inode and mark the inode
1830 * for writeback. Note that this function is meant exclusively for
1831 * usage in the file write path of filesystems, and filesystems may
1832 * choose to explicitly ignore update via this function with the
1833 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1834 * timestamps are handled by the server. This can return an error for
1835 * file systems who need to allocate space in order to update an inode.
1836 */
1837
1838 int file_update_time(struct file *file)
1839 {
1840 struct inode *inode = file_inode(file);
1841 struct timespec now;
1842 int sync_it = 0;
1843 int ret;
1844
1845 /* First try to exhaust all avenues to not sync */
1846 if (IS_NOCMTIME(inode))
1847 return 0;
1848
1849 now = current_time(inode);
1850 if (!timespec_equal(&inode->i_mtime, &now))
1851 sync_it = S_MTIME;
1852
1853 if (!timespec_equal(&inode->i_ctime, &now))
1854 sync_it |= S_CTIME;
1855
1856 if (IS_I_VERSION(inode))
1857 sync_it |= S_VERSION;
1858
1859 if (!sync_it)
1860 return 0;
1861
1862 /* Finally allowed to write? Takes lock. */
1863 if (__mnt_want_write_file(file))
1864 return 0;
1865
1866 ret = update_time(inode, &now, sync_it);
1867 __mnt_drop_write_file(file);
1868
1869 return ret;
1870 }
1871 EXPORT_SYMBOL(file_update_time);
1872
1873 int inode_needs_sync(struct inode *inode)
1874 {
1875 if (IS_SYNC(inode))
1876 return 1;
1877 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1878 return 1;
1879 return 0;
1880 }
1881 EXPORT_SYMBOL(inode_needs_sync);
1882
1883 /*
1884 * If we try to find an inode in the inode hash while it is being
1885 * deleted, we have to wait until the filesystem completes its
1886 * deletion before reporting that it isn't found. This function waits
1887 * until the deletion _might_ have completed. Callers are responsible
1888 * to recheck inode state.
1889 *
1890 * It doesn't matter if I_NEW is not set initially, a call to
1891 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1892 * will DTRT.
1893 */
1894 static void __wait_on_freeing_inode(struct inode *inode)
1895 {
1896 wait_queue_head_t *wq;
1897 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1898 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1899 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1900 spin_unlock(&inode->i_lock);
1901 spin_unlock(&inode_hash_lock);
1902 schedule();
1903 finish_wait(wq, &wait.wq_entry);
1904 spin_lock(&inode_hash_lock);
1905 }
1906
1907 static __initdata unsigned long ihash_entries;
1908 static int __init set_ihash_entries(char *str)
1909 {
1910 if (!str)
1911 return 0;
1912 ihash_entries = simple_strtoul(str, &str, 0);
1913 return 1;
1914 }
1915 __setup("ihash_entries=", set_ihash_entries);
1916
1917 /*
1918 * Initialize the waitqueues and inode hash table.
1919 */
1920 void __init inode_init_early(void)
1921 {
1922 /* If hashes are distributed across NUMA nodes, defer
1923 * hash allocation until vmalloc space is available.
1924 */
1925 if (hashdist)
1926 return;
1927
1928 inode_hashtable =
1929 alloc_large_system_hash("Inode-cache",
1930 sizeof(struct hlist_head),
1931 ihash_entries,
1932 14,
1933 HASH_EARLY | HASH_ZERO,
1934 &i_hash_shift,
1935 &i_hash_mask,
1936 0,
1937 0);
1938 }
1939
1940 void __init inode_init(void)
1941 {
1942 /* inode slab cache */
1943 inode_cachep = kmem_cache_create("inode_cache",
1944 sizeof(struct inode),
1945 0,
1946 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1947 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1948 init_once);
1949
1950 /* Hash may have been set up in inode_init_early */
1951 if (!hashdist)
1952 return;
1953
1954 inode_hashtable =
1955 alloc_large_system_hash("Inode-cache",
1956 sizeof(struct hlist_head),
1957 ihash_entries,
1958 14,
1959 HASH_ZERO,
1960 &i_hash_shift,
1961 &i_hash_mask,
1962 0,
1963 0);
1964 }
1965
1966 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1967 {
1968 inode->i_mode = mode;
1969 if (S_ISCHR(mode)) {
1970 inode->i_fop = &def_chr_fops;
1971 inode->i_rdev = rdev;
1972 } else if (S_ISBLK(mode)) {
1973 inode->i_fop = &def_blk_fops;
1974 inode->i_rdev = rdev;
1975 } else if (S_ISFIFO(mode))
1976 inode->i_fop = &pipefifo_fops;
1977 else if (S_ISSOCK(mode))
1978 ; /* leave it no_open_fops */
1979 else
1980 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1981 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1982 inode->i_ino);
1983 }
1984 EXPORT_SYMBOL(init_special_inode);
1985
1986 /**
1987 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1988 * @inode: New inode
1989 * @dir: Directory inode
1990 * @mode: mode of the new inode
1991 */
1992 void inode_init_owner(struct inode *inode, const struct inode *dir,
1993 umode_t mode)
1994 {
1995 inode->i_uid = current_fsuid();
1996 if (dir && dir->i_mode & S_ISGID) {
1997 inode->i_gid = dir->i_gid;
1998 if (S_ISDIR(mode))
1999 mode |= S_ISGID;
2000 } else
2001 inode->i_gid = current_fsgid();
2002 inode->i_mode = mode;
2003 }
2004 EXPORT_SYMBOL(inode_init_owner);
2005
2006 /**
2007 * inode_owner_or_capable - check current task permissions to inode
2008 * @inode: inode being checked
2009 *
2010 * Return true if current either has CAP_FOWNER in a namespace with the
2011 * inode owner uid mapped, or owns the file.
2012 */
2013 bool inode_owner_or_capable(const struct inode *inode)
2014 {
2015 struct user_namespace *ns;
2016
2017 if (uid_eq(current_fsuid(), inode->i_uid))
2018 return true;
2019
2020 ns = current_user_ns();
2021 if (kuid_has_mapping(ns, inode->i_uid) && ns_capable(ns, CAP_FOWNER))
2022 return true;
2023 return false;
2024 }
2025 EXPORT_SYMBOL(inode_owner_or_capable);
2026
2027 /*
2028 * Direct i/o helper functions
2029 */
2030 static void __inode_dio_wait(struct inode *inode)
2031 {
2032 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2033 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2034
2035 do {
2036 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2037 if (atomic_read(&inode->i_dio_count))
2038 schedule();
2039 } while (atomic_read(&inode->i_dio_count));
2040 finish_wait(wq, &q.wq_entry);
2041 }
2042
2043 /**
2044 * inode_dio_wait - wait for outstanding DIO requests to finish
2045 * @inode: inode to wait for
2046 *
2047 * Waits for all pending direct I/O requests to finish so that we can
2048 * proceed with a truncate or equivalent operation.
2049 *
2050 * Must be called under a lock that serializes taking new references
2051 * to i_dio_count, usually by inode->i_mutex.
2052 */
2053 void inode_dio_wait(struct inode *inode)
2054 {
2055 if (atomic_read(&inode->i_dio_count))
2056 __inode_dio_wait(inode);
2057 }
2058 EXPORT_SYMBOL(inode_dio_wait);
2059
2060 /*
2061 * inode_set_flags - atomically set some inode flags
2062 *
2063 * Note: the caller should be holding i_mutex, or else be sure that
2064 * they have exclusive access to the inode structure (i.e., while the
2065 * inode is being instantiated). The reason for the cmpxchg() loop
2066 * --- which wouldn't be necessary if all code paths which modify
2067 * i_flags actually followed this rule, is that there is at least one
2068 * code path which doesn't today so we use cmpxchg() out of an abundance
2069 * of caution.
2070 *
2071 * In the long run, i_mutex is overkill, and we should probably look
2072 * at using the i_lock spinlock to protect i_flags, and then make sure
2073 * it is so documented in include/linux/fs.h and that all code follows
2074 * the locking convention!!
2075 */
2076 void inode_set_flags(struct inode *inode, unsigned int flags,
2077 unsigned int mask)
2078 {
2079 unsigned int old_flags, new_flags;
2080
2081 WARN_ON_ONCE(flags & ~mask);
2082 do {
2083 old_flags = ACCESS_ONCE(inode->i_flags);
2084 new_flags = (old_flags & ~mask) | flags;
2085 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2086 new_flags) != old_flags));
2087 }
2088 EXPORT_SYMBOL(inode_set_flags);
2089
2090 void inode_nohighmem(struct inode *inode)
2091 {
2092 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2093 }
2094 EXPORT_SYMBOL(inode_nohighmem);
2095
2096 /**
2097 * current_time - Return FS time
2098 * @inode: inode.
2099 *
2100 * Return the current time truncated to the time granularity supported by
2101 * the fs.
2102 *
2103 * Note that inode and inode->sb cannot be NULL.
2104 * Otherwise, the function warns and returns time without truncation.
2105 */
2106 struct timespec current_time(struct inode *inode)
2107 {
2108 struct timespec now = current_kernel_time();
2109
2110 if (unlikely(!inode->i_sb)) {
2111 WARN(1, "current_time() called with uninitialized super_block in the inode");
2112 return now;
2113 }
2114
2115 return timespec_trunc(now, inode->i_sb->s_time_gran);
2116 }
2117 EXPORT_SYMBOL(current_time);