blob: e6d93070f1402e249f9f1be017b4831033eab1a4 [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))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800417 reap += invalidate_mapping_pages(&inode->i_data,
418 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 iput(inode);
420 spin_lock(&inode_lock);
421
422 if (inode != list_entry(inode_unused.next,
423 struct inode, i_list))
424 continue; /* wrong inode or list_empty */
425 if (!can_unuse(inode))
426 continue;
427 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 list_move(&inode->i_list, &freeable);
429 inode->i_state |= I_FREEING;
430 nr_pruned++;
431 }
432 inodes_stat.nr_unused -= nr_pruned;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700433 if (current_is_kswapd())
434 __count_vm_events(KSWAPD_INODESTEAL, reap);
435 else
436 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 spin_unlock(&inode_lock);
438
439 dispose_list(&freeable);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800440 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441}
442
443/*
444 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
445 * "unused" means that no dentries are referring to the inodes: the files are
446 * not open and the dcache references to those inodes have already been
447 * reclaimed.
448 *
449 * This function is passed the number of inodes to scan, and it returns the
450 * total number of remaining possibly-reclaimable inodes.
451 */
Al Viro27496a82005-10-21 03:20:48 -0400452static int shrink_icache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
454 if (nr) {
455 /*
456 * Nasty deadlock avoidance. We may hold various FS locks,
457 * and we don't want to recurse into the FS that called us
458 * in clear_inode() and friends..
459 */
460 if (!(gfp_mask & __GFP_FS))
461 return -1;
462 prune_icache(nr);
463 }
464 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
465}
466
467static void __wait_on_freeing_inode(struct inode *inode);
468/*
469 * Called with the inode lock held.
470 * NOTE: we are not increasing the inode-refcount, you must call __iget()
471 * by hand after calling find_inode now! This simplifies iunique and won't
472 * add any additional branch in the common code.
473 */
474static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
475{
476 struct hlist_node *node;
477 struct inode * inode = NULL;
478
479repeat:
480 hlist_for_each (node, head) {
481 inode = hlist_entry(node, struct inode, i_hash);
482 if (inode->i_sb != sb)
483 continue;
484 if (!test(inode, data))
485 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700486 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 __wait_on_freeing_inode(inode);
488 goto repeat;
489 }
490 break;
491 }
492 return node ? inode : NULL;
493}
494
495/*
496 * find_inode_fast is the fast path version of find_inode, see the comment at
497 * iget_locked for details.
498 */
499static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
500{
501 struct hlist_node *node;
502 struct inode * inode = NULL;
503
504repeat:
505 hlist_for_each (node, head) {
506 inode = hlist_entry(node, struct inode, i_hash);
507 if (inode->i_ino != ino)
508 continue;
509 if (inode->i_sb != sb)
510 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700511 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 __wait_on_freeing_inode(inode);
513 goto repeat;
514 }
515 break;
516 }
517 return node ? inode : NULL;
518}
519
520/**
521 * new_inode - obtain an inode
522 * @sb: superblock
523 *
524 * Allocates a new inode for given superblock.
525 */
526struct inode *new_inode(struct super_block *sb)
527{
528 static unsigned long last_ino;
529 struct inode * inode;
530
531 spin_lock_prefetch(&inode_lock);
532
533 inode = alloc_inode(sb);
534 if (inode) {
535 spin_lock(&inode_lock);
536 inodes_stat.nr_inodes++;
537 list_add(&inode->i_list, &inode_in_use);
538 list_add(&inode->i_sb_list, &sb->s_inodes);
539 inode->i_ino = ++last_ino;
540 inode->i_state = 0;
541 spin_unlock(&inode_lock);
542 }
543 return inode;
544}
545
546EXPORT_SYMBOL(new_inode);
547
548void unlock_new_inode(struct inode *inode)
549{
550 /*
551 * This is special! We do not need the spinlock
552 * when clearing I_LOCK, because we're guaranteed
553 * that nobody else tries to do anything about the
554 * state of the inode when it is locked, as we
555 * just created it (so there can be no old holders
556 * that haven't tested I_LOCK).
557 */
558 inode->i_state &= ~(I_LOCK|I_NEW);
559 wake_up_inode(inode);
560}
561
562EXPORT_SYMBOL(unlock_new_inode);
563
564/*
565 * This is called without the inode lock held.. Be careful.
566 *
567 * We no longer cache the sb_flags in i_flags - see fs.h
568 * -- rmk@arm.uk.linux.org
569 */
570static 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)
571{
572 struct inode * inode;
573
574 inode = alloc_inode(sb);
575 if (inode) {
576 struct inode * old;
577
578 spin_lock(&inode_lock);
579 /* We released the lock, so.. */
580 old = find_inode(sb, head, test, data);
581 if (!old) {
582 if (set(inode, data))
583 goto set_failed;
584
585 inodes_stat.nr_inodes++;
586 list_add(&inode->i_list, &inode_in_use);
587 list_add(&inode->i_sb_list, &sb->s_inodes);
588 hlist_add_head(&inode->i_hash, head);
589 inode->i_state = I_LOCK|I_NEW;
590 spin_unlock(&inode_lock);
591
592 /* Return the locked inode with I_NEW set, the
593 * caller is responsible for filling in the contents
594 */
595 return inode;
596 }
597
598 /*
599 * Uhhuh, somebody else created the same inode under
600 * us. Use the old inode instead of the one we just
601 * allocated.
602 */
603 __iget(old);
604 spin_unlock(&inode_lock);
605 destroy_inode(inode);
606 inode = old;
607 wait_on_inode(inode);
608 }
609 return inode;
610
611set_failed:
612 spin_unlock(&inode_lock);
613 destroy_inode(inode);
614 return NULL;
615}
616
617/*
618 * get_new_inode_fast is the fast path version of get_new_inode, see the
619 * comment at iget_locked for details.
620 */
621static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
622{
623 struct inode * inode;
624
625 inode = alloc_inode(sb);
626 if (inode) {
627 struct inode * old;
628
629 spin_lock(&inode_lock);
630 /* We released the lock, so.. */
631 old = find_inode_fast(sb, head, ino);
632 if (!old) {
633 inode->i_ino = ino;
634 inodes_stat.nr_inodes++;
635 list_add(&inode->i_list, &inode_in_use);
636 list_add(&inode->i_sb_list, &sb->s_inodes);
637 hlist_add_head(&inode->i_hash, head);
638 inode->i_state = I_LOCK|I_NEW;
639 spin_unlock(&inode_lock);
640
641 /* Return the locked inode with I_NEW set, the
642 * caller is responsible for filling in the contents
643 */
644 return inode;
645 }
646
647 /*
648 * Uhhuh, somebody else created the same inode under
649 * us. Use the old inode instead of the one we just
650 * allocated.
651 */
652 __iget(old);
653 spin_unlock(&inode_lock);
654 destroy_inode(inode);
655 inode = old;
656 wait_on_inode(inode);
657 }
658 return inode;
659}
660
Andreas Mohred97bd32006-10-02 02:17:17 -0700661static unsigned long hash(struct super_block *sb, unsigned long hashval)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662{
663 unsigned long tmp;
664
665 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
666 L1_CACHE_BYTES;
667 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
668 return tmp & I_HASHMASK;
669}
670
671/**
672 * iunique - get a unique inode number
673 * @sb: superblock
674 * @max_reserved: highest reserved inode number
675 *
676 * Obtain an inode number that is unique on the system for a given
677 * superblock. This is used by file systems that have no natural
678 * permanent inode numbering system. An inode number is returned that
679 * is higher than the reserved limit but unique.
680 *
681 * BUGS:
682 * With a large number of inodes live on the file system this function
683 * currently becomes quite slow.
684 */
685ino_t iunique(struct super_block *sb, ino_t max_reserved)
686{
687 static ino_t counter;
688 struct inode *inode;
689 struct hlist_head * head;
690 ino_t res;
691 spin_lock(&inode_lock);
692retry:
693 if (counter > max_reserved) {
694 head = inode_hashtable + hash(sb,counter);
695 res = counter++;
696 inode = find_inode_fast(sb, head, res);
697 if (!inode) {
698 spin_unlock(&inode_lock);
699 return res;
700 }
701 } else {
702 counter = max_reserved + 1;
703 }
704 goto retry;
705
706}
707
708EXPORT_SYMBOL(iunique);
709
710struct inode *igrab(struct inode *inode)
711{
712 spin_lock(&inode_lock);
Jan Blunck4a3b0a42007-02-10 01:44:59 -0800713 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 __iget(inode);
715 else
716 /*
717 * Handle the case where s_op->clear_inode is not been
718 * called yet, and somebody is calling igrab
719 * while the inode is getting freed.
720 */
721 inode = NULL;
722 spin_unlock(&inode_lock);
723 return inode;
724}
725
726EXPORT_SYMBOL(igrab);
727
728/**
729 * ifind - internal function, you want ilookup5() or iget5().
730 * @sb: super block of file system to search
731 * @head: the head of the list to search
732 * @test: callback used for comparisons between inodes
733 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700734 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 *
736 * ifind() searches for the inode specified by @data in the inode
737 * cache. This is a generalized version of ifind_fast() for file systems where
738 * the inode number is not sufficient for unique identification of an inode.
739 *
740 * If the inode is in the cache, the inode is returned with an incremented
741 * reference count.
742 *
743 * Otherwise NULL is returned.
744 *
745 * Note, @test is called with the inode_lock held, so can't sleep.
746 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800747static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700749 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
751 struct inode *inode;
752
753 spin_lock(&inode_lock);
754 inode = find_inode(sb, head, test, data);
755 if (inode) {
756 __iget(inode);
757 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700758 if (likely(wait))
759 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 return inode;
761 }
762 spin_unlock(&inode_lock);
763 return NULL;
764}
765
766/**
767 * ifind_fast - internal function, you want ilookup() or iget().
768 * @sb: super block of file system to search
769 * @head: head of the list to search
770 * @ino: inode number to search for
771 *
772 * ifind_fast() searches for the inode @ino in the inode cache. This is for
773 * file systems where the inode number is sufficient for unique identification
774 * of an inode.
775 *
776 * If the inode is in the cache, the inode is returned with an incremented
777 * reference count.
778 *
779 * Otherwise NULL is returned.
780 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800781static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 struct hlist_head *head, unsigned long ino)
783{
784 struct inode *inode;
785
786 spin_lock(&inode_lock);
787 inode = find_inode_fast(sb, head, ino);
788 if (inode) {
789 __iget(inode);
790 spin_unlock(&inode_lock);
791 wait_on_inode(inode);
792 return inode;
793 }
794 spin_unlock(&inode_lock);
795 return NULL;
796}
797
798/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700799 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 * @sb: super block of file system to search
801 * @hashval: hash value (usually inode number) to search for
802 * @test: callback used for comparisons between inodes
803 * @data: opaque data pointer to pass to @test
804 *
805 * ilookup5() uses ifind() to search for the inode specified by @hashval and
806 * @data in the inode cache. This is a generalized version of ilookup() for
807 * file systems where the inode number is not sufficient for unique
808 * identification of an inode.
809 *
810 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700811 * reference count. Note, the inode lock is not waited upon so you have to be
812 * very careful what you do with the returned inode. You probably should be
813 * using ilookup5() instead.
814 *
815 * Otherwise NULL is returned.
816 *
817 * Note, @test is called with the inode_lock held, so can't sleep.
818 */
819struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
820 int (*test)(struct inode *, void *), void *data)
821{
822 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
823
824 return ifind(sb, head, test, data, 0);
825}
826
827EXPORT_SYMBOL(ilookup5_nowait);
828
829/**
830 * ilookup5 - search for an inode in the inode cache
831 * @sb: super block of file system to search
832 * @hashval: hash value (usually inode number) to search for
833 * @test: callback used for comparisons between inodes
834 * @data: opaque data pointer to pass to @test
835 *
836 * ilookup5() uses ifind() to search for the inode specified by @hashval and
837 * @data in the inode cache. This is a generalized version of ilookup() for
838 * file systems where the inode number is not sufficient for unique
839 * identification of an inode.
840 *
841 * If the inode is in the cache, the inode lock is waited upon and the inode is
842 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 *
844 * Otherwise NULL is returned.
845 *
846 * Note, @test is called with the inode_lock held, so can't sleep.
847 */
848struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
849 int (*test)(struct inode *, void *), void *data)
850{
851 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
852
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700853 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854}
855
856EXPORT_SYMBOL(ilookup5);
857
858/**
859 * ilookup - search for an inode in the inode cache
860 * @sb: super block of file system to search
861 * @ino: inode number to search for
862 *
863 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
864 * This is for file systems where the inode number is sufficient for unique
865 * identification of an inode.
866 *
867 * If the inode is in the cache, the inode is returned with an incremented
868 * reference count.
869 *
870 * Otherwise NULL is returned.
871 */
872struct inode *ilookup(struct super_block *sb, unsigned long ino)
873{
874 struct hlist_head *head = inode_hashtable + hash(sb, ino);
875
876 return ifind_fast(sb, head, ino);
877}
878
879EXPORT_SYMBOL(ilookup);
880
881/**
882 * iget5_locked - obtain an inode from a mounted file system
883 * @sb: super block of file system
884 * @hashval: hash value (usually inode number) to get
885 * @test: callback used for comparisons between inodes
886 * @set: callback used to initialize a new struct inode
887 * @data: opaque data pointer to pass to @test and @set
888 *
889 * This is iget() without the read_inode() portion of get_new_inode().
890 *
891 * iget5_locked() uses ifind() to search for the inode specified by @hashval
892 * and @data in the inode cache and if present it is returned with an increased
893 * reference count. This is a generalized version of iget_locked() for file
894 * systems where the inode number is not sufficient for unique identification
895 * of an inode.
896 *
897 * If the inode is not in cache, get_new_inode() is called to allocate a new
898 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
899 * file system gets to fill it in before unlocking it via unlock_new_inode().
900 *
901 * Note both @test and @set are called with the inode_lock held, so can't sleep.
902 */
903struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
904 int (*test)(struct inode *, void *),
905 int (*set)(struct inode *, void *), void *data)
906{
907 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
908 struct inode *inode;
909
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700910 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 if (inode)
912 return inode;
913 /*
914 * get_new_inode() will do the right thing, re-trying the search
915 * in case it had to block at any point.
916 */
917 return get_new_inode(sb, head, test, set, data);
918}
919
920EXPORT_SYMBOL(iget5_locked);
921
922/**
923 * iget_locked - obtain an inode from a mounted file system
924 * @sb: super block of file system
925 * @ino: inode number to get
926 *
927 * This is iget() without the read_inode() portion of get_new_inode_fast().
928 *
929 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
930 * the inode cache and if present it is returned with an increased reference
931 * count. This is for file systems where the inode number is sufficient for
932 * unique identification of an inode.
933 *
934 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
935 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
936 * The file system gets to fill it in before unlocking it via
937 * unlock_new_inode().
938 */
939struct inode *iget_locked(struct super_block *sb, unsigned long ino)
940{
941 struct hlist_head *head = inode_hashtable + hash(sb, ino);
942 struct inode *inode;
943
944 inode = ifind_fast(sb, head, ino);
945 if (inode)
946 return inode;
947 /*
948 * get_new_inode_fast() will do the right thing, re-trying the search
949 * in case it had to block at any point.
950 */
951 return get_new_inode_fast(sb, head, ino);
952}
953
954EXPORT_SYMBOL(iget_locked);
955
956/**
957 * __insert_inode_hash - hash an inode
958 * @inode: unhashed inode
959 * @hashval: unsigned long value used to locate this object in the
960 * inode_hashtable.
961 *
962 * Add an inode to the inode hash for this superblock.
963 */
964void __insert_inode_hash(struct inode *inode, unsigned long hashval)
965{
966 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
967 spin_lock(&inode_lock);
968 hlist_add_head(&inode->i_hash, head);
969 spin_unlock(&inode_lock);
970}
971
972EXPORT_SYMBOL(__insert_inode_hash);
973
974/**
975 * remove_inode_hash - remove an inode from the hash
976 * @inode: inode to unhash
977 *
978 * Remove an inode from the superblock.
979 */
980void remove_inode_hash(struct inode *inode)
981{
982 spin_lock(&inode_lock);
983 hlist_del_init(&inode->i_hash);
984 spin_unlock(&inode_lock);
985}
986
987EXPORT_SYMBOL(remove_inode_hash);
988
989/*
990 * Tell the filesystem that this inode is no longer of any interest and should
991 * be completely destroyed.
992 *
993 * We leave the inode in the inode hash table until *after* the filesystem's
994 * ->delete_inode completes. This ensures that an iget (such as nfsd might
995 * instigate) will always find up-to-date information either in the hash or on
996 * disk.
997 *
998 * I_FREEING is set so that no-one will take a new reference to the inode while
999 * it is being deleted.
1000 */
1001void generic_delete_inode(struct inode *inode)
1002{
1003 struct super_operations *op = inode->i_sb->s_op;
1004
1005 list_del_init(&inode->i_list);
1006 list_del_init(&inode->i_sb_list);
Andreas Mohred97bd32006-10-02 02:17:17 -07001007 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 inodes_stat.nr_inodes--;
1009 spin_unlock(&inode_lock);
1010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 security_inode_delete(inode);
1012
1013 if (op->delete_inode) {
1014 void (*delete)(struct inode *) = op->delete_inode;
1015 if (!is_bad_inode(inode))
1016 DQUOT_INIT(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001017 /* Filesystems implementing their own
1018 * s_op->delete_inode are required to call
1019 * truncate_inode_pages and clear_inode()
1020 * internally */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 delete(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001022 } else {
1023 truncate_inode_pages(&inode->i_data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 clear_inode(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001025 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 spin_lock(&inode_lock);
1027 hlist_del_init(&inode->i_hash);
1028 spin_unlock(&inode_lock);
1029 wake_up_inode(inode);
Eric Sesterhennb7542f82006-04-02 13:38:18 +02001030 BUG_ON(inode->i_state != I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 destroy_inode(inode);
1032}
1033
1034EXPORT_SYMBOL(generic_delete_inode);
1035
1036static void generic_forget_inode(struct inode *inode)
1037{
1038 struct super_block *sb = inode->i_sb;
1039
1040 if (!hlist_unhashed(&inode->i_hash)) {
1041 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1042 list_move(&inode->i_list, &inode_unused);
1043 inodes_stat.nr_unused++;
Alexander Viro991114c2005-06-23 00:09:01 -07001044 if (!sb || (sb->s_flags & MS_ACTIVE)) {
1045 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001047 }
1048 inode->i_state |= I_WILL_FREE;
1049 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 write_inode_now(inode, 1);
1051 spin_lock(&inode_lock);
Alexander Viro991114c2005-06-23 00:09:01 -07001052 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 inodes_stat.nr_unused--;
1054 hlist_del_init(&inode->i_hash);
1055 }
1056 list_del_init(&inode->i_list);
1057 list_del_init(&inode->i_sb_list);
Alexander Viro991114c2005-06-23 00:09:01 -07001058 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 inodes_stat.nr_inodes--;
1060 spin_unlock(&inode_lock);
1061 if (inode->i_data.nrpages)
1062 truncate_inode_pages(&inode->i_data, 0);
1063 clear_inode(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001064 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 destroy_inode(inode);
1066}
1067
1068/*
1069 * Normal UNIX filesystem behaviour: delete the
1070 * inode when the usage count drops to zero, and
1071 * i_nlink is zero.
1072 */
Mark Fashehcb2c0232005-07-07 17:56:03 -07001073void generic_drop_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074{
1075 if (!inode->i_nlink)
1076 generic_delete_inode(inode);
1077 else
1078 generic_forget_inode(inode);
1079}
1080
Mark Fashehcb2c0232005-07-07 17:56:03 -07001081EXPORT_SYMBOL_GPL(generic_drop_inode);
1082
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083/*
1084 * Called when we're dropping the last reference
1085 * to an inode.
1086 *
1087 * Call the FS "drop()" function, defaulting to
1088 * the legacy UNIX filesystem behaviour..
1089 *
1090 * NOTE! NOTE! NOTE! We're called with the inode lock
1091 * held, and the drop function is supposed to release
1092 * the lock!
1093 */
1094static inline void iput_final(struct inode *inode)
1095{
1096 struct super_operations *op = inode->i_sb->s_op;
1097 void (*drop)(struct inode *) = generic_drop_inode;
1098
1099 if (op && op->drop_inode)
1100 drop = op->drop_inode;
1101 drop(inode);
1102}
1103
1104/**
1105 * iput - put an inode
1106 * @inode: inode to put
1107 *
1108 * Puts an inode, dropping its usage count. If the inode use count hits
1109 * zero, the inode is then freed and may also be destroyed.
1110 *
1111 * Consequently, iput() can sleep.
1112 */
1113void iput(struct inode *inode)
1114{
1115 if (inode) {
1116 struct super_operations *op = inode->i_sb->s_op;
1117
1118 BUG_ON(inode->i_state == I_CLEAR);
1119
1120 if (op && op->put_inode)
1121 op->put_inode(inode);
1122
1123 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1124 iput_final(inode);
1125 }
1126}
1127
1128EXPORT_SYMBOL(iput);
1129
1130/**
1131 * bmap - find a block number in a file
1132 * @inode: inode of file
1133 * @block: block to find
1134 *
1135 * Returns the block number on the device holding the inode that
1136 * is the disk block number for the block of the file requested.
1137 * That is, asked for block 4 of inode 1 the function will return the
1138 * disk block relative to the disk start that holds that block of the
1139 * file.
1140 */
1141sector_t bmap(struct inode * inode, sector_t block)
1142{
1143 sector_t res = 0;
1144 if (inode->i_mapping->a_ops->bmap)
1145 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1146 return res;
1147}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148EXPORT_SYMBOL(bmap);
1149
1150/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001151 * touch_atime - update the access time
1152 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001153 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 *
1155 * Update the accessed time on an inode and mark it for writeback.
1156 * This function automatically handles read only file systems and media,
1157 * as well as the "noatime" flag and inode specific "noatime" markers.
1158 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001159void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001161 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 struct timespec now;
1163
Andrew Mortonb2276132006-12-13 00:34:33 -08001164 if (inode->i_flags & S_NOATIME)
1165 return;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001166 if (IS_NOATIME(inode))
Andrew Mortonb2276132006-12-13 00:34:33 -08001167 return;
1168 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001169 return;
1170
1171 /*
1172 * We may have a NULL vfsmount when coming from NFSD
1173 */
Andrew Mortonb2276132006-12-13 00:34:33 -08001174 if (mnt) {
1175 if (mnt->mnt_flags & MNT_NOATIME)
1176 return;
1177 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1178 return;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001179
1180 if (mnt->mnt_flags & MNT_RELATIME) {
1181 /*
1182 * With relative atime, only update atime if the
1183 * previous atime is earlier than either the ctime or
1184 * mtime.
1185 */
1186 if (timespec_compare(&inode->i_mtime,
1187 &inode->i_atime) < 0 &&
1188 timespec_compare(&inode->i_ctime,
1189 &inode->i_atime) < 0)
1190 return;
1191 }
Andrew Mortonb2276132006-12-13 00:34:33 -08001192 }
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001193
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 now = current_fs_time(inode->i_sb);
Valerie Henson47ae32d2006-12-13 00:34:34 -08001195 if (timespec_equal(&inode->i_atime, &now))
1196 return;
1197
1198 inode->i_atime = now;
1199 mark_inode_dirty_sync(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001201EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202
1203/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001204 * file_update_time - update mtime and ctime time
1205 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001207 * Update the mtime and ctime members of an inode and mark the inode
1208 * for writeback. Note that this function is meant exclusively for
1209 * usage in the file write path of filesystems, and filesystems may
1210 * choose to explicitly ignore update via this function with the
1211 * S_NOCTIME inode flag, e.g. for network filesystem where these
1212 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 */
1214
Christoph Hellwig870f4812006-01-09 20:52:01 -08001215void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001217 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 struct timespec now;
1219 int sync_it = 0;
1220
1221 if (IS_NOCMTIME(inode))
1222 return;
1223 if (IS_RDONLY(inode))
1224 return;
1225
1226 now = current_fs_time(inode->i_sb);
Andreas Mohred97bd32006-10-02 02:17:17 -07001227 if (!timespec_equal(&inode->i_mtime, &now)) {
1228 inode->i_mtime = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
Andreas Mohred97bd32006-10-02 02:17:17 -07001232 if (!timespec_equal(&inode->i_ctime, &now)) {
1233 inode->i_ctime = now;
Christoph Hellwig870f4812006-01-09 20:52:01 -08001234 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001235 }
Christoph Hellwig870f4812006-01-09 20:52:01 -08001236
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 if (sync_it)
1238 mark_inode_dirty_sync(inode);
1239}
1240
Christoph Hellwig870f4812006-01-09 20:52:01 -08001241EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
1243int inode_needs_sync(struct inode *inode)
1244{
1245 if (IS_SYNC(inode))
1246 return 1;
1247 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1248 return 1;
1249 return 0;
1250}
1251
1252EXPORT_SYMBOL(inode_needs_sync);
1253
1254/*
1255 * Quota functions that want to walk the inode lists..
1256 */
1257#ifdef CONFIG_QUOTA
1258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259void remove_dquot_ref(struct super_block *sb, int type,
1260 struct list_head *tofree_head)
1261{
1262 struct inode *inode;
1263
1264 if (!sb->dq_op)
1265 return; /* nothing to do */
1266 spin_lock(&inode_lock); /* This lock is for inodes code */
1267
1268 /*
1269 * We don't have to lock against quota code - test IS_QUOTAINIT is
1270 * just for speedup...
1271 */
1272 list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1273 if (!IS_NOQUOTA(inode))
1274 remove_inode_dquot_ref(inode, type, tofree_head);
1275
1276 spin_unlock(&inode_lock);
1277}
1278
1279#endif
1280
1281int inode_wait(void *word)
1282{
1283 schedule();
1284 return 0;
1285}
1286
1287/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001288 * If we try to find an inode in the inode hash while it is being
1289 * deleted, we have to wait until the filesystem completes its
1290 * deletion before reporting that it isn't found. This function waits
1291 * until the deletion _might_ have completed. Callers are responsible
1292 * to recheck inode state.
1293 *
1294 * It doesn't matter if I_LOCK is not set initially, a call to
1295 * wake_up_inode() after removing from the hash list will DTRT.
1296 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 * This is called with inode_lock held.
1298 */
1299static void __wait_on_freeing_inode(struct inode *inode)
1300{
1301 wait_queue_head_t *wq;
1302 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1304 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1305 spin_unlock(&inode_lock);
1306 schedule();
1307 finish_wait(wq, &wait.wait);
1308 spin_lock(&inode_lock);
1309}
1310
1311void wake_up_inode(struct inode *inode)
1312{
1313 /*
1314 * Prevent speculative execution through spin_unlock(&inode_lock);
1315 */
1316 smp_mb();
1317 wake_up_bit(&inode->i_state, __I_LOCK);
1318}
1319
Mark Fasheh62752ee2006-10-17 10:31:38 +02001320/*
1321 * We rarely want to lock two inodes that do not have a parent/child
1322 * relationship (such as directory, child inode) simultaneously. The
1323 * vast majority of file systems should be able to get along fine
1324 * without this. Do not use these functions except as a last resort.
1325 */
1326void inode_double_lock(struct inode *inode1, struct inode *inode2)
1327{
1328 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1329 if (inode1)
1330 mutex_lock(&inode1->i_mutex);
1331 else if (inode2)
1332 mutex_lock(&inode2->i_mutex);
1333 return;
1334 }
1335
1336 if (inode1 < inode2) {
1337 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1338 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1339 } else {
1340 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1341 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1342 }
1343}
1344EXPORT_SYMBOL(inode_double_lock);
1345
1346void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1347{
1348 if (inode1)
1349 mutex_unlock(&inode1->i_mutex);
1350
1351 if (inode2 && inode2 != inode1)
1352 mutex_unlock(&inode2->i_mutex);
1353}
1354EXPORT_SYMBOL(inode_double_unlock);
1355
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356static __initdata unsigned long ihash_entries;
1357static int __init set_ihash_entries(char *str)
1358{
1359 if (!str)
1360 return 0;
1361 ihash_entries = simple_strtoul(str, &str, 0);
1362 return 1;
1363}
1364__setup("ihash_entries=", set_ihash_entries);
1365
1366/*
1367 * Initialize the waitqueues and inode hash table.
1368 */
1369void __init inode_init_early(void)
1370{
1371 int loop;
1372
1373 /* If hashes are distributed across NUMA nodes, defer
1374 * hash allocation until vmalloc space is available.
1375 */
1376 if (hashdist)
1377 return;
1378
1379 inode_hashtable =
1380 alloc_large_system_hash("Inode-cache",
1381 sizeof(struct hlist_head),
1382 ihash_entries,
1383 14,
1384 HASH_EARLY,
1385 &i_hash_shift,
1386 &i_hash_mask,
1387 0);
1388
1389 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1390 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1391}
1392
1393void __init inode_init(unsigned long mempages)
1394{
1395 int loop;
1396
1397 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001398 inode_cachep = kmem_cache_create("inode_cache",
1399 sizeof(struct inode),
1400 0,
1401 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1402 SLAB_MEM_SPREAD),
1403 init_once,
1404 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1406
1407 /* Hash may have been set up in inode_init_early */
1408 if (!hashdist)
1409 return;
1410
1411 inode_hashtable =
1412 alloc_large_system_hash("Inode-cache",
1413 sizeof(struct hlist_head),
1414 ihash_entries,
1415 14,
1416 0,
1417 &i_hash_shift,
1418 &i_hash_mask,
1419 0);
1420
1421 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1422 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1423}
1424
1425void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1426{
1427 inode->i_mode = mode;
1428 if (S_ISCHR(mode)) {
1429 inode->i_fop = &def_chr_fops;
1430 inode->i_rdev = rdev;
1431 } else if (S_ISBLK(mode)) {
1432 inode->i_fop = &def_blk_fops;
1433 inode->i_rdev = rdev;
1434 } else if (S_ISFIFO(mode))
1435 inode->i_fop = &def_fifo_fops;
1436 else if (S_ISSOCK(mode))
1437 inode->i_fop = &bad_sock_fops;
1438 else
1439 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1440 mode);
1441}
1442EXPORT_SYMBOL(init_special_inode);