]>
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> |
a178d202 | 27 | #include <linux/ima.h> |
e795b717 | 28 | #include <linux/cred.h> |
1da177e4 | 29 | |
250df6ed DC |
30 | /* |
31 | * inode locking rules. | |
32 | * | |
33 | * inode->i_lock protects: | |
34 | * inode->i_state, inode->i_hash, __iget() | |
02afc410 DC |
35 | * inode_lru_lock protects: |
36 | * inode_lru, inode->i_lru | |
55fa6091 DC |
37 | * inode_sb_list_lock protects: |
38 | * sb->s_inodes, inode->i_sb_list | |
250df6ed DC |
39 | * |
40 | * Lock ordering: | |
41 | * inode_lock | |
42 | * inode->i_lock | |
55fa6091 DC |
43 | * |
44 | * inode_sb_list_lock | |
45 | * inode->i_lock | |
02afc410 | 46 | * inode_lru_lock |
250df6ed DC |
47 | */ |
48 | ||
1da177e4 LT |
49 | /* |
50 | * This is needed for the following functions: | |
51 | * - inode_has_buffers | |
1da177e4 LT |
52 | * - invalidate_bdev |
53 | * | |
54 | * FIXME: remove all knowledge of the buffer layer from this file | |
55 | */ | |
56 | #include <linux/buffer_head.h> | |
57 | ||
58 | /* | |
59 | * New inode.c implementation. | |
60 | * | |
61 | * This implementation has the basic premise of trying | |
62 | * to be extremely low-overhead and SMP-safe, yet be | |
63 | * simple enough to be "obviously correct". | |
64 | * | |
65 | * Famous last words. | |
66 | */ | |
67 | ||
68 | /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */ | |
69 | ||
70 | /* #define INODE_PARANOIA 1 */ | |
71 | /* #define INODE_DEBUG 1 */ | |
72 | ||
73 | /* | |
74 | * Inode lookup is no longer as critical as it used to be: | |
75 | * most of the lookups are going to be through the dcache. | |
76 | */ | |
77 | #define I_HASHBITS i_hash_shift | |
78 | #define I_HASHMASK i_hash_mask | |
79 | ||
fa3536cc ED |
80 | static unsigned int i_hash_mask __read_mostly; |
81 | static unsigned int i_hash_shift __read_mostly; | |
1da177e4 LT |
82 | |
83 | /* | |
84 | * Each inode can be on two separate lists. One is | |
85 | * the hash list of the inode, used for lookups. The | |
86 | * other linked list is the "type" list: | |
87 | * "in_use" - valid inode, i_count > 0, i_nlink > 0 | |
88 | * "dirty" - as "in_use" but also dirty | |
89 | * "unused" - valid inode, i_count = 0 | |
90 | * | |
91 | * A "dirty" list is maintained for each super block, | |
92 | * allowing for low-overhead inode sync() operations. | |
93 | */ | |
94 | ||
7ccf19a8 | 95 | static LIST_HEAD(inode_lru); |
02afc410 | 96 | static DEFINE_SPINLOCK(inode_lru_lock); |
fa3536cc | 97 | static struct hlist_head *inode_hashtable __read_mostly; |
1da177e4 LT |
98 | |
99 | /* | |
100 | * A simple spinlock to protect the list manipulations. | |
101 | * | |
102 | * NOTE! You also have to own the lock if you change | |
103 | * the i_state of an inode while it is in use.. | |
104 | */ | |
105 | DEFINE_SPINLOCK(inode_lock); | |
106 | ||
55fa6091 DC |
107 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock); |
108 | ||
1da177e4 | 109 | /* |
bab1d944 CH |
110 | * iprune_sem provides exclusion between the icache shrinking and the |
111 | * umount path. | |
88e0fbc4 | 112 | * |
bab1d944 CH |
113 | * We don't actually need it to protect anything in the umount path, |
114 | * but only need to cycle through it to make sure any inode that | |
115 | * prune_icache took off the LRU list has been fully torn down by the | |
116 | * time we are past evict_inodes. | |
1da177e4 | 117 | */ |
88e0fbc4 | 118 | static DECLARE_RWSEM(iprune_sem); |
1da177e4 LT |
119 | |
120 | /* | |
121 | * Statistics gathering.. | |
122 | */ | |
123 | struct inodes_stat_t inodes_stat; | |
124 | ||
3e880fb5 | 125 | static DEFINE_PER_CPU(unsigned int, nr_inodes); |
cffbc8aa | 126 | |
6b3304b5 | 127 | static struct kmem_cache *inode_cachep __read_mostly; |
1da177e4 | 128 | |
3e880fb5 | 129 | static int get_nr_inodes(void) |
cffbc8aa | 130 | { |
3e880fb5 NP |
131 | int i; |
132 | int sum = 0; | |
133 | for_each_possible_cpu(i) | |
134 | sum += per_cpu(nr_inodes, i); | |
135 | return sum < 0 ? 0 : sum; | |
cffbc8aa DC |
136 | } |
137 | ||
138 | static inline int get_nr_inodes_unused(void) | |
139 | { | |
86c8749e | 140 | return inodes_stat.nr_unused; |
cffbc8aa DC |
141 | } |
142 | ||
143 | int get_nr_dirty_inodes(void) | |
144 | { | |
3e880fb5 | 145 | /* not actually dirty inodes, but a wild approximation */ |
cffbc8aa DC |
146 | int nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); |
147 | return nr_dirty > 0 ? nr_dirty : 0; | |
cffbc8aa DC |
148 | } |
149 | ||
150 | /* | |
151 | * Handle nr_inode sysctl | |
152 | */ | |
153 | #ifdef CONFIG_SYSCTL | |
154 | int proc_nr_inodes(ctl_table *table, int write, | |
155 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
156 | { | |
157 | inodes_stat.nr_inodes = get_nr_inodes(); | |
cffbc8aa DC |
158 | return proc_dointvec(table, write, buffer, lenp, ppos); |
159 | } | |
160 | #endif | |
161 | ||
2cb1599f DC |
162 | /** |
163 | * inode_init_always - perform inode structure intialisation | |
0bc02f3f RD |
164 | * @sb: superblock inode belongs to |
165 | * @inode: inode to initialise | |
2cb1599f DC |
166 | * |
167 | * These are initializations that need to be done on every inode | |
168 | * allocation as the fields are not initialised by slab allocation. | |
169 | */ | |
54e34621 | 170 | int inode_init_always(struct super_block *sb, struct inode *inode) |
1da177e4 | 171 | { |
f5e54d6e | 172 | static const struct address_space_operations empty_aops; |
6e1d5dcc | 173 | static const struct inode_operations empty_iops; |
99ac48f5 | 174 | static const struct file_operations empty_fops; |
6b3304b5 | 175 | struct address_space *const mapping = &inode->i_data; |
2cb1599f DC |
176 | |
177 | inode->i_sb = sb; | |
178 | inode->i_blkbits = sb->s_blocksize_bits; | |
179 | inode->i_flags = 0; | |
180 | atomic_set(&inode->i_count, 1); | |
181 | inode->i_op = &empty_iops; | |
182 | inode->i_fop = &empty_fops; | |
183 | inode->i_nlink = 1; | |
56ff5efa AV |
184 | inode->i_uid = 0; |
185 | inode->i_gid = 0; | |
2cb1599f DC |
186 | atomic_set(&inode->i_writecount, 0); |
187 | inode->i_size = 0; | |
188 | inode->i_blocks = 0; | |
189 | inode->i_bytes = 0; | |
190 | inode->i_generation = 0; | |
1da177e4 | 191 | #ifdef CONFIG_QUOTA |
2cb1599f | 192 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); |
1da177e4 | 193 | #endif |
2cb1599f DC |
194 | inode->i_pipe = NULL; |
195 | inode->i_bdev = NULL; | |
196 | inode->i_cdev = NULL; | |
197 | inode->i_rdev = 0; | |
198 | inode->dirtied_when = 0; | |
6146f0d5 MZ |
199 | |
200 | if (security_inode_alloc(inode)) | |
54e34621 | 201 | goto out; |
2cb1599f DC |
202 | spin_lock_init(&inode->i_lock); |
203 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
204 | ||
205 | mutex_init(&inode->i_mutex); | |
206 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); | |
207 | ||
208 | init_rwsem(&inode->i_alloc_sem); | |
209 | lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key); | |
210 | ||
211 | mapping->a_ops = &empty_aops; | |
212 | mapping->host = inode; | |
213 | mapping->flags = 0; | |
3c1d4378 | 214 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); |
2cb1599f DC |
215 | mapping->assoc_mapping = NULL; |
216 | mapping->backing_dev_info = &default_backing_dev_info; | |
217 | mapping->writeback_index = 0; | |
218 | ||
219 | /* | |
220 | * If the block_device provides a backing_dev_info for client | |
221 | * inodes then use that. Otherwise the inode share the bdev's | |
222 | * backing_dev_info. | |
223 | */ | |
224 | if (sb->s_bdev) { | |
225 | struct backing_dev_info *bdi; | |
226 | ||
2c96ce9f | 227 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; |
2cb1599f DC |
228 | mapping->backing_dev_info = bdi; |
229 | } | |
230 | inode->i_private = NULL; | |
231 | inode->i_mapping = mapping; | |
f19d4a8f AV |
232 | #ifdef CONFIG_FS_POSIX_ACL |
233 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; | |
234 | #endif | |
2cb1599f | 235 | |
3be25f49 EP |
236 | #ifdef CONFIG_FSNOTIFY |
237 | inode->i_fsnotify_mask = 0; | |
238 | #endif | |
239 | ||
3e880fb5 | 240 | this_cpu_inc(nr_inodes); |
cffbc8aa | 241 | |
54e34621 | 242 | return 0; |
54e34621 CH |
243 | out: |
244 | return -ENOMEM; | |
1da177e4 | 245 | } |
2cb1599f DC |
246 | EXPORT_SYMBOL(inode_init_always); |
247 | ||
248 | static struct inode *alloc_inode(struct super_block *sb) | |
249 | { | |
250 | struct inode *inode; | |
251 | ||
252 | if (sb->s_op->alloc_inode) | |
253 | inode = sb->s_op->alloc_inode(sb); | |
254 | else | |
255 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
256 | ||
54e34621 CH |
257 | if (!inode) |
258 | return NULL; | |
259 | ||
260 | if (unlikely(inode_init_always(sb, inode))) { | |
261 | if (inode->i_sb->s_op->destroy_inode) | |
262 | inode->i_sb->s_op->destroy_inode(inode); | |
263 | else | |
264 | kmem_cache_free(inode_cachep, inode); | |
265 | return NULL; | |
266 | } | |
267 | ||
268 | return inode; | |
2cb1599f | 269 | } |
1da177e4 | 270 | |
ff0c7d15 NP |
271 | void free_inode_nonrcu(struct inode *inode) |
272 | { | |
273 | kmem_cache_free(inode_cachep, inode); | |
274 | } | |
275 | EXPORT_SYMBOL(free_inode_nonrcu); | |
276 | ||
2e00c97e | 277 | void __destroy_inode(struct inode *inode) |
1da177e4 | 278 | { |
b7542f8c | 279 | BUG_ON(inode_has_buffers(inode)); |
1da177e4 | 280 | security_inode_free(inode); |
3be25f49 | 281 | fsnotify_inode_delete(inode); |
f19d4a8f AV |
282 | #ifdef CONFIG_FS_POSIX_ACL |
283 | if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) | |
284 | posix_acl_release(inode->i_acl); | |
285 | if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) | |
286 | posix_acl_release(inode->i_default_acl); | |
287 | #endif | |
3e880fb5 | 288 | this_cpu_dec(nr_inodes); |
2e00c97e CH |
289 | } |
290 | EXPORT_SYMBOL(__destroy_inode); | |
291 | ||
fa0d7e3d NP |
292 | static void i_callback(struct rcu_head *head) |
293 | { | |
294 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
295 | INIT_LIST_HEAD(&inode->i_dentry); | |
296 | kmem_cache_free(inode_cachep, inode); | |
297 | } | |
298 | ||
56b0dacf | 299 | static void destroy_inode(struct inode *inode) |
2e00c97e | 300 | { |
7ccf19a8 | 301 | BUG_ON(!list_empty(&inode->i_lru)); |
2e00c97e | 302 | __destroy_inode(inode); |
1da177e4 LT |
303 | if (inode->i_sb->s_op->destroy_inode) |
304 | inode->i_sb->s_op->destroy_inode(inode); | |
305 | else | |
fa0d7e3d | 306 | call_rcu(&inode->i_rcu, i_callback); |
1da177e4 | 307 | } |
1da177e4 | 308 | |
2aa15890 MS |
309 | void address_space_init_once(struct address_space *mapping) |
310 | { | |
311 | memset(mapping, 0, sizeof(*mapping)); | |
312 | INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC); | |
313 | spin_lock_init(&mapping->tree_lock); | |
314 | spin_lock_init(&mapping->i_mmap_lock); | |
315 | INIT_LIST_HEAD(&mapping->private_list); | |
316 | spin_lock_init(&mapping->private_lock); | |
317 | INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap); | |
318 | INIT_LIST_HEAD(&mapping->i_mmap_nonlinear); | |
319 | mutex_init(&mapping->unmap_mutex); | |
320 | } | |
321 | EXPORT_SYMBOL(address_space_init_once); | |
322 | ||
1da177e4 LT |
323 | /* |
324 | * These are initializations that only need to be done | |
325 | * once, because the fields are idempotent across use | |
326 | * of the inode, so let the slab aware of that. | |
327 | */ | |
328 | void inode_init_once(struct inode *inode) | |
329 | { | |
330 | memset(inode, 0, sizeof(*inode)); | |
331 | INIT_HLIST_NODE(&inode->i_hash); | |
332 | INIT_LIST_HEAD(&inode->i_dentry); | |
333 | INIT_LIST_HEAD(&inode->i_devices); | |
7ccf19a8 NP |
334 | INIT_LIST_HEAD(&inode->i_wb_list); |
335 | INIT_LIST_HEAD(&inode->i_lru); | |
2aa15890 | 336 | address_space_init_once(&inode->i_data); |
1da177e4 | 337 | i_size_ordered_init(inode); |
3be25f49 | 338 | #ifdef CONFIG_FSNOTIFY |
e61ce867 | 339 | INIT_HLIST_HEAD(&inode->i_fsnotify_marks); |
3be25f49 | 340 | #endif |
1da177e4 | 341 | } |
1da177e4 LT |
342 | EXPORT_SYMBOL(inode_init_once); |
343 | ||
51cc5068 | 344 | static void init_once(void *foo) |
1da177e4 | 345 | { |
6b3304b5 | 346 | struct inode *inode = (struct inode *) foo; |
1da177e4 | 347 | |
a35afb83 | 348 | inode_init_once(inode); |
1da177e4 LT |
349 | } |
350 | ||
351 | /* | |
250df6ed | 352 | * inode->i_lock must be held |
1da177e4 | 353 | */ |
6b3304b5 | 354 | void __iget(struct inode *inode) |
1da177e4 | 355 | { |
9e38d86f NP |
356 | atomic_inc(&inode->i_count); |
357 | } | |
2e147f1e | 358 | |
7de9c6ee AV |
359 | /* |
360 | * get additional reference to inode; caller must already hold one. | |
361 | */ | |
362 | void ihold(struct inode *inode) | |
363 | { | |
364 | WARN_ON(atomic_inc_return(&inode->i_count) < 2); | |
365 | } | |
366 | EXPORT_SYMBOL(ihold); | |
367 | ||
9e38d86f NP |
368 | static void inode_lru_list_add(struct inode *inode) |
369 | { | |
02afc410 | 370 | spin_lock(&inode_lru_lock); |
7ccf19a8 NP |
371 | if (list_empty(&inode->i_lru)) { |
372 | list_add(&inode->i_lru, &inode_lru); | |
86c8749e | 373 | inodes_stat.nr_unused++; |
9e38d86f | 374 | } |
02afc410 | 375 | spin_unlock(&inode_lru_lock); |
9e38d86f | 376 | } |
2e147f1e | 377 | |
9e38d86f NP |
378 | static void inode_lru_list_del(struct inode *inode) |
379 | { | |
02afc410 | 380 | spin_lock(&inode_lru_lock); |
7ccf19a8 NP |
381 | if (!list_empty(&inode->i_lru)) { |
382 | list_del_init(&inode->i_lru); | |
86c8749e | 383 | inodes_stat.nr_unused--; |
9e38d86f | 384 | } |
02afc410 | 385 | spin_unlock(&inode_lru_lock); |
1da177e4 LT |
386 | } |
387 | ||
646ec461 CH |
388 | /** |
389 | * inode_sb_list_add - add inode to the superblock list of inodes | |
390 | * @inode: inode to add | |
391 | */ | |
392 | void inode_sb_list_add(struct inode *inode) | |
393 | { | |
55fa6091 DC |
394 | spin_lock(&inode_sb_list_lock); |
395 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); | |
396 | spin_unlock(&inode_sb_list_lock); | |
646ec461 CH |
397 | } |
398 | EXPORT_SYMBOL_GPL(inode_sb_list_add); | |
399 | ||
55fa6091 | 400 | static inline void inode_sb_list_del(struct inode *inode) |
646ec461 | 401 | { |
55fa6091 | 402 | spin_lock(&inode_sb_list_lock); |
646ec461 | 403 | list_del_init(&inode->i_sb_list); |
55fa6091 | 404 | spin_unlock(&inode_sb_list_lock); |
646ec461 CH |
405 | } |
406 | ||
4c51acbc DC |
407 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
408 | { | |
409 | unsigned long tmp; | |
410 | ||
411 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
412 | L1_CACHE_BYTES; | |
413 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS); | |
414 | return tmp & I_HASHMASK; | |
415 | } | |
416 | ||
417 | /** | |
418 | * __insert_inode_hash - hash an inode | |
419 | * @inode: unhashed inode | |
420 | * @hashval: unsigned long value used to locate this object in the | |
421 | * inode_hashtable. | |
422 | * | |
423 | * Add an inode to the inode hash for this superblock. | |
424 | */ | |
425 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
426 | { | |
646ec461 CH |
427 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); |
428 | ||
4c51acbc | 429 | spin_lock(&inode_lock); |
250df6ed | 430 | spin_lock(&inode->i_lock); |
646ec461 | 431 | hlist_add_head(&inode->i_hash, b); |
250df6ed | 432 | spin_unlock(&inode->i_lock); |
4c51acbc DC |
433 | spin_unlock(&inode_lock); |
434 | } | |
435 | EXPORT_SYMBOL(__insert_inode_hash); | |
436 | ||
4c51acbc DC |
437 | /** |
438 | * remove_inode_hash - remove an inode from the hash | |
439 | * @inode: inode to unhash | |
440 | * | |
441 | * Remove an inode from the superblock. | |
442 | */ | |
443 | void remove_inode_hash(struct inode *inode) | |
444 | { | |
445 | spin_lock(&inode_lock); | |
250df6ed | 446 | spin_lock(&inode->i_lock); |
4c51acbc | 447 | hlist_del_init(&inode->i_hash); |
250df6ed | 448 | spin_unlock(&inode->i_lock); |
4c51acbc DC |
449 | spin_unlock(&inode_lock); |
450 | } | |
451 | EXPORT_SYMBOL(remove_inode_hash); | |
452 | ||
b0683aa6 AV |
453 | void end_writeback(struct inode *inode) |
454 | { | |
455 | might_sleep(); | |
456 | BUG_ON(inode->i_data.nrpages); | |
457 | BUG_ON(!list_empty(&inode->i_data.private_list)); | |
458 | BUG_ON(!(inode->i_state & I_FREEING)); | |
459 | BUG_ON(inode->i_state & I_CLEAR); | |
460 | inode_sync_wait(inode); | |
fa0d7e3d | 461 | /* don't need i_lock here, no concurrent mods to i_state */ |
b0683aa6 AV |
462 | inode->i_state = I_FREEING | I_CLEAR; |
463 | } | |
464 | EXPORT_SYMBOL(end_writeback); | |
465 | ||
b2b2af8e DC |
466 | /* |
467 | * Free the inode passed in, removing it from the lists it is still connected | |
468 | * to. We remove any pages still attached to the inode and wait for any IO that | |
469 | * is still in progress before finally destroying the inode. | |
470 | * | |
471 | * An inode must already be marked I_FREEING so that we avoid the inode being | |
472 | * moved back onto lists if we race with other code that manipulates the lists | |
473 | * (e.g. writeback_single_inode). The caller is responsible for setting this. | |
474 | * | |
475 | * An inode must already be removed from the LRU list before being evicted from | |
476 | * the cache. This should occur atomically with setting the I_FREEING state | |
477 | * flag, so no inodes here should ever be on the LRU when being evicted. | |
478 | */ | |
644da596 | 479 | static void evict(struct inode *inode) |
b4272d4c AV |
480 | { |
481 | const struct super_operations *op = inode->i_sb->s_op; | |
482 | ||
b2b2af8e DC |
483 | BUG_ON(!(inode->i_state & I_FREEING)); |
484 | BUG_ON(!list_empty(&inode->i_lru)); | |
485 | ||
486 | spin_lock(&inode_lock); | |
487 | list_del_init(&inode->i_wb_list); | |
b2b2af8e DC |
488 | spin_unlock(&inode_lock); |
489 | ||
55fa6091 DC |
490 | inode_sb_list_del(inode); |
491 | ||
be7ce416 AV |
492 | if (op->evict_inode) { |
493 | op->evict_inode(inode); | |
b4272d4c AV |
494 | } else { |
495 | if (inode->i_data.nrpages) | |
496 | truncate_inode_pages(&inode->i_data, 0); | |
30140837 | 497 | end_writeback(inode); |
b4272d4c | 498 | } |
661074e9 AV |
499 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
500 | bd_forget(inode); | |
501 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) | |
502 | cd_forget(inode); | |
b2b2af8e DC |
503 | |
504 | remove_inode_hash(inode); | |
505 | ||
506 | spin_lock(&inode->i_lock); | |
507 | wake_up_bit(&inode->i_state, __I_NEW); | |
508 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); | |
509 | spin_unlock(&inode->i_lock); | |
510 | ||
511 | destroy_inode(inode); | |
b4272d4c AV |
512 | } |
513 | ||
1da177e4 LT |
514 | /* |
515 | * dispose_list - dispose of the contents of a local list | |
516 | * @head: the head of the list to free | |
517 | * | |
518 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
519 | * need to worry about list corruption and SMP locks. | |
520 | */ | |
521 | static void dispose_list(struct list_head *head) | |
522 | { | |
1da177e4 LT |
523 | while (!list_empty(head)) { |
524 | struct inode *inode; | |
525 | ||
7ccf19a8 NP |
526 | inode = list_first_entry(head, struct inode, i_lru); |
527 | list_del_init(&inode->i_lru); | |
1da177e4 | 528 | |
644da596 | 529 | evict(inode); |
1da177e4 | 530 | } |
1da177e4 LT |
531 | } |
532 | ||
63997e98 AV |
533 | /** |
534 | * evict_inodes - evict all evictable inodes for a superblock | |
535 | * @sb: superblock to operate on | |
536 | * | |
537 | * Make sure that no inodes with zero refcount are retained. This is | |
538 | * called by superblock shutdown after having MS_ACTIVE flag removed, | |
539 | * so any inode reaching zero refcount during or after that call will | |
540 | * be immediately evicted. | |
1da177e4 | 541 | */ |
63997e98 | 542 | void evict_inodes(struct super_block *sb) |
1da177e4 | 543 | { |
63997e98 AV |
544 | struct inode *inode, *next; |
545 | LIST_HEAD(dispose); | |
1da177e4 | 546 | |
55fa6091 | 547 | spin_lock(&inode_sb_list_lock); |
63997e98 AV |
548 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
549 | if (atomic_read(&inode->i_count)) | |
aabb8fdb | 550 | continue; |
250df6ed DC |
551 | |
552 | spin_lock(&inode->i_lock); | |
553 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
554 | spin_unlock(&inode->i_lock); | |
1da177e4 | 555 | continue; |
250df6ed | 556 | } |
63997e98 AV |
557 | |
558 | inode->i_state |= I_FREEING; | |
02afc410 | 559 | inode_lru_list_del(inode); |
250df6ed | 560 | spin_unlock(&inode->i_lock); |
02afc410 | 561 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 562 | } |
55fa6091 | 563 | spin_unlock(&inode_sb_list_lock); |
63997e98 AV |
564 | |
565 | dispose_list(&dispose); | |
bab1d944 CH |
566 | |
567 | /* | |
568 | * Cycle through iprune_sem to make sure any inode that prune_icache | |
569 | * moved off the list before we took the lock has been fully torn | |
570 | * down. | |
571 | */ | |
572 | down_write(&iprune_sem); | |
63997e98 | 573 | up_write(&iprune_sem); |
1da177e4 LT |
574 | } |
575 | ||
1da177e4 | 576 | /** |
a0318786 CH |
577 | * invalidate_inodes - attempt to free all inodes on a superblock |
578 | * @sb: superblock to operate on | |
93b270f7 | 579 | * @kill_dirty: flag to guide handling of dirty inodes |
1da177e4 | 580 | * |
a0318786 CH |
581 | * Attempts to free all inodes for a given superblock. If there were any |
582 | * busy inodes return a non-zero value, else zero. | |
93b270f7 N |
583 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat |
584 | * them as busy. | |
1da177e4 | 585 | */ |
93b270f7 | 586 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) |
1da177e4 | 587 | { |
cffbc8aa | 588 | int busy = 0; |
a0318786 CH |
589 | struct inode *inode, *next; |
590 | LIST_HEAD(dispose); | |
1da177e4 | 591 | |
55fa6091 | 592 | spin_lock(&inode_sb_list_lock); |
a0318786 | 593 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
250df6ed DC |
594 | spin_lock(&inode->i_lock); |
595 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
596 | spin_unlock(&inode->i_lock); | |
aabb8fdb | 597 | continue; |
250df6ed | 598 | } |
93b270f7 | 599 | if (inode->i_state & I_DIRTY && !kill_dirty) { |
250df6ed | 600 | spin_unlock(&inode->i_lock); |
93b270f7 N |
601 | busy = 1; |
602 | continue; | |
603 | } | |
99a38919 | 604 | if (atomic_read(&inode->i_count)) { |
250df6ed | 605 | spin_unlock(&inode->i_lock); |
99a38919 | 606 | busy = 1; |
1da177e4 LT |
607 | continue; |
608 | } | |
99a38919 | 609 | |
99a38919 | 610 | inode->i_state |= I_FREEING; |
02afc410 | 611 | inode_lru_list_del(inode); |
250df6ed | 612 | spin_unlock(&inode->i_lock); |
02afc410 | 613 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 614 | } |
55fa6091 | 615 | spin_unlock(&inode_sb_list_lock); |
1da177e4 | 616 | |
a0318786 | 617 | dispose_list(&dispose); |
1da177e4 LT |
618 | |
619 | return busy; | |
620 | } | |
1da177e4 LT |
621 | |
622 | static int can_unuse(struct inode *inode) | |
623 | { | |
9e38d86f | 624 | if (inode->i_state & ~I_REFERENCED) |
1da177e4 LT |
625 | return 0; |
626 | if (inode_has_buffers(inode)) | |
627 | return 0; | |
628 | if (atomic_read(&inode->i_count)) | |
629 | return 0; | |
630 | if (inode->i_data.nrpages) | |
631 | return 0; | |
632 | return 1; | |
633 | } | |
634 | ||
635 | /* | |
9e38d86f | 636 | * Scan `goal' inodes on the unused list for freeable ones. They are moved to a |
02afc410 | 637 | * temporary list and then are freed outside inode_lru_lock by dispose_list(). |
1da177e4 LT |
638 | * |
639 | * Any inodes which are pinned purely because of attached pagecache have their | |
9e38d86f NP |
640 | * pagecache removed. If the inode has metadata buffers attached to |
641 | * mapping->private_list then try to remove them. | |
1da177e4 | 642 | * |
9e38d86f NP |
643 | * If the inode has the I_REFERENCED flag set, then it means that it has been |
644 | * used recently - the flag is set in iput_final(). When we encounter such an | |
645 | * inode, clear the flag and move it to the back of the LRU so it gets another | |
646 | * pass through the LRU before it gets reclaimed. This is necessary because of | |
647 | * the fact we are doing lazy LRU updates to minimise lock contention so the | |
648 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes | |
649 | * with this flag set because they are the inodes that are out of order. | |
1da177e4 LT |
650 | */ |
651 | static void prune_icache(int nr_to_scan) | |
652 | { | |
653 | LIST_HEAD(freeable); | |
1da177e4 LT |
654 | int nr_scanned; |
655 | unsigned long reap = 0; | |
656 | ||
88e0fbc4 | 657 | down_read(&iprune_sem); |
02afc410 | 658 | spin_lock(&inode_lru_lock); |
1da177e4 LT |
659 | for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) { |
660 | struct inode *inode; | |
661 | ||
7ccf19a8 | 662 | if (list_empty(&inode_lru)) |
1da177e4 LT |
663 | break; |
664 | ||
7ccf19a8 | 665 | inode = list_entry(inode_lru.prev, struct inode, i_lru); |
1da177e4 | 666 | |
02afc410 DC |
667 | /* |
668 | * we are inverting the inode_lru_lock/inode->i_lock here, | |
669 | * so use a trylock. If we fail to get the lock, just move the | |
670 | * inode to the back of the list so we don't spin on it. | |
671 | */ | |
672 | if (!spin_trylock(&inode->i_lock)) { | |
673 | list_move(&inode->i_lru, &inode_lru); | |
674 | continue; | |
675 | } | |
676 | ||
9e38d86f NP |
677 | /* |
678 | * Referenced or dirty inodes are still in use. Give them | |
679 | * another pass through the LRU as we canot reclaim them now. | |
680 | */ | |
681 | if (atomic_read(&inode->i_count) || | |
682 | (inode->i_state & ~I_REFERENCED)) { | |
7ccf19a8 | 683 | list_del_init(&inode->i_lru); |
f283c86a | 684 | spin_unlock(&inode->i_lock); |
86c8749e | 685 | inodes_stat.nr_unused--; |
9e38d86f NP |
686 | continue; |
687 | } | |
688 | ||
689 | /* recently referenced inodes get one more pass */ | |
690 | if (inode->i_state & I_REFERENCED) { | |
9e38d86f | 691 | inode->i_state &= ~I_REFERENCED; |
250df6ed | 692 | list_move(&inode->i_lru, &inode_lru); |
f283c86a | 693 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
694 | continue; |
695 | } | |
696 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { | |
697 | __iget(inode); | |
250df6ed | 698 | spin_unlock(&inode->i_lock); |
02afc410 | 699 | spin_unlock(&inode_lru_lock); |
1da177e4 | 700 | if (remove_inode_buffers(inode)) |
fc0ecff6 AM |
701 | reap += invalidate_mapping_pages(&inode->i_data, |
702 | 0, -1); | |
1da177e4 | 703 | iput(inode); |
02afc410 | 704 | spin_lock(&inode_lru_lock); |
1da177e4 | 705 | |
7ccf19a8 NP |
706 | if (inode != list_entry(inode_lru.next, |
707 | struct inode, i_lru)) | |
1da177e4 | 708 | continue; /* wrong inode or list_empty */ |
02afc410 DC |
709 | /* avoid lock inversions with trylock */ |
710 | if (!spin_trylock(&inode->i_lock)) | |
711 | continue; | |
250df6ed DC |
712 | if (!can_unuse(inode)) { |
713 | spin_unlock(&inode->i_lock); | |
1da177e4 | 714 | continue; |
250df6ed | 715 | } |
1da177e4 | 716 | } |
7ef0d737 | 717 | WARN_ON(inode->i_state & I_NEW); |
1da177e4 | 718 | inode->i_state |= I_FREEING; |
250df6ed | 719 | spin_unlock(&inode->i_lock); |
7ccf19a8 | 720 | |
7ccf19a8 | 721 | list_move(&inode->i_lru, &freeable); |
86c8749e | 722 | inodes_stat.nr_unused--; |
1da177e4 | 723 | } |
f8891e5e CL |
724 | if (current_is_kswapd()) |
725 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
726 | else | |
727 | __count_vm_events(PGINODESTEAL, reap); | |
02afc410 | 728 | spin_unlock(&inode_lru_lock); |
1da177e4 LT |
729 | |
730 | dispose_list(&freeable); | |
88e0fbc4 | 731 | up_read(&iprune_sem); |
1da177e4 LT |
732 | } |
733 | ||
734 | /* | |
735 | * shrink_icache_memory() will attempt to reclaim some unused inodes. Here, | |
736 | * "unused" means that no dentries are referring to the inodes: the files are | |
737 | * not open and the dcache references to those inodes have already been | |
738 | * reclaimed. | |
739 | * | |
740 | * This function is passed the number of inodes to scan, and it returns the | |
741 | * total number of remaining possibly-reclaimable inodes. | |
742 | */ | |
7f8275d0 | 743 | static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
1da177e4 LT |
744 | { |
745 | if (nr) { | |
746 | /* | |
747 | * Nasty deadlock avoidance. We may hold various FS locks, | |
748 | * and we don't want to recurse into the FS that called us | |
749 | * in clear_inode() and friends.. | |
6b3304b5 | 750 | */ |
1da177e4 LT |
751 | if (!(gfp_mask & __GFP_FS)) |
752 | return -1; | |
753 | prune_icache(nr); | |
754 | } | |
cffbc8aa | 755 | return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure; |
1da177e4 LT |
756 | } |
757 | ||
8e1f936b RR |
758 | static struct shrinker icache_shrinker = { |
759 | .shrink = shrink_icache_memory, | |
760 | .seeks = DEFAULT_SEEKS, | |
761 | }; | |
762 | ||
1da177e4 LT |
763 | static void __wait_on_freeing_inode(struct inode *inode); |
764 | /* | |
765 | * Called with the inode lock held. | |
1da177e4 | 766 | */ |
6b3304b5 MK |
767 | static struct inode *find_inode(struct super_block *sb, |
768 | struct hlist_head *head, | |
769 | int (*test)(struct inode *, void *), | |
770 | void *data) | |
1da177e4 LT |
771 | { |
772 | struct hlist_node *node; | |
6b3304b5 | 773 | struct inode *inode = NULL; |
1da177e4 LT |
774 | |
775 | repeat: | |
c5c8be3c | 776 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
777 | if (inode->i_sb != sb) |
778 | continue; | |
779 | if (!test(inode, data)) | |
780 | continue; | |
250df6ed | 781 | spin_lock(&inode->i_lock); |
a4ffdde6 | 782 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
783 | __wait_on_freeing_inode(inode); |
784 | goto repeat; | |
785 | } | |
f7899bd5 | 786 | __iget(inode); |
250df6ed | 787 | spin_unlock(&inode->i_lock); |
f7899bd5 | 788 | return inode; |
1da177e4 | 789 | } |
f7899bd5 | 790 | return NULL; |
1da177e4 LT |
791 | } |
792 | ||
793 | /* | |
794 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
795 | * iget_locked for details. | |
796 | */ | |
6b3304b5 MK |
797 | static struct inode *find_inode_fast(struct super_block *sb, |
798 | struct hlist_head *head, unsigned long ino) | |
1da177e4 LT |
799 | { |
800 | struct hlist_node *node; | |
6b3304b5 | 801 | struct inode *inode = NULL; |
1da177e4 LT |
802 | |
803 | repeat: | |
c5c8be3c | 804 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
805 | if (inode->i_ino != ino) |
806 | continue; | |
807 | if (inode->i_sb != sb) | |
808 | continue; | |
250df6ed | 809 | spin_lock(&inode->i_lock); |
a4ffdde6 | 810 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
811 | __wait_on_freeing_inode(inode); |
812 | goto repeat; | |
813 | } | |
f7899bd5 | 814 | __iget(inode); |
250df6ed | 815 | spin_unlock(&inode->i_lock); |
f7899bd5 | 816 | return inode; |
1da177e4 | 817 | } |
f7899bd5 | 818 | return NULL; |
8290c35f DC |
819 | } |
820 | ||
f991bd2e ED |
821 | /* |
822 | * Each cpu owns a range of LAST_INO_BATCH numbers. | |
823 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, | |
824 | * to renew the exhausted range. | |
8290c35f | 825 | * |
f991bd2e ED |
826 | * This does not significantly increase overflow rate because every CPU can |
827 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is | |
828 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the | |
829 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase | |
830 | * overflow rate by 2x, which does not seem too significant. | |
831 | * | |
832 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
833 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
834 | * here to attempt to avoid that. | |
8290c35f | 835 | */ |
f991bd2e ED |
836 | #define LAST_INO_BATCH 1024 |
837 | static DEFINE_PER_CPU(unsigned int, last_ino); | |
838 | ||
85fe4025 | 839 | unsigned int get_next_ino(void) |
8290c35f | 840 | { |
f991bd2e ED |
841 | unsigned int *p = &get_cpu_var(last_ino); |
842 | unsigned int res = *p; | |
8290c35f | 843 | |
f991bd2e ED |
844 | #ifdef CONFIG_SMP |
845 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { | |
846 | static atomic_t shared_last_ino; | |
847 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); | |
848 | ||
849 | res = next - LAST_INO_BATCH; | |
850 | } | |
851 | #endif | |
852 | ||
853 | *p = ++res; | |
854 | put_cpu_var(last_ino); | |
855 | return res; | |
8290c35f | 856 | } |
85fe4025 | 857 | EXPORT_SYMBOL(get_next_ino); |
8290c35f | 858 | |
1da177e4 LT |
859 | /** |
860 | * new_inode - obtain an inode | |
861 | * @sb: superblock | |
862 | * | |
769848c0 | 863 | * Allocates a new inode for given superblock. The default gfp_mask |
3c1d4378 | 864 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
769848c0 MG |
865 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
866 | * for the page cache are not reclaimable or migratable, | |
867 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
868 | * newly created inode's mapping | |
869 | * | |
1da177e4 LT |
870 | */ |
871 | struct inode *new_inode(struct super_block *sb) | |
872 | { | |
6b3304b5 | 873 | struct inode *inode; |
1da177e4 | 874 | |
55fa6091 | 875 | spin_lock_prefetch(&inode_sb_list_lock); |
6b3304b5 | 876 | |
1da177e4 LT |
877 | inode = alloc_inode(sb); |
878 | if (inode) { | |
250df6ed | 879 | spin_lock(&inode->i_lock); |
1da177e4 | 880 | inode->i_state = 0; |
250df6ed | 881 | spin_unlock(&inode->i_lock); |
55fa6091 | 882 | inode_sb_list_add(inode); |
1da177e4 LT |
883 | } |
884 | return inode; | |
885 | } | |
1da177e4 LT |
886 | EXPORT_SYMBOL(new_inode); |
887 | ||
250df6ed DC |
888 | /** |
889 | * unlock_new_inode - clear the I_NEW state and wake up any waiters | |
890 | * @inode: new inode to unlock | |
891 | * | |
892 | * Called when the inode is fully initialised to clear the new state of the | |
893 | * inode and wake up anyone waiting for the inode to finish initialisation. | |
894 | */ | |
1da177e4 LT |
895 | void unlock_new_inode(struct inode *inode) |
896 | { | |
14358e6d | 897 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
a3314a0e | 898 | if (S_ISDIR(inode->i_mode)) { |
1e89a5e1 PZ |
899 | struct file_system_type *type = inode->i_sb->s_type; |
900 | ||
9a7aa12f JK |
901 | /* Set new key only if filesystem hasn't already changed it */ |
902 | if (!lockdep_match_class(&inode->i_mutex, | |
903 | &type->i_mutex_key)) { | |
904 | /* | |
905 | * ensure nobody is actually holding i_mutex | |
906 | */ | |
907 | mutex_destroy(&inode->i_mutex); | |
908 | mutex_init(&inode->i_mutex); | |
909 | lockdep_set_class(&inode->i_mutex, | |
910 | &type->i_mutex_dir_key); | |
911 | } | |
1e89a5e1 | 912 | } |
14358e6d | 913 | #endif |
250df6ed | 914 | spin_lock(&inode->i_lock); |
eaff8079 CH |
915 | WARN_ON(!(inode->i_state & I_NEW)); |
916 | inode->i_state &= ~I_NEW; | |
250df6ed DC |
917 | wake_up_bit(&inode->i_state, __I_NEW); |
918 | spin_unlock(&inode->i_lock); | |
1da177e4 | 919 | } |
1da177e4 LT |
920 | EXPORT_SYMBOL(unlock_new_inode); |
921 | ||
922 | /* | |
923 | * This is called without the inode lock held.. Be careful. | |
924 | * | |
925 | * We no longer cache the sb_flags in i_flags - see fs.h | |
926 | * -- rmk@arm.uk.linux.org | |
927 | */ | |
6b3304b5 MK |
928 | static struct inode *get_new_inode(struct super_block *sb, |
929 | struct hlist_head *head, | |
930 | int (*test)(struct inode *, void *), | |
931 | int (*set)(struct inode *, void *), | |
932 | void *data) | |
1da177e4 | 933 | { |
6b3304b5 | 934 | struct inode *inode; |
1da177e4 LT |
935 | |
936 | inode = alloc_inode(sb); | |
937 | if (inode) { | |
6b3304b5 | 938 | struct inode *old; |
1da177e4 LT |
939 | |
940 | spin_lock(&inode_lock); | |
941 | /* We released the lock, so.. */ | |
942 | old = find_inode(sb, head, test, data); | |
943 | if (!old) { | |
944 | if (set(inode, data)) | |
945 | goto set_failed; | |
946 | ||
250df6ed DC |
947 | spin_lock(&inode->i_lock); |
948 | inode->i_state = I_NEW; | |
646ec461 | 949 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 950 | spin_unlock(&inode->i_lock); |
55fa6091 | 951 | inode_sb_list_add(inode); |
1da177e4 LT |
952 | spin_unlock(&inode_lock); |
953 | ||
954 | /* Return the locked inode with I_NEW set, the | |
955 | * caller is responsible for filling in the contents | |
956 | */ | |
957 | return inode; | |
958 | } | |
959 | ||
960 | /* | |
961 | * Uhhuh, somebody else created the same inode under | |
962 | * us. Use the old inode instead of the one we just | |
963 | * allocated. | |
964 | */ | |
1da177e4 LT |
965 | spin_unlock(&inode_lock); |
966 | destroy_inode(inode); | |
967 | inode = old; | |
968 | wait_on_inode(inode); | |
969 | } | |
970 | return inode; | |
971 | ||
972 | set_failed: | |
973 | spin_unlock(&inode_lock); | |
974 | destroy_inode(inode); | |
975 | return NULL; | |
976 | } | |
977 | ||
978 | /* | |
979 | * get_new_inode_fast is the fast path version of get_new_inode, see the | |
980 | * comment at iget_locked for details. | |
981 | */ | |
6b3304b5 MK |
982 | static struct inode *get_new_inode_fast(struct super_block *sb, |
983 | struct hlist_head *head, unsigned long ino) | |
1da177e4 | 984 | { |
6b3304b5 | 985 | struct inode *inode; |
1da177e4 LT |
986 | |
987 | inode = alloc_inode(sb); | |
988 | if (inode) { | |
6b3304b5 | 989 | struct inode *old; |
1da177e4 LT |
990 | |
991 | spin_lock(&inode_lock); | |
992 | /* We released the lock, so.. */ | |
993 | old = find_inode_fast(sb, head, ino); | |
994 | if (!old) { | |
995 | inode->i_ino = ino; | |
250df6ed DC |
996 | spin_lock(&inode->i_lock); |
997 | inode->i_state = I_NEW; | |
646ec461 | 998 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 999 | spin_unlock(&inode->i_lock); |
55fa6091 | 1000 | inode_sb_list_add(inode); |
1da177e4 LT |
1001 | spin_unlock(&inode_lock); |
1002 | ||
1003 | /* Return the locked inode with I_NEW set, the | |
1004 | * caller is responsible for filling in the contents | |
1005 | */ | |
1006 | return inode; | |
1007 | } | |
1008 | ||
1009 | /* | |
1010 | * Uhhuh, somebody else created the same inode under | |
1011 | * us. Use the old inode instead of the one we just | |
1012 | * allocated. | |
1013 | */ | |
1da177e4 LT |
1014 | spin_unlock(&inode_lock); |
1015 | destroy_inode(inode); | |
1016 | inode = old; | |
1017 | wait_on_inode(inode); | |
1018 | } | |
1019 | return inode; | |
1020 | } | |
1021 | ||
ad5e195a CH |
1022 | /* |
1023 | * search the inode cache for a matching inode number. | |
1024 | * If we find one, then the inode number we are trying to | |
1025 | * allocate is not unique and so we should not use it. | |
1026 | * | |
1027 | * Returns 1 if the inode number is unique, 0 if it is not. | |
1028 | */ | |
1029 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) | |
1030 | { | |
1031 | struct hlist_head *b = inode_hashtable + hash(sb, ino); | |
1032 | struct hlist_node *node; | |
1033 | struct inode *inode; | |
1034 | ||
1035 | hlist_for_each_entry(inode, node, b, i_hash) { | |
1036 | if (inode->i_ino == ino && inode->i_sb == sb) | |
1037 | return 0; | |
1038 | } | |
1039 | ||
1040 | return 1; | |
1041 | } | |
1042 | ||
1da177e4 LT |
1043 | /** |
1044 | * iunique - get a unique inode number | |
1045 | * @sb: superblock | |
1046 | * @max_reserved: highest reserved inode number | |
1047 | * | |
1048 | * Obtain an inode number that is unique on the system for a given | |
1049 | * superblock. This is used by file systems that have no natural | |
1050 | * permanent inode numbering system. An inode number is returned that | |
1051 | * is higher than the reserved limit but unique. | |
1052 | * | |
1053 | * BUGS: | |
1054 | * With a large number of inodes live on the file system this function | |
1055 | * currently becomes quite slow. | |
1056 | */ | |
1057 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
1058 | { | |
866b04fc JL |
1059 | /* |
1060 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
1061 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
1062 | * here to attempt to avoid that. | |
1063 | */ | |
ad5e195a | 1064 | static DEFINE_SPINLOCK(iunique_lock); |
866b04fc | 1065 | static unsigned int counter; |
1da177e4 | 1066 | ino_t res; |
3361c7be | 1067 | |
1da177e4 | 1068 | spin_lock(&inode_lock); |
ad5e195a | 1069 | spin_lock(&iunique_lock); |
3361c7be JL |
1070 | do { |
1071 | if (counter <= max_reserved) | |
1072 | counter = max_reserved + 1; | |
1da177e4 | 1073 | res = counter++; |
ad5e195a CH |
1074 | } while (!test_inode_iunique(sb, res)); |
1075 | spin_unlock(&iunique_lock); | |
3361c7be | 1076 | spin_unlock(&inode_lock); |
1da177e4 | 1077 | |
3361c7be JL |
1078 | return res; |
1079 | } | |
1da177e4 LT |
1080 | EXPORT_SYMBOL(iunique); |
1081 | ||
1082 | struct inode *igrab(struct inode *inode) | |
1083 | { | |
250df6ed DC |
1084 | spin_lock(&inode->i_lock); |
1085 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { | |
1da177e4 | 1086 | __iget(inode); |
250df6ed DC |
1087 | spin_unlock(&inode->i_lock); |
1088 | } else { | |
1089 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1090 | /* |
1091 | * Handle the case where s_op->clear_inode is not been | |
1092 | * called yet, and somebody is calling igrab | |
1093 | * while the inode is getting freed. | |
1094 | */ | |
1095 | inode = NULL; | |
250df6ed | 1096 | } |
1da177e4 LT |
1097 | return inode; |
1098 | } | |
1da177e4 LT |
1099 | EXPORT_SYMBOL(igrab); |
1100 | ||
1101 | /** | |
1102 | * ifind - internal function, you want ilookup5() or iget5(). | |
1103 | * @sb: super block of file system to search | |
1104 | * @head: the head of the list to search | |
1105 | * @test: callback used for comparisons between inodes | |
1106 | * @data: opaque data pointer to pass to @test | |
88bd5121 | 1107 | * @wait: if true wait for the inode to be unlocked, if false do not |
1da177e4 LT |
1108 | * |
1109 | * ifind() searches for the inode specified by @data in the inode | |
1110 | * cache. This is a generalized version of ifind_fast() for file systems where | |
1111 | * the inode number is not sufficient for unique identification of an inode. | |
1112 | * | |
1113 | * If the inode is in the cache, the inode is returned with an incremented | |
1114 | * reference count. | |
1115 | * | |
1116 | * Otherwise NULL is returned. | |
1117 | * | |
1118 | * Note, @test is called with the inode_lock held, so can't sleep. | |
1119 | */ | |
5d2bea45 | 1120 | static struct inode *ifind(struct super_block *sb, |
1da177e4 | 1121 | struct hlist_head *head, int (*test)(struct inode *, void *), |
88bd5121 | 1122 | void *data, const int wait) |
1da177e4 LT |
1123 | { |
1124 | struct inode *inode; | |
1125 | ||
1126 | spin_lock(&inode_lock); | |
1127 | inode = find_inode(sb, head, test, data); | |
1128 | if (inode) { | |
1da177e4 | 1129 | spin_unlock(&inode_lock); |
88bd5121 AA |
1130 | if (likely(wait)) |
1131 | wait_on_inode(inode); | |
1da177e4 LT |
1132 | return inode; |
1133 | } | |
1134 | spin_unlock(&inode_lock); | |
1135 | return NULL; | |
1136 | } | |
1137 | ||
1138 | /** | |
1139 | * ifind_fast - internal function, you want ilookup() or iget(). | |
1140 | * @sb: super block of file system to search | |
1141 | * @head: head of the list to search | |
1142 | * @ino: inode number to search for | |
1143 | * | |
1144 | * ifind_fast() searches for the inode @ino in the inode cache. This is for | |
1145 | * file systems where the inode number is sufficient for unique identification | |
1146 | * of an inode. | |
1147 | * | |
1148 | * If the inode is in the cache, the inode is returned with an incremented | |
1149 | * reference count. | |
1150 | * | |
1151 | * Otherwise NULL is returned. | |
1152 | */ | |
5d2bea45 | 1153 | static struct inode *ifind_fast(struct super_block *sb, |
1da177e4 LT |
1154 | struct hlist_head *head, unsigned long ino) |
1155 | { | |
1156 | struct inode *inode; | |
1157 | ||
1158 | spin_lock(&inode_lock); | |
1159 | inode = find_inode_fast(sb, head, ino); | |
1160 | if (inode) { | |
1da177e4 LT |
1161 | spin_unlock(&inode_lock); |
1162 | wait_on_inode(inode); | |
1163 | return inode; | |
1164 | } | |
1165 | spin_unlock(&inode_lock); | |
1166 | return NULL; | |
1167 | } | |
1168 | ||
1169 | /** | |
88bd5121 | 1170 | * ilookup5_nowait - search for an inode in the inode cache |
1da177e4 LT |
1171 | * @sb: super block of file system to search |
1172 | * @hashval: hash value (usually inode number) to search for | |
1173 | * @test: callback used for comparisons between inodes | |
1174 | * @data: opaque data pointer to pass to @test | |
1175 | * | |
1176 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
1177 | * @data in the inode cache. This is a generalized version of ilookup() for | |
1178 | * file systems where the inode number is not sufficient for unique | |
1179 | * identification of an inode. | |
1180 | * | |
1181 | * If the inode is in the cache, the inode is returned with an incremented | |
88bd5121 AA |
1182 | * reference count. Note, the inode lock is not waited upon so you have to be |
1183 | * very careful what you do with the returned inode. You probably should be | |
1184 | * using ilookup5() instead. | |
1185 | * | |
1186 | * Otherwise NULL is returned. | |
1187 | * | |
1188 | * Note, @test is called with the inode_lock held, so can't sleep. | |
1189 | */ | |
1190 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, | |
1191 | int (*test)(struct inode *, void *), void *data) | |
1192 | { | |
1193 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1194 | ||
1195 | return ifind(sb, head, test, data, 0); | |
1196 | } | |
88bd5121 AA |
1197 | EXPORT_SYMBOL(ilookup5_nowait); |
1198 | ||
1199 | /** | |
1200 | * ilookup5 - search for an inode in the inode cache | |
1201 | * @sb: super block of file system to search | |
1202 | * @hashval: hash value (usually inode number) to search for | |
1203 | * @test: callback used for comparisons between inodes | |
1204 | * @data: opaque data pointer to pass to @test | |
1205 | * | |
1206 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
1207 | * @data in the inode cache. This is a generalized version of ilookup() for | |
1208 | * file systems where the inode number is not sufficient for unique | |
1209 | * identification of an inode. | |
1210 | * | |
1211 | * If the inode is in the cache, the inode lock is waited upon and the inode is | |
1212 | * returned with an incremented reference count. | |
1da177e4 LT |
1213 | * |
1214 | * Otherwise NULL is returned. | |
1215 | * | |
1216 | * Note, @test is called with the inode_lock held, so can't sleep. | |
1217 | */ | |
1218 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
1219 | int (*test)(struct inode *, void *), void *data) | |
1220 | { | |
1221 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1222 | ||
88bd5121 | 1223 | return ifind(sb, head, test, data, 1); |
1da177e4 | 1224 | } |
1da177e4 LT |
1225 | EXPORT_SYMBOL(ilookup5); |
1226 | ||
1227 | /** | |
1228 | * ilookup - search for an inode in the inode cache | |
1229 | * @sb: super block of file system to search | |
1230 | * @ino: inode number to search for | |
1231 | * | |
1232 | * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache. | |
1233 | * This is for file systems where the inode number is sufficient for unique | |
1234 | * identification of an inode. | |
1235 | * | |
1236 | * If the inode is in the cache, the inode is returned with an incremented | |
1237 | * reference count. | |
1238 | * | |
1239 | * Otherwise NULL is returned. | |
1240 | */ | |
1241 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1242 | { | |
1243 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1244 | ||
1245 | return ifind_fast(sb, head, ino); | |
1246 | } | |
1da177e4 LT |
1247 | EXPORT_SYMBOL(ilookup); |
1248 | ||
1249 | /** | |
1250 | * iget5_locked - obtain an inode from a mounted file system | |
1251 | * @sb: super block of file system | |
1252 | * @hashval: hash value (usually inode number) to get | |
1253 | * @test: callback used for comparisons between inodes | |
1254 | * @set: callback used to initialize a new struct inode | |
1255 | * @data: opaque data pointer to pass to @test and @set | |
1256 | * | |
1da177e4 LT |
1257 | * iget5_locked() uses ifind() to search for the inode specified by @hashval |
1258 | * and @data in the inode cache and if present it is returned with an increased | |
1259 | * reference count. This is a generalized version of iget_locked() for file | |
1260 | * systems where the inode number is not sufficient for unique identification | |
1261 | * of an inode. | |
1262 | * | |
1263 | * If the inode is not in cache, get_new_inode() is called to allocate a new | |
1264 | * inode and this is returned locked, hashed, and with the I_NEW flag set. The | |
1265 | * file system gets to fill it in before unlocking it via unlock_new_inode(). | |
1266 | * | |
1267 | * Note both @test and @set are called with the inode_lock held, so can't sleep. | |
1268 | */ | |
1269 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, | |
1270 | int (*test)(struct inode *, void *), | |
1271 | int (*set)(struct inode *, void *), void *data) | |
1272 | { | |
1273 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1274 | struct inode *inode; | |
1275 | ||
88bd5121 | 1276 | inode = ifind(sb, head, test, data, 1); |
1da177e4 LT |
1277 | if (inode) |
1278 | return inode; | |
1279 | /* | |
1280 | * get_new_inode() will do the right thing, re-trying the search | |
1281 | * in case it had to block at any point. | |
1282 | */ | |
1283 | return get_new_inode(sb, head, test, set, data); | |
1284 | } | |
1da177e4 LT |
1285 | EXPORT_SYMBOL(iget5_locked); |
1286 | ||
1287 | /** | |
1288 | * iget_locked - obtain an inode from a mounted file system | |
1289 | * @sb: super block of file system | |
1290 | * @ino: inode number to get | |
1291 | * | |
1da177e4 LT |
1292 | * iget_locked() uses ifind_fast() to search for the inode specified by @ino in |
1293 | * the inode cache and if present it is returned with an increased reference | |
1294 | * count. This is for file systems where the inode number is sufficient for | |
1295 | * unique identification of an inode. | |
1296 | * | |
1297 | * If the inode is not in cache, get_new_inode_fast() is called to allocate a | |
1298 | * new inode and this is returned locked, hashed, and with the I_NEW flag set. | |
1299 | * The file system gets to fill it in before unlocking it via | |
1300 | * unlock_new_inode(). | |
1301 | */ | |
1302 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) | |
1303 | { | |
1304 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1305 | struct inode *inode; | |
1306 | ||
1307 | inode = ifind_fast(sb, head, ino); | |
1308 | if (inode) | |
1309 | return inode; | |
1310 | /* | |
1311 | * get_new_inode_fast() will do the right thing, re-trying the search | |
1312 | * in case it had to block at any point. | |
1313 | */ | |
1314 | return get_new_inode_fast(sb, head, ino); | |
1315 | } | |
1da177e4 LT |
1316 | EXPORT_SYMBOL(iget_locked); |
1317 | ||
261bca86 AV |
1318 | int insert_inode_locked(struct inode *inode) |
1319 | { | |
1320 | struct super_block *sb = inode->i_sb; | |
1321 | ino_t ino = inode->i_ino; | |
1322 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
261bca86 | 1323 | |
261bca86 | 1324 | while (1) { |
72a43d63 AV |
1325 | struct hlist_node *node; |
1326 | struct inode *old = NULL; | |
261bca86 | 1327 | spin_lock(&inode_lock); |
72a43d63 AV |
1328 | hlist_for_each_entry(old, node, head, i_hash) { |
1329 | if (old->i_ino != ino) | |
1330 | continue; | |
1331 | if (old->i_sb != sb) | |
1332 | continue; | |
250df6ed DC |
1333 | spin_lock(&old->i_lock); |
1334 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1335 | spin_unlock(&old->i_lock); | |
72a43d63 | 1336 | continue; |
250df6ed | 1337 | } |
72a43d63 AV |
1338 | break; |
1339 | } | |
1340 | if (likely(!node)) { | |
250df6ed DC |
1341 | spin_lock(&inode->i_lock); |
1342 | inode->i_state |= I_NEW; | |
261bca86 | 1343 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1344 | spin_unlock(&inode->i_lock); |
261bca86 AV |
1345 | spin_unlock(&inode_lock); |
1346 | return 0; | |
1347 | } | |
1348 | __iget(old); | |
250df6ed | 1349 | spin_unlock(&old->i_lock); |
261bca86 AV |
1350 | spin_unlock(&inode_lock); |
1351 | wait_on_inode(old); | |
1d3382cb | 1352 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1353 | iput(old); |
1354 | return -EBUSY; | |
1355 | } | |
1356 | iput(old); | |
1357 | } | |
1358 | } | |
261bca86 AV |
1359 | EXPORT_SYMBOL(insert_inode_locked); |
1360 | ||
1361 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1362 | int (*test)(struct inode *, void *), void *data) | |
1363 | { | |
1364 | struct super_block *sb = inode->i_sb; | |
1365 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
261bca86 | 1366 | |
261bca86 | 1367 | while (1) { |
72a43d63 AV |
1368 | struct hlist_node *node; |
1369 | struct inode *old = NULL; | |
1370 | ||
261bca86 | 1371 | spin_lock(&inode_lock); |
72a43d63 AV |
1372 | hlist_for_each_entry(old, node, head, i_hash) { |
1373 | if (old->i_sb != sb) | |
1374 | continue; | |
1375 | if (!test(old, data)) | |
1376 | continue; | |
250df6ed DC |
1377 | spin_lock(&old->i_lock); |
1378 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1379 | spin_unlock(&old->i_lock); | |
72a43d63 | 1380 | continue; |
250df6ed | 1381 | } |
72a43d63 AV |
1382 | break; |
1383 | } | |
1384 | if (likely(!node)) { | |
250df6ed DC |
1385 | spin_lock(&inode->i_lock); |
1386 | inode->i_state |= I_NEW; | |
261bca86 | 1387 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1388 | spin_unlock(&inode->i_lock); |
261bca86 AV |
1389 | spin_unlock(&inode_lock); |
1390 | return 0; | |
1391 | } | |
1392 | __iget(old); | |
250df6ed | 1393 | spin_unlock(&old->i_lock); |
261bca86 AV |
1394 | spin_unlock(&inode_lock); |
1395 | wait_on_inode(old); | |
1d3382cb | 1396 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1397 | iput(old); |
1398 | return -EBUSY; | |
1399 | } | |
1400 | iput(old); | |
1401 | } | |
1402 | } | |
261bca86 AV |
1403 | EXPORT_SYMBOL(insert_inode_locked4); |
1404 | ||
1da177e4 | 1405 | |
45321ac5 AV |
1406 | int generic_delete_inode(struct inode *inode) |
1407 | { | |
1408 | return 1; | |
1409 | } | |
1410 | EXPORT_SYMBOL(generic_delete_inode); | |
1411 | ||
1da177e4 | 1412 | /* |
45321ac5 AV |
1413 | * Normal UNIX filesystem behaviour: delete the |
1414 | * inode when the usage count drops to zero, and | |
1415 | * i_nlink is zero. | |
1da177e4 | 1416 | */ |
45321ac5 | 1417 | int generic_drop_inode(struct inode *inode) |
1da177e4 | 1418 | { |
1d3382cb | 1419 | return !inode->i_nlink || inode_unhashed(inode); |
1da177e4 | 1420 | } |
45321ac5 | 1421 | EXPORT_SYMBOL_GPL(generic_drop_inode); |
1da177e4 | 1422 | |
45321ac5 AV |
1423 | /* |
1424 | * Called when we're dropping the last reference | |
1425 | * to an inode. | |
22fe4042 | 1426 | * |
45321ac5 AV |
1427 | * Call the FS "drop_inode()" function, defaulting to |
1428 | * the legacy UNIX filesystem behaviour. If it tells | |
1429 | * us to evict inode, do so. Otherwise, retain inode | |
1430 | * in cache if fs is alive, sync and evict if fs is | |
1431 | * shutting down. | |
22fe4042 | 1432 | */ |
45321ac5 | 1433 | static void iput_final(struct inode *inode) |
1da177e4 LT |
1434 | { |
1435 | struct super_block *sb = inode->i_sb; | |
45321ac5 AV |
1436 | const struct super_operations *op = inode->i_sb->s_op; |
1437 | int drop; | |
1438 | ||
250df6ed DC |
1439 | WARN_ON(inode->i_state & I_NEW); |
1440 | ||
45321ac5 AV |
1441 | if (op && op->drop_inode) |
1442 | drop = op->drop_inode(inode); | |
1443 | else | |
1444 | drop = generic_drop_inode(inode); | |
1da177e4 | 1445 | |
b2b2af8e DC |
1446 | if (!drop && (sb->s_flags & MS_ACTIVE)) { |
1447 | inode->i_state |= I_REFERENCED; | |
1448 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) | |
1449 | inode_lru_list_add(inode); | |
1450 | spin_unlock(&inode->i_lock); | |
b2b2af8e DC |
1451 | return; |
1452 | } | |
1453 | ||
45321ac5 | 1454 | if (!drop) { |
991114c6 | 1455 | inode->i_state |= I_WILL_FREE; |
250df6ed | 1456 | spin_unlock(&inode->i_lock); |
1da177e4 | 1457 | write_inode_now(inode, 1); |
250df6ed | 1458 | spin_lock(&inode->i_lock); |
7ef0d737 | 1459 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1460 | inode->i_state &= ~I_WILL_FREE; |
1da177e4 | 1461 | } |
7ccf19a8 | 1462 | |
991114c6 | 1463 | inode->i_state |= I_FREEING; |
9e38d86f | 1464 | inode_lru_list_del(inode); |
b2b2af8e | 1465 | spin_unlock(&inode->i_lock); |
b2b2af8e | 1466 | |
644da596 | 1467 | evict(inode); |
1da177e4 LT |
1468 | } |
1469 | ||
1da177e4 | 1470 | /** |
6b3304b5 | 1471 | * iput - put an inode |
1da177e4 LT |
1472 | * @inode: inode to put |
1473 | * | |
1474 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1475 | * zero, the inode is then freed and may also be destroyed. | |
1476 | * | |
1477 | * Consequently, iput() can sleep. | |
1478 | */ | |
1479 | void iput(struct inode *inode) | |
1480 | { | |
1481 | if (inode) { | |
a4ffdde6 | 1482 | BUG_ON(inode->i_state & I_CLEAR); |
1da177e4 | 1483 | |
f283c86a | 1484 | if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) |
1da177e4 LT |
1485 | iput_final(inode); |
1486 | } | |
1487 | } | |
1da177e4 LT |
1488 | EXPORT_SYMBOL(iput); |
1489 | ||
1490 | /** | |
1491 | * bmap - find a block number in a file | |
1492 | * @inode: inode of file | |
1493 | * @block: block to find | |
1494 | * | |
1495 | * Returns the block number on the device holding the inode that | |
1496 | * is the disk block number for the block of the file requested. | |
1497 | * That is, asked for block 4 of inode 1 the function will return the | |
6b3304b5 | 1498 | * disk block relative to the disk start that holds that block of the |
1da177e4 LT |
1499 | * file. |
1500 | */ | |
6b3304b5 | 1501 | sector_t bmap(struct inode *inode, sector_t block) |
1da177e4 LT |
1502 | { |
1503 | sector_t res = 0; | |
1504 | if (inode->i_mapping->a_ops->bmap) | |
1505 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1506 | return res; | |
1507 | } | |
1da177e4 LT |
1508 | EXPORT_SYMBOL(bmap); |
1509 | ||
11ff6f05 MG |
1510 | /* |
1511 | * With relative atime, only update atime if the previous atime is | |
1512 | * earlier than either the ctime or mtime or if at least a day has | |
1513 | * passed since the last atime update. | |
1514 | */ | |
1515 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1516 | struct timespec now) | |
1517 | { | |
1518 | ||
1519 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1520 | return 1; | |
1521 | /* | |
1522 | * Is mtime younger than atime? If yes, update atime: | |
1523 | */ | |
1524 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1525 | return 1; | |
1526 | /* | |
1527 | * Is ctime younger than atime? If yes, update atime: | |
1528 | */ | |
1529 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1530 | return 1; | |
1531 | ||
1532 | /* | |
1533 | * Is the previous atime value older than a day? If yes, | |
1534 | * update atime: | |
1535 | */ | |
1536 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1537 | return 1; | |
1538 | /* | |
1539 | * Good, we can skip the atime update: | |
1540 | */ | |
1541 | return 0; | |
1542 | } | |
1543 | ||
1da177e4 | 1544 | /** |
869243a0 CH |
1545 | * touch_atime - update the access time |
1546 | * @mnt: mount the inode is accessed on | |
7045f37b | 1547 | * @dentry: dentry accessed |
1da177e4 LT |
1548 | * |
1549 | * Update the accessed time on an inode and mark it for writeback. | |
1550 | * This function automatically handles read only file systems and media, | |
1551 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1552 | */ | |
869243a0 | 1553 | void touch_atime(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 1554 | { |
869243a0 | 1555 | struct inode *inode = dentry->d_inode; |
1da177e4 LT |
1556 | struct timespec now; |
1557 | ||
cdb70f3f | 1558 | if (inode->i_flags & S_NOATIME) |
b12536c2 | 1559 | return; |
37756ced | 1560 | if (IS_NOATIME(inode)) |
b12536c2 | 1561 | return; |
b2276138 | 1562 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1563 | return; |
47ae32d6 | 1564 | |
cdb70f3f | 1565 | if (mnt->mnt_flags & MNT_NOATIME) |
b12536c2 | 1566 | return; |
cdb70f3f | 1567 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1568 | return; |
1da177e4 LT |
1569 | |
1570 | now = current_fs_time(inode->i_sb); | |
11ff6f05 MG |
1571 | |
1572 | if (!relatime_need_update(mnt, inode, now)) | |
b12536c2 | 1573 | return; |
11ff6f05 | 1574 | |
47ae32d6 | 1575 | if (timespec_equal(&inode->i_atime, &now)) |
b12536c2 AK |
1576 | return; |
1577 | ||
1578 | if (mnt_want_write(mnt)) | |
1579 | return; | |
47ae32d6 VH |
1580 | |
1581 | inode->i_atime = now; | |
1582 | mark_inode_dirty_sync(inode); | |
cdb70f3f | 1583 | mnt_drop_write(mnt); |
1da177e4 | 1584 | } |
869243a0 | 1585 | EXPORT_SYMBOL(touch_atime); |
1da177e4 LT |
1586 | |
1587 | /** | |
870f4817 CH |
1588 | * file_update_time - update mtime and ctime time |
1589 | * @file: file accessed | |
1da177e4 | 1590 | * |
870f4817 CH |
1591 | * Update the mtime and ctime members of an inode and mark the inode |
1592 | * for writeback. Note that this function is meant exclusively for | |
1593 | * usage in the file write path of filesystems, and filesystems may | |
1594 | * choose to explicitly ignore update via this function with the | |
2eadfc0e | 1595 | * S_NOCMTIME inode flag, e.g. for network filesystem where these |
870f4817 | 1596 | * timestamps are handled by the server. |
1da177e4 LT |
1597 | */ |
1598 | ||
870f4817 | 1599 | void file_update_time(struct file *file) |
1da177e4 | 1600 | { |
0f7fc9e4 | 1601 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 | 1602 | struct timespec now; |
ce06e0b2 | 1603 | enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0; |
1da177e4 | 1604 | |
ce06e0b2 | 1605 | /* First try to exhaust all avenues to not sync */ |
1da177e4 LT |
1606 | if (IS_NOCMTIME(inode)) |
1607 | return; | |
20ddee2c | 1608 | |
1da177e4 | 1609 | now = current_fs_time(inode->i_sb); |
ce06e0b2 AK |
1610 | if (!timespec_equal(&inode->i_mtime, &now)) |
1611 | sync_it = S_MTIME; | |
1da177e4 | 1612 | |
ce06e0b2 AK |
1613 | if (!timespec_equal(&inode->i_ctime, &now)) |
1614 | sync_it |= S_CTIME; | |
870f4817 | 1615 | |
ce06e0b2 AK |
1616 | if (IS_I_VERSION(inode)) |
1617 | sync_it |= S_VERSION; | |
7a224228 | 1618 | |
ce06e0b2 AK |
1619 | if (!sync_it) |
1620 | return; | |
1621 | ||
1622 | /* Finally allowed to write? Takes lock. */ | |
1623 | if (mnt_want_write_file(file)) | |
1624 | return; | |
1625 | ||
1626 | /* Only change inode inside the lock region */ | |
1627 | if (sync_it & S_VERSION) | |
1628 | inode_inc_iversion(inode); | |
1629 | if (sync_it & S_CTIME) | |
1630 | inode->i_ctime = now; | |
1631 | if (sync_it & S_MTIME) | |
1632 | inode->i_mtime = now; | |
1633 | mark_inode_dirty_sync(inode); | |
20ddee2c | 1634 | mnt_drop_write(file->f_path.mnt); |
1da177e4 | 1635 | } |
870f4817 | 1636 | EXPORT_SYMBOL(file_update_time); |
1da177e4 LT |
1637 | |
1638 | int inode_needs_sync(struct inode *inode) | |
1639 | { | |
1640 | if (IS_SYNC(inode)) | |
1641 | return 1; | |
1642 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1643 | return 1; | |
1644 | return 0; | |
1645 | } | |
1da177e4 LT |
1646 | EXPORT_SYMBOL(inode_needs_sync); |
1647 | ||
1da177e4 LT |
1648 | int inode_wait(void *word) |
1649 | { | |
1650 | schedule(); | |
1651 | return 0; | |
1652 | } | |
d44dab8d | 1653 | EXPORT_SYMBOL(inode_wait); |
1da177e4 LT |
1654 | |
1655 | /* | |
168a9fd6 MS |
1656 | * If we try to find an inode in the inode hash while it is being |
1657 | * deleted, we have to wait until the filesystem completes its | |
1658 | * deletion before reporting that it isn't found. This function waits | |
1659 | * until the deletion _might_ have completed. Callers are responsible | |
1660 | * to recheck inode state. | |
1661 | * | |
eaff8079 | 1662 | * It doesn't matter if I_NEW is not set initially, a call to |
250df6ed DC |
1663 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list |
1664 | * will DTRT. | |
1da177e4 LT |
1665 | */ |
1666 | static void __wait_on_freeing_inode(struct inode *inode) | |
1667 | { | |
1668 | wait_queue_head_t *wq; | |
eaff8079 CH |
1669 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); |
1670 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1da177e4 | 1671 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
250df6ed | 1672 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1673 | spin_unlock(&inode_lock); |
1674 | schedule(); | |
1675 | finish_wait(wq, &wait.wait); | |
1676 | spin_lock(&inode_lock); | |
1677 | } | |
1678 | ||
1da177e4 LT |
1679 | static __initdata unsigned long ihash_entries; |
1680 | static int __init set_ihash_entries(char *str) | |
1681 | { | |
1682 | if (!str) | |
1683 | return 0; | |
1684 | ihash_entries = simple_strtoul(str, &str, 0); | |
1685 | return 1; | |
1686 | } | |
1687 | __setup("ihash_entries=", set_ihash_entries); | |
1688 | ||
1689 | /* | |
1690 | * Initialize the waitqueues and inode hash table. | |
1691 | */ | |
1692 | void __init inode_init_early(void) | |
1693 | { | |
1694 | int loop; | |
1695 | ||
1696 | /* If hashes are distributed across NUMA nodes, defer | |
1697 | * hash allocation until vmalloc space is available. | |
1698 | */ | |
1699 | if (hashdist) | |
1700 | return; | |
1701 | ||
1702 | inode_hashtable = | |
1703 | alloc_large_system_hash("Inode-cache", | |
1704 | sizeof(struct hlist_head), | |
1705 | ihash_entries, | |
1706 | 14, | |
1707 | HASH_EARLY, | |
1708 | &i_hash_shift, | |
1709 | &i_hash_mask, | |
1710 | 0); | |
1711 | ||
1712 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1713 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1714 | } | |
1715 | ||
74bf17cf | 1716 | void __init inode_init(void) |
1da177e4 LT |
1717 | { |
1718 | int loop; | |
1719 | ||
1720 | /* inode slab cache */ | |
b0196009 PJ |
1721 | inode_cachep = kmem_cache_create("inode_cache", |
1722 | sizeof(struct inode), | |
1723 | 0, | |
1724 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1725 | SLAB_MEM_SPREAD), | |
20c2df83 | 1726 | init_once); |
8e1f936b | 1727 | register_shrinker(&icache_shrinker); |
1da177e4 LT |
1728 | |
1729 | /* Hash may have been set up in inode_init_early */ | |
1730 | if (!hashdist) | |
1731 | return; | |
1732 | ||
1733 | inode_hashtable = | |
1734 | alloc_large_system_hash("Inode-cache", | |
1735 | sizeof(struct hlist_head), | |
1736 | ihash_entries, | |
1737 | 14, | |
1738 | 0, | |
1739 | &i_hash_shift, | |
1740 | &i_hash_mask, | |
1741 | 0); | |
1742 | ||
1743 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1744 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1745 | } | |
1746 | ||
1747 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1748 | { | |
1749 | inode->i_mode = mode; | |
1750 | if (S_ISCHR(mode)) { | |
1751 | inode->i_fop = &def_chr_fops; | |
1752 | inode->i_rdev = rdev; | |
1753 | } else if (S_ISBLK(mode)) { | |
1754 | inode->i_fop = &def_blk_fops; | |
1755 | inode->i_rdev = rdev; | |
1756 | } else if (S_ISFIFO(mode)) | |
1757 | inode->i_fop = &def_fifo_fops; | |
1758 | else if (S_ISSOCK(mode)) | |
1759 | inode->i_fop = &bad_sock_fops; | |
1760 | else | |
af0d9ae8 MK |
1761 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" |
1762 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1763 | inode->i_ino); | |
1da177e4 LT |
1764 | } |
1765 | EXPORT_SYMBOL(init_special_inode); | |
a1bd120d DM |
1766 | |
1767 | /** | |
eaae668d | 1768 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards |
a1bd120d DM |
1769 | * @inode: New inode |
1770 | * @dir: Directory inode | |
1771 | * @mode: mode of the new inode | |
1772 | */ | |
1773 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
1774 | mode_t mode) | |
1775 | { | |
1776 | inode->i_uid = current_fsuid(); | |
1777 | if (dir && dir->i_mode & S_ISGID) { | |
1778 | inode->i_gid = dir->i_gid; | |
1779 | if (S_ISDIR(mode)) | |
1780 | mode |= S_ISGID; | |
1781 | } else | |
1782 | inode->i_gid = current_fsgid(); | |
1783 | inode->i_mode = mode; | |
1784 | } | |
1785 | EXPORT_SYMBOL(inode_init_owner); | |
e795b717 | 1786 | |
2e149670 SH |
1787 | /** |
1788 | * inode_owner_or_capable - check current task permissions to inode | |
1789 | * @inode: inode being checked | |
1790 | * | |
1791 | * Return true if current either has CAP_FOWNER to the inode, or | |
1792 | * owns the file. | |
e795b717 | 1793 | */ |
2e149670 | 1794 | bool inode_owner_or_capable(const struct inode *inode) |
e795b717 SH |
1795 | { |
1796 | struct user_namespace *ns = inode_userns(inode); | |
1797 | ||
1798 | if (current_user_ns() == ns && current_fsuid() == inode->i_uid) | |
1799 | return true; | |
1800 | if (ns_capable(ns, CAP_FOWNER)) | |
1801 | return true; | |
1802 | return false; | |
1803 | } | |
2e149670 | 1804 | EXPORT_SYMBOL(inode_owner_or_capable); |