blob: bf21dc6d0dbd752d243a2d3b6d92d2d554dc0381 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/inode.c
3 *
4 * (C) 1997 Linus Torvalds
5 */
6
Linus Torvalds1da177e2005-04-16 15:20:36 -07007#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/pagemap.h>
21#include <linux/cdev.h>
22#include <linux/bootmem.h>
Robert Love0eeca282005-07-12 17:06:03 -040023#include <linux/inotify.h>
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -080024#include <linux/mount.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26/*
27 * This is needed for the following functions:
28 * - inode_has_buffers
29 * - invalidate_inode_buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 * - invalidate_bdev
31 *
32 * FIXME: remove all knowledge of the buffer layer from this file
33 */
34#include <linux/buffer_head.h>
35
36/*
37 * New inode.c implementation.
38 *
39 * This implementation has the basic premise of trying
40 * to be extremely low-overhead and SMP-safe, yet be
41 * simple enough to be "obviously correct".
42 *
43 * Famous last words.
44 */
45
46/* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
47
48/* #define INODE_PARANOIA 1 */
49/* #define INODE_DEBUG 1 */
50
51/*
52 * Inode lookup is no longer as critical as it used to be:
53 * most of the lookups are going to be through the dcache.
54 */
55#define I_HASHBITS i_hash_shift
56#define I_HASHMASK i_hash_mask
57
Eric Dumazetfa3536c2006-03-26 01:37:24 -080058static unsigned int i_hash_mask __read_mostly;
59static unsigned int i_hash_shift __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61/*
62 * Each inode can be on two separate lists. One is
63 * the hash list of the inode, used for lookups. The
64 * other linked list is the "type" list:
65 * "in_use" - valid inode, i_count > 0, i_nlink > 0
66 * "dirty" - as "in_use" but also dirty
67 * "unused" - valid inode, i_count = 0
68 *
69 * A "dirty" list is maintained for each super block,
70 * allowing for low-overhead inode sync() operations.
71 */
72
73LIST_HEAD(inode_in_use);
74LIST_HEAD(inode_unused);
Eric Dumazetfa3536c2006-03-26 01:37:24 -080075static struct hlist_head *inode_hashtable __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
77/*
78 * A simple spinlock to protect the list manipulations.
79 *
80 * NOTE! You also have to own the lock if you change
81 * the i_state of an inode while it is in use..
82 */
83DEFINE_SPINLOCK(inode_lock);
84
85/*
Ingo Molnarf24075b2006-03-23 03:00:34 -080086 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 * icache shrinking path, and the umount path. Without this exclusion,
88 * by the time prune_icache calls iput for the inode whose pages it has
89 * been invalidating, or by the time it calls clear_inode & destroy_inode
90 * from its final dispose_list, the struct super_block they refer to
91 * (for inode->i_sb->s_op) may already have been freed and reused.
92 */
Adrian Bunkbdfc3262006-03-25 03:06:56 -080093static DEFINE_MUTEX(iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
95/*
96 * Statistics gathering..
97 */
98struct inodes_stat_t inodes_stat;
99
Christoph Lametere18b8902006-12-06 20:33:20 -0800100static struct kmem_cache * inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
102static struct inode *alloc_inode(struct super_block *sb)
103{
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700104 static const struct address_space_operations empty_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 static struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800106 static const struct file_operations empty_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 struct inode *inode;
108
109 if (sb->s_op->alloc_inode)
110 inode = sb->s_op->alloc_inode(sb);
111 else
Christoph Lametere94b1762006-12-06 20:33:17 -0800112 inode = (struct inode *) kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113
114 if (inode) {
115 struct address_space * const mapping = &inode->i_data;
116
117 inode->i_sb = sb;
118 inode->i_blkbits = sb->s_blocksize_bits;
119 inode->i_flags = 0;
120 atomic_set(&inode->i_count, 1);
121 inode->i_op = &empty_iops;
122 inode->i_fop = &empty_fops;
123 inode->i_nlink = 1;
124 atomic_set(&inode->i_writecount, 0);
125 inode->i_size = 0;
126 inode->i_blocks = 0;
127 inode->i_bytes = 0;
128 inode->i_generation = 0;
129#ifdef CONFIG_QUOTA
130 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
131#endif
132 inode->i_pipe = NULL;
133 inode->i_bdev = NULL;
134 inode->i_cdev = NULL;
135 inode->i_rdev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 inode->dirtied_when = 0;
137 if (security_inode_alloc(inode)) {
138 if (inode->i_sb->s_op->destroy_inode)
139 inode->i_sb->s_op->destroy_inode(inode);
140 else
141 kmem_cache_free(inode_cachep, (inode));
142 return NULL;
143 }
144
145 mapping->a_ops = &empty_aops;
146 mapping->host = inode;
147 mapping->flags = 0;
148 mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
149 mapping->assoc_mapping = NULL;
150 mapping->backing_dev_info = &default_backing_dev_info;
151
152 /*
153 * If the block_device provides a backing_dev_info for client
154 * inodes then use that. Otherwise the inode share the bdev's
155 * backing_dev_info.
156 */
157 if (sb->s_bdev) {
158 struct backing_dev_info *bdi;
159
160 bdi = sb->s_bdev->bd_inode_backing_dev_info;
161 if (!bdi)
162 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
163 mapping->backing_dev_info = bdi;
164 }
Al Viroe6c6e642006-10-10 22:48:47 +0100165 inode->i_private = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 inode->i_mapping = mapping;
167 }
168 return inode;
169}
170
171void destroy_inode(struct inode *inode)
172{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200173 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 security_inode_free(inode);
175 if (inode->i_sb->s_op->destroy_inode)
176 inode->i_sb->s_op->destroy_inode(inode);
177 else
178 kmem_cache_free(inode_cachep, (inode));
179}
180
181
182/*
183 * These are initializations that only need to be done
184 * once, because the fields are idempotent across use
185 * of the inode, so let the slab aware of that.
186 */
187void inode_init_once(struct inode *inode)
188{
189 memset(inode, 0, sizeof(*inode));
190 INIT_HLIST_NODE(&inode->i_hash);
191 INIT_LIST_HEAD(&inode->i_dentry);
192 INIT_LIST_HEAD(&inode->i_devices);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800193 mutex_init(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 init_rwsem(&inode->i_alloc_sem);
195 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
196 rwlock_init(&inode->i_data.tree_lock);
197 spin_lock_init(&inode->i_data.i_mmap_lock);
198 INIT_LIST_HEAD(&inode->i_data.private_list);
199 spin_lock_init(&inode->i_data.private_lock);
200 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
201 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
202 spin_lock_init(&inode->i_lock);
203 i_size_ordered_init(inode);
Robert Love0eeca282005-07-12 17:06:03 -0400204#ifdef CONFIG_INOTIFY
205 INIT_LIST_HEAD(&inode->inotify_watches);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800206 mutex_init(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400207#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
209
210EXPORT_SYMBOL(inode_init_once);
211
Christoph Lametere18b8902006-12-06 20:33:20 -0800212static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213{
214 struct inode * inode = (struct inode *) foo;
215
216 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
217 SLAB_CTOR_CONSTRUCTOR)
218 inode_init_once(inode);
219}
220
221/*
222 * inode_lock must be held
223 */
224void __iget(struct inode * inode)
225{
226 if (atomic_read(&inode->i_count)) {
227 atomic_inc(&inode->i_count);
228 return;
229 }
230 atomic_inc(&inode->i_count);
231 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
232 list_move(&inode->i_list, &inode_in_use);
233 inodes_stat.nr_unused--;
234}
235
236/**
237 * clear_inode - clear an inode
238 * @inode: inode to clear
239 *
240 * This is called by the filesystem to tell us
241 * that the inode is no longer useful. We just
242 * terminate it with extreme prejudice.
243 */
244void clear_inode(struct inode *inode)
245{
246 might_sleep();
247 invalidate_inode_buffers(inode);
248
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200249 BUG_ON(inode->i_data.nrpages);
250 BUG_ON(!(inode->i_state & I_FREEING));
251 BUG_ON(inode->i_state & I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 wait_on_inode(inode);
253 DQUOT_DROP(inode);
254 if (inode->i_sb && inode->i_sb->s_op->clear_inode)
255 inode->i_sb->s_op->clear_inode(inode);
Theodore Ts'oeaf796e2006-09-27 01:50:48 -0700256 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 bd_forget(inode);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700258 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 cd_forget(inode);
260 inode->i_state = I_CLEAR;
261}
262
263EXPORT_SYMBOL(clear_inode);
264
265/*
266 * dispose_list - dispose of the contents of a local list
267 * @head: the head of the list to free
268 *
269 * Dispose-list gets a local list with local inodes in it, so it doesn't
270 * need to worry about list corruption and SMP locks.
271 */
272static void dispose_list(struct list_head *head)
273{
274 int nr_disposed = 0;
275
276 while (!list_empty(head)) {
277 struct inode *inode;
278
279 inode = list_entry(head->next, struct inode, i_list);
280 list_del(&inode->i_list);
281
282 if (inode->i_data.nrpages)
283 truncate_inode_pages(&inode->i_data, 0);
284 clear_inode(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700285
286 spin_lock(&inode_lock);
287 hlist_del_init(&inode->i_hash);
288 list_del_init(&inode->i_sb_list);
289 spin_unlock(&inode_lock);
290
291 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 destroy_inode(inode);
293 nr_disposed++;
294 }
295 spin_lock(&inode_lock);
296 inodes_stat.nr_inodes -= nr_disposed;
297 spin_unlock(&inode_lock);
298}
299
300/*
301 * Invalidate all inodes for a device.
302 */
303static int invalidate_list(struct list_head *head, struct list_head *dispose)
304{
305 struct list_head *next;
306 int busy = 0, count = 0;
307
308 next = head->next;
309 for (;;) {
310 struct list_head * tmp = next;
311 struct inode * inode;
312
313 /*
314 * We can reschedule here without worrying about the list's
315 * consistency because the per-sb list of inodes must not
Ingo Molnarf24075b2006-03-23 03:00:34 -0800316 * change during umount anymore, and because iprune_mutex keeps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 * shrink_icache_memory() away.
318 */
319 cond_resched_lock(&inode_lock);
320
321 next = next->next;
322 if (tmp == head)
323 break;
324 inode = list_entry(tmp, struct inode, i_sb_list);
325 invalidate_inode_buffers(inode);
326 if (!atomic_read(&inode->i_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 list_move(&inode->i_list, dispose);
328 inode->i_state |= I_FREEING;
329 count++;
330 continue;
331 }
332 busy = 1;
333 }
334 /* only unused inodes may be cached with i_count zero */
335 inodes_stat.nr_unused -= count;
336 return busy;
337}
338
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339/**
340 * invalidate_inodes - discard the inodes on a device
341 * @sb: superblock
342 *
343 * Discard all of the inodes for a given superblock. If the discard
344 * fails because there are busy inodes then a non zero value is returned.
345 * If the discard is successful all the inodes have been discarded.
346 */
347int invalidate_inodes(struct super_block * sb)
348{
349 int busy;
350 LIST_HEAD(throw_away);
351
Ingo Molnarf24075b2006-03-23 03:00:34 -0800352 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 spin_lock(&inode_lock);
Robert Love0eeca282005-07-12 17:06:03 -0400354 inotify_unmount_inodes(&sb->s_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 busy = invalidate_list(&sb->s_inodes, &throw_away);
356 spin_unlock(&inode_lock);
357
358 dispose_list(&throw_away);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800359 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360
361 return busy;
362}
363
364EXPORT_SYMBOL(invalidate_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
366static int can_unuse(struct inode *inode)
367{
368 if (inode->i_state)
369 return 0;
370 if (inode_has_buffers(inode))
371 return 0;
372 if (atomic_read(&inode->i_count))
373 return 0;
374 if (inode->i_data.nrpages)
375 return 0;
376 return 1;
377}
378
379/*
380 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
381 * a temporary list and then are freed outside inode_lock by dispose_list().
382 *
383 * Any inodes which are pinned purely because of attached pagecache have their
384 * pagecache removed. We expect the final iput() on that inode to add it to
385 * the front of the inode_unused list. So look for it there and if the
386 * inode is still freeable, proceed. The right inode is found 99.9% of the
387 * time in testing on a 4-way.
388 *
389 * If the inode has metadata buffers attached to mapping->private_list then
390 * try to remove them.
391 */
392static void prune_icache(int nr_to_scan)
393{
394 LIST_HEAD(freeable);
395 int nr_pruned = 0;
396 int nr_scanned;
397 unsigned long reap = 0;
398
Ingo Molnarf24075b2006-03-23 03:00:34 -0800399 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 spin_lock(&inode_lock);
401 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
402 struct inode *inode;
403
404 if (list_empty(&inode_unused))
405 break;
406
407 inode = list_entry(inode_unused.prev, struct inode, i_list);
408
409 if (inode->i_state || atomic_read(&inode->i_count)) {
410 list_move(&inode->i_list, &inode_unused);
411 continue;
412 }
413 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
414 __iget(inode);
415 spin_unlock(&inode_lock);
416 if (remove_inode_buffers(inode))
417 reap += invalidate_inode_pages(&inode->i_data);
418 iput(inode);
419 spin_lock(&inode_lock);
420
421 if (inode != list_entry(inode_unused.next,
422 struct inode, i_list))
423 continue; /* wrong inode or list_empty */
424 if (!can_unuse(inode))
425 continue;
426 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 list_move(&inode->i_list, &freeable);
428 inode->i_state |= I_FREEING;
429 nr_pruned++;
430 }
431 inodes_stat.nr_unused -= nr_pruned;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700432 if (current_is_kswapd())
433 __count_vm_events(KSWAPD_INODESTEAL, reap);
434 else
435 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 spin_unlock(&inode_lock);
437
438 dispose_list(&freeable);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800439 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440}
441
442/*
443 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
444 * "unused" means that no dentries are referring to the inodes: the files are
445 * not open and the dcache references to those inodes have already been
446 * reclaimed.
447 *
448 * This function is passed the number of inodes to scan, and it returns the
449 * total number of remaining possibly-reclaimable inodes.
450 */
Al Viro27496a82005-10-21 03:20:48 -0400451static int shrink_icache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452{
453 if (nr) {
454 /*
455 * Nasty deadlock avoidance. We may hold various FS locks,
456 * and we don't want to recurse into the FS that called us
457 * in clear_inode() and friends..
458 */
459 if (!(gfp_mask & __GFP_FS))
460 return -1;
461 prune_icache(nr);
462 }
463 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
464}
465
466static void __wait_on_freeing_inode(struct inode *inode);
467/*
468 * Called with the inode lock held.
469 * NOTE: we are not increasing the inode-refcount, you must call __iget()
470 * by hand after calling find_inode now! This simplifies iunique and won't
471 * add any additional branch in the common code.
472 */
473static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
474{
475 struct hlist_node *node;
476 struct inode * inode = NULL;
477
478repeat:
479 hlist_for_each (node, head) {
480 inode = hlist_entry(node, struct inode, i_hash);
481 if (inode->i_sb != sb)
482 continue;
483 if (!test(inode, data))
484 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700485 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 __wait_on_freeing_inode(inode);
487 goto repeat;
488 }
489 break;
490 }
491 return node ? inode : NULL;
492}
493
494/*
495 * find_inode_fast is the fast path version of find_inode, see the comment at
496 * iget_locked for details.
497 */
498static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
499{
500 struct hlist_node *node;
501 struct inode * inode = NULL;
502
503repeat:
504 hlist_for_each (node, head) {
505 inode = hlist_entry(node, struct inode, i_hash);
506 if (inode->i_ino != ino)
507 continue;
508 if (inode->i_sb != sb)
509 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700510 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 __wait_on_freeing_inode(inode);
512 goto repeat;
513 }
514 break;
515 }
516 return node ? inode : NULL;
517}
518
519/**
520 * new_inode - obtain an inode
521 * @sb: superblock
522 *
523 * Allocates a new inode for given superblock.
524 */
525struct inode *new_inode(struct super_block *sb)
526{
527 static unsigned long last_ino;
528 struct inode * inode;
529
530 spin_lock_prefetch(&inode_lock);
531
532 inode = alloc_inode(sb);
533 if (inode) {
534 spin_lock(&inode_lock);
535 inodes_stat.nr_inodes++;
536 list_add(&inode->i_list, &inode_in_use);
537 list_add(&inode->i_sb_list, &sb->s_inodes);
538 inode->i_ino = ++last_ino;
539 inode->i_state = 0;
540 spin_unlock(&inode_lock);
541 }
542 return inode;
543}
544
545EXPORT_SYMBOL(new_inode);
546
547void unlock_new_inode(struct inode *inode)
548{
549 /*
550 * This is special! We do not need the spinlock
551 * when clearing I_LOCK, because we're guaranteed
552 * that nobody else tries to do anything about the
553 * state of the inode when it is locked, as we
554 * just created it (so there can be no old holders
555 * that haven't tested I_LOCK).
556 */
557 inode->i_state &= ~(I_LOCK|I_NEW);
558 wake_up_inode(inode);
559}
560
561EXPORT_SYMBOL(unlock_new_inode);
562
563/*
564 * This is called without the inode lock held.. Be careful.
565 *
566 * We no longer cache the sb_flags in i_flags - see fs.h
567 * -- rmk@arm.uk.linux.org
568 */
569static struct inode * get_new_inode(struct super_block *sb, struct hlist_head *head, int (*test)(struct inode *, void *), int (*set)(struct inode *, void *), void *data)
570{
571 struct inode * inode;
572
573 inode = alloc_inode(sb);
574 if (inode) {
575 struct inode * old;
576
577 spin_lock(&inode_lock);
578 /* We released the lock, so.. */
579 old = find_inode(sb, head, test, data);
580 if (!old) {
581 if (set(inode, data))
582 goto set_failed;
583
584 inodes_stat.nr_inodes++;
585 list_add(&inode->i_list, &inode_in_use);
586 list_add(&inode->i_sb_list, &sb->s_inodes);
587 hlist_add_head(&inode->i_hash, head);
588 inode->i_state = I_LOCK|I_NEW;
589 spin_unlock(&inode_lock);
590
591 /* Return the locked inode with I_NEW set, the
592 * caller is responsible for filling in the contents
593 */
594 return inode;
595 }
596
597 /*
598 * Uhhuh, somebody else created the same inode under
599 * us. Use the old inode instead of the one we just
600 * allocated.
601 */
602 __iget(old);
603 spin_unlock(&inode_lock);
604 destroy_inode(inode);
605 inode = old;
606 wait_on_inode(inode);
607 }
608 return inode;
609
610set_failed:
611 spin_unlock(&inode_lock);
612 destroy_inode(inode);
613 return NULL;
614}
615
616/*
617 * get_new_inode_fast is the fast path version of get_new_inode, see the
618 * comment at iget_locked for details.
619 */
620static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
621{
622 struct inode * inode;
623
624 inode = alloc_inode(sb);
625 if (inode) {
626 struct inode * old;
627
628 spin_lock(&inode_lock);
629 /* We released the lock, so.. */
630 old = find_inode_fast(sb, head, ino);
631 if (!old) {
632 inode->i_ino = ino;
633 inodes_stat.nr_inodes++;
634 list_add(&inode->i_list, &inode_in_use);
635 list_add(&inode->i_sb_list, &sb->s_inodes);
636 hlist_add_head(&inode->i_hash, head);
637 inode->i_state = I_LOCK|I_NEW;
638 spin_unlock(&inode_lock);
639
640 /* Return the locked inode with I_NEW set, the
641 * caller is responsible for filling in the contents
642 */
643 return inode;
644 }
645
646 /*
647 * Uhhuh, somebody else created the same inode under
648 * us. Use the old inode instead of the one we just
649 * allocated.
650 */
651 __iget(old);
652 spin_unlock(&inode_lock);
653 destroy_inode(inode);
654 inode = old;
655 wait_on_inode(inode);
656 }
657 return inode;
658}
659
Andreas Mohred97bd32006-10-02 02:17:17 -0700660static unsigned long hash(struct super_block *sb, unsigned long hashval)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661{
662 unsigned long tmp;
663
664 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
665 L1_CACHE_BYTES;
666 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
667 return tmp & I_HASHMASK;
668}
669
670/**
671 * iunique - get a unique inode number
672 * @sb: superblock
673 * @max_reserved: highest reserved inode number
674 *
675 * Obtain an inode number that is unique on the system for a given
676 * superblock. This is used by file systems that have no natural
677 * permanent inode numbering system. An inode number is returned that
678 * is higher than the reserved limit but unique.
679 *
680 * BUGS:
681 * With a large number of inodes live on the file system this function
682 * currently becomes quite slow.
683 */
684ino_t iunique(struct super_block *sb, ino_t max_reserved)
685{
686 static ino_t counter;
687 struct inode *inode;
688 struct hlist_head * head;
689 ino_t res;
690 spin_lock(&inode_lock);
691retry:
692 if (counter > max_reserved) {
693 head = inode_hashtable + hash(sb,counter);
694 res = counter++;
695 inode = find_inode_fast(sb, head, res);
696 if (!inode) {
697 spin_unlock(&inode_lock);
698 return res;
699 }
700 } else {
701 counter = max_reserved + 1;
702 }
703 goto retry;
704
705}
706
707EXPORT_SYMBOL(iunique);
708
709struct inode *igrab(struct inode *inode)
710{
711 spin_lock(&inode_lock);
Alexander Viro991114c2005-06-23 00:09:01 -0700712 if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 __iget(inode);
714 else
715 /*
716 * Handle the case where s_op->clear_inode is not been
717 * called yet, and somebody is calling igrab
718 * while the inode is getting freed.
719 */
720 inode = NULL;
721 spin_unlock(&inode_lock);
722 return inode;
723}
724
725EXPORT_SYMBOL(igrab);
726
727/**
728 * ifind - internal function, you want ilookup5() or iget5().
729 * @sb: super block of file system to search
730 * @head: the head of the list to search
731 * @test: callback used for comparisons between inodes
732 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700733 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 *
735 * ifind() searches for the inode specified by @data in the inode
736 * cache. This is a generalized version of ifind_fast() for file systems where
737 * the inode number is not sufficient for unique identification of an inode.
738 *
739 * If the inode is in the cache, the inode is returned with an incremented
740 * reference count.
741 *
742 * Otherwise NULL is returned.
743 *
744 * Note, @test is called with the inode_lock held, so can't sleep.
745 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800746static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700748 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
750 struct inode *inode;
751
752 spin_lock(&inode_lock);
753 inode = find_inode(sb, head, test, data);
754 if (inode) {
755 __iget(inode);
756 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700757 if (likely(wait))
758 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 return inode;
760 }
761 spin_unlock(&inode_lock);
762 return NULL;
763}
764
765/**
766 * ifind_fast - internal function, you want ilookup() or iget().
767 * @sb: super block of file system to search
768 * @head: head of the list to search
769 * @ino: inode number to search for
770 *
771 * ifind_fast() searches for the inode @ino in the inode cache. This is for
772 * file systems where the inode number is sufficient for unique identification
773 * of an inode.
774 *
775 * If the inode is in the cache, the inode is returned with an incremented
776 * reference count.
777 *
778 * Otherwise NULL is returned.
779 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800780static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 struct hlist_head *head, unsigned long ino)
782{
783 struct inode *inode;
784
785 spin_lock(&inode_lock);
786 inode = find_inode_fast(sb, head, ino);
787 if (inode) {
788 __iget(inode);
789 spin_unlock(&inode_lock);
790 wait_on_inode(inode);
791 return inode;
792 }
793 spin_unlock(&inode_lock);
794 return NULL;
795}
796
797/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700798 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 * @sb: super block of file system to search
800 * @hashval: hash value (usually inode number) to search for
801 * @test: callback used for comparisons between inodes
802 * @data: opaque data pointer to pass to @test
803 *
804 * ilookup5() uses ifind() to search for the inode specified by @hashval and
805 * @data in the inode cache. This is a generalized version of ilookup() for
806 * file systems where the inode number is not sufficient for unique
807 * identification of an inode.
808 *
809 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700810 * reference count. Note, the inode lock is not waited upon so you have to be
811 * very careful what you do with the returned inode. You probably should be
812 * using ilookup5() instead.
813 *
814 * Otherwise NULL is returned.
815 *
816 * Note, @test is called with the inode_lock held, so can't sleep.
817 */
818struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
819 int (*test)(struct inode *, void *), void *data)
820{
821 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
822
823 return ifind(sb, head, test, data, 0);
824}
825
826EXPORT_SYMBOL(ilookup5_nowait);
827
828/**
829 * ilookup5 - search for an inode in the inode cache
830 * @sb: super block of file system to search
831 * @hashval: hash value (usually inode number) to search for
832 * @test: callback used for comparisons between inodes
833 * @data: opaque data pointer to pass to @test
834 *
835 * ilookup5() uses ifind() to search for the inode specified by @hashval and
836 * @data in the inode cache. This is a generalized version of ilookup() for
837 * file systems where the inode number is not sufficient for unique
838 * identification of an inode.
839 *
840 * If the inode is in the cache, the inode lock is waited upon and the inode is
841 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 *
843 * Otherwise NULL is returned.
844 *
845 * Note, @test is called with the inode_lock held, so can't sleep.
846 */
847struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
848 int (*test)(struct inode *, void *), void *data)
849{
850 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
851
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700852 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853}
854
855EXPORT_SYMBOL(ilookup5);
856
857/**
858 * ilookup - search for an inode in the inode cache
859 * @sb: super block of file system to search
860 * @ino: inode number to search for
861 *
862 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
863 * This is for file systems where the inode number is sufficient for unique
864 * identification of an inode.
865 *
866 * If the inode is in the cache, the inode is returned with an incremented
867 * reference count.
868 *
869 * Otherwise NULL is returned.
870 */
871struct inode *ilookup(struct super_block *sb, unsigned long ino)
872{
873 struct hlist_head *head = inode_hashtable + hash(sb, ino);
874
875 return ifind_fast(sb, head, ino);
876}
877
878EXPORT_SYMBOL(ilookup);
879
880/**
881 * iget5_locked - obtain an inode from a mounted file system
882 * @sb: super block of file system
883 * @hashval: hash value (usually inode number) to get
884 * @test: callback used for comparisons between inodes
885 * @set: callback used to initialize a new struct inode
886 * @data: opaque data pointer to pass to @test and @set
887 *
888 * This is iget() without the read_inode() portion of get_new_inode().
889 *
890 * iget5_locked() uses ifind() to search for the inode specified by @hashval
891 * and @data in the inode cache and if present it is returned with an increased
892 * reference count. This is a generalized version of iget_locked() for file
893 * systems where the inode number is not sufficient for unique identification
894 * of an inode.
895 *
896 * If the inode is not in cache, get_new_inode() is called to allocate a new
897 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
898 * file system gets to fill it in before unlocking it via unlock_new_inode().
899 *
900 * Note both @test and @set are called with the inode_lock held, so can't sleep.
901 */
902struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
903 int (*test)(struct inode *, void *),
904 int (*set)(struct inode *, void *), void *data)
905{
906 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
907 struct inode *inode;
908
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700909 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 if (inode)
911 return inode;
912 /*
913 * get_new_inode() will do the right thing, re-trying the search
914 * in case it had to block at any point.
915 */
916 return get_new_inode(sb, head, test, set, data);
917}
918
919EXPORT_SYMBOL(iget5_locked);
920
921/**
922 * iget_locked - obtain an inode from a mounted file system
923 * @sb: super block of file system
924 * @ino: inode number to get
925 *
926 * This is iget() without the read_inode() portion of get_new_inode_fast().
927 *
928 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
929 * the inode cache and if present it is returned with an increased reference
930 * count. This is for file systems where the inode number is sufficient for
931 * unique identification of an inode.
932 *
933 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
934 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
935 * The file system gets to fill it in before unlocking it via
936 * unlock_new_inode().
937 */
938struct inode *iget_locked(struct super_block *sb, unsigned long ino)
939{
940 struct hlist_head *head = inode_hashtable + hash(sb, ino);
941 struct inode *inode;
942
943 inode = ifind_fast(sb, head, ino);
944 if (inode)
945 return inode;
946 /*
947 * get_new_inode_fast() will do the right thing, re-trying the search
948 * in case it had to block at any point.
949 */
950 return get_new_inode_fast(sb, head, ino);
951}
952
953EXPORT_SYMBOL(iget_locked);
954
955/**
956 * __insert_inode_hash - hash an inode
957 * @inode: unhashed inode
958 * @hashval: unsigned long value used to locate this object in the
959 * inode_hashtable.
960 *
961 * Add an inode to the inode hash for this superblock.
962 */
963void __insert_inode_hash(struct inode *inode, unsigned long hashval)
964{
965 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
966 spin_lock(&inode_lock);
967 hlist_add_head(&inode->i_hash, head);
968 spin_unlock(&inode_lock);
969}
970
971EXPORT_SYMBOL(__insert_inode_hash);
972
973/**
974 * remove_inode_hash - remove an inode from the hash
975 * @inode: inode to unhash
976 *
977 * Remove an inode from the superblock.
978 */
979void remove_inode_hash(struct inode *inode)
980{
981 spin_lock(&inode_lock);
982 hlist_del_init(&inode->i_hash);
983 spin_unlock(&inode_lock);
984}
985
986EXPORT_SYMBOL(remove_inode_hash);
987
988/*
989 * Tell the filesystem that this inode is no longer of any interest and should
990 * be completely destroyed.
991 *
992 * We leave the inode in the inode hash table until *after* the filesystem's
993 * ->delete_inode completes. This ensures that an iget (such as nfsd might
994 * instigate) will always find up-to-date information either in the hash or on
995 * disk.
996 *
997 * I_FREEING is set so that no-one will take a new reference to the inode while
998 * it is being deleted.
999 */
1000void generic_delete_inode(struct inode *inode)
1001{
1002 struct super_operations *op = inode->i_sb->s_op;
1003
1004 list_del_init(&inode->i_list);
1005 list_del_init(&inode->i_sb_list);
Andreas Mohred97bd32006-10-02 02:17:17 -07001006 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 inodes_stat.nr_inodes--;
1008 spin_unlock(&inode_lock);
1009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 security_inode_delete(inode);
1011
1012 if (op->delete_inode) {
1013 void (*delete)(struct inode *) = op->delete_inode;
1014 if (!is_bad_inode(inode))
1015 DQUOT_INIT(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001016 /* Filesystems implementing their own
1017 * s_op->delete_inode are required to call
1018 * truncate_inode_pages and clear_inode()
1019 * internally */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 delete(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001021 } else {
1022 truncate_inode_pages(&inode->i_data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 clear_inode(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001024 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 spin_lock(&inode_lock);
1026 hlist_del_init(&inode->i_hash);
1027 spin_unlock(&inode_lock);
1028 wake_up_inode(inode);
Eric Sesterhennb7542f82006-04-02 13:38:18 +02001029 BUG_ON(inode->i_state != I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 destroy_inode(inode);
1031}
1032
1033EXPORT_SYMBOL(generic_delete_inode);
1034
1035static void generic_forget_inode(struct inode *inode)
1036{
1037 struct super_block *sb = inode->i_sb;
1038
1039 if (!hlist_unhashed(&inode->i_hash)) {
1040 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1041 list_move(&inode->i_list, &inode_unused);
1042 inodes_stat.nr_unused++;
Alexander Viro991114c2005-06-23 00:09:01 -07001043 if (!sb || (sb->s_flags & MS_ACTIVE)) {
1044 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001046 }
1047 inode->i_state |= I_WILL_FREE;
1048 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 write_inode_now(inode, 1);
1050 spin_lock(&inode_lock);
Alexander Viro991114c2005-06-23 00:09:01 -07001051 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 inodes_stat.nr_unused--;
1053 hlist_del_init(&inode->i_hash);
1054 }
1055 list_del_init(&inode->i_list);
1056 list_del_init(&inode->i_sb_list);
Alexander Viro991114c2005-06-23 00:09:01 -07001057 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 inodes_stat.nr_inodes--;
1059 spin_unlock(&inode_lock);
1060 if (inode->i_data.nrpages)
1061 truncate_inode_pages(&inode->i_data, 0);
1062 clear_inode(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001063 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 destroy_inode(inode);
1065}
1066
1067/*
1068 * Normal UNIX filesystem behaviour: delete the
1069 * inode when the usage count drops to zero, and
1070 * i_nlink is zero.
1071 */
Mark Fashehcb2c0232005-07-07 17:56:03 -07001072void generic_drop_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
1074 if (!inode->i_nlink)
1075 generic_delete_inode(inode);
1076 else
1077 generic_forget_inode(inode);
1078}
1079
Mark Fashehcb2c0232005-07-07 17:56:03 -07001080EXPORT_SYMBOL_GPL(generic_drop_inode);
1081
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082/*
1083 * Called when we're dropping the last reference
1084 * to an inode.
1085 *
1086 * Call the FS "drop()" function, defaulting to
1087 * the legacy UNIX filesystem behaviour..
1088 *
1089 * NOTE! NOTE! NOTE! We're called with the inode lock
1090 * held, and the drop function is supposed to release
1091 * the lock!
1092 */
1093static inline void iput_final(struct inode *inode)
1094{
1095 struct super_operations *op = inode->i_sb->s_op;
1096 void (*drop)(struct inode *) = generic_drop_inode;
1097
1098 if (op && op->drop_inode)
1099 drop = op->drop_inode;
1100 drop(inode);
1101}
1102
1103/**
1104 * iput - put an inode
1105 * @inode: inode to put
1106 *
1107 * Puts an inode, dropping its usage count. If the inode use count hits
1108 * zero, the inode is then freed and may also be destroyed.
1109 *
1110 * Consequently, iput() can sleep.
1111 */
1112void iput(struct inode *inode)
1113{
1114 if (inode) {
1115 struct super_operations *op = inode->i_sb->s_op;
1116
1117 BUG_ON(inode->i_state == I_CLEAR);
1118
1119 if (op && op->put_inode)
1120 op->put_inode(inode);
1121
1122 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1123 iput_final(inode);
1124 }
1125}
1126
1127EXPORT_SYMBOL(iput);
1128
1129/**
1130 * bmap - find a block number in a file
1131 * @inode: inode of file
1132 * @block: block to find
1133 *
1134 * Returns the block number on the device holding the inode that
1135 * is the disk block number for the block of the file requested.
1136 * That is, asked for block 4 of inode 1 the function will return the
1137 * disk block relative to the disk start that holds that block of the
1138 * file.
1139 */
1140sector_t bmap(struct inode * inode, sector_t block)
1141{
1142 sector_t res = 0;
1143 if (inode->i_mapping->a_ops->bmap)
1144 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1145 return res;
1146}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147EXPORT_SYMBOL(bmap);
1148
1149/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001150 * touch_atime - update the access time
1151 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001152 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 *
1154 * Update the accessed time on an inode and mark it for writeback.
1155 * This function automatically handles read only file systems and media,
1156 * as well as the "noatime" flag and inode specific "noatime" markers.
1157 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001158void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001160 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 struct timespec now;
1162
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 if (IS_RDONLY(inode))
1164 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001165 if (inode->i_flags & S_NOATIME)
1166 return;
1167 if (inode->i_sb->s_flags & MS_NOATIME)
1168 return;
1169 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001170 return;
1171
1172 /*
1173 * We may have a NULL vfsmount when coming from NFSD
1174 */
Andrew Mortonb2276132006-12-13 00:34:33 -08001175 if (mnt) {
1176 if (mnt->mnt_flags & MNT_NOATIME)
1177 return;
1178 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1179 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001180
1181 if (mnt->mnt_flags & MNT_RELATIME) {
1182 /*
1183 * With relative atime, only update atime if the
1184 * previous atime is earlier than either the ctime or
1185 * mtime.
1186 */
1187 if (timespec_compare(&inode->i_mtime,
1188 &inode->i_atime) < 0 &&
1189 timespec_compare(&inode->i_ctime,
1190 &inode->i_atime) < 0)
1191 return;
1192 }
Andrew Mortonb2276132006-12-13 00:34:33 -08001193 }
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001194
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 now = current_fs_time(inode->i_sb);
Valerie Henson47ae32d2006-12-13 00:34:34 -08001196 if (timespec_equal(&inode->i_atime, &now))
1197 return;
1198
1199 inode->i_atime = now;
1200 mark_inode_dirty_sync(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001202EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
1204/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001205 * file_update_time - update mtime and ctime time
1206 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001208 * Update the mtime and ctime members of an inode and mark the inode
1209 * for writeback. Note that this function is meant exclusively for
1210 * usage in the file write path of filesystems, and filesystems may
1211 * choose to explicitly ignore update via this function with the
1212 * S_NOCTIME inode flag, e.g. for network filesystem where these
1213 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 */
1215
Christoph Hellwig870f4812006-01-09 20:52:01 -08001216void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001218 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 struct timespec now;
1220 int sync_it = 0;
1221
1222 if (IS_NOCMTIME(inode))
1223 return;
1224 if (IS_RDONLY(inode))
1225 return;
1226
1227 now = current_fs_time(inode->i_sb);
Andreas Mohred97bd32006-10-02 02:17:17 -07001228 if (!timespec_equal(&inode->i_mtime, &now)) {
1229 inode->i_mtime = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001231 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
Andreas Mohred97bd32006-10-02 02:17:17 -07001233 if (!timespec_equal(&inode->i_ctime, &now)) {
1234 inode->i_ctime = now;
Christoph Hellwig870f4812006-01-09 20:52:01 -08001235 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001236 }
Christoph Hellwig870f4812006-01-09 20:52:01 -08001237
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 if (sync_it)
1239 mark_inode_dirty_sync(inode);
1240}
1241
Christoph Hellwig870f4812006-01-09 20:52:01 -08001242EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
1244int inode_needs_sync(struct inode *inode)
1245{
1246 if (IS_SYNC(inode))
1247 return 1;
1248 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1249 return 1;
1250 return 0;
1251}
1252
1253EXPORT_SYMBOL(inode_needs_sync);
1254
1255/*
1256 * Quota functions that want to walk the inode lists..
1257 */
1258#ifdef CONFIG_QUOTA
1259
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260void remove_dquot_ref(struct super_block *sb, int type,
1261 struct list_head *tofree_head)
1262{
1263 struct inode *inode;
1264
1265 if (!sb->dq_op)
1266 return; /* nothing to do */
1267 spin_lock(&inode_lock); /* This lock is for inodes code */
1268
1269 /*
1270 * We don't have to lock against quota code - test IS_QUOTAINIT is
1271 * just for speedup...
1272 */
1273 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1274 if (!IS_NOQUOTA(inode))
1275 remove_inode_dquot_ref(inode, type, tofree_head);
1276
1277 spin_unlock(&inode_lock);
1278}
1279
1280#endif
1281
1282int inode_wait(void *word)
1283{
1284 schedule();
1285 return 0;
1286}
1287
1288/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001289 * If we try to find an inode in the inode hash while it is being
1290 * deleted, we have to wait until the filesystem completes its
1291 * deletion before reporting that it isn't found. This function waits
1292 * until the deletion _might_ have completed. Callers are responsible
1293 * to recheck inode state.
1294 *
1295 * It doesn't matter if I_LOCK is not set initially, a call to
1296 * wake_up_inode() after removing from the hash list will DTRT.
1297 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 * This is called with inode_lock held.
1299 */
1300static void __wait_on_freeing_inode(struct inode *inode)
1301{
1302 wait_queue_head_t *wq;
1303 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1305 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1306 spin_unlock(&inode_lock);
1307 schedule();
1308 finish_wait(wq, &wait.wait);
1309 spin_lock(&inode_lock);
1310}
1311
1312void wake_up_inode(struct inode *inode)
1313{
1314 /*
1315 * Prevent speculative execution through spin_unlock(&inode_lock);
1316 */
1317 smp_mb();
1318 wake_up_bit(&inode->i_state, __I_LOCK);
1319}
1320
Mark Fasheh62752ee2006-10-17 10:31:38 +02001321/*
1322 * We rarely want to lock two inodes that do not have a parent/child
1323 * relationship (such as directory, child inode) simultaneously. The
1324 * vast majority of file systems should be able to get along fine
1325 * without this. Do not use these functions except as a last resort.
1326 */
1327void inode_double_lock(struct inode *inode1, struct inode *inode2)
1328{
1329 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1330 if (inode1)
1331 mutex_lock(&inode1->i_mutex);
1332 else if (inode2)
1333 mutex_lock(&inode2->i_mutex);
1334 return;
1335 }
1336
1337 if (inode1 < inode2) {
1338 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1339 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1340 } else {
1341 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1342 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1343 }
1344}
1345EXPORT_SYMBOL(inode_double_lock);
1346
1347void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1348{
1349 if (inode1)
1350 mutex_unlock(&inode1->i_mutex);
1351
1352 if (inode2 && inode2 != inode1)
1353 mutex_unlock(&inode2->i_mutex);
1354}
1355EXPORT_SYMBOL(inode_double_unlock);
1356
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357static __initdata unsigned long ihash_entries;
1358static int __init set_ihash_entries(char *str)
1359{
1360 if (!str)
1361 return 0;
1362 ihash_entries = simple_strtoul(str, &str, 0);
1363 return 1;
1364}
1365__setup("ihash_entries=", set_ihash_entries);
1366
1367/*
1368 * Initialize the waitqueues and inode hash table.
1369 */
1370void __init inode_init_early(void)
1371{
1372 int loop;
1373
1374 /* If hashes are distributed across NUMA nodes, defer
1375 * hash allocation until vmalloc space is available.
1376 */
1377 if (hashdist)
1378 return;
1379
1380 inode_hashtable =
1381 alloc_large_system_hash("Inode-cache",
1382 sizeof(struct hlist_head),
1383 ihash_entries,
1384 14,
1385 HASH_EARLY,
1386 &i_hash_shift,
1387 &i_hash_mask,
1388 0);
1389
1390 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1391 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1392}
1393
1394void __init inode_init(unsigned long mempages)
1395{
1396 int loop;
1397
1398 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001399 inode_cachep = kmem_cache_create("inode_cache",
1400 sizeof(struct inode),
1401 0,
1402 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1403 SLAB_MEM_SPREAD),
1404 init_once,
1405 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1407
1408 /* Hash may have been set up in inode_init_early */
1409 if (!hashdist)
1410 return;
1411
1412 inode_hashtable =
1413 alloc_large_system_hash("Inode-cache",
1414 sizeof(struct hlist_head),
1415 ihash_entries,
1416 14,
1417 0,
1418 &i_hash_shift,
1419 &i_hash_mask,
1420 0);
1421
1422 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1423 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1424}
1425
1426void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1427{
1428 inode->i_mode = mode;
1429 if (S_ISCHR(mode)) {
1430 inode->i_fop = &def_chr_fops;
1431 inode->i_rdev = rdev;
1432 } else if (S_ISBLK(mode)) {
1433 inode->i_fop = &def_blk_fops;
1434 inode->i_rdev = rdev;
1435 } else if (S_ISFIFO(mode))
1436 inode->i_fop = &def_fifo_fops;
1437 else if (S_ISSOCK(mode))
1438 inode->i_fop = &bad_sock_fops;
1439 else
1440 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1441 mode);
1442}
1443EXPORT_SYMBOL(init_special_inode);