Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 | 2 | * (C) 1997 Linus Torvalds |
4b4563dc | 3 | * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation) |
1da177e4 | 4 | */ |
1da177e4 LT |
5 | #include <linux/fs.h> |
6 | #include <linux/mm.h> | |
7 | #include <linux/dcache.h> | |
8 | #include <linux/init.h> | |
1da177e4 LT |
9 | #include <linux/slab.h> |
10 | #include <linux/writeback.h> | |
11 | #include <linux/module.h> | |
12 | #include <linux/backing-dev.h> | |
13 | #include <linux/wait.h> | |
88e0fbc4 | 14 | #include <linux/rwsem.h> |
1da177e4 LT |
15 | #include <linux/hash.h> |
16 | #include <linux/swap.h> | |
17 | #include <linux/security.h> | |
18 | #include <linux/pagemap.h> | |
19 | #include <linux/cdev.h> | |
20 | #include <linux/bootmem.h> | |
3be25f49 | 21 | #include <linux/fsnotify.h> |
fc33a7bb | 22 | #include <linux/mount.h> |
efaee192 | 23 | #include <linux/async.h> |
f19d4a8f | 24 | #include <linux/posix_acl.h> |
9ce6e0be | 25 | #include <linux/prefetch.h> |
a178d202 | 26 | #include <linux/ima.h> |
e795b717 | 27 | #include <linux/cred.h> |
4b4563dc | 28 | #include <linux/buffer_head.h> /* for inode_has_buffers */ |
a66979ab | 29 | #include "internal.h" |
1da177e4 | 30 | |
250df6ed | 31 | /* |
4b4563dc | 32 | * Inode locking rules: |
250df6ed DC |
33 | * |
34 | * inode->i_lock protects: | |
35 | * inode->i_state, inode->i_hash, __iget() | |
09cc9fc7 | 36 | * inode->i_sb->s_inode_lru_lock protects: |
98b745c6 | 37 | * inode->i_sb->s_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 | |
09cc9fc7 | 49 | * inode->i_sb->s_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 | ||
fa3536cc ED |
62 | static unsigned int i_hash_mask __read_mostly; |
63 | static unsigned int i_hash_shift __read_mostly; | |
67a23c49 DC |
64 | static struct hlist_head *inode_hashtable __read_mostly; |
65 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock); | |
1da177e4 | 66 | |
55fa6091 | 67 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock); |
a66979ab | 68 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_wb_list_lock); |
55fa6091 | 69 | |
1da177e4 | 70 | /* |
bab1d944 CH |
71 | * iprune_sem provides exclusion between the icache shrinking and the |
72 | * umount path. | |
88e0fbc4 | 73 | * |
bab1d944 CH |
74 | * We don't actually need it to protect anything in the umount path, |
75 | * but only need to cycle through it to make sure any inode that | |
76 | * prune_icache took off the LRU list has been fully torn down by the | |
77 | * time we are past evict_inodes. | |
1da177e4 | 78 | */ |
88e0fbc4 | 79 | static DECLARE_RWSEM(iprune_sem); |
1da177e4 | 80 | |
7dcda1c9 JA |
81 | /* |
82 | * Empty aops. Can be used for the cases where the user does not | |
83 | * define any of the address_space operations. | |
84 | */ | |
85 | const struct address_space_operations empty_aops = { | |
86 | }; | |
87 | EXPORT_SYMBOL(empty_aops); | |
88 | ||
1da177e4 LT |
89 | /* |
90 | * Statistics gathering.. | |
91 | */ | |
92 | struct inodes_stat_t inodes_stat; | |
93 | ||
3e880fb5 | 94 | static DEFINE_PER_CPU(unsigned int, nr_inodes); |
fcb94f72 | 95 | static DEFINE_PER_CPU(unsigned int, nr_unused); |
cffbc8aa | 96 | |
6b3304b5 | 97 | static struct kmem_cache *inode_cachep __read_mostly; |
1da177e4 | 98 | |
3e880fb5 | 99 | static int get_nr_inodes(void) |
cffbc8aa | 100 | { |
3e880fb5 NP |
101 | int i; |
102 | int sum = 0; | |
103 | for_each_possible_cpu(i) | |
104 | sum += per_cpu(nr_inodes, i); | |
105 | return sum < 0 ? 0 : sum; | |
cffbc8aa DC |
106 | } |
107 | ||
108 | static inline int get_nr_inodes_unused(void) | |
109 | { | |
fcb94f72 DC |
110 | int i; |
111 | int sum = 0; | |
112 | for_each_possible_cpu(i) | |
113 | sum += per_cpu(nr_unused, i); | |
114 | return sum < 0 ? 0 : sum; | |
cffbc8aa DC |
115 | } |
116 | ||
117 | int get_nr_dirty_inodes(void) | |
118 | { | |
3e880fb5 | 119 | /* not actually dirty inodes, but a wild approximation */ |
cffbc8aa DC |
120 | int nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); |
121 | return nr_dirty > 0 ? nr_dirty : 0; | |
cffbc8aa DC |
122 | } |
123 | ||
124 | /* | |
125 | * Handle nr_inode sysctl | |
126 | */ | |
127 | #ifdef CONFIG_SYSCTL | |
128 | int proc_nr_inodes(ctl_table *table, int write, | |
129 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
130 | { | |
131 | inodes_stat.nr_inodes = get_nr_inodes(); | |
fcb94f72 | 132 | inodes_stat.nr_unused = get_nr_inodes_unused(); |
cffbc8aa DC |
133 | return proc_dointvec(table, write, buffer, lenp, ppos); |
134 | } | |
135 | #endif | |
136 | ||
2cb1599f DC |
137 | /** |
138 | * inode_init_always - perform inode structure intialisation | |
0bc02f3f RD |
139 | * @sb: superblock inode belongs to |
140 | * @inode: inode to initialise | |
2cb1599f DC |
141 | * |
142 | * These are initializations that need to be done on every inode | |
143 | * allocation as the fields are not initialised by slab allocation. | |
144 | */ | |
54e34621 | 145 | int inode_init_always(struct super_block *sb, struct inode *inode) |
1da177e4 | 146 | { |
6e1d5dcc | 147 | static const struct inode_operations empty_iops; |
99ac48f5 | 148 | static const struct file_operations empty_fops; |
6b3304b5 | 149 | struct address_space *const mapping = &inode->i_data; |
2cb1599f DC |
150 | |
151 | inode->i_sb = sb; | |
152 | inode->i_blkbits = sb->s_blocksize_bits; | |
153 | inode->i_flags = 0; | |
154 | atomic_set(&inode->i_count, 1); | |
155 | inode->i_op = &empty_iops; | |
156 | inode->i_fop = &empty_fops; | |
157 | inode->i_nlink = 1; | |
56ff5efa AV |
158 | inode->i_uid = 0; |
159 | inode->i_gid = 0; | |
2cb1599f DC |
160 | atomic_set(&inode->i_writecount, 0); |
161 | inode->i_size = 0; | |
162 | inode->i_blocks = 0; | |
163 | inode->i_bytes = 0; | |
164 | inode->i_generation = 0; | |
1da177e4 | 165 | #ifdef CONFIG_QUOTA |
2cb1599f | 166 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); |
1da177e4 | 167 | #endif |
2cb1599f DC |
168 | inode->i_pipe = NULL; |
169 | inode->i_bdev = NULL; | |
170 | inode->i_cdev = NULL; | |
171 | inode->i_rdev = 0; | |
172 | inode->dirtied_when = 0; | |
6146f0d5 MZ |
173 | |
174 | if (security_inode_alloc(inode)) | |
54e34621 | 175 | goto out; |
2cb1599f DC |
176 | spin_lock_init(&inode->i_lock); |
177 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
178 | ||
179 | mutex_init(&inode->i_mutex); | |
180 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); | |
181 | ||
182 | init_rwsem(&inode->i_alloc_sem); | |
183 | lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key); | |
184 | ||
185 | mapping->a_ops = &empty_aops; | |
186 | mapping->host = inode; | |
187 | mapping->flags = 0; | |
3c1d4378 | 188 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); |
2cb1599f DC |
189 | mapping->assoc_mapping = NULL; |
190 | mapping->backing_dev_info = &default_backing_dev_info; | |
191 | mapping->writeback_index = 0; | |
192 | ||
193 | /* | |
194 | * If the block_device provides a backing_dev_info for client | |
195 | * inodes then use that. Otherwise the inode share the bdev's | |
196 | * backing_dev_info. | |
197 | */ | |
198 | if (sb->s_bdev) { | |
199 | struct backing_dev_info *bdi; | |
200 | ||
2c96ce9f | 201 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; |
2cb1599f DC |
202 | mapping->backing_dev_info = bdi; |
203 | } | |
204 | inode->i_private = NULL; | |
205 | inode->i_mapping = mapping; | |
f19d4a8f AV |
206 | #ifdef CONFIG_FS_POSIX_ACL |
207 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; | |
208 | #endif | |
2cb1599f | 209 | |
3be25f49 EP |
210 | #ifdef CONFIG_FSNOTIFY |
211 | inode->i_fsnotify_mask = 0; | |
212 | #endif | |
213 | ||
3e880fb5 | 214 | this_cpu_inc(nr_inodes); |
cffbc8aa | 215 | |
54e34621 | 216 | return 0; |
54e34621 CH |
217 | out: |
218 | return -ENOMEM; | |
1da177e4 | 219 | } |
2cb1599f DC |
220 | EXPORT_SYMBOL(inode_init_always); |
221 | ||
222 | static struct inode *alloc_inode(struct super_block *sb) | |
223 | { | |
224 | struct inode *inode; | |
225 | ||
226 | if (sb->s_op->alloc_inode) | |
227 | inode = sb->s_op->alloc_inode(sb); | |
228 | else | |
229 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
230 | ||
54e34621 CH |
231 | if (!inode) |
232 | return NULL; | |
233 | ||
234 | if (unlikely(inode_init_always(sb, inode))) { | |
235 | if (inode->i_sb->s_op->destroy_inode) | |
236 | inode->i_sb->s_op->destroy_inode(inode); | |
237 | else | |
238 | kmem_cache_free(inode_cachep, inode); | |
239 | return NULL; | |
240 | } | |
241 | ||
242 | return inode; | |
2cb1599f | 243 | } |
1da177e4 | 244 | |
ff0c7d15 NP |
245 | void free_inode_nonrcu(struct inode *inode) |
246 | { | |
247 | kmem_cache_free(inode_cachep, inode); | |
248 | } | |
249 | EXPORT_SYMBOL(free_inode_nonrcu); | |
250 | ||
2e00c97e | 251 | void __destroy_inode(struct inode *inode) |
1da177e4 | 252 | { |
b7542f8c | 253 | BUG_ON(inode_has_buffers(inode)); |
1da177e4 | 254 | security_inode_free(inode); |
3be25f49 | 255 | fsnotify_inode_delete(inode); |
f19d4a8f AV |
256 | #ifdef CONFIG_FS_POSIX_ACL |
257 | if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) | |
258 | posix_acl_release(inode->i_acl); | |
259 | if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) | |
260 | posix_acl_release(inode->i_default_acl); | |
261 | #endif | |
3e880fb5 | 262 | this_cpu_dec(nr_inodes); |
2e00c97e CH |
263 | } |
264 | EXPORT_SYMBOL(__destroy_inode); | |
265 | ||
fa0d7e3d NP |
266 | static void i_callback(struct rcu_head *head) |
267 | { | |
268 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
269 | INIT_LIST_HEAD(&inode->i_dentry); | |
270 | kmem_cache_free(inode_cachep, inode); | |
271 | } | |
272 | ||
56b0dacf | 273 | static void destroy_inode(struct inode *inode) |
2e00c97e | 274 | { |
7ccf19a8 | 275 | BUG_ON(!list_empty(&inode->i_lru)); |
2e00c97e | 276 | __destroy_inode(inode); |
1da177e4 LT |
277 | if (inode->i_sb->s_op->destroy_inode) |
278 | inode->i_sb->s_op->destroy_inode(inode); | |
279 | else | |
fa0d7e3d | 280 | call_rcu(&inode->i_rcu, i_callback); |
1da177e4 | 281 | } |
1da177e4 | 282 | |
2aa15890 MS |
283 | void address_space_init_once(struct address_space *mapping) |
284 | { | |
285 | memset(mapping, 0, sizeof(*mapping)); | |
286 | INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC); | |
287 | spin_lock_init(&mapping->tree_lock); | |
3d48ae45 | 288 | mutex_init(&mapping->i_mmap_mutex); |
2aa15890 MS |
289 | INIT_LIST_HEAD(&mapping->private_list); |
290 | spin_lock_init(&mapping->private_lock); | |
291 | INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap); | |
292 | INIT_LIST_HEAD(&mapping->i_mmap_nonlinear); | |
2aa15890 MS |
293 | } |
294 | EXPORT_SYMBOL(address_space_init_once); | |
295 | ||
1da177e4 LT |
296 | /* |
297 | * These are initializations that only need to be done | |
298 | * once, because the fields are idempotent across use | |
299 | * of the inode, so let the slab aware of that. | |
300 | */ | |
301 | void inode_init_once(struct inode *inode) | |
302 | { | |
303 | memset(inode, 0, sizeof(*inode)); | |
304 | INIT_HLIST_NODE(&inode->i_hash); | |
305 | INIT_LIST_HEAD(&inode->i_dentry); | |
306 | INIT_LIST_HEAD(&inode->i_devices); | |
7ccf19a8 NP |
307 | INIT_LIST_HEAD(&inode->i_wb_list); |
308 | INIT_LIST_HEAD(&inode->i_lru); | |
2aa15890 | 309 | address_space_init_once(&inode->i_data); |
1da177e4 | 310 | i_size_ordered_init(inode); |
3be25f49 | 311 | #ifdef CONFIG_FSNOTIFY |
e61ce867 | 312 | INIT_HLIST_HEAD(&inode->i_fsnotify_marks); |
3be25f49 | 313 | #endif |
1da177e4 | 314 | } |
1da177e4 LT |
315 | EXPORT_SYMBOL(inode_init_once); |
316 | ||
51cc5068 | 317 | static void init_once(void *foo) |
1da177e4 | 318 | { |
6b3304b5 | 319 | struct inode *inode = (struct inode *) foo; |
1da177e4 | 320 | |
a35afb83 | 321 | inode_init_once(inode); |
1da177e4 LT |
322 | } |
323 | ||
324 | /* | |
250df6ed | 325 | * inode->i_lock must be held |
1da177e4 | 326 | */ |
6b3304b5 | 327 | void __iget(struct inode *inode) |
1da177e4 | 328 | { |
9e38d86f NP |
329 | atomic_inc(&inode->i_count); |
330 | } | |
2e147f1e | 331 | |
7de9c6ee AV |
332 | /* |
333 | * get additional reference to inode; caller must already hold one. | |
334 | */ | |
335 | void ihold(struct inode *inode) | |
336 | { | |
337 | WARN_ON(atomic_inc_return(&inode->i_count) < 2); | |
338 | } | |
339 | EXPORT_SYMBOL(ihold); | |
340 | ||
9e38d86f NP |
341 | static void inode_lru_list_add(struct inode *inode) |
342 | { | |
09cc9fc7 | 343 | spin_lock(&inode->i_sb->s_inode_lru_lock); |
7ccf19a8 | 344 | if (list_empty(&inode->i_lru)) { |
98b745c6 DC |
345 | list_add(&inode->i_lru, &inode->i_sb->s_inode_lru); |
346 | inode->i_sb->s_nr_inodes_unused++; | |
fcb94f72 | 347 | this_cpu_inc(nr_unused); |
9e38d86f | 348 | } |
09cc9fc7 | 349 | spin_unlock(&inode->i_sb->s_inode_lru_lock); |
9e38d86f | 350 | } |
2e147f1e | 351 | |
9e38d86f NP |
352 | static void inode_lru_list_del(struct inode *inode) |
353 | { | |
09cc9fc7 | 354 | spin_lock(&inode->i_sb->s_inode_lru_lock); |
7ccf19a8 NP |
355 | if (!list_empty(&inode->i_lru)) { |
356 | list_del_init(&inode->i_lru); | |
98b745c6 | 357 | inode->i_sb->s_nr_inodes_unused--; |
fcb94f72 | 358 | this_cpu_dec(nr_unused); |
9e38d86f | 359 | } |
09cc9fc7 | 360 | spin_unlock(&inode->i_sb->s_inode_lru_lock); |
1da177e4 LT |
361 | } |
362 | ||
646ec461 CH |
363 | /** |
364 | * inode_sb_list_add - add inode to the superblock list of inodes | |
365 | * @inode: inode to add | |
366 | */ | |
367 | void inode_sb_list_add(struct inode *inode) | |
368 | { | |
55fa6091 DC |
369 | spin_lock(&inode_sb_list_lock); |
370 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); | |
371 | spin_unlock(&inode_sb_list_lock); | |
646ec461 CH |
372 | } |
373 | EXPORT_SYMBOL_GPL(inode_sb_list_add); | |
374 | ||
55fa6091 | 375 | static inline void inode_sb_list_del(struct inode *inode) |
646ec461 | 376 | { |
55fa6091 | 377 | spin_lock(&inode_sb_list_lock); |
646ec461 | 378 | list_del_init(&inode->i_sb_list); |
55fa6091 | 379 | spin_unlock(&inode_sb_list_lock); |
646ec461 CH |
380 | } |
381 | ||
4c51acbc DC |
382 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
383 | { | |
384 | unsigned long tmp; | |
385 | ||
386 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
387 | L1_CACHE_BYTES; | |
4b4563dc CH |
388 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift); |
389 | return tmp & i_hash_mask; | |
4c51acbc DC |
390 | } |
391 | ||
392 | /** | |
393 | * __insert_inode_hash - hash an inode | |
394 | * @inode: unhashed inode | |
395 | * @hashval: unsigned long value used to locate this object in the | |
396 | * inode_hashtable. | |
397 | * | |
398 | * Add an inode to the inode hash for this superblock. | |
399 | */ | |
400 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
401 | { | |
646ec461 CH |
402 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); |
403 | ||
67a23c49 | 404 | spin_lock(&inode_hash_lock); |
250df6ed | 405 | spin_lock(&inode->i_lock); |
646ec461 | 406 | hlist_add_head(&inode->i_hash, b); |
250df6ed | 407 | spin_unlock(&inode->i_lock); |
67a23c49 | 408 | spin_unlock(&inode_hash_lock); |
4c51acbc DC |
409 | } |
410 | EXPORT_SYMBOL(__insert_inode_hash); | |
411 | ||
4c51acbc DC |
412 | /** |
413 | * remove_inode_hash - remove an inode from the hash | |
414 | * @inode: inode to unhash | |
415 | * | |
416 | * Remove an inode from the superblock. | |
417 | */ | |
418 | void remove_inode_hash(struct inode *inode) | |
419 | { | |
67a23c49 | 420 | spin_lock(&inode_hash_lock); |
250df6ed | 421 | spin_lock(&inode->i_lock); |
4c51acbc | 422 | hlist_del_init(&inode->i_hash); |
250df6ed | 423 | spin_unlock(&inode->i_lock); |
67a23c49 | 424 | spin_unlock(&inode_hash_lock); |
4c51acbc DC |
425 | } |
426 | EXPORT_SYMBOL(remove_inode_hash); | |
427 | ||
b0683aa6 AV |
428 | void end_writeback(struct inode *inode) |
429 | { | |
430 | might_sleep(); | |
08142579 JK |
431 | /* |
432 | * We have to cycle tree_lock here because reclaim can be still in the | |
433 | * process of removing the last page (in __delete_from_page_cache()) | |
434 | * and we must not free mapping under it. | |
435 | */ | |
436 | spin_lock_irq(&inode->i_data.tree_lock); | |
b0683aa6 | 437 | BUG_ON(inode->i_data.nrpages); |
08142579 | 438 | spin_unlock_irq(&inode->i_data.tree_lock); |
b0683aa6 AV |
439 | BUG_ON(!list_empty(&inode->i_data.private_list)); |
440 | BUG_ON(!(inode->i_state & I_FREEING)); | |
441 | BUG_ON(inode->i_state & I_CLEAR); | |
442 | inode_sync_wait(inode); | |
fa0d7e3d | 443 | /* don't need i_lock here, no concurrent mods to i_state */ |
b0683aa6 AV |
444 | inode->i_state = I_FREEING | I_CLEAR; |
445 | } | |
446 | EXPORT_SYMBOL(end_writeback); | |
447 | ||
b2b2af8e DC |
448 | /* |
449 | * Free the inode passed in, removing it from the lists it is still connected | |
450 | * to. We remove any pages still attached to the inode and wait for any IO that | |
451 | * is still in progress before finally destroying the inode. | |
452 | * | |
453 | * An inode must already be marked I_FREEING so that we avoid the inode being | |
454 | * moved back onto lists if we race with other code that manipulates the lists | |
455 | * (e.g. writeback_single_inode). The caller is responsible for setting this. | |
456 | * | |
457 | * An inode must already be removed from the LRU list before being evicted from | |
458 | * the cache. This should occur atomically with setting the I_FREEING state | |
459 | * flag, so no inodes here should ever be on the LRU when being evicted. | |
460 | */ | |
644da596 | 461 | static void evict(struct inode *inode) |
b4272d4c AV |
462 | { |
463 | const struct super_operations *op = inode->i_sb->s_op; | |
464 | ||
b2b2af8e DC |
465 | BUG_ON(!(inode->i_state & I_FREEING)); |
466 | BUG_ON(!list_empty(&inode->i_lru)); | |
467 | ||
a66979ab | 468 | inode_wb_list_del(inode); |
55fa6091 DC |
469 | inode_sb_list_del(inode); |
470 | ||
be7ce416 AV |
471 | if (op->evict_inode) { |
472 | op->evict_inode(inode); | |
b4272d4c AV |
473 | } else { |
474 | if (inode->i_data.nrpages) | |
475 | truncate_inode_pages(&inode->i_data, 0); | |
30140837 | 476 | end_writeback(inode); |
b4272d4c | 477 | } |
661074e9 AV |
478 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
479 | bd_forget(inode); | |
480 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) | |
481 | cd_forget(inode); | |
b2b2af8e DC |
482 | |
483 | remove_inode_hash(inode); | |
484 | ||
485 | spin_lock(&inode->i_lock); | |
486 | wake_up_bit(&inode->i_state, __I_NEW); | |
487 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); | |
488 | spin_unlock(&inode->i_lock); | |
489 | ||
490 | destroy_inode(inode); | |
b4272d4c AV |
491 | } |
492 | ||
1da177e4 LT |
493 | /* |
494 | * dispose_list - dispose of the contents of a local list | |
495 | * @head: the head of the list to free | |
496 | * | |
497 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
498 | * need to worry about list corruption and SMP locks. | |
499 | */ | |
500 | static void dispose_list(struct list_head *head) | |
501 | { | |
1da177e4 LT |
502 | while (!list_empty(head)) { |
503 | struct inode *inode; | |
504 | ||
7ccf19a8 NP |
505 | inode = list_first_entry(head, struct inode, i_lru); |
506 | list_del_init(&inode->i_lru); | |
1da177e4 | 507 | |
644da596 | 508 | evict(inode); |
1da177e4 | 509 | } |
1da177e4 LT |
510 | } |
511 | ||
63997e98 AV |
512 | /** |
513 | * evict_inodes - evict all evictable inodes for a superblock | |
514 | * @sb: superblock to operate on | |
515 | * | |
516 | * Make sure that no inodes with zero refcount are retained. This is | |
517 | * called by superblock shutdown after having MS_ACTIVE flag removed, | |
518 | * so any inode reaching zero refcount during or after that call will | |
519 | * be immediately evicted. | |
1da177e4 | 520 | */ |
63997e98 | 521 | void evict_inodes(struct super_block *sb) |
1da177e4 | 522 | { |
63997e98 AV |
523 | struct inode *inode, *next; |
524 | LIST_HEAD(dispose); | |
1da177e4 | 525 | |
55fa6091 | 526 | spin_lock(&inode_sb_list_lock); |
63997e98 AV |
527 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
528 | if (atomic_read(&inode->i_count)) | |
aabb8fdb | 529 | continue; |
250df6ed DC |
530 | |
531 | spin_lock(&inode->i_lock); | |
532 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
533 | spin_unlock(&inode->i_lock); | |
1da177e4 | 534 | continue; |
250df6ed | 535 | } |
63997e98 AV |
536 | |
537 | inode->i_state |= I_FREEING; | |
02afc410 | 538 | inode_lru_list_del(inode); |
250df6ed | 539 | spin_unlock(&inode->i_lock); |
02afc410 | 540 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 541 | } |
55fa6091 | 542 | spin_unlock(&inode_sb_list_lock); |
63997e98 AV |
543 | |
544 | dispose_list(&dispose); | |
bab1d944 CH |
545 | |
546 | /* | |
547 | * Cycle through iprune_sem to make sure any inode that prune_icache | |
548 | * moved off the list before we took the lock has been fully torn | |
549 | * down. | |
550 | */ | |
551 | down_write(&iprune_sem); | |
63997e98 | 552 | up_write(&iprune_sem); |
1da177e4 LT |
553 | } |
554 | ||
1da177e4 | 555 | /** |
a0318786 CH |
556 | * invalidate_inodes - attempt to free all inodes on a superblock |
557 | * @sb: superblock to operate on | |
93b270f7 | 558 | * @kill_dirty: flag to guide handling of dirty inodes |
1da177e4 | 559 | * |
a0318786 CH |
560 | * Attempts to free all inodes for a given superblock. If there were any |
561 | * busy inodes return a non-zero value, else zero. | |
93b270f7 N |
562 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat |
563 | * them as busy. | |
1da177e4 | 564 | */ |
93b270f7 | 565 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) |
1da177e4 | 566 | { |
cffbc8aa | 567 | int busy = 0; |
a0318786 CH |
568 | struct inode *inode, *next; |
569 | LIST_HEAD(dispose); | |
1da177e4 | 570 | |
55fa6091 | 571 | spin_lock(&inode_sb_list_lock); |
a0318786 | 572 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
250df6ed DC |
573 | spin_lock(&inode->i_lock); |
574 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
575 | spin_unlock(&inode->i_lock); | |
aabb8fdb | 576 | continue; |
250df6ed | 577 | } |
93b270f7 | 578 | if (inode->i_state & I_DIRTY && !kill_dirty) { |
250df6ed | 579 | spin_unlock(&inode->i_lock); |
93b270f7 N |
580 | busy = 1; |
581 | continue; | |
582 | } | |
99a38919 | 583 | if (atomic_read(&inode->i_count)) { |
250df6ed | 584 | spin_unlock(&inode->i_lock); |
99a38919 | 585 | busy = 1; |
1da177e4 LT |
586 | continue; |
587 | } | |
99a38919 | 588 | |
99a38919 | 589 | inode->i_state |= I_FREEING; |
02afc410 | 590 | inode_lru_list_del(inode); |
250df6ed | 591 | spin_unlock(&inode->i_lock); |
02afc410 | 592 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 593 | } |
55fa6091 | 594 | spin_unlock(&inode_sb_list_lock); |
1da177e4 | 595 | |
a0318786 | 596 | dispose_list(&dispose); |
1da177e4 LT |
597 | |
598 | return busy; | |
599 | } | |
1da177e4 LT |
600 | |
601 | static int can_unuse(struct inode *inode) | |
602 | { | |
9e38d86f | 603 | if (inode->i_state & ~I_REFERENCED) |
1da177e4 LT |
604 | return 0; |
605 | if (inode_has_buffers(inode)) | |
606 | return 0; | |
607 | if (atomic_read(&inode->i_count)) | |
608 | return 0; | |
609 | if (inode->i_data.nrpages) | |
610 | return 0; | |
611 | return 1; | |
612 | } | |
613 | ||
614 | /* | |
9e38d86f | 615 | * Scan `goal' inodes on the unused list for freeable ones. They are moved to a |
09cc9fc7 DC |
616 | * temporary list and then are freed outside sb->s_inode_lru_lock by |
617 | * dispose_list(). | |
1da177e4 LT |
618 | * |
619 | * Any inodes which are pinned purely because of attached pagecache have their | |
9e38d86f NP |
620 | * pagecache removed. If the inode has metadata buffers attached to |
621 | * mapping->private_list then try to remove them. | |
1da177e4 | 622 | * |
9e38d86f NP |
623 | * If the inode has the I_REFERENCED flag set, then it means that it has been |
624 | * used recently - the flag is set in iput_final(). When we encounter such an | |
625 | * inode, clear the flag and move it to the back of the LRU so it gets another | |
626 | * pass through the LRU before it gets reclaimed. This is necessary because of | |
627 | * the fact we are doing lazy LRU updates to minimise lock contention so the | |
628 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes | |
629 | * with this flag set because they are the inodes that are out of order. | |
1da177e4 | 630 | */ |
98b745c6 | 631 | static void shrink_icache_sb(struct super_block *sb, int *nr_to_scan) |
1da177e4 LT |
632 | { |
633 | LIST_HEAD(freeable); | |
1da177e4 LT |
634 | int nr_scanned; |
635 | unsigned long reap = 0; | |
636 | ||
09cc9fc7 | 637 | spin_lock(&sb->s_inode_lru_lock); |
98b745c6 | 638 | for (nr_scanned = *nr_to_scan; nr_scanned >= 0; nr_scanned--) { |
1da177e4 LT |
639 | struct inode *inode; |
640 | ||
98b745c6 | 641 | if (list_empty(&sb->s_inode_lru)) |
1da177e4 LT |
642 | break; |
643 | ||
98b745c6 | 644 | inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru); |
1da177e4 | 645 | |
02afc410 | 646 | /* |
09cc9fc7 | 647 | * we are inverting the sb->s_inode_lru_lock/inode->i_lock here, |
02afc410 DC |
648 | * so use a trylock. If we fail to get the lock, just move the |
649 | * inode to the back of the list so we don't spin on it. | |
650 | */ | |
651 | if (!spin_trylock(&inode->i_lock)) { | |
98b745c6 | 652 | list_move(&inode->i_lru, &sb->s_inode_lru); |
02afc410 DC |
653 | continue; |
654 | } | |
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 | */ | |
660 | if (atomic_read(&inode->i_count) || | |
661 | (inode->i_state & ~I_REFERENCED)) { | |
7ccf19a8 | 662 | list_del_init(&inode->i_lru); |
f283c86a | 663 | spin_unlock(&inode->i_lock); |
98b745c6 | 664 | sb->s_nr_inodes_unused--; |
fcb94f72 | 665 | this_cpu_dec(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; |
98b745c6 | 672 | list_move(&inode->i_lru, &sb->s_inode_lru); |
f283c86a | 673 | spin_unlock(&inode->i_lock); |
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); |
09cc9fc7 | 679 | spin_unlock(&sb->s_inode_lru_lock); |
1da177e4 | 680 | if (remove_inode_buffers(inode)) |
fc0ecff6 AM |
681 | reap += invalidate_mapping_pages(&inode->i_data, |
682 | 0, -1); | |
1da177e4 | 683 | iput(inode); |
09cc9fc7 | 684 | spin_lock(&sb->s_inode_lru_lock); |
1da177e4 | 685 | |
98b745c6 | 686 | if (inode != list_entry(sb->s_inode_lru.next, |
7ccf19a8 | 687 | struct inode, i_lru)) |
1da177e4 | 688 | continue; /* wrong inode or list_empty */ |
02afc410 DC |
689 | /* avoid lock inversions with trylock */ |
690 | if (!spin_trylock(&inode->i_lock)) | |
691 | continue; | |
250df6ed DC |
692 | if (!can_unuse(inode)) { |
693 | spin_unlock(&inode->i_lock); | |
1da177e4 | 694 | continue; |
250df6ed | 695 | } |
1da177e4 | 696 | } |
7ef0d737 | 697 | WARN_ON(inode->i_state & I_NEW); |
1da177e4 | 698 | inode->i_state |= I_FREEING; |
250df6ed | 699 | spin_unlock(&inode->i_lock); |
7ccf19a8 | 700 | |
7ccf19a8 | 701 | list_move(&inode->i_lru, &freeable); |
98b745c6 | 702 | sb->s_nr_inodes_unused--; |
fcb94f72 | 703 | this_cpu_dec(nr_unused); |
1da177e4 | 704 | } |
f8891e5e CL |
705 | if (current_is_kswapd()) |
706 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
707 | else | |
708 | __count_vm_events(PGINODESTEAL, reap); | |
09cc9fc7 | 709 | spin_unlock(&sb->s_inode_lru_lock); |
98b745c6 | 710 | *nr_to_scan = nr_scanned; |
1da177e4 LT |
711 | |
712 | dispose_list(&freeable); | |
98b745c6 DC |
713 | } |
714 | ||
715 | static void prune_icache(int count) | |
716 | { | |
717 | struct super_block *sb, *p = NULL; | |
718 | int w_count; | |
719 | int unused = inodes_stat.nr_unused; | |
720 | int prune_ratio; | |
721 | int pruned; | |
722 | ||
723 | if (unused == 0 || count == 0) | |
724 | return; | |
725 | down_read(&iprune_sem); | |
726 | if (count >= unused) | |
727 | prune_ratio = 1; | |
728 | else | |
729 | prune_ratio = unused / count; | |
730 | spin_lock(&sb_lock); | |
731 | list_for_each_entry(sb, &super_blocks, s_list) { | |
732 | if (list_empty(&sb->s_instances)) | |
733 | continue; | |
734 | if (sb->s_nr_inodes_unused == 0) | |
735 | continue; | |
736 | sb->s_count++; | |
737 | /* Now, we reclaim unused dentrins with fairness. | |
738 | * We reclaim them same percentage from each superblock. | |
739 | * We calculate number of dentries to scan on this sb | |
740 | * as follows, but the implementation is arranged to avoid | |
741 | * overflows: | |
742 | * number of dentries to scan on this sb = | |
743 | * count * (number of dentries on this sb / | |
744 | * number of dentries in the machine) | |
745 | */ | |
746 | spin_unlock(&sb_lock); | |
747 | if (prune_ratio != 1) | |
748 | w_count = (sb->s_nr_inodes_unused / prune_ratio) + 1; | |
749 | else | |
750 | w_count = sb->s_nr_inodes_unused; | |
751 | pruned = w_count; | |
752 | /* | |
753 | * We need to be sure this filesystem isn't being unmounted, | |
754 | * otherwise we could race with generic_shutdown_super(), and | |
755 | * end up holding a reference to an inode while the filesystem | |
756 | * is unmounted. So we try to get s_umount, and make sure | |
757 | * s_root isn't NULL. | |
758 | */ | |
759 | if (down_read_trylock(&sb->s_umount)) { | |
760 | if ((sb->s_root != NULL) && | |
761 | (!list_empty(&sb->s_dentry_lru))) { | |
762 | shrink_icache_sb(sb, &w_count); | |
763 | pruned -= w_count; | |
764 | } | |
765 | up_read(&sb->s_umount); | |
766 | } | |
767 | spin_lock(&sb_lock); | |
768 | if (p) | |
769 | __put_super(p); | |
770 | count -= pruned; | |
771 | p = sb; | |
772 | /* more work left to do? */ | |
773 | if (count <= 0) | |
774 | break; | |
775 | } | |
776 | if (p) | |
777 | __put_super(p); | |
778 | spin_unlock(&sb_lock); | |
88e0fbc4 | 779 | up_read(&iprune_sem); |
1da177e4 LT |
780 | } |
781 | ||
782 | /* | |
783 | * shrink_icache_memory() will attempt to reclaim some unused inodes. Here, | |
784 | * "unused" means that no dentries are referring to the inodes: the files are | |
785 | * not open and the dcache references to those inodes have already been | |
786 | * reclaimed. | |
787 | * | |
788 | * This function is passed the number of inodes to scan, and it returns the | |
789 | * total number of remaining possibly-reclaimable inodes. | |
790 | */ | |
1495f230 YH |
791 | static int shrink_icache_memory(struct shrinker *shrink, |
792 | struct shrink_control *sc) | |
1da177e4 | 793 | { |
1495f230 YH |
794 | int nr = sc->nr_to_scan; |
795 | gfp_t gfp_mask = sc->gfp_mask; | |
796 | ||
1da177e4 LT |
797 | if (nr) { |
798 | /* | |
799 | * Nasty deadlock avoidance. We may hold various FS locks, | |
800 | * and we don't want to recurse into the FS that called us | |
801 | * in clear_inode() and friends.. | |
6b3304b5 | 802 | */ |
1da177e4 LT |
803 | if (!(gfp_mask & __GFP_FS)) |
804 | return -1; | |
805 | prune_icache(nr); | |
806 | } | |
cffbc8aa | 807 | return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure; |
1da177e4 LT |
808 | } |
809 | ||
8e1f936b RR |
810 | static struct shrinker icache_shrinker = { |
811 | .shrink = shrink_icache_memory, | |
812 | .seeks = DEFAULT_SEEKS, | |
813 | }; | |
814 | ||
1da177e4 LT |
815 | static void __wait_on_freeing_inode(struct inode *inode); |
816 | /* | |
817 | * Called with the inode lock held. | |
1da177e4 | 818 | */ |
6b3304b5 MK |
819 | static struct inode *find_inode(struct super_block *sb, |
820 | struct hlist_head *head, | |
821 | int (*test)(struct inode *, void *), | |
822 | void *data) | |
1da177e4 LT |
823 | { |
824 | struct hlist_node *node; | |
6b3304b5 | 825 | struct inode *inode = NULL; |
1da177e4 LT |
826 | |
827 | repeat: | |
c5c8be3c | 828 | hlist_for_each_entry(inode, node, head, i_hash) { |
67a23c49 DC |
829 | spin_lock(&inode->i_lock); |
830 | if (inode->i_sb != sb) { | |
831 | spin_unlock(&inode->i_lock); | |
1da177e4 | 832 | continue; |
67a23c49 DC |
833 | } |
834 | if (!test(inode, data)) { | |
835 | spin_unlock(&inode->i_lock); | |
1da177e4 | 836 | continue; |
67a23c49 | 837 | } |
a4ffdde6 | 838 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
839 | __wait_on_freeing_inode(inode); |
840 | goto repeat; | |
841 | } | |
f7899bd5 | 842 | __iget(inode); |
250df6ed | 843 | spin_unlock(&inode->i_lock); |
f7899bd5 | 844 | return inode; |
1da177e4 | 845 | } |
f7899bd5 | 846 | return NULL; |
1da177e4 LT |
847 | } |
848 | ||
849 | /* | |
850 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
851 | * iget_locked for details. | |
852 | */ | |
6b3304b5 MK |
853 | static struct inode *find_inode_fast(struct super_block *sb, |
854 | struct hlist_head *head, unsigned long ino) | |
1da177e4 LT |
855 | { |
856 | struct hlist_node *node; | |
6b3304b5 | 857 | struct inode *inode = NULL; |
1da177e4 LT |
858 | |
859 | repeat: | |
c5c8be3c | 860 | hlist_for_each_entry(inode, node, head, i_hash) { |
67a23c49 DC |
861 | spin_lock(&inode->i_lock); |
862 | if (inode->i_ino != ino) { | |
863 | spin_unlock(&inode->i_lock); | |
1da177e4 | 864 | continue; |
67a23c49 DC |
865 | } |
866 | if (inode->i_sb != sb) { | |
867 | spin_unlock(&inode->i_lock); | |
1da177e4 | 868 | continue; |
67a23c49 | 869 | } |
a4ffdde6 | 870 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
871 | __wait_on_freeing_inode(inode); |
872 | goto repeat; | |
873 | } | |
f7899bd5 | 874 | __iget(inode); |
250df6ed | 875 | spin_unlock(&inode->i_lock); |
f7899bd5 | 876 | return inode; |
1da177e4 | 877 | } |
f7899bd5 | 878 | return NULL; |
8290c35f DC |
879 | } |
880 | ||
f991bd2e ED |
881 | /* |
882 | * Each cpu owns a range of LAST_INO_BATCH numbers. | |
883 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, | |
884 | * to renew the exhausted range. | |
8290c35f | 885 | * |
f991bd2e ED |
886 | * This does not significantly increase overflow rate because every CPU can |
887 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is | |
888 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the | |
889 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase | |
890 | * overflow rate by 2x, which does not seem too significant. | |
891 | * | |
892 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
893 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
894 | * here to attempt to avoid that. | |
8290c35f | 895 | */ |
f991bd2e ED |
896 | #define LAST_INO_BATCH 1024 |
897 | static DEFINE_PER_CPU(unsigned int, last_ino); | |
898 | ||
85fe4025 | 899 | unsigned int get_next_ino(void) |
8290c35f | 900 | { |
f991bd2e ED |
901 | unsigned int *p = &get_cpu_var(last_ino); |
902 | unsigned int res = *p; | |
8290c35f | 903 | |
f991bd2e ED |
904 | #ifdef CONFIG_SMP |
905 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { | |
906 | static atomic_t shared_last_ino; | |
907 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); | |
908 | ||
909 | res = next - LAST_INO_BATCH; | |
910 | } | |
911 | #endif | |
912 | ||
913 | *p = ++res; | |
914 | put_cpu_var(last_ino); | |
915 | return res; | |
8290c35f | 916 | } |
85fe4025 | 917 | EXPORT_SYMBOL(get_next_ino); |
8290c35f | 918 | |
1da177e4 LT |
919 | /** |
920 | * new_inode - obtain an inode | |
921 | * @sb: superblock | |
922 | * | |
769848c0 | 923 | * Allocates a new inode for given superblock. The default gfp_mask |
3c1d4378 | 924 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
769848c0 MG |
925 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
926 | * for the page cache are not reclaimable or migratable, | |
927 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
928 | * newly created inode's mapping | |
929 | * | |
1da177e4 LT |
930 | */ |
931 | struct inode *new_inode(struct super_block *sb) | |
932 | { | |
6b3304b5 | 933 | struct inode *inode; |
1da177e4 | 934 | |
55fa6091 | 935 | spin_lock_prefetch(&inode_sb_list_lock); |
6b3304b5 | 936 | |
1da177e4 LT |
937 | inode = alloc_inode(sb); |
938 | if (inode) { | |
250df6ed | 939 | spin_lock(&inode->i_lock); |
1da177e4 | 940 | inode->i_state = 0; |
250df6ed | 941 | spin_unlock(&inode->i_lock); |
55fa6091 | 942 | inode_sb_list_add(inode); |
1da177e4 LT |
943 | } |
944 | return inode; | |
945 | } | |
1da177e4 LT |
946 | EXPORT_SYMBOL(new_inode); |
947 | ||
250df6ed DC |
948 | /** |
949 | * unlock_new_inode - clear the I_NEW state and wake up any waiters | |
950 | * @inode: new inode to unlock | |
951 | * | |
952 | * Called when the inode is fully initialised to clear the new state of the | |
953 | * inode and wake up anyone waiting for the inode to finish initialisation. | |
954 | */ | |
1da177e4 LT |
955 | void unlock_new_inode(struct inode *inode) |
956 | { | |
14358e6d | 957 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
a3314a0e | 958 | if (S_ISDIR(inode->i_mode)) { |
1e89a5e1 PZ |
959 | struct file_system_type *type = inode->i_sb->s_type; |
960 | ||
9a7aa12f JK |
961 | /* Set new key only if filesystem hasn't already changed it */ |
962 | if (!lockdep_match_class(&inode->i_mutex, | |
963 | &type->i_mutex_key)) { | |
964 | /* | |
965 | * ensure nobody is actually holding i_mutex | |
966 | */ | |
967 | mutex_destroy(&inode->i_mutex); | |
968 | mutex_init(&inode->i_mutex); | |
969 | lockdep_set_class(&inode->i_mutex, | |
970 | &type->i_mutex_dir_key); | |
971 | } | |
1e89a5e1 | 972 | } |
14358e6d | 973 | #endif |
250df6ed | 974 | spin_lock(&inode->i_lock); |
eaff8079 CH |
975 | WARN_ON(!(inode->i_state & I_NEW)); |
976 | inode->i_state &= ~I_NEW; | |
250df6ed DC |
977 | wake_up_bit(&inode->i_state, __I_NEW); |
978 | spin_unlock(&inode->i_lock); | |
1da177e4 | 979 | } |
1da177e4 LT |
980 | EXPORT_SYMBOL(unlock_new_inode); |
981 | ||
0b2d0724 CH |
982 | /** |
983 | * iget5_locked - obtain an inode from a mounted file system | |
984 | * @sb: super block of file system | |
985 | * @hashval: hash value (usually inode number) to get | |
986 | * @test: callback used for comparisons between inodes | |
987 | * @set: callback used to initialize a new struct inode | |
988 | * @data: opaque data pointer to pass to @test and @set | |
989 | * | |
990 | * Search for the inode specified by @hashval and @data in the inode cache, | |
991 | * and if present it is return it with an increased reference count. This is | |
992 | * a generalized version of iget_locked() for file systems where the inode | |
993 | * number is not sufficient for unique identification of an inode. | |
994 | * | |
995 | * If the inode is not in cache, allocate a new inode and return it locked, | |
996 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
997 | * before unlocking it via unlock_new_inode(). | |
1da177e4 | 998 | * |
0b2d0724 CH |
999 | * Note both @test and @set are called with the inode_hash_lock held, so can't |
1000 | * sleep. | |
1da177e4 | 1001 | */ |
0b2d0724 CH |
1002 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, |
1003 | int (*test)(struct inode *, void *), | |
1004 | int (*set)(struct inode *, void *), void *data) | |
1da177e4 | 1005 | { |
0b2d0724 | 1006 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
6b3304b5 | 1007 | struct inode *inode; |
1da177e4 | 1008 | |
0b2d0724 CH |
1009 | spin_lock(&inode_hash_lock); |
1010 | inode = find_inode(sb, head, test, data); | |
1011 | spin_unlock(&inode_hash_lock); | |
1012 | ||
1013 | if (inode) { | |
1014 | wait_on_inode(inode); | |
1015 | return inode; | |
1016 | } | |
1017 | ||
1da177e4 LT |
1018 | inode = alloc_inode(sb); |
1019 | if (inode) { | |
6b3304b5 | 1020 | struct inode *old; |
1da177e4 | 1021 | |
67a23c49 | 1022 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1023 | /* We released the lock, so.. */ |
1024 | old = find_inode(sb, head, test, data); | |
1025 | if (!old) { | |
1026 | if (set(inode, data)) | |
1027 | goto set_failed; | |
1028 | ||
250df6ed DC |
1029 | spin_lock(&inode->i_lock); |
1030 | inode->i_state = I_NEW; | |
646ec461 | 1031 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1032 | spin_unlock(&inode->i_lock); |
55fa6091 | 1033 | inode_sb_list_add(inode); |
67a23c49 | 1034 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1035 | |
1036 | /* Return the locked inode with I_NEW set, the | |
1037 | * caller is responsible for filling in the contents | |
1038 | */ | |
1039 | return inode; | |
1040 | } | |
1041 | ||
1042 | /* | |
1043 | * Uhhuh, somebody else created the same inode under | |
1044 | * us. Use the old inode instead of the one we just | |
1045 | * allocated. | |
1046 | */ | |
67a23c49 | 1047 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1048 | destroy_inode(inode); |
1049 | inode = old; | |
1050 | wait_on_inode(inode); | |
1051 | } | |
1052 | return inode; | |
1053 | ||
1054 | set_failed: | |
67a23c49 | 1055 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1056 | destroy_inode(inode); |
1057 | return NULL; | |
1058 | } | |
0b2d0724 | 1059 | EXPORT_SYMBOL(iget5_locked); |
1da177e4 | 1060 | |
0b2d0724 CH |
1061 | /** |
1062 | * iget_locked - obtain an inode from a mounted file system | |
1063 | * @sb: super block of file system | |
1064 | * @ino: inode number to get | |
1065 | * | |
1066 | * Search for the inode specified by @ino in the inode cache and if present | |
1067 | * return it with an increased reference count. This is for file systems | |
1068 | * where the inode number is sufficient for unique identification of an inode. | |
1069 | * | |
1070 | * If the inode is not in cache, allocate a new inode and return it locked, | |
1071 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
1072 | * before unlocking it via unlock_new_inode(). | |
1da177e4 | 1073 | */ |
0b2d0724 | 1074 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) |
1da177e4 | 1075 | { |
0b2d0724 | 1076 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
6b3304b5 | 1077 | struct inode *inode; |
1da177e4 | 1078 | |
0b2d0724 CH |
1079 | spin_lock(&inode_hash_lock); |
1080 | inode = find_inode_fast(sb, head, ino); | |
1081 | spin_unlock(&inode_hash_lock); | |
1082 | if (inode) { | |
1083 | wait_on_inode(inode); | |
1084 | return inode; | |
1085 | } | |
1086 | ||
1da177e4 LT |
1087 | inode = alloc_inode(sb); |
1088 | if (inode) { | |
6b3304b5 | 1089 | struct inode *old; |
1da177e4 | 1090 | |
67a23c49 | 1091 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1092 | /* We released the lock, so.. */ |
1093 | old = find_inode_fast(sb, head, ino); | |
1094 | if (!old) { | |
1095 | inode->i_ino = ino; | |
250df6ed DC |
1096 | spin_lock(&inode->i_lock); |
1097 | inode->i_state = I_NEW; | |
646ec461 | 1098 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1099 | spin_unlock(&inode->i_lock); |
55fa6091 | 1100 | inode_sb_list_add(inode); |
67a23c49 | 1101 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1102 | |
1103 | /* Return the locked inode with I_NEW set, the | |
1104 | * caller is responsible for filling in the contents | |
1105 | */ | |
1106 | return inode; | |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * Uhhuh, somebody else created the same inode under | |
1111 | * us. Use the old inode instead of the one we just | |
1112 | * allocated. | |
1113 | */ | |
67a23c49 | 1114 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1115 | destroy_inode(inode); |
1116 | inode = old; | |
1117 | wait_on_inode(inode); | |
1118 | } | |
1119 | return inode; | |
1120 | } | |
0b2d0724 | 1121 | EXPORT_SYMBOL(iget_locked); |
1da177e4 | 1122 | |
ad5e195a CH |
1123 | /* |
1124 | * search the inode cache for a matching inode number. | |
1125 | * If we find one, then the inode number we are trying to | |
1126 | * allocate is not unique and so we should not use it. | |
1127 | * | |
1128 | * Returns 1 if the inode number is unique, 0 if it is not. | |
1129 | */ | |
1130 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) | |
1131 | { | |
1132 | struct hlist_head *b = inode_hashtable + hash(sb, ino); | |
1133 | struct hlist_node *node; | |
1134 | struct inode *inode; | |
1135 | ||
67a23c49 | 1136 | spin_lock(&inode_hash_lock); |
ad5e195a | 1137 | hlist_for_each_entry(inode, node, b, i_hash) { |
67a23c49 DC |
1138 | if (inode->i_ino == ino && inode->i_sb == sb) { |
1139 | spin_unlock(&inode_hash_lock); | |
ad5e195a | 1140 | return 0; |
67a23c49 | 1141 | } |
ad5e195a | 1142 | } |
67a23c49 | 1143 | spin_unlock(&inode_hash_lock); |
ad5e195a CH |
1144 | |
1145 | return 1; | |
1146 | } | |
1147 | ||
1da177e4 LT |
1148 | /** |
1149 | * iunique - get a unique inode number | |
1150 | * @sb: superblock | |
1151 | * @max_reserved: highest reserved inode number | |
1152 | * | |
1153 | * Obtain an inode number that is unique on the system for a given | |
1154 | * superblock. This is used by file systems that have no natural | |
1155 | * permanent inode numbering system. An inode number is returned that | |
1156 | * is higher than the reserved limit but unique. | |
1157 | * | |
1158 | * BUGS: | |
1159 | * With a large number of inodes live on the file system this function | |
1160 | * currently becomes quite slow. | |
1161 | */ | |
1162 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
1163 | { | |
866b04fc JL |
1164 | /* |
1165 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
1166 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
1167 | * here to attempt to avoid that. | |
1168 | */ | |
ad5e195a | 1169 | static DEFINE_SPINLOCK(iunique_lock); |
866b04fc | 1170 | static unsigned int counter; |
1da177e4 | 1171 | ino_t res; |
3361c7be | 1172 | |
ad5e195a | 1173 | spin_lock(&iunique_lock); |
3361c7be JL |
1174 | do { |
1175 | if (counter <= max_reserved) | |
1176 | counter = max_reserved + 1; | |
1da177e4 | 1177 | res = counter++; |
ad5e195a CH |
1178 | } while (!test_inode_iunique(sb, res)); |
1179 | spin_unlock(&iunique_lock); | |
1da177e4 | 1180 | |
3361c7be JL |
1181 | return res; |
1182 | } | |
1da177e4 LT |
1183 | EXPORT_SYMBOL(iunique); |
1184 | ||
1185 | struct inode *igrab(struct inode *inode) | |
1186 | { | |
250df6ed DC |
1187 | spin_lock(&inode->i_lock); |
1188 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { | |
1da177e4 | 1189 | __iget(inode); |
250df6ed DC |
1190 | spin_unlock(&inode->i_lock); |
1191 | } else { | |
1192 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1193 | /* |
1194 | * Handle the case where s_op->clear_inode is not been | |
1195 | * called yet, and somebody is calling igrab | |
1196 | * while the inode is getting freed. | |
1197 | */ | |
1198 | inode = NULL; | |
250df6ed | 1199 | } |
1da177e4 LT |
1200 | return inode; |
1201 | } | |
1da177e4 LT |
1202 | EXPORT_SYMBOL(igrab); |
1203 | ||
1204 | /** | |
0b2d0724 | 1205 | * ilookup5_nowait - search for an inode in the inode cache |
1da177e4 | 1206 | * @sb: super block of file system to search |
0b2d0724 | 1207 | * @hashval: hash value (usually inode number) to search for |
1da177e4 LT |
1208 | * @test: callback used for comparisons between inodes |
1209 | * @data: opaque data pointer to pass to @test | |
1da177e4 | 1210 | * |
0b2d0724 | 1211 | * Search for the inode specified by @hashval and @data in the inode cache. |
1da177e4 LT |
1212 | * If the inode is in the cache, the inode is returned with an incremented |
1213 | * reference count. | |
1214 | * | |
0b2d0724 CH |
1215 | * Note: I_NEW is not waited upon so you have to be very careful what you do |
1216 | * with the returned inode. You probably should be using ilookup5() instead. | |
1da177e4 | 1217 | * |
b6d0ad68 | 1218 | * Note2: @test is called with the inode_hash_lock held, so can't sleep. |
1da177e4 | 1219 | */ |
0b2d0724 CH |
1220 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, |
1221 | int (*test)(struct inode *, void *), void *data) | |
1da177e4 | 1222 | { |
0b2d0724 | 1223 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
1da177e4 LT |
1224 | struct inode *inode; |
1225 | ||
67a23c49 | 1226 | spin_lock(&inode_hash_lock); |
1da177e4 | 1227 | inode = find_inode(sb, head, test, data); |
67a23c49 | 1228 | spin_unlock(&inode_hash_lock); |
88bd5121 | 1229 | |
0b2d0724 | 1230 | return inode; |
88bd5121 | 1231 | } |
88bd5121 AA |
1232 | EXPORT_SYMBOL(ilookup5_nowait); |
1233 | ||
1234 | /** | |
1235 | * ilookup5 - search for an inode in the inode cache | |
1236 | * @sb: super block of file system to search | |
1237 | * @hashval: hash value (usually inode number) to search for | |
1238 | * @test: callback used for comparisons between inodes | |
1239 | * @data: opaque data pointer to pass to @test | |
1240 | * | |
0b2d0724 CH |
1241 | * Search for the inode specified by @hashval and @data in the inode cache, |
1242 | * and if the inode is in the cache, return the inode with an incremented | |
1243 | * reference count. Waits on I_NEW before returning the inode. | |
88bd5121 | 1244 | * returned with an incremented reference count. |
1da177e4 | 1245 | * |
0b2d0724 CH |
1246 | * This is a generalized version of ilookup() for file systems where the |
1247 | * inode number is not sufficient for unique identification of an inode. | |
1da177e4 | 1248 | * |
0b2d0724 | 1249 | * Note: @test is called with the inode_hash_lock held, so can't sleep. |
1da177e4 LT |
1250 | */ |
1251 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
1252 | int (*test)(struct inode *, void *), void *data) | |
1253 | { | |
0b2d0724 | 1254 | struct inode *inode = ilookup5_nowait(sb, hashval, test, data); |
1da177e4 | 1255 | |
0b2d0724 CH |
1256 | if (inode) |
1257 | wait_on_inode(inode); | |
1258 | return inode; | |
1da177e4 | 1259 | } |
1da177e4 LT |
1260 | EXPORT_SYMBOL(ilookup5); |
1261 | ||
1262 | /** | |
1263 | * ilookup - search for an inode in the inode cache | |
1264 | * @sb: super block of file system to search | |
1265 | * @ino: inode number to search for | |
1266 | * | |
0b2d0724 CH |
1267 | * Search for the inode @ino in the inode cache, and if the inode is in the |
1268 | * cache, the inode is returned with an incremented reference count. | |
1da177e4 LT |
1269 | */ |
1270 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1271 | { | |
1272 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1da177e4 LT |
1273 | struct inode *inode; |
1274 | ||
0b2d0724 CH |
1275 | spin_lock(&inode_hash_lock); |
1276 | inode = find_inode_fast(sb, head, ino); | |
1277 | spin_unlock(&inode_hash_lock); | |
1da177e4 | 1278 | |
1da177e4 | 1279 | if (inode) |
0b2d0724 CH |
1280 | wait_on_inode(inode); |
1281 | return inode; | |
1da177e4 | 1282 | } |
0b2d0724 | 1283 | EXPORT_SYMBOL(ilookup); |
1da177e4 | 1284 | |
261bca86 AV |
1285 | int insert_inode_locked(struct inode *inode) |
1286 | { | |
1287 | struct super_block *sb = inode->i_sb; | |
1288 | ino_t ino = inode->i_ino; | |
1289 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
261bca86 | 1290 | |
261bca86 | 1291 | while (1) { |
72a43d63 AV |
1292 | struct hlist_node *node; |
1293 | struct inode *old = NULL; | |
67a23c49 | 1294 | spin_lock(&inode_hash_lock); |
72a43d63 AV |
1295 | hlist_for_each_entry(old, node, head, i_hash) { |
1296 | if (old->i_ino != ino) | |
1297 | continue; | |
1298 | if (old->i_sb != sb) | |
1299 | continue; | |
250df6ed DC |
1300 | spin_lock(&old->i_lock); |
1301 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1302 | spin_unlock(&old->i_lock); | |
72a43d63 | 1303 | continue; |
250df6ed | 1304 | } |
72a43d63 AV |
1305 | break; |
1306 | } | |
1307 | if (likely(!node)) { | |
250df6ed DC |
1308 | spin_lock(&inode->i_lock); |
1309 | inode->i_state |= I_NEW; | |
261bca86 | 1310 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1311 | spin_unlock(&inode->i_lock); |
67a23c49 | 1312 | spin_unlock(&inode_hash_lock); |
261bca86 AV |
1313 | return 0; |
1314 | } | |
1315 | __iget(old); | |
250df6ed | 1316 | spin_unlock(&old->i_lock); |
67a23c49 | 1317 | spin_unlock(&inode_hash_lock); |
261bca86 | 1318 | wait_on_inode(old); |
1d3382cb | 1319 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1320 | iput(old); |
1321 | return -EBUSY; | |
1322 | } | |
1323 | iput(old); | |
1324 | } | |
1325 | } | |
261bca86 AV |
1326 | EXPORT_SYMBOL(insert_inode_locked); |
1327 | ||
1328 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1329 | int (*test)(struct inode *, void *), void *data) | |
1330 | { | |
1331 | struct super_block *sb = inode->i_sb; | |
1332 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
261bca86 | 1333 | |
261bca86 | 1334 | while (1) { |
72a43d63 AV |
1335 | struct hlist_node *node; |
1336 | struct inode *old = NULL; | |
1337 | ||
67a23c49 | 1338 | spin_lock(&inode_hash_lock); |
72a43d63 AV |
1339 | hlist_for_each_entry(old, node, head, i_hash) { |
1340 | if (old->i_sb != sb) | |
1341 | continue; | |
1342 | if (!test(old, data)) | |
1343 | continue; | |
250df6ed DC |
1344 | spin_lock(&old->i_lock); |
1345 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1346 | spin_unlock(&old->i_lock); | |
72a43d63 | 1347 | continue; |
250df6ed | 1348 | } |
72a43d63 AV |
1349 | break; |
1350 | } | |
1351 | if (likely(!node)) { | |
250df6ed DC |
1352 | spin_lock(&inode->i_lock); |
1353 | inode->i_state |= I_NEW; | |
261bca86 | 1354 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1355 | spin_unlock(&inode->i_lock); |
67a23c49 | 1356 | spin_unlock(&inode_hash_lock); |
261bca86 AV |
1357 | return 0; |
1358 | } | |
1359 | __iget(old); | |
250df6ed | 1360 | spin_unlock(&old->i_lock); |
67a23c49 | 1361 | spin_unlock(&inode_hash_lock); |
261bca86 | 1362 | wait_on_inode(old); |
1d3382cb | 1363 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1364 | iput(old); |
1365 | return -EBUSY; | |
1366 | } | |
1367 | iput(old); | |
1368 | } | |
1369 | } | |
261bca86 AV |
1370 | EXPORT_SYMBOL(insert_inode_locked4); |
1371 | ||
1da177e4 | 1372 | |
45321ac5 AV |
1373 | int generic_delete_inode(struct inode *inode) |
1374 | { | |
1375 | return 1; | |
1376 | } | |
1377 | EXPORT_SYMBOL(generic_delete_inode); | |
1378 | ||
1da177e4 | 1379 | /* |
45321ac5 AV |
1380 | * Normal UNIX filesystem behaviour: delete the |
1381 | * inode when the usage count drops to zero, and | |
1382 | * i_nlink is zero. | |
1da177e4 | 1383 | */ |
45321ac5 | 1384 | int generic_drop_inode(struct inode *inode) |
1da177e4 | 1385 | { |
1d3382cb | 1386 | return !inode->i_nlink || inode_unhashed(inode); |
1da177e4 | 1387 | } |
45321ac5 | 1388 | EXPORT_SYMBOL_GPL(generic_drop_inode); |
1da177e4 | 1389 | |
45321ac5 AV |
1390 | /* |
1391 | * Called when we're dropping the last reference | |
1392 | * to an inode. | |
22fe4042 | 1393 | * |
45321ac5 AV |
1394 | * Call the FS "drop_inode()" function, defaulting to |
1395 | * the legacy UNIX filesystem behaviour. If it tells | |
1396 | * us to evict inode, do so. Otherwise, retain inode | |
1397 | * in cache if fs is alive, sync and evict if fs is | |
1398 | * shutting down. | |
22fe4042 | 1399 | */ |
45321ac5 | 1400 | static void iput_final(struct inode *inode) |
1da177e4 LT |
1401 | { |
1402 | struct super_block *sb = inode->i_sb; | |
45321ac5 AV |
1403 | const struct super_operations *op = inode->i_sb->s_op; |
1404 | int drop; | |
1405 | ||
250df6ed DC |
1406 | WARN_ON(inode->i_state & I_NEW); |
1407 | ||
e7f59097 | 1408 | if (op->drop_inode) |
45321ac5 AV |
1409 | drop = op->drop_inode(inode); |
1410 | else | |
1411 | drop = generic_drop_inode(inode); | |
1da177e4 | 1412 | |
b2b2af8e DC |
1413 | if (!drop && (sb->s_flags & MS_ACTIVE)) { |
1414 | inode->i_state |= I_REFERENCED; | |
1415 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) | |
1416 | inode_lru_list_add(inode); | |
1417 | spin_unlock(&inode->i_lock); | |
b2b2af8e DC |
1418 | return; |
1419 | } | |
1420 | ||
45321ac5 | 1421 | if (!drop) { |
991114c6 | 1422 | inode->i_state |= I_WILL_FREE; |
250df6ed | 1423 | spin_unlock(&inode->i_lock); |
1da177e4 | 1424 | write_inode_now(inode, 1); |
250df6ed | 1425 | spin_lock(&inode->i_lock); |
7ef0d737 | 1426 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1427 | inode->i_state &= ~I_WILL_FREE; |
1da177e4 | 1428 | } |
7ccf19a8 | 1429 | |
991114c6 | 1430 | inode->i_state |= I_FREEING; |
9e38d86f | 1431 | inode_lru_list_del(inode); |
b2b2af8e | 1432 | spin_unlock(&inode->i_lock); |
b2b2af8e | 1433 | |
644da596 | 1434 | evict(inode); |
1da177e4 LT |
1435 | } |
1436 | ||
1da177e4 | 1437 | /** |
6b3304b5 | 1438 | * iput - put an inode |
1da177e4 LT |
1439 | * @inode: inode to put |
1440 | * | |
1441 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1442 | * zero, the inode is then freed and may also be destroyed. | |
1443 | * | |
1444 | * Consequently, iput() can sleep. | |
1445 | */ | |
1446 | void iput(struct inode *inode) | |
1447 | { | |
1448 | if (inode) { | |
a4ffdde6 | 1449 | BUG_ON(inode->i_state & I_CLEAR); |
1da177e4 | 1450 | |
f283c86a | 1451 | if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) |
1da177e4 LT |
1452 | iput_final(inode); |
1453 | } | |
1454 | } | |
1da177e4 LT |
1455 | EXPORT_SYMBOL(iput); |
1456 | ||
1457 | /** | |
1458 | * bmap - find a block number in a file | |
1459 | * @inode: inode of file | |
1460 | * @block: block to find | |
1461 | * | |
1462 | * Returns the block number on the device holding the inode that | |
1463 | * is the disk block number for the block of the file requested. | |
1464 | * That is, asked for block 4 of inode 1 the function will return the | |
6b3304b5 | 1465 | * disk block relative to the disk start that holds that block of the |
1da177e4 LT |
1466 | * file. |
1467 | */ | |
6b3304b5 | 1468 | sector_t bmap(struct inode *inode, sector_t block) |
1da177e4 LT |
1469 | { |
1470 | sector_t res = 0; | |
1471 | if (inode->i_mapping->a_ops->bmap) | |
1472 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1473 | return res; | |
1474 | } | |
1da177e4 LT |
1475 | EXPORT_SYMBOL(bmap); |
1476 | ||
11ff6f05 MG |
1477 | /* |
1478 | * With relative atime, only update atime if the previous atime is | |
1479 | * earlier than either the ctime or mtime or if at least a day has | |
1480 | * passed since the last atime update. | |
1481 | */ | |
1482 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1483 | struct timespec now) | |
1484 | { | |
1485 | ||
1486 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1487 | return 1; | |
1488 | /* | |
1489 | * Is mtime younger than atime? If yes, update atime: | |
1490 | */ | |
1491 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1492 | return 1; | |
1493 | /* | |
1494 | * Is ctime younger than atime? If yes, update atime: | |
1495 | */ | |
1496 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1497 | return 1; | |
1498 | ||
1499 | /* | |
1500 | * Is the previous atime value older than a day? If yes, | |
1501 | * update atime: | |
1502 | */ | |
1503 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1504 | return 1; | |
1505 | /* | |
1506 | * Good, we can skip the atime update: | |
1507 | */ | |
1508 | return 0; | |
1509 | } | |
1510 | ||
1da177e4 | 1511 | /** |
869243a0 CH |
1512 | * touch_atime - update the access time |
1513 | * @mnt: mount the inode is accessed on | |
7045f37b | 1514 | * @dentry: dentry accessed |
1da177e4 LT |
1515 | * |
1516 | * Update the accessed time on an inode and mark it for writeback. | |
1517 | * This function automatically handles read only file systems and media, | |
1518 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1519 | */ | |
869243a0 | 1520 | void touch_atime(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 1521 | { |
869243a0 | 1522 | struct inode *inode = dentry->d_inode; |
1da177e4 LT |
1523 | struct timespec now; |
1524 | ||
cdb70f3f | 1525 | if (inode->i_flags & S_NOATIME) |
b12536c2 | 1526 | return; |
37756ced | 1527 | if (IS_NOATIME(inode)) |
b12536c2 | 1528 | return; |
b2276138 | 1529 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1530 | return; |
47ae32d6 | 1531 | |
cdb70f3f | 1532 | if (mnt->mnt_flags & MNT_NOATIME) |
b12536c2 | 1533 | return; |
cdb70f3f | 1534 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1535 | return; |
1da177e4 LT |
1536 | |
1537 | now = current_fs_time(inode->i_sb); | |
11ff6f05 MG |
1538 | |
1539 | if (!relatime_need_update(mnt, inode, now)) | |
b12536c2 | 1540 | return; |
11ff6f05 | 1541 | |
47ae32d6 | 1542 | if (timespec_equal(&inode->i_atime, &now)) |
b12536c2 AK |
1543 | return; |
1544 | ||
1545 | if (mnt_want_write(mnt)) | |
1546 | return; | |
47ae32d6 VH |
1547 | |
1548 | inode->i_atime = now; | |
1549 | mark_inode_dirty_sync(inode); | |
cdb70f3f | 1550 | mnt_drop_write(mnt); |
1da177e4 | 1551 | } |
869243a0 | 1552 | EXPORT_SYMBOL(touch_atime); |
1da177e4 LT |
1553 | |
1554 | /** | |
870f4817 CH |
1555 | * file_update_time - update mtime and ctime time |
1556 | * @file: file accessed | |
1da177e4 | 1557 | * |
870f4817 CH |
1558 | * Update the mtime and ctime members of an inode and mark the inode |
1559 | * for writeback. Note that this function is meant exclusively for | |
1560 | * usage in the file write path of filesystems, and filesystems may | |
1561 | * choose to explicitly ignore update via this function with the | |
2eadfc0e | 1562 | * S_NOCMTIME inode flag, e.g. for network filesystem where these |
870f4817 | 1563 | * timestamps are handled by the server. |
1da177e4 LT |
1564 | */ |
1565 | ||
870f4817 | 1566 | void file_update_time(struct file *file) |
1da177e4 | 1567 | { |
0f7fc9e4 | 1568 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 | 1569 | struct timespec now; |
ce06e0b2 | 1570 | enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0; |
1da177e4 | 1571 | |
ce06e0b2 | 1572 | /* First try to exhaust all avenues to not sync */ |
1da177e4 LT |
1573 | if (IS_NOCMTIME(inode)) |
1574 | return; | |
20ddee2c | 1575 | |
1da177e4 | 1576 | now = current_fs_time(inode->i_sb); |
ce06e0b2 AK |
1577 | if (!timespec_equal(&inode->i_mtime, &now)) |
1578 | sync_it = S_MTIME; | |
1da177e4 | 1579 | |
ce06e0b2 AK |
1580 | if (!timespec_equal(&inode->i_ctime, &now)) |
1581 | sync_it |= S_CTIME; | |
870f4817 | 1582 | |
ce06e0b2 AK |
1583 | if (IS_I_VERSION(inode)) |
1584 | sync_it |= S_VERSION; | |
7a224228 | 1585 | |
ce06e0b2 AK |
1586 | if (!sync_it) |
1587 | return; | |
1588 | ||
1589 | /* Finally allowed to write? Takes lock. */ | |
1590 | if (mnt_want_write_file(file)) | |
1591 | return; | |
1592 | ||
1593 | /* Only change inode inside the lock region */ | |
1594 | if (sync_it & S_VERSION) | |
1595 | inode_inc_iversion(inode); | |
1596 | if (sync_it & S_CTIME) | |
1597 | inode->i_ctime = now; | |
1598 | if (sync_it & S_MTIME) | |
1599 | inode->i_mtime = now; | |
1600 | mark_inode_dirty_sync(inode); | |
20ddee2c | 1601 | mnt_drop_write(file->f_path.mnt); |
1da177e4 | 1602 | } |
870f4817 | 1603 | EXPORT_SYMBOL(file_update_time); |
1da177e4 LT |
1604 | |
1605 | int inode_needs_sync(struct inode *inode) | |
1606 | { | |
1607 | if (IS_SYNC(inode)) | |
1608 | return 1; | |
1609 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1610 | return 1; | |
1611 | return 0; | |
1612 | } | |
1da177e4 LT |
1613 | EXPORT_SYMBOL(inode_needs_sync); |
1614 | ||
1da177e4 LT |
1615 | int inode_wait(void *word) |
1616 | { | |
1617 | schedule(); | |
1618 | return 0; | |
1619 | } | |
d44dab8d | 1620 | EXPORT_SYMBOL(inode_wait); |
1da177e4 LT |
1621 | |
1622 | /* | |
168a9fd6 MS |
1623 | * If we try to find an inode in the inode hash while it is being |
1624 | * deleted, we have to wait until the filesystem completes its | |
1625 | * deletion before reporting that it isn't found. This function waits | |
1626 | * until the deletion _might_ have completed. Callers are responsible | |
1627 | * to recheck inode state. | |
1628 | * | |
eaff8079 | 1629 | * It doesn't matter if I_NEW is not set initially, a call to |
250df6ed DC |
1630 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list |
1631 | * will DTRT. | |
1da177e4 LT |
1632 | */ |
1633 | static void __wait_on_freeing_inode(struct inode *inode) | |
1634 | { | |
1635 | wait_queue_head_t *wq; | |
eaff8079 CH |
1636 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); |
1637 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1da177e4 | 1638 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
250df6ed | 1639 | spin_unlock(&inode->i_lock); |
67a23c49 | 1640 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1641 | schedule(); |
1642 | finish_wait(wq, &wait.wait); | |
67a23c49 | 1643 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1644 | } |
1645 | ||
1da177e4 LT |
1646 | static __initdata unsigned long ihash_entries; |
1647 | static int __init set_ihash_entries(char *str) | |
1648 | { | |
1649 | if (!str) | |
1650 | return 0; | |
1651 | ihash_entries = simple_strtoul(str, &str, 0); | |
1652 | return 1; | |
1653 | } | |
1654 | __setup("ihash_entries=", set_ihash_entries); | |
1655 | ||
1656 | /* | |
1657 | * Initialize the waitqueues and inode hash table. | |
1658 | */ | |
1659 | void __init inode_init_early(void) | |
1660 | { | |
1661 | int loop; | |
1662 | ||
1663 | /* If hashes are distributed across NUMA nodes, defer | |
1664 | * hash allocation until vmalloc space is available. | |
1665 | */ | |
1666 | if (hashdist) | |
1667 | return; | |
1668 | ||
1669 | inode_hashtable = | |
1670 | alloc_large_system_hash("Inode-cache", | |
1671 | sizeof(struct hlist_head), | |
1672 | ihash_entries, | |
1673 | 14, | |
1674 | HASH_EARLY, | |
1675 | &i_hash_shift, | |
1676 | &i_hash_mask, | |
1677 | 0); | |
1678 | ||
1679 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1680 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1681 | } | |
1682 | ||
74bf17cf | 1683 | void __init inode_init(void) |
1da177e4 LT |
1684 | { |
1685 | int loop; | |
1686 | ||
1687 | /* inode slab cache */ | |
b0196009 PJ |
1688 | inode_cachep = kmem_cache_create("inode_cache", |
1689 | sizeof(struct inode), | |
1690 | 0, | |
1691 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1692 | SLAB_MEM_SPREAD), | |
20c2df83 | 1693 | init_once); |
8e1f936b | 1694 | register_shrinker(&icache_shrinker); |
1da177e4 LT |
1695 | |
1696 | /* Hash may have been set up in inode_init_early */ | |
1697 | if (!hashdist) | |
1698 | return; | |
1699 | ||
1700 | inode_hashtable = | |
1701 | alloc_large_system_hash("Inode-cache", | |
1702 | sizeof(struct hlist_head), | |
1703 | ihash_entries, | |
1704 | 14, | |
1705 | 0, | |
1706 | &i_hash_shift, | |
1707 | &i_hash_mask, | |
1708 | 0); | |
1709 | ||
1710 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1711 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1712 | } | |
1713 | ||
1714 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1715 | { | |
1716 | inode->i_mode = mode; | |
1717 | if (S_ISCHR(mode)) { | |
1718 | inode->i_fop = &def_chr_fops; | |
1719 | inode->i_rdev = rdev; | |
1720 | } else if (S_ISBLK(mode)) { | |
1721 | inode->i_fop = &def_blk_fops; | |
1722 | inode->i_rdev = rdev; | |
1723 | } else if (S_ISFIFO(mode)) | |
1724 | inode->i_fop = &def_fifo_fops; | |
1725 | else if (S_ISSOCK(mode)) | |
1726 | inode->i_fop = &bad_sock_fops; | |
1727 | else | |
af0d9ae8 MK |
1728 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" |
1729 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1730 | inode->i_ino); | |
1da177e4 LT |
1731 | } |
1732 | EXPORT_SYMBOL(init_special_inode); | |
a1bd120d DM |
1733 | |
1734 | /** | |
eaae668d | 1735 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards |
a1bd120d DM |
1736 | * @inode: New inode |
1737 | * @dir: Directory inode | |
1738 | * @mode: mode of the new inode | |
1739 | */ | |
1740 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
1741 | mode_t mode) | |
1742 | { | |
1743 | inode->i_uid = current_fsuid(); | |
1744 | if (dir && dir->i_mode & S_ISGID) { | |
1745 | inode->i_gid = dir->i_gid; | |
1746 | if (S_ISDIR(mode)) | |
1747 | mode |= S_ISGID; | |
1748 | } else | |
1749 | inode->i_gid = current_fsgid(); | |
1750 | inode->i_mode = mode; | |
1751 | } | |
1752 | EXPORT_SYMBOL(inode_init_owner); | |
e795b717 | 1753 | |
2e149670 SH |
1754 | /** |
1755 | * inode_owner_or_capable - check current task permissions to inode | |
1756 | * @inode: inode being checked | |
1757 | * | |
1758 | * Return true if current either has CAP_FOWNER to the inode, or | |
1759 | * owns the file. | |
e795b717 | 1760 | */ |
2e149670 | 1761 | bool inode_owner_or_capable(const struct inode *inode) |
e795b717 SH |
1762 | { |
1763 | struct user_namespace *ns = inode_userns(inode); | |
1764 | ||
1765 | if (current_user_ns() == ns && current_fsuid() == inode->i_uid) | |
1766 | return true; | |
1767 | if (ns_capable(ns, CAP_FOWNER)) | |
1768 | return true; | |
1769 | return false; | |
1770 | } | |
2e149670 | 1771 | EXPORT_SYMBOL(inode_owner_or_capable); |