blob: d06d6d268de9a12c5fbe9d66bd618c293729e8e9 [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>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050020#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/pagemap.h>
22#include <linux/cdev.h>
23#include <linux/bootmem.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/inotify.h>
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -080025#include <linux/mount.h>
Arjan van de Venefaee192009-01-06 07:20:54 -080026#include <linux/async.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
28/*
29 * This is needed for the following functions:
30 * - inode_has_buffers
31 * - invalidate_inode_buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 * - invalidate_bdev
33 *
34 * FIXME: remove all knowledge of the buffer layer from this file
35 */
36#include <linux/buffer_head.h>
37
38/*
39 * New inode.c implementation.
40 *
41 * This implementation has the basic premise of trying
42 * to be extremely low-overhead and SMP-safe, yet be
43 * simple enough to be "obviously correct".
44 *
45 * Famous last words.
46 */
47
48/* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
49
50/* #define INODE_PARANOIA 1 */
51/* #define INODE_DEBUG 1 */
52
53/*
54 * Inode lookup is no longer as critical as it used to be:
55 * most of the lookups are going to be through the dcache.
56 */
57#define I_HASHBITS i_hash_shift
58#define I_HASHMASK i_hash_mask
59
Eric Dumazetfa3536c2006-03-26 01:37:24 -080060static unsigned int i_hash_mask __read_mostly;
61static unsigned int i_hash_shift __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
63/*
64 * Each inode can be on two separate lists. One is
65 * the hash list of the inode, used for lookups. The
66 * other linked list is the "type" list:
67 * "in_use" - valid inode, i_count > 0, i_nlink > 0
68 * "dirty" - as "in_use" but also dirty
69 * "unused" - valid inode, i_count = 0
70 *
71 * A "dirty" list is maintained for each super block,
72 * allowing for low-overhead inode sync() operations.
73 */
74
75LIST_HEAD(inode_in_use);
76LIST_HEAD(inode_unused);
Eric Dumazetfa3536c2006-03-26 01:37:24 -080077static struct hlist_head *inode_hashtable __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
79/*
80 * A simple spinlock to protect the list manipulations.
81 *
82 * NOTE! You also have to own the lock if you change
83 * the i_state of an inode while it is in use..
84 */
85DEFINE_SPINLOCK(inode_lock);
86
87/*
Ingo Molnarf24075b2006-03-23 03:00:34 -080088 * iprune_mutex provides exclusion between the kswapd or try_to_free_pages
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 * icache shrinking path, and the umount path. Without this exclusion,
90 * by the time prune_icache calls iput for the inode whose pages it has
91 * been invalidating, or by the time it calls clear_inode & destroy_inode
92 * from its final dispose_list, the struct super_block they refer to
93 * (for inode->i_sb->s_op) may already have been freed and reused.
94 */
Adrian Bunkbdfc3262006-03-25 03:06:56 -080095static DEFINE_MUTEX(iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
97/*
98 * Statistics gathering..
99 */
100struct inodes_stat_t inodes_stat;
101
Christoph Lametere18b8902006-12-06 20:33:20 -0800102static struct kmem_cache * inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700104static void wake_up_inode(struct inode *inode)
105{
106 /*
107 * Prevent speculative execution through spin_unlock(&inode_lock);
108 */
109 smp_mb();
110 wake_up_bit(&inode->i_state, __I_LOCK);
111}
112
David Chinner2cb15992008-10-30 17:32:23 +1100113/**
114 * inode_init_always - perform inode structure intialisation
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800115 * @sb: superblock inode belongs to
116 * @inode: inode to initialise
David Chinner2cb15992008-10-30 17:32:23 +1100117 *
118 * These are initializations that need to be done on every inode
119 * allocation as the fields are not initialised by slab allocation.
120 */
121struct inode *inode_init_always(struct super_block *sb, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122{
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700123 static const struct address_space_operations empty_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 static struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800125 static const struct file_operations empty_fops;
David Chinner2cb15992008-10-30 17:32:23 +1100126
127 struct address_space * const mapping = &inode->i_data;
128
129 inode->i_sb = sb;
130 inode->i_blkbits = sb->s_blocksize_bits;
131 inode->i_flags = 0;
132 atomic_set(&inode->i_count, 1);
133 inode->i_op = &empty_iops;
134 inode->i_fop = &empty_fops;
135 inode->i_nlink = 1;
Al Viro56ff5ef2008-12-09 09:34:39 -0500136 inode->i_uid = 0;
137 inode->i_gid = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100138 atomic_set(&inode->i_writecount, 0);
139 inode->i_size = 0;
140 inode->i_blocks = 0;
141 inode->i_bytes = 0;
142 inode->i_generation = 0;
143#ifdef CONFIG_QUOTA
144 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
145#endif
146 inode->i_pipe = NULL;
147 inode->i_bdev = NULL;
148 inode->i_cdev = NULL;
149 inode->i_rdev = 0;
150 inode->dirtied_when = 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500151
152 if (security_inode_alloc(inode))
153 goto out_free_inode;
154
155 /* allocate and initialize an i_integrity */
156 if (ima_inode_alloc(inode))
157 goto out_free_security;
David Chinner2cb15992008-10-30 17:32:23 +1100158
159 spin_lock_init(&inode->i_lock);
160 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
161
162 mutex_init(&inode->i_mutex);
163 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
164
165 init_rwsem(&inode->i_alloc_sem);
166 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
167
168 mapping->a_ops = &empty_aops;
169 mapping->host = inode;
170 mapping->flags = 0;
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800171 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
David Chinner2cb15992008-10-30 17:32:23 +1100172 mapping->assoc_mapping = NULL;
173 mapping->backing_dev_info = &default_backing_dev_info;
174 mapping->writeback_index = 0;
175
176 /*
177 * If the block_device provides a backing_dev_info for client
178 * inodes then use that. Otherwise the inode share the bdev's
179 * backing_dev_info.
180 */
181 if (sb->s_bdev) {
182 struct backing_dev_info *bdi;
183
184 bdi = sb->s_bdev->bd_inode_backing_dev_info;
185 if (!bdi)
186 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
187 mapping->backing_dev_info = bdi;
188 }
189 inode->i_private = NULL;
190 inode->i_mapping = mapping;
191
192 return inode;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500193
194out_free_security:
195 security_inode_free(inode);
196out_free_inode:
197 if (inode->i_sb->s_op->destroy_inode)
198 inode->i_sb->s_op->destroy_inode(inode);
199 else
200 kmem_cache_free(inode_cachep, (inode));
201 return NULL;
David Chinner2cb15992008-10-30 17:32:23 +1100202}
203EXPORT_SYMBOL(inode_init_always);
204
205static struct inode *alloc_inode(struct super_block *sb)
206{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 struct inode *inode;
208
209 if (sb->s_op->alloc_inode)
210 inode = sb->s_op->alloc_inode(sb);
211 else
David Chinner2cb15992008-10-30 17:32:23 +1100212 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
David Chinner2cb15992008-10-30 17:32:23 +1100214 if (inode)
215 return inode_init_always(sb, inode);
216 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217}
218
219void destroy_inode(struct inode *inode)
220{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200221 BUG_ON(inode_has_buffers(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 security_inode_free(inode);
223 if (inode->i_sb->s_op->destroy_inode)
224 inode->i_sb->s_op->destroy_inode(inode);
225 else
226 kmem_cache_free(inode_cachep, (inode));
227}
Christoph Hellwig087e3b02008-10-30 18:24:37 +1100228EXPORT_SYMBOL(destroy_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
230
231/*
232 * These are initializations that only need to be done
233 * once, because the fields are idempotent across use
234 * of the inode, so let the slab aware of that.
235 */
236void inode_init_once(struct inode *inode)
237{
238 memset(inode, 0, sizeof(*inode));
239 INIT_HLIST_NODE(&inode->i_hash);
240 INIT_LIST_HEAD(&inode->i_dentry);
241 INIT_LIST_HEAD(&inode->i_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
Nick Piggin19fd6232008-07-25 19:45:32 -0700243 spin_lock_init(&inode->i_data.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 spin_lock_init(&inode->i_data.i_mmap_lock);
245 INIT_LIST_HEAD(&inode->i_data.private_list);
246 spin_lock_init(&inode->i_data.private_lock);
247 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
248 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 i_size_ordered_init(inode);
Robert Love0eeca282005-07-12 17:06:03 -0400250#ifdef CONFIG_INOTIFY
251 INIT_LIST_HEAD(&inode->inotify_watches);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800252 mutex_init(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400253#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254}
255
256EXPORT_SYMBOL(inode_init_once);
257
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700258static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259{
260 struct inode * inode = (struct inode *) foo;
261
Christoph Lametera35afb82007-05-16 22:10:57 -0700262 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263}
264
265/*
266 * inode_lock must be held
267 */
268void __iget(struct inode * inode)
269{
270 if (atomic_read(&inode->i_count)) {
271 atomic_inc(&inode->i_count);
272 return;
273 }
274 atomic_inc(&inode->i_count);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700275 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 list_move(&inode->i_list, &inode_in_use);
277 inodes_stat.nr_unused--;
278}
279
280/**
281 * clear_inode - clear an inode
282 * @inode: inode to clear
283 *
284 * This is called by the filesystem to tell us
285 * that the inode is no longer useful. We just
286 * terminate it with extreme prejudice.
287 */
288void clear_inode(struct inode *inode)
289{
290 might_sleep();
291 invalidate_inode_buffers(inode);
292
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200293 BUG_ON(inode->i_data.nrpages);
294 BUG_ON(!(inode->i_state & I_FREEING));
295 BUG_ON(inode->i_state & I_CLEAR);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700296 inode_sync_wait(inode);
Jan Kara9e3509e2009-01-26 16:45:12 +0100297 vfs_dq_drop(inode);
Christoph Hellwigacb0c852007-05-08 00:25:52 -0700298 if (inode->i_sb->s_op->clear_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 inode->i_sb->s_op->clear_inode(inode);
Theodore Ts'oeaf796e2006-09-27 01:50:48 -0700300 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 bd_forget(inode);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700302 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 cd_forget(inode);
304 inode->i_state = I_CLEAR;
305}
306
307EXPORT_SYMBOL(clear_inode);
308
309/*
310 * dispose_list - dispose of the contents of a local list
311 * @head: the head of the list to free
312 *
313 * Dispose-list gets a local list with local inodes in it, so it doesn't
314 * need to worry about list corruption and SMP locks.
315 */
316static void dispose_list(struct list_head *head)
317{
318 int nr_disposed = 0;
319
320 while (!list_empty(head)) {
321 struct inode *inode;
322
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700323 inode = list_first_entry(head, struct inode, i_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 list_del(&inode->i_list);
325
326 if (inode->i_data.nrpages)
327 truncate_inode_pages(&inode->i_data, 0);
328 clear_inode(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700329
330 spin_lock(&inode_lock);
331 hlist_del_init(&inode->i_hash);
332 list_del_init(&inode->i_sb_list);
333 spin_unlock(&inode_lock);
334
335 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 destroy_inode(inode);
337 nr_disposed++;
338 }
339 spin_lock(&inode_lock);
340 inodes_stat.nr_inodes -= nr_disposed;
341 spin_unlock(&inode_lock);
342}
343
344/*
345 * Invalidate all inodes for a device.
346 */
347static int invalidate_list(struct list_head *head, struct list_head *dispose)
348{
349 struct list_head *next;
350 int busy = 0, count = 0;
351
352 next = head->next;
353 for (;;) {
354 struct list_head * tmp = next;
355 struct inode * inode;
356
357 /*
358 * We can reschedule here without worrying about the list's
359 * consistency because the per-sb list of inodes must not
Ingo Molnarf24075b2006-03-23 03:00:34 -0800360 * change during umount anymore, and because iprune_mutex keeps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 * shrink_icache_memory() away.
362 */
363 cond_resched_lock(&inode_lock);
364
365 next = next->next;
366 if (tmp == head)
367 break;
368 inode = list_entry(tmp, struct inode, i_sb_list);
Nick Pigginaabb8fd2009-03-11 13:17:36 -0700369 if (inode->i_state & I_NEW)
370 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 invalidate_inode_buffers(inode);
372 if (!atomic_read(&inode->i_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 list_move(&inode->i_list, dispose);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700374 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 inode->i_state |= I_FREEING;
376 count++;
377 continue;
378 }
379 busy = 1;
380 }
381 /* only unused inodes may be cached with i_count zero */
382 inodes_stat.nr_unused -= count;
383 return busy;
384}
385
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386/**
387 * invalidate_inodes - discard the inodes on a device
388 * @sb: superblock
389 *
390 * Discard all of the inodes for a given superblock. If the discard
391 * fails because there are busy inodes then a non zero value is returned.
392 * If the discard is successful all the inodes have been discarded.
393 */
394int invalidate_inodes(struct super_block * sb)
395{
396 int busy;
397 LIST_HEAD(throw_away);
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);
Robert Love0eeca282005-07-12 17:06:03 -0400401 inotify_unmount_inodes(&sb->s_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 busy = invalidate_list(&sb->s_inodes, &throw_away);
403 spin_unlock(&inode_lock);
404
405 dispose_list(&throw_away);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800406 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
408 return busy;
409}
410
411EXPORT_SYMBOL(invalidate_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
413static int can_unuse(struct inode *inode)
414{
415 if (inode->i_state)
416 return 0;
417 if (inode_has_buffers(inode))
418 return 0;
419 if (atomic_read(&inode->i_count))
420 return 0;
421 if (inode->i_data.nrpages)
422 return 0;
423 return 1;
424}
425
426/*
427 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
428 * a temporary list and then are freed outside inode_lock by dispose_list().
429 *
430 * Any inodes which are pinned purely because of attached pagecache have their
431 * pagecache removed. We expect the final iput() on that inode to add it to
432 * the front of the inode_unused list. So look for it there and if the
433 * inode is still freeable, proceed. The right inode is found 99.9% of the
434 * time in testing on a 4-way.
435 *
436 * If the inode has metadata buffers attached to mapping->private_list then
437 * try to remove them.
438 */
439static void prune_icache(int nr_to_scan)
440{
441 LIST_HEAD(freeable);
442 int nr_pruned = 0;
443 int nr_scanned;
444 unsigned long reap = 0;
445
Ingo Molnarf24075b2006-03-23 03:00:34 -0800446 mutex_lock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 spin_lock(&inode_lock);
448 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
449 struct inode *inode;
450
451 if (list_empty(&inode_unused))
452 break;
453
454 inode = list_entry(inode_unused.prev, struct inode, i_list);
455
456 if (inode->i_state || atomic_read(&inode->i_count)) {
457 list_move(&inode->i_list, &inode_unused);
458 continue;
459 }
460 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
461 __iget(inode);
462 spin_unlock(&inode_lock);
463 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800464 reap += invalidate_mapping_pages(&inode->i_data,
465 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 iput(inode);
467 spin_lock(&inode_lock);
468
469 if (inode != list_entry(inode_unused.next,
470 struct inode, i_list))
471 continue; /* wrong inode or list_empty */
472 if (!can_unuse(inode))
473 continue;
474 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 list_move(&inode->i_list, &freeable);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700476 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 inode->i_state |= I_FREEING;
478 nr_pruned++;
479 }
480 inodes_stat.nr_unused -= nr_pruned;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700481 if (current_is_kswapd())
482 __count_vm_events(KSWAPD_INODESTEAL, reap);
483 else
484 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 spin_unlock(&inode_lock);
486
487 dispose_list(&freeable);
Ingo Molnarf24075b2006-03-23 03:00:34 -0800488 mutex_unlock(&iprune_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489}
490
491/*
492 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
493 * "unused" means that no dentries are referring to the inodes: the files are
494 * not open and the dcache references to those inodes have already been
495 * reclaimed.
496 *
497 * This function is passed the number of inodes to scan, and it returns the
498 * total number of remaining possibly-reclaimable inodes.
499 */
Al Viro27496a82005-10-21 03:20:48 -0400500static int shrink_icache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501{
502 if (nr) {
503 /*
504 * Nasty deadlock avoidance. We may hold various FS locks,
505 * and we don't want to recurse into the FS that called us
506 * in clear_inode() and friends..
507 */
508 if (!(gfp_mask & __GFP_FS))
509 return -1;
510 prune_icache(nr);
511 }
512 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
513}
514
Rusty Russell8e1f9362007-07-17 04:03:17 -0700515static struct shrinker icache_shrinker = {
516 .shrink = shrink_icache_memory,
517 .seeks = DEFAULT_SEEKS,
518};
519
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520static void __wait_on_freeing_inode(struct inode *inode);
521/*
522 * Called with the inode lock held.
523 * NOTE: we are not increasing the inode-refcount, you must call __iget()
524 * by hand after calling find_inode now! This simplifies iunique and won't
525 * add any additional branch in the common code.
526 */
527static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
528{
529 struct hlist_node *node;
530 struct inode * inode = NULL;
531
532repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700533 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 if (inode->i_sb != sb)
535 continue;
536 if (!test(inode, data))
537 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700538 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 __wait_on_freeing_inode(inode);
540 goto repeat;
541 }
542 break;
543 }
544 return node ? inode : NULL;
545}
546
547/*
548 * find_inode_fast is the fast path version of find_inode, see the comment at
549 * iget_locked for details.
550 */
551static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
552{
553 struct hlist_node *node;
554 struct inode * inode = NULL;
555
556repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700557 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 if (inode->i_ino != ino)
559 continue;
560 if (inode->i_sb != sb)
561 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700562 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 __wait_on_freeing_inode(inode);
564 goto repeat;
565 }
566 break;
567 }
568 return node ? inode : NULL;
569}
570
David Chinner8290c352008-10-30 17:35:24 +1100571static unsigned long hash(struct super_block *sb, unsigned long hashval)
572{
573 unsigned long tmp;
574
575 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
576 L1_CACHE_BYTES;
577 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
578 return tmp & I_HASHMASK;
579}
580
581static inline void
582__inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
583 struct inode *inode)
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 if (head)
589 hlist_add_head(&inode->i_hash, head);
590}
591
592/**
593 * inode_add_to_lists - add a new inode to relevant lists
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800594 * @sb: superblock inode belongs to
595 * @inode: inode to mark in use
David Chinner8290c352008-10-30 17:35:24 +1100596 *
597 * When an inode is allocated it needs to be accounted for, added to the in use
598 * list, the owning superblock and the inode hash. This needs to be done under
599 * the inode_lock, so export a function to do this rather than the inode lock
600 * itself. We calculate the hash list to add to here so it is all internal
601 * which requires the caller to have already set up the inode number in the
602 * inode to add.
603 */
604void inode_add_to_lists(struct super_block *sb, struct inode *inode)
605{
606 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
607
608 spin_lock(&inode_lock);
609 __inode_add_to_lists(sb, head, inode);
610 spin_unlock(&inode_lock);
611}
612EXPORT_SYMBOL_GPL(inode_add_to_lists);
613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614/**
615 * new_inode - obtain an inode
616 * @sb: superblock
617 *
Mel Gorman769848c2007-07-17 04:03:05 -0700618 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800619 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700620 * If HIGHMEM pages are unsuitable or it is known that pages allocated
621 * for the page cache are not reclaimable or migratable,
622 * mapping_set_gfp_mask() must be called with suitable flags on the
623 * newly created inode's mapping
624 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 */
626struct inode *new_inode(struct super_block *sb)
627{
Jeff Layton866b04f2007-05-08 00:32:29 -0700628 /*
629 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
630 * error if st_ino won't fit in target struct field. Use 32bit counter
631 * here to attempt to avoid that.
632 */
633 static unsigned int last_ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 struct inode * inode;
635
636 spin_lock_prefetch(&inode_lock);
637
638 inode = alloc_inode(sb);
639 if (inode) {
640 spin_lock(&inode_lock);
David Chinner8290c352008-10-30 17:35:24 +1100641 __inode_add_to_lists(sb, NULL, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 inode->i_ino = ++last_ino;
643 inode->i_state = 0;
644 spin_unlock(&inode_lock);
645 }
646 return inode;
647}
648
649EXPORT_SYMBOL(new_inode);
650
651void unlock_new_inode(struct inode *inode)
652{
Peter Zijlstra14358e62007-10-14 01:38:33 +0200653#ifdef CONFIG_DEBUG_LOCK_ALLOC
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200654 if (inode->i_mode & S_IFDIR) {
655 struct file_system_type *type = inode->i_sb->s_type;
656
657 /*
658 * ensure nobody is actually holding i_mutex
659 */
660 mutex_destroy(&inode->i_mutex);
661 mutex_init(&inode->i_mutex);
Peter Zijlstra14358e62007-10-14 01:38:33 +0200662 lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key);
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200663 }
Peter Zijlstra14358e62007-10-14 01:38:33 +0200664#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 /*
666 * This is special! We do not need the spinlock
667 * when clearing I_LOCK, because we're guaranteed
668 * that nobody else tries to do anything about the
669 * state of the inode when it is locked, as we
670 * just created it (so there can be no old holders
671 * that haven't tested I_LOCK).
672 */
Nick Piggin7ef0d732009-03-12 14:31:38 -0700673 WARN_ON((inode->i_state & (I_LOCK|I_NEW)) != (I_LOCK|I_NEW));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 inode->i_state &= ~(I_LOCK|I_NEW);
675 wake_up_inode(inode);
676}
677
678EXPORT_SYMBOL(unlock_new_inode);
679
680/*
681 * This is called without the inode lock held.. Be careful.
682 *
683 * We no longer cache the sb_flags in i_flags - see fs.h
684 * -- rmk@arm.uk.linux.org
685 */
686static 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)
687{
688 struct inode * inode;
689
690 inode = alloc_inode(sb);
691 if (inode) {
692 struct inode * old;
693
694 spin_lock(&inode_lock);
695 /* We released the lock, so.. */
696 old = find_inode(sb, head, test, data);
697 if (!old) {
698 if (set(inode, data))
699 goto set_failed;
700
David Chinner8290c352008-10-30 17:35:24 +1100701 __inode_add_to_lists(sb, head, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 inode->i_state = I_LOCK|I_NEW;
703 spin_unlock(&inode_lock);
704
705 /* Return the locked inode with I_NEW set, the
706 * caller is responsible for filling in the contents
707 */
708 return inode;
709 }
710
711 /*
712 * Uhhuh, somebody else created the same inode under
713 * us. Use the old inode instead of the one we just
714 * allocated.
715 */
716 __iget(old);
717 spin_unlock(&inode_lock);
718 destroy_inode(inode);
719 inode = old;
720 wait_on_inode(inode);
721 }
722 return inode;
723
724set_failed:
725 spin_unlock(&inode_lock);
726 destroy_inode(inode);
727 return NULL;
728}
729
730/*
731 * get_new_inode_fast is the fast path version of get_new_inode, see the
732 * comment at iget_locked for details.
733 */
734static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
735{
736 struct inode * inode;
737
738 inode = alloc_inode(sb);
739 if (inode) {
740 struct inode * old;
741
742 spin_lock(&inode_lock);
743 /* We released the lock, so.. */
744 old = find_inode_fast(sb, head, ino);
745 if (!old) {
746 inode->i_ino = ino;
David Chinner8290c352008-10-30 17:35:24 +1100747 __inode_add_to_lists(sb, head, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 inode->i_state = I_LOCK|I_NEW;
749 spin_unlock(&inode_lock);
750
751 /* Return the locked inode with I_NEW set, the
752 * caller is responsible for filling in the contents
753 */
754 return inode;
755 }
756
757 /*
758 * Uhhuh, somebody else created the same inode under
759 * us. Use the old inode instead of the one we just
760 * allocated.
761 */
762 __iget(old);
763 spin_unlock(&inode_lock);
764 destroy_inode(inode);
765 inode = old;
766 wait_on_inode(inode);
767 }
768 return inode;
769}
770
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771/**
772 * iunique - get a unique inode number
773 * @sb: superblock
774 * @max_reserved: highest reserved inode number
775 *
776 * Obtain an inode number that is unique on the system for a given
777 * superblock. This is used by file systems that have no natural
778 * permanent inode numbering system. An inode number is returned that
779 * is higher than the reserved limit but unique.
780 *
781 * BUGS:
782 * With a large number of inodes live on the file system this function
783 * currently becomes quite slow.
784 */
785ino_t iunique(struct super_block *sb, ino_t max_reserved)
786{
Jeff Layton866b04f2007-05-08 00:32:29 -0700787 /*
788 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
789 * error if st_ino won't fit in target struct field. Use 32bit counter
790 * here to attempt to avoid that.
791 */
792 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 struct inode *inode;
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700794 struct hlist_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700797 spin_lock(&inode_lock);
798 do {
799 if (counter <= max_reserved)
800 counter = max_reserved + 1;
801 res = counter++;
802 head = inode_hashtable + hash(sb, res);
803 inode = find_inode_fast(sb, head, res);
804 } while (inode != NULL);
805 spin_unlock(&inode_lock);
806
807 return res;
808}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809EXPORT_SYMBOL(iunique);
810
811struct inode *igrab(struct inode *inode)
812{
813 spin_lock(&inode_lock);
Jan Blunck4a3b0a42007-02-10 01:44:59 -0800814 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 __iget(inode);
816 else
817 /*
818 * Handle the case where s_op->clear_inode is not been
819 * called yet, and somebody is calling igrab
820 * while the inode is getting freed.
821 */
822 inode = NULL;
823 spin_unlock(&inode_lock);
824 return inode;
825}
826
827EXPORT_SYMBOL(igrab);
828
829/**
830 * ifind - internal function, you want ilookup5() or iget5().
831 * @sb: super block of file system to search
832 * @head: the head of the list to search
833 * @test: callback used for comparisons between inodes
834 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700835 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 *
837 * ifind() searches for the inode specified by @data in the inode
838 * cache. This is a generalized version of ifind_fast() for file systems where
839 * the inode number is not sufficient for unique identification of an inode.
840 *
841 * If the inode is in the cache, the inode is returned with an incremented
842 * reference count.
843 *
844 * Otherwise NULL is returned.
845 *
846 * Note, @test is called with the inode_lock held, so can't sleep.
847 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800848static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700850 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851{
852 struct inode *inode;
853
854 spin_lock(&inode_lock);
855 inode = find_inode(sb, head, test, data);
856 if (inode) {
857 __iget(inode);
858 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700859 if (likely(wait))
860 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 return inode;
862 }
863 spin_unlock(&inode_lock);
864 return NULL;
865}
866
867/**
868 * ifind_fast - internal function, you want ilookup() or iget().
869 * @sb: super block of file system to search
870 * @head: head of the list to search
871 * @ino: inode number to search for
872 *
873 * ifind_fast() searches for the inode @ino in the inode cache. This is for
874 * file systems where the inode number is sufficient for unique identification
875 * of an inode.
876 *
877 * If the inode is in the cache, the inode is returned with an incremented
878 * reference count.
879 *
880 * Otherwise NULL is returned.
881 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800882static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 struct hlist_head *head, unsigned long ino)
884{
885 struct inode *inode;
886
887 spin_lock(&inode_lock);
888 inode = find_inode_fast(sb, head, ino);
889 if (inode) {
890 __iget(inode);
891 spin_unlock(&inode_lock);
892 wait_on_inode(inode);
893 return inode;
894 }
895 spin_unlock(&inode_lock);
896 return NULL;
897}
898
899/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700900 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 * @sb: super block of file system to search
902 * @hashval: hash value (usually inode number) to search for
903 * @test: callback used for comparisons between inodes
904 * @data: opaque data pointer to pass to @test
905 *
906 * ilookup5() uses ifind() to search for the inode specified by @hashval and
907 * @data in the inode cache. This is a generalized version of ilookup() for
908 * file systems where the inode number is not sufficient for unique
909 * identification of an inode.
910 *
911 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700912 * reference count. Note, the inode lock is not waited upon so you have to be
913 * very careful what you do with the returned inode. You probably should be
914 * using ilookup5() instead.
915 *
916 * Otherwise NULL is returned.
917 *
918 * Note, @test is called with the inode_lock held, so can't sleep.
919 */
920struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
921 int (*test)(struct inode *, void *), void *data)
922{
923 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
924
925 return ifind(sb, head, test, data, 0);
926}
927
928EXPORT_SYMBOL(ilookup5_nowait);
929
930/**
931 * ilookup5 - search for an inode in the inode cache
932 * @sb: super block of file system to search
933 * @hashval: hash value (usually inode number) to search for
934 * @test: callback used for comparisons between inodes
935 * @data: opaque data pointer to pass to @test
936 *
937 * ilookup5() uses ifind() to search for the inode specified by @hashval and
938 * @data in the inode cache. This is a generalized version of ilookup() for
939 * file systems where the inode number is not sufficient for unique
940 * identification of an inode.
941 *
942 * If the inode is in the cache, the inode lock is waited upon and the inode is
943 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 *
945 * Otherwise NULL is returned.
946 *
947 * Note, @test is called with the inode_lock held, so can't sleep.
948 */
949struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
950 int (*test)(struct inode *, void *), void *data)
951{
952 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
953
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700954 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955}
956
957EXPORT_SYMBOL(ilookup5);
958
959/**
960 * ilookup - search for an inode in the inode cache
961 * @sb: super block of file system to search
962 * @ino: inode number to search for
963 *
964 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
965 * This is for file systems where the inode number is sufficient for unique
966 * identification of an inode.
967 *
968 * If the inode is in the cache, the inode is returned with an incremented
969 * reference count.
970 *
971 * Otherwise NULL is returned.
972 */
973struct inode *ilookup(struct super_block *sb, unsigned long ino)
974{
975 struct hlist_head *head = inode_hashtable + hash(sb, ino);
976
977 return ifind_fast(sb, head, ino);
978}
979
980EXPORT_SYMBOL(ilookup);
981
982/**
983 * iget5_locked - obtain an inode from a mounted file system
984 * @sb: super block of file system
985 * @hashval: hash value (usually inode number) to get
986 * @test: callback used for comparisons between inodes
987 * @set: callback used to initialize a new struct inode
988 * @data: opaque data pointer to pass to @test and @set
989 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 * iget5_locked() uses ifind() to search for the inode specified by @hashval
991 * and @data in the inode cache and if present it is returned with an increased
992 * reference count. This is a generalized version of iget_locked() for file
993 * systems where the inode number is not sufficient for unique identification
994 * of an inode.
995 *
996 * If the inode is not in cache, get_new_inode() is called to allocate a new
997 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
998 * file system gets to fill it in before unlocking it via unlock_new_inode().
999 *
1000 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1001 */
1002struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1003 int (*test)(struct inode *, void *),
1004 int (*set)(struct inode *, void *), void *data)
1005{
1006 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1007 struct inode *inode;
1008
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001009 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 if (inode)
1011 return inode;
1012 /*
1013 * get_new_inode() will do the right thing, re-trying the search
1014 * in case it had to block at any point.
1015 */
1016 return get_new_inode(sb, head, test, set, data);
1017}
1018
1019EXPORT_SYMBOL(iget5_locked);
1020
1021/**
1022 * iget_locked - obtain an inode from a mounted file system
1023 * @sb: super block of file system
1024 * @ino: inode number to get
1025 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1027 * the inode cache and if present it is returned with an increased reference
1028 * count. This is for file systems where the inode number is sufficient for
1029 * unique identification of an inode.
1030 *
1031 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1032 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1033 * The file system gets to fill it in before unlocking it via
1034 * unlock_new_inode().
1035 */
1036struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1037{
1038 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1039 struct inode *inode;
1040
1041 inode = ifind_fast(sb, head, ino);
1042 if (inode)
1043 return inode;
1044 /*
1045 * get_new_inode_fast() will do the right thing, re-trying the search
1046 * in case it had to block at any point.
1047 */
1048 return get_new_inode_fast(sb, head, ino);
1049}
1050
1051EXPORT_SYMBOL(iget_locked);
1052
Al Viro261bca82008-12-30 01:48:21 -05001053int insert_inode_locked(struct inode *inode)
1054{
1055 struct super_block *sb = inode->i_sb;
1056 ino_t ino = inode->i_ino;
1057 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1058 struct inode *old;
1059
1060 inode->i_state |= I_LOCK|I_NEW;
1061 while (1) {
1062 spin_lock(&inode_lock);
1063 old = find_inode_fast(sb, head, ino);
1064 if (likely(!old)) {
1065 hlist_add_head(&inode->i_hash, head);
1066 spin_unlock(&inode_lock);
1067 return 0;
1068 }
1069 __iget(old);
1070 spin_unlock(&inode_lock);
1071 wait_on_inode(old);
1072 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1073 iput(old);
1074 return -EBUSY;
1075 }
1076 iput(old);
1077 }
1078}
1079
1080EXPORT_SYMBOL(insert_inode_locked);
1081
1082int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1083 int (*test)(struct inode *, void *), void *data)
1084{
1085 struct super_block *sb = inode->i_sb;
1086 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1087 struct inode *old;
1088
1089 inode->i_state |= I_LOCK|I_NEW;
1090
1091 while (1) {
1092 spin_lock(&inode_lock);
1093 old = find_inode(sb, head, test, data);
1094 if (likely(!old)) {
1095 hlist_add_head(&inode->i_hash, head);
1096 spin_unlock(&inode_lock);
1097 return 0;
1098 }
1099 __iget(old);
1100 spin_unlock(&inode_lock);
1101 wait_on_inode(old);
1102 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1103 iput(old);
1104 return -EBUSY;
1105 }
1106 iput(old);
1107 }
1108}
1109
1110EXPORT_SYMBOL(insert_inode_locked4);
1111
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112/**
1113 * __insert_inode_hash - hash an inode
1114 * @inode: unhashed inode
1115 * @hashval: unsigned long value used to locate this object in the
1116 * inode_hashtable.
1117 *
1118 * Add an inode to the inode hash for this superblock.
1119 */
1120void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1121{
1122 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1123 spin_lock(&inode_lock);
1124 hlist_add_head(&inode->i_hash, head);
1125 spin_unlock(&inode_lock);
1126}
1127
1128EXPORT_SYMBOL(__insert_inode_hash);
1129
1130/**
1131 * remove_inode_hash - remove an inode from the hash
1132 * @inode: inode to unhash
1133 *
1134 * Remove an inode from the superblock.
1135 */
1136void remove_inode_hash(struct inode *inode)
1137{
1138 spin_lock(&inode_lock);
1139 hlist_del_init(&inode->i_hash);
1140 spin_unlock(&inode_lock);
1141}
1142
1143EXPORT_SYMBOL(remove_inode_hash);
1144
1145/*
1146 * Tell the filesystem that this inode is no longer of any interest and should
1147 * be completely destroyed.
1148 *
1149 * We leave the inode in the inode hash table until *after* the filesystem's
1150 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1151 * instigate) will always find up-to-date information either in the hash or on
1152 * disk.
1153 *
1154 * I_FREEING is set so that no-one will take a new reference to the inode while
1155 * it is being deleted.
1156 */
Arjan van de Venb32714b2009-01-09 07:04:15 -08001157void generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001159 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
Arjan van de Venb32714b2009-01-09 07:04:15 -08001161 list_del_init(&inode->i_list);
1162 list_del_init(&inode->i_sb_list);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001163 WARN_ON(inode->i_state & I_NEW);
Arjan van de Venb32714b2009-01-09 07:04:15 -08001164 inode->i_state |= I_FREEING;
1165 inodes_stat.nr_inodes--;
1166 spin_unlock(&inode_lock);
1167
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 security_inode_delete(inode);
1169
1170 if (op->delete_inode) {
1171 void (*delete)(struct inode *) = op->delete_inode;
1172 if (!is_bad_inode(inode))
Jan Kara9e3509e2009-01-26 16:45:12 +01001173 vfs_dq_init(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001174 /* Filesystems implementing their own
1175 * s_op->delete_inode are required to call
1176 * truncate_inode_pages and clear_inode()
1177 * internally */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 delete(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001179 } else {
1180 truncate_inode_pages(&inode->i_data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 clear_inode(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001182 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 spin_lock(&inode_lock);
1184 hlist_del_init(&inode->i_hash);
1185 spin_unlock(&inode_lock);
1186 wake_up_inode(inode);
Eric Sesterhennb7542f82006-04-02 13:38:18 +02001187 BUG_ON(inode->i_state != I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 destroy_inode(inode);
1189}
1190
1191EXPORT_SYMBOL(generic_delete_inode);
1192
1193static void generic_forget_inode(struct inode *inode)
1194{
1195 struct super_block *sb = inode->i_sb;
1196
1197 if (!hlist_unhashed(&inode->i_hash)) {
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001198 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 list_move(&inode->i_list, &inode_unused);
1200 inodes_stat.nr_unused++;
Christoph Hellwigacb0c852007-05-08 00:25:52 -07001201 if (sb->s_flags & MS_ACTIVE) {
Alexander Viro991114c2005-06-23 00:09:01 -07001202 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001204 }
Nick Piggin7ef0d732009-03-12 14:31:38 -07001205 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001206 inode->i_state |= I_WILL_FREE;
1207 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 write_inode_now(inode, 1);
1209 spin_lock(&inode_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001210 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001211 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 inodes_stat.nr_unused--;
1213 hlist_del_init(&inode->i_hash);
1214 }
1215 list_del_init(&inode->i_list);
1216 list_del_init(&inode->i_sb_list);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001217 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001218 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 inodes_stat.nr_inodes--;
1220 spin_unlock(&inode_lock);
1221 if (inode->i_data.nrpages)
1222 truncate_inode_pages(&inode->i_data, 0);
1223 clear_inode(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001224 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 destroy_inode(inode);
1226}
1227
1228/*
1229 * Normal UNIX filesystem behaviour: delete the
1230 * inode when the usage count drops to zero, and
1231 * i_nlink is zero.
1232 */
Mark Fashehcb2c0232005-07-07 17:56:03 -07001233void generic_drop_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234{
1235 if (!inode->i_nlink)
1236 generic_delete_inode(inode);
1237 else
1238 generic_forget_inode(inode);
1239}
1240
Mark Fashehcb2c0232005-07-07 17:56:03 -07001241EXPORT_SYMBOL_GPL(generic_drop_inode);
1242
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243/*
1244 * Called when we're dropping the last reference
1245 * to an inode.
1246 *
1247 * Call the FS "drop()" function, defaulting to
1248 * the legacy UNIX filesystem behaviour..
1249 *
1250 * NOTE! NOTE! NOTE! We're called with the inode lock
1251 * held, and the drop function is supposed to release
1252 * the lock!
1253 */
1254static inline void iput_final(struct inode *inode)
1255{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001256 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 void (*drop)(struct inode *) = generic_drop_inode;
1258
1259 if (op && op->drop_inode)
1260 drop = op->drop_inode;
1261 drop(inode);
1262}
1263
1264/**
1265 * iput - put an inode
1266 * @inode: inode to put
1267 *
1268 * Puts an inode, dropping its usage count. If the inode use count hits
1269 * zero, the inode is then freed and may also be destroyed.
1270 *
1271 * Consequently, iput() can sleep.
1272 */
1273void iput(struct inode *inode)
1274{
1275 if (inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 BUG_ON(inode->i_state == I_CLEAR);
1277
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1279 iput_final(inode);
1280 }
1281}
1282
1283EXPORT_SYMBOL(iput);
1284
1285/**
1286 * bmap - find a block number in a file
1287 * @inode: inode of file
1288 * @block: block to find
1289 *
1290 * Returns the block number on the device holding the inode that
1291 * is the disk block number for the block of the file requested.
1292 * That is, asked for block 4 of inode 1 the function will return the
1293 * disk block relative to the disk start that holds that block of the
1294 * file.
1295 */
1296sector_t bmap(struct inode * inode, sector_t block)
1297{
1298 sector_t res = 0;
1299 if (inode->i_mapping->a_ops->bmap)
1300 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1301 return res;
1302}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303EXPORT_SYMBOL(bmap);
1304
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001305/*
1306 * With relative atime, only update atime if the previous atime is
1307 * earlier than either the ctime or mtime or if at least a day has
1308 * passed since the last atime update.
1309 */
1310static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1311 struct timespec now)
1312{
1313
1314 if (!(mnt->mnt_flags & MNT_RELATIME))
1315 return 1;
1316 /*
1317 * Is mtime younger than atime? If yes, update atime:
1318 */
1319 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1320 return 1;
1321 /*
1322 * Is ctime younger than atime? If yes, update atime:
1323 */
1324 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1325 return 1;
1326
1327 /*
1328 * Is the previous atime value older than a day? If yes,
1329 * update atime:
1330 */
1331 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1332 return 1;
1333 /*
1334 * Good, we can skip the atime update:
1335 */
1336 return 0;
1337}
1338
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001340 * touch_atime - update the access time
1341 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001342 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 *
1344 * Update the accessed time on an inode and mark it for writeback.
1345 * This function automatically handles read only file systems and media,
1346 * as well as the "noatime" flag and inode specific "noatime" markers.
1347 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001348void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001350 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 struct timespec now;
1352
Dave Hansencdb70f32008-02-15 14:37:41 -08001353 if (mnt_want_write(mnt))
1354 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001355 if (inode->i_flags & S_NOATIME)
Dave Hansencdb70f32008-02-15 14:37:41 -08001356 goto out;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001357 if (IS_NOATIME(inode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001358 goto out;
Andrew Mortonb2276132006-12-13 00:34:33 -08001359 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001360 goto out;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001361
Dave Hansencdb70f32008-02-15 14:37:41 -08001362 if (mnt->mnt_flags & MNT_NOATIME)
1363 goto out;
1364 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1365 goto out;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001366
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 now = current_fs_time(inode->i_sb);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001368
1369 if (!relatime_need_update(mnt, inode, now))
1370 goto out;
1371
Valerie Henson47ae32d2006-12-13 00:34:34 -08001372 if (timespec_equal(&inode->i_atime, &now))
Dave Hansencdb70f32008-02-15 14:37:41 -08001373 goto out;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001374
1375 inode->i_atime = now;
1376 mark_inode_dirty_sync(inode);
Dave Hansencdb70f32008-02-15 14:37:41 -08001377out:
1378 mnt_drop_write(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001380EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
1382/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001383 * file_update_time - update mtime and ctime time
1384 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001386 * Update the mtime and ctime members of an inode and mark the inode
1387 * for writeback. Note that this function is meant exclusively for
1388 * usage in the file write path of filesystems, and filesystems may
1389 * choose to explicitly ignore update via this function with the
1390 * S_NOCTIME inode flag, e.g. for network filesystem where these
1391 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 */
1393
Christoph Hellwig870f4812006-01-09 20:52:01 -08001394void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001396 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 struct timespec now;
1398 int sync_it = 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001399 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400
1401 if (IS_NOCMTIME(inode))
1402 return;
Dave Hansen20ddee22008-02-15 14:37:43 -08001403
1404 err = mnt_want_write(file->f_path.mnt);
1405 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 return;
1407
1408 now = current_fs_time(inode->i_sb);
Andreas Mohred97bd32006-10-02 02:17:17 -07001409 if (!timespec_equal(&inode->i_mtime, &now)) {
1410 inode->i_mtime = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001412 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
Andreas Mohred97bd32006-10-02 02:17:17 -07001414 if (!timespec_equal(&inode->i_ctime, &now)) {
1415 inode->i_ctime = now;
Christoph Hellwig870f4812006-01-09 20:52:01 -08001416 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001417 }
Christoph Hellwig870f4812006-01-09 20:52:01 -08001418
Jean Noel Cordenner7a224222008-01-28 23:58:27 -05001419 if (IS_I_VERSION(inode)) {
1420 inode_inc_iversion(inode);
1421 sync_it = 1;
1422 }
1423
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 if (sync_it)
1425 mark_inode_dirty_sync(inode);
Dave Hansen20ddee22008-02-15 14:37:43 -08001426 mnt_drop_write(file->f_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427}
1428
Christoph Hellwig870f4812006-01-09 20:52:01 -08001429EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
1431int inode_needs_sync(struct inode *inode)
1432{
1433 if (IS_SYNC(inode))
1434 return 1;
1435 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1436 return 1;
1437 return 0;
1438}
1439
1440EXPORT_SYMBOL(inode_needs_sync);
1441
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442int inode_wait(void *word)
1443{
1444 schedule();
1445 return 0;
1446}
Stephen Rothwelld44dab82008-11-10 17:06:05 +11001447EXPORT_SYMBOL(inode_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
1449/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001450 * If we try to find an inode in the inode hash while it is being
1451 * deleted, we have to wait until the filesystem completes its
1452 * deletion before reporting that it isn't found. This function waits
1453 * until the deletion _might_ have completed. Callers are responsible
1454 * to recheck inode state.
1455 *
1456 * It doesn't matter if I_LOCK is not set initially, a call to
1457 * wake_up_inode() after removing from the hash list will DTRT.
1458 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 * This is called with inode_lock held.
1460 */
1461static void __wait_on_freeing_inode(struct inode *inode)
1462{
1463 wait_queue_head_t *wq;
1464 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1466 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1467 spin_unlock(&inode_lock);
1468 schedule();
1469 finish_wait(wq, &wait.wait);
1470 spin_lock(&inode_lock);
1471}
1472
Mark Fasheh62752ee2006-10-17 10:31:38 +02001473/*
1474 * We rarely want to lock two inodes that do not have a parent/child
1475 * relationship (such as directory, child inode) simultaneously. The
1476 * vast majority of file systems should be able to get along fine
1477 * without this. Do not use these functions except as a last resort.
1478 */
1479void inode_double_lock(struct inode *inode1, struct inode *inode2)
1480{
1481 if (inode1 == NULL || inode2 == NULL || inode1 == inode2) {
1482 if (inode1)
1483 mutex_lock(&inode1->i_mutex);
1484 else if (inode2)
1485 mutex_lock(&inode2->i_mutex);
1486 return;
1487 }
1488
1489 if (inode1 < inode2) {
1490 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1491 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1492 } else {
1493 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1494 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1495 }
1496}
1497EXPORT_SYMBOL(inode_double_lock);
1498
1499void inode_double_unlock(struct inode *inode1, struct inode *inode2)
1500{
1501 if (inode1)
1502 mutex_unlock(&inode1->i_mutex);
1503
1504 if (inode2 && inode2 != inode1)
1505 mutex_unlock(&inode2->i_mutex);
1506}
1507EXPORT_SYMBOL(inode_double_unlock);
1508
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509static __initdata unsigned long ihash_entries;
1510static int __init set_ihash_entries(char *str)
1511{
1512 if (!str)
1513 return 0;
1514 ihash_entries = simple_strtoul(str, &str, 0);
1515 return 1;
1516}
1517__setup("ihash_entries=", set_ihash_entries);
1518
1519/*
1520 * Initialize the waitqueues and inode hash table.
1521 */
1522void __init inode_init_early(void)
1523{
1524 int loop;
1525
1526 /* If hashes are distributed across NUMA nodes, defer
1527 * hash allocation until vmalloc space is available.
1528 */
1529 if (hashdist)
1530 return;
1531
1532 inode_hashtable =
1533 alloc_large_system_hash("Inode-cache",
1534 sizeof(struct hlist_head),
1535 ihash_entries,
1536 14,
1537 HASH_EARLY,
1538 &i_hash_shift,
1539 &i_hash_mask,
1540 0);
1541
1542 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1543 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1544}
1545
Denis Cheng74bf17c2007-10-16 23:26:30 -07001546void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547{
1548 int loop;
1549
1550 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001551 inode_cachep = kmem_cache_create("inode_cache",
1552 sizeof(struct inode),
1553 0,
1554 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1555 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001556 init_once);
Rusty Russell8e1f9362007-07-17 04:03:17 -07001557 register_shrinker(&icache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558
1559 /* Hash may have been set up in inode_init_early */
1560 if (!hashdist)
1561 return;
1562
1563 inode_hashtable =
1564 alloc_large_system_hash("Inode-cache",
1565 sizeof(struct hlist_head),
1566 ihash_entries,
1567 14,
1568 0,
1569 &i_hash_shift,
1570 &i_hash_mask,
1571 0);
1572
1573 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1574 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1575}
1576
1577void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1578{
1579 inode->i_mode = mode;
1580 if (S_ISCHR(mode)) {
1581 inode->i_fop = &def_chr_fops;
1582 inode->i_rdev = rdev;
1583 } else if (S_ISBLK(mode)) {
1584 inode->i_fop = &def_blk_fops;
1585 inode->i_rdev = rdev;
1586 } else if (S_ISFIFO(mode))
1587 inode->i_fop = &def_fifo_fops;
1588 else if (S_ISSOCK(mode))
1589 inode->i_fop = &bad_sock_fops;
1590 else
1591 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1592 mode);
1593}
1594EXPORT_SYMBOL(init_special_inode);