]>
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> |
a66979ab | 29 | #include "internal.h" |
1da177e4 | 30 | |
250df6ed DC |
31 | /* |
32 | * inode locking rules. | |
33 | * | |
34 | * inode->i_lock protects: | |
35 | * inode->i_state, inode->i_hash, __iget() | |
02afc410 DC |
36 | * inode_lru_lock protects: |
37 | * inode_lru, inode->i_lru | |
55fa6091 DC |
38 | * inode_sb_list_lock protects: |
39 | * sb->s_inodes, inode->i_sb_list | |
a66979ab DC |
40 | * inode_wb_list_lock protects: |
41 | * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list | |
67a23c49 DC |
42 | * inode_hash_lock protects: |
43 | * inode_hashtable, inode->i_hash | |
250df6ed DC |
44 | * |
45 | * Lock ordering: | |
55fa6091 DC |
46 | * |
47 | * inode_sb_list_lock | |
48 | * inode->i_lock | |
02afc410 | 49 | * inode_lru_lock |
a66979ab DC |
50 | * |
51 | * inode_wb_list_lock | |
52 | * inode->i_lock | |
67a23c49 DC |
53 | * |
54 | * inode_hash_lock | |
55 | * inode_sb_list_lock | |
56 | * inode->i_lock | |
57 | * | |
58 | * iunique_lock | |
59 | * inode_hash_lock | |
250df6ed DC |
60 | */ |
61 | ||
1da177e4 LT |
62 | /* |
63 | * This is needed for the following functions: | |
64 | * - inode_has_buffers | |
1da177e4 LT |
65 | * - invalidate_bdev |
66 | * | |
67 | * FIXME: remove all knowledge of the buffer layer from this file | |
68 | */ | |
69 | #include <linux/buffer_head.h> | |
70 | ||
71 | /* | |
72 | * New inode.c implementation. | |
73 | * | |
74 | * This implementation has the basic premise of trying | |
75 | * to be extremely low-overhead and SMP-safe, yet be | |
76 | * simple enough to be "obviously correct". | |
77 | * | |
78 | * Famous last words. | |
79 | */ | |
80 | ||
81 | /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */ | |
82 | ||
83 | /* #define INODE_PARANOIA 1 */ | |
84 | /* #define INODE_DEBUG 1 */ | |
85 | ||
86 | /* | |
87 | * Inode lookup is no longer as critical as it used to be: | |
88 | * most of the lookups are going to be through the dcache. | |
89 | */ | |
90 | #define I_HASHBITS i_hash_shift | |
91 | #define I_HASHMASK i_hash_mask | |
92 | ||
fa3536cc ED |
93 | static unsigned int i_hash_mask __read_mostly; |
94 | static unsigned int i_hash_shift __read_mostly; | |
67a23c49 DC |
95 | static struct hlist_head *inode_hashtable __read_mostly; |
96 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock); | |
1da177e4 LT |
97 | |
98 | /* | |
99 | * Each inode can be on two separate lists. One is | |
100 | * the hash list of the inode, used for lookups. The | |
101 | * other linked list is the "type" list: | |
102 | * "in_use" - valid inode, i_count > 0, i_nlink > 0 | |
103 | * "dirty" - as "in_use" but also dirty | |
104 | * "unused" - valid inode, i_count = 0 | |
105 | * | |
106 | * A "dirty" list is maintained for each super block, | |
107 | * allowing for low-overhead inode sync() operations. | |
108 | */ | |
109 | ||
7ccf19a8 | 110 | static LIST_HEAD(inode_lru); |
02afc410 | 111 | static DEFINE_SPINLOCK(inode_lru_lock); |
1da177e4 | 112 | |
55fa6091 | 113 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock); |
a66979ab | 114 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_wb_list_lock); |
55fa6091 | 115 | |
1da177e4 | 116 | /* |
bab1d944 CH |
117 | * iprune_sem provides exclusion between the icache shrinking and the |
118 | * umount path. | |
88e0fbc4 | 119 | * |
bab1d944 CH |
120 | * We don't actually need it to protect anything in the umount path, |
121 | * but only need to cycle through it to make sure any inode that | |
122 | * prune_icache took off the LRU list has been fully torn down by the | |
123 | * time we are past evict_inodes. | |
1da177e4 | 124 | */ |
88e0fbc4 | 125 | static DECLARE_RWSEM(iprune_sem); |
1da177e4 LT |
126 | |
127 | /* | |
128 | * Statistics gathering.. | |
129 | */ | |
130 | struct inodes_stat_t inodes_stat; | |
131 | ||
3e880fb5 | 132 | static DEFINE_PER_CPU(unsigned int, nr_inodes); |
cffbc8aa | 133 | |
6b3304b5 | 134 | static struct kmem_cache *inode_cachep __read_mostly; |
1da177e4 | 135 | |
3e880fb5 | 136 | static int get_nr_inodes(void) |
cffbc8aa | 137 | { |
3e880fb5 NP |
138 | int i; |
139 | int sum = 0; | |
140 | for_each_possible_cpu(i) | |
141 | sum += per_cpu(nr_inodes, i); | |
142 | return sum < 0 ? 0 : sum; | |
cffbc8aa DC |
143 | } |
144 | ||
145 | static inline int get_nr_inodes_unused(void) | |
146 | { | |
86c8749e | 147 | return inodes_stat.nr_unused; |
cffbc8aa DC |
148 | } |
149 | ||
150 | int get_nr_dirty_inodes(void) | |
151 | { | |
3e880fb5 | 152 | /* not actually dirty inodes, but a wild approximation */ |
cffbc8aa DC |
153 | int nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); |
154 | return nr_dirty > 0 ? nr_dirty : 0; | |
cffbc8aa DC |
155 | } |
156 | ||
157 | /* | |
158 | * Handle nr_inode sysctl | |
159 | */ | |
160 | #ifdef CONFIG_SYSCTL | |
161 | int proc_nr_inodes(ctl_table *table, int write, | |
162 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
163 | { | |
164 | inodes_stat.nr_inodes = get_nr_inodes(); | |
cffbc8aa DC |
165 | return proc_dointvec(table, write, buffer, lenp, ppos); |
166 | } | |
167 | #endif | |
168 | ||
2cb1599f DC |
169 | /** |
170 | * inode_init_always - perform inode structure intialisation | |
0bc02f3f RD |
171 | * @sb: superblock inode belongs to |
172 | * @inode: inode to initialise | |
2cb1599f DC |
173 | * |
174 | * These are initializations that need to be done on every inode | |
175 | * allocation as the fields are not initialised by slab allocation. | |
176 | */ | |
54e34621 | 177 | int inode_init_always(struct super_block *sb, struct inode *inode) |
1da177e4 | 178 | { |
f5e54d6e | 179 | static const struct address_space_operations empty_aops; |
6e1d5dcc | 180 | static const struct inode_operations empty_iops; |
99ac48f5 | 181 | static const struct file_operations empty_fops; |
6b3304b5 | 182 | struct address_space *const mapping = &inode->i_data; |
2cb1599f DC |
183 | |
184 | inode->i_sb = sb; | |
185 | inode->i_blkbits = sb->s_blocksize_bits; | |
186 | inode->i_flags = 0; | |
187 | atomic_set(&inode->i_count, 1); | |
188 | inode->i_op = &empty_iops; | |
189 | inode->i_fop = &empty_fops; | |
190 | inode->i_nlink = 1; | |
56ff5efa AV |
191 | inode->i_uid = 0; |
192 | inode->i_gid = 0; | |
2cb1599f DC |
193 | atomic_set(&inode->i_writecount, 0); |
194 | inode->i_size = 0; | |
195 | inode->i_blocks = 0; | |
196 | inode->i_bytes = 0; | |
197 | inode->i_generation = 0; | |
1da177e4 | 198 | #ifdef CONFIG_QUOTA |
2cb1599f | 199 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); |
1da177e4 | 200 | #endif |
2cb1599f DC |
201 | inode->i_pipe = NULL; |
202 | inode->i_bdev = NULL; | |
203 | inode->i_cdev = NULL; | |
204 | inode->i_rdev = 0; | |
205 | inode->dirtied_when = 0; | |
6146f0d5 MZ |
206 | |
207 | if (security_inode_alloc(inode)) | |
54e34621 | 208 | goto out; |
2cb1599f DC |
209 | spin_lock_init(&inode->i_lock); |
210 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
211 | ||
212 | mutex_init(&inode->i_mutex); | |
213 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); | |
214 | ||
215 | init_rwsem(&inode->i_alloc_sem); | |
216 | lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key); | |
217 | ||
218 | mapping->a_ops = &empty_aops; | |
219 | mapping->host = inode; | |
220 | mapping->flags = 0; | |
3c1d4378 | 221 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); |
2cb1599f DC |
222 | mapping->assoc_mapping = NULL; |
223 | mapping->backing_dev_info = &default_backing_dev_info; | |
224 | mapping->writeback_index = 0; | |
225 | ||
226 | /* | |
227 | * If the block_device provides a backing_dev_info for client | |
228 | * inodes then use that. Otherwise the inode share the bdev's | |
229 | * backing_dev_info. | |
230 | */ | |
231 | if (sb->s_bdev) { | |
232 | struct backing_dev_info *bdi; | |
233 | ||
2c96ce9f | 234 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; |
2cb1599f DC |
235 | mapping->backing_dev_info = bdi; |
236 | } | |
237 | inode->i_private = NULL; | |
238 | inode->i_mapping = mapping; | |
f19d4a8f AV |
239 | #ifdef CONFIG_FS_POSIX_ACL |
240 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; | |
241 | #endif | |
2cb1599f | 242 | |
3be25f49 EP |
243 | #ifdef CONFIG_FSNOTIFY |
244 | inode->i_fsnotify_mask = 0; | |
245 | #endif | |
246 | ||
3e880fb5 | 247 | this_cpu_inc(nr_inodes); |
cffbc8aa | 248 | |
54e34621 | 249 | return 0; |
54e34621 CH |
250 | out: |
251 | return -ENOMEM; | |
1da177e4 | 252 | } |
2cb1599f DC |
253 | EXPORT_SYMBOL(inode_init_always); |
254 | ||
255 | static struct inode *alloc_inode(struct super_block *sb) | |
256 | { | |
257 | struct inode *inode; | |
258 | ||
259 | if (sb->s_op->alloc_inode) | |
260 | inode = sb->s_op->alloc_inode(sb); | |
261 | else | |
262 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
263 | ||
54e34621 CH |
264 | if (!inode) |
265 | return NULL; | |
266 | ||
267 | if (unlikely(inode_init_always(sb, inode))) { | |
268 | if (inode->i_sb->s_op->destroy_inode) | |
269 | inode->i_sb->s_op->destroy_inode(inode); | |
270 | else | |
271 | kmem_cache_free(inode_cachep, inode); | |
272 | return NULL; | |
273 | } | |
274 | ||
275 | return inode; | |
2cb1599f | 276 | } |
1da177e4 | 277 | |
ff0c7d15 NP |
278 | void free_inode_nonrcu(struct inode *inode) |
279 | { | |
280 | kmem_cache_free(inode_cachep, inode); | |
281 | } | |
282 | EXPORT_SYMBOL(free_inode_nonrcu); | |
283 | ||
2e00c97e | 284 | void __destroy_inode(struct inode *inode) |
1da177e4 | 285 | { |
b7542f8c | 286 | BUG_ON(inode_has_buffers(inode)); |
1da177e4 | 287 | security_inode_free(inode); |
3be25f49 | 288 | fsnotify_inode_delete(inode); |
f19d4a8f AV |
289 | #ifdef CONFIG_FS_POSIX_ACL |
290 | if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) | |
291 | posix_acl_release(inode->i_acl); | |
292 | if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) | |
293 | posix_acl_release(inode->i_default_acl); | |
294 | #endif | |
3e880fb5 | 295 | this_cpu_dec(nr_inodes); |
2e00c97e CH |
296 | } |
297 | EXPORT_SYMBOL(__destroy_inode); | |
298 | ||
fa0d7e3d NP |
299 | static void i_callback(struct rcu_head *head) |
300 | { | |
301 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
302 | INIT_LIST_HEAD(&inode->i_dentry); | |
303 | kmem_cache_free(inode_cachep, inode); | |
304 | } | |
305 | ||
56b0dacf | 306 | static void destroy_inode(struct inode *inode) |
2e00c97e | 307 | { |
7ccf19a8 | 308 | BUG_ON(!list_empty(&inode->i_lru)); |
2e00c97e | 309 | __destroy_inode(inode); |
1da177e4 LT |
310 | if (inode->i_sb->s_op->destroy_inode) |
311 | inode->i_sb->s_op->destroy_inode(inode); | |
312 | else | |
fa0d7e3d | 313 | call_rcu(&inode->i_rcu, i_callback); |
1da177e4 | 314 | } |
1da177e4 | 315 | |
2aa15890 MS |
316 | void address_space_init_once(struct address_space *mapping) |
317 | { | |
318 | memset(mapping, 0, sizeof(*mapping)); | |
319 | INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC); | |
320 | spin_lock_init(&mapping->tree_lock); | |
321 | spin_lock_init(&mapping->i_mmap_lock); | |
322 | INIT_LIST_HEAD(&mapping->private_list); | |
323 | spin_lock_init(&mapping->private_lock); | |
324 | INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap); | |
325 | INIT_LIST_HEAD(&mapping->i_mmap_nonlinear); | |
326 | mutex_init(&mapping->unmap_mutex); | |
327 | } | |
328 | EXPORT_SYMBOL(address_space_init_once); | |
329 | ||
1da177e4 LT |
330 | /* |
331 | * These are initializations that only need to be done | |
332 | * once, because the fields are idempotent across use | |
333 | * of the inode, so let the slab aware of that. | |
334 | */ | |
335 | void inode_init_once(struct inode *inode) | |
336 | { | |
337 | memset(inode, 0, sizeof(*inode)); | |
338 | INIT_HLIST_NODE(&inode->i_hash); | |
339 | INIT_LIST_HEAD(&inode->i_dentry); | |
340 | INIT_LIST_HEAD(&inode->i_devices); | |
7ccf19a8 NP |
341 | INIT_LIST_HEAD(&inode->i_wb_list); |
342 | INIT_LIST_HEAD(&inode->i_lru); | |
2aa15890 | 343 | address_space_init_once(&inode->i_data); |
1da177e4 | 344 | i_size_ordered_init(inode); |
3be25f49 | 345 | #ifdef CONFIG_FSNOTIFY |
e61ce867 | 346 | INIT_HLIST_HEAD(&inode->i_fsnotify_marks); |
3be25f49 | 347 | #endif |
1da177e4 | 348 | } |
1da177e4 LT |
349 | EXPORT_SYMBOL(inode_init_once); |
350 | ||
51cc5068 | 351 | static void init_once(void *foo) |
1da177e4 | 352 | { |
6b3304b5 | 353 | struct inode *inode = (struct inode *) foo; |
1da177e4 | 354 | |
a35afb83 | 355 | inode_init_once(inode); |
1da177e4 LT |
356 | } |
357 | ||
358 | /* | |
250df6ed | 359 | * inode->i_lock must be held |
1da177e4 | 360 | */ |
6b3304b5 | 361 | void __iget(struct inode *inode) |
1da177e4 | 362 | { |
9e38d86f NP |
363 | atomic_inc(&inode->i_count); |
364 | } | |
2e147f1e | 365 | |
7de9c6ee AV |
366 | /* |
367 | * get additional reference to inode; caller must already hold one. | |
368 | */ | |
369 | void ihold(struct inode *inode) | |
370 | { | |
371 | WARN_ON(atomic_inc_return(&inode->i_count) < 2); | |
372 | } | |
373 | EXPORT_SYMBOL(ihold); | |
374 | ||
9e38d86f NP |
375 | static void inode_lru_list_add(struct inode *inode) |
376 | { | |
02afc410 | 377 | spin_lock(&inode_lru_lock); |
7ccf19a8 NP |
378 | if (list_empty(&inode->i_lru)) { |
379 | list_add(&inode->i_lru, &inode_lru); | |
86c8749e | 380 | inodes_stat.nr_unused++; |
9e38d86f | 381 | } |
02afc410 | 382 | spin_unlock(&inode_lru_lock); |
9e38d86f | 383 | } |
2e147f1e | 384 | |
9e38d86f NP |
385 | static void inode_lru_list_del(struct inode *inode) |
386 | { | |
02afc410 | 387 | spin_lock(&inode_lru_lock); |
7ccf19a8 NP |
388 | if (!list_empty(&inode->i_lru)) { |
389 | list_del_init(&inode->i_lru); | |
86c8749e | 390 | inodes_stat.nr_unused--; |
9e38d86f | 391 | } |
02afc410 | 392 | spin_unlock(&inode_lru_lock); |
1da177e4 LT |
393 | } |
394 | ||
646ec461 CH |
395 | /** |
396 | * inode_sb_list_add - add inode to the superblock list of inodes | |
397 | * @inode: inode to add | |
398 | */ | |
399 | void inode_sb_list_add(struct inode *inode) | |
400 | { | |
55fa6091 DC |
401 | spin_lock(&inode_sb_list_lock); |
402 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); | |
403 | spin_unlock(&inode_sb_list_lock); | |
646ec461 CH |
404 | } |
405 | EXPORT_SYMBOL_GPL(inode_sb_list_add); | |
406 | ||
55fa6091 | 407 | static inline void inode_sb_list_del(struct inode *inode) |
646ec461 | 408 | { |
55fa6091 | 409 | spin_lock(&inode_sb_list_lock); |
646ec461 | 410 | list_del_init(&inode->i_sb_list); |
55fa6091 | 411 | spin_unlock(&inode_sb_list_lock); |
646ec461 CH |
412 | } |
413 | ||
4c51acbc DC |
414 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
415 | { | |
416 | unsigned long tmp; | |
417 | ||
418 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
419 | L1_CACHE_BYTES; | |
420 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS); | |
421 | return tmp & I_HASHMASK; | |
422 | } | |
423 | ||
424 | /** | |
425 | * __insert_inode_hash - hash an inode | |
426 | * @inode: unhashed inode | |
427 | * @hashval: unsigned long value used to locate this object in the | |
428 | * inode_hashtable. | |
429 | * | |
430 | * Add an inode to the inode hash for this superblock. | |
431 | */ | |
432 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
433 | { | |
646ec461 CH |
434 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); |
435 | ||
67a23c49 | 436 | spin_lock(&inode_hash_lock); |
250df6ed | 437 | spin_lock(&inode->i_lock); |
646ec461 | 438 | hlist_add_head(&inode->i_hash, b); |
250df6ed | 439 | spin_unlock(&inode->i_lock); |
67a23c49 | 440 | spin_unlock(&inode_hash_lock); |
4c51acbc DC |
441 | } |
442 | EXPORT_SYMBOL(__insert_inode_hash); | |
443 | ||
4c51acbc DC |
444 | /** |
445 | * remove_inode_hash - remove an inode from the hash | |
446 | * @inode: inode to unhash | |
447 | * | |
448 | * Remove an inode from the superblock. | |
449 | */ | |
450 | void remove_inode_hash(struct inode *inode) | |
451 | { | |
67a23c49 | 452 | spin_lock(&inode_hash_lock); |
250df6ed | 453 | spin_lock(&inode->i_lock); |
4c51acbc | 454 | hlist_del_init(&inode->i_hash); |
250df6ed | 455 | spin_unlock(&inode->i_lock); |
67a23c49 | 456 | spin_unlock(&inode_hash_lock); |
4c51acbc DC |
457 | } |
458 | EXPORT_SYMBOL(remove_inode_hash); | |
459 | ||
b0683aa6 AV |
460 | void end_writeback(struct inode *inode) |
461 | { | |
462 | might_sleep(); | |
463 | BUG_ON(inode->i_data.nrpages); | |
464 | BUG_ON(!list_empty(&inode->i_data.private_list)); | |
465 | BUG_ON(!(inode->i_state & I_FREEING)); | |
466 | BUG_ON(inode->i_state & I_CLEAR); | |
467 | inode_sync_wait(inode); | |
fa0d7e3d | 468 | /* don't need i_lock here, no concurrent mods to i_state */ |
b0683aa6 AV |
469 | inode->i_state = I_FREEING | I_CLEAR; |
470 | } | |
471 | EXPORT_SYMBOL(end_writeback); | |
472 | ||
b2b2af8e DC |
473 | /* |
474 | * Free the inode passed in, removing it from the lists it is still connected | |
475 | * to. We remove any pages still attached to the inode and wait for any IO that | |
476 | * is still in progress before finally destroying the inode. | |
477 | * | |
478 | * An inode must already be marked I_FREEING so that we avoid the inode being | |
479 | * moved back onto lists if we race with other code that manipulates the lists | |
480 | * (e.g. writeback_single_inode). The caller is responsible for setting this. | |
481 | * | |
482 | * An inode must already be removed from the LRU list before being evicted from | |
483 | * the cache. This should occur atomically with setting the I_FREEING state | |
484 | * flag, so no inodes here should ever be on the LRU when being evicted. | |
485 | */ | |
644da596 | 486 | static void evict(struct inode *inode) |
b4272d4c AV |
487 | { |
488 | const struct super_operations *op = inode->i_sb->s_op; | |
489 | ||
b2b2af8e DC |
490 | BUG_ON(!(inode->i_state & I_FREEING)); |
491 | BUG_ON(!list_empty(&inode->i_lru)); | |
492 | ||
a66979ab | 493 | inode_wb_list_del(inode); |
55fa6091 DC |
494 | inode_sb_list_del(inode); |
495 | ||
be7ce416 AV |
496 | if (op->evict_inode) { |
497 | op->evict_inode(inode); | |
b4272d4c AV |
498 | } else { |
499 | if (inode->i_data.nrpages) | |
500 | truncate_inode_pages(&inode->i_data, 0); | |
30140837 | 501 | end_writeback(inode); |
b4272d4c | 502 | } |
661074e9 AV |
503 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
504 | bd_forget(inode); | |
505 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) | |
506 | cd_forget(inode); | |
b2b2af8e DC |
507 | |
508 | remove_inode_hash(inode); | |
509 | ||
510 | spin_lock(&inode->i_lock); | |
511 | wake_up_bit(&inode->i_state, __I_NEW); | |
512 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); | |
513 | spin_unlock(&inode->i_lock); | |
514 | ||
515 | destroy_inode(inode); | |
b4272d4c AV |
516 | } |
517 | ||
1da177e4 LT |
518 | /* |
519 | * dispose_list - dispose of the contents of a local list | |
520 | * @head: the head of the list to free | |
521 | * | |
522 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
523 | * need to worry about list corruption and SMP locks. | |
524 | */ | |
525 | static void dispose_list(struct list_head *head) | |
526 | { | |
1da177e4 LT |
527 | while (!list_empty(head)) { |
528 | struct inode *inode; | |
529 | ||
7ccf19a8 NP |
530 | inode = list_first_entry(head, struct inode, i_lru); |
531 | list_del_init(&inode->i_lru); | |
1da177e4 | 532 | |
644da596 | 533 | evict(inode); |
1da177e4 | 534 | } |
1da177e4 LT |
535 | } |
536 | ||
63997e98 AV |
537 | /** |
538 | * evict_inodes - evict all evictable inodes for a superblock | |
539 | * @sb: superblock to operate on | |
540 | * | |
541 | * Make sure that no inodes with zero refcount are retained. This is | |
542 | * called by superblock shutdown after having MS_ACTIVE flag removed, | |
543 | * so any inode reaching zero refcount during or after that call will | |
544 | * be immediately evicted. | |
1da177e4 | 545 | */ |
63997e98 | 546 | void evict_inodes(struct super_block *sb) |
1da177e4 | 547 | { |
63997e98 AV |
548 | struct inode *inode, *next; |
549 | LIST_HEAD(dispose); | |
1da177e4 | 550 | |
55fa6091 | 551 | spin_lock(&inode_sb_list_lock); |
63997e98 AV |
552 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
553 | if (atomic_read(&inode->i_count)) | |
aabb8fdb | 554 | continue; |
250df6ed DC |
555 | |
556 | spin_lock(&inode->i_lock); | |
557 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
558 | spin_unlock(&inode->i_lock); | |
1da177e4 | 559 | continue; |
250df6ed | 560 | } |
63997e98 AV |
561 | |
562 | inode->i_state |= I_FREEING; | |
02afc410 | 563 | inode_lru_list_del(inode); |
250df6ed | 564 | spin_unlock(&inode->i_lock); |
02afc410 | 565 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 566 | } |
55fa6091 | 567 | spin_unlock(&inode_sb_list_lock); |
63997e98 AV |
568 | |
569 | dispose_list(&dispose); | |
bab1d944 CH |
570 | |
571 | /* | |
572 | * Cycle through iprune_sem to make sure any inode that prune_icache | |
573 | * moved off the list before we took the lock has been fully torn | |
574 | * down. | |
575 | */ | |
576 | down_write(&iprune_sem); | |
63997e98 | 577 | up_write(&iprune_sem); |
1da177e4 LT |
578 | } |
579 | ||
1da177e4 | 580 | /** |
a0318786 CH |
581 | * invalidate_inodes - attempt to free all inodes on a superblock |
582 | * @sb: superblock to operate on | |
93b270f7 | 583 | * @kill_dirty: flag to guide handling of dirty inodes |
1da177e4 | 584 | * |
a0318786 CH |
585 | * Attempts to free all inodes for a given superblock. If there were any |
586 | * busy inodes return a non-zero value, else zero. | |
93b270f7 N |
587 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat |
588 | * them as busy. | |
1da177e4 | 589 | */ |
93b270f7 | 590 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) |
1da177e4 | 591 | { |
cffbc8aa | 592 | int busy = 0; |
a0318786 CH |
593 | struct inode *inode, *next; |
594 | LIST_HEAD(dispose); | |
1da177e4 | 595 | |
55fa6091 | 596 | spin_lock(&inode_sb_list_lock); |
a0318786 | 597 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
250df6ed DC |
598 | spin_lock(&inode->i_lock); |
599 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
600 | spin_unlock(&inode->i_lock); | |
aabb8fdb | 601 | continue; |
250df6ed | 602 | } |
93b270f7 | 603 | if (inode->i_state & I_DIRTY && !kill_dirty) { |
250df6ed | 604 | spin_unlock(&inode->i_lock); |
93b270f7 N |
605 | busy = 1; |
606 | continue; | |
607 | } | |
99a38919 | 608 | if (atomic_read(&inode->i_count)) { |
250df6ed | 609 | spin_unlock(&inode->i_lock); |
99a38919 | 610 | busy = 1; |
1da177e4 LT |
611 | continue; |
612 | } | |
99a38919 | 613 | |
99a38919 | 614 | inode->i_state |= I_FREEING; |
02afc410 | 615 | inode_lru_list_del(inode); |
250df6ed | 616 | spin_unlock(&inode->i_lock); |
02afc410 | 617 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 618 | } |
55fa6091 | 619 | spin_unlock(&inode_sb_list_lock); |
1da177e4 | 620 | |
a0318786 | 621 | dispose_list(&dispose); |
1da177e4 LT |
622 | |
623 | return busy; | |
624 | } | |
1da177e4 LT |
625 | |
626 | static int can_unuse(struct inode *inode) | |
627 | { | |
9e38d86f | 628 | if (inode->i_state & ~I_REFERENCED) |
1da177e4 LT |
629 | return 0; |
630 | if (inode_has_buffers(inode)) | |
631 | return 0; | |
632 | if (atomic_read(&inode->i_count)) | |
633 | return 0; | |
634 | if (inode->i_data.nrpages) | |
635 | return 0; | |
636 | return 1; | |
637 | } | |
638 | ||
639 | /* | |
9e38d86f | 640 | * Scan `goal' inodes on the unused list for freeable ones. They are moved to a |
02afc410 | 641 | * temporary list and then are freed outside inode_lru_lock by dispose_list(). |
1da177e4 LT |
642 | * |
643 | * Any inodes which are pinned purely because of attached pagecache have their | |
9e38d86f NP |
644 | * pagecache removed. If the inode has metadata buffers attached to |
645 | * mapping->private_list then try to remove them. | |
1da177e4 | 646 | * |
9e38d86f NP |
647 | * If the inode has the I_REFERENCED flag set, then it means that it has been |
648 | * used recently - the flag is set in iput_final(). When we encounter such an | |
649 | * inode, clear the flag and move it to the back of the LRU so it gets another | |
650 | * pass through the LRU before it gets reclaimed. This is necessary because of | |
651 | * the fact we are doing lazy LRU updates to minimise lock contention so the | |
652 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes | |
653 | * with this flag set because they are the inodes that are out of order. | |
1da177e4 LT |
654 | */ |
655 | static void prune_icache(int nr_to_scan) | |
656 | { | |
657 | LIST_HEAD(freeable); | |
1da177e4 LT |
658 | int nr_scanned; |
659 | unsigned long reap = 0; | |
660 | ||
88e0fbc4 | 661 | down_read(&iprune_sem); |
02afc410 | 662 | spin_lock(&inode_lru_lock); |
1da177e4 LT |
663 | for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) { |
664 | struct inode *inode; | |
665 | ||
7ccf19a8 | 666 | if (list_empty(&inode_lru)) |
1da177e4 LT |
667 | break; |
668 | ||
7ccf19a8 | 669 | inode = list_entry(inode_lru.prev, struct inode, i_lru); |
1da177e4 | 670 | |
02afc410 DC |
671 | /* |
672 | * we are inverting the inode_lru_lock/inode->i_lock here, | |
673 | * so use a trylock. If we fail to get the lock, just move the | |
674 | * inode to the back of the list so we don't spin on it. | |
675 | */ | |
676 | if (!spin_trylock(&inode->i_lock)) { | |
677 | list_move(&inode->i_lru, &inode_lru); | |
678 | continue; | |
679 | } | |
680 | ||
9e38d86f NP |
681 | /* |
682 | * Referenced or dirty inodes are still in use. Give them | |
683 | * another pass through the LRU as we canot reclaim them now. | |
684 | */ | |
685 | if (atomic_read(&inode->i_count) || | |
686 | (inode->i_state & ~I_REFERENCED)) { | |
7ccf19a8 | 687 | list_del_init(&inode->i_lru); |
f283c86a | 688 | spin_unlock(&inode->i_lock); |
86c8749e | 689 | inodes_stat.nr_unused--; |
9e38d86f NP |
690 | continue; |
691 | } | |
692 | ||
693 | /* recently referenced inodes get one more pass */ | |
694 | if (inode->i_state & I_REFERENCED) { | |
9e38d86f | 695 | inode->i_state &= ~I_REFERENCED; |
250df6ed | 696 | list_move(&inode->i_lru, &inode_lru); |
f283c86a | 697 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
698 | continue; |
699 | } | |
700 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { | |
701 | __iget(inode); | |
250df6ed | 702 | spin_unlock(&inode->i_lock); |
02afc410 | 703 | spin_unlock(&inode_lru_lock); |
1da177e4 | 704 | if (remove_inode_buffers(inode)) |
fc0ecff6 AM |
705 | reap += invalidate_mapping_pages(&inode->i_data, |
706 | 0, -1); | |
1da177e4 | 707 | iput(inode); |
02afc410 | 708 | spin_lock(&inode_lru_lock); |
1da177e4 | 709 | |
7ccf19a8 NP |
710 | if (inode != list_entry(inode_lru.next, |
711 | struct inode, i_lru)) | |
1da177e4 | 712 | continue; /* wrong inode or list_empty */ |
02afc410 DC |
713 | /* avoid lock inversions with trylock */ |
714 | if (!spin_trylock(&inode->i_lock)) | |
715 | continue; | |
250df6ed DC |
716 | if (!can_unuse(inode)) { |
717 | spin_unlock(&inode->i_lock); | |
1da177e4 | 718 | continue; |
250df6ed | 719 | } |
1da177e4 | 720 | } |
7ef0d737 | 721 | WARN_ON(inode->i_state & I_NEW); |
1da177e4 | 722 | inode->i_state |= I_FREEING; |
250df6ed | 723 | spin_unlock(&inode->i_lock); |
7ccf19a8 | 724 | |
7ccf19a8 | 725 | list_move(&inode->i_lru, &freeable); |
86c8749e | 726 | inodes_stat.nr_unused--; |
1da177e4 | 727 | } |
f8891e5e CL |
728 | if (current_is_kswapd()) |
729 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
730 | else | |
731 | __count_vm_events(PGINODESTEAL, reap); | |
02afc410 | 732 | spin_unlock(&inode_lru_lock); |
1da177e4 LT |
733 | |
734 | dispose_list(&freeable); | |
88e0fbc4 | 735 | up_read(&iprune_sem); |
1da177e4 LT |
736 | } |
737 | ||
738 | /* | |
739 | * shrink_icache_memory() will attempt to reclaim some unused inodes. Here, | |
740 | * "unused" means that no dentries are referring to the inodes: the files are | |
741 | * not open and the dcache references to those inodes have already been | |
742 | * reclaimed. | |
743 | * | |
744 | * This function is passed the number of inodes to scan, and it returns the | |
745 | * total number of remaining possibly-reclaimable inodes. | |
746 | */ | |
7f8275d0 | 747 | static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
1da177e4 LT |
748 | { |
749 | if (nr) { | |
750 | /* | |
751 | * Nasty deadlock avoidance. We may hold various FS locks, | |
752 | * and we don't want to recurse into the FS that called us | |
753 | * in clear_inode() and friends.. | |
6b3304b5 | 754 | */ |
1da177e4 LT |
755 | if (!(gfp_mask & __GFP_FS)) |
756 | return -1; | |
757 | prune_icache(nr); | |
758 | } | |
cffbc8aa | 759 | return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure; |
1da177e4 LT |
760 | } |
761 | ||
8e1f936b RR |
762 | static struct shrinker icache_shrinker = { |
763 | .shrink = shrink_icache_memory, | |
764 | .seeks = DEFAULT_SEEKS, | |
765 | }; | |
766 | ||
1da177e4 LT |
767 | static void __wait_on_freeing_inode(struct inode *inode); |
768 | /* | |
769 | * Called with the inode lock held. | |
1da177e4 | 770 | */ |
6b3304b5 MK |
771 | static struct inode *find_inode(struct super_block *sb, |
772 | struct hlist_head *head, | |
773 | int (*test)(struct inode *, void *), | |
774 | void *data) | |
1da177e4 LT |
775 | { |
776 | struct hlist_node *node; | |
6b3304b5 | 777 | struct inode *inode = NULL; |
1da177e4 LT |
778 | |
779 | repeat: | |
c5c8be3c | 780 | hlist_for_each_entry(inode, node, head, i_hash) { |
67a23c49 DC |
781 | spin_lock(&inode->i_lock); |
782 | if (inode->i_sb != sb) { | |
783 | spin_unlock(&inode->i_lock); | |
1da177e4 | 784 | continue; |
67a23c49 DC |
785 | } |
786 | if (!test(inode, data)) { | |
787 | spin_unlock(&inode->i_lock); | |
1da177e4 | 788 | continue; |
67a23c49 | 789 | } |
a4ffdde6 | 790 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
791 | __wait_on_freeing_inode(inode); |
792 | goto repeat; | |
793 | } | |
f7899bd5 | 794 | __iget(inode); |
250df6ed | 795 | spin_unlock(&inode->i_lock); |
f7899bd5 | 796 | return inode; |
1da177e4 | 797 | } |
f7899bd5 | 798 | return NULL; |
1da177e4 LT |
799 | } |
800 | ||
801 | /* | |
802 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
803 | * iget_locked for details. | |
804 | */ | |
6b3304b5 MK |
805 | static struct inode *find_inode_fast(struct super_block *sb, |
806 | struct hlist_head *head, unsigned long ino) | |
1da177e4 LT |
807 | { |
808 | struct hlist_node *node; | |
6b3304b5 | 809 | struct inode *inode = NULL; |
1da177e4 LT |
810 | |
811 | repeat: | |
c5c8be3c | 812 | hlist_for_each_entry(inode, node, head, i_hash) { |
67a23c49 DC |
813 | spin_lock(&inode->i_lock); |
814 | if (inode->i_ino != ino) { | |
815 | spin_unlock(&inode->i_lock); | |
1da177e4 | 816 | continue; |
67a23c49 DC |
817 | } |
818 | if (inode->i_sb != sb) { | |
819 | spin_unlock(&inode->i_lock); | |
1da177e4 | 820 | continue; |
67a23c49 | 821 | } |
a4ffdde6 | 822 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
823 | __wait_on_freeing_inode(inode); |
824 | goto repeat; | |
825 | } | |
f7899bd5 | 826 | __iget(inode); |
250df6ed | 827 | spin_unlock(&inode->i_lock); |
f7899bd5 | 828 | return inode; |
1da177e4 | 829 | } |
f7899bd5 | 830 | return NULL; |
8290c35f DC |
831 | } |
832 | ||
f991bd2e ED |
833 | /* |
834 | * Each cpu owns a range of LAST_INO_BATCH numbers. | |
835 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, | |
836 | * to renew the exhausted range. | |
8290c35f | 837 | * |
f991bd2e ED |
838 | * This does not significantly increase overflow rate because every CPU can |
839 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is | |
840 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the | |
841 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase | |
842 | * overflow rate by 2x, which does not seem too significant. | |
843 | * | |
844 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
845 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
846 | * here to attempt to avoid that. | |
8290c35f | 847 | */ |
f991bd2e ED |
848 | #define LAST_INO_BATCH 1024 |
849 | static DEFINE_PER_CPU(unsigned int, last_ino); | |
850 | ||
85fe4025 | 851 | unsigned int get_next_ino(void) |
8290c35f | 852 | { |
f991bd2e ED |
853 | unsigned int *p = &get_cpu_var(last_ino); |
854 | unsigned int res = *p; | |
8290c35f | 855 | |
f991bd2e ED |
856 | #ifdef CONFIG_SMP |
857 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { | |
858 | static atomic_t shared_last_ino; | |
859 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); | |
860 | ||
861 | res = next - LAST_INO_BATCH; | |
862 | } | |
863 | #endif | |
864 | ||
865 | *p = ++res; | |
866 | put_cpu_var(last_ino); | |
867 | return res; | |
8290c35f | 868 | } |
85fe4025 | 869 | EXPORT_SYMBOL(get_next_ino); |
8290c35f | 870 | |
1da177e4 LT |
871 | /** |
872 | * new_inode - obtain an inode | |
873 | * @sb: superblock | |
874 | * | |
769848c0 | 875 | * Allocates a new inode for given superblock. The default gfp_mask |
3c1d4378 | 876 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
769848c0 MG |
877 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
878 | * for the page cache are not reclaimable or migratable, | |
879 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
880 | * newly created inode's mapping | |
881 | * | |
1da177e4 LT |
882 | */ |
883 | struct inode *new_inode(struct super_block *sb) | |
884 | { | |
6b3304b5 | 885 | struct inode *inode; |
1da177e4 | 886 | |
55fa6091 | 887 | spin_lock_prefetch(&inode_sb_list_lock); |
6b3304b5 | 888 | |
1da177e4 LT |
889 | inode = alloc_inode(sb); |
890 | if (inode) { | |
250df6ed | 891 | spin_lock(&inode->i_lock); |
1da177e4 | 892 | inode->i_state = 0; |
250df6ed | 893 | spin_unlock(&inode->i_lock); |
55fa6091 | 894 | inode_sb_list_add(inode); |
1da177e4 LT |
895 | } |
896 | return inode; | |
897 | } | |
1da177e4 LT |
898 | EXPORT_SYMBOL(new_inode); |
899 | ||
250df6ed DC |
900 | /** |
901 | * unlock_new_inode - clear the I_NEW state and wake up any waiters | |
902 | * @inode: new inode to unlock | |
903 | * | |
904 | * Called when the inode is fully initialised to clear the new state of the | |
905 | * inode and wake up anyone waiting for the inode to finish initialisation. | |
906 | */ | |
1da177e4 LT |
907 | void unlock_new_inode(struct inode *inode) |
908 | { | |
14358e6d | 909 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
a3314a0e | 910 | if (S_ISDIR(inode->i_mode)) { |
1e89a5e1 PZ |
911 | struct file_system_type *type = inode->i_sb->s_type; |
912 | ||
9a7aa12f JK |
913 | /* Set new key only if filesystem hasn't already changed it */ |
914 | if (!lockdep_match_class(&inode->i_mutex, | |
915 | &type->i_mutex_key)) { | |
916 | /* | |
917 | * ensure nobody is actually holding i_mutex | |
918 | */ | |
919 | mutex_destroy(&inode->i_mutex); | |
920 | mutex_init(&inode->i_mutex); | |
921 | lockdep_set_class(&inode->i_mutex, | |
922 | &type->i_mutex_dir_key); | |
923 | } | |
1e89a5e1 | 924 | } |
14358e6d | 925 | #endif |
250df6ed | 926 | spin_lock(&inode->i_lock); |
eaff8079 CH |
927 | WARN_ON(!(inode->i_state & I_NEW)); |
928 | inode->i_state &= ~I_NEW; | |
250df6ed DC |
929 | wake_up_bit(&inode->i_state, __I_NEW); |
930 | spin_unlock(&inode->i_lock); | |
1da177e4 | 931 | } |
1da177e4 LT |
932 | EXPORT_SYMBOL(unlock_new_inode); |
933 | ||
934 | /* | |
67a23c49 | 935 | * This is called without the inode hash lock held.. Be careful. |
1da177e4 LT |
936 | * |
937 | * We no longer cache the sb_flags in i_flags - see fs.h | |
938 | * -- rmk@arm.uk.linux.org | |
939 | */ | |
6b3304b5 MK |
940 | static struct inode *get_new_inode(struct super_block *sb, |
941 | struct hlist_head *head, | |
942 | int (*test)(struct inode *, void *), | |
943 | int (*set)(struct inode *, void *), | |
944 | void *data) | |
1da177e4 | 945 | { |
6b3304b5 | 946 | struct inode *inode; |
1da177e4 LT |
947 | |
948 | inode = alloc_inode(sb); | |
949 | if (inode) { | |
6b3304b5 | 950 | struct inode *old; |
1da177e4 | 951 | |
67a23c49 | 952 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
953 | /* We released the lock, so.. */ |
954 | old = find_inode(sb, head, test, data); | |
955 | if (!old) { | |
956 | if (set(inode, data)) | |
957 | goto set_failed; | |
958 | ||
250df6ed DC |
959 | spin_lock(&inode->i_lock); |
960 | inode->i_state = I_NEW; | |
646ec461 | 961 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 962 | spin_unlock(&inode->i_lock); |
55fa6091 | 963 | inode_sb_list_add(inode); |
67a23c49 | 964 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
965 | |
966 | /* Return the locked inode with I_NEW set, the | |
967 | * caller is responsible for filling in the contents | |
968 | */ | |
969 | return inode; | |
970 | } | |
971 | ||
972 | /* | |
973 | * Uhhuh, somebody else created the same inode under | |
974 | * us. Use the old inode instead of the one we just | |
975 | * allocated. | |
976 | */ | |
67a23c49 | 977 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
978 | destroy_inode(inode); |
979 | inode = old; | |
980 | wait_on_inode(inode); | |
981 | } | |
982 | return inode; | |
983 | ||
984 | set_failed: | |
67a23c49 | 985 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
986 | destroy_inode(inode); |
987 | return NULL; | |
988 | } | |
989 | ||
990 | /* | |
991 | * get_new_inode_fast is the fast path version of get_new_inode, see the | |
992 | * comment at iget_locked for details. | |
993 | */ | |
6b3304b5 MK |
994 | static struct inode *get_new_inode_fast(struct super_block *sb, |
995 | struct hlist_head *head, unsigned long ino) | |
1da177e4 | 996 | { |
6b3304b5 | 997 | struct inode *inode; |
1da177e4 LT |
998 | |
999 | inode = alloc_inode(sb); | |
1000 | if (inode) { | |
6b3304b5 | 1001 | struct inode *old; |
1da177e4 | 1002 | |
67a23c49 | 1003 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1004 | /* We released the lock, so.. */ |
1005 | old = find_inode_fast(sb, head, ino); | |
1006 | if (!old) { | |
1007 | inode->i_ino = ino; | |
250df6ed DC |
1008 | spin_lock(&inode->i_lock); |
1009 | inode->i_state = I_NEW; | |
646ec461 | 1010 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1011 | spin_unlock(&inode->i_lock); |
55fa6091 | 1012 | inode_sb_list_add(inode); |
67a23c49 | 1013 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1014 | |
1015 | /* Return the locked inode with I_NEW set, the | |
1016 | * caller is responsible for filling in the contents | |
1017 | */ | |
1018 | return inode; | |
1019 | } | |
1020 | ||
1021 | /* | |
1022 | * Uhhuh, somebody else created the same inode under | |
1023 | * us. Use the old inode instead of the one we just | |
1024 | * allocated. | |
1025 | */ | |
67a23c49 | 1026 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1027 | destroy_inode(inode); |
1028 | inode = old; | |
1029 | wait_on_inode(inode); | |
1030 | } | |
1031 | return inode; | |
1032 | } | |
1033 | ||
ad5e195a CH |
1034 | /* |
1035 | * search the inode cache for a matching inode number. | |
1036 | * If we find one, then the inode number we are trying to | |
1037 | * allocate is not unique and so we should not use it. | |
1038 | * | |
1039 | * Returns 1 if the inode number is unique, 0 if it is not. | |
1040 | */ | |
1041 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) | |
1042 | { | |
1043 | struct hlist_head *b = inode_hashtable + hash(sb, ino); | |
1044 | struct hlist_node *node; | |
1045 | struct inode *inode; | |
1046 | ||
67a23c49 | 1047 | spin_lock(&inode_hash_lock); |
ad5e195a | 1048 | hlist_for_each_entry(inode, node, b, i_hash) { |
67a23c49 DC |
1049 | if (inode->i_ino == ino && inode->i_sb == sb) { |
1050 | spin_unlock(&inode_hash_lock); | |
ad5e195a | 1051 | return 0; |
67a23c49 | 1052 | } |
ad5e195a | 1053 | } |
67a23c49 | 1054 | spin_unlock(&inode_hash_lock); |
ad5e195a CH |
1055 | |
1056 | return 1; | |
1057 | } | |
1058 | ||
1da177e4 LT |
1059 | /** |
1060 | * iunique - get a unique inode number | |
1061 | * @sb: superblock | |
1062 | * @max_reserved: highest reserved inode number | |
1063 | * | |
1064 | * Obtain an inode number that is unique on the system for a given | |
1065 | * superblock. This is used by file systems that have no natural | |
1066 | * permanent inode numbering system. An inode number is returned that | |
1067 | * is higher than the reserved limit but unique. | |
1068 | * | |
1069 | * BUGS: | |
1070 | * With a large number of inodes live on the file system this function | |
1071 | * currently becomes quite slow. | |
1072 | */ | |
1073 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
1074 | { | |
866b04fc JL |
1075 | /* |
1076 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
1077 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
1078 | * here to attempt to avoid that. | |
1079 | */ | |
ad5e195a | 1080 | static DEFINE_SPINLOCK(iunique_lock); |
866b04fc | 1081 | static unsigned int counter; |
1da177e4 | 1082 | ino_t res; |
3361c7be | 1083 | |
ad5e195a | 1084 | spin_lock(&iunique_lock); |
3361c7be JL |
1085 | do { |
1086 | if (counter <= max_reserved) | |
1087 | counter = max_reserved + 1; | |
1da177e4 | 1088 | res = counter++; |
ad5e195a CH |
1089 | } while (!test_inode_iunique(sb, res)); |
1090 | spin_unlock(&iunique_lock); | |
1da177e4 | 1091 | |
3361c7be JL |
1092 | return res; |
1093 | } | |
1da177e4 LT |
1094 | EXPORT_SYMBOL(iunique); |
1095 | ||
1096 | struct inode *igrab(struct inode *inode) | |
1097 | { | |
250df6ed DC |
1098 | spin_lock(&inode->i_lock); |
1099 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { | |
1da177e4 | 1100 | __iget(inode); |
250df6ed DC |
1101 | spin_unlock(&inode->i_lock); |
1102 | } else { | |
1103 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1104 | /* |
1105 | * Handle the case where s_op->clear_inode is not been | |
1106 | * called yet, and somebody is calling igrab | |
1107 | * while the inode is getting freed. | |
1108 | */ | |
1109 | inode = NULL; | |
250df6ed | 1110 | } |
1da177e4 LT |
1111 | return inode; |
1112 | } | |
1da177e4 LT |
1113 | EXPORT_SYMBOL(igrab); |
1114 | ||
1115 | /** | |
1116 | * ifind - internal function, you want ilookup5() or iget5(). | |
1117 | * @sb: super block of file system to search | |
1118 | * @head: the head of the list to search | |
1119 | * @test: callback used for comparisons between inodes | |
1120 | * @data: opaque data pointer to pass to @test | |
88bd5121 | 1121 | * @wait: if true wait for the inode to be unlocked, if false do not |
1da177e4 LT |
1122 | * |
1123 | * ifind() searches for the inode specified by @data in the inode | |
1124 | * cache. This is a generalized version of ifind_fast() for file systems where | |
1125 | * the inode number is not sufficient for unique identification of an inode. | |
1126 | * | |
1127 | * If the inode is in the cache, the inode is returned with an incremented | |
1128 | * reference count. | |
1129 | * | |
1130 | * Otherwise NULL is returned. | |
1131 | * | |
67a23c49 | 1132 | * Note, @test is called with the inode_hash_lock held, so can't sleep. |
1da177e4 | 1133 | */ |
5d2bea45 | 1134 | static struct inode *ifind(struct super_block *sb, |
1da177e4 | 1135 | struct hlist_head *head, int (*test)(struct inode *, void *), |
88bd5121 | 1136 | void *data, const int wait) |
1da177e4 LT |
1137 | { |
1138 | struct inode *inode; | |
1139 | ||
67a23c49 | 1140 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1141 | inode = find_inode(sb, head, test, data); |
1142 | if (inode) { | |
67a23c49 | 1143 | spin_unlock(&inode_hash_lock); |
88bd5121 AA |
1144 | if (likely(wait)) |
1145 | wait_on_inode(inode); | |
1da177e4 LT |
1146 | return inode; |
1147 | } | |
67a23c49 | 1148 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1149 | return NULL; |
1150 | } | |
1151 | ||
1152 | /** | |
1153 | * ifind_fast - internal function, you want ilookup() or iget(). | |
1154 | * @sb: super block of file system to search | |
1155 | * @head: head of the list to search | |
1156 | * @ino: inode number to search for | |
1157 | * | |
1158 | * ifind_fast() searches for the inode @ino in the inode cache. This is for | |
1159 | * file systems where the inode number is sufficient for unique identification | |
1160 | * of an inode. | |
1161 | * | |
1162 | * If the inode is in the cache, the inode is returned with an incremented | |
1163 | * reference count. | |
1164 | * | |
1165 | * Otherwise NULL is returned. | |
1166 | */ | |
5d2bea45 | 1167 | static struct inode *ifind_fast(struct super_block *sb, |
1da177e4 LT |
1168 | struct hlist_head *head, unsigned long ino) |
1169 | { | |
1170 | struct inode *inode; | |
1171 | ||
67a23c49 | 1172 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1173 | inode = find_inode_fast(sb, head, ino); |
1174 | if (inode) { | |
67a23c49 | 1175 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1176 | wait_on_inode(inode); |
1177 | return inode; | |
1178 | } | |
67a23c49 | 1179 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1180 | return NULL; |
1181 | } | |
1182 | ||
1183 | /** | |
88bd5121 | 1184 | * ilookup5_nowait - search for an inode in the inode cache |
1da177e4 LT |
1185 | * @sb: super block of file system to search |
1186 | * @hashval: hash value (usually inode number) to search for | |
1187 | * @test: callback used for comparisons between inodes | |
1188 | * @data: opaque data pointer to pass to @test | |
1189 | * | |
1190 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
1191 | * @data in the inode cache. This is a generalized version of ilookup() for | |
1192 | * file systems where the inode number is not sufficient for unique | |
1193 | * identification of an inode. | |
1194 | * | |
1195 | * If the inode is in the cache, the inode is returned with an incremented | |
88bd5121 AA |
1196 | * reference count. Note, the inode lock is not waited upon so you have to be |
1197 | * very careful what you do with the returned inode. You probably should be | |
1198 | * using ilookup5() instead. | |
1199 | * | |
1200 | * Otherwise NULL is returned. | |
1201 | * | |
67a23c49 | 1202 | * Note, @test is called with the inode_hash_lock held, so can't sleep. |
88bd5121 AA |
1203 | */ |
1204 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, | |
1205 | int (*test)(struct inode *, void *), void *data) | |
1206 | { | |
1207 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1208 | ||
1209 | return ifind(sb, head, test, data, 0); | |
1210 | } | |
88bd5121 AA |
1211 | EXPORT_SYMBOL(ilookup5_nowait); |
1212 | ||
1213 | /** | |
1214 | * ilookup5 - search for an inode in the inode cache | |
1215 | * @sb: super block of file system to search | |
1216 | * @hashval: hash value (usually inode number) to search for | |
1217 | * @test: callback used for comparisons between inodes | |
1218 | * @data: opaque data pointer to pass to @test | |
1219 | * | |
1220 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
1221 | * @data in the inode cache. This is a generalized version of ilookup() for | |
1222 | * file systems where the inode number is not sufficient for unique | |
1223 | * identification of an inode. | |
1224 | * | |
1225 | * If the inode is in the cache, the inode lock is waited upon and the inode is | |
1226 | * returned with an incremented reference count. | |
1da177e4 LT |
1227 | * |
1228 | * Otherwise NULL is returned. | |
1229 | * | |
67a23c49 | 1230 | * Note, @test is called with the inode_hash_lock held, so can't sleep. |
1da177e4 LT |
1231 | */ |
1232 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
1233 | int (*test)(struct inode *, void *), void *data) | |
1234 | { | |
1235 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1236 | ||
88bd5121 | 1237 | return ifind(sb, head, test, data, 1); |
1da177e4 | 1238 | } |
1da177e4 LT |
1239 | EXPORT_SYMBOL(ilookup5); |
1240 | ||
1241 | /** | |
1242 | * ilookup - search for an inode in the inode cache | |
1243 | * @sb: super block of file system to search | |
1244 | * @ino: inode number to search for | |
1245 | * | |
1246 | * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache. | |
1247 | * This is for file systems where the inode number is sufficient for unique | |
1248 | * identification of an inode. | |
1249 | * | |
1250 | * If the inode is in the cache, the inode is returned with an incremented | |
1251 | * reference count. | |
1252 | * | |
1253 | * Otherwise NULL is returned. | |
1254 | */ | |
1255 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1256 | { | |
1257 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1258 | ||
1259 | return ifind_fast(sb, head, ino); | |
1260 | } | |
1da177e4 LT |
1261 | EXPORT_SYMBOL(ilookup); |
1262 | ||
1263 | /** | |
1264 | * iget5_locked - obtain an inode from a mounted file system | |
1265 | * @sb: super block of file system | |
1266 | * @hashval: hash value (usually inode number) to get | |
1267 | * @test: callback used for comparisons between inodes | |
1268 | * @set: callback used to initialize a new struct inode | |
1269 | * @data: opaque data pointer to pass to @test and @set | |
1270 | * | |
1da177e4 LT |
1271 | * iget5_locked() uses ifind() to search for the inode specified by @hashval |
1272 | * and @data in the inode cache and if present it is returned with an increased | |
1273 | * reference count. This is a generalized version of iget_locked() for file | |
1274 | * systems where the inode number is not sufficient for unique identification | |
1275 | * of an inode. | |
1276 | * | |
1277 | * If the inode is not in cache, get_new_inode() is called to allocate a new | |
1278 | * inode and this is returned locked, hashed, and with the I_NEW flag set. The | |
1279 | * file system gets to fill it in before unlocking it via unlock_new_inode(). | |
1280 | * | |
67a23c49 DC |
1281 | * Note both @test and @set are called with the inode_hash_lock held, so can't |
1282 | * sleep. | |
1da177e4 LT |
1283 | */ |
1284 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, | |
1285 | int (*test)(struct inode *, void *), | |
1286 | int (*set)(struct inode *, void *), void *data) | |
1287 | { | |
1288 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1289 | struct inode *inode; | |
1290 | ||
88bd5121 | 1291 | inode = ifind(sb, head, test, data, 1); |
1da177e4 LT |
1292 | if (inode) |
1293 | return inode; | |
1294 | /* | |
1295 | * get_new_inode() will do the right thing, re-trying the search | |
1296 | * in case it had to block at any point. | |
1297 | */ | |
1298 | return get_new_inode(sb, head, test, set, data); | |
1299 | } | |
1da177e4 LT |
1300 | EXPORT_SYMBOL(iget5_locked); |
1301 | ||
1302 | /** | |
1303 | * iget_locked - obtain an inode from a mounted file system | |
1304 | * @sb: super block of file system | |
1305 | * @ino: inode number to get | |
1306 | * | |
1da177e4 LT |
1307 | * iget_locked() uses ifind_fast() to search for the inode specified by @ino in |
1308 | * the inode cache and if present it is returned with an increased reference | |
1309 | * count. This is for file systems where the inode number is sufficient for | |
1310 | * unique identification of an inode. | |
1311 | * | |
1312 | * If the inode is not in cache, get_new_inode_fast() is called to allocate a | |
1313 | * new inode and this is returned locked, hashed, and with the I_NEW flag set. | |
1314 | * The file system gets to fill it in before unlocking it via | |
1315 | * unlock_new_inode(). | |
1316 | */ | |
1317 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) | |
1318 | { | |
1319 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1320 | struct inode *inode; | |
1321 | ||
1322 | inode = ifind_fast(sb, head, ino); | |
1323 | if (inode) | |
1324 | return inode; | |
1325 | /* | |
1326 | * get_new_inode_fast() will do the right thing, re-trying the search | |
1327 | * in case it had to block at any point. | |
1328 | */ | |
1329 | return get_new_inode_fast(sb, head, ino); | |
1330 | } | |
1da177e4 LT |
1331 | EXPORT_SYMBOL(iget_locked); |
1332 | ||
261bca86 AV |
1333 | int insert_inode_locked(struct inode *inode) |
1334 | { | |
1335 | struct super_block *sb = inode->i_sb; | |
1336 | ino_t ino = inode->i_ino; | |
1337 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
261bca86 | 1338 | |
261bca86 | 1339 | while (1) { |
72a43d63 AV |
1340 | struct hlist_node *node; |
1341 | struct inode *old = NULL; | |
67a23c49 | 1342 | spin_lock(&inode_hash_lock); |
72a43d63 AV |
1343 | hlist_for_each_entry(old, node, head, i_hash) { |
1344 | if (old->i_ino != ino) | |
1345 | continue; | |
1346 | if (old->i_sb != sb) | |
1347 | continue; | |
250df6ed DC |
1348 | spin_lock(&old->i_lock); |
1349 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1350 | spin_unlock(&old->i_lock); | |
72a43d63 | 1351 | continue; |
250df6ed | 1352 | } |
72a43d63 AV |
1353 | break; |
1354 | } | |
1355 | if (likely(!node)) { | |
250df6ed DC |
1356 | spin_lock(&inode->i_lock); |
1357 | inode->i_state |= I_NEW; | |
261bca86 | 1358 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1359 | spin_unlock(&inode->i_lock); |
67a23c49 | 1360 | spin_unlock(&inode_hash_lock); |
261bca86 AV |
1361 | return 0; |
1362 | } | |
1363 | __iget(old); | |
250df6ed | 1364 | spin_unlock(&old->i_lock); |
67a23c49 | 1365 | spin_unlock(&inode_hash_lock); |
261bca86 | 1366 | wait_on_inode(old); |
1d3382cb | 1367 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1368 | iput(old); |
1369 | return -EBUSY; | |
1370 | } | |
1371 | iput(old); | |
1372 | } | |
1373 | } | |
261bca86 AV |
1374 | EXPORT_SYMBOL(insert_inode_locked); |
1375 | ||
1376 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1377 | int (*test)(struct inode *, void *), void *data) | |
1378 | { | |
1379 | struct super_block *sb = inode->i_sb; | |
1380 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
261bca86 | 1381 | |
261bca86 | 1382 | while (1) { |
72a43d63 AV |
1383 | struct hlist_node *node; |
1384 | struct inode *old = NULL; | |
1385 | ||
67a23c49 | 1386 | spin_lock(&inode_hash_lock); |
72a43d63 AV |
1387 | hlist_for_each_entry(old, node, head, i_hash) { |
1388 | if (old->i_sb != sb) | |
1389 | continue; | |
1390 | if (!test(old, data)) | |
1391 | continue; | |
250df6ed DC |
1392 | spin_lock(&old->i_lock); |
1393 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1394 | spin_unlock(&old->i_lock); | |
72a43d63 | 1395 | continue; |
250df6ed | 1396 | } |
72a43d63 AV |
1397 | break; |
1398 | } | |
1399 | if (likely(!node)) { | |
250df6ed DC |
1400 | spin_lock(&inode->i_lock); |
1401 | inode->i_state |= I_NEW; | |
261bca86 | 1402 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1403 | spin_unlock(&inode->i_lock); |
67a23c49 | 1404 | spin_unlock(&inode_hash_lock); |
261bca86 AV |
1405 | return 0; |
1406 | } | |
1407 | __iget(old); | |
250df6ed | 1408 | spin_unlock(&old->i_lock); |
67a23c49 | 1409 | spin_unlock(&inode_hash_lock); |
261bca86 | 1410 | wait_on_inode(old); |
1d3382cb | 1411 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1412 | iput(old); |
1413 | return -EBUSY; | |
1414 | } | |
1415 | iput(old); | |
1416 | } | |
1417 | } | |
261bca86 AV |
1418 | EXPORT_SYMBOL(insert_inode_locked4); |
1419 | ||
1da177e4 | 1420 | |
45321ac5 AV |
1421 | int generic_delete_inode(struct inode *inode) |
1422 | { | |
1423 | return 1; | |
1424 | } | |
1425 | EXPORT_SYMBOL(generic_delete_inode); | |
1426 | ||
1da177e4 | 1427 | /* |
45321ac5 AV |
1428 | * Normal UNIX filesystem behaviour: delete the |
1429 | * inode when the usage count drops to zero, and | |
1430 | * i_nlink is zero. | |
1da177e4 | 1431 | */ |
45321ac5 | 1432 | int generic_drop_inode(struct inode *inode) |
1da177e4 | 1433 | { |
1d3382cb | 1434 | return !inode->i_nlink || inode_unhashed(inode); |
1da177e4 | 1435 | } |
45321ac5 | 1436 | EXPORT_SYMBOL_GPL(generic_drop_inode); |
1da177e4 | 1437 | |
45321ac5 AV |
1438 | /* |
1439 | * Called when we're dropping the last reference | |
1440 | * to an inode. | |
22fe4042 | 1441 | * |
45321ac5 AV |
1442 | * Call the FS "drop_inode()" function, defaulting to |
1443 | * the legacy UNIX filesystem behaviour. If it tells | |
1444 | * us to evict inode, do so. Otherwise, retain inode | |
1445 | * in cache if fs is alive, sync and evict if fs is | |
1446 | * shutting down. | |
22fe4042 | 1447 | */ |
45321ac5 | 1448 | static void iput_final(struct inode *inode) |
1da177e4 LT |
1449 | { |
1450 | struct super_block *sb = inode->i_sb; | |
45321ac5 AV |
1451 | const struct super_operations *op = inode->i_sb->s_op; |
1452 | int drop; | |
1453 | ||
250df6ed DC |
1454 | WARN_ON(inode->i_state & I_NEW); |
1455 | ||
45321ac5 AV |
1456 | if (op && op->drop_inode) |
1457 | drop = op->drop_inode(inode); | |
1458 | else | |
1459 | drop = generic_drop_inode(inode); | |
1da177e4 | 1460 | |
b2b2af8e DC |
1461 | if (!drop && (sb->s_flags & MS_ACTIVE)) { |
1462 | inode->i_state |= I_REFERENCED; | |
1463 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) | |
1464 | inode_lru_list_add(inode); | |
1465 | spin_unlock(&inode->i_lock); | |
b2b2af8e DC |
1466 | return; |
1467 | } | |
1468 | ||
45321ac5 | 1469 | if (!drop) { |
991114c6 | 1470 | inode->i_state |= I_WILL_FREE; |
250df6ed | 1471 | spin_unlock(&inode->i_lock); |
1da177e4 | 1472 | write_inode_now(inode, 1); |
250df6ed | 1473 | spin_lock(&inode->i_lock); |
7ef0d737 | 1474 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1475 | inode->i_state &= ~I_WILL_FREE; |
1da177e4 | 1476 | } |
7ccf19a8 | 1477 | |
991114c6 | 1478 | inode->i_state |= I_FREEING; |
9e38d86f | 1479 | inode_lru_list_del(inode); |
b2b2af8e | 1480 | spin_unlock(&inode->i_lock); |
b2b2af8e | 1481 | |
644da596 | 1482 | evict(inode); |
1da177e4 LT |
1483 | } |
1484 | ||
1da177e4 | 1485 | /** |
6b3304b5 | 1486 | * iput - put an inode |
1da177e4 LT |
1487 | * @inode: inode to put |
1488 | * | |
1489 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1490 | * zero, the inode is then freed and may also be destroyed. | |
1491 | * | |
1492 | * Consequently, iput() can sleep. | |
1493 | */ | |
1494 | void iput(struct inode *inode) | |
1495 | { | |
1496 | if (inode) { | |
a4ffdde6 | 1497 | BUG_ON(inode->i_state & I_CLEAR); |
1da177e4 | 1498 | |
f283c86a | 1499 | if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) |
1da177e4 LT |
1500 | iput_final(inode); |
1501 | } | |
1502 | } | |
1da177e4 LT |
1503 | EXPORT_SYMBOL(iput); |
1504 | ||
1505 | /** | |
1506 | * bmap - find a block number in a file | |
1507 | * @inode: inode of file | |
1508 | * @block: block to find | |
1509 | * | |
1510 | * Returns the block number on the device holding the inode that | |
1511 | * is the disk block number for the block of the file requested. | |
1512 | * That is, asked for block 4 of inode 1 the function will return the | |
6b3304b5 | 1513 | * disk block relative to the disk start that holds that block of the |
1da177e4 LT |
1514 | * file. |
1515 | */ | |
6b3304b5 | 1516 | sector_t bmap(struct inode *inode, sector_t block) |
1da177e4 LT |
1517 | { |
1518 | sector_t res = 0; | |
1519 | if (inode->i_mapping->a_ops->bmap) | |
1520 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1521 | return res; | |
1522 | } | |
1da177e4 LT |
1523 | EXPORT_SYMBOL(bmap); |
1524 | ||
11ff6f05 MG |
1525 | /* |
1526 | * With relative atime, only update atime if the previous atime is | |
1527 | * earlier than either the ctime or mtime or if at least a day has | |
1528 | * passed since the last atime update. | |
1529 | */ | |
1530 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1531 | struct timespec now) | |
1532 | { | |
1533 | ||
1534 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1535 | return 1; | |
1536 | /* | |
1537 | * Is mtime younger than atime? If yes, update atime: | |
1538 | */ | |
1539 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1540 | return 1; | |
1541 | /* | |
1542 | * Is ctime younger than atime? If yes, update atime: | |
1543 | */ | |
1544 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1545 | return 1; | |
1546 | ||
1547 | /* | |
1548 | * Is the previous atime value older than a day? If yes, | |
1549 | * update atime: | |
1550 | */ | |
1551 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1552 | return 1; | |
1553 | /* | |
1554 | * Good, we can skip the atime update: | |
1555 | */ | |
1556 | return 0; | |
1557 | } | |
1558 | ||
1da177e4 | 1559 | /** |
869243a0 CH |
1560 | * touch_atime - update the access time |
1561 | * @mnt: mount the inode is accessed on | |
7045f37b | 1562 | * @dentry: dentry accessed |
1da177e4 LT |
1563 | * |
1564 | * Update the accessed time on an inode and mark it for writeback. | |
1565 | * This function automatically handles read only file systems and media, | |
1566 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1567 | */ | |
869243a0 | 1568 | void touch_atime(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 1569 | { |
869243a0 | 1570 | struct inode *inode = dentry->d_inode; |
1da177e4 LT |
1571 | struct timespec now; |
1572 | ||
cdb70f3f | 1573 | if (inode->i_flags & S_NOATIME) |
b12536c2 | 1574 | return; |
37756ced | 1575 | if (IS_NOATIME(inode)) |
b12536c2 | 1576 | return; |
b2276138 | 1577 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1578 | return; |
47ae32d6 | 1579 | |
cdb70f3f | 1580 | if (mnt->mnt_flags & MNT_NOATIME) |
b12536c2 | 1581 | return; |
cdb70f3f | 1582 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1583 | return; |
1da177e4 LT |
1584 | |
1585 | now = current_fs_time(inode->i_sb); | |
11ff6f05 MG |
1586 | |
1587 | if (!relatime_need_update(mnt, inode, now)) | |
b12536c2 | 1588 | return; |
11ff6f05 | 1589 | |
47ae32d6 | 1590 | if (timespec_equal(&inode->i_atime, &now)) |
b12536c2 AK |
1591 | return; |
1592 | ||
1593 | if (mnt_want_write(mnt)) | |
1594 | return; | |
47ae32d6 VH |
1595 | |
1596 | inode->i_atime = now; | |
1597 | mark_inode_dirty_sync(inode); | |
cdb70f3f | 1598 | mnt_drop_write(mnt); |
1da177e4 | 1599 | } |
869243a0 | 1600 | EXPORT_SYMBOL(touch_atime); |
1da177e4 LT |
1601 | |
1602 | /** | |
870f4817 CH |
1603 | * file_update_time - update mtime and ctime time |
1604 | * @file: file accessed | |
1da177e4 | 1605 | * |
870f4817 CH |
1606 | * Update the mtime and ctime members of an inode and mark the inode |
1607 | * for writeback. Note that this function is meant exclusively for | |
1608 | * usage in the file write path of filesystems, and filesystems may | |
1609 | * choose to explicitly ignore update via this function with the | |
2eadfc0e | 1610 | * S_NOCMTIME inode flag, e.g. for network filesystem where these |
870f4817 | 1611 | * timestamps are handled by the server. |
1da177e4 LT |
1612 | */ |
1613 | ||
870f4817 | 1614 | void file_update_time(struct file *file) |
1da177e4 | 1615 | { |
0f7fc9e4 | 1616 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 | 1617 | struct timespec now; |
ce06e0b2 | 1618 | enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0; |
1da177e4 | 1619 | |
ce06e0b2 | 1620 | /* First try to exhaust all avenues to not sync */ |
1da177e4 LT |
1621 | if (IS_NOCMTIME(inode)) |
1622 | return; | |
20ddee2c | 1623 | |
1da177e4 | 1624 | now = current_fs_time(inode->i_sb); |
ce06e0b2 AK |
1625 | if (!timespec_equal(&inode->i_mtime, &now)) |
1626 | sync_it = S_MTIME; | |
1da177e4 | 1627 | |
ce06e0b2 AK |
1628 | if (!timespec_equal(&inode->i_ctime, &now)) |
1629 | sync_it |= S_CTIME; | |
870f4817 | 1630 | |
ce06e0b2 AK |
1631 | if (IS_I_VERSION(inode)) |
1632 | sync_it |= S_VERSION; | |
7a224228 | 1633 | |
ce06e0b2 AK |
1634 | if (!sync_it) |
1635 | return; | |
1636 | ||
1637 | /* Finally allowed to write? Takes lock. */ | |
1638 | if (mnt_want_write_file(file)) | |
1639 | return; | |
1640 | ||
1641 | /* Only change inode inside the lock region */ | |
1642 | if (sync_it & S_VERSION) | |
1643 | inode_inc_iversion(inode); | |
1644 | if (sync_it & S_CTIME) | |
1645 | inode->i_ctime = now; | |
1646 | if (sync_it & S_MTIME) | |
1647 | inode->i_mtime = now; | |
1648 | mark_inode_dirty_sync(inode); | |
20ddee2c | 1649 | mnt_drop_write(file->f_path.mnt); |
1da177e4 | 1650 | } |
870f4817 | 1651 | EXPORT_SYMBOL(file_update_time); |
1da177e4 LT |
1652 | |
1653 | int inode_needs_sync(struct inode *inode) | |
1654 | { | |
1655 | if (IS_SYNC(inode)) | |
1656 | return 1; | |
1657 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1658 | return 1; | |
1659 | return 0; | |
1660 | } | |
1da177e4 LT |
1661 | EXPORT_SYMBOL(inode_needs_sync); |
1662 | ||
1da177e4 LT |
1663 | int inode_wait(void *word) |
1664 | { | |
1665 | schedule(); | |
1666 | return 0; | |
1667 | } | |
d44dab8d | 1668 | EXPORT_SYMBOL(inode_wait); |
1da177e4 LT |
1669 | |
1670 | /* | |
168a9fd6 MS |
1671 | * If we try to find an inode in the inode hash while it is being |
1672 | * deleted, we have to wait until the filesystem completes its | |
1673 | * deletion before reporting that it isn't found. This function waits | |
1674 | * until the deletion _might_ have completed. Callers are responsible | |
1675 | * to recheck inode state. | |
1676 | * | |
eaff8079 | 1677 | * It doesn't matter if I_NEW is not set initially, a call to |
250df6ed DC |
1678 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list |
1679 | * will DTRT. | |
1da177e4 LT |
1680 | */ |
1681 | static void __wait_on_freeing_inode(struct inode *inode) | |
1682 | { | |
1683 | wait_queue_head_t *wq; | |
eaff8079 CH |
1684 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); |
1685 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1da177e4 | 1686 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
250df6ed | 1687 | spin_unlock(&inode->i_lock); |
67a23c49 | 1688 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1689 | schedule(); |
1690 | finish_wait(wq, &wait.wait); | |
67a23c49 | 1691 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1692 | } |
1693 | ||
1da177e4 LT |
1694 | static __initdata unsigned long ihash_entries; |
1695 | static int __init set_ihash_entries(char *str) | |
1696 | { | |
1697 | if (!str) | |
1698 | return 0; | |
1699 | ihash_entries = simple_strtoul(str, &str, 0); | |
1700 | return 1; | |
1701 | } | |
1702 | __setup("ihash_entries=", set_ihash_entries); | |
1703 | ||
1704 | /* | |
1705 | * Initialize the waitqueues and inode hash table. | |
1706 | */ | |
1707 | void __init inode_init_early(void) | |
1708 | { | |
1709 | int loop; | |
1710 | ||
1711 | /* If hashes are distributed across NUMA nodes, defer | |
1712 | * hash allocation until vmalloc space is available. | |
1713 | */ | |
1714 | if (hashdist) | |
1715 | return; | |
1716 | ||
1717 | inode_hashtable = | |
1718 | alloc_large_system_hash("Inode-cache", | |
1719 | sizeof(struct hlist_head), | |
1720 | ihash_entries, | |
1721 | 14, | |
1722 | HASH_EARLY, | |
1723 | &i_hash_shift, | |
1724 | &i_hash_mask, | |
1725 | 0); | |
1726 | ||
1727 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1728 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1729 | } | |
1730 | ||
74bf17cf | 1731 | void __init inode_init(void) |
1da177e4 LT |
1732 | { |
1733 | int loop; | |
1734 | ||
1735 | /* inode slab cache */ | |
b0196009 PJ |
1736 | inode_cachep = kmem_cache_create("inode_cache", |
1737 | sizeof(struct inode), | |
1738 | 0, | |
1739 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1740 | SLAB_MEM_SPREAD), | |
20c2df83 | 1741 | init_once); |
8e1f936b | 1742 | register_shrinker(&icache_shrinker); |
1da177e4 LT |
1743 | |
1744 | /* Hash may have been set up in inode_init_early */ | |
1745 | if (!hashdist) | |
1746 | return; | |
1747 | ||
1748 | inode_hashtable = | |
1749 | alloc_large_system_hash("Inode-cache", | |
1750 | sizeof(struct hlist_head), | |
1751 | ihash_entries, | |
1752 | 14, | |
1753 | 0, | |
1754 | &i_hash_shift, | |
1755 | &i_hash_mask, | |
1756 | 0); | |
1757 | ||
1758 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1759 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1760 | } | |
1761 | ||
1762 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1763 | { | |
1764 | inode->i_mode = mode; | |
1765 | if (S_ISCHR(mode)) { | |
1766 | inode->i_fop = &def_chr_fops; | |
1767 | inode->i_rdev = rdev; | |
1768 | } else if (S_ISBLK(mode)) { | |
1769 | inode->i_fop = &def_blk_fops; | |
1770 | inode->i_rdev = rdev; | |
1771 | } else if (S_ISFIFO(mode)) | |
1772 | inode->i_fop = &def_fifo_fops; | |
1773 | else if (S_ISSOCK(mode)) | |
1774 | inode->i_fop = &bad_sock_fops; | |
1775 | else | |
af0d9ae8 MK |
1776 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" |
1777 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1778 | inode->i_ino); | |
1da177e4 LT |
1779 | } |
1780 | EXPORT_SYMBOL(init_special_inode); | |
a1bd120d DM |
1781 | |
1782 | /** | |
eaae668d | 1783 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards |
a1bd120d DM |
1784 | * @inode: New inode |
1785 | * @dir: Directory inode | |
1786 | * @mode: mode of the new inode | |
1787 | */ | |
1788 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
1789 | mode_t mode) | |
1790 | { | |
1791 | inode->i_uid = current_fsuid(); | |
1792 | if (dir && dir->i_mode & S_ISGID) { | |
1793 | inode->i_gid = dir->i_gid; | |
1794 | if (S_ISDIR(mode)) | |
1795 | mode |= S_ISGID; | |
1796 | } else | |
1797 | inode->i_gid = current_fsgid(); | |
1798 | inode->i_mode = mode; | |
1799 | } | |
1800 | EXPORT_SYMBOL(inode_init_owner); | |
e795b717 | 1801 | |
2e149670 SH |
1802 | /** |
1803 | * inode_owner_or_capable - check current task permissions to inode | |
1804 | * @inode: inode being checked | |
1805 | * | |
1806 | * Return true if current either has CAP_FOWNER to the inode, or | |
1807 | * owns the file. | |
e795b717 | 1808 | */ |
2e149670 | 1809 | bool inode_owner_or_capable(const struct inode *inode) |
e795b717 SH |
1810 | { |
1811 | struct user_namespace *ns = inode_userns(inode); | |
1812 | ||
1813 | if (current_user_ns() == ns && current_fsuid() == inode->i_uid) | |
1814 | return true; | |
1815 | if (ns_capable(ns, CAP_FOWNER)) | |
1816 | return true; | |
1817 | return false; | |
1818 | } | |
2e149670 | 1819 | EXPORT_SYMBOL(inode_owner_or_capable); |