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 | |
250df6ed DC |
37 | * |
38 | * Lock ordering: | |
39 | * inode_lock | |
40 | * inode->i_lock | |
02afc410 | 41 | * inode_lru_lock |
250df6ed DC |
42 | */ |
43 | ||
1da177e4 LT |
44 | /* |
45 | * This is needed for the following functions: | |
46 | * - inode_has_buffers | |
1da177e4 LT |
47 | * - invalidate_bdev |
48 | * | |
49 | * FIXME: remove all knowledge of the buffer layer from this file | |
50 | */ | |
51 | #include <linux/buffer_head.h> | |
52 | ||
53 | /* | |
54 | * New inode.c implementation. | |
55 | * | |
56 | * This implementation has the basic premise of trying | |
57 | * to be extremely low-overhead and SMP-safe, yet be | |
58 | * simple enough to be "obviously correct". | |
59 | * | |
60 | * Famous last words. | |
61 | */ | |
62 | ||
63 | /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */ | |
64 | ||
65 | /* #define INODE_PARANOIA 1 */ | |
66 | /* #define INODE_DEBUG 1 */ | |
67 | ||
68 | /* | |
69 | * Inode lookup is no longer as critical as it used to be: | |
70 | * most of the lookups are going to be through the dcache. | |
71 | */ | |
72 | #define I_HASHBITS i_hash_shift | |
73 | #define I_HASHMASK i_hash_mask | |
74 | ||
fa3536cc ED |
75 | static unsigned int i_hash_mask __read_mostly; |
76 | static unsigned int i_hash_shift __read_mostly; | |
1da177e4 LT |
77 | |
78 | /* | |
79 | * Each inode can be on two separate lists. One is | |
80 | * the hash list of the inode, used for lookups. The | |
81 | * other linked list is the "type" list: | |
82 | * "in_use" - valid inode, i_count > 0, i_nlink > 0 | |
83 | * "dirty" - as "in_use" but also dirty | |
84 | * "unused" - valid inode, i_count = 0 | |
85 | * | |
86 | * A "dirty" list is maintained for each super block, | |
87 | * allowing for low-overhead inode sync() operations. | |
88 | */ | |
89 | ||
7ccf19a8 | 90 | static LIST_HEAD(inode_lru); |
02afc410 | 91 | static DEFINE_SPINLOCK(inode_lru_lock); |
fa3536cc | 92 | static struct hlist_head *inode_hashtable __read_mostly; |
1da177e4 LT |
93 | |
94 | /* | |
95 | * A simple spinlock to protect the list manipulations. | |
96 | * | |
97 | * NOTE! You also have to own the lock if you change | |
98 | * the i_state of an inode while it is in use.. | |
99 | */ | |
100 | DEFINE_SPINLOCK(inode_lock); | |
101 | ||
102 | /* | |
bab1d944 CH |
103 | * iprune_sem provides exclusion between the icache shrinking and the |
104 | * umount path. | |
88e0fbc4 | 105 | * |
bab1d944 CH |
106 | * We don't actually need it to protect anything in the umount path, |
107 | * but only need to cycle through it to make sure any inode that | |
108 | * prune_icache took off the LRU list has been fully torn down by the | |
109 | * time we are past evict_inodes. | |
1da177e4 | 110 | */ |
88e0fbc4 | 111 | static DECLARE_RWSEM(iprune_sem); |
1da177e4 LT |
112 | |
113 | /* | |
114 | * Statistics gathering.. | |
115 | */ | |
116 | struct inodes_stat_t inodes_stat; | |
117 | ||
3e880fb5 | 118 | static DEFINE_PER_CPU(unsigned int, nr_inodes); |
cffbc8aa | 119 | |
6b3304b5 | 120 | static struct kmem_cache *inode_cachep __read_mostly; |
1da177e4 | 121 | |
3e880fb5 | 122 | static int get_nr_inodes(void) |
cffbc8aa | 123 | { |
3e880fb5 NP |
124 | int i; |
125 | int sum = 0; | |
126 | for_each_possible_cpu(i) | |
127 | sum += per_cpu(nr_inodes, i); | |
128 | return sum < 0 ? 0 : sum; | |
cffbc8aa DC |
129 | } |
130 | ||
131 | static inline int get_nr_inodes_unused(void) | |
132 | { | |
86c8749e | 133 | return inodes_stat.nr_unused; |
cffbc8aa DC |
134 | } |
135 | ||
136 | int get_nr_dirty_inodes(void) | |
137 | { | |
3e880fb5 | 138 | /* not actually dirty inodes, but a wild approximation */ |
cffbc8aa DC |
139 | int nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); |
140 | return nr_dirty > 0 ? nr_dirty : 0; | |
cffbc8aa DC |
141 | } |
142 | ||
143 | /* | |
144 | * Handle nr_inode sysctl | |
145 | */ | |
146 | #ifdef CONFIG_SYSCTL | |
147 | int proc_nr_inodes(ctl_table *table, int write, | |
148 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
149 | { | |
150 | inodes_stat.nr_inodes = get_nr_inodes(); | |
cffbc8aa DC |
151 | return proc_dointvec(table, write, buffer, lenp, ppos); |
152 | } | |
153 | #endif | |
154 | ||
2cb1599f DC |
155 | /** |
156 | * inode_init_always - perform inode structure intialisation | |
0bc02f3f RD |
157 | * @sb: superblock inode belongs to |
158 | * @inode: inode to initialise | |
2cb1599f DC |
159 | * |
160 | * These are initializations that need to be done on every inode | |
161 | * allocation as the fields are not initialised by slab allocation. | |
162 | */ | |
54e34621 | 163 | int inode_init_always(struct super_block *sb, struct inode *inode) |
1da177e4 | 164 | { |
f5e54d6e | 165 | static const struct address_space_operations empty_aops; |
6e1d5dcc | 166 | static const struct inode_operations empty_iops; |
99ac48f5 | 167 | static const struct file_operations empty_fops; |
6b3304b5 | 168 | struct address_space *const mapping = &inode->i_data; |
2cb1599f DC |
169 | |
170 | inode->i_sb = sb; | |
171 | inode->i_blkbits = sb->s_blocksize_bits; | |
172 | inode->i_flags = 0; | |
173 | atomic_set(&inode->i_count, 1); | |
174 | inode->i_op = &empty_iops; | |
175 | inode->i_fop = &empty_fops; | |
176 | inode->i_nlink = 1; | |
56ff5efa AV |
177 | inode->i_uid = 0; |
178 | inode->i_gid = 0; | |
2cb1599f DC |
179 | atomic_set(&inode->i_writecount, 0); |
180 | inode->i_size = 0; | |
181 | inode->i_blocks = 0; | |
182 | inode->i_bytes = 0; | |
183 | inode->i_generation = 0; | |
1da177e4 | 184 | #ifdef CONFIG_QUOTA |
2cb1599f | 185 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); |
1da177e4 | 186 | #endif |
2cb1599f DC |
187 | inode->i_pipe = NULL; |
188 | inode->i_bdev = NULL; | |
189 | inode->i_cdev = NULL; | |
190 | inode->i_rdev = 0; | |
191 | inode->dirtied_when = 0; | |
6146f0d5 MZ |
192 | |
193 | if (security_inode_alloc(inode)) | |
54e34621 | 194 | goto out; |
2cb1599f DC |
195 | spin_lock_init(&inode->i_lock); |
196 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
197 | ||
198 | mutex_init(&inode->i_mutex); | |
199 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); | |
200 | ||
201 | init_rwsem(&inode->i_alloc_sem); | |
202 | lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key); | |
203 | ||
204 | mapping->a_ops = &empty_aops; | |
205 | mapping->host = inode; | |
206 | mapping->flags = 0; | |
3c1d4378 | 207 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); |
2cb1599f DC |
208 | mapping->assoc_mapping = NULL; |
209 | mapping->backing_dev_info = &default_backing_dev_info; | |
210 | mapping->writeback_index = 0; | |
211 | ||
212 | /* | |
213 | * If the block_device provides a backing_dev_info for client | |
214 | * inodes then use that. Otherwise the inode share the bdev's | |
215 | * backing_dev_info. | |
216 | */ | |
217 | if (sb->s_bdev) { | |
218 | struct backing_dev_info *bdi; | |
219 | ||
2c96ce9f | 220 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; |
2cb1599f DC |
221 | mapping->backing_dev_info = bdi; |
222 | } | |
223 | inode->i_private = NULL; | |
224 | inode->i_mapping = mapping; | |
f19d4a8f AV |
225 | #ifdef CONFIG_FS_POSIX_ACL |
226 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; | |
227 | #endif | |
2cb1599f | 228 | |
3be25f49 EP |
229 | #ifdef CONFIG_FSNOTIFY |
230 | inode->i_fsnotify_mask = 0; | |
231 | #endif | |
232 | ||
3e880fb5 | 233 | this_cpu_inc(nr_inodes); |
cffbc8aa | 234 | |
54e34621 | 235 | return 0; |
54e34621 CH |
236 | out: |
237 | return -ENOMEM; | |
1da177e4 | 238 | } |
2cb1599f DC |
239 | EXPORT_SYMBOL(inode_init_always); |
240 | ||
241 | static struct inode *alloc_inode(struct super_block *sb) | |
242 | { | |
243 | struct inode *inode; | |
244 | ||
245 | if (sb->s_op->alloc_inode) | |
246 | inode = sb->s_op->alloc_inode(sb); | |
247 | else | |
248 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
249 | ||
54e34621 CH |
250 | if (!inode) |
251 | return NULL; | |
252 | ||
253 | if (unlikely(inode_init_always(sb, inode))) { | |
254 | if (inode->i_sb->s_op->destroy_inode) | |
255 | inode->i_sb->s_op->destroy_inode(inode); | |
256 | else | |
257 | kmem_cache_free(inode_cachep, inode); | |
258 | return NULL; | |
259 | } | |
260 | ||
261 | return inode; | |
2cb1599f | 262 | } |
1da177e4 | 263 | |
ff0c7d15 NP |
264 | void free_inode_nonrcu(struct inode *inode) |
265 | { | |
266 | kmem_cache_free(inode_cachep, inode); | |
267 | } | |
268 | EXPORT_SYMBOL(free_inode_nonrcu); | |
269 | ||
2e00c97e | 270 | void __destroy_inode(struct inode *inode) |
1da177e4 | 271 | { |
b7542f8c | 272 | BUG_ON(inode_has_buffers(inode)); |
1da177e4 | 273 | security_inode_free(inode); |
3be25f49 | 274 | fsnotify_inode_delete(inode); |
f19d4a8f AV |
275 | #ifdef CONFIG_FS_POSIX_ACL |
276 | if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) | |
277 | posix_acl_release(inode->i_acl); | |
278 | if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) | |
279 | posix_acl_release(inode->i_default_acl); | |
280 | #endif | |
3e880fb5 | 281 | this_cpu_dec(nr_inodes); |
2e00c97e CH |
282 | } |
283 | EXPORT_SYMBOL(__destroy_inode); | |
284 | ||
fa0d7e3d NP |
285 | static void i_callback(struct rcu_head *head) |
286 | { | |
287 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
288 | INIT_LIST_HEAD(&inode->i_dentry); | |
289 | kmem_cache_free(inode_cachep, inode); | |
290 | } | |
291 | ||
56b0dacf | 292 | static void destroy_inode(struct inode *inode) |
2e00c97e | 293 | { |
7ccf19a8 | 294 | BUG_ON(!list_empty(&inode->i_lru)); |
2e00c97e | 295 | __destroy_inode(inode); |
1da177e4 LT |
296 | if (inode->i_sb->s_op->destroy_inode) |
297 | inode->i_sb->s_op->destroy_inode(inode); | |
298 | else | |
fa0d7e3d | 299 | call_rcu(&inode->i_rcu, i_callback); |
1da177e4 | 300 | } |
1da177e4 | 301 | |
2aa15890 MS |
302 | void address_space_init_once(struct address_space *mapping) |
303 | { | |
304 | memset(mapping, 0, sizeof(*mapping)); | |
305 | INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC); | |
306 | spin_lock_init(&mapping->tree_lock); | |
307 | spin_lock_init(&mapping->i_mmap_lock); | |
308 | INIT_LIST_HEAD(&mapping->private_list); | |
309 | spin_lock_init(&mapping->private_lock); | |
310 | INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap); | |
311 | INIT_LIST_HEAD(&mapping->i_mmap_nonlinear); | |
312 | mutex_init(&mapping->unmap_mutex); | |
313 | } | |
314 | EXPORT_SYMBOL(address_space_init_once); | |
315 | ||
1da177e4 LT |
316 | /* |
317 | * These are initializations that only need to be done | |
318 | * once, because the fields are idempotent across use | |
319 | * of the inode, so let the slab aware of that. | |
320 | */ | |
321 | void inode_init_once(struct inode *inode) | |
322 | { | |
323 | memset(inode, 0, sizeof(*inode)); | |
324 | INIT_HLIST_NODE(&inode->i_hash); | |
325 | INIT_LIST_HEAD(&inode->i_dentry); | |
326 | INIT_LIST_HEAD(&inode->i_devices); | |
7ccf19a8 NP |
327 | INIT_LIST_HEAD(&inode->i_wb_list); |
328 | INIT_LIST_HEAD(&inode->i_lru); | |
2aa15890 | 329 | address_space_init_once(&inode->i_data); |
1da177e4 | 330 | i_size_ordered_init(inode); |
3be25f49 | 331 | #ifdef CONFIG_FSNOTIFY |
e61ce867 | 332 | INIT_HLIST_HEAD(&inode->i_fsnotify_marks); |
3be25f49 | 333 | #endif |
1da177e4 | 334 | } |
1da177e4 LT |
335 | EXPORT_SYMBOL(inode_init_once); |
336 | ||
51cc5068 | 337 | static void init_once(void *foo) |
1da177e4 | 338 | { |
6b3304b5 | 339 | struct inode *inode = (struct inode *) foo; |
1da177e4 | 340 | |
a35afb83 | 341 | inode_init_once(inode); |
1da177e4 LT |
342 | } |
343 | ||
344 | /* | |
250df6ed | 345 | * inode->i_lock must be held |
1da177e4 | 346 | */ |
6b3304b5 | 347 | void __iget(struct inode *inode) |
1da177e4 | 348 | { |
9e38d86f NP |
349 | atomic_inc(&inode->i_count); |
350 | } | |
2e147f1e | 351 | |
7de9c6ee AV |
352 | /* |
353 | * get additional reference to inode; caller must already hold one. | |
354 | */ | |
355 | void ihold(struct inode *inode) | |
356 | { | |
357 | WARN_ON(atomic_inc_return(&inode->i_count) < 2); | |
358 | } | |
359 | EXPORT_SYMBOL(ihold); | |
360 | ||
9e38d86f NP |
361 | static void inode_lru_list_add(struct inode *inode) |
362 | { | |
02afc410 | 363 | spin_lock(&inode_lru_lock); |
7ccf19a8 NP |
364 | if (list_empty(&inode->i_lru)) { |
365 | list_add(&inode->i_lru, &inode_lru); | |
86c8749e | 366 | inodes_stat.nr_unused++; |
9e38d86f | 367 | } |
02afc410 | 368 | spin_unlock(&inode_lru_lock); |
9e38d86f | 369 | } |
2e147f1e | 370 | |
9e38d86f NP |
371 | static void inode_lru_list_del(struct inode *inode) |
372 | { | |
02afc410 | 373 | spin_lock(&inode_lru_lock); |
7ccf19a8 NP |
374 | if (!list_empty(&inode->i_lru)) { |
375 | list_del_init(&inode->i_lru); | |
86c8749e | 376 | inodes_stat.nr_unused--; |
9e38d86f | 377 | } |
02afc410 | 378 | spin_unlock(&inode_lru_lock); |
1da177e4 LT |
379 | } |
380 | ||
646ec461 CH |
381 | static inline void __inode_sb_list_add(struct inode *inode) |
382 | { | |
383 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); | |
1da177e4 LT |
384 | } |
385 | ||
646ec461 CH |
386 | /** |
387 | * inode_sb_list_add - add inode to the superblock list of inodes | |
388 | * @inode: inode to add | |
389 | */ | |
390 | void inode_sb_list_add(struct inode *inode) | |
391 | { | |
392 | spin_lock(&inode_lock); | |
393 | __inode_sb_list_add(inode); | |
394 | spin_unlock(&inode_lock); | |
395 | } | |
396 | EXPORT_SYMBOL_GPL(inode_sb_list_add); | |
397 | ||
398 | static inline void __inode_sb_list_del(struct inode *inode) | |
399 | { | |
400 | list_del_init(&inode->i_sb_list); | |
401 | } | |
402 | ||
4c51acbc DC |
403 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
404 | { | |
405 | unsigned long tmp; | |
406 | ||
407 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
408 | L1_CACHE_BYTES; | |
409 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS); | |
410 | return tmp & I_HASHMASK; | |
411 | } | |
412 | ||
413 | /** | |
414 | * __insert_inode_hash - hash an inode | |
415 | * @inode: unhashed inode | |
416 | * @hashval: unsigned long value used to locate this object in the | |
417 | * inode_hashtable. | |
418 | * | |
419 | * Add an inode to the inode hash for this superblock. | |
420 | */ | |
421 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
422 | { | |
646ec461 CH |
423 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); |
424 | ||
4c51acbc | 425 | spin_lock(&inode_lock); |
250df6ed | 426 | spin_lock(&inode->i_lock); |
646ec461 | 427 | hlist_add_head(&inode->i_hash, b); |
250df6ed | 428 | spin_unlock(&inode->i_lock); |
4c51acbc DC |
429 | spin_unlock(&inode_lock); |
430 | } | |
431 | EXPORT_SYMBOL(__insert_inode_hash); | |
432 | ||
4c51acbc DC |
433 | /** |
434 | * remove_inode_hash - remove an inode from the hash | |
435 | * @inode: inode to unhash | |
436 | * | |
437 | * Remove an inode from the superblock. | |
438 | */ | |
439 | void remove_inode_hash(struct inode *inode) | |
440 | { | |
441 | spin_lock(&inode_lock); | |
250df6ed | 442 | spin_lock(&inode->i_lock); |
4c51acbc | 443 | hlist_del_init(&inode->i_hash); |
250df6ed | 444 | spin_unlock(&inode->i_lock); |
4c51acbc DC |
445 | spin_unlock(&inode_lock); |
446 | } | |
447 | EXPORT_SYMBOL(remove_inode_hash); | |
448 | ||
b0683aa6 AV |
449 | void end_writeback(struct inode *inode) |
450 | { | |
451 | might_sleep(); | |
452 | BUG_ON(inode->i_data.nrpages); | |
453 | BUG_ON(!list_empty(&inode->i_data.private_list)); | |
454 | BUG_ON(!(inode->i_state & I_FREEING)); | |
455 | BUG_ON(inode->i_state & I_CLEAR); | |
456 | inode_sync_wait(inode); | |
fa0d7e3d | 457 | /* don't need i_lock here, no concurrent mods to i_state */ |
b0683aa6 AV |
458 | inode->i_state = I_FREEING | I_CLEAR; |
459 | } | |
460 | EXPORT_SYMBOL(end_writeback); | |
461 | ||
b2b2af8e DC |
462 | /* |
463 | * Free the inode passed in, removing it from the lists it is still connected | |
464 | * to. We remove any pages still attached to the inode and wait for any IO that | |
465 | * is still in progress before finally destroying the inode. | |
466 | * | |
467 | * An inode must already be marked I_FREEING so that we avoid the inode being | |
468 | * moved back onto lists if we race with other code that manipulates the lists | |
469 | * (e.g. writeback_single_inode). The caller is responsible for setting this. | |
470 | * | |
471 | * An inode must already be removed from the LRU list before being evicted from | |
472 | * the cache. This should occur atomically with setting the I_FREEING state | |
473 | * flag, so no inodes here should ever be on the LRU when being evicted. | |
474 | */ | |
644da596 | 475 | static void evict(struct inode *inode) |
b4272d4c AV |
476 | { |
477 | const struct super_operations *op = inode->i_sb->s_op; | |
478 | ||
b2b2af8e DC |
479 | BUG_ON(!(inode->i_state & I_FREEING)); |
480 | BUG_ON(!list_empty(&inode->i_lru)); | |
481 | ||
482 | spin_lock(&inode_lock); | |
483 | list_del_init(&inode->i_wb_list); | |
484 | __inode_sb_list_del(inode); | |
485 | spin_unlock(&inode_lock); | |
486 | ||
be7ce416 AV |
487 | if (op->evict_inode) { |
488 | op->evict_inode(inode); | |
b4272d4c AV |
489 | } else { |
490 | if (inode->i_data.nrpages) | |
491 | truncate_inode_pages(&inode->i_data, 0); | |
30140837 | 492 | end_writeback(inode); |
b4272d4c | 493 | } |
661074e9 AV |
494 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
495 | bd_forget(inode); | |
496 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) | |
497 | cd_forget(inode); | |
b2b2af8e DC |
498 | |
499 | remove_inode_hash(inode); | |
500 | ||
501 | spin_lock(&inode->i_lock); | |
502 | wake_up_bit(&inode->i_state, __I_NEW); | |
503 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); | |
504 | spin_unlock(&inode->i_lock); | |
505 | ||
506 | destroy_inode(inode); | |
b4272d4c AV |
507 | } |
508 | ||
1da177e4 LT |
509 | /* |
510 | * dispose_list - dispose of the contents of a local list | |
511 | * @head: the head of the list to free | |
512 | * | |
513 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
514 | * need to worry about list corruption and SMP locks. | |
515 | */ | |
516 | static void dispose_list(struct list_head *head) | |
517 | { | |
1da177e4 LT |
518 | while (!list_empty(head)) { |
519 | struct inode *inode; | |
520 | ||
7ccf19a8 NP |
521 | inode = list_first_entry(head, struct inode, i_lru); |
522 | list_del_init(&inode->i_lru); | |
1da177e4 | 523 | |
644da596 | 524 | evict(inode); |
1da177e4 | 525 | } |
1da177e4 LT |
526 | } |
527 | ||
63997e98 AV |
528 | /** |
529 | * evict_inodes - evict all evictable inodes for a superblock | |
530 | * @sb: superblock to operate on | |
531 | * | |
532 | * Make sure that no inodes with zero refcount are retained. This is | |
533 | * called by superblock shutdown after having MS_ACTIVE flag removed, | |
534 | * so any inode reaching zero refcount during or after that call will | |
535 | * be immediately evicted. | |
1da177e4 | 536 | */ |
63997e98 | 537 | void evict_inodes(struct super_block *sb) |
1da177e4 | 538 | { |
63997e98 AV |
539 | struct inode *inode, *next; |
540 | LIST_HEAD(dispose); | |
1da177e4 | 541 | |
63997e98 AV |
542 | spin_lock(&inode_lock); |
543 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { | |
544 | if (atomic_read(&inode->i_count)) | |
aabb8fdb | 545 | continue; |
250df6ed DC |
546 | |
547 | spin_lock(&inode->i_lock); | |
548 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
549 | spin_unlock(&inode->i_lock); | |
1da177e4 | 550 | continue; |
250df6ed | 551 | } |
63997e98 AV |
552 | |
553 | inode->i_state |= I_FREEING; | |
02afc410 | 554 | inode_lru_list_del(inode); |
250df6ed | 555 | spin_unlock(&inode->i_lock); |
02afc410 | 556 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 557 | } |
63997e98 AV |
558 | spin_unlock(&inode_lock); |
559 | ||
560 | dispose_list(&dispose); | |
bab1d944 CH |
561 | |
562 | /* | |
563 | * Cycle through iprune_sem to make sure any inode that prune_icache | |
564 | * moved off the list before we took the lock has been fully torn | |
565 | * down. | |
566 | */ | |
567 | down_write(&iprune_sem); | |
63997e98 | 568 | up_write(&iprune_sem); |
1da177e4 LT |
569 | } |
570 | ||
1da177e4 | 571 | /** |
a0318786 CH |
572 | * invalidate_inodes - attempt to free all inodes on a superblock |
573 | * @sb: superblock to operate on | |
93b270f7 | 574 | * @kill_dirty: flag to guide handling of dirty inodes |
1da177e4 | 575 | * |
a0318786 CH |
576 | * Attempts to free all inodes for a given superblock. If there were any |
577 | * busy inodes return a non-zero value, else zero. | |
93b270f7 N |
578 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat |
579 | * them as busy. | |
1da177e4 | 580 | */ |
93b270f7 | 581 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) |
1da177e4 | 582 | { |
cffbc8aa | 583 | int busy = 0; |
a0318786 CH |
584 | struct inode *inode, *next; |
585 | LIST_HEAD(dispose); | |
1da177e4 | 586 | |
1da177e4 | 587 | spin_lock(&inode_lock); |
a0318786 | 588 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
250df6ed DC |
589 | spin_lock(&inode->i_lock); |
590 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
591 | spin_unlock(&inode->i_lock); | |
aabb8fdb | 592 | continue; |
250df6ed | 593 | } |
93b270f7 | 594 | if (inode->i_state & I_DIRTY && !kill_dirty) { |
250df6ed | 595 | spin_unlock(&inode->i_lock); |
93b270f7 N |
596 | busy = 1; |
597 | continue; | |
598 | } | |
99a38919 | 599 | if (atomic_read(&inode->i_count)) { |
250df6ed | 600 | spin_unlock(&inode->i_lock); |
99a38919 | 601 | busy = 1; |
1da177e4 LT |
602 | continue; |
603 | } | |
99a38919 | 604 | |
99a38919 | 605 | inode->i_state |= I_FREEING; |
02afc410 | 606 | inode_lru_list_del(inode); |
250df6ed | 607 | spin_unlock(&inode->i_lock); |
02afc410 | 608 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 609 | } |
1da177e4 LT |
610 | spin_unlock(&inode_lock); |
611 | ||
a0318786 | 612 | dispose_list(&dispose); |
1da177e4 LT |
613 | |
614 | return busy; | |
615 | } | |
1da177e4 LT |
616 | |
617 | static int can_unuse(struct inode *inode) | |
618 | { | |
9e38d86f | 619 | if (inode->i_state & ~I_REFERENCED) |
1da177e4 LT |
620 | return 0; |
621 | if (inode_has_buffers(inode)) | |
622 | return 0; | |
623 | if (atomic_read(&inode->i_count)) | |
624 | return 0; | |
625 | if (inode->i_data.nrpages) | |
626 | return 0; | |
627 | return 1; | |
628 | } | |
629 | ||
630 | /* | |
9e38d86f | 631 | * Scan `goal' inodes on the unused list for freeable ones. They are moved to a |
02afc410 | 632 | * temporary list and then are freed outside inode_lru_lock by dispose_list(). |
1da177e4 LT |
633 | * |
634 | * Any inodes which are pinned purely because of attached pagecache have their | |
9e38d86f NP |
635 | * pagecache removed. If the inode has metadata buffers attached to |
636 | * mapping->private_list then try to remove them. | |
1da177e4 | 637 | * |
9e38d86f NP |
638 | * If the inode has the I_REFERENCED flag set, then it means that it has been |
639 | * used recently - the flag is set in iput_final(). When we encounter such an | |
640 | * inode, clear the flag and move it to the back of the LRU so it gets another | |
641 | * pass through the LRU before it gets reclaimed. This is necessary because of | |
642 | * the fact we are doing lazy LRU updates to minimise lock contention so the | |
643 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes | |
644 | * with this flag set because they are the inodes that are out of order. | |
1da177e4 LT |
645 | */ |
646 | static void prune_icache(int nr_to_scan) | |
647 | { | |
648 | LIST_HEAD(freeable); | |
1da177e4 LT |
649 | int nr_scanned; |
650 | unsigned long reap = 0; | |
651 | ||
88e0fbc4 | 652 | down_read(&iprune_sem); |
1da177e4 | 653 | spin_lock(&inode_lock); |
02afc410 | 654 | spin_lock(&inode_lru_lock); |
1da177e4 LT |
655 | for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) { |
656 | struct inode *inode; | |
657 | ||
7ccf19a8 | 658 | if (list_empty(&inode_lru)) |
1da177e4 LT |
659 | break; |
660 | ||
7ccf19a8 | 661 | inode = list_entry(inode_lru.prev, struct inode, i_lru); |
1da177e4 | 662 | |
02afc410 DC |
663 | /* |
664 | * we are inverting the inode_lru_lock/inode->i_lock here, | |
665 | * so use a trylock. If we fail to get the lock, just move the | |
666 | * inode to the back of the list so we don't spin on it. | |
667 | */ | |
668 | if (!spin_trylock(&inode->i_lock)) { | |
669 | list_move(&inode->i_lru, &inode_lru); | |
670 | continue; | |
671 | } | |
672 | ||
9e38d86f NP |
673 | /* |
674 | * Referenced or dirty inodes are still in use. Give them | |
675 | * another pass through the LRU as we canot reclaim them now. | |
676 | */ | |
677 | if (atomic_read(&inode->i_count) || | |
678 | (inode->i_state & ~I_REFERENCED)) { | |
250df6ed | 679 | spin_unlock(&inode->i_lock); |
7ccf19a8 | 680 | list_del_init(&inode->i_lru); |
86c8749e | 681 | inodes_stat.nr_unused--; |
9e38d86f NP |
682 | continue; |
683 | } | |
684 | ||
685 | /* recently referenced inodes get one more pass */ | |
686 | if (inode->i_state & I_REFERENCED) { | |
9e38d86f | 687 | inode->i_state &= ~I_REFERENCED; |
250df6ed DC |
688 | spin_unlock(&inode->i_lock); |
689 | list_move(&inode->i_lru, &inode_lru); | |
1da177e4 LT |
690 | continue; |
691 | } | |
692 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { | |
693 | __iget(inode); | |
250df6ed | 694 | spin_unlock(&inode->i_lock); |
02afc410 | 695 | spin_unlock(&inode_lru_lock); |
1da177e4 LT |
696 | spin_unlock(&inode_lock); |
697 | if (remove_inode_buffers(inode)) | |
fc0ecff6 AM |
698 | reap += invalidate_mapping_pages(&inode->i_data, |
699 | 0, -1); | |
1da177e4 LT |
700 | iput(inode); |
701 | spin_lock(&inode_lock); | |
02afc410 | 702 | spin_lock(&inode_lru_lock); |
1da177e4 | 703 | |
7ccf19a8 NP |
704 | if (inode != list_entry(inode_lru.next, |
705 | struct inode, i_lru)) | |
1da177e4 | 706 | continue; /* wrong inode or list_empty */ |
02afc410 DC |
707 | /* avoid lock inversions with trylock */ |
708 | if (!spin_trylock(&inode->i_lock)) | |
709 | continue; | |
250df6ed DC |
710 | if (!can_unuse(inode)) { |
711 | spin_unlock(&inode->i_lock); | |
1da177e4 | 712 | continue; |
250df6ed | 713 | } |
1da177e4 | 714 | } |
7ef0d737 | 715 | WARN_ON(inode->i_state & I_NEW); |
1da177e4 | 716 | inode->i_state |= I_FREEING; |
250df6ed | 717 | spin_unlock(&inode->i_lock); |
7ccf19a8 | 718 | |
7ccf19a8 | 719 | list_move(&inode->i_lru, &freeable); |
86c8749e | 720 | inodes_stat.nr_unused--; |
1da177e4 | 721 | } |
f8891e5e CL |
722 | if (current_is_kswapd()) |
723 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
724 | else | |
725 | __count_vm_events(PGINODESTEAL, reap); | |
02afc410 | 726 | spin_unlock(&inode_lru_lock); |
1da177e4 LT |
727 | spin_unlock(&inode_lock); |
728 | ||
729 | dispose_list(&freeable); | |
88e0fbc4 | 730 | up_read(&iprune_sem); |
1da177e4 LT |
731 | } |
732 | ||
733 | /* | |
734 | * shrink_icache_memory() will attempt to reclaim some unused inodes. Here, | |
735 | * "unused" means that no dentries are referring to the inodes: the files are | |
736 | * not open and the dcache references to those inodes have already been | |
737 | * reclaimed. | |
738 | * | |
739 | * This function is passed the number of inodes to scan, and it returns the | |
740 | * total number of remaining possibly-reclaimable inodes. | |
741 | */ | |
7f8275d0 | 742 | static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
1da177e4 LT |
743 | { |
744 | if (nr) { | |
745 | /* | |
746 | * Nasty deadlock avoidance. We may hold various FS locks, | |
747 | * and we don't want to recurse into the FS that called us | |
748 | * in clear_inode() and friends.. | |
6b3304b5 | 749 | */ |
1da177e4 LT |
750 | if (!(gfp_mask & __GFP_FS)) |
751 | return -1; | |
752 | prune_icache(nr); | |
753 | } | |
cffbc8aa | 754 | return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure; |
1da177e4 LT |
755 | } |
756 | ||
8e1f936b RR |
757 | static struct shrinker icache_shrinker = { |
758 | .shrink = shrink_icache_memory, | |
759 | .seeks = DEFAULT_SEEKS, | |
760 | }; | |
761 | ||
1da177e4 LT |
762 | static void __wait_on_freeing_inode(struct inode *inode); |
763 | /* | |
764 | * Called with the inode lock held. | |
1da177e4 | 765 | */ |
6b3304b5 MK |
766 | static struct inode *find_inode(struct super_block *sb, |
767 | struct hlist_head *head, | |
768 | int (*test)(struct inode *, void *), | |
769 | void *data) | |
1da177e4 LT |
770 | { |
771 | struct hlist_node *node; | |
6b3304b5 | 772 | struct inode *inode = NULL; |
1da177e4 LT |
773 | |
774 | repeat: | |
c5c8be3c | 775 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
776 | if (inode->i_sb != sb) |
777 | continue; | |
778 | if (!test(inode, data)) | |
779 | continue; | |
250df6ed | 780 | spin_lock(&inode->i_lock); |
a4ffdde6 | 781 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
782 | __wait_on_freeing_inode(inode); |
783 | goto repeat; | |
784 | } | |
f7899bd5 | 785 | __iget(inode); |
250df6ed | 786 | spin_unlock(&inode->i_lock); |
f7899bd5 | 787 | return inode; |
1da177e4 | 788 | } |
f7899bd5 | 789 | return NULL; |
1da177e4 LT |
790 | } |
791 | ||
792 | /* | |
793 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
794 | * iget_locked for details. | |
795 | */ | |
6b3304b5 MK |
796 | static struct inode *find_inode_fast(struct super_block *sb, |
797 | struct hlist_head *head, unsigned long ino) | |
1da177e4 LT |
798 | { |
799 | struct hlist_node *node; | |
6b3304b5 | 800 | struct inode *inode = NULL; |
1da177e4 LT |
801 | |
802 | repeat: | |
c5c8be3c | 803 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
804 | if (inode->i_ino != ino) |
805 | continue; | |
806 | if (inode->i_sb != sb) | |
807 | continue; | |
250df6ed | 808 | spin_lock(&inode->i_lock); |
a4ffdde6 | 809 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
810 | __wait_on_freeing_inode(inode); |
811 | goto repeat; | |
812 | } | |
f7899bd5 | 813 | __iget(inode); |
250df6ed | 814 | spin_unlock(&inode->i_lock); |
f7899bd5 | 815 | return inode; |
1da177e4 | 816 | } |
f7899bd5 | 817 | return NULL; |
8290c35f DC |
818 | } |
819 | ||
f991bd2e ED |
820 | /* |
821 | * Each cpu owns a range of LAST_INO_BATCH numbers. | |
822 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, | |
823 | * to renew the exhausted range. | |
8290c35f | 824 | * |
f991bd2e ED |
825 | * This does not significantly increase overflow rate because every CPU can |
826 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is | |
827 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the | |
828 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase | |
829 | * overflow rate by 2x, which does not seem too significant. | |
830 | * | |
831 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
832 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
833 | * here to attempt to avoid that. | |
8290c35f | 834 | */ |
f991bd2e ED |
835 | #define LAST_INO_BATCH 1024 |
836 | static DEFINE_PER_CPU(unsigned int, last_ino); | |
837 | ||
85fe4025 | 838 | unsigned int get_next_ino(void) |
8290c35f | 839 | { |
f991bd2e ED |
840 | unsigned int *p = &get_cpu_var(last_ino); |
841 | unsigned int res = *p; | |
8290c35f | 842 | |
f991bd2e ED |
843 | #ifdef CONFIG_SMP |
844 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { | |
845 | static atomic_t shared_last_ino; | |
846 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); | |
847 | ||
848 | res = next - LAST_INO_BATCH; | |
849 | } | |
850 | #endif | |
851 | ||
852 | *p = ++res; | |
853 | put_cpu_var(last_ino); | |
854 | return res; | |
8290c35f | 855 | } |
85fe4025 | 856 | EXPORT_SYMBOL(get_next_ino); |
8290c35f | 857 | |
1da177e4 LT |
858 | /** |
859 | * new_inode - obtain an inode | |
860 | * @sb: superblock | |
861 | * | |
769848c0 | 862 | * Allocates a new inode for given superblock. The default gfp_mask |
3c1d4378 | 863 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
769848c0 MG |
864 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
865 | * for the page cache are not reclaimable or migratable, | |
866 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
867 | * newly created inode's mapping | |
868 | * | |
1da177e4 LT |
869 | */ |
870 | struct inode *new_inode(struct super_block *sb) | |
871 | { | |
6b3304b5 | 872 | struct inode *inode; |
1da177e4 LT |
873 | |
874 | spin_lock_prefetch(&inode_lock); | |
6b3304b5 | 875 | |
1da177e4 LT |
876 | inode = alloc_inode(sb); |
877 | if (inode) { | |
878 | spin_lock(&inode_lock); | |
250df6ed | 879 | spin_lock(&inode->i_lock); |
1da177e4 | 880 | inode->i_state = 0; |
250df6ed DC |
881 | spin_unlock(&inode->i_lock); |
882 | __inode_sb_list_add(inode); | |
1da177e4 LT |
883 | spin_unlock(&inode_lock); |
884 | } | |
885 | return inode; | |
886 | } | |
1da177e4 LT |
887 | EXPORT_SYMBOL(new_inode); |
888 | ||
250df6ed DC |
889 | /** |
890 | * unlock_new_inode - clear the I_NEW state and wake up any waiters | |
891 | * @inode: new inode to unlock | |
892 | * | |
893 | * Called when the inode is fully initialised to clear the new state of the | |
894 | * inode and wake up anyone waiting for the inode to finish initialisation. | |
895 | */ | |
1da177e4 LT |
896 | void unlock_new_inode(struct inode *inode) |
897 | { | |
14358e6d | 898 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
a3314a0e | 899 | if (S_ISDIR(inode->i_mode)) { |
1e89a5e1 PZ |
900 | struct file_system_type *type = inode->i_sb->s_type; |
901 | ||
9a7aa12f JK |
902 | /* Set new key only if filesystem hasn't already changed it */ |
903 | if (!lockdep_match_class(&inode->i_mutex, | |
904 | &type->i_mutex_key)) { | |
905 | /* | |
906 | * ensure nobody is actually holding i_mutex | |
907 | */ | |
908 | mutex_destroy(&inode->i_mutex); | |
909 | mutex_init(&inode->i_mutex); | |
910 | lockdep_set_class(&inode->i_mutex, | |
911 | &type->i_mutex_dir_key); | |
912 | } | |
1e89a5e1 | 913 | } |
14358e6d | 914 | #endif |
250df6ed | 915 | spin_lock(&inode->i_lock); |
eaff8079 CH |
916 | WARN_ON(!(inode->i_state & I_NEW)); |
917 | inode->i_state &= ~I_NEW; | |
250df6ed DC |
918 | wake_up_bit(&inode->i_state, __I_NEW); |
919 | spin_unlock(&inode->i_lock); | |
1da177e4 | 920 | } |
1da177e4 LT |
921 | EXPORT_SYMBOL(unlock_new_inode); |
922 | ||
923 | /* | |
924 | * This is called without the inode lock held.. Be careful. | |
925 | * | |
926 | * We no longer cache the sb_flags in i_flags - see fs.h | |
927 | * -- rmk@arm.uk.linux.org | |
928 | */ | |
6b3304b5 MK |
929 | static struct inode *get_new_inode(struct super_block *sb, |
930 | struct hlist_head *head, | |
931 | int (*test)(struct inode *, void *), | |
932 | int (*set)(struct inode *, void *), | |
933 | void *data) | |
1da177e4 | 934 | { |
6b3304b5 | 935 | struct inode *inode; |
1da177e4 LT |
936 | |
937 | inode = alloc_inode(sb); | |
938 | if (inode) { | |
6b3304b5 | 939 | struct inode *old; |
1da177e4 LT |
940 | |
941 | spin_lock(&inode_lock); | |
942 | /* We released the lock, so.. */ | |
943 | old = find_inode(sb, head, test, data); | |
944 | if (!old) { | |
945 | if (set(inode, data)) | |
946 | goto set_failed; | |
947 | ||
250df6ed DC |
948 | spin_lock(&inode->i_lock); |
949 | inode->i_state = I_NEW; | |
646ec461 | 950 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 951 | spin_unlock(&inode->i_lock); |
646ec461 | 952 | __inode_sb_list_add(inode); |
1da177e4 LT |
953 | spin_unlock(&inode_lock); |
954 | ||
955 | /* Return the locked inode with I_NEW set, the | |
956 | * caller is responsible for filling in the contents | |
957 | */ | |
958 | return inode; | |
959 | } | |
960 | ||
961 | /* | |
962 | * Uhhuh, somebody else created the same inode under | |
963 | * us. Use the old inode instead of the one we just | |
964 | * allocated. | |
965 | */ | |
1da177e4 LT |
966 | spin_unlock(&inode_lock); |
967 | destroy_inode(inode); | |
968 | inode = old; | |
969 | wait_on_inode(inode); | |
970 | } | |
971 | return inode; | |
972 | ||
973 | set_failed: | |
974 | spin_unlock(&inode_lock); | |
975 | destroy_inode(inode); | |
976 | return NULL; | |
977 | } | |
978 | ||
979 | /* | |
980 | * get_new_inode_fast is the fast path version of get_new_inode, see the | |
981 | * comment at iget_locked for details. | |
982 | */ | |
6b3304b5 MK |
983 | static struct inode *get_new_inode_fast(struct super_block *sb, |
984 | struct hlist_head *head, unsigned long ino) | |
1da177e4 | 985 | { |
6b3304b5 | 986 | struct inode *inode; |
1da177e4 LT |
987 | |
988 | inode = alloc_inode(sb); | |
989 | if (inode) { | |
6b3304b5 | 990 | struct inode *old; |
1da177e4 LT |
991 | |
992 | spin_lock(&inode_lock); | |
993 | /* We released the lock, so.. */ | |
994 | old = find_inode_fast(sb, head, ino); | |
995 | if (!old) { | |
996 | inode->i_ino = ino; | |
250df6ed DC |
997 | spin_lock(&inode->i_lock); |
998 | inode->i_state = I_NEW; | |
646ec461 | 999 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1000 | spin_unlock(&inode->i_lock); |
646ec461 | 1001 | __inode_sb_list_add(inode); |
1da177e4 LT |
1002 | spin_unlock(&inode_lock); |
1003 | ||
1004 | /* Return the locked inode with I_NEW set, the | |
1005 | * caller is responsible for filling in the contents | |
1006 | */ | |
1007 | return inode; | |
1008 | } | |
1009 | ||
1010 | /* | |
1011 | * Uhhuh, somebody else created the same inode under | |
1012 | * us. Use the old inode instead of the one we just | |
1013 | * allocated. | |
1014 | */ | |
1da177e4 LT |
1015 | spin_unlock(&inode_lock); |
1016 | destroy_inode(inode); | |
1017 | inode = old; | |
1018 | wait_on_inode(inode); | |
1019 | } | |
1020 | return inode; | |
1021 | } | |
1022 | ||
ad5e195a CH |
1023 | /* |
1024 | * search the inode cache for a matching inode number. | |
1025 | * If we find one, then the inode number we are trying to | |
1026 | * allocate is not unique and so we should not use it. | |
1027 | * | |
1028 | * Returns 1 if the inode number is unique, 0 if it is not. | |
1029 | */ | |
1030 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) | |
1031 | { | |
1032 | struct hlist_head *b = inode_hashtable + hash(sb, ino); | |
1033 | struct hlist_node *node; | |
1034 | struct inode *inode; | |
1035 | ||
1036 | hlist_for_each_entry(inode, node, b, i_hash) { | |
1037 | if (inode->i_ino == ino && inode->i_sb == sb) | |
1038 | return 0; | |
1039 | } | |
1040 | ||
1041 | return 1; | |
1042 | } | |
1043 | ||
1da177e4 LT |
1044 | /** |
1045 | * iunique - get a unique inode number | |
1046 | * @sb: superblock | |
1047 | * @max_reserved: highest reserved inode number | |
1048 | * | |
1049 | * Obtain an inode number that is unique on the system for a given | |
1050 | * superblock. This is used by file systems that have no natural | |
1051 | * permanent inode numbering system. An inode number is returned that | |
1052 | * is higher than the reserved limit but unique. | |
1053 | * | |
1054 | * BUGS: | |
1055 | * With a large number of inodes live on the file system this function | |
1056 | * currently becomes quite slow. | |
1057 | */ | |
1058 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
1059 | { | |
866b04fc JL |
1060 | /* |
1061 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
1062 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
1063 | * here to attempt to avoid that. | |
1064 | */ | |
ad5e195a | 1065 | static DEFINE_SPINLOCK(iunique_lock); |
866b04fc | 1066 | static unsigned int counter; |
1da177e4 | 1067 | ino_t res; |
3361c7be | 1068 | |
1da177e4 | 1069 | spin_lock(&inode_lock); |
ad5e195a | 1070 | spin_lock(&iunique_lock); |
3361c7be JL |
1071 | do { |
1072 | if (counter <= max_reserved) | |
1073 | counter = max_reserved + 1; | |
1da177e4 | 1074 | res = counter++; |
ad5e195a CH |
1075 | } while (!test_inode_iunique(sb, res)); |
1076 | spin_unlock(&iunique_lock); | |
3361c7be | 1077 | spin_unlock(&inode_lock); |
1da177e4 | 1078 | |
3361c7be JL |
1079 | return res; |
1080 | } | |
1da177e4 LT |
1081 | EXPORT_SYMBOL(iunique); |
1082 | ||
1083 | struct inode *igrab(struct inode *inode) | |
1084 | { | |
1085 | spin_lock(&inode_lock); | |
250df6ed DC |
1086 | spin_lock(&inode->i_lock); |
1087 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { | |
1da177e4 | 1088 | __iget(inode); |
250df6ed DC |
1089 | spin_unlock(&inode->i_lock); |
1090 | } else { | |
1091 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1092 | /* |
1093 | * Handle the case where s_op->clear_inode is not been | |
1094 | * called yet, and somebody is calling igrab | |
1095 | * while the inode is getting freed. | |
1096 | */ | |
1097 | inode = NULL; | |
250df6ed | 1098 | } |
1da177e4 LT |
1099 | spin_unlock(&inode_lock); |
1100 | return inode; | |
1101 | } | |
1da177e4 LT |
1102 | EXPORT_SYMBOL(igrab); |
1103 | ||
1104 | /** | |
1105 | * ifind - internal function, you want ilookup5() or iget5(). | |
1106 | * @sb: super block of file system to search | |
1107 | * @head: the head of the list to search | |
1108 | * @test: callback used for comparisons between inodes | |
1109 | * @data: opaque data pointer to pass to @test | |
88bd5121 | 1110 | * @wait: if true wait for the inode to be unlocked, if false do not |
1da177e4 LT |
1111 | * |
1112 | * ifind() searches for the inode specified by @data in the inode | |
1113 | * cache. This is a generalized version of ifind_fast() for file systems where | |
1114 | * the inode number is not sufficient for unique identification of an inode. | |
1115 | * | |
1116 | * If the inode is in the cache, the inode is returned with an incremented | |
1117 | * reference count. | |
1118 | * | |
1119 | * Otherwise NULL is returned. | |
1120 | * | |
1121 | * Note, @test is called with the inode_lock held, so can't sleep. | |
1122 | */ | |
5d2bea45 | 1123 | static struct inode *ifind(struct super_block *sb, |
1da177e4 | 1124 | struct hlist_head *head, int (*test)(struct inode *, void *), |
88bd5121 | 1125 | void *data, const int wait) |
1da177e4 LT |
1126 | { |
1127 | struct inode *inode; | |
1128 | ||
1129 | spin_lock(&inode_lock); | |
1130 | inode = find_inode(sb, head, test, data); | |
1131 | if (inode) { | |
1da177e4 | 1132 | spin_unlock(&inode_lock); |
88bd5121 AA |
1133 | if (likely(wait)) |
1134 | wait_on_inode(inode); | |
1da177e4 LT |
1135 | return inode; |
1136 | } | |
1137 | spin_unlock(&inode_lock); | |
1138 | return NULL; | |
1139 | } | |
1140 | ||
1141 | /** | |
1142 | * ifind_fast - internal function, you want ilookup() or iget(). | |
1143 | * @sb: super block of file system to search | |
1144 | * @head: head of the list to search | |
1145 | * @ino: inode number to search for | |
1146 | * | |
1147 | * ifind_fast() searches for the inode @ino in the inode cache. This is for | |
1148 | * file systems where the inode number is sufficient for unique identification | |
1149 | * of an inode. | |
1150 | * | |
1151 | * If the inode is in the cache, the inode is returned with an incremented | |
1152 | * reference count. | |
1153 | * | |
1154 | * Otherwise NULL is returned. | |
1155 | */ | |
5d2bea45 | 1156 | static struct inode *ifind_fast(struct super_block *sb, |
1da177e4 LT |
1157 | struct hlist_head *head, unsigned long ino) |
1158 | { | |
1159 | struct inode *inode; | |
1160 | ||
1161 | spin_lock(&inode_lock); | |
1162 | inode = find_inode_fast(sb, head, ino); | |
1163 | if (inode) { | |
1da177e4 LT |
1164 | spin_unlock(&inode_lock); |
1165 | wait_on_inode(inode); | |
1166 | return inode; | |
1167 | } | |
1168 | spin_unlock(&inode_lock); | |
1169 | return NULL; | |
1170 | } | |
1171 | ||
1172 | /** | |
88bd5121 | 1173 | * ilookup5_nowait - search for an inode in the inode cache |
1da177e4 LT |
1174 | * @sb: super block of file system to search |
1175 | * @hashval: hash value (usually inode number) to search for | |
1176 | * @test: callback used for comparisons between inodes | |
1177 | * @data: opaque data pointer to pass to @test | |
1178 | * | |
1179 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
1180 | * @data in the inode cache. This is a generalized version of ilookup() for | |
1181 | * file systems where the inode number is not sufficient for unique | |
1182 | * identification of an inode. | |
1183 | * | |
1184 | * If the inode is in the cache, the inode is returned with an incremented | |
88bd5121 AA |
1185 | * reference count. Note, the inode lock is not waited upon so you have to be |
1186 | * very careful what you do with the returned inode. You probably should be | |
1187 | * using ilookup5() instead. | |
1188 | * | |
1189 | * Otherwise NULL is returned. | |
1190 | * | |
1191 | * Note, @test is called with the inode_lock held, so can't sleep. | |
1192 | */ | |
1193 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, | |
1194 | int (*test)(struct inode *, void *), void *data) | |
1195 | { | |
1196 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1197 | ||
1198 | return ifind(sb, head, test, data, 0); | |
1199 | } | |
88bd5121 AA |
1200 | EXPORT_SYMBOL(ilookup5_nowait); |
1201 | ||
1202 | /** | |
1203 | * ilookup5 - search for an inode in the inode cache | |
1204 | * @sb: super block of file system to search | |
1205 | * @hashval: hash value (usually inode number) to search for | |
1206 | * @test: callback used for comparisons between inodes | |
1207 | * @data: opaque data pointer to pass to @test | |
1208 | * | |
1209 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
1210 | * @data in the inode cache. This is a generalized version of ilookup() for | |
1211 | * file systems where the inode number is not sufficient for unique | |
1212 | * identification of an inode. | |
1213 | * | |
1214 | * If the inode is in the cache, the inode lock is waited upon and the inode is | |
1215 | * returned with an incremented reference count. | |
1da177e4 LT |
1216 | * |
1217 | * Otherwise NULL is returned. | |
1218 | * | |
1219 | * Note, @test is called with the inode_lock held, so can't sleep. | |
1220 | */ | |
1221 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
1222 | int (*test)(struct inode *, void *), void *data) | |
1223 | { | |
1224 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1225 | ||
88bd5121 | 1226 | return ifind(sb, head, test, data, 1); |
1da177e4 | 1227 | } |
1da177e4 LT |
1228 | EXPORT_SYMBOL(ilookup5); |
1229 | ||
1230 | /** | |
1231 | * ilookup - search for an inode in the inode cache | |
1232 | * @sb: super block of file system to search | |
1233 | * @ino: inode number to search for | |
1234 | * | |
1235 | * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache. | |
1236 | * This is for file systems where the inode number is sufficient for unique | |
1237 | * identification of an inode. | |
1238 | * | |
1239 | * If the inode is in the cache, the inode is returned with an incremented | |
1240 | * reference count. | |
1241 | * | |
1242 | * Otherwise NULL is returned. | |
1243 | */ | |
1244 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1245 | { | |
1246 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1247 | ||
1248 | return ifind_fast(sb, head, ino); | |
1249 | } | |
1da177e4 LT |
1250 | EXPORT_SYMBOL(ilookup); |
1251 | ||
1252 | /** | |
1253 | * iget5_locked - obtain an inode from a mounted file system | |
1254 | * @sb: super block of file system | |
1255 | * @hashval: hash value (usually inode number) to get | |
1256 | * @test: callback used for comparisons between inodes | |
1257 | * @set: callback used to initialize a new struct inode | |
1258 | * @data: opaque data pointer to pass to @test and @set | |
1259 | * | |
1da177e4 LT |
1260 | * iget5_locked() uses ifind() to search for the inode specified by @hashval |
1261 | * and @data in the inode cache and if present it is returned with an increased | |
1262 | * reference count. This is a generalized version of iget_locked() for file | |
1263 | * systems where the inode number is not sufficient for unique identification | |
1264 | * of an inode. | |
1265 | * | |
1266 | * If the inode is not in cache, get_new_inode() is called to allocate a new | |
1267 | * inode and this is returned locked, hashed, and with the I_NEW flag set. The | |
1268 | * file system gets to fill it in before unlocking it via unlock_new_inode(). | |
1269 | * | |
1270 | * Note both @test and @set are called with the inode_lock held, so can't sleep. | |
1271 | */ | |
1272 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, | |
1273 | int (*test)(struct inode *, void *), | |
1274 | int (*set)(struct inode *, void *), void *data) | |
1275 | { | |
1276 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1277 | struct inode *inode; | |
1278 | ||
88bd5121 | 1279 | inode = ifind(sb, head, test, data, 1); |
1da177e4 LT |
1280 | if (inode) |
1281 | return inode; | |
1282 | /* | |
1283 | * get_new_inode() will do the right thing, re-trying the search | |
1284 | * in case it had to block at any point. | |
1285 | */ | |
1286 | return get_new_inode(sb, head, test, set, data); | |
1287 | } | |
1da177e4 LT |
1288 | EXPORT_SYMBOL(iget5_locked); |
1289 | ||
1290 | /** | |
1291 | * iget_locked - obtain an inode from a mounted file system | |
1292 | * @sb: super block of file system | |
1293 | * @ino: inode number to get | |
1294 | * | |
1da177e4 LT |
1295 | * iget_locked() uses ifind_fast() to search for the inode specified by @ino in |
1296 | * the inode cache and if present it is returned with an increased reference | |
1297 | * count. This is for file systems where the inode number is sufficient for | |
1298 | * unique identification of an inode. | |
1299 | * | |
1300 | * If the inode is not in cache, get_new_inode_fast() is called to allocate a | |
1301 | * new inode and this is returned locked, hashed, and with the I_NEW flag set. | |
1302 | * The file system gets to fill it in before unlocking it via | |
1303 | * unlock_new_inode(). | |
1304 | */ | |
1305 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) | |
1306 | { | |
1307 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1308 | struct inode *inode; | |
1309 | ||
1310 | inode = ifind_fast(sb, head, ino); | |
1311 | if (inode) | |
1312 | return inode; | |
1313 | /* | |
1314 | * get_new_inode_fast() will do the right thing, re-trying the search | |
1315 | * in case it had to block at any point. | |
1316 | */ | |
1317 | return get_new_inode_fast(sb, head, ino); | |
1318 | } | |
1da177e4 LT |
1319 | EXPORT_SYMBOL(iget_locked); |
1320 | ||
261bca86 AV |
1321 | int insert_inode_locked(struct inode *inode) |
1322 | { | |
1323 | struct super_block *sb = inode->i_sb; | |
1324 | ino_t ino = inode->i_ino; | |
1325 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
261bca86 | 1326 | |
261bca86 | 1327 | while (1) { |
72a43d63 AV |
1328 | struct hlist_node *node; |
1329 | struct inode *old = NULL; | |
261bca86 | 1330 | spin_lock(&inode_lock); |
72a43d63 AV |
1331 | hlist_for_each_entry(old, node, head, i_hash) { |
1332 | if (old->i_ino != ino) | |
1333 | continue; | |
1334 | if (old->i_sb != sb) | |
1335 | continue; | |
250df6ed DC |
1336 | spin_lock(&old->i_lock); |
1337 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1338 | spin_unlock(&old->i_lock); | |
72a43d63 | 1339 | continue; |
250df6ed | 1340 | } |
72a43d63 AV |
1341 | break; |
1342 | } | |
1343 | if (likely(!node)) { | |
250df6ed DC |
1344 | spin_lock(&inode->i_lock); |
1345 | inode->i_state |= I_NEW; | |
261bca86 | 1346 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1347 | spin_unlock(&inode->i_lock); |
261bca86 AV |
1348 | spin_unlock(&inode_lock); |
1349 | return 0; | |
1350 | } | |
1351 | __iget(old); | |
250df6ed | 1352 | spin_unlock(&old->i_lock); |
261bca86 AV |
1353 | spin_unlock(&inode_lock); |
1354 | wait_on_inode(old); | |
1d3382cb | 1355 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1356 | iput(old); |
1357 | return -EBUSY; | |
1358 | } | |
1359 | iput(old); | |
1360 | } | |
1361 | } | |
261bca86 AV |
1362 | EXPORT_SYMBOL(insert_inode_locked); |
1363 | ||
1364 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1365 | int (*test)(struct inode *, void *), void *data) | |
1366 | { | |
1367 | struct super_block *sb = inode->i_sb; | |
1368 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
261bca86 | 1369 | |
261bca86 | 1370 | while (1) { |
72a43d63 AV |
1371 | struct hlist_node *node; |
1372 | struct inode *old = NULL; | |
1373 | ||
261bca86 | 1374 | spin_lock(&inode_lock); |
72a43d63 AV |
1375 | hlist_for_each_entry(old, node, head, i_hash) { |
1376 | if (old->i_sb != sb) | |
1377 | continue; | |
1378 | if (!test(old, data)) | |
1379 | continue; | |
250df6ed DC |
1380 | spin_lock(&old->i_lock); |
1381 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1382 | spin_unlock(&old->i_lock); | |
72a43d63 | 1383 | continue; |
250df6ed | 1384 | } |
72a43d63 AV |
1385 | break; |
1386 | } | |
1387 | if (likely(!node)) { | |
250df6ed DC |
1388 | spin_lock(&inode->i_lock); |
1389 | inode->i_state |= I_NEW; | |
261bca86 | 1390 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1391 | spin_unlock(&inode->i_lock); |
261bca86 AV |
1392 | spin_unlock(&inode_lock); |
1393 | return 0; | |
1394 | } | |
1395 | __iget(old); | |
250df6ed | 1396 | spin_unlock(&old->i_lock); |
261bca86 AV |
1397 | spin_unlock(&inode_lock); |
1398 | wait_on_inode(old); | |
1d3382cb | 1399 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1400 | iput(old); |
1401 | return -EBUSY; | |
1402 | } | |
1403 | iput(old); | |
1404 | } | |
1405 | } | |
261bca86 AV |
1406 | EXPORT_SYMBOL(insert_inode_locked4); |
1407 | ||
1da177e4 | 1408 | |
45321ac5 AV |
1409 | int generic_delete_inode(struct inode *inode) |
1410 | { | |
1411 | return 1; | |
1412 | } | |
1413 | EXPORT_SYMBOL(generic_delete_inode); | |
1414 | ||
1da177e4 | 1415 | /* |
45321ac5 AV |
1416 | * Normal UNIX filesystem behaviour: delete the |
1417 | * inode when the usage count drops to zero, and | |
1418 | * i_nlink is zero. | |
1da177e4 | 1419 | */ |
45321ac5 | 1420 | int generic_drop_inode(struct inode *inode) |
1da177e4 | 1421 | { |
1d3382cb | 1422 | return !inode->i_nlink || inode_unhashed(inode); |
1da177e4 | 1423 | } |
45321ac5 | 1424 | EXPORT_SYMBOL_GPL(generic_drop_inode); |
1da177e4 | 1425 | |
45321ac5 AV |
1426 | /* |
1427 | * Called when we're dropping the last reference | |
1428 | * to an inode. | |
22fe4042 | 1429 | * |
45321ac5 AV |
1430 | * Call the FS "drop_inode()" function, defaulting to |
1431 | * the legacy UNIX filesystem behaviour. If it tells | |
1432 | * us to evict inode, do so. Otherwise, retain inode | |
1433 | * in cache if fs is alive, sync and evict if fs is | |
1434 | * shutting down. | |
22fe4042 | 1435 | */ |
45321ac5 | 1436 | static void iput_final(struct inode *inode) |
1da177e4 LT |
1437 | { |
1438 | struct super_block *sb = inode->i_sb; | |
45321ac5 AV |
1439 | const struct super_operations *op = inode->i_sb->s_op; |
1440 | int drop; | |
1441 | ||
250df6ed DC |
1442 | spin_lock(&inode->i_lock); |
1443 | WARN_ON(inode->i_state & I_NEW); | |
1444 | ||
45321ac5 AV |
1445 | if (op && op->drop_inode) |
1446 | drop = op->drop_inode(inode); | |
1447 | else | |
1448 | drop = generic_drop_inode(inode); | |
1da177e4 | 1449 | |
b2b2af8e DC |
1450 | if (!drop && (sb->s_flags & MS_ACTIVE)) { |
1451 | inode->i_state |= I_REFERENCED; | |
1452 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) | |
1453 | inode_lru_list_add(inode); | |
1454 | spin_unlock(&inode->i_lock); | |
1455 | spin_unlock(&inode_lock); | |
1456 | return; | |
1457 | } | |
1458 | ||
45321ac5 | 1459 | if (!drop) { |
991114c6 | 1460 | inode->i_state |= I_WILL_FREE; |
250df6ed | 1461 | spin_unlock(&inode->i_lock); |
991114c6 | 1462 | spin_unlock(&inode_lock); |
1da177e4 LT |
1463 | write_inode_now(inode, 1); |
1464 | spin_lock(&inode_lock); | |
250df6ed | 1465 | spin_lock(&inode->i_lock); |
7ef0d737 | 1466 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1467 | inode->i_state &= ~I_WILL_FREE; |
1da177e4 | 1468 | } |
7ccf19a8 | 1469 | |
991114c6 | 1470 | inode->i_state |= I_FREEING; |
9e38d86f | 1471 | inode_lru_list_del(inode); |
b2b2af8e | 1472 | spin_unlock(&inode->i_lock); |
1da177e4 | 1473 | spin_unlock(&inode_lock); |
b2b2af8e | 1474 | |
644da596 | 1475 | evict(inode); |
1da177e4 LT |
1476 | } |
1477 | ||
1da177e4 | 1478 | /** |
6b3304b5 | 1479 | * iput - put an inode |
1da177e4 LT |
1480 | * @inode: inode to put |
1481 | * | |
1482 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1483 | * zero, the inode is then freed and may also be destroyed. | |
1484 | * | |
1485 | * Consequently, iput() can sleep. | |
1486 | */ | |
1487 | void iput(struct inode *inode) | |
1488 | { | |
1489 | if (inode) { | |
a4ffdde6 | 1490 | BUG_ON(inode->i_state & I_CLEAR); |
1da177e4 | 1491 | |
1da177e4 LT |
1492 | if (atomic_dec_and_lock(&inode->i_count, &inode_lock)) |
1493 | iput_final(inode); | |
1494 | } | |
1495 | } | |
1da177e4 LT |
1496 | EXPORT_SYMBOL(iput); |
1497 | ||
1498 | /** | |
1499 | * bmap - find a block number in a file | |
1500 | * @inode: inode of file | |
1501 | * @block: block to find | |
1502 | * | |
1503 | * Returns the block number on the device holding the inode that | |
1504 | * is the disk block number for the block of the file requested. | |
1505 | * That is, asked for block 4 of inode 1 the function will return the | |
6b3304b5 | 1506 | * disk block relative to the disk start that holds that block of the |
1da177e4 LT |
1507 | * file. |
1508 | */ | |
6b3304b5 | 1509 | sector_t bmap(struct inode *inode, sector_t block) |
1da177e4 LT |
1510 | { |
1511 | sector_t res = 0; | |
1512 | if (inode->i_mapping->a_ops->bmap) | |
1513 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1514 | return res; | |
1515 | } | |
1da177e4 LT |
1516 | EXPORT_SYMBOL(bmap); |
1517 | ||
11ff6f05 MG |
1518 | /* |
1519 | * With relative atime, only update atime if the previous atime is | |
1520 | * earlier than either the ctime or mtime or if at least a day has | |
1521 | * passed since the last atime update. | |
1522 | */ | |
1523 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1524 | struct timespec now) | |
1525 | { | |
1526 | ||
1527 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1528 | return 1; | |
1529 | /* | |
1530 | * Is mtime younger than atime? If yes, update atime: | |
1531 | */ | |
1532 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1533 | return 1; | |
1534 | /* | |
1535 | * Is ctime younger than atime? If yes, update atime: | |
1536 | */ | |
1537 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1538 | return 1; | |
1539 | ||
1540 | /* | |
1541 | * Is the previous atime value older than a day? If yes, | |
1542 | * update atime: | |
1543 | */ | |
1544 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1545 | return 1; | |
1546 | /* | |
1547 | * Good, we can skip the atime update: | |
1548 | */ | |
1549 | return 0; | |
1550 | } | |
1551 | ||
1da177e4 | 1552 | /** |
869243a0 CH |
1553 | * touch_atime - update the access time |
1554 | * @mnt: mount the inode is accessed on | |
7045f37b | 1555 | * @dentry: dentry accessed |
1da177e4 LT |
1556 | * |
1557 | * Update the accessed time on an inode and mark it for writeback. | |
1558 | * This function automatically handles read only file systems and media, | |
1559 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1560 | */ | |
869243a0 | 1561 | void touch_atime(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 1562 | { |
869243a0 | 1563 | struct inode *inode = dentry->d_inode; |
1da177e4 LT |
1564 | struct timespec now; |
1565 | ||
cdb70f3f | 1566 | if (inode->i_flags & S_NOATIME) |
b12536c2 | 1567 | return; |
37756ced | 1568 | if (IS_NOATIME(inode)) |
b12536c2 | 1569 | return; |
b2276138 | 1570 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1571 | return; |
47ae32d6 | 1572 | |
cdb70f3f | 1573 | if (mnt->mnt_flags & MNT_NOATIME) |
b12536c2 | 1574 | return; |
cdb70f3f | 1575 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1576 | return; |
1da177e4 LT |
1577 | |
1578 | now = current_fs_time(inode->i_sb); | |
11ff6f05 MG |
1579 | |
1580 | if (!relatime_need_update(mnt, inode, now)) | |
b12536c2 | 1581 | return; |
11ff6f05 | 1582 | |
47ae32d6 | 1583 | if (timespec_equal(&inode->i_atime, &now)) |
b12536c2 AK |
1584 | return; |
1585 | ||
1586 | if (mnt_want_write(mnt)) | |
1587 | return; | |
47ae32d6 VH |
1588 | |
1589 | inode->i_atime = now; | |
1590 | mark_inode_dirty_sync(inode); | |
cdb70f3f | 1591 | mnt_drop_write(mnt); |
1da177e4 | 1592 | } |
869243a0 | 1593 | EXPORT_SYMBOL(touch_atime); |
1da177e4 LT |
1594 | |
1595 | /** | |
870f4817 CH |
1596 | * file_update_time - update mtime and ctime time |
1597 | * @file: file accessed | |
1da177e4 | 1598 | * |
870f4817 CH |
1599 | * Update the mtime and ctime members of an inode and mark the inode |
1600 | * for writeback. Note that this function is meant exclusively for | |
1601 | * usage in the file write path of filesystems, and filesystems may | |
1602 | * choose to explicitly ignore update via this function with the | |
2eadfc0e | 1603 | * S_NOCMTIME inode flag, e.g. for network filesystem where these |
870f4817 | 1604 | * timestamps are handled by the server. |
1da177e4 LT |
1605 | */ |
1606 | ||
870f4817 | 1607 | void file_update_time(struct file *file) |
1da177e4 | 1608 | { |
0f7fc9e4 | 1609 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 | 1610 | struct timespec now; |
ce06e0b2 | 1611 | enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0; |
1da177e4 | 1612 | |
ce06e0b2 | 1613 | /* First try to exhaust all avenues to not sync */ |
1da177e4 LT |
1614 | if (IS_NOCMTIME(inode)) |
1615 | return; | |
20ddee2c | 1616 | |
1da177e4 | 1617 | now = current_fs_time(inode->i_sb); |
ce06e0b2 AK |
1618 | if (!timespec_equal(&inode->i_mtime, &now)) |
1619 | sync_it = S_MTIME; | |
1da177e4 | 1620 | |
ce06e0b2 AK |
1621 | if (!timespec_equal(&inode->i_ctime, &now)) |
1622 | sync_it |= S_CTIME; | |
870f4817 | 1623 | |
ce06e0b2 AK |
1624 | if (IS_I_VERSION(inode)) |
1625 | sync_it |= S_VERSION; | |
7a224228 | 1626 | |
ce06e0b2 AK |
1627 | if (!sync_it) |
1628 | return; | |
1629 | ||
1630 | /* Finally allowed to write? Takes lock. */ | |
1631 | if (mnt_want_write_file(file)) | |
1632 | return; | |
1633 | ||
1634 | /* Only change inode inside the lock region */ | |
1635 | if (sync_it & S_VERSION) | |
1636 | inode_inc_iversion(inode); | |
1637 | if (sync_it & S_CTIME) | |
1638 | inode->i_ctime = now; | |
1639 | if (sync_it & S_MTIME) | |
1640 | inode->i_mtime = now; | |
1641 | mark_inode_dirty_sync(inode); | |
20ddee2c | 1642 | mnt_drop_write(file->f_path.mnt); |
1da177e4 | 1643 | } |
870f4817 | 1644 | EXPORT_SYMBOL(file_update_time); |
1da177e4 LT |
1645 | |
1646 | int inode_needs_sync(struct inode *inode) | |
1647 | { | |
1648 | if (IS_SYNC(inode)) | |
1649 | return 1; | |
1650 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1651 | return 1; | |
1652 | return 0; | |
1653 | } | |
1da177e4 LT |
1654 | EXPORT_SYMBOL(inode_needs_sync); |
1655 | ||
1da177e4 LT |
1656 | int inode_wait(void *word) |
1657 | { | |
1658 | schedule(); | |
1659 | return 0; | |
1660 | } | |
d44dab8d | 1661 | EXPORT_SYMBOL(inode_wait); |
1da177e4 LT |
1662 | |
1663 | /* | |
168a9fd6 MS |
1664 | * If we try to find an inode in the inode hash while it is being |
1665 | * deleted, we have to wait until the filesystem completes its | |
1666 | * deletion before reporting that it isn't found. This function waits | |
1667 | * until the deletion _might_ have completed. Callers are responsible | |
1668 | * to recheck inode state. | |
1669 | * | |
eaff8079 | 1670 | * It doesn't matter if I_NEW is not set initially, a call to |
250df6ed DC |
1671 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list |
1672 | * will DTRT. | |
1da177e4 LT |
1673 | */ |
1674 | static void __wait_on_freeing_inode(struct inode *inode) | |
1675 | { | |
1676 | wait_queue_head_t *wq; | |
eaff8079 CH |
1677 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); |
1678 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1da177e4 | 1679 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
250df6ed | 1680 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1681 | spin_unlock(&inode_lock); |
1682 | schedule(); | |
1683 | finish_wait(wq, &wait.wait); | |
1684 | spin_lock(&inode_lock); | |
1685 | } | |
1686 | ||
1da177e4 LT |
1687 | static __initdata unsigned long ihash_entries; |
1688 | static int __init set_ihash_entries(char *str) | |
1689 | { | |
1690 | if (!str) | |
1691 | return 0; | |
1692 | ihash_entries = simple_strtoul(str, &str, 0); | |
1693 | return 1; | |
1694 | } | |
1695 | __setup("ihash_entries=", set_ihash_entries); | |
1696 | ||
1697 | /* | |
1698 | * Initialize the waitqueues and inode hash table. | |
1699 | */ | |
1700 | void __init inode_init_early(void) | |
1701 | { | |
1702 | int loop; | |
1703 | ||
1704 | /* If hashes are distributed across NUMA nodes, defer | |
1705 | * hash allocation until vmalloc space is available. | |
1706 | */ | |
1707 | if (hashdist) | |
1708 | return; | |
1709 | ||
1710 | inode_hashtable = | |
1711 | alloc_large_system_hash("Inode-cache", | |
1712 | sizeof(struct hlist_head), | |
1713 | ihash_entries, | |
1714 | 14, | |
1715 | HASH_EARLY, | |
1716 | &i_hash_shift, | |
1717 | &i_hash_mask, | |
1718 | 0); | |
1719 | ||
1720 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1721 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1722 | } | |
1723 | ||
74bf17cf | 1724 | void __init inode_init(void) |
1da177e4 LT |
1725 | { |
1726 | int loop; | |
1727 | ||
1728 | /* inode slab cache */ | |
b0196009 PJ |
1729 | inode_cachep = kmem_cache_create("inode_cache", |
1730 | sizeof(struct inode), | |
1731 | 0, | |
1732 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1733 | SLAB_MEM_SPREAD), | |
20c2df83 | 1734 | init_once); |
8e1f936b | 1735 | register_shrinker(&icache_shrinker); |
1da177e4 LT |
1736 | |
1737 | /* Hash may have been set up in inode_init_early */ | |
1738 | if (!hashdist) | |
1739 | return; | |
1740 | ||
1741 | inode_hashtable = | |
1742 | alloc_large_system_hash("Inode-cache", | |
1743 | sizeof(struct hlist_head), | |
1744 | ihash_entries, | |
1745 | 14, | |
1746 | 0, | |
1747 | &i_hash_shift, | |
1748 | &i_hash_mask, | |
1749 | 0); | |
1750 | ||
1751 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1752 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1753 | } | |
1754 | ||
1755 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1756 | { | |
1757 | inode->i_mode = mode; | |
1758 | if (S_ISCHR(mode)) { | |
1759 | inode->i_fop = &def_chr_fops; | |
1760 | inode->i_rdev = rdev; | |
1761 | } else if (S_ISBLK(mode)) { | |
1762 | inode->i_fop = &def_blk_fops; | |
1763 | inode->i_rdev = rdev; | |
1764 | } else if (S_ISFIFO(mode)) | |
1765 | inode->i_fop = &def_fifo_fops; | |
1766 | else if (S_ISSOCK(mode)) | |
1767 | inode->i_fop = &bad_sock_fops; | |
1768 | else | |
af0d9ae8 MK |
1769 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" |
1770 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1771 | inode->i_ino); | |
1da177e4 LT |
1772 | } |
1773 | EXPORT_SYMBOL(init_special_inode); | |
a1bd120d DM |
1774 | |
1775 | /** | |
eaae668d | 1776 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards |
a1bd120d DM |
1777 | * @inode: New inode |
1778 | * @dir: Directory inode | |
1779 | * @mode: mode of the new inode | |
1780 | */ | |
1781 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
1782 | mode_t mode) | |
1783 | { | |
1784 | inode->i_uid = current_fsuid(); | |
1785 | if (dir && dir->i_mode & S_ISGID) { | |
1786 | inode->i_gid = dir->i_gid; | |
1787 | if (S_ISDIR(mode)) | |
1788 | mode |= S_ISGID; | |
1789 | } else | |
1790 | inode->i_gid = current_fsgid(); | |
1791 | inode->i_mode = mode; | |
1792 | } | |
1793 | EXPORT_SYMBOL(inode_init_owner); | |
e795b717 | 1794 | |
2e149670 SH |
1795 | /** |
1796 | * inode_owner_or_capable - check current task permissions to inode | |
1797 | * @inode: inode being checked | |
1798 | * | |
1799 | * Return true if current either has CAP_FOWNER to the inode, or | |
1800 | * owns the file. | |
e795b717 | 1801 | */ |
2e149670 | 1802 | bool inode_owner_or_capable(const struct inode *inode) |
e795b717 SH |
1803 | { |
1804 | struct user_namespace *ns = inode_userns(inode); | |
1805 | ||
1806 | if (current_user_ns() == ns && current_fsuid() == inode->i_uid) | |
1807 | return true; | |
1808 | if (ns_capable(ns, CAP_FOWNER)) | |
1809 | return true; | |
1810 | return false; | |
1811 | } | |
2e149670 | 1812 | EXPORT_SYMBOL(inode_owner_or_capable); |