]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/inode.c | |
3 | * | |
4 | * (C) 1997 Linus Torvalds | |
5 | */ | |
6 | ||
1da177e4 LT |
7 | #include <linux/fs.h> |
8 | #include <linux/mm.h> | |
9 | #include <linux/dcache.h> | |
10 | #include <linux/init.h> | |
1da177e4 LT |
11 | #include <linux/slab.h> |
12 | #include <linux/writeback.h> | |
13 | #include <linux/module.h> | |
14 | #include <linux/backing-dev.h> | |
15 | #include <linux/wait.h> | |
88e0fbc4 | 16 | #include <linux/rwsem.h> |
1da177e4 LT |
17 | #include <linux/hash.h> |
18 | #include <linux/swap.h> | |
19 | #include <linux/security.h> | |
20 | #include <linux/pagemap.h> | |
21 | #include <linux/cdev.h> | |
22 | #include <linux/bootmem.h> | |
3be25f49 | 23 | #include <linux/fsnotify.h> |
fc33a7bb | 24 | #include <linux/mount.h> |
efaee192 | 25 | #include <linux/async.h> |
f19d4a8f | 26 | #include <linux/posix_acl.h> |
1da177e4 LT |
27 | |
28 | /* | |
29 | * This is needed for the following functions: | |
30 | * - inode_has_buffers | |
31 | * - invalidate_inode_buffers | |
1da177e4 LT |
32 | * - 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 | ||
fa3536cc ED |
60 | static unsigned int i_hash_mask __read_mostly; |
61 | static unsigned int i_hash_shift __read_mostly; | |
1da177e4 LT |
62 | |
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 | ||
9e38d86f | 75 | static LIST_HEAD(inode_unused); |
fa3536cc | 76 | static struct hlist_head *inode_hashtable __read_mostly; |
1da177e4 LT |
77 | |
78 | /* | |
79 | * A simple spinlock to protect the list manipulations. | |
80 | * | |
81 | * NOTE! You also have to own the lock if you change | |
82 | * the i_state of an inode while it is in use.. | |
83 | */ | |
84 | DEFINE_SPINLOCK(inode_lock); | |
85 | ||
86 | /* | |
88e0fbc4 | 87 | * iprune_sem provides exclusion between the kswapd or try_to_free_pages |
1da177e4 LT |
88 | * icache shrinking path, and the umount path. Without this exclusion, |
89 | * by the time prune_icache calls iput for the inode whose pages it has | |
90 | * been invalidating, or by the time it calls clear_inode & destroy_inode | |
91 | * from its final dispose_list, the struct super_block they refer to | |
92 | * (for inode->i_sb->s_op) may already have been freed and reused. | |
88e0fbc4 NP |
93 | * |
94 | * We make this an rwsem because the fastpath is icache shrinking. In | |
95 | * some cases a filesystem may be doing a significant amount of work in | |
96 | * its inode reclaim code, so this should improve parallelism. | |
1da177e4 | 97 | */ |
88e0fbc4 | 98 | static DECLARE_RWSEM(iprune_sem); |
1da177e4 LT |
99 | |
100 | /* | |
101 | * Statistics gathering.. | |
102 | */ | |
103 | struct inodes_stat_t inodes_stat; | |
104 | ||
cffbc8aa DC |
105 | static struct percpu_counter nr_inodes __cacheline_aligned_in_smp; |
106 | static struct percpu_counter nr_inodes_unused __cacheline_aligned_in_smp; | |
107 | ||
6b3304b5 | 108 | static struct kmem_cache *inode_cachep __read_mostly; |
1da177e4 | 109 | |
cffbc8aa DC |
110 | static inline int get_nr_inodes(void) |
111 | { | |
112 | return percpu_counter_sum_positive(&nr_inodes); | |
113 | } | |
114 | ||
115 | static inline int get_nr_inodes_unused(void) | |
116 | { | |
117 | return percpu_counter_sum_positive(&nr_inodes_unused); | |
118 | } | |
119 | ||
120 | int get_nr_dirty_inodes(void) | |
121 | { | |
122 | int nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); | |
123 | return nr_dirty > 0 ? nr_dirty : 0; | |
124 | ||
125 | } | |
126 | ||
127 | /* | |
128 | * Handle nr_inode sysctl | |
129 | */ | |
130 | #ifdef CONFIG_SYSCTL | |
131 | int proc_nr_inodes(ctl_table *table, int write, | |
132 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
133 | { | |
134 | inodes_stat.nr_inodes = get_nr_inodes(); | |
135 | inodes_stat.nr_unused = get_nr_inodes_unused(); | |
136 | return proc_dointvec(table, write, buffer, lenp, ppos); | |
137 | } | |
138 | #endif | |
139 | ||
1c0eeaf5 JE |
140 | static void wake_up_inode(struct inode *inode) |
141 | { | |
142 | /* | |
143 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
144 | */ | |
145 | smp_mb(); | |
eaff8079 | 146 | wake_up_bit(&inode->i_state, __I_NEW); |
1c0eeaf5 JE |
147 | } |
148 | ||
2cb1599f DC |
149 | /** |
150 | * inode_init_always - perform inode structure intialisation | |
0bc02f3f RD |
151 | * @sb: superblock inode belongs to |
152 | * @inode: inode to initialise | |
2cb1599f DC |
153 | * |
154 | * These are initializations that need to be done on every inode | |
155 | * allocation as the fields are not initialised by slab allocation. | |
156 | */ | |
54e34621 | 157 | int inode_init_always(struct super_block *sb, struct inode *inode) |
1da177e4 | 158 | { |
f5e54d6e | 159 | static const struct address_space_operations empty_aops; |
6e1d5dcc | 160 | static const struct inode_operations empty_iops; |
99ac48f5 | 161 | static const struct file_operations empty_fops; |
6b3304b5 | 162 | struct address_space *const mapping = &inode->i_data; |
2cb1599f DC |
163 | |
164 | inode->i_sb = sb; | |
165 | inode->i_blkbits = sb->s_blocksize_bits; | |
166 | inode->i_flags = 0; | |
167 | atomic_set(&inode->i_count, 1); | |
168 | inode->i_op = &empty_iops; | |
169 | inode->i_fop = &empty_fops; | |
170 | inode->i_nlink = 1; | |
56ff5efa AV |
171 | inode->i_uid = 0; |
172 | inode->i_gid = 0; | |
2cb1599f DC |
173 | atomic_set(&inode->i_writecount, 0); |
174 | inode->i_size = 0; | |
175 | inode->i_blocks = 0; | |
176 | inode->i_bytes = 0; | |
177 | inode->i_generation = 0; | |
1da177e4 | 178 | #ifdef CONFIG_QUOTA |
2cb1599f | 179 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); |
1da177e4 | 180 | #endif |
2cb1599f DC |
181 | inode->i_pipe = NULL; |
182 | inode->i_bdev = NULL; | |
183 | inode->i_cdev = NULL; | |
184 | inode->i_rdev = 0; | |
185 | inode->dirtied_when = 0; | |
6146f0d5 MZ |
186 | |
187 | if (security_inode_alloc(inode)) | |
54e34621 | 188 | goto out; |
2cb1599f DC |
189 | spin_lock_init(&inode->i_lock); |
190 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
191 | ||
192 | mutex_init(&inode->i_mutex); | |
193 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); | |
194 | ||
195 | init_rwsem(&inode->i_alloc_sem); | |
196 | lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key); | |
197 | ||
198 | mapping->a_ops = &empty_aops; | |
199 | mapping->host = inode; | |
200 | mapping->flags = 0; | |
3c1d4378 | 201 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); |
2cb1599f DC |
202 | mapping->assoc_mapping = NULL; |
203 | mapping->backing_dev_info = &default_backing_dev_info; | |
204 | mapping->writeback_index = 0; | |
205 | ||
206 | /* | |
207 | * If the block_device provides a backing_dev_info for client | |
208 | * inodes then use that. Otherwise the inode share the bdev's | |
209 | * backing_dev_info. | |
210 | */ | |
211 | if (sb->s_bdev) { | |
212 | struct backing_dev_info *bdi; | |
213 | ||
2c96ce9f | 214 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; |
2cb1599f DC |
215 | mapping->backing_dev_info = bdi; |
216 | } | |
217 | inode->i_private = NULL; | |
218 | inode->i_mapping = mapping; | |
f19d4a8f AV |
219 | #ifdef CONFIG_FS_POSIX_ACL |
220 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; | |
221 | #endif | |
2cb1599f | 222 | |
3be25f49 EP |
223 | #ifdef CONFIG_FSNOTIFY |
224 | inode->i_fsnotify_mask = 0; | |
225 | #endif | |
226 | ||
cffbc8aa DC |
227 | percpu_counter_inc(&nr_inodes); |
228 | ||
54e34621 | 229 | return 0; |
54e34621 CH |
230 | out: |
231 | return -ENOMEM; | |
1da177e4 | 232 | } |
2cb1599f DC |
233 | EXPORT_SYMBOL(inode_init_always); |
234 | ||
235 | static struct inode *alloc_inode(struct super_block *sb) | |
236 | { | |
237 | struct inode *inode; | |
238 | ||
239 | if (sb->s_op->alloc_inode) | |
240 | inode = sb->s_op->alloc_inode(sb); | |
241 | else | |
242 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
243 | ||
54e34621 CH |
244 | if (!inode) |
245 | return NULL; | |
246 | ||
247 | if (unlikely(inode_init_always(sb, inode))) { | |
248 | if (inode->i_sb->s_op->destroy_inode) | |
249 | inode->i_sb->s_op->destroy_inode(inode); | |
250 | else | |
251 | kmem_cache_free(inode_cachep, inode); | |
252 | return NULL; | |
253 | } | |
254 | ||
255 | return inode; | |
2cb1599f | 256 | } |
1da177e4 | 257 | |
2e00c97e | 258 | void __destroy_inode(struct inode *inode) |
1da177e4 | 259 | { |
b7542f8c | 260 | BUG_ON(inode_has_buffers(inode)); |
1da177e4 | 261 | security_inode_free(inode); |
3be25f49 | 262 | fsnotify_inode_delete(inode); |
f19d4a8f AV |
263 | #ifdef CONFIG_FS_POSIX_ACL |
264 | if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) | |
265 | posix_acl_release(inode->i_acl); | |
266 | if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) | |
267 | posix_acl_release(inode->i_default_acl); | |
268 | #endif | |
cffbc8aa | 269 | percpu_counter_dec(&nr_inodes); |
2e00c97e CH |
270 | } |
271 | EXPORT_SYMBOL(__destroy_inode); | |
272 | ||
56b0dacf | 273 | static void destroy_inode(struct inode *inode) |
2e00c97e CH |
274 | { |
275 | __destroy_inode(inode); | |
1da177e4 LT |
276 | if (inode->i_sb->s_op->destroy_inode) |
277 | inode->i_sb->s_op->destroy_inode(inode); | |
278 | else | |
279 | kmem_cache_free(inode_cachep, (inode)); | |
280 | } | |
1da177e4 LT |
281 | |
282 | /* | |
283 | * These are initializations that only need to be done | |
284 | * once, because the fields are idempotent across use | |
285 | * of the inode, so let the slab aware of that. | |
286 | */ | |
287 | void inode_init_once(struct inode *inode) | |
288 | { | |
289 | memset(inode, 0, sizeof(*inode)); | |
290 | INIT_HLIST_NODE(&inode->i_hash); | |
291 | INIT_LIST_HEAD(&inode->i_dentry); | |
292 | INIT_LIST_HEAD(&inode->i_devices); | |
9e38d86f | 293 | INIT_LIST_HEAD(&inode->i_list); |
1da177e4 | 294 | INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC); |
19fd6231 | 295 | spin_lock_init(&inode->i_data.tree_lock); |
1da177e4 LT |
296 | spin_lock_init(&inode->i_data.i_mmap_lock); |
297 | INIT_LIST_HEAD(&inode->i_data.private_list); | |
298 | spin_lock_init(&inode->i_data.private_lock); | |
299 | INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap); | |
300 | INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear); | |
1da177e4 | 301 | i_size_ordered_init(inode); |
3be25f49 | 302 | #ifdef CONFIG_FSNOTIFY |
e61ce867 | 303 | INIT_HLIST_HEAD(&inode->i_fsnotify_marks); |
3be25f49 | 304 | #endif |
1da177e4 | 305 | } |
1da177e4 LT |
306 | EXPORT_SYMBOL(inode_init_once); |
307 | ||
51cc5068 | 308 | static void init_once(void *foo) |
1da177e4 | 309 | { |
6b3304b5 | 310 | struct inode *inode = (struct inode *) foo; |
1da177e4 | 311 | |
a35afb83 | 312 | inode_init_once(inode); |
1da177e4 LT |
313 | } |
314 | ||
315 | /* | |
316 | * inode_lock must be held | |
317 | */ | |
6b3304b5 | 318 | void __iget(struct inode *inode) |
1da177e4 | 319 | { |
9e38d86f NP |
320 | atomic_inc(&inode->i_count); |
321 | } | |
2e147f1e | 322 | |
7de9c6ee AV |
323 | /* |
324 | * get additional reference to inode; caller must already hold one. | |
325 | */ | |
326 | void ihold(struct inode *inode) | |
327 | { | |
328 | WARN_ON(atomic_inc_return(&inode->i_count) < 2); | |
329 | } | |
330 | EXPORT_SYMBOL(ihold); | |
331 | ||
9e38d86f NP |
332 | static void inode_lru_list_add(struct inode *inode) |
333 | { | |
334 | if (list_empty(&inode->i_list)) { | |
335 | list_add(&inode->i_list, &inode_unused); | |
336 | percpu_counter_inc(&nr_inodes_unused); | |
337 | } | |
338 | } | |
339 | ||
340 | static void inode_lru_list_del(struct inode *inode) | |
341 | { | |
342 | if (!list_empty(&inode->i_list)) { | |
343 | list_del_init(&inode->i_list); | |
344 | percpu_counter_dec(&nr_inodes_unused); | |
345 | } | |
1da177e4 LT |
346 | } |
347 | ||
646ec461 CH |
348 | static inline void __inode_sb_list_add(struct inode *inode) |
349 | { | |
350 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); | |
351 | } | |
352 | ||
353 | /** | |
354 | * inode_sb_list_add - add inode to the superblock list of inodes | |
355 | * @inode: inode to add | |
356 | */ | |
357 | void inode_sb_list_add(struct inode *inode) | |
358 | { | |
359 | spin_lock(&inode_lock); | |
360 | __inode_sb_list_add(inode); | |
361 | spin_unlock(&inode_lock); | |
362 | } | |
363 | EXPORT_SYMBOL_GPL(inode_sb_list_add); | |
364 | ||
365 | static inline void __inode_sb_list_del(struct inode *inode) | |
366 | { | |
367 | list_del_init(&inode->i_sb_list); | |
368 | } | |
369 | ||
4c51acbc DC |
370 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
371 | { | |
372 | unsigned long tmp; | |
373 | ||
374 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
375 | L1_CACHE_BYTES; | |
376 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS); | |
377 | return tmp & I_HASHMASK; | |
378 | } | |
379 | ||
380 | /** | |
381 | * __insert_inode_hash - hash an inode | |
382 | * @inode: unhashed inode | |
383 | * @hashval: unsigned long value used to locate this object in the | |
384 | * inode_hashtable. | |
385 | * | |
386 | * Add an inode to the inode hash for this superblock. | |
387 | */ | |
388 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
389 | { | |
646ec461 CH |
390 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); |
391 | ||
4c51acbc | 392 | spin_lock(&inode_lock); |
646ec461 | 393 | hlist_add_head(&inode->i_hash, b); |
4c51acbc DC |
394 | spin_unlock(&inode_lock); |
395 | } | |
396 | EXPORT_SYMBOL(__insert_inode_hash); | |
397 | ||
398 | /** | |
399 | * __remove_inode_hash - remove an inode from the hash | |
400 | * @inode: inode to unhash | |
401 | * | |
402 | * Remove an inode from the superblock. | |
403 | */ | |
404 | static void __remove_inode_hash(struct inode *inode) | |
405 | { | |
406 | hlist_del_init(&inode->i_hash); | |
407 | } | |
408 | ||
409 | /** | |
410 | * remove_inode_hash - remove an inode from the hash | |
411 | * @inode: inode to unhash | |
412 | * | |
413 | * Remove an inode from the superblock. | |
414 | */ | |
415 | void remove_inode_hash(struct inode *inode) | |
416 | { | |
417 | spin_lock(&inode_lock); | |
418 | hlist_del_init(&inode->i_hash); | |
419 | spin_unlock(&inode_lock); | |
420 | } | |
421 | EXPORT_SYMBOL(remove_inode_hash); | |
422 | ||
b0683aa6 AV |
423 | void end_writeback(struct inode *inode) |
424 | { | |
425 | might_sleep(); | |
426 | BUG_ON(inode->i_data.nrpages); | |
427 | BUG_ON(!list_empty(&inode->i_data.private_list)); | |
428 | BUG_ON(!(inode->i_state & I_FREEING)); | |
429 | BUG_ON(inode->i_state & I_CLEAR); | |
430 | inode_sync_wait(inode); | |
431 | inode->i_state = I_FREEING | I_CLEAR; | |
432 | } | |
433 | EXPORT_SYMBOL(end_writeback); | |
434 | ||
644da596 | 435 | static void evict(struct inode *inode) |
b4272d4c AV |
436 | { |
437 | const struct super_operations *op = inode->i_sb->s_op; | |
438 | ||
be7ce416 AV |
439 | if (op->evict_inode) { |
440 | op->evict_inode(inode); | |
b4272d4c AV |
441 | } else { |
442 | if (inode->i_data.nrpages) | |
443 | truncate_inode_pages(&inode->i_data, 0); | |
30140837 | 444 | end_writeback(inode); |
b4272d4c | 445 | } |
661074e9 AV |
446 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
447 | bd_forget(inode); | |
448 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) | |
449 | cd_forget(inode); | |
b4272d4c AV |
450 | } |
451 | ||
1da177e4 LT |
452 | /* |
453 | * dispose_list - dispose of the contents of a local list | |
454 | * @head: the head of the list to free | |
455 | * | |
456 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
457 | * need to worry about list corruption and SMP locks. | |
458 | */ | |
459 | static void dispose_list(struct list_head *head) | |
460 | { | |
1da177e4 LT |
461 | while (!list_empty(head)) { |
462 | struct inode *inode; | |
463 | ||
b5e61818 | 464 | inode = list_first_entry(head, struct inode, i_list); |
9e38d86f | 465 | list_del_init(&inode->i_list); |
1da177e4 | 466 | |
644da596 | 467 | evict(inode); |
4120db47 AB |
468 | |
469 | spin_lock(&inode_lock); | |
4c51acbc | 470 | __remove_inode_hash(inode); |
646ec461 | 471 | __inode_sb_list_del(inode); |
4120db47 AB |
472 | spin_unlock(&inode_lock); |
473 | ||
474 | wake_up_inode(inode); | |
1da177e4 | 475 | destroy_inode(inode); |
1da177e4 | 476 | } |
1da177e4 LT |
477 | } |
478 | ||
479 | /* | |
480 | * Invalidate all inodes for a device. | |
481 | */ | |
482 | static int invalidate_list(struct list_head *head, struct list_head *dispose) | |
483 | { | |
484 | struct list_head *next; | |
cffbc8aa | 485 | int busy = 0; |
1da177e4 LT |
486 | |
487 | next = head->next; | |
488 | for (;;) { | |
6b3304b5 MK |
489 | struct list_head *tmp = next; |
490 | struct inode *inode; | |
1da177e4 LT |
491 | |
492 | /* | |
493 | * We can reschedule here without worrying about the list's | |
494 | * consistency because the per-sb list of inodes must not | |
88e0fbc4 | 495 | * change during umount anymore, and because iprune_sem keeps |
1da177e4 LT |
496 | * shrink_icache_memory() away. |
497 | */ | |
498 | cond_resched_lock(&inode_lock); | |
499 | ||
500 | next = next->next; | |
501 | if (tmp == head) | |
502 | break; | |
503 | inode = list_entry(tmp, struct inode, i_sb_list); | |
aabb8fdb NP |
504 | if (inode->i_state & I_NEW) |
505 | continue; | |
1da177e4 LT |
506 | invalidate_inode_buffers(inode); |
507 | if (!atomic_read(&inode->i_count)) { | |
1da177e4 | 508 | list_move(&inode->i_list, dispose); |
7ef0d737 | 509 | WARN_ON(inode->i_state & I_NEW); |
1da177e4 | 510 | inode->i_state |= I_FREEING; |
9e38d86f NP |
511 | if (!(inode->i_state & (I_DIRTY | I_SYNC))) |
512 | percpu_counter_dec(&nr_inodes_unused); | |
1da177e4 LT |
513 | continue; |
514 | } | |
515 | busy = 1; | |
516 | } | |
1da177e4 LT |
517 | return busy; |
518 | } | |
519 | ||
1da177e4 LT |
520 | /** |
521 | * invalidate_inodes - discard the inodes on a device | |
522 | * @sb: superblock | |
523 | * | |
524 | * Discard all of the inodes for a given superblock. If the discard | |
525 | * fails because there are busy inodes then a non zero value is returned. | |
526 | * If the discard is successful all the inodes have been discarded. | |
527 | */ | |
6b3304b5 | 528 | int invalidate_inodes(struct super_block *sb) |
1da177e4 LT |
529 | { |
530 | int busy; | |
531 | LIST_HEAD(throw_away); | |
532 | ||
88e0fbc4 | 533 | down_write(&iprune_sem); |
1da177e4 | 534 | spin_lock(&inode_lock); |
164bc619 | 535 | fsnotify_unmount_inodes(&sb->s_inodes); |
1da177e4 LT |
536 | busy = invalidate_list(&sb->s_inodes, &throw_away); |
537 | spin_unlock(&inode_lock); | |
538 | ||
539 | dispose_list(&throw_away); | |
88e0fbc4 | 540 | up_write(&iprune_sem); |
1da177e4 LT |
541 | |
542 | return busy; | |
543 | } | |
1da177e4 LT |
544 | |
545 | static int can_unuse(struct inode *inode) | |
546 | { | |
9e38d86f | 547 | if (inode->i_state & ~I_REFERENCED) |
1da177e4 LT |
548 | return 0; |
549 | if (inode_has_buffers(inode)) | |
550 | return 0; | |
551 | if (atomic_read(&inode->i_count)) | |
552 | return 0; | |
553 | if (inode->i_data.nrpages) | |
554 | return 0; | |
555 | return 1; | |
556 | } | |
557 | ||
558 | /* | |
9e38d86f NP |
559 | * Scan `goal' inodes on the unused list for freeable ones. They are moved to a |
560 | * temporary list and then are freed outside inode_lock by dispose_list(). | |
1da177e4 LT |
561 | * |
562 | * Any inodes which are pinned purely because of attached pagecache have their | |
9e38d86f NP |
563 | * pagecache removed. If the inode has metadata buffers attached to |
564 | * mapping->private_list then try to remove them. | |
1da177e4 | 565 | * |
9e38d86f NP |
566 | * If the inode has the I_REFERENCED flag set, then it means that it has been |
567 | * used recently - the flag is set in iput_final(). When we encounter such an | |
568 | * inode, clear the flag and move it to the back of the LRU so it gets another | |
569 | * pass through the LRU before it gets reclaimed. This is necessary because of | |
570 | * the fact we are doing lazy LRU updates to minimise lock contention so the | |
571 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes | |
572 | * with this flag set because they are the inodes that are out of order. | |
1da177e4 LT |
573 | */ |
574 | static void prune_icache(int nr_to_scan) | |
575 | { | |
576 | LIST_HEAD(freeable); | |
1da177e4 LT |
577 | int nr_scanned; |
578 | unsigned long reap = 0; | |
579 | ||
88e0fbc4 | 580 | down_read(&iprune_sem); |
1da177e4 LT |
581 | spin_lock(&inode_lock); |
582 | for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) { | |
583 | struct inode *inode; | |
584 | ||
585 | if (list_empty(&inode_unused)) | |
586 | break; | |
587 | ||
588 | inode = list_entry(inode_unused.prev, struct inode, i_list); | |
589 | ||
9e38d86f NP |
590 | /* |
591 | * Referenced or dirty inodes are still in use. Give them | |
592 | * another pass through the LRU as we canot reclaim them now. | |
593 | */ | |
594 | if (atomic_read(&inode->i_count) || | |
595 | (inode->i_state & ~I_REFERENCED)) { | |
596 | list_del_init(&inode->i_list); | |
597 | percpu_counter_dec(&nr_inodes_unused); | |
598 | continue; | |
599 | } | |
600 | ||
601 | /* recently referenced inodes get one more pass */ | |
602 | if (inode->i_state & I_REFERENCED) { | |
1da177e4 | 603 | list_move(&inode->i_list, &inode_unused); |
9e38d86f | 604 | inode->i_state &= ~I_REFERENCED; |
1da177e4 LT |
605 | continue; |
606 | } | |
607 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { | |
608 | __iget(inode); | |
609 | spin_unlock(&inode_lock); | |
610 | if (remove_inode_buffers(inode)) | |
fc0ecff6 AM |
611 | reap += invalidate_mapping_pages(&inode->i_data, |
612 | 0, -1); | |
1da177e4 LT |
613 | iput(inode); |
614 | spin_lock(&inode_lock); | |
615 | ||
616 | if (inode != list_entry(inode_unused.next, | |
617 | struct inode, i_list)) | |
618 | continue; /* wrong inode or list_empty */ | |
619 | if (!can_unuse(inode)) | |
620 | continue; | |
621 | } | |
1da177e4 | 622 | list_move(&inode->i_list, &freeable); |
7ef0d737 | 623 | WARN_ON(inode->i_state & I_NEW); |
1da177e4 | 624 | inode->i_state |= I_FREEING; |
cffbc8aa | 625 | percpu_counter_dec(&nr_inodes_unused); |
1da177e4 | 626 | } |
f8891e5e CL |
627 | if (current_is_kswapd()) |
628 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
629 | else | |
630 | __count_vm_events(PGINODESTEAL, reap); | |
1da177e4 LT |
631 | spin_unlock(&inode_lock); |
632 | ||
633 | dispose_list(&freeable); | |
88e0fbc4 | 634 | up_read(&iprune_sem); |
1da177e4 LT |
635 | } |
636 | ||
637 | /* | |
638 | * shrink_icache_memory() will attempt to reclaim some unused inodes. Here, | |
639 | * "unused" means that no dentries are referring to the inodes: the files are | |
640 | * not open and the dcache references to those inodes have already been | |
641 | * reclaimed. | |
642 | * | |
643 | * This function is passed the number of inodes to scan, and it returns the | |
644 | * total number of remaining possibly-reclaimable inodes. | |
645 | */ | |
7f8275d0 | 646 | static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
1da177e4 LT |
647 | { |
648 | if (nr) { | |
649 | /* | |
650 | * Nasty deadlock avoidance. We may hold various FS locks, | |
651 | * and we don't want to recurse into the FS that called us | |
652 | * in clear_inode() and friends.. | |
6b3304b5 | 653 | */ |
1da177e4 LT |
654 | if (!(gfp_mask & __GFP_FS)) |
655 | return -1; | |
656 | prune_icache(nr); | |
657 | } | |
cffbc8aa | 658 | return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure; |
1da177e4 LT |
659 | } |
660 | ||
8e1f936b RR |
661 | static struct shrinker icache_shrinker = { |
662 | .shrink = shrink_icache_memory, | |
663 | .seeks = DEFAULT_SEEKS, | |
664 | }; | |
665 | ||
1da177e4 LT |
666 | static void __wait_on_freeing_inode(struct inode *inode); |
667 | /* | |
668 | * Called with the inode lock held. | |
1da177e4 | 669 | */ |
6b3304b5 MK |
670 | static struct inode *find_inode(struct super_block *sb, |
671 | struct hlist_head *head, | |
672 | int (*test)(struct inode *, void *), | |
673 | void *data) | |
1da177e4 LT |
674 | { |
675 | struct hlist_node *node; | |
6b3304b5 | 676 | struct inode *inode = NULL; |
1da177e4 LT |
677 | |
678 | repeat: | |
c5c8be3c | 679 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
680 | if (inode->i_sb != sb) |
681 | continue; | |
682 | if (!test(inode, data)) | |
683 | continue; | |
a4ffdde6 | 684 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
685 | __wait_on_freeing_inode(inode); |
686 | goto repeat; | |
687 | } | |
f7899bd5 CH |
688 | __iget(inode); |
689 | return inode; | |
1da177e4 | 690 | } |
f7899bd5 | 691 | return NULL; |
1da177e4 LT |
692 | } |
693 | ||
694 | /* | |
695 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
696 | * iget_locked for details. | |
697 | */ | |
6b3304b5 MK |
698 | static struct inode *find_inode_fast(struct super_block *sb, |
699 | struct hlist_head *head, unsigned long ino) | |
1da177e4 LT |
700 | { |
701 | struct hlist_node *node; | |
6b3304b5 | 702 | struct inode *inode = NULL; |
1da177e4 LT |
703 | |
704 | repeat: | |
c5c8be3c | 705 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
706 | if (inode->i_ino != ino) |
707 | continue; | |
708 | if (inode->i_sb != sb) | |
709 | continue; | |
a4ffdde6 | 710 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
711 | __wait_on_freeing_inode(inode); |
712 | goto repeat; | |
713 | } | |
f7899bd5 CH |
714 | __iget(inode); |
715 | return inode; | |
1da177e4 | 716 | } |
f7899bd5 | 717 | return NULL; |
1da177e4 LT |
718 | } |
719 | ||
f991bd2e ED |
720 | /* |
721 | * Each cpu owns a range of LAST_INO_BATCH numbers. | |
722 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, | |
723 | * to renew the exhausted range. | |
724 | * | |
725 | * This does not significantly increase overflow rate because every CPU can | |
726 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is | |
727 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the | |
728 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase | |
729 | * overflow rate by 2x, which does not seem too significant. | |
730 | * | |
731 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
732 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
733 | * here to attempt to avoid that. | |
734 | */ | |
735 | #define LAST_INO_BATCH 1024 | |
736 | static DEFINE_PER_CPU(unsigned int, last_ino); | |
737 | ||
85fe4025 | 738 | unsigned int get_next_ino(void) |
f991bd2e ED |
739 | { |
740 | unsigned int *p = &get_cpu_var(last_ino); | |
741 | unsigned int res = *p; | |
742 | ||
743 | #ifdef CONFIG_SMP | |
744 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { | |
745 | static atomic_t shared_last_ino; | |
746 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); | |
747 | ||
748 | res = next - LAST_INO_BATCH; | |
749 | } | |
750 | #endif | |
751 | ||
752 | *p = ++res; | |
753 | put_cpu_var(last_ino); | |
754 | return res; | |
755 | } | |
85fe4025 | 756 | EXPORT_SYMBOL(get_next_ino); |
f991bd2e | 757 | |
1da177e4 LT |
758 | /** |
759 | * new_inode - obtain an inode | |
760 | * @sb: superblock | |
761 | * | |
769848c0 | 762 | * Allocates a new inode for given superblock. The default gfp_mask |
3c1d4378 | 763 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
769848c0 MG |
764 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
765 | * for the page cache are not reclaimable or migratable, | |
766 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
767 | * newly created inode's mapping | |
768 | * | |
1da177e4 LT |
769 | */ |
770 | struct inode *new_inode(struct super_block *sb) | |
771 | { | |
6b3304b5 | 772 | struct inode *inode; |
1da177e4 LT |
773 | |
774 | spin_lock_prefetch(&inode_lock); | |
6b3304b5 | 775 | |
1da177e4 LT |
776 | inode = alloc_inode(sb); |
777 | if (inode) { | |
778 | spin_lock(&inode_lock); | |
646ec461 | 779 | __inode_sb_list_add(inode); |
1da177e4 LT |
780 | inode->i_state = 0; |
781 | spin_unlock(&inode_lock); | |
782 | } | |
783 | return inode; | |
784 | } | |
1da177e4 LT |
785 | EXPORT_SYMBOL(new_inode); |
786 | ||
787 | void unlock_new_inode(struct inode *inode) | |
788 | { | |
14358e6d | 789 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
a3314a0e | 790 | if (S_ISDIR(inode->i_mode)) { |
1e89a5e1 PZ |
791 | struct file_system_type *type = inode->i_sb->s_type; |
792 | ||
9a7aa12f JK |
793 | /* Set new key only if filesystem hasn't already changed it */ |
794 | if (!lockdep_match_class(&inode->i_mutex, | |
795 | &type->i_mutex_key)) { | |
796 | /* | |
797 | * ensure nobody is actually holding i_mutex | |
798 | */ | |
799 | mutex_destroy(&inode->i_mutex); | |
800 | mutex_init(&inode->i_mutex); | |
801 | lockdep_set_class(&inode->i_mutex, | |
802 | &type->i_mutex_dir_key); | |
803 | } | |
1e89a5e1 | 804 | } |
14358e6d | 805 | #endif |
1da177e4 | 806 | /* |
eaff8079 | 807 | * This is special! We do not need the spinlock when clearing I_NEW, |
580be083 JK |
808 | * because we're guaranteed that nobody else tries to do anything about |
809 | * the state of the inode when it is locked, as we just created it (so | |
eaff8079 | 810 | * there can be no old holders that haven't tested I_NEW). |
580be083 | 811 | * However we must emit the memory barrier so that other CPUs reliably |
eaff8079 | 812 | * see the clearing of I_NEW after the other inode initialisation has |
580be083 | 813 | * completed. |
1da177e4 | 814 | */ |
580be083 | 815 | smp_mb(); |
eaff8079 CH |
816 | WARN_ON(!(inode->i_state & I_NEW)); |
817 | inode->i_state &= ~I_NEW; | |
1da177e4 LT |
818 | wake_up_inode(inode); |
819 | } | |
1da177e4 LT |
820 | EXPORT_SYMBOL(unlock_new_inode); |
821 | ||
822 | /* | |
823 | * This is called without the inode lock held.. Be careful. | |
824 | * | |
825 | * We no longer cache the sb_flags in i_flags - see fs.h | |
826 | * -- rmk@arm.uk.linux.org | |
827 | */ | |
6b3304b5 MK |
828 | static struct inode *get_new_inode(struct super_block *sb, |
829 | struct hlist_head *head, | |
830 | int (*test)(struct inode *, void *), | |
831 | int (*set)(struct inode *, void *), | |
832 | void *data) | |
1da177e4 | 833 | { |
6b3304b5 | 834 | struct inode *inode; |
1da177e4 LT |
835 | |
836 | inode = alloc_inode(sb); | |
837 | if (inode) { | |
6b3304b5 | 838 | struct inode *old; |
1da177e4 LT |
839 | |
840 | spin_lock(&inode_lock); | |
841 | /* We released the lock, so.. */ | |
842 | old = find_inode(sb, head, test, data); | |
843 | if (!old) { | |
844 | if (set(inode, data)) | |
845 | goto set_failed; | |
846 | ||
646ec461 CH |
847 | hlist_add_head(&inode->i_hash, head); |
848 | __inode_sb_list_add(inode); | |
eaff8079 | 849 | inode->i_state = I_NEW; |
1da177e4 LT |
850 | spin_unlock(&inode_lock); |
851 | ||
852 | /* Return the locked inode with I_NEW set, the | |
853 | * caller is responsible for filling in the contents | |
854 | */ | |
855 | return inode; | |
856 | } | |
857 | ||
858 | /* | |
859 | * Uhhuh, somebody else created the same inode under | |
860 | * us. Use the old inode instead of the one we just | |
861 | * allocated. | |
862 | */ | |
1da177e4 LT |
863 | spin_unlock(&inode_lock); |
864 | destroy_inode(inode); | |
865 | inode = old; | |
866 | wait_on_inode(inode); | |
867 | } | |
868 | return inode; | |
869 | ||
870 | set_failed: | |
871 | spin_unlock(&inode_lock); | |
872 | destroy_inode(inode); | |
873 | return NULL; | |
874 | } | |
875 | ||
876 | /* | |
877 | * get_new_inode_fast is the fast path version of get_new_inode, see the | |
878 | * comment at iget_locked for details. | |
879 | */ | |
6b3304b5 MK |
880 | static struct inode *get_new_inode_fast(struct super_block *sb, |
881 | struct hlist_head *head, unsigned long ino) | |
1da177e4 | 882 | { |
6b3304b5 | 883 | struct inode *inode; |
1da177e4 LT |
884 | |
885 | inode = alloc_inode(sb); | |
886 | if (inode) { | |
6b3304b5 | 887 | struct inode *old; |
1da177e4 LT |
888 | |
889 | spin_lock(&inode_lock); | |
890 | /* We released the lock, so.. */ | |
891 | old = find_inode_fast(sb, head, ino); | |
892 | if (!old) { | |
893 | inode->i_ino = ino; | |
646ec461 CH |
894 | hlist_add_head(&inode->i_hash, head); |
895 | __inode_sb_list_add(inode); | |
eaff8079 | 896 | inode->i_state = I_NEW; |
1da177e4 LT |
897 | spin_unlock(&inode_lock); |
898 | ||
899 | /* Return the locked inode with I_NEW set, the | |
900 | * caller is responsible for filling in the contents | |
901 | */ | |
902 | return inode; | |
903 | } | |
904 | ||
905 | /* | |
906 | * Uhhuh, somebody else created the same inode under | |
907 | * us. Use the old inode instead of the one we just | |
908 | * allocated. | |
909 | */ | |
1da177e4 LT |
910 | spin_unlock(&inode_lock); |
911 | destroy_inode(inode); | |
912 | inode = old; | |
913 | wait_on_inode(inode); | |
914 | } | |
915 | return inode; | |
916 | } | |
917 | ||
ad5e195a CH |
918 | /* |
919 | * search the inode cache for a matching inode number. | |
920 | * If we find one, then the inode number we are trying to | |
921 | * allocate is not unique and so we should not use it. | |
922 | * | |
923 | * Returns 1 if the inode number is unique, 0 if it is not. | |
924 | */ | |
925 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) | |
926 | { | |
927 | struct hlist_head *b = inode_hashtable + hash(sb, ino); | |
928 | struct hlist_node *node; | |
929 | struct inode *inode; | |
930 | ||
931 | hlist_for_each_entry(inode, node, b, i_hash) { | |
932 | if (inode->i_ino == ino && inode->i_sb == sb) | |
933 | return 0; | |
934 | } | |
935 | ||
936 | return 1; | |
937 | } | |
938 | ||
1da177e4 LT |
939 | /** |
940 | * iunique - get a unique inode number | |
941 | * @sb: superblock | |
942 | * @max_reserved: highest reserved inode number | |
943 | * | |
944 | * Obtain an inode number that is unique on the system for a given | |
945 | * superblock. This is used by file systems that have no natural | |
946 | * permanent inode numbering system. An inode number is returned that | |
947 | * is higher than the reserved limit but unique. | |
948 | * | |
949 | * BUGS: | |
950 | * With a large number of inodes live on the file system this function | |
951 | * currently becomes quite slow. | |
952 | */ | |
953 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
954 | { | |
866b04fc JL |
955 | /* |
956 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
957 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
958 | * here to attempt to avoid that. | |
959 | */ | |
ad5e195a | 960 | static DEFINE_SPINLOCK(iunique_lock); |
866b04fc | 961 | static unsigned int counter; |
1da177e4 | 962 | ino_t res; |
3361c7be | 963 | |
1da177e4 | 964 | spin_lock(&inode_lock); |
ad5e195a | 965 | spin_lock(&iunique_lock); |
3361c7be JL |
966 | do { |
967 | if (counter <= max_reserved) | |
968 | counter = max_reserved + 1; | |
1da177e4 | 969 | res = counter++; |
ad5e195a CH |
970 | } while (!test_inode_iunique(sb, res)); |
971 | spin_unlock(&iunique_lock); | |
3361c7be | 972 | spin_unlock(&inode_lock); |
1da177e4 | 973 | |
3361c7be JL |
974 | return res; |
975 | } | |
1da177e4 LT |
976 | EXPORT_SYMBOL(iunique); |
977 | ||
978 | struct inode *igrab(struct inode *inode) | |
979 | { | |
980 | spin_lock(&inode_lock); | |
a4ffdde6 | 981 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) |
1da177e4 LT |
982 | __iget(inode); |
983 | else | |
984 | /* | |
985 | * Handle the case where s_op->clear_inode is not been | |
986 | * called yet, and somebody is calling igrab | |
987 | * while the inode is getting freed. | |
988 | */ | |
989 | inode = NULL; | |
990 | spin_unlock(&inode_lock); | |
991 | return inode; | |
992 | } | |
1da177e4 LT |
993 | EXPORT_SYMBOL(igrab); |
994 | ||
995 | /** | |
996 | * ifind - internal function, you want ilookup5() or iget5(). | |
997 | * @sb: super block of file system to search | |
998 | * @head: the head of the list to search | |
999 | * @test: callback used for comparisons between inodes | |
1000 | * @data: opaque data pointer to pass to @test | |
88bd5121 | 1001 | * @wait: if true wait for the inode to be unlocked, if false do not |
1da177e4 LT |
1002 | * |
1003 | * ifind() searches for the inode specified by @data in the inode | |
1004 | * cache. This is a generalized version of ifind_fast() for file systems where | |
1005 | * the inode number is not sufficient for unique identification of an inode. | |
1006 | * | |
1007 | * If the inode is in the cache, the inode is returned with an incremented | |
1008 | * reference count. | |
1009 | * | |
1010 | * Otherwise NULL is returned. | |
1011 | * | |
1012 | * Note, @test is called with the inode_lock held, so can't sleep. | |
1013 | */ | |
5d2bea45 | 1014 | static struct inode *ifind(struct super_block *sb, |
1da177e4 | 1015 | struct hlist_head *head, int (*test)(struct inode *, void *), |
88bd5121 | 1016 | void *data, const int wait) |
1da177e4 LT |
1017 | { |
1018 | struct inode *inode; | |
1019 | ||
1020 | spin_lock(&inode_lock); | |
1021 | inode = find_inode(sb, head, test, data); | |
1022 | if (inode) { | |
1da177e4 | 1023 | spin_unlock(&inode_lock); |
88bd5121 AA |
1024 | if (likely(wait)) |
1025 | wait_on_inode(inode); | |
1da177e4 LT |
1026 | return inode; |
1027 | } | |
1028 | spin_unlock(&inode_lock); | |
1029 | return NULL; | |
1030 | } | |
1031 | ||
1032 | /** | |
1033 | * ifind_fast - internal function, you want ilookup() or iget(). | |
1034 | * @sb: super block of file system to search | |
1035 | * @head: head of the list to search | |
1036 | * @ino: inode number to search for | |
1037 | * | |
1038 | * ifind_fast() searches for the inode @ino in the inode cache. This is for | |
1039 | * file systems where the inode number is sufficient for unique identification | |
1040 | * of an inode. | |
1041 | * | |
1042 | * If the inode is in the cache, the inode is returned with an incremented | |
1043 | * reference count. | |
1044 | * | |
1045 | * Otherwise NULL is returned. | |
1046 | */ | |
5d2bea45 | 1047 | static struct inode *ifind_fast(struct super_block *sb, |
1da177e4 LT |
1048 | struct hlist_head *head, unsigned long ino) |
1049 | { | |
1050 | struct inode *inode; | |
1051 | ||
1052 | spin_lock(&inode_lock); | |
1053 | inode = find_inode_fast(sb, head, ino); | |
1054 | if (inode) { | |
1da177e4 LT |
1055 | spin_unlock(&inode_lock); |
1056 | wait_on_inode(inode); | |
1057 | return inode; | |
1058 | } | |
1059 | spin_unlock(&inode_lock); | |
1060 | return NULL; | |
1061 | } | |
1062 | ||
1063 | /** | |
88bd5121 | 1064 | * ilookup5_nowait - search for an inode in the inode cache |
1da177e4 LT |
1065 | * @sb: super block of file system to search |
1066 | * @hashval: hash value (usually inode number) to search for | |
1067 | * @test: callback used for comparisons between inodes | |
1068 | * @data: opaque data pointer to pass to @test | |
1069 | * | |
1070 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
1071 | * @data in the inode cache. This is a generalized version of ilookup() for | |
1072 | * file systems where the inode number is not sufficient for unique | |
1073 | * identification of an inode. | |
1074 | * | |
1075 | * If the inode is in the cache, the inode is returned with an incremented | |
88bd5121 AA |
1076 | * reference count. Note, the inode lock is not waited upon so you have to be |
1077 | * very careful what you do with the returned inode. You probably should be | |
1078 | * using ilookup5() instead. | |
1079 | * | |
1080 | * Otherwise NULL is returned. | |
1081 | * | |
1082 | * Note, @test is called with the inode_lock held, so can't sleep. | |
1083 | */ | |
1084 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, | |
1085 | int (*test)(struct inode *, void *), void *data) | |
1086 | { | |
1087 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1088 | ||
1089 | return ifind(sb, head, test, data, 0); | |
1090 | } | |
88bd5121 AA |
1091 | EXPORT_SYMBOL(ilookup5_nowait); |
1092 | ||
1093 | /** | |
1094 | * ilookup5 - search for an inode in the inode cache | |
1095 | * @sb: super block of file system to search | |
1096 | * @hashval: hash value (usually inode number) to search for | |
1097 | * @test: callback used for comparisons between inodes | |
1098 | * @data: opaque data pointer to pass to @test | |
1099 | * | |
1100 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
1101 | * @data in the inode cache. This is a generalized version of ilookup() for | |
1102 | * file systems where the inode number is not sufficient for unique | |
1103 | * identification of an inode. | |
1104 | * | |
1105 | * If the inode is in the cache, the inode lock is waited upon and the inode is | |
1106 | * returned with an incremented reference count. | |
1da177e4 LT |
1107 | * |
1108 | * Otherwise NULL is returned. | |
1109 | * | |
1110 | * Note, @test is called with the inode_lock held, so can't sleep. | |
1111 | */ | |
1112 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
1113 | int (*test)(struct inode *, void *), void *data) | |
1114 | { | |
1115 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1116 | ||
88bd5121 | 1117 | return ifind(sb, head, test, data, 1); |
1da177e4 | 1118 | } |
1da177e4 LT |
1119 | EXPORT_SYMBOL(ilookup5); |
1120 | ||
1121 | /** | |
1122 | * ilookup - search for an inode in the inode cache | |
1123 | * @sb: super block of file system to search | |
1124 | * @ino: inode number to search for | |
1125 | * | |
1126 | * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache. | |
1127 | * This is for file systems where the inode number is sufficient for unique | |
1128 | * identification of an inode. | |
1129 | * | |
1130 | * If the inode is in the cache, the inode is returned with an incremented | |
1131 | * reference count. | |
1132 | * | |
1133 | * Otherwise NULL is returned. | |
1134 | */ | |
1135 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1136 | { | |
1137 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1138 | ||
1139 | return ifind_fast(sb, head, ino); | |
1140 | } | |
1da177e4 LT |
1141 | EXPORT_SYMBOL(ilookup); |
1142 | ||
1143 | /** | |
1144 | * iget5_locked - obtain an inode from a mounted file system | |
1145 | * @sb: super block of file system | |
1146 | * @hashval: hash value (usually inode number) to get | |
1147 | * @test: callback used for comparisons between inodes | |
1148 | * @set: callback used to initialize a new struct inode | |
1149 | * @data: opaque data pointer to pass to @test and @set | |
1150 | * | |
1da177e4 LT |
1151 | * iget5_locked() uses ifind() to search for the inode specified by @hashval |
1152 | * and @data in the inode cache and if present it is returned with an increased | |
1153 | * reference count. This is a generalized version of iget_locked() for file | |
1154 | * systems where the inode number is not sufficient for unique identification | |
1155 | * of an inode. | |
1156 | * | |
1157 | * If the inode is not in cache, get_new_inode() is called to allocate a new | |
1158 | * inode and this is returned locked, hashed, and with the I_NEW flag set. The | |
1159 | * file system gets to fill it in before unlocking it via unlock_new_inode(). | |
1160 | * | |
1161 | * Note both @test and @set are called with the inode_lock held, so can't sleep. | |
1162 | */ | |
1163 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, | |
1164 | int (*test)(struct inode *, void *), | |
1165 | int (*set)(struct inode *, void *), void *data) | |
1166 | { | |
1167 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1168 | struct inode *inode; | |
1169 | ||
88bd5121 | 1170 | inode = ifind(sb, head, test, data, 1); |
1da177e4 LT |
1171 | if (inode) |
1172 | return inode; | |
1173 | /* | |
1174 | * get_new_inode() will do the right thing, re-trying the search | |
1175 | * in case it had to block at any point. | |
1176 | */ | |
1177 | return get_new_inode(sb, head, test, set, data); | |
1178 | } | |
1da177e4 LT |
1179 | EXPORT_SYMBOL(iget5_locked); |
1180 | ||
1181 | /** | |
1182 | * iget_locked - obtain an inode from a mounted file system | |
1183 | * @sb: super block of file system | |
1184 | * @ino: inode number to get | |
1185 | * | |
1da177e4 LT |
1186 | * iget_locked() uses ifind_fast() to search for the inode specified by @ino in |
1187 | * the inode cache and if present it is returned with an increased reference | |
1188 | * count. This is for file systems where the inode number is sufficient for | |
1189 | * unique identification of an inode. | |
1190 | * | |
1191 | * If the inode is not in cache, get_new_inode_fast() is called to allocate a | |
1192 | * new inode and this is returned locked, hashed, and with the I_NEW flag set. | |
1193 | * The file system gets to fill it in before unlocking it via | |
1194 | * unlock_new_inode(). | |
1195 | */ | |
1196 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) | |
1197 | { | |
1198 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1199 | struct inode *inode; | |
1200 | ||
1201 | inode = ifind_fast(sb, head, ino); | |
1202 | if (inode) | |
1203 | return inode; | |
1204 | /* | |
1205 | * get_new_inode_fast() will do the right thing, re-trying the search | |
1206 | * in case it had to block at any point. | |
1207 | */ | |
1208 | return get_new_inode_fast(sb, head, ino); | |
1209 | } | |
1da177e4 LT |
1210 | EXPORT_SYMBOL(iget_locked); |
1211 | ||
261bca86 AV |
1212 | int insert_inode_locked(struct inode *inode) |
1213 | { | |
1214 | struct super_block *sb = inode->i_sb; | |
1215 | ino_t ino = inode->i_ino; | |
1216 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
261bca86 | 1217 | |
eaff8079 | 1218 | inode->i_state |= I_NEW; |
261bca86 | 1219 | while (1) { |
72a43d63 AV |
1220 | struct hlist_node *node; |
1221 | struct inode *old = NULL; | |
261bca86 | 1222 | spin_lock(&inode_lock); |
72a43d63 AV |
1223 | hlist_for_each_entry(old, node, head, i_hash) { |
1224 | if (old->i_ino != ino) | |
1225 | continue; | |
1226 | if (old->i_sb != sb) | |
1227 | continue; | |
a4ffdde6 | 1228 | if (old->i_state & (I_FREEING|I_WILL_FREE)) |
72a43d63 AV |
1229 | continue; |
1230 | break; | |
1231 | } | |
1232 | if (likely(!node)) { | |
261bca86 AV |
1233 | hlist_add_head(&inode->i_hash, head); |
1234 | spin_unlock(&inode_lock); | |
1235 | return 0; | |
1236 | } | |
1237 | __iget(old); | |
1238 | spin_unlock(&inode_lock); | |
1239 | wait_on_inode(old); | |
1d3382cb | 1240 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1241 | iput(old); |
1242 | return -EBUSY; | |
1243 | } | |
1244 | iput(old); | |
1245 | } | |
1246 | } | |
261bca86 AV |
1247 | EXPORT_SYMBOL(insert_inode_locked); |
1248 | ||
1249 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1250 | int (*test)(struct inode *, void *), void *data) | |
1251 | { | |
1252 | struct super_block *sb = inode->i_sb; | |
1253 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
261bca86 | 1254 | |
eaff8079 | 1255 | inode->i_state |= I_NEW; |
261bca86 AV |
1256 | |
1257 | while (1) { | |
72a43d63 AV |
1258 | struct hlist_node *node; |
1259 | struct inode *old = NULL; | |
1260 | ||
261bca86 | 1261 | spin_lock(&inode_lock); |
72a43d63 AV |
1262 | hlist_for_each_entry(old, node, head, i_hash) { |
1263 | if (old->i_sb != sb) | |
1264 | continue; | |
1265 | if (!test(old, data)) | |
1266 | continue; | |
a4ffdde6 | 1267 | if (old->i_state & (I_FREEING|I_WILL_FREE)) |
72a43d63 AV |
1268 | continue; |
1269 | break; | |
1270 | } | |
1271 | if (likely(!node)) { | |
261bca86 AV |
1272 | hlist_add_head(&inode->i_hash, head); |
1273 | spin_unlock(&inode_lock); | |
1274 | return 0; | |
1275 | } | |
1276 | __iget(old); | |
1277 | spin_unlock(&inode_lock); | |
1278 | wait_on_inode(old); | |
1d3382cb | 1279 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1280 | iput(old); |
1281 | return -EBUSY; | |
1282 | } | |
1283 | iput(old); | |
1284 | } | |
1285 | } | |
261bca86 AV |
1286 | EXPORT_SYMBOL(insert_inode_locked4); |
1287 | ||
1da177e4 | 1288 | |
45321ac5 AV |
1289 | int generic_delete_inode(struct inode *inode) |
1290 | { | |
1291 | return 1; | |
1292 | } | |
1293 | EXPORT_SYMBOL(generic_delete_inode); | |
1294 | ||
1da177e4 | 1295 | /* |
45321ac5 AV |
1296 | * Normal UNIX filesystem behaviour: delete the |
1297 | * inode when the usage count drops to zero, and | |
1298 | * i_nlink is zero. | |
1da177e4 | 1299 | */ |
45321ac5 | 1300 | int generic_drop_inode(struct inode *inode) |
1da177e4 | 1301 | { |
1d3382cb | 1302 | return !inode->i_nlink || inode_unhashed(inode); |
1da177e4 | 1303 | } |
45321ac5 | 1304 | EXPORT_SYMBOL_GPL(generic_drop_inode); |
1da177e4 | 1305 | |
45321ac5 AV |
1306 | /* |
1307 | * Called when we're dropping the last reference | |
1308 | * to an inode. | |
22fe4042 | 1309 | * |
45321ac5 AV |
1310 | * Call the FS "drop_inode()" function, defaulting to |
1311 | * the legacy UNIX filesystem behaviour. If it tells | |
1312 | * us to evict inode, do so. Otherwise, retain inode | |
1313 | * in cache if fs is alive, sync and evict if fs is | |
1314 | * shutting down. | |
22fe4042 | 1315 | */ |
45321ac5 | 1316 | static void iput_final(struct inode *inode) |
1da177e4 LT |
1317 | { |
1318 | struct super_block *sb = inode->i_sb; | |
45321ac5 AV |
1319 | const struct super_operations *op = inode->i_sb->s_op; |
1320 | int drop; | |
1321 | ||
1322 | if (op && op->drop_inode) | |
1323 | drop = op->drop_inode(inode); | |
1324 | else | |
1325 | drop = generic_drop_inode(inode); | |
1da177e4 | 1326 | |
45321ac5 | 1327 | if (!drop) { |
acb0c854 | 1328 | if (sb->s_flags & MS_ACTIVE) { |
9e38d86f NP |
1329 | inode->i_state |= I_REFERENCED; |
1330 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) { | |
1331 | inode_lru_list_add(inode); | |
1332 | } | |
991114c6 | 1333 | spin_unlock(&inode_lock); |
45321ac5 | 1334 | return; |
991114c6 | 1335 | } |
7ef0d737 | 1336 | WARN_ON(inode->i_state & I_NEW); |
991114c6 AV |
1337 | inode->i_state |= I_WILL_FREE; |
1338 | spin_unlock(&inode_lock); | |
1da177e4 LT |
1339 | write_inode_now(inode, 1); |
1340 | spin_lock(&inode_lock); | |
7ef0d737 | 1341 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1342 | inode->i_state &= ~I_WILL_FREE; |
4c51acbc | 1343 | __remove_inode_hash(inode); |
1da177e4 | 1344 | } |
7ef0d737 | 1345 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1346 | inode->i_state |= I_FREEING; |
9e38d86f NP |
1347 | |
1348 | /* | |
1349 | * After we delete the inode from the LRU here, we avoid moving dirty | |
1350 | * inodes back onto the LRU now because I_FREEING is set and hence | |
1351 | * writeback_single_inode() won't move the inode around. | |
1352 | */ | |
1353 | inode_lru_list_del(inode); | |
1354 | ||
646ec461 | 1355 | __inode_sb_list_del(inode); |
1da177e4 | 1356 | spin_unlock(&inode_lock); |
644da596 | 1357 | evict(inode); |
4c51acbc | 1358 | remove_inode_hash(inode); |
7f04c26d | 1359 | wake_up_inode(inode); |
45321ac5 | 1360 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); |
1da177e4 LT |
1361 | destroy_inode(inode); |
1362 | } | |
1363 | ||
1da177e4 | 1364 | /** |
6b3304b5 | 1365 | * iput - put an inode |
1da177e4 LT |
1366 | * @inode: inode to put |
1367 | * | |
1368 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1369 | * zero, the inode is then freed and may also be destroyed. | |
1370 | * | |
1371 | * Consequently, iput() can sleep. | |
1372 | */ | |
1373 | void iput(struct inode *inode) | |
1374 | { | |
1375 | if (inode) { | |
a4ffdde6 | 1376 | BUG_ON(inode->i_state & I_CLEAR); |
1da177e4 | 1377 | |
1da177e4 LT |
1378 | if (atomic_dec_and_lock(&inode->i_count, &inode_lock)) |
1379 | iput_final(inode); | |
1380 | } | |
1381 | } | |
1da177e4 LT |
1382 | EXPORT_SYMBOL(iput); |
1383 | ||
1384 | /** | |
1385 | * bmap - find a block number in a file | |
1386 | * @inode: inode of file | |
1387 | * @block: block to find | |
1388 | * | |
1389 | * Returns the block number on the device holding the inode that | |
1390 | * is the disk block number for the block of the file requested. | |
1391 | * That is, asked for block 4 of inode 1 the function will return the | |
6b3304b5 | 1392 | * disk block relative to the disk start that holds that block of the |
1da177e4 LT |
1393 | * file. |
1394 | */ | |
6b3304b5 | 1395 | sector_t bmap(struct inode *inode, sector_t block) |
1da177e4 LT |
1396 | { |
1397 | sector_t res = 0; | |
1398 | if (inode->i_mapping->a_ops->bmap) | |
1399 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1400 | return res; | |
1401 | } | |
1da177e4 LT |
1402 | EXPORT_SYMBOL(bmap); |
1403 | ||
11ff6f05 MG |
1404 | /* |
1405 | * With relative atime, only update atime if the previous atime is | |
1406 | * earlier than either the ctime or mtime or if at least a day has | |
1407 | * passed since the last atime update. | |
1408 | */ | |
1409 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1410 | struct timespec now) | |
1411 | { | |
1412 | ||
1413 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1414 | return 1; | |
1415 | /* | |
1416 | * Is mtime younger than atime? If yes, update atime: | |
1417 | */ | |
1418 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1419 | return 1; | |
1420 | /* | |
1421 | * Is ctime younger than atime? If yes, update atime: | |
1422 | */ | |
1423 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1424 | return 1; | |
1425 | ||
1426 | /* | |
1427 | * Is the previous atime value older than a day? If yes, | |
1428 | * update atime: | |
1429 | */ | |
1430 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1431 | return 1; | |
1432 | /* | |
1433 | * Good, we can skip the atime update: | |
1434 | */ | |
1435 | return 0; | |
1436 | } | |
1437 | ||
1da177e4 | 1438 | /** |
869243a0 CH |
1439 | * touch_atime - update the access time |
1440 | * @mnt: mount the inode is accessed on | |
7045f37b | 1441 | * @dentry: dentry accessed |
1da177e4 LT |
1442 | * |
1443 | * Update the accessed time on an inode and mark it for writeback. | |
1444 | * This function automatically handles read only file systems and media, | |
1445 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1446 | */ | |
869243a0 | 1447 | void touch_atime(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 1448 | { |
869243a0 | 1449 | struct inode *inode = dentry->d_inode; |
1da177e4 LT |
1450 | struct timespec now; |
1451 | ||
cdb70f3f | 1452 | if (inode->i_flags & S_NOATIME) |
b12536c2 | 1453 | return; |
37756ced | 1454 | if (IS_NOATIME(inode)) |
b12536c2 | 1455 | return; |
b2276138 | 1456 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1457 | return; |
47ae32d6 | 1458 | |
cdb70f3f | 1459 | if (mnt->mnt_flags & MNT_NOATIME) |
b12536c2 | 1460 | return; |
cdb70f3f | 1461 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1462 | return; |
1da177e4 LT |
1463 | |
1464 | now = current_fs_time(inode->i_sb); | |
11ff6f05 MG |
1465 | |
1466 | if (!relatime_need_update(mnt, inode, now)) | |
b12536c2 | 1467 | return; |
11ff6f05 | 1468 | |
47ae32d6 | 1469 | if (timespec_equal(&inode->i_atime, &now)) |
b12536c2 AK |
1470 | return; |
1471 | ||
1472 | if (mnt_want_write(mnt)) | |
1473 | return; | |
47ae32d6 VH |
1474 | |
1475 | inode->i_atime = now; | |
1476 | mark_inode_dirty_sync(inode); | |
cdb70f3f | 1477 | mnt_drop_write(mnt); |
1da177e4 | 1478 | } |
869243a0 | 1479 | EXPORT_SYMBOL(touch_atime); |
1da177e4 LT |
1480 | |
1481 | /** | |
870f4817 CH |
1482 | * file_update_time - update mtime and ctime time |
1483 | * @file: file accessed | |
1da177e4 | 1484 | * |
870f4817 CH |
1485 | * Update the mtime and ctime members of an inode and mark the inode |
1486 | * for writeback. Note that this function is meant exclusively for | |
1487 | * usage in the file write path of filesystems, and filesystems may | |
1488 | * choose to explicitly ignore update via this function with the | |
2eadfc0e | 1489 | * S_NOCMTIME inode flag, e.g. for network filesystem where these |
870f4817 | 1490 | * timestamps are handled by the server. |
1da177e4 LT |
1491 | */ |
1492 | ||
870f4817 | 1493 | void file_update_time(struct file *file) |
1da177e4 | 1494 | { |
0f7fc9e4 | 1495 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 | 1496 | struct timespec now; |
ce06e0b2 | 1497 | enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0; |
1da177e4 | 1498 | |
ce06e0b2 | 1499 | /* First try to exhaust all avenues to not sync */ |
1da177e4 LT |
1500 | if (IS_NOCMTIME(inode)) |
1501 | return; | |
20ddee2c | 1502 | |
1da177e4 | 1503 | now = current_fs_time(inode->i_sb); |
ce06e0b2 AK |
1504 | if (!timespec_equal(&inode->i_mtime, &now)) |
1505 | sync_it = S_MTIME; | |
1da177e4 | 1506 | |
ce06e0b2 AK |
1507 | if (!timespec_equal(&inode->i_ctime, &now)) |
1508 | sync_it |= S_CTIME; | |
870f4817 | 1509 | |
ce06e0b2 AK |
1510 | if (IS_I_VERSION(inode)) |
1511 | sync_it |= S_VERSION; | |
7a224228 | 1512 | |
ce06e0b2 AK |
1513 | if (!sync_it) |
1514 | return; | |
1515 | ||
1516 | /* Finally allowed to write? Takes lock. */ | |
1517 | if (mnt_want_write_file(file)) | |
1518 | return; | |
1519 | ||
1520 | /* Only change inode inside the lock region */ | |
1521 | if (sync_it & S_VERSION) | |
1522 | inode_inc_iversion(inode); | |
1523 | if (sync_it & S_CTIME) | |
1524 | inode->i_ctime = now; | |
1525 | if (sync_it & S_MTIME) | |
1526 | inode->i_mtime = now; | |
1527 | mark_inode_dirty_sync(inode); | |
20ddee2c | 1528 | mnt_drop_write(file->f_path.mnt); |
1da177e4 | 1529 | } |
870f4817 | 1530 | EXPORT_SYMBOL(file_update_time); |
1da177e4 LT |
1531 | |
1532 | int inode_needs_sync(struct inode *inode) | |
1533 | { | |
1534 | if (IS_SYNC(inode)) | |
1535 | return 1; | |
1536 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1537 | return 1; | |
1538 | return 0; | |
1539 | } | |
1da177e4 LT |
1540 | EXPORT_SYMBOL(inode_needs_sync); |
1541 | ||
1da177e4 LT |
1542 | int inode_wait(void *word) |
1543 | { | |
1544 | schedule(); | |
1545 | return 0; | |
1546 | } | |
d44dab8d | 1547 | EXPORT_SYMBOL(inode_wait); |
1da177e4 LT |
1548 | |
1549 | /* | |
168a9fd6 MS |
1550 | * If we try to find an inode in the inode hash while it is being |
1551 | * deleted, we have to wait until the filesystem completes its | |
1552 | * deletion before reporting that it isn't found. This function waits | |
1553 | * until the deletion _might_ have completed. Callers are responsible | |
1554 | * to recheck inode state. | |
1555 | * | |
eaff8079 | 1556 | * It doesn't matter if I_NEW is not set initially, a call to |
168a9fd6 MS |
1557 | * wake_up_inode() after removing from the hash list will DTRT. |
1558 | * | |
1da177e4 LT |
1559 | * This is called with inode_lock held. |
1560 | */ | |
1561 | static void __wait_on_freeing_inode(struct inode *inode) | |
1562 | { | |
1563 | wait_queue_head_t *wq; | |
eaff8079 CH |
1564 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); |
1565 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1da177e4 LT |
1566 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
1567 | spin_unlock(&inode_lock); | |
1568 | schedule(); | |
1569 | finish_wait(wq, &wait.wait); | |
1570 | spin_lock(&inode_lock); | |
1571 | } | |
1572 | ||
1da177e4 LT |
1573 | static __initdata unsigned long ihash_entries; |
1574 | static int __init set_ihash_entries(char *str) | |
1575 | { | |
1576 | if (!str) | |
1577 | return 0; | |
1578 | ihash_entries = simple_strtoul(str, &str, 0); | |
1579 | return 1; | |
1580 | } | |
1581 | __setup("ihash_entries=", set_ihash_entries); | |
1582 | ||
1583 | /* | |
1584 | * Initialize the waitqueues and inode hash table. | |
1585 | */ | |
1586 | void __init inode_init_early(void) | |
1587 | { | |
1588 | int loop; | |
1589 | ||
1590 | /* If hashes are distributed across NUMA nodes, defer | |
1591 | * hash allocation until vmalloc space is available. | |
1592 | */ | |
1593 | if (hashdist) | |
1594 | return; | |
1595 | ||
1596 | inode_hashtable = | |
1597 | alloc_large_system_hash("Inode-cache", | |
1598 | sizeof(struct hlist_head), | |
1599 | ihash_entries, | |
1600 | 14, | |
1601 | HASH_EARLY, | |
1602 | &i_hash_shift, | |
1603 | &i_hash_mask, | |
1604 | 0); | |
1605 | ||
1606 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1607 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1608 | } | |
1609 | ||
74bf17cf | 1610 | void __init inode_init(void) |
1da177e4 LT |
1611 | { |
1612 | int loop; | |
1613 | ||
1614 | /* inode slab cache */ | |
b0196009 PJ |
1615 | inode_cachep = kmem_cache_create("inode_cache", |
1616 | sizeof(struct inode), | |
1617 | 0, | |
1618 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1619 | SLAB_MEM_SPREAD), | |
20c2df83 | 1620 | init_once); |
8e1f936b | 1621 | register_shrinker(&icache_shrinker); |
cffbc8aa DC |
1622 | percpu_counter_init(&nr_inodes, 0); |
1623 | percpu_counter_init(&nr_inodes_unused, 0); | |
1da177e4 LT |
1624 | |
1625 | /* Hash may have been set up in inode_init_early */ | |
1626 | if (!hashdist) | |
1627 | return; | |
1628 | ||
1629 | inode_hashtable = | |
1630 | alloc_large_system_hash("Inode-cache", | |
1631 | sizeof(struct hlist_head), | |
1632 | ihash_entries, | |
1633 | 14, | |
1634 | 0, | |
1635 | &i_hash_shift, | |
1636 | &i_hash_mask, | |
1637 | 0); | |
1638 | ||
1639 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1640 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1641 | } | |
1642 | ||
1643 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1644 | { | |
1645 | inode->i_mode = mode; | |
1646 | if (S_ISCHR(mode)) { | |
1647 | inode->i_fop = &def_chr_fops; | |
1648 | inode->i_rdev = rdev; | |
1649 | } else if (S_ISBLK(mode)) { | |
1650 | inode->i_fop = &def_blk_fops; | |
1651 | inode->i_rdev = rdev; | |
1652 | } else if (S_ISFIFO(mode)) | |
1653 | inode->i_fop = &def_fifo_fops; | |
1654 | else if (S_ISSOCK(mode)) | |
1655 | inode->i_fop = &bad_sock_fops; | |
1656 | else | |
af0d9ae8 MK |
1657 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" |
1658 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1659 | inode->i_ino); | |
1da177e4 LT |
1660 | } |
1661 | EXPORT_SYMBOL(init_special_inode); | |
a1bd120d DM |
1662 | |
1663 | /** | |
1664 | * Init uid,gid,mode for new inode according to posix standards | |
1665 | * @inode: New inode | |
1666 | * @dir: Directory inode | |
1667 | * @mode: mode of the new inode | |
1668 | */ | |
1669 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
1670 | mode_t mode) | |
1671 | { | |
1672 | inode->i_uid = current_fsuid(); | |
1673 | if (dir && dir->i_mode & S_ISGID) { | |
1674 | inode->i_gid = dir->i_gid; | |
1675 | if (S_ISDIR(mode)) | |
1676 | mode |= S_ISGID; | |
1677 | } else | |
1678 | inode->i_gid = current_fsgid(); | |
1679 | inode->i_mode = mode; | |
1680 | } | |
1681 | EXPORT_SYMBOL(inode_init_owner); |