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