1 /*
2 * linux/fs/inode.c
3 *
4 * (C) 1997 Linus Torvalds
5 */
6
7 #include <linux/fs.h>
8 #include <linux/mm.h>
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/quotaops.h>
12 #include <linux/slab.h>
13 #include <linux/writeback.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/wait.h>
17 #include <linux/hash.h>
18 #include <linux/swap.h>
19 #include <linux/security.h>
20 #include <linux/ima.h>
21 #include <linux/pagemap.h>
22 #include <linux/cdev.h>
23 #include <linux/bootmem.h>
24 #include <linux/inotify.h>
25 #include <linux/fsnotify.h>
26 #include <linux/mount.h>
27 #include <linux/async.h>
28 #include <linux/posix_acl.h>
29
30 /*
31 * This is needed for the following functions:
32 * - inode_has_buffers
33 * - invalidate_inode_buffers
34 * - invalidate_bdev
35 *
36 * FIXME: remove all knowledge of the buffer layer from this file
37 */
38 #include <linux/buffer_head.h>
39
40 /*
41 * New inode.c implementation.
42 *
43 * This implementation has the basic premise of trying
44 * to be extremely low-overhead and SMP-safe, yet be
45 * simple enough to be "obviously correct".
46 *
47 * Famous last words.
48 */
49
50 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
51
52 /* #define INODE_PARANOIA 1 */
53 /* #define INODE_DEBUG 1 */
54
55 /*
56 * Inode lookup is no longer as critical as it used to be:
57 * most of the lookups are going to be through the dcache.
58 */
59 #define I_HASHBITS i_hash_shift
60 #define I_HASHMASK i_hash_mask
61
62 static unsigned int i_hash_mask __read_mostly;
63 static unsigned int i_hash_shift __read_mostly;
64
65 /*
66 * Each inode can be on two separate lists. One is
67 * the hash list of the inode, used for lookups. The
68 * other linked list is the "type" list:
69 * "in_use" - valid inode, i_count > 0, i_nlink > 0
70 * "dirty" - as "in_use" but also dirty
71 * "unused" - valid inode, i_count = 0
72 *
73 * A "dirty" list is maintained for each super block,
74 * allowing for low-overhead inode sync() operations.
75 */
76
77 LIST_HEAD(inode_in_use);
78 LIST_HEAD(inode_unused);
79 static struct hlist_head *inode_hashtable __read_mostly;
80
81 /*
82 * A simple spinlock to protect the list manipulations.
83 *
84 * NOTE! You also have to own the lock if you change
85 * the i_state of an inode while it is in use..
86 */
87 DEFINE_SPINLOCK(inode_lock);
88
89 /*
90 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
91 * icache shrinking path, and the umount path. Without this exclusion,
92 * by the time prune_icache calls iput for the inode whose pages it has
93 * been invalidating, or by the time it calls clear_inode & destroy_inode
94 * from its final dispose_list, the struct super_block they refer to
95 * (for inode->i_sb->s_op) may already have been freed and reused.
96 */
97 static DEFINE_MUTEX(iprune_mutex);
98
99 /*
100 * Statistics gathering..
101 */
102 struct inodes_stat_t inodes_stat;
103
104 static struct kmem_cache *inode_cachep __read_mostly;
105
106 static void wake_up_inode(struct inode *inode)
107 {
108 /*
109 * Prevent speculative execution through spin_unlock(&inode_lock);
110 */
111 smp_mb();
112 wake_up_bit(&inode->i_state, __I_LOCK);
113 }
114
115 /**
116 * inode_init_always - perform inode structure intialisation
117 * @sb: superblock inode belongs to
118 * @inode: inode to initialise
119 *
120 * These are initializations that need to be done on every inode
121 * allocation as the fields are not initialised by slab allocation.
122 */
123 int inode_init_always(struct super_block *sb, struct inode *inode)
124 {
125 static const struct address_space_operations empty_aops;
126 static struct inode_operations empty_iops;
127 static const struct file_operations empty_fops;
128 struct address_space *const mapping = &inode->i_data;
129
130 inode->i_sb = sb;
131 inode->i_blkbits = sb->s_blocksize_bits;
132 inode->i_flags = 0;
133 atomic_set(&inode->i_count, 1);
134 inode->i_op = &empty_iops;
135 inode->i_fop = &empty_fops;
136 inode->i_nlink = 1;
137 inode->i_uid = 0;
138 inode->i_gid = 0;
139 atomic_set(&inode->i_writecount, 0);
140 inode->i_size = 0;
141 inode->i_blocks = 0;
142 inode->i_bytes = 0;
143 inode->i_generation = 0;
144 #ifdef CONFIG_QUOTA
145 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
146 #endif
147 inode->i_pipe = NULL;
148 inode->i_bdev = NULL;
149 inode->i_cdev = NULL;
150 inode->i_rdev = 0;
151 inode->dirtied_when = 0;
152
153 if (security_inode_alloc(inode))
154 goto out;
155
156 /* allocate and initialize an i_integrity */
157 if (ima_inode_alloc(inode))
158 goto out_free_security;
159
160 spin_lock_init(&inode->i_lock);
161 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
162
163 mutex_init(&inode->i_mutex);
164 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
165
166 init_rwsem(&inode->i_alloc_sem);
167 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
168
169 mapping->a_ops = &empty_aops;
170 mapping->host = inode;
171 mapping->flags = 0;
172 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
173 mapping->assoc_mapping = NULL;
174 mapping->backing_dev_info = &default_backing_dev_info;
175 mapping->writeback_index = 0;
176
177 /*
178 * If the block_device provides a backing_dev_info for client
179 * inodes then use that. Otherwise the inode share the bdev's
180 * backing_dev_info.
181 */
182 if (sb->s_bdev) {
183 struct backing_dev_info *bdi;
184
185 bdi = sb->s_bdev->bd_inode_backing_dev_info;
186 if (!bdi)
187 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
188 mapping->backing_dev_info = bdi;
189 }
190 inode->i_private = NULL;
191 inode->i_mapping = mapping;
192 #ifdef CONFIG_FS_POSIX_ACL
193 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
194 #endif
195
196 #ifdef CONFIG_FSNOTIFY
197 inode->i_fsnotify_mask = 0;
198 #endif
199
200 return 0;
201
202 out_free_security:
203 security_inode_free(inode);
204 out:
205 return -ENOMEM;
206 }
207 EXPORT_SYMBOL(inode_init_always);
208
209 static struct inode *alloc_inode(struct super_block *sb)
210 {
211 struct inode *inode;
212
213 if (sb->s_op->alloc_inode)
214 inode = sb->s_op->alloc_inode(sb);
215 else
216 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
217
218 if (!inode)
219 return NULL;
220
221 if (unlikely(inode_init_always(sb, inode))) {
222 if (inode->i_sb->s_op->destroy_inode)
223 inode->i_sb->s_op->destroy_inode(inode);
224 else
225 kmem_cache_free(inode_cachep, inode);
226 return NULL;
227 }
228
229 return inode;
230 }
231
232 void __destroy_inode(struct inode *inode)
233 {
234 BUG_ON(inode_has_buffers(inode));
235 ima_inode_free(inode);
236 security_inode_free(inode);
237 fsnotify_inode_delete(inode);
238 #ifdef CONFIG_FS_POSIX_ACL
239 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
240 posix_acl_release(inode->i_acl);
241 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
242 posix_acl_release(inode->i_default_acl);
243 #endif
244 }
245 EXPORT_SYMBOL(__destroy_inode);
246
247 void destroy_inode(struct inode *inode)
248 {
249 __destroy_inode(inode);
250 if (inode->i_sb->s_op->destroy_inode)
251 inode->i_sb->s_op->destroy_inode(inode);
252 else
253 kmem_cache_free(inode_cachep, (inode));
254 }
255
256 /*
257 * These are initializations that only need to be done
258 * once, because the fields are idempotent across use
259 * of the inode, so let the slab aware of that.
260 */
261 void inode_init_once(struct inode *inode)
262 {
263 memset(inode, 0, sizeof(*inode));
264 INIT_HLIST_NODE(&inode->i_hash);
265 INIT_LIST_HEAD(&inode->i_dentry);
266 INIT_LIST_HEAD(&inode->i_devices);
267 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
268 spin_lock_init(&inode->i_data.tree_lock);
269 spin_lock_init(&inode->i_data.i_mmap_lock);
270 INIT_LIST_HEAD(&inode->i_data.private_list);
271 spin_lock_init(&inode->i_data.private_lock);
272 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
273 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
274 i_size_ordered_init(inode);
275 #ifdef CONFIG_INOTIFY
276 INIT_LIST_HEAD(&inode->inotify_watches);
277 mutex_init(&inode->inotify_mutex);
278 #endif
279 #ifdef CONFIG_FSNOTIFY
280 INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries);
281 #endif
282 }
283 EXPORT_SYMBOL(inode_init_once);
284
285 static void init_once(void *foo)
286 {
287 struct inode *inode = (struct inode *) foo;
288
289 inode_init_once(inode);
290 }
291
292 /*
293 * inode_lock must be held
294 */
295 void __iget(struct inode *inode)
296 {
297 if (atomic_read(&inode->i_count)) {
298 atomic_inc(&inode->i_count);
299 return;
300 }
301 atomic_inc(&inode->i_count);
302 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
303 list_move(&inode->i_list, &inode_in_use);
304 inodes_stat.nr_unused--;
305 }
306
307 /**
308 * clear_inode - clear an inode
309 * @inode: inode to clear
310 *
311 * This is called by the filesystem to tell us
312 * that the inode is no longer useful. We just
313 * terminate it with extreme prejudice.
314 */
315 void clear_inode(struct inode *inode)
316 {
317 might_sleep();
318 invalidate_inode_buffers(inode);
319
320 BUG_ON(inode->i_data.nrpages);
321 BUG_ON(!(inode->i_state & I_FREEING));
322 BUG_ON(inode->i_state & I_CLEAR);
323 inode_sync_wait(inode);
324 vfs_dq_drop(inode);
325 if (inode->i_sb->s_op->clear_inode)
326 inode->i_sb->s_op->clear_inode(inode);
327 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
328 bd_forget(inode);
329 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
330 cd_forget(inode);
331 inode->i_state = I_CLEAR;
332 }
333 EXPORT_SYMBOL(clear_inode);
334
335 /*
336 * dispose_list - dispose of the contents of a local list
337 * @head: the head of the list to free
338 *
339 * Dispose-list gets a local list with local inodes in it, so it doesn't
340 * need to worry about list corruption and SMP locks.
341 */
342 static void dispose_list(struct list_head *head)
343 {
344 int nr_disposed = 0;
345
346 while (!list_empty(head)) {
347 struct inode *inode;
348
349 inode = list_first_entry(head, struct inode, i_list);
350 list_del(&inode->i_list);
351
352 if (inode->i_data.nrpages)
353 truncate_inode_pages(&inode->i_data, 0);
354 clear_inode(inode);
355
356 spin_lock(&inode_lock);
357 hlist_del_init(&inode->i_hash);
358 list_del_init(&inode->i_sb_list);
359 spin_unlock(&inode_lock);
360
361 wake_up_inode(inode);
362 destroy_inode(inode);
363 nr_disposed++;
364 }
365 spin_lock(&inode_lock);
366 inodes_stat.nr_inodes -= nr_disposed;
367 spin_unlock(&inode_lock);
368 }
369
370 /*
371 * Invalidate all inodes for a device.
372 */
373 static int invalidate_list(struct list_head *head, struct list_head *dispose)
374 {
375 struct list_head *next;
376 int busy = 0, count = 0;
377
378 next = head->next;
379 for (;;) {
380 struct list_head *tmp = next;
381 struct inode *inode;
382
383 /*
384 * We can reschedule here without worrying about the list's
385 * consistency because the per-sb list of inodes must not
386 * change during umount anymore, and because iprune_mutex keeps
387 * shrink_icache_memory() away.
388 */
389 cond_resched_lock(&inode_lock);
390
391 next = next->next;
392 if (tmp == head)
393 break;
394 inode = list_entry(tmp, struct inode, i_sb_list);
395 if (inode->i_state & I_NEW)
396 continue;
397 invalidate_inode_buffers(inode);
398 if (!atomic_read(&inode->i_count)) {
399 list_move(&inode->i_list, dispose);
400 WARN_ON(inode->i_state & I_NEW);
401 inode->i_state |= I_FREEING;
402 count++;
403 continue;
404 }
405 busy = 1;
406 }
407 /* only unused inodes may be cached with i_count zero */
408 inodes_stat.nr_unused -= count;
409 return busy;
410 }
411
412 /**
413 * invalidate_inodes - discard the inodes on a device
414 * @sb: superblock
415 *
416 * Discard all of the inodes for a given superblock. If the discard
417 * fails because there are busy inodes then a non zero value is returned.
418 * If the discard is successful all the inodes have been discarded.
419 */
420 int invalidate_inodes(struct super_block *sb)
421 {
422 int busy;
423 LIST_HEAD(throw_away);
424
425 mutex_lock(&iprune_mutex);
426 spin_lock(&inode_lock);
427 inotify_unmount_inodes(&sb->s_inodes);
428 fsnotify_unmount_inodes(&sb->s_inodes);
429 busy = invalidate_list(&sb->s_inodes, &throw_away);
430 spin_unlock(&inode_lock);
431
432 dispose_list(&throw_away);
433 mutex_unlock(&iprune_mutex);
434
435 return busy;
436 }
437 EXPORT_SYMBOL(invalidate_inodes);
438
439 static int can_unuse(struct inode *inode)
440 {
441 if (inode->i_state)
442 return 0;
443 if (inode_has_buffers(inode))
444 return 0;
445 if (atomic_read(&inode->i_count))
446 return 0;
447 if (inode->i_data.nrpages)
448 return 0;
449 return 1;
450 }
451
452 /*
453 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
454 * a temporary list and then are freed outside inode_lock by dispose_list().
455 *
456 * Any inodes which are pinned purely because of attached pagecache have their
457 * pagecache removed. We expect the final iput() on that inode to add it to
458 * the front of the inode_unused list. So look for it there and if the
459 * inode is still freeable, proceed. The right inode is found 99.9% of the
460 * time in testing on a 4-way.
461 *
462 * If the inode has metadata buffers attached to mapping->private_list then
463 * try to remove them.
464 */
465 static void prune_icache(int nr_to_scan)
466 {
467 LIST_HEAD(freeable);
468 int nr_pruned = 0;
469 int nr_scanned;
470 unsigned long reap = 0;
471
472 mutex_lock(&iprune_mutex);
473 spin_lock(&inode_lock);
474 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
475 struct inode *inode;
476
477 if (list_empty(&inode_unused))
478 break;
479
480 inode = list_entry(inode_unused.prev, struct inode, i_list);
481
482 if (inode->i_state || atomic_read(&inode->i_count)) {
483 list_move(&inode->i_list, &inode_unused);
484 continue;
485 }
486 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
487 __iget(inode);
488 spin_unlock(&inode_lock);
489 if (remove_inode_buffers(inode))
490 reap += invalidate_mapping_pages(&inode->i_data,
491 0, -1);
492 iput(inode);
493 spin_lock(&inode_lock);
494
495 if (inode != list_entry(inode_unused.next,
496 struct inode, i_list))
497 continue; /* wrong inode or list_empty */
498 if (!can_unuse(inode))
499 continue;
500 }
501 list_move(&inode->i_list, &freeable);
502 WARN_ON(inode->i_state & I_NEW);
503 inode->i_state |= I_FREEING;
504 nr_pruned++;
505 }
506 inodes_stat.nr_unused -= nr_pruned;
507 if (current_is_kswapd())
508 __count_vm_events(KSWAPD_INODESTEAL, reap);
509 else
510 __count_vm_events(PGINODESTEAL, reap);
511 spin_unlock(&inode_lock);
512
513 dispose_list(&freeable);
514 mutex_unlock(&iprune_mutex);
515 }
516
517 /*
518 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
519 * "unused" means that no dentries are referring to the inodes: the files are
520 * not open and the dcache references to those inodes have already been
521 * reclaimed.
522 *
523 * This function is passed the number of inodes to scan, and it returns the
524 * total number of remaining possibly-reclaimable inodes.
525 */
526 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
527 {
528 if (nr) {
529 /*
530 * Nasty deadlock avoidance. We may hold various FS locks,
531 * and we don't want to recurse into the FS that called us
532 * in clear_inode() and friends..
533 */
534 if (!(gfp_mask & __GFP_FS))
535 return -1;
536 prune_icache(nr);
537 }
538 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
539 }
540
541 static struct shrinker icache_shrinker = {
542 .shrink = shrink_icache_memory,
543 .seeks = DEFAULT_SEEKS,
544 };
545
546 static void __wait_on_freeing_inode(struct inode *inode);
547 /*
548 * Called with the inode lock held.
549 * NOTE: we are not increasing the inode-refcount, you must call __iget()
550 * by hand after calling find_inode now! This simplifies iunique and won't
551 * add any additional branch in the common code.
552 */
553 static struct inode *find_inode(struct super_block *sb,
554 struct hlist_head *head,
555 int (*test)(struct inode *, void *),
556 void *data)
557 {
558 struct hlist_node *node;
559 struct inode *inode = NULL;
560
561 repeat:
562 hlist_for_each_entry(inode, node, head, i_hash) {
563 if (inode->i_sb != sb)
564 continue;
565 if (!test(inode, data))
566 continue;
567 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
568 __wait_on_freeing_inode(inode);
569 goto repeat;
570 }
571 break;
572 }
573 return node ? inode : NULL;
574 }
575
576 /*
577 * find_inode_fast is the fast path version of find_inode, see the comment at
578 * iget_locked for details.
579 */
580 static struct inode *find_inode_fast(struct super_block *sb,
581 struct hlist_head *head, unsigned long ino)
582 {
583 struct hlist_node *node;
584 struct inode *inode = NULL;
585
586 repeat:
587 hlist_for_each_entry(inode, node, head, i_hash) {
588 if (inode->i_ino != ino)
589 continue;
590 if (inode->i_sb != sb)
591 continue;
592 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
593 __wait_on_freeing_inode(inode);
594 goto repeat;
595 }
596 break;
597 }
598 return node ? inode : NULL;
599 }
600
601 static unsigned long hash(struct super_block *sb, unsigned long hashval)
602 {
603 unsigned long tmp;
604
605 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
606 L1_CACHE_BYTES;
607 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
608 return tmp & I_HASHMASK;
609 }
610
611 static inline void
612 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
613 struct inode *inode)
614 {
615 inodes_stat.nr_inodes++;
616 list_add(&inode->i_list, &inode_in_use);
617 list_add(&inode->i_sb_list, &sb->s_inodes);
618 if (head)
619 hlist_add_head(&inode->i_hash, head);
620 }
621
622 /**
623 * inode_add_to_lists - add a new inode to relevant lists
624 * @sb: superblock inode belongs to
625 * @inode: inode to mark in use
626 *
627 * When an inode is allocated it needs to be accounted for, added to the in use
628 * list, the owning superblock and the inode hash. This needs to be done under
629 * the inode_lock, so export a function to do this rather than the inode lock
630 * itself. We calculate the hash list to add to here so it is all internal
631 * which requires the caller to have already set up the inode number in the
632 * inode to add.
633 */
634 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
635 {
636 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
637
638 spin_lock(&inode_lock);
639 __inode_add_to_lists(sb, head, inode);
640 spin_unlock(&inode_lock);
641 }
642 EXPORT_SYMBOL_GPL(inode_add_to_lists);
643
644 /**
645 * new_inode - obtain an inode
646 * @sb: superblock
647 *
648 * Allocates a new inode for given superblock. The default gfp_mask
649 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
650 * If HIGHMEM pages are unsuitable or it is known that pages allocated
651 * for the page cache are not reclaimable or migratable,
652 * mapping_set_gfp_mask() must be called with suitable flags on the
653 * newly created inode's mapping
654 *
655 */
656 struct inode *new_inode(struct super_block *sb)
657 {
658 /*
659 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
660 * error if st_ino won't fit in target struct field. Use 32bit counter
661 * here to attempt to avoid that.
662 */
663 static unsigned int last_ino;
664 struct inode *inode;
665
666 spin_lock_prefetch(&inode_lock);
667
668 inode = alloc_inode(sb);
669 if (inode) {
670 spin_lock(&inode_lock);
671 __inode_add_to_lists(sb, NULL, inode);
672 inode->i_ino = ++last_ino;
673 inode->i_state = 0;
674 spin_unlock(&inode_lock);
675 }
676 return inode;
677 }
678 EXPORT_SYMBOL(new_inode);
679
680 void unlock_new_inode(struct inode *inode)
681 {
682 #ifdef CONFIG_DEBUG_LOCK_ALLOC
683 if (inode->i_mode & S_IFDIR) {
684 struct file_system_type *type = inode->i_sb->s_type;
685
686 /* Set new key only if filesystem hasn't already changed it */
687 if (!lockdep_match_class(&inode->i_mutex,
688 &type->i_mutex_key)) {
689 /*
690 * ensure nobody is actually holding i_mutex
691 */
692 mutex_destroy(&inode->i_mutex);
693 mutex_init(&inode->i_mutex);
694 lockdep_set_class(&inode->i_mutex,
695 &type->i_mutex_dir_key);
696 }
697 }
698 #endif
699 /*
700 * This is special! We do not need the spinlock when clearing I_LOCK,
701 * because we're guaranteed that nobody else tries to do anything about
702 * the state of the inode when it is locked, as we just created it (so
703 * there can be no old holders that haven't tested I_LOCK).
704 * However we must emit the memory barrier so that other CPUs reliably
705 * see the clearing of I_LOCK after the other inode initialisation has
706 * completed.
707 */
708 smp_mb();
709 WARN_ON((inode->i_state & (I_LOCK|I_NEW)) != (I_LOCK|I_NEW));
710 inode->i_state &= ~(I_LOCK|I_NEW);
711 wake_up_inode(inode);
712 }
713 EXPORT_SYMBOL(unlock_new_inode);
714
715 /*
716 * This is called without the inode lock held.. Be careful.
717 *
718 * We no longer cache the sb_flags in i_flags - see fs.h
719 * -- rmk@arm.uk.linux.org
720 */
721 static struct inode *get_new_inode(struct super_block *sb,
722 struct hlist_head *head,
723 int (*test)(struct inode *, void *),
724 int (*set)(struct inode *, void *),
725 void *data)
726 {
727 struct inode *inode;
728
729 inode = alloc_inode(sb);
730 if (inode) {
731 struct inode *old;
732
733 spin_lock(&inode_lock);
734 /* We released the lock, so.. */
735 old = find_inode(sb, head, test, data);
736 if (!old) {
737 if (set(inode, data))
738 goto set_failed;
739
740 __inode_add_to_lists(sb, head, inode);
741 inode->i_state = I_LOCK|I_NEW;
742 spin_unlock(&inode_lock);
743
744 /* Return the locked inode with I_NEW set, the
745 * caller is responsible for filling in the contents
746 */
747 return inode;
748 }
749
750 /*
751 * Uhhuh, somebody else created the same inode under
752 * us. Use the old inode instead of the one we just
753 * allocated.
754 */
755 __iget(old);
756 spin_unlock(&inode_lock);
757 destroy_inode(inode);
758 inode = old;
759 wait_on_inode(inode);
760 }
761 return inode;
762
763 set_failed:
764 spin_unlock(&inode_lock);
765 destroy_inode(inode);
766 return NULL;
767 }
768
769 /*
770 * get_new_inode_fast is the fast path version of get_new_inode, see the
771 * comment at iget_locked for details.
772 */
773 static struct inode *get_new_inode_fast(struct super_block *sb,
774 struct hlist_head *head, unsigned long ino)
775 {
776 struct inode *inode;
777
778 inode = alloc_inode(sb);
779 if (inode) {
780 struct inode *old;
781
782 spin_lock(&inode_lock);
783 /* We released the lock, so.. */
784 old = find_inode_fast(sb, head, ino);
785 if (!old) {
786 inode->i_ino = ino;
787 __inode_add_to_lists(sb, head, inode);
788 inode->i_state = I_LOCK|I_NEW;
789 spin_unlock(&inode_lock);
790
791 /* Return the locked inode with I_NEW set, the
792 * caller is responsible for filling in the contents
793 */
794 return inode;
795 }
796
797 /*
798 * Uhhuh, somebody else created the same inode under
799 * us. Use the old inode instead of the one we just
800 * allocated.
801 */
802 __iget(old);
803 spin_unlock(&inode_lock);
804 destroy_inode(inode);
805 inode = old;
806 wait_on_inode(inode);
807 }
808 return inode;
809 }
810
811 /**
812 * iunique - get a unique inode number
813 * @sb: superblock
814 * @max_reserved: highest reserved inode number
815 *
816 * Obtain an inode number that is unique on the system for a given
817 * superblock. This is used by file systems that have no natural
818 * permanent inode numbering system. An inode number is returned that
819 * is higher than the reserved limit but unique.
820 *
821 * BUGS:
822 * With a large number of inodes live on the file system this function
823 * currently becomes quite slow.
824 */
825 ino_t iunique(struct super_block *sb, ino_t max_reserved)
826 {
827 /*
828 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
829 * error if st_ino won't fit in target struct field. Use 32bit counter
830 * here to attempt to avoid that.
831 */
832 static unsigned int counter;
833 struct inode *inode;
834 struct hlist_head *head;
835 ino_t res;
836
837 spin_lock(&inode_lock);
838 do {
839 if (counter <= max_reserved)
840 counter = max_reserved + 1;
841 res = counter++;
842 head = inode_hashtable + hash(sb, res);
843 inode = find_inode_fast(sb, head, res);
844 } while (inode != NULL);
845 spin_unlock(&inode_lock);
846
847 return res;
848 }
849 EXPORT_SYMBOL(iunique);
850
851 struct inode *igrab(struct inode *inode)
852 {
853 spin_lock(&inode_lock);
854 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
855 __iget(inode);
856 else
857 /*
858 * Handle the case where s_op->clear_inode is not been
859 * called yet, and somebody is calling igrab
860 * while the inode is getting freed.
861 */
862 inode = NULL;
863 spin_unlock(&inode_lock);
864 return inode;
865 }
866 EXPORT_SYMBOL(igrab);
867
868 /**
869 * ifind - internal function, you want ilookup5() or iget5().
870 * @sb: super block of file system to search
871 * @head: the head of the list to search
872 * @test: callback used for comparisons between inodes
873 * @data: opaque data pointer to pass to @test
874 * @wait: if true wait for the inode to be unlocked, if false do not
875 *
876 * ifind() searches for the inode specified by @data in the inode
877 * cache. This is a generalized version of ifind_fast() for file systems where
878 * the inode number is not sufficient for unique identification of an inode.
879 *
880 * If the inode is in the cache, the inode is returned with an incremented
881 * reference count.
882 *
883 * Otherwise NULL is returned.
884 *
885 * Note, @test is called with the inode_lock held, so can't sleep.
886 */
887 static struct inode *ifind(struct super_block *sb,
888 struct hlist_head *head, int (*test)(struct inode *, void *),
889 void *data, const int wait)
890 {
891 struct inode *inode;
892
893 spin_lock(&inode_lock);
894 inode = find_inode(sb, head, test, data);
895 if (inode) {
896 __iget(inode);
897 spin_unlock(&inode_lock);
898 if (likely(wait))
899 wait_on_inode(inode);
900 return inode;
901 }
902 spin_unlock(&inode_lock);
903 return NULL;
904 }
905
906 /**
907 * ifind_fast - internal function, you want ilookup() or iget().
908 * @sb: super block of file system to search
909 * @head: head of the list to search
910 * @ino: inode number to search for
911 *
912 * ifind_fast() searches for the inode @ino in the inode cache. This is for
913 * file systems where the inode number is sufficient for unique identification
914 * of an inode.
915 *
916 * If the inode is in the cache, the inode is returned with an incremented
917 * reference count.
918 *
919 * Otherwise NULL is returned.
920 */
921 static struct inode *ifind_fast(struct super_block *sb,
922 struct hlist_head *head, unsigned long ino)
923 {
924 struct inode *inode;
925
926 spin_lock(&inode_lock);
927 inode = find_inode_fast(sb, head, ino);
928 if (inode) {
929 __iget(inode);
930 spin_unlock(&inode_lock);
931 wait_on_inode(inode);
932 return inode;
933 }
934 spin_unlock(&inode_lock);
935 return NULL;
936 }
937
938 /**
939 * ilookup5_nowait - search for an inode in the inode cache
940 * @sb: super block of file system to search
941 * @hashval: hash value (usually inode number) to search for
942 * @test: callback used for comparisons between inodes
943 * @data: opaque data pointer to pass to @test
944 *
945 * ilookup5() uses ifind() to search for the inode specified by @hashval and
946 * @data in the inode cache. This is a generalized version of ilookup() for
947 * file systems where the inode number is not sufficient for unique
948 * identification of an inode.
949 *
950 * If the inode is in the cache, the inode is returned with an incremented
951 * reference count. Note, the inode lock is not waited upon so you have to be
952 * very careful what you do with the returned inode. You probably should be
953 * using ilookup5() instead.
954 *
955 * Otherwise NULL is returned.
956 *
957 * Note, @test is called with the inode_lock held, so can't sleep.
958 */
959 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
960 int (*test)(struct inode *, void *), void *data)
961 {
962 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
963
964 return ifind(sb, head, test, data, 0);
965 }
966 EXPORT_SYMBOL(ilookup5_nowait);
967
968 /**
969 * ilookup5 - search for an inode in the inode cache
970 * @sb: super block of file system to search
971 * @hashval: hash value (usually inode number) to search for
972 * @test: callback used for comparisons between inodes
973 * @data: opaque data pointer to pass to @test
974 *
975 * ilookup5() uses ifind() to search for the inode specified by @hashval and
976 * @data in the inode cache. This is a generalized version of ilookup() for
977 * file systems where the inode number is not sufficient for unique
978 * identification of an inode.
979 *
980 * If the inode is in the cache, the inode lock is waited upon and the inode is
981 * returned with an incremented reference count.
982 *
983 * Otherwise NULL is returned.
984 *
985 * Note, @test is called with the inode_lock held, so can't sleep.
986 */
987 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
988 int (*test)(struct inode *, void *), void *data)
989 {
990 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
991
992 return ifind(sb, head, test, data, 1);
993 }
994 EXPORT_SYMBOL(ilookup5);
995
996 /**
997 * ilookup - search for an inode in the inode cache
998 * @sb: super block of file system to search
999 * @ino: inode number to search for
1000 *
1001 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1002 * This is for file systems where the inode number is sufficient for unique
1003 * identification of an inode.
1004 *
1005 * If the inode is in the cache, the inode is returned with an incremented
1006 * reference count.
1007 *
1008 * Otherwise NULL is returned.
1009 */
1010 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1011 {
1012 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1013
1014 return ifind_fast(sb, head, ino);
1015 }
1016 EXPORT_SYMBOL(ilookup);
1017
1018 /**
1019 * iget5_locked - obtain an inode from a mounted file system
1020 * @sb: super block of file system
1021 * @hashval: hash value (usually inode number) to get
1022 * @test: callback used for comparisons between inodes
1023 * @set: callback used to initialize a new struct inode
1024 * @data: opaque data pointer to pass to @test and @set
1025 *
1026 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1027 * and @data in the inode cache and if present it is returned with an increased
1028 * reference count. This is a generalized version of iget_locked() for file
1029 * systems where the inode number is not sufficient for unique identification
1030 * of an inode.
1031 *
1032 * If the inode is not in cache, get_new_inode() is called to allocate a new
1033 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1034 * file system gets to fill it in before unlocking it via unlock_new_inode().
1035 *
1036 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1037 */
1038 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1039 int (*test)(struct inode *, void *),
1040 int (*set)(struct inode *, void *), void *data)
1041 {
1042 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1043 struct inode *inode;
1044
1045 inode = ifind(sb, head, test, data, 1);
1046 if (inode)
1047 return inode;
1048 /*
1049 * get_new_inode() will do the right thing, re-trying the search
1050 * in case it had to block at any point.
1051 */
1052 return get_new_inode(sb, head, test, set, data);
1053 }
1054 EXPORT_SYMBOL(iget5_locked);
1055
1056 /**
1057 * iget_locked - obtain an inode from a mounted file system
1058 * @sb: super block of file system
1059 * @ino: inode number to get
1060 *
1061 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1062 * the inode cache and if present it is returned with an increased reference
1063 * count. This is for file systems where the inode number is sufficient for
1064 * unique identification of an inode.
1065 *
1066 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1067 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1068 * The file system gets to fill it in before unlocking it via
1069 * unlock_new_inode().
1070 */
1071 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1072 {
1073 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1074 struct inode *inode;
1075
1076 inode = ifind_fast(sb, head, ino);
1077 if (inode)
1078 return inode;
1079 /*
1080 * get_new_inode_fast() will do the right thing, re-trying the search
1081 * in case it had to block at any point.
1082 */
1083 return get_new_inode_fast(sb, head, ino);
1084 }
1085 EXPORT_SYMBOL(iget_locked);
1086
1087 int insert_inode_locked(struct inode *inode)
1088 {
1089 struct super_block *sb = inode->i_sb;
1090 ino_t ino = inode->i_ino;
1091 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1092
1093 inode->i_state |= I_LOCK|I_NEW;
1094 while (1) {
1095 struct hlist_node *node;
1096 struct inode *old = NULL;
1097 spin_lock(&inode_lock);
1098 hlist_for_each_entry(old, node, head, i_hash) {
1099 if (old->i_ino != ino)
1100 continue;
1101 if (old->i_sb != sb)
1102 continue;
1103 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1104 continue;
1105 break;
1106 }
1107 if (likely(!node)) {
1108 hlist_add_head(&inode->i_hash, head);
1109 spin_unlock(&inode_lock);
1110 return 0;
1111 }
1112 __iget(old);
1113 spin_unlock(&inode_lock);
1114 wait_on_inode(old);
1115 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1116 iput(old);
1117 return -EBUSY;
1118 }
1119 iput(old);
1120 }
1121 }
1122 EXPORT_SYMBOL(insert_inode_locked);
1123
1124 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1125 int (*test)(struct inode *, void *), void *data)
1126 {
1127 struct super_block *sb = inode->i_sb;
1128 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1129
1130 inode->i_state |= I_LOCK|I_NEW;
1131
1132 while (1) {
1133 struct hlist_node *node;
1134 struct inode *old = NULL;
1135
1136 spin_lock(&inode_lock);
1137 hlist_for_each_entry(old, node, head, i_hash) {
1138 if (old->i_sb != sb)
1139 continue;
1140 if (!test(old, data))
1141 continue;
1142 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1143 continue;
1144 break;
1145 }
1146 if (likely(!node)) {
1147 hlist_add_head(&inode->i_hash, head);
1148 spin_unlock(&inode_lock);
1149 return 0;
1150 }
1151 __iget(old);
1152 spin_unlock(&inode_lock);
1153 wait_on_inode(old);
1154 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1155 iput(old);
1156 return -EBUSY;
1157 }
1158 iput(old);
1159 }
1160 }
1161 EXPORT_SYMBOL(insert_inode_locked4);
1162
1163 /**
1164 * __insert_inode_hash - hash an inode
1165 * @inode: unhashed inode
1166 * @hashval: unsigned long value used to locate this object in the
1167 * inode_hashtable.
1168 *
1169 * Add an inode to the inode hash for this superblock.
1170 */
1171 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1172 {
1173 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1174 spin_lock(&inode_lock);
1175 hlist_add_head(&inode->i_hash, head);
1176 spin_unlock(&inode_lock);
1177 }
1178 EXPORT_SYMBOL(__insert_inode_hash);
1179
1180 /**
1181 * remove_inode_hash - remove an inode from the hash
1182 * @inode: inode to unhash
1183 *
1184 * Remove an inode from the superblock.
1185 */
1186 void remove_inode_hash(struct inode *inode)
1187 {
1188 spin_lock(&inode_lock);
1189 hlist_del_init(&inode->i_hash);
1190 spin_unlock(&inode_lock);
1191 }
1192 EXPORT_SYMBOL(remove_inode_hash);
1193
1194 /*
1195 * Tell the filesystem that this inode is no longer of any interest and should
1196 * be completely destroyed.
1197 *
1198 * We leave the inode in the inode hash table until *after* the filesystem's
1199 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1200 * instigate) will always find up-to-date information either in the hash or on
1201 * disk.
1202 *
1203 * I_FREEING is set so that no-one will take a new reference to the inode while
1204 * it is being deleted.
1205 */
1206 void generic_delete_inode(struct inode *inode)
1207 {
1208 const struct super_operations *op = inode->i_sb->s_op;
1209
1210 list_del_init(&inode->i_list);
1211 list_del_init(&inode->i_sb_list);
1212 WARN_ON(inode->i_state & I_NEW);
1213 inode->i_state |= I_FREEING;
1214 inodes_stat.nr_inodes--;
1215 spin_unlock(&inode_lock);
1216
1217 security_inode_delete(inode);
1218
1219 if (op->delete_inode) {
1220 void (*delete)(struct inode *) = op->delete_inode;
1221 if (!is_bad_inode(inode))
1222 vfs_dq_init(inode);
1223 /* Filesystems implementing their own
1224 * s_op->delete_inode are required to call
1225 * truncate_inode_pages and clear_inode()
1226 * internally */
1227 delete(inode);
1228 } else {
1229 truncate_inode_pages(&inode->i_data, 0);
1230 clear_inode(inode);
1231 }
1232 spin_lock(&inode_lock);
1233 hlist_del_init(&inode->i_hash);
1234 spin_unlock(&inode_lock);
1235 wake_up_inode(inode);
1236 BUG_ON(inode->i_state != I_CLEAR);
1237 destroy_inode(inode);
1238 }
1239 EXPORT_SYMBOL(generic_delete_inode);
1240
1241 static void generic_forget_inode(struct inode *inode)
1242 {
1243 struct super_block *sb = inode->i_sb;
1244
1245 if (!hlist_unhashed(&inode->i_hash)) {
1246 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1247 list_move(&inode->i_list, &inode_unused);
1248 inodes_stat.nr_unused++;
1249 if (sb->s_flags & MS_ACTIVE) {
1250 spin_unlock(&inode_lock);
1251 return;
1252 }
1253 WARN_ON(inode->i_state & I_NEW);
1254 inode->i_state |= I_WILL_FREE;
1255 spin_unlock(&inode_lock);
1256 write_inode_now(inode, 1);
1257 spin_lock(&inode_lock);
1258 WARN_ON(inode->i_state & I_NEW);
1259 inode->i_state &= ~I_WILL_FREE;
1260 inodes_stat.nr_unused--;
1261 hlist_del_init(&inode->i_hash);
1262 }
1263 list_del_init(&inode->i_list);
1264 list_del_init(&inode->i_sb_list);
1265 WARN_ON(inode->i_state & I_NEW);
1266 inode->i_state |= I_FREEING;
1267 inodes_stat.nr_inodes--;
1268 spin_unlock(&inode_lock);
1269 if (inode->i_data.nrpages)
1270 truncate_inode_pages(&inode->i_data, 0);
1271 clear_inode(inode);
1272 wake_up_inode(inode);
1273 destroy_inode(inode);
1274 }
1275
1276 /*
1277 * Normal UNIX filesystem behaviour: delete the
1278 * inode when the usage count drops to zero, and
1279 * i_nlink is zero.
1280 */
1281 void generic_drop_inode(struct inode *inode)
1282 {
1283 if (!inode->i_nlink)
1284 generic_delete_inode(inode);
1285 else
1286 generic_forget_inode(inode);
1287 }
1288 EXPORT_SYMBOL_GPL(generic_drop_inode);
1289
1290 /*
1291 * Called when we're dropping the last reference
1292 * to an inode.
1293 *
1294 * Call the FS "drop()" function, defaulting to
1295 * the legacy UNIX filesystem behaviour..
1296 *
1297 * NOTE! NOTE! NOTE! We're called with the inode lock
1298 * held, and the drop function is supposed to release
1299 * the lock!
1300 */
1301 static inline void iput_final(struct inode *inode)
1302 {
1303 const struct super_operations *op = inode->i_sb->s_op;
1304 void (*drop)(struct inode *) = generic_drop_inode;
1305
1306 if (op && op->drop_inode)
1307 drop = op->drop_inode;
1308 drop(inode);
1309 }
1310
1311 /**
1312 * iput - put an inode
1313 * @inode: inode to put
1314 *
1315 * Puts an inode, dropping its usage count. If the inode use count hits
1316 * zero, the inode is then freed and may also be destroyed.
1317 *
1318 * Consequently, iput() can sleep.
1319 */
1320 void iput(struct inode *inode)
1321 {
1322 if (inode) {
1323 BUG_ON(inode->i_state == I_CLEAR);
1324
1325 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1326 iput_final(inode);
1327 }
1328 }
1329 EXPORT_SYMBOL(iput);
1330
1331 /**
1332 * bmap - find a block number in a file
1333 * @inode: inode of file
1334 * @block: block to find
1335 *
1336 * Returns the block number on the device holding the inode that
1337 * is the disk block number for the block of the file requested.
1338 * That is, asked for block 4 of inode 1 the function will return the
1339 * disk block relative to the disk start that holds that block of the
1340 * file.
1341 */
1342 sector_t bmap(struct inode *inode, sector_t block)
1343 {
1344 sector_t res = 0;
1345 if (inode->i_mapping->a_ops->bmap)
1346 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1347 return res;
1348 }
1349 EXPORT_SYMBOL(bmap);
1350
1351 /*
1352 * With relative atime, only update atime if the previous atime is
1353 * earlier than either the ctime or mtime or if at least a day has
1354 * passed since the last atime update.
1355 */
1356 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1357 struct timespec now)
1358 {
1359
1360 if (!(mnt->mnt_flags & MNT_RELATIME))
1361 return 1;
1362 /*
1363 * Is mtime younger than atime? If yes, update atime:
1364 */
1365 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1366 return 1;
1367 /*
1368 * Is ctime younger than atime? If yes, update atime:
1369 */
1370 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1371 return 1;
1372
1373 /*
1374 * Is the previous atime value older than a day? If yes,
1375 * update atime:
1376 */
1377 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1378 return 1;
1379 /*
1380 * Good, we can skip the atime update:
1381 */
1382 return 0;
1383 }
1384
1385 /**
1386 * touch_atime - update the access time
1387 * @mnt: mount the inode is accessed on
1388 * @dentry: dentry accessed
1389 *
1390 * Update the accessed time on an inode and mark it for writeback.
1391 * This function automatically handles read only file systems and media,
1392 * as well as the "noatime" flag and inode specific "noatime" markers.
1393 */
1394 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1395 {
1396 struct inode *inode = dentry->d_inode;
1397 struct timespec now;
1398
1399 if (mnt_want_write(mnt))
1400 return;
1401 if (inode->i_flags & S_NOATIME)
1402 goto out;
1403 if (IS_NOATIME(inode))
1404 goto out;
1405 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1406 goto out;
1407
1408 if (mnt->mnt_flags & MNT_NOATIME)
1409 goto out;
1410 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1411 goto out;
1412
1413 now = current_fs_time(inode->i_sb);
1414
1415 if (!relatime_need_update(mnt, inode, now))
1416 goto out;
1417
1418 if (timespec_equal(&inode->i_atime, &now))
1419 goto out;
1420
1421 inode->i_atime = now;
1422 mark_inode_dirty_sync(inode);
1423 out:
1424 mnt_drop_write(mnt);
1425 }
1426 EXPORT_SYMBOL(touch_atime);
1427
1428 /**
1429 * file_update_time - update mtime and ctime time
1430 * @file: file accessed
1431 *
1432 * Update the mtime and ctime members of an inode and mark the inode
1433 * for writeback. Note that this function is meant exclusively for
1434 * usage in the file write path of filesystems, and filesystems may
1435 * choose to explicitly ignore update via this function with the
1436 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1437 * timestamps are handled by the server.
1438 */
1439
1440 void file_update_time(struct file *file)
1441 {
1442 struct inode *inode = file->f_path.dentry->d_inode;
1443 struct timespec now;
1444 int sync_it = 0;
1445 int err;
1446
1447 if (IS_NOCMTIME(inode))
1448 return;
1449
1450 err = mnt_want_write_file(file);
1451 if (err)
1452 return;
1453
1454 now = current_fs_time(inode->i_sb);
1455 if (!timespec_equal(&inode->i_mtime, &now)) {
1456 inode->i_mtime = now;
1457 sync_it = 1;
1458 }
1459
1460 if (!timespec_equal(&inode->i_ctime, &now)) {
1461 inode->i_ctime = now;
1462 sync_it = 1;
1463 }
1464
1465 if (IS_I_VERSION(inode)) {
1466 inode_inc_iversion(inode);
1467 sync_it = 1;
1468 }
1469
1470 if (sync_it)
1471 mark_inode_dirty_sync(inode);
1472 mnt_drop_write(file->f_path.mnt);
1473 }
1474 EXPORT_SYMBOL(file_update_time);
1475
1476 int inode_needs_sync(struct inode *inode)
1477 {
1478 if (IS_SYNC(inode))
1479 return 1;
1480 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1481 return 1;
1482 return 0;
1483 }
1484 EXPORT_SYMBOL(inode_needs_sync);
1485
1486 int inode_wait(void *word)
1487 {
1488 schedule();
1489 return 0;
1490 }
1491 EXPORT_SYMBOL(inode_wait);
1492
1493 /*
1494 * If we try to find an inode in the inode hash while it is being
1495 * deleted, we have to wait until the filesystem completes its
1496 * deletion before reporting that it isn't found. This function waits
1497 * until the deletion _might_ have completed. Callers are responsible
1498 * to recheck inode state.
1499 *
1500 * It doesn't matter if I_LOCK is not set initially, a call to
1501 * wake_up_inode() after removing from the hash list will DTRT.
1502 *
1503 * This is called with inode_lock held.
1504 */
1505 static void __wait_on_freeing_inode(struct inode *inode)
1506 {
1507 wait_queue_head_t *wq;
1508 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1509 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1510 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1511 spin_unlock(&inode_lock);
1512 schedule();
1513 finish_wait(wq, &wait.wait);
1514 spin_lock(&inode_lock);
1515 }
1516
1517 static __initdata unsigned long ihash_entries;
1518 static int __init set_ihash_entries(char *str)
1519 {
1520 if (!str)
1521 return 0;
1522 ihash_entries = simple_strtoul(str, &str, 0);
1523 return 1;
1524 }
1525 __setup("ihash_entries=", set_ihash_entries);
1526
1527 /*
1528 * Initialize the waitqueues and inode hash table.
1529 */
1530 void __init inode_init_early(void)
1531 {
1532 int loop;
1533
1534 /* If hashes are distributed across NUMA nodes, defer
1535 * hash allocation until vmalloc space is available.
1536 */
1537 if (hashdist)
1538 return;
1539
1540 inode_hashtable =
1541 alloc_large_system_hash("Inode-cache",
1542 sizeof(struct hlist_head),
1543 ihash_entries,
1544 14,
1545 HASH_EARLY,
1546 &i_hash_shift,
1547 &i_hash_mask,
1548 0);
1549
1550 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1551 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1552 }
1553
1554 void __init inode_init(void)
1555 {
1556 int loop;
1557
1558 /* inode slab cache */
1559 inode_cachep = kmem_cache_create("inode_cache",
1560 sizeof(struct inode),
1561 0,
1562 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1563 SLAB_MEM_SPREAD),
1564 init_once);
1565 register_shrinker(&icache_shrinker);
1566
1567 /* Hash may have been set up in inode_init_early */
1568 if (!hashdist)
1569 return;
1570
1571 inode_hashtable =
1572 alloc_large_system_hash("Inode-cache",
1573 sizeof(struct hlist_head),
1574 ihash_entries,
1575 14,
1576 0,
1577 &i_hash_shift,
1578 &i_hash_mask,
1579 0);
1580
1581 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1582 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1583 }
1584
1585 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1586 {
1587 inode->i_mode = mode;
1588 if (S_ISCHR(mode)) {
1589 inode->i_fop = &def_chr_fops;
1590 inode->i_rdev = rdev;
1591 } else if (S_ISBLK(mode)) {
1592 inode->i_fop = &def_blk_fops;
1593 inode->i_rdev = rdev;
1594 } else if (S_ISFIFO(mode))
1595 inode->i_fop = &def_fifo_fops;
1596 else if (S_ISSOCK(mode))
1597 inode->i_fop = &bad_sock_fops;
1598 else
1599 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1600 mode);
1601 }
1602 EXPORT_SYMBOL(init_special_inode);
1603
|
This page was automatically generated by the
LXR engine.
|