blob: 76582b06ab975d76afcba7f5800d6c8e3873ebc6 [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>
Nick Piggin88e0fbc2009-09-22 16:43:50 -070017#include <linux/rwsem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/hash.h>
19#include <linux/swap.h>
20#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050021#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/pagemap.h>
23#include <linux/cdev.h>
24#include <linux/bootmem.h>
Robert Love0eeca282005-07-12 17:06:03 -040025#include <linux/inotify.h>
Eric Paris3be25f42009-05-21 17:01:26 -040026#include <linux/fsnotify.h>
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -080027#include <linux/mount.h>
Arjan van de Venefaee192009-01-06 07:20:54 -080028#include <linux/async.h>
Al Virof19d4a82009-06-08 19:50:45 -040029#include <linux/posix_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31/*
32 * This is needed for the following functions:
33 * - inode_has_buffers
34 * - invalidate_inode_buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -070035 * - invalidate_bdev
36 *
37 * FIXME: remove all knowledge of the buffer layer from this file
38 */
39#include <linux/buffer_head.h>
40
41/*
42 * New inode.c implementation.
43 *
44 * This implementation has the basic premise of trying
45 * to be extremely low-overhead and SMP-safe, yet be
46 * simple enough to be "obviously correct".
47 *
48 * Famous last words.
49 */
50
51/* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
52
53/* #define INODE_PARANOIA 1 */
54/* #define INODE_DEBUG 1 */
55
56/*
57 * Inode lookup is no longer as critical as it used to be:
58 * most of the lookups are going to be through the dcache.
59 */
60#define I_HASHBITS i_hash_shift
61#define I_HASHMASK i_hash_mask
62
Eric Dumazetfa3536c2006-03-26 01:37:24 -080063static unsigned int i_hash_mask __read_mostly;
64static unsigned int i_hash_shift __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
66/*
67 * Each inode can be on two separate lists. One is
68 * the hash list of the inode, used for lookups. The
69 * other linked list is the "type" list:
70 * "in_use" - valid inode, i_count > 0, i_nlink > 0
71 * "dirty" - as "in_use" but also dirty
72 * "unused" - valid inode, i_count = 0
73 *
74 * A "dirty" list is maintained for each super block,
75 * allowing for low-overhead inode sync() operations.
76 */
77
78LIST_HEAD(inode_in_use);
79LIST_HEAD(inode_unused);
Eric Dumazetfa3536c2006-03-26 01:37:24 -080080static struct hlist_head *inode_hashtable __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
82/*
83 * A simple spinlock to protect the list manipulations.
84 *
85 * NOTE! You also have to own the lock if you change
86 * the i_state of an inode while it is in use..
87 */
88DEFINE_SPINLOCK(inode_lock);
89
90/*
Nick Piggin88e0fbc2009-09-22 16:43:50 -070091 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 * icache shrinking path, and the umount path. Without this exclusion,
93 * by the time prune_icache calls iput for the inode whose pages it has
94 * been invalidating, or by the time it calls clear_inode & destroy_inode
95 * from its final dispose_list, the struct super_block they refer to
96 * (for inode->i_sb->s_op) may already have been freed and reused.
Nick Piggin88e0fbc2009-09-22 16:43:50 -070097 *
98 * We make this an rwsem because the fastpath is icache shrinking. In
99 * some cases a filesystem may be doing a significant amount of work in
100 * its inode reclaim code, so this should improve parallelism.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 */
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700102static DECLARE_RWSEM(iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
104/*
105 * Statistics gathering..
106 */
107struct inodes_stat_t inodes_stat;
108
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530109static struct kmem_cache *inode_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700111static void wake_up_inode(struct inode *inode)
112{
113 /*
114 * Prevent speculative execution through spin_unlock(&inode_lock);
115 */
116 smp_mb();
117 wake_up_bit(&inode->i_state, __I_LOCK);
118}
119
David Chinner2cb15992008-10-30 17:32:23 +1100120/**
121 * inode_init_always - perform inode structure intialisation
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800122 * @sb: superblock inode belongs to
123 * @inode: inode to initialise
David Chinner2cb15992008-10-30 17:32:23 +1100124 *
125 * These are initializations that need to be done on every inode
126 * allocation as the fields are not initialised by slab allocation.
127 */
Christoph Hellwig54e34622009-08-07 14:38:25 -0300128int inode_init_always(struct super_block *sb, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129{
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700130 static const struct address_space_operations empty_aops;
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700131 static const struct inode_operations empty_iops;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800132 static const struct file_operations empty_fops;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530133 struct address_space *const mapping = &inode->i_data;
David Chinner2cb15992008-10-30 17:32:23 +1100134
135 inode->i_sb = sb;
136 inode->i_blkbits = sb->s_blocksize_bits;
137 inode->i_flags = 0;
138 atomic_set(&inode->i_count, 1);
139 inode->i_op = &empty_iops;
140 inode->i_fop = &empty_fops;
141 inode->i_nlink = 1;
Al Viro56ff5ef2008-12-09 09:34:39 -0500142 inode->i_uid = 0;
143 inode->i_gid = 0;
David Chinner2cb15992008-10-30 17:32:23 +1100144 atomic_set(&inode->i_writecount, 0);
145 inode->i_size = 0;
146 inode->i_blocks = 0;
147 inode->i_bytes = 0;
148 inode->i_generation = 0;
149#ifdef CONFIG_QUOTA
150 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
151#endif
152 inode->i_pipe = NULL;
153 inode->i_bdev = NULL;
154 inode->i_cdev = NULL;
155 inode->i_rdev = 0;
156 inode->dirtied_when = 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500157
158 if (security_inode_alloc(inode))
Christoph Hellwig54e34622009-08-07 14:38:25 -0300159 goto out;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500160
161 /* allocate and initialize an i_integrity */
162 if (ima_inode_alloc(inode))
163 goto out_free_security;
David Chinner2cb15992008-10-30 17:32:23 +1100164
165 spin_lock_init(&inode->i_lock);
166 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
167
168 mutex_init(&inode->i_mutex);
169 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
170
171 init_rwsem(&inode->i_alloc_sem);
172 lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
173
174 mapping->a_ops = &empty_aops;
175 mapping->host = inode;
176 mapping->flags = 0;
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800177 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
David Chinner2cb15992008-10-30 17:32:23 +1100178 mapping->assoc_mapping = NULL;
179 mapping->backing_dev_info = &default_backing_dev_info;
180 mapping->writeback_index = 0;
181
182 /*
183 * If the block_device provides a backing_dev_info for client
184 * inodes then use that. Otherwise the inode share the bdev's
185 * backing_dev_info.
186 */
187 if (sb->s_bdev) {
188 struct backing_dev_info *bdi;
189
Jens Axboe2c96ce92009-09-15 09:43:56 +0200190 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
David Chinner2cb15992008-10-30 17:32:23 +1100191 mapping->backing_dev_info = bdi;
192 }
193 inode->i_private = NULL;
194 inode->i_mapping = mapping;
Al Virof19d4a82009-06-08 19:50:45 -0400195#ifdef CONFIG_FS_POSIX_ACL
196 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
197#endif
David Chinner2cb15992008-10-30 17:32:23 +1100198
Eric Paris3be25f42009-05-21 17:01:26 -0400199#ifdef CONFIG_FSNOTIFY
200 inode->i_fsnotify_mask = 0;
201#endif
202
Christoph Hellwig54e34622009-08-07 14:38:25 -0300203 return 0;
Mimi Zohar6146f0d2009-02-04 09:06:57 -0500204
205out_free_security:
206 security_inode_free(inode);
Christoph Hellwig54e34622009-08-07 14:38:25 -0300207out:
208 return -ENOMEM;
David Chinner2cb15992008-10-30 17:32:23 +1100209}
210EXPORT_SYMBOL(inode_init_always);
211
212static struct inode *alloc_inode(struct super_block *sb)
213{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 struct inode *inode;
215
216 if (sb->s_op->alloc_inode)
217 inode = sb->s_op->alloc_inode(sb);
218 else
David Chinner2cb15992008-10-30 17:32:23 +1100219 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
Christoph Hellwig54e34622009-08-07 14:38:25 -0300221 if (!inode)
222 return NULL;
223
224 if (unlikely(inode_init_always(sb, inode))) {
225 if (inode->i_sb->s_op->destroy_inode)
226 inode->i_sb->s_op->destroy_inode(inode);
227 else
228 kmem_cache_free(inode_cachep, inode);
229 return NULL;
230 }
231
232 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
234
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300235void __destroy_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236{
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200237 BUG_ON(inode_has_buffers(inode));
Hugh Dickinsf07502d2009-06-06 21:18:09 +0100238 ima_inode_free(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 security_inode_free(inode);
Eric Paris3be25f42009-05-21 17:01:26 -0400240 fsnotify_inode_delete(inode);
Al Virof19d4a82009-06-08 19:50:45 -0400241#ifdef CONFIG_FS_POSIX_ACL
242 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
243 posix_acl_release(inode->i_acl);
244 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
245 posix_acl_release(inode->i_default_acl);
246#endif
Christoph Hellwig2e00c972009-08-07 14:38:29 -0300247}
248EXPORT_SYMBOL(__destroy_inode);
249
250void destroy_inode(struct inode *inode)
251{
252 __destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 if (inode->i_sb->s_op->destroy_inode)
254 inode->i_sb->s_op->destroy_inode(inode);
255 else
256 kmem_cache_free(inode_cachep, (inode));
257}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
259/*
260 * These are initializations that only need to be done
261 * once, because the fields are idempotent across use
262 * of the inode, so let the slab aware of that.
263 */
264void inode_init_once(struct inode *inode)
265{
266 memset(inode, 0, sizeof(*inode));
267 INIT_HLIST_NODE(&inode->i_hash);
268 INIT_LIST_HEAD(&inode->i_dentry);
269 INIT_LIST_HEAD(&inode->i_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
Nick Piggin19fd6232008-07-25 19:45:32 -0700271 spin_lock_init(&inode->i_data.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 spin_lock_init(&inode->i_data.i_mmap_lock);
273 INIT_LIST_HEAD(&inode->i_data.private_list);
274 spin_lock_init(&inode->i_data.private_lock);
275 INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
276 INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 i_size_ordered_init(inode);
Robert Love0eeca282005-07-12 17:06:03 -0400278#ifdef CONFIG_INOTIFY
279 INIT_LIST_HEAD(&inode->inotify_watches);
Ingo Molnard4f9af92006-03-23 03:00:30 -0800280 mutex_init(&inode->inotify_mutex);
Robert Love0eeca282005-07-12 17:06:03 -0400281#endif
Eric Paris3be25f42009-05-21 17:01:26 -0400282#ifdef CONFIG_FSNOTIFY
283 INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries);
284#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286EXPORT_SYMBOL(inode_init_once);
287
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700288static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530290 struct inode *inode = (struct inode *) foo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291
Christoph Lametera35afb82007-05-16 22:10:57 -0700292 inode_init_once(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293}
294
295/*
296 * inode_lock must be held
297 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530298void __iget(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
300 if (atomic_read(&inode->i_count)) {
301 atomic_inc(&inode->i_count);
302 return;
303 }
304 atomic_inc(&inode->i_count);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700305 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 list_move(&inode->i_list, &inode_in_use);
307 inodes_stat.nr_unused--;
308}
309
310/**
311 * clear_inode - clear an inode
312 * @inode: inode to clear
313 *
314 * This is called by the filesystem to tell us
315 * that the inode is no longer useful. We just
316 * terminate it with extreme prejudice.
317 */
318void clear_inode(struct inode *inode)
319{
320 might_sleep();
321 invalidate_inode_buffers(inode);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530322
Eric Sesterhennb7542f82006-04-02 13:38:18 +0200323 BUG_ON(inode->i_data.nrpages);
324 BUG_ON(!(inode->i_state & I_FREEING));
325 BUG_ON(inode->i_state & I_CLEAR);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700326 inode_sync_wait(inode);
Jan Kara9e3509e2009-01-26 16:45:12 +0100327 vfs_dq_drop(inode);
Christoph Hellwigacb0c852007-05-08 00:25:52 -0700328 if (inode->i_sb->s_op->clear_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 inode->i_sb->s_op->clear_inode(inode);
Theodore Ts'oeaf796e2006-09-27 01:50:48 -0700330 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 bd_forget(inode);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700332 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 cd_forget(inode);
334 inode->i_state = I_CLEAR;
335}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336EXPORT_SYMBOL(clear_inode);
337
338/*
339 * dispose_list - dispose of the contents of a local list
340 * @head: the head of the list to free
341 *
342 * Dispose-list gets a local list with local inodes in it, so it doesn't
343 * need to worry about list corruption and SMP locks.
344 */
345static void dispose_list(struct list_head *head)
346{
347 int nr_disposed = 0;
348
349 while (!list_empty(head)) {
350 struct inode *inode;
351
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700352 inode = list_first_entry(head, struct inode, i_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 list_del(&inode->i_list);
354
355 if (inode->i_data.nrpages)
356 truncate_inode_pages(&inode->i_data, 0);
357 clear_inode(inode);
Artem B. Bityuckiy4120db42005-07-12 13:58:12 -0700358
359 spin_lock(&inode_lock);
360 hlist_del_init(&inode->i_hash);
361 list_del_init(&inode->i_sb_list);
362 spin_unlock(&inode_lock);
363
364 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 destroy_inode(inode);
366 nr_disposed++;
367 }
368 spin_lock(&inode_lock);
369 inodes_stat.nr_inodes -= nr_disposed;
370 spin_unlock(&inode_lock);
371}
372
373/*
374 * Invalidate all inodes for a device.
375 */
376static int invalidate_list(struct list_head *head, struct list_head *dispose)
377{
378 struct list_head *next;
379 int busy = 0, count = 0;
380
381 next = head->next;
382 for (;;) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530383 struct list_head *tmp = next;
384 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
386 /*
387 * We can reschedule here without worrying about the list's
388 * consistency because the per-sb list of inodes must not
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700389 * change during umount anymore, and because iprune_sem keeps
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 * shrink_icache_memory() away.
391 */
392 cond_resched_lock(&inode_lock);
393
394 next = next->next;
395 if (tmp == head)
396 break;
397 inode = list_entry(tmp, struct inode, i_sb_list);
Nick Pigginaabb8fd2009-03-11 13:17:36 -0700398 if (inode->i_state & I_NEW)
399 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 invalidate_inode_buffers(inode);
401 if (!atomic_read(&inode->i_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 list_move(&inode->i_list, dispose);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700403 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 inode->i_state |= I_FREEING;
405 count++;
406 continue;
407 }
408 busy = 1;
409 }
410 /* only unused inodes may be cached with i_count zero */
411 inodes_stat.nr_unused -= count;
412 return busy;
413}
414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415/**
416 * invalidate_inodes - discard the inodes on a device
417 * @sb: superblock
418 *
419 * Discard all of the inodes for a given superblock. If the discard
420 * fails because there are busy inodes then a non zero value is returned.
421 * If the discard is successful all the inodes have been discarded.
422 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530423int invalidate_inodes(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424{
425 int busy;
426 LIST_HEAD(throw_away);
427
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700428 down_write(&iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 spin_lock(&inode_lock);
Robert Love0eeca282005-07-12 17:06:03 -0400430 inotify_unmount_inodes(&sb->s_inodes);
Eric Paris164bc612009-05-21 17:01:58 -0400431 fsnotify_unmount_inodes(&sb->s_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 busy = invalidate_list(&sb->s_inodes, &throw_away);
433 spin_unlock(&inode_lock);
434
435 dispose_list(&throw_away);
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700436 up_write(&iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
438 return busy;
439}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440EXPORT_SYMBOL(invalidate_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
442static int can_unuse(struct inode *inode)
443{
444 if (inode->i_state)
445 return 0;
446 if (inode_has_buffers(inode))
447 return 0;
448 if (atomic_read(&inode->i_count))
449 return 0;
450 if (inode->i_data.nrpages)
451 return 0;
452 return 1;
453}
454
455/*
456 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
457 * a temporary list and then are freed outside inode_lock by dispose_list().
458 *
459 * Any inodes which are pinned purely because of attached pagecache have their
460 * pagecache removed. We expect the final iput() on that inode to add it to
461 * the front of the inode_unused list. So look for it there and if the
462 * inode is still freeable, proceed. The right inode is found 99.9% of the
463 * time in testing on a 4-way.
464 *
465 * If the inode has metadata buffers attached to mapping->private_list then
466 * try to remove them.
467 */
468static void prune_icache(int nr_to_scan)
469{
470 LIST_HEAD(freeable);
471 int nr_pruned = 0;
472 int nr_scanned;
473 unsigned long reap = 0;
474
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700475 down_read(&iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 spin_lock(&inode_lock);
477 for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
478 struct inode *inode;
479
480 if (list_empty(&inode_unused))
481 break;
482
483 inode = list_entry(inode_unused.prev, struct inode, i_list);
484
485 if (inode->i_state || atomic_read(&inode->i_count)) {
486 list_move(&inode->i_list, &inode_unused);
487 continue;
488 }
489 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
490 __iget(inode);
491 spin_unlock(&inode_lock);
492 if (remove_inode_buffers(inode))
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800493 reap += invalidate_mapping_pages(&inode->i_data,
494 0, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 iput(inode);
496 spin_lock(&inode_lock);
497
498 if (inode != list_entry(inode_unused.next,
499 struct inode, i_list))
500 continue; /* wrong inode or list_empty */
501 if (!can_unuse(inode))
502 continue;
503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 list_move(&inode->i_list, &freeable);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700505 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 inode->i_state |= I_FREEING;
507 nr_pruned++;
508 }
509 inodes_stat.nr_unused -= nr_pruned;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700510 if (current_is_kswapd())
511 __count_vm_events(KSWAPD_INODESTEAL, reap);
512 else
513 __count_vm_events(PGINODESTEAL, reap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 spin_unlock(&inode_lock);
515
516 dispose_list(&freeable);
Nick Piggin88e0fbc2009-09-22 16:43:50 -0700517 up_read(&iprune_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518}
519
520/*
521 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
522 * "unused" means that no dentries are referring to the inodes: the files are
523 * not open and the dcache references to those inodes have already been
524 * reclaimed.
525 *
526 * This function is passed the number of inodes to scan, and it returns the
527 * total number of remaining possibly-reclaimable inodes.
528 */
Al Viro27496a82005-10-21 03:20:48 -0400529static int shrink_icache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530{
531 if (nr) {
532 /*
533 * Nasty deadlock avoidance. We may hold various FS locks,
534 * and we don't want to recurse into the FS that called us
535 * in clear_inode() and friends..
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530536 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 if (!(gfp_mask & __GFP_FS))
538 return -1;
539 prune_icache(nr);
540 }
541 return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
542}
543
Rusty Russell8e1f9362007-07-17 04:03:17 -0700544static struct shrinker icache_shrinker = {
545 .shrink = shrink_icache_memory,
546 .seeks = DEFAULT_SEEKS,
547};
548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549static void __wait_on_freeing_inode(struct inode *inode);
550/*
551 * Called with the inode lock held.
552 * NOTE: we are not increasing the inode-refcount, you must call __iget()
553 * by hand after calling find_inode now! This simplifies iunique and won't
554 * add any additional branch in the common code.
555 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530556static struct inode *find_inode(struct super_block *sb,
557 struct hlist_head *head,
558 int (*test)(struct inode *, void *),
559 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
561 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530562 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
564repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700565 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 if (inode->i_sb != sb)
567 continue;
568 if (!test(inode, data))
569 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700570 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 __wait_on_freeing_inode(inode);
572 goto repeat;
573 }
574 break;
575 }
576 return node ? inode : NULL;
577}
578
579/*
580 * find_inode_fast is the fast path version of find_inode, see the comment at
581 * iget_locked for details.
582 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530583static struct inode *find_inode_fast(struct super_block *sb,
584 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585{
586 struct hlist_node *node;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530587 struct inode *inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
589repeat:
Matthias Kaehlckec5c8be32008-04-29 00:59:40 -0700590 hlist_for_each_entry(inode, node, head, i_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 if (inode->i_ino != ino)
592 continue;
593 if (inode->i_sb != sb)
594 continue;
Alexander Viro991114c2005-06-23 00:09:01 -0700595 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 __wait_on_freeing_inode(inode);
597 goto repeat;
598 }
599 break;
600 }
601 return node ? inode : NULL;
602}
603
David Chinner8290c352008-10-30 17:35:24 +1100604static unsigned long hash(struct super_block *sb, unsigned long hashval)
605{
606 unsigned long tmp;
607
608 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
609 L1_CACHE_BYTES;
610 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
611 return tmp & I_HASHMASK;
612}
613
614static inline void
615__inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
616 struct inode *inode)
617{
618 inodes_stat.nr_inodes++;
619 list_add(&inode->i_list, &inode_in_use);
620 list_add(&inode->i_sb_list, &sb->s_inodes);
621 if (head)
622 hlist_add_head(&inode->i_hash, head);
623}
624
625/**
626 * inode_add_to_lists - add a new inode to relevant lists
Randy Dunlap0bc02f32009-01-06 14:41:13 -0800627 * @sb: superblock inode belongs to
628 * @inode: inode to mark in use
David Chinner8290c352008-10-30 17:35:24 +1100629 *
630 * When an inode is allocated it needs to be accounted for, added to the in use
631 * list, the owning superblock and the inode hash. This needs to be done under
632 * the inode_lock, so export a function to do this rather than the inode lock
633 * itself. We calculate the hash list to add to here so it is all internal
634 * which requires the caller to have already set up the inode number in the
635 * inode to add.
636 */
637void inode_add_to_lists(struct super_block *sb, struct inode *inode)
638{
639 struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
640
641 spin_lock(&inode_lock);
642 __inode_add_to_lists(sb, head, inode);
643 spin_unlock(&inode_lock);
644}
645EXPORT_SYMBOL_GPL(inode_add_to_lists);
646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647/**
648 * new_inode - obtain an inode
649 * @sb: superblock
650 *
Mel Gorman769848c2007-07-17 04:03:05 -0700651 * Allocates a new inode for given superblock. The default gfp_mask
Hugh Dickins3c1d4372009-01-06 14:39:23 -0800652 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
Mel Gorman769848c2007-07-17 04:03:05 -0700653 * If HIGHMEM pages are unsuitable or it is known that pages allocated
654 * for the page cache are not reclaimable or migratable,
655 * mapping_set_gfp_mask() must be called with suitable flags on the
656 * newly created inode's mapping
657 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 */
659struct inode *new_inode(struct super_block *sb)
660{
Jeff Layton866b04f2007-05-08 00:32:29 -0700661 /*
662 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
663 * error if st_ino won't fit in target struct field. Use 32bit counter
664 * here to attempt to avoid that.
665 */
666 static unsigned int last_ino;
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530667 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
669 spin_lock_prefetch(&inode_lock);
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530670
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 inode = alloc_inode(sb);
672 if (inode) {
673 spin_lock(&inode_lock);
David Chinner8290c352008-10-30 17:35:24 +1100674 __inode_add_to_lists(sb, NULL, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 inode->i_ino = ++last_ino;
676 inode->i_state = 0;
677 spin_unlock(&inode_lock);
678 }
679 return inode;
680}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681EXPORT_SYMBOL(new_inode);
682
683void unlock_new_inode(struct inode *inode)
684{
Peter Zijlstra14358e62007-10-14 01:38:33 +0200685#ifdef CONFIG_DEBUG_LOCK_ALLOC
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200686 if (inode->i_mode & S_IFDIR) {
687 struct file_system_type *type = inode->i_sb->s_type;
688
Jan Kara9a7aa122009-06-04 15:26:49 +0200689 /* Set new key only if filesystem hasn't already changed it */
690 if (!lockdep_match_class(&inode->i_mutex,
691 &type->i_mutex_key)) {
692 /*
693 * ensure nobody is actually holding i_mutex
694 */
695 mutex_destroy(&inode->i_mutex);
696 mutex_init(&inode->i_mutex);
697 lockdep_set_class(&inode->i_mutex,
698 &type->i_mutex_dir_key);
699 }
Peter Zijlstra1e89a5e2007-10-16 06:47:54 +0200700 }
Peter Zijlstra14358e62007-10-14 01:38:33 +0200701#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 /*
Jan Kara580be082009-09-21 17:01:06 -0700703 * This is special! We do not need the spinlock when clearing I_LOCK,
704 * because we're guaranteed that nobody else tries to do anything about
705 * the state of the inode when it is locked, as we just created it (so
706 * there can be no old holders that haven't tested I_LOCK).
707 * However we must emit the memory barrier so that other CPUs reliably
708 * see the clearing of I_LOCK after the other inode initialisation has
709 * completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 */
Jan Kara580be082009-09-21 17:01:06 -0700711 smp_mb();
Nick Piggin7ef0d732009-03-12 14:31:38 -0700712 WARN_ON((inode->i_state & (I_LOCK|I_NEW)) != (I_LOCK|I_NEW));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 inode->i_state &= ~(I_LOCK|I_NEW);
714 wake_up_inode(inode);
715}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716EXPORT_SYMBOL(unlock_new_inode);
717
718/*
719 * This is called without the inode lock held.. Be careful.
720 *
721 * We no longer cache the sb_flags in i_flags - see fs.h
722 * -- rmk@arm.uk.linux.org
723 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530724static struct inode *get_new_inode(struct super_block *sb,
725 struct hlist_head *head,
726 int (*test)(struct inode *, void *),
727 int (*set)(struct inode *, void *),
728 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530730 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
732 inode = alloc_inode(sb);
733 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530734 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
736 spin_lock(&inode_lock);
737 /* We released the lock, so.. */
738 old = find_inode(sb, head, test, data);
739 if (!old) {
740 if (set(inode, data))
741 goto set_failed;
742
David Chinner8290c352008-10-30 17:35:24 +1100743 __inode_add_to_lists(sb, head, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 inode->i_state = I_LOCK|I_NEW;
745 spin_unlock(&inode_lock);
746
747 /* Return the locked inode with I_NEW set, the
748 * caller is responsible for filling in the contents
749 */
750 return inode;
751 }
752
753 /*
754 * Uhhuh, somebody else created the same inode under
755 * us. Use the old inode instead of the one we just
756 * allocated.
757 */
758 __iget(old);
759 spin_unlock(&inode_lock);
760 destroy_inode(inode);
761 inode = old;
762 wait_on_inode(inode);
763 }
764 return inode;
765
766set_failed:
767 spin_unlock(&inode_lock);
768 destroy_inode(inode);
769 return NULL;
770}
771
772/*
773 * get_new_inode_fast is the fast path version of get_new_inode, see the
774 * comment at iget_locked for details.
775 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530776static struct inode *get_new_inode_fast(struct super_block *sb,
777 struct hlist_head *head, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778{
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530779 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
781 inode = alloc_inode(sb);
782 if (inode) {
Manish Katiyar6b3304b2009-03-31 19:35:54 +0530783 struct inode *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
785 spin_lock(&inode_lock);
786 /* We released the lock, so.. */
787 old = find_inode_fast(sb, head, ino);
788 if (!old) {
789 inode->i_ino = ino;
David Chinner8290c352008-10-30 17:35:24 +1100790 __inode_add_to_lists(sb, head, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 inode->i_state = I_LOCK|I_NEW;
792 spin_unlock(&inode_lock);
793
794 /* Return the locked inode with I_NEW set, the
795 * caller is responsible for filling in the contents
796 */
797 return inode;
798 }
799
800 /*
801 * Uhhuh, somebody else created the same inode under
802 * us. Use the old inode instead of the one we just
803 * allocated.
804 */
805 __iget(old);
806 spin_unlock(&inode_lock);
807 destroy_inode(inode);
808 inode = old;
809 wait_on_inode(inode);
810 }
811 return inode;
812}
813
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814/**
815 * iunique - get a unique inode number
816 * @sb: superblock
817 * @max_reserved: highest reserved inode number
818 *
819 * Obtain an inode number that is unique on the system for a given
820 * superblock. This is used by file systems that have no natural
821 * permanent inode numbering system. An inode number is returned that
822 * is higher than the reserved limit but unique.
823 *
824 * BUGS:
825 * With a large number of inodes live on the file system this function
826 * currently becomes quite slow.
827 */
828ino_t iunique(struct super_block *sb, ino_t max_reserved)
829{
Jeff Layton866b04f2007-05-08 00:32:29 -0700830 /*
831 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
832 * error if st_ino won't fit in target struct field. Use 32bit counter
833 * here to attempt to avoid that.
834 */
835 static unsigned int counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 struct inode *inode;
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700837 struct hlist_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 ino_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
Jeffrey Layton3361c7b2007-05-08 00:29:48 -0700840 spin_lock(&inode_lock);
841 do {
842 if (counter <= max_reserved)
843 counter = max_reserved + 1;
844 res = counter++;
845 head = inode_hashtable + hash(sb, res);
846 inode = find_inode_fast(sb, head, res);
847 } while (inode != NULL);
848 spin_unlock(&inode_lock);
849
850 return res;
851}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852EXPORT_SYMBOL(iunique);
853
854struct inode *igrab(struct inode *inode)
855{
856 spin_lock(&inode_lock);
Jan Blunck4a3b0a42007-02-10 01:44:59 -0800857 if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 __iget(inode);
859 else
860 /*
861 * Handle the case where s_op->clear_inode is not been
862 * called yet, and somebody is calling igrab
863 * while the inode is getting freed.
864 */
865 inode = NULL;
866 spin_unlock(&inode_lock);
867 return inode;
868}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869EXPORT_SYMBOL(igrab);
870
871/**
872 * ifind - internal function, you want ilookup5() or iget5().
873 * @sb: super block of file system to search
874 * @head: the head of the list to search
875 * @test: callback used for comparisons between inodes
876 * @data: opaque data pointer to pass to @test
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700877 * @wait: if true wait for the inode to be unlocked, if false do not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 *
879 * ifind() searches for the inode specified by @data in the inode
880 * cache. This is a generalized version of ifind_fast() for file systems where
881 * the inode number is not sufficient for unique identification of an inode.
882 *
883 * If the inode is in the cache, the inode is returned with an incremented
884 * reference count.
885 *
886 * Otherwise NULL is returned.
887 *
888 * Note, @test is called with the inode_lock held, so can't sleep.
889 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800890static struct inode *ifind(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 struct hlist_head *head, int (*test)(struct inode *, void *),
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700892 void *data, const int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893{
894 struct inode *inode;
895
896 spin_lock(&inode_lock);
897 inode = find_inode(sb, head, test, data);
898 if (inode) {
899 __iget(inode);
900 spin_unlock(&inode_lock);
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700901 if (likely(wait))
902 wait_on_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 return inode;
904 }
905 spin_unlock(&inode_lock);
906 return NULL;
907}
908
909/**
910 * ifind_fast - internal function, you want ilookup() or iget().
911 * @sb: super block of file system to search
912 * @head: head of the list to search
913 * @ino: inode number to search for
914 *
915 * ifind_fast() searches for the inode @ino in the inode cache. This is for
916 * file systems where the inode number is sufficient for unique identification
917 * of an inode.
918 *
919 * If the inode is in the cache, the inode is returned with an incremented
920 * reference count.
921 *
922 * Otherwise NULL is returned.
923 */
Matt Mackall5d2bea42006-01-08 01:05:21 -0800924static struct inode *ifind_fast(struct super_block *sb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 struct hlist_head *head, unsigned long ino)
926{
927 struct inode *inode;
928
929 spin_lock(&inode_lock);
930 inode = find_inode_fast(sb, head, ino);
931 if (inode) {
932 __iget(inode);
933 spin_unlock(&inode_lock);
934 wait_on_inode(inode);
935 return inode;
936 }
937 spin_unlock(&inode_lock);
938 return NULL;
939}
940
941/**
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700942 * ilookup5_nowait - search for an inode in the inode cache
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 * @sb: super block of file system to search
944 * @hashval: hash value (usually inode number) to search for
945 * @test: callback used for comparisons between inodes
946 * @data: opaque data pointer to pass to @test
947 *
948 * ilookup5() uses ifind() to search for the inode specified by @hashval and
949 * @data in the inode cache. This is a generalized version of ilookup() for
950 * file systems where the inode number is not sufficient for unique
951 * identification of an inode.
952 *
953 * If the inode is in the cache, the inode is returned with an incremented
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700954 * reference count. Note, the inode lock is not waited upon so you have to be
955 * very careful what you do with the returned inode. You probably should be
956 * using ilookup5() instead.
957 *
958 * Otherwise NULL is returned.
959 *
960 * Note, @test is called with the inode_lock held, so can't sleep.
961 */
962struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
963 int (*test)(struct inode *, void *), void *data)
964{
965 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
966
967 return ifind(sb, head, test, data, 0);
968}
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700969EXPORT_SYMBOL(ilookup5_nowait);
970
971/**
972 * ilookup5 - search for an inode in the inode cache
973 * @sb: super block of file system to search
974 * @hashval: hash value (usually inode number) to search for
975 * @test: callback used for comparisons between inodes
976 * @data: opaque data pointer to pass to @test
977 *
978 * ilookup5() uses ifind() to search for the inode specified by @hashval and
979 * @data in the inode cache. This is a generalized version of ilookup() for
980 * file systems where the inode number is not sufficient for unique
981 * identification of an inode.
982 *
983 * If the inode is in the cache, the inode lock is waited upon and the inode is
984 * returned with an incremented reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 *
986 * Otherwise NULL is returned.
987 *
988 * Note, @test is called with the inode_lock held, so can't sleep.
989 */
990struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
991 int (*test)(struct inode *, void *), void *data)
992{
993 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
994
Anton Altaparmakov88bd5122005-07-13 01:10:44 -0700995 return ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997EXPORT_SYMBOL(ilookup5);
998
999/**
1000 * ilookup - search for an inode in the inode cache
1001 * @sb: super block of file system to search
1002 * @ino: inode number to search for
1003 *
1004 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
1005 * This is for file systems where the inode number is sufficient for unique
1006 * identification of an inode.
1007 *
1008 * If the inode is in the cache, the inode is returned with an incremented
1009 * reference count.
1010 *
1011 * Otherwise NULL is returned.
1012 */
1013struct inode *ilookup(struct super_block *sb, unsigned long ino)
1014{
1015 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1016
1017 return ifind_fast(sb, head, ino);
1018}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019EXPORT_SYMBOL(ilookup);
1020
1021/**
1022 * iget5_locked - obtain an inode from a mounted file system
1023 * @sb: super block of file system
1024 * @hashval: hash value (usually inode number) to get
1025 * @test: callback used for comparisons between inodes
1026 * @set: callback used to initialize a new struct inode
1027 * @data: opaque data pointer to pass to @test and @set
1028 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1030 * and @data in the inode cache and if present it is returned with an increased
1031 * reference count. This is a generalized version of iget_locked() for file
1032 * systems where the inode number is not sufficient for unique identification
1033 * of an inode.
1034 *
1035 * If the inode is not in cache, get_new_inode() is called to allocate a new
1036 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1037 * file system gets to fill it in before unlocking it via unlock_new_inode().
1038 *
1039 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1040 */
1041struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1042 int (*test)(struct inode *, void *),
1043 int (*set)(struct inode *, void *), void *data)
1044{
1045 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1046 struct inode *inode;
1047
Anton Altaparmakov88bd5122005-07-13 01:10:44 -07001048 inode = ifind(sb, head, test, data, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 if (inode)
1050 return inode;
1051 /*
1052 * get_new_inode() will do the right thing, re-trying the search
1053 * in case it had to block at any point.
1054 */
1055 return get_new_inode(sb, head, test, set, data);
1056}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057EXPORT_SYMBOL(iget5_locked);
1058
1059/**
1060 * iget_locked - obtain an inode from a mounted file system
1061 * @sb: super block of file system
1062 * @ino: inode number to get
1063 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1065 * the inode cache and if present it is returned with an increased reference
1066 * count. This is for file systems where the inode number is sufficient for
1067 * unique identification of an inode.
1068 *
1069 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1070 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1071 * The file system gets to fill it in before unlocking it via
1072 * unlock_new_inode().
1073 */
1074struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1075{
1076 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1077 struct inode *inode;
1078
1079 inode = ifind_fast(sb, head, ino);
1080 if (inode)
1081 return inode;
1082 /*
1083 * get_new_inode_fast() will do the right thing, re-trying the search
1084 * in case it had to block at any point.
1085 */
1086 return get_new_inode_fast(sb, head, ino);
1087}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088EXPORT_SYMBOL(iget_locked);
1089
Al Viro261bca82008-12-30 01:48:21 -05001090int insert_inode_locked(struct inode *inode)
1091{
1092 struct super_block *sb = inode->i_sb;
1093 ino_t ino = inode->i_ino;
1094 struct hlist_head *head = inode_hashtable + hash(sb, ino);
Al Viro261bca82008-12-30 01:48:21 -05001095
1096 inode->i_state |= I_LOCK|I_NEW;
1097 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001098 struct hlist_node *node;
1099 struct inode *old = NULL;
Al Viro261bca82008-12-30 01:48:21 -05001100 spin_lock(&inode_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001101 hlist_for_each_entry(old, node, head, i_hash) {
1102 if (old->i_ino != ino)
1103 continue;
1104 if (old->i_sb != sb)
1105 continue;
1106 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1107 continue;
1108 break;
1109 }
1110 if (likely(!node)) {
Al Viro261bca82008-12-30 01:48:21 -05001111 hlist_add_head(&inode->i_hash, head);
1112 spin_unlock(&inode_lock);
1113 return 0;
1114 }
1115 __iget(old);
1116 spin_unlock(&inode_lock);
1117 wait_on_inode(old);
1118 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1119 iput(old);
1120 return -EBUSY;
1121 }
1122 iput(old);
1123 }
1124}
Al Viro261bca82008-12-30 01:48:21 -05001125EXPORT_SYMBOL(insert_inode_locked);
1126
1127int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1128 int (*test)(struct inode *, void *), void *data)
1129{
1130 struct super_block *sb = inode->i_sb;
1131 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
Al Viro261bca82008-12-30 01:48:21 -05001132
1133 inode->i_state |= I_LOCK|I_NEW;
1134
1135 while (1) {
Al Viro72a43d62009-05-13 19:13:40 +01001136 struct hlist_node *node;
1137 struct inode *old = NULL;
1138
Al Viro261bca82008-12-30 01:48:21 -05001139 spin_lock(&inode_lock);
Al Viro72a43d62009-05-13 19:13:40 +01001140 hlist_for_each_entry(old, node, head, i_hash) {
1141 if (old->i_sb != sb)
1142 continue;
1143 if (!test(old, data))
1144 continue;
1145 if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1146 continue;
1147 break;
1148 }
1149 if (likely(!node)) {
Al Viro261bca82008-12-30 01:48:21 -05001150 hlist_add_head(&inode->i_hash, head);
1151 spin_unlock(&inode_lock);
1152 return 0;
1153 }
1154 __iget(old);
1155 spin_unlock(&inode_lock);
1156 wait_on_inode(old);
1157 if (unlikely(!hlist_unhashed(&old->i_hash))) {
1158 iput(old);
1159 return -EBUSY;
1160 }
1161 iput(old);
1162 }
1163}
Al Viro261bca82008-12-30 01:48:21 -05001164EXPORT_SYMBOL(insert_inode_locked4);
1165
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166/**
1167 * __insert_inode_hash - hash an inode
1168 * @inode: unhashed inode
1169 * @hashval: unsigned long value used to locate this object in the
1170 * inode_hashtable.
1171 *
1172 * Add an inode to the inode hash for this superblock.
1173 */
1174void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1175{
1176 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1177 spin_lock(&inode_lock);
1178 hlist_add_head(&inode->i_hash, head);
1179 spin_unlock(&inode_lock);
1180}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181EXPORT_SYMBOL(__insert_inode_hash);
1182
1183/**
1184 * remove_inode_hash - remove an inode from the hash
1185 * @inode: inode to unhash
1186 *
1187 * Remove an inode from the superblock.
1188 */
1189void remove_inode_hash(struct inode *inode)
1190{
1191 spin_lock(&inode_lock);
1192 hlist_del_init(&inode->i_hash);
1193 spin_unlock(&inode_lock);
1194}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195EXPORT_SYMBOL(remove_inode_hash);
1196
1197/*
1198 * Tell the filesystem that this inode is no longer of any interest and should
1199 * be completely destroyed.
1200 *
1201 * We leave the inode in the inode hash table until *after* the filesystem's
1202 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1203 * instigate) will always find up-to-date information either in the hash or on
1204 * disk.
1205 *
1206 * I_FREEING is set so that no-one will take a new reference to the inode while
1207 * it is being deleted.
1208 */
Arjan van de Venb32714b2009-01-09 07:04:15 -08001209void generic_delete_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001211 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
Arjan van de Venb32714b2009-01-09 07:04:15 -08001213 list_del_init(&inode->i_list);
1214 list_del_init(&inode->i_sb_list);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001215 WARN_ON(inode->i_state & I_NEW);
Arjan van de Venb32714b2009-01-09 07:04:15 -08001216 inode->i_state |= I_FREEING;
1217 inodes_stat.nr_inodes--;
1218 spin_unlock(&inode_lock);
1219
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 security_inode_delete(inode);
1221
1222 if (op->delete_inode) {
1223 void (*delete)(struct inode *) = op->delete_inode;
1224 if (!is_bad_inode(inode))
Jan Kara9e3509e2009-01-26 16:45:12 +01001225 vfs_dq_init(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001226 /* Filesystems implementing their own
1227 * s_op->delete_inode are required to call
1228 * truncate_inode_pages and clear_inode()
1229 * internally */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 delete(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001231 } else {
1232 truncate_inode_pages(&inode->i_data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 clear_inode(inode);
Mark Fashehe85b5652005-09-09 13:01:29 -07001234 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 spin_lock(&inode_lock);
1236 hlist_del_init(&inode->i_hash);
1237 spin_unlock(&inode_lock);
1238 wake_up_inode(inode);
Eric Sesterhennb7542f82006-04-02 13:38:18 +02001239 BUG_ON(inode->i_state != I_CLEAR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 destroy_inode(inode);
1241}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242EXPORT_SYMBOL(generic_delete_inode);
1243
1244static void generic_forget_inode(struct inode *inode)
1245{
1246 struct super_block *sb = inode->i_sb;
1247
1248 if (!hlist_unhashed(&inode->i_hash)) {
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001249 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 list_move(&inode->i_list, &inode_unused);
1251 inodes_stat.nr_unused++;
Christoph Hellwigacb0c852007-05-08 00:25:52 -07001252 if (sb->s_flags & MS_ACTIVE) {
Alexander Viro991114c2005-06-23 00:09:01 -07001253 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 return;
Alexander Viro991114c2005-06-23 00:09:01 -07001255 }
Nick Piggin7ef0d732009-03-12 14:31:38 -07001256 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001257 inode->i_state |= I_WILL_FREE;
1258 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 write_inode_now(inode, 1);
1260 spin_lock(&inode_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001261 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001262 inode->i_state &= ~I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 inodes_stat.nr_unused--;
1264 hlist_del_init(&inode->i_hash);
1265 }
1266 list_del_init(&inode->i_list);
1267 list_del_init(&inode->i_sb_list);
Nick Piggin7ef0d732009-03-12 14:31:38 -07001268 WARN_ON(inode->i_state & I_NEW);
Alexander Viro991114c2005-06-23 00:09:01 -07001269 inode->i_state |= I_FREEING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 inodes_stat.nr_inodes--;
1271 spin_unlock(&inode_lock);
1272 if (inode->i_data.nrpages)
1273 truncate_inode_pages(&inode->i_data, 0);
1274 clear_inode(inode);
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001275 wake_up_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 destroy_inode(inode);
1277}
1278
1279/*
1280 * Normal UNIX filesystem behaviour: delete the
1281 * inode when the usage count drops to zero, and
1282 * i_nlink is zero.
1283 */
Mark Fashehcb2c0232005-07-07 17:56:03 -07001284void generic_drop_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285{
1286 if (!inode->i_nlink)
1287 generic_delete_inode(inode);
1288 else
1289 generic_forget_inode(inode);
1290}
Mark Fashehcb2c0232005-07-07 17:56:03 -07001291EXPORT_SYMBOL_GPL(generic_drop_inode);
1292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293/*
1294 * Called when we're dropping the last reference
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301295 * to an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 *
1297 * Call the FS "drop()" function, defaulting to
1298 * the legacy UNIX filesystem behaviour..
1299 *
1300 * NOTE! NOTE! NOTE! We're called with the inode lock
1301 * held, and the drop function is supposed to release
1302 * the lock!
1303 */
1304static inline void iput_final(struct inode *inode)
1305{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -08001306 const struct super_operations *op = inode->i_sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 void (*drop)(struct inode *) = generic_drop_inode;
1308
1309 if (op && op->drop_inode)
1310 drop = op->drop_inode;
1311 drop(inode);
1312}
1313
1314/**
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301315 * iput - put an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 * @inode: inode to put
1317 *
1318 * Puts an inode, dropping its usage count. If the inode use count hits
1319 * zero, the inode is then freed and may also be destroyed.
1320 *
1321 * Consequently, iput() can sleep.
1322 */
1323void iput(struct inode *inode)
1324{
1325 if (inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 BUG_ON(inode->i_state == I_CLEAR);
1327
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1329 iput_final(inode);
1330 }
1331}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332EXPORT_SYMBOL(iput);
1333
1334/**
1335 * bmap - find a block number in a file
1336 * @inode: inode of file
1337 * @block: block to find
1338 *
1339 * Returns the block number on the device holding the inode that
1340 * is the disk block number for the block of the file requested.
1341 * That is, asked for block 4 of inode 1 the function will return the
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301342 * disk block relative to the disk start that holds that block of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 * file.
1344 */
Manish Katiyar6b3304b2009-03-31 19:35:54 +05301345sector_t bmap(struct inode *inode, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346{
1347 sector_t res = 0;
1348 if (inode->i_mapping->a_ops->bmap)
1349 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1350 return res;
1351}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352EXPORT_SYMBOL(bmap);
1353
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001354/*
1355 * With relative atime, only update atime if the previous atime is
1356 * earlier than either the ctime or mtime or if at least a day has
1357 * passed since the last atime update.
1358 */
1359static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1360 struct timespec now)
1361{
1362
1363 if (!(mnt->mnt_flags & MNT_RELATIME))
1364 return 1;
1365 /*
1366 * Is mtime younger than atime? If yes, update atime:
1367 */
1368 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1369 return 1;
1370 /*
1371 * Is ctime younger than atime? If yes, update atime:
1372 */
1373 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1374 return 1;
1375
1376 /*
1377 * Is the previous atime value older than a day? If yes,
1378 * update atime:
1379 */
1380 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1381 return 1;
1382 /*
1383 * Good, we can skip the atime update:
1384 */
1385 return 0;
1386}
1387
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388/**
Christoph Hellwig869243a2006-01-09 20:52:03 -08001389 * touch_atime - update the access time
1390 * @mnt: mount the inode is accessed on
Martin Waitz7045f372006-02-01 03:06:57 -08001391 * @dentry: dentry accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 *
1393 * Update the accessed time on an inode and mark it for writeback.
1394 * This function automatically handles read only file systems and media,
1395 * as well as the "noatime" flag and inode specific "noatime" markers.
1396 */
Christoph Hellwig869243a2006-01-09 20:52:03 -08001397void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398{
Christoph Hellwig869243a2006-01-09 20:52:03 -08001399 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 struct timespec now;
1401
Dave Hansencdb70f32008-02-15 14:37:41 -08001402 if (mnt_want_write(mnt))
1403 return;
Andrew Mortonb2276132006-12-13 00:34:33 -08001404 if (inode->i_flags & S_NOATIME)
Dave Hansencdb70f32008-02-15 14:37:41 -08001405 goto out;
Eric Dumazet37756ce2007-02-10 01:44:49 -08001406 if (IS_NOATIME(inode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001407 goto out;
Andrew Mortonb2276132006-12-13 00:34:33 -08001408 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
Dave Hansencdb70f32008-02-15 14:37:41 -08001409 goto out;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001410
Dave Hansencdb70f32008-02-15 14:37:41 -08001411 if (mnt->mnt_flags & MNT_NOATIME)
1412 goto out;
1413 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1414 goto out;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001415
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 now = current_fs_time(inode->i_sb);
Matthew Garrett11ff6f052009-03-26 17:32:14 +00001417
1418 if (!relatime_need_update(mnt, inode, now))
1419 goto out;
1420
Valerie Henson47ae32d2006-12-13 00:34:34 -08001421 if (timespec_equal(&inode->i_atime, &now))
Dave Hansencdb70f32008-02-15 14:37:41 -08001422 goto out;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001423
1424 inode->i_atime = now;
1425 mark_inode_dirty_sync(inode);
Dave Hansencdb70f32008-02-15 14:37:41 -08001426out:
1427 mnt_drop_write(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428}
Christoph Hellwig869243a2006-01-09 20:52:03 -08001429EXPORT_SYMBOL(touch_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
1431/**
Christoph Hellwig870f4812006-01-09 20:52:01 -08001432 * file_update_time - update mtime and ctime time
1433 * @file: file accessed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 *
Christoph Hellwig870f4812006-01-09 20:52:01 -08001435 * Update the mtime and ctime members of an inode and mark the inode
1436 * for writeback. Note that this function is meant exclusively for
1437 * usage in the file write path of filesystems, and filesystems may
1438 * choose to explicitly ignore update via this function with the
Wolfram Sang2eadfc02009-04-02 15:23:37 +02001439 * S_NOCMTIME inode flag, e.g. for network filesystem where these
Christoph Hellwig870f4812006-01-09 20:52:01 -08001440 * timestamps are handled by the server.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 */
1442
Christoph Hellwig870f4812006-01-09 20:52:01 -08001443void file_update_time(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001445 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 struct timespec now;
1447 int sync_it = 0;
Dave Hansen20ddee22008-02-15 14:37:43 -08001448 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
1450 if (IS_NOCMTIME(inode))
1451 return;
Dave Hansen20ddee22008-02-15 14:37:43 -08001452
npiggin@suse.de96029c42009-04-26 20:25:55 +10001453 err = mnt_want_write_file(file);
Dave Hansen20ddee22008-02-15 14:37:43 -08001454 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 return;
1456
1457 now = current_fs_time(inode->i_sb);
Andreas Mohred97bd32006-10-02 02:17:17 -07001458 if (!timespec_equal(&inode->i_mtime, &now)) {
1459 inode->i_mtime = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001461 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
Andreas Mohred97bd32006-10-02 02:17:17 -07001463 if (!timespec_equal(&inode->i_ctime, &now)) {
1464 inode->i_ctime = now;
Christoph Hellwig870f4812006-01-09 20:52:01 -08001465 sync_it = 1;
Andreas Mohred97bd32006-10-02 02:17:17 -07001466 }
Christoph Hellwig870f4812006-01-09 20:52:01 -08001467
Jean Noel Cordenner7a224222008-01-28 23:58:27 -05001468 if (IS_I_VERSION(inode)) {
1469 inode_inc_iversion(inode);
1470 sync_it = 1;
1471 }
1472
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 if (sync_it)
1474 mark_inode_dirty_sync(inode);
Dave Hansen20ddee22008-02-15 14:37:43 -08001475 mnt_drop_write(file->f_path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476}
Christoph Hellwig870f4812006-01-09 20:52:01 -08001477EXPORT_SYMBOL(file_update_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
1479int inode_needs_sync(struct inode *inode)
1480{
1481 if (IS_SYNC(inode))
1482 return 1;
1483 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1484 return 1;
1485 return 0;
1486}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487EXPORT_SYMBOL(inode_needs_sync);
1488
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489int inode_wait(void *word)
1490{
1491 schedule();
1492 return 0;
1493}
Stephen Rothwelld44dab82008-11-10 17:06:05 +11001494EXPORT_SYMBOL(inode_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495
1496/*
Miklos Szeredi168a9fd2005-07-12 13:58:10 -07001497 * If we try to find an inode in the inode hash while it is being
1498 * deleted, we have to wait until the filesystem completes its
1499 * deletion before reporting that it isn't found. This function waits
1500 * until the deletion _might_ have completed. Callers are responsible
1501 * to recheck inode state.
1502 *
1503 * It doesn't matter if I_LOCK is not set initially, a call to
1504 * wake_up_inode() after removing from the hash list will DTRT.
1505 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 * This is called with inode_lock held.
1507 */
1508static void __wait_on_freeing_inode(struct inode *inode)
1509{
1510 wait_queue_head_t *wq;
1511 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1513 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1514 spin_unlock(&inode_lock);
1515 schedule();
1516 finish_wait(wq, &wait.wait);
1517 spin_lock(&inode_lock);
1518}
1519
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520static __initdata unsigned long ihash_entries;
1521static int __init set_ihash_entries(char *str)
1522{
1523 if (!str)
1524 return 0;
1525 ihash_entries = simple_strtoul(str, &str, 0);
1526 return 1;
1527}
1528__setup("ihash_entries=", set_ihash_entries);
1529
1530/*
1531 * Initialize the waitqueues and inode hash table.
1532 */
1533void __init inode_init_early(void)
1534{
1535 int loop;
1536
1537 /* If hashes are distributed across NUMA nodes, defer
1538 * hash allocation until vmalloc space is available.
1539 */
1540 if (hashdist)
1541 return;
1542
1543 inode_hashtable =
1544 alloc_large_system_hash("Inode-cache",
1545 sizeof(struct hlist_head),
1546 ihash_entries,
1547 14,
1548 HASH_EARLY,
1549 &i_hash_shift,
1550 &i_hash_mask,
1551 0);
1552
1553 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1554 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1555}
1556
Denis Cheng74bf17c2007-10-16 23:26:30 -07001557void __init inode_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558{
1559 int loop;
1560
1561 /* inode slab cache */
Paul Jacksonb0196002006-03-24 03:16:09 -08001562 inode_cachep = kmem_cache_create("inode_cache",
1563 sizeof(struct inode),
1564 0,
1565 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1566 SLAB_MEM_SPREAD),
Paul Mundt20c2df82007-07-20 10:11:58 +09001567 init_once);
Rusty Russell8e1f9362007-07-17 04:03:17 -07001568 register_shrinker(&icache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
1570 /* Hash may have been set up in inode_init_early */
1571 if (!hashdist)
1572 return;
1573
1574 inode_hashtable =
1575 alloc_large_system_hash("Inode-cache",
1576 sizeof(struct hlist_head),
1577 ihash_entries,
1578 14,
1579 0,
1580 &i_hash_shift,
1581 &i_hash_mask,
1582 0);
1583
1584 for (loop = 0; loop < (1 << i_hash_shift); loop++)
1585 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1586}
1587
1588void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1589{
1590 inode->i_mode = mode;
1591 if (S_ISCHR(mode)) {
1592 inode->i_fop = &def_chr_fops;
1593 inode->i_rdev = rdev;
1594 } else if (S_ISBLK(mode)) {
1595 inode->i_fop = &def_blk_fops;
1596 inode->i_rdev = rdev;
1597 } else if (S_ISFIFO(mode))
1598 inode->i_fop = &def_fifo_fops;
1599 else if (S_ISSOCK(mode))
1600 inode->i_fop = &bad_sock_fops;
1601 else
1602 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1603 mode);
1604}
1605EXPORT_SYMBOL(init_special_inode);