]>
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() | |
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 | ||
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 | |
7ccf19a8 | 67 | static LIST_HEAD(inode_lru); |
02afc410 | 68 | static DEFINE_SPINLOCK(inode_lru_lock); |
1da177e4 | 69 | |
55fa6091 | 70 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock); |
a66979ab | 71 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_wb_list_lock); |
55fa6091 | 72 | |
1da177e4 | 73 | /* |
bab1d944 CH |
74 | * iprune_sem provides exclusion between the icache shrinking and the |
75 | * umount path. | |
88e0fbc4 | 76 | * |
bab1d944 CH |
77 | * We don't actually need it to protect anything in the umount path, |
78 | * but only need to cycle through it to make sure any inode that | |
79 | * prune_icache took off the LRU list has been fully torn down by the | |
80 | * time we are past evict_inodes. | |
1da177e4 | 81 | */ |
88e0fbc4 | 82 | static DECLARE_RWSEM(iprune_sem); |
1da177e4 | 83 | |
7dcda1c9 JA |
84 | /* |
85 | * Empty aops. Can be used for the cases where the user does not | |
86 | * define any of the address_space operations. | |
87 | */ | |
88 | const struct address_space_operations empty_aops = { | |
89 | }; | |
90 | EXPORT_SYMBOL(empty_aops); | |
91 | ||
1da177e4 LT |
92 | /* |
93 | * Statistics gathering.. | |
94 | */ | |
95 | struct inodes_stat_t inodes_stat; | |
96 | ||
3e880fb5 | 97 | static DEFINE_PER_CPU(unsigned int, nr_inodes); |
fcb94f72 | 98 | static DEFINE_PER_CPU(unsigned int, nr_unused); |
cffbc8aa | 99 | |
6b3304b5 | 100 | static struct kmem_cache *inode_cachep __read_mostly; |
1da177e4 | 101 | |
3e880fb5 | 102 | static int get_nr_inodes(void) |
cffbc8aa | 103 | { |
3e880fb5 NP |
104 | int i; |
105 | int sum = 0; | |
106 | for_each_possible_cpu(i) | |
107 | sum += per_cpu(nr_inodes, i); | |
108 | return sum < 0 ? 0 : sum; | |
cffbc8aa DC |
109 | } |
110 | ||
111 | static inline int get_nr_inodes_unused(void) | |
112 | { | |
fcb94f72 DC |
113 | int i; |
114 | int sum = 0; | |
115 | for_each_possible_cpu(i) | |
116 | sum += per_cpu(nr_unused, i); | |
117 | return sum < 0 ? 0 : sum; | |
cffbc8aa DC |
118 | } |
119 | ||
120 | int get_nr_dirty_inodes(void) | |
121 | { | |
3e880fb5 | 122 | /* not actually dirty inodes, but a wild approximation */ |
cffbc8aa DC |
123 | int nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); |
124 | return nr_dirty > 0 ? nr_dirty : 0; | |
cffbc8aa DC |
125 | } |
126 | ||
127 | /* | |
128 | * Handle nr_inode sysctl | |
129 | */ | |
130 | #ifdef CONFIG_SYSCTL | |
131 | int proc_nr_inodes(ctl_table *table, int write, | |
132 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
133 | { | |
134 | inodes_stat.nr_inodes = get_nr_inodes(); | |
fcb94f72 | 135 | inodes_stat.nr_unused = get_nr_inodes_unused(); |
cffbc8aa DC |
136 | return proc_dointvec(table, write, buffer, lenp, ppos); |
137 | } | |
138 | #endif | |
139 | ||
2cb1599f DC |
140 | /** |
141 | * inode_init_always - perform inode structure intialisation | |
0bc02f3f RD |
142 | * @sb: superblock inode belongs to |
143 | * @inode: inode to initialise | |
2cb1599f DC |
144 | * |
145 | * These are initializations that need to be done on every inode | |
146 | * allocation as the fields are not initialised by slab allocation. | |
147 | */ | |
54e34621 | 148 | int inode_init_always(struct super_block *sb, struct inode *inode) |
1da177e4 | 149 | { |
6e1d5dcc | 150 | static const struct inode_operations empty_iops; |
99ac48f5 | 151 | static const struct file_operations empty_fops; |
6b3304b5 | 152 | struct address_space *const mapping = &inode->i_data; |
2cb1599f DC |
153 | |
154 | inode->i_sb = sb; | |
155 | inode->i_blkbits = sb->s_blocksize_bits; | |
156 | inode->i_flags = 0; | |
157 | atomic_set(&inode->i_count, 1); | |
158 | inode->i_op = &empty_iops; | |
159 | inode->i_fop = &empty_fops; | |
160 | inode->i_nlink = 1; | |
56ff5efa AV |
161 | inode->i_uid = 0; |
162 | inode->i_gid = 0; | |
2cb1599f DC |
163 | atomic_set(&inode->i_writecount, 0); |
164 | inode->i_size = 0; | |
165 | inode->i_blocks = 0; | |
166 | inode->i_bytes = 0; | |
167 | inode->i_generation = 0; | |
1da177e4 | 168 | #ifdef CONFIG_QUOTA |
2cb1599f | 169 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); |
1da177e4 | 170 | #endif |
2cb1599f DC |
171 | inode->i_pipe = NULL; |
172 | inode->i_bdev = NULL; | |
173 | inode->i_cdev = NULL; | |
174 | inode->i_rdev = 0; | |
175 | inode->dirtied_when = 0; | |
6146f0d5 MZ |
176 | |
177 | if (security_inode_alloc(inode)) | |
54e34621 | 178 | goto out; |
2cb1599f DC |
179 | spin_lock_init(&inode->i_lock); |
180 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
181 | ||
182 | mutex_init(&inode->i_mutex); | |
183 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); | |
184 | ||
185 | init_rwsem(&inode->i_alloc_sem); | |
186 | lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key); | |
187 | ||
188 | mapping->a_ops = &empty_aops; | |
189 | mapping->host = inode; | |
190 | mapping->flags = 0; | |
3c1d4378 | 191 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); |
2cb1599f DC |
192 | mapping->assoc_mapping = NULL; |
193 | mapping->backing_dev_info = &default_backing_dev_info; | |
194 | mapping->writeback_index = 0; | |
195 | ||
196 | /* | |
197 | * If the block_device provides a backing_dev_info for client | |
198 | * inodes then use that. Otherwise the inode share the bdev's | |
199 | * backing_dev_info. | |
200 | */ | |
201 | if (sb->s_bdev) { | |
202 | struct backing_dev_info *bdi; | |
203 | ||
2c96ce9f | 204 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; |
2cb1599f DC |
205 | mapping->backing_dev_info = bdi; |
206 | } | |
207 | inode->i_private = NULL; | |
208 | inode->i_mapping = mapping; | |
f19d4a8f AV |
209 | #ifdef CONFIG_FS_POSIX_ACL |
210 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; | |
211 | #endif | |
2cb1599f | 212 | |
3be25f49 EP |
213 | #ifdef CONFIG_FSNOTIFY |
214 | inode->i_fsnotify_mask = 0; | |
215 | #endif | |
216 | ||
3e880fb5 | 217 | this_cpu_inc(nr_inodes); |
cffbc8aa | 218 | |
54e34621 | 219 | return 0; |
54e34621 CH |
220 | out: |
221 | return -ENOMEM; | |
1da177e4 | 222 | } |
2cb1599f DC |
223 | EXPORT_SYMBOL(inode_init_always); |
224 | ||
225 | static struct inode *alloc_inode(struct super_block *sb) | |
226 | { | |
227 | struct inode *inode; | |
228 | ||
229 | if (sb->s_op->alloc_inode) | |
230 | inode = sb->s_op->alloc_inode(sb); | |
231 | else | |
232 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
233 | ||
54e34621 CH |
234 | if (!inode) |
235 | return NULL; | |
236 | ||
237 | if (unlikely(inode_init_always(sb, inode))) { | |
238 | if (inode->i_sb->s_op->destroy_inode) | |
239 | inode->i_sb->s_op->destroy_inode(inode); | |
240 | else | |
241 | kmem_cache_free(inode_cachep, inode); | |
242 | return NULL; | |
243 | } | |
244 | ||
245 | return inode; | |
2cb1599f | 246 | } |
1da177e4 | 247 | |
ff0c7d15 NP |
248 | void free_inode_nonrcu(struct inode *inode) |
249 | { | |
250 | kmem_cache_free(inode_cachep, inode); | |
251 | } | |
252 | EXPORT_SYMBOL(free_inode_nonrcu); | |
253 | ||
2e00c97e | 254 | void __destroy_inode(struct inode *inode) |
1da177e4 | 255 | { |
b7542f8c | 256 | BUG_ON(inode_has_buffers(inode)); |
1da177e4 | 257 | security_inode_free(inode); |
3be25f49 | 258 | fsnotify_inode_delete(inode); |
f19d4a8f AV |
259 | #ifdef CONFIG_FS_POSIX_ACL |
260 | if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) | |
261 | posix_acl_release(inode->i_acl); | |
262 | if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) | |
263 | posix_acl_release(inode->i_default_acl); | |
264 | #endif | |
3e880fb5 | 265 | this_cpu_dec(nr_inodes); |
2e00c97e CH |
266 | } |
267 | EXPORT_SYMBOL(__destroy_inode); | |
268 | ||
fa0d7e3d NP |
269 | static void i_callback(struct rcu_head *head) |
270 | { | |
271 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
272 | INIT_LIST_HEAD(&inode->i_dentry); | |
273 | kmem_cache_free(inode_cachep, inode); | |
274 | } | |
275 | ||
56b0dacf | 276 | static void destroy_inode(struct inode *inode) |
2e00c97e | 277 | { |
7ccf19a8 | 278 | BUG_ON(!list_empty(&inode->i_lru)); |
2e00c97e | 279 | __destroy_inode(inode); |
1da177e4 LT |
280 | if (inode->i_sb->s_op->destroy_inode) |
281 | inode->i_sb->s_op->destroy_inode(inode); | |
282 | else | |
fa0d7e3d | 283 | call_rcu(&inode->i_rcu, i_callback); |
1da177e4 | 284 | } |
1da177e4 | 285 | |
2aa15890 MS |
286 | void address_space_init_once(struct address_space *mapping) |
287 | { | |
288 | memset(mapping, 0, sizeof(*mapping)); | |
289 | INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC); | |
290 | spin_lock_init(&mapping->tree_lock); | |
3d48ae45 | 291 | mutex_init(&mapping->i_mmap_mutex); |
2aa15890 MS |
292 | INIT_LIST_HEAD(&mapping->private_list); |
293 | spin_lock_init(&mapping->private_lock); | |
294 | INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap); | |
295 | INIT_LIST_HEAD(&mapping->i_mmap_nonlinear); | |
2aa15890 MS |
296 | } |
297 | EXPORT_SYMBOL(address_space_init_once); | |
298 | ||
1da177e4 LT |
299 | /* |
300 | * These are initializations that only need to be done | |
301 | * once, because the fields are idempotent across use | |
302 | * of the inode, so let the slab aware of that. | |
303 | */ | |
304 | void inode_init_once(struct inode *inode) | |
305 | { | |
306 | memset(inode, 0, sizeof(*inode)); | |
307 | INIT_HLIST_NODE(&inode->i_hash); | |
308 | INIT_LIST_HEAD(&inode->i_dentry); | |
309 | INIT_LIST_HEAD(&inode->i_devices); | |
7ccf19a8 NP |
310 | INIT_LIST_HEAD(&inode->i_wb_list); |
311 | INIT_LIST_HEAD(&inode->i_lru); | |
2aa15890 | 312 | address_space_init_once(&inode->i_data); |
1da177e4 | 313 | i_size_ordered_init(inode); |
3be25f49 | 314 | #ifdef CONFIG_FSNOTIFY |
e61ce867 | 315 | INIT_HLIST_HEAD(&inode->i_fsnotify_marks); |
3be25f49 | 316 | #endif |
1da177e4 | 317 | } |
1da177e4 LT |
318 | EXPORT_SYMBOL(inode_init_once); |
319 | ||
51cc5068 | 320 | static void init_once(void *foo) |
1da177e4 | 321 | { |
6b3304b5 | 322 | struct inode *inode = (struct inode *) foo; |
1da177e4 | 323 | |
a35afb83 | 324 | inode_init_once(inode); |
1da177e4 LT |
325 | } |
326 | ||
327 | /* | |
250df6ed | 328 | * inode->i_lock must be held |
1da177e4 | 329 | */ |
6b3304b5 | 330 | void __iget(struct inode *inode) |
1da177e4 | 331 | { |
9e38d86f NP |
332 | atomic_inc(&inode->i_count); |
333 | } | |
2e147f1e | 334 | |
7de9c6ee AV |
335 | /* |
336 | * get additional reference to inode; caller must already hold one. | |
337 | */ | |
338 | void ihold(struct inode *inode) | |
339 | { | |
340 | WARN_ON(atomic_inc_return(&inode->i_count) < 2); | |
341 | } | |
342 | EXPORT_SYMBOL(ihold); | |
343 | ||
9e38d86f NP |
344 | static void inode_lru_list_add(struct inode *inode) |
345 | { | |
02afc410 | 346 | spin_lock(&inode_lru_lock); |
7ccf19a8 NP |
347 | if (list_empty(&inode->i_lru)) { |
348 | list_add(&inode->i_lru, &inode_lru); | |
fcb94f72 | 349 | this_cpu_inc(nr_unused); |
9e38d86f | 350 | } |
02afc410 | 351 | spin_unlock(&inode_lru_lock); |
9e38d86f | 352 | } |
2e147f1e | 353 | |
9e38d86f NP |
354 | static void inode_lru_list_del(struct inode *inode) |
355 | { | |
02afc410 | 356 | spin_lock(&inode_lru_lock); |
7ccf19a8 NP |
357 | if (!list_empty(&inode->i_lru)) { |
358 | list_del_init(&inode->i_lru); | |
fcb94f72 | 359 | this_cpu_dec(nr_unused); |
9e38d86f | 360 | } |
02afc410 | 361 | spin_unlock(&inode_lru_lock); |
1da177e4 LT |
362 | } |
363 | ||
646ec461 CH |
364 | /** |
365 | * inode_sb_list_add - add inode to the superblock list of inodes | |
366 | * @inode: inode to add | |
367 | */ | |
368 | void inode_sb_list_add(struct inode *inode) | |
369 | { | |
55fa6091 DC |
370 | spin_lock(&inode_sb_list_lock); |
371 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); | |
372 | spin_unlock(&inode_sb_list_lock); | |
646ec461 CH |
373 | } |
374 | EXPORT_SYMBOL_GPL(inode_sb_list_add); | |
375 | ||
55fa6091 | 376 | static inline void inode_sb_list_del(struct inode *inode) |
646ec461 | 377 | { |
55fa6091 | 378 | spin_lock(&inode_sb_list_lock); |
646ec461 | 379 | list_del_init(&inode->i_sb_list); |
55fa6091 | 380 | spin_unlock(&inode_sb_list_lock); |
646ec461 CH |
381 | } |
382 | ||
4c51acbc DC |
383 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
384 | { | |
385 | unsigned long tmp; | |
386 | ||
387 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
388 | L1_CACHE_BYTES; | |
4b4563dc CH |
389 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift); |
390 | return tmp & i_hash_mask; | |
4c51acbc DC |
391 | } |
392 | ||
393 | /** | |
394 | * __insert_inode_hash - hash an inode | |
395 | * @inode: unhashed inode | |
396 | * @hashval: unsigned long value used to locate this object in the | |
397 | * inode_hashtable. | |
398 | * | |
399 | * Add an inode to the inode hash for this superblock. | |
400 | */ | |
401 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
402 | { | |
646ec461 CH |
403 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); |
404 | ||
67a23c49 | 405 | spin_lock(&inode_hash_lock); |
250df6ed | 406 | spin_lock(&inode->i_lock); |
646ec461 | 407 | hlist_add_head(&inode->i_hash, b); |
250df6ed | 408 | spin_unlock(&inode->i_lock); |
67a23c49 | 409 | spin_unlock(&inode_hash_lock); |
4c51acbc DC |
410 | } |
411 | EXPORT_SYMBOL(__insert_inode_hash); | |
412 | ||
4c51acbc DC |
413 | /** |
414 | * remove_inode_hash - remove an inode from the hash | |
415 | * @inode: inode to unhash | |
416 | * | |
417 | * Remove an inode from the superblock. | |
418 | */ | |
419 | void remove_inode_hash(struct inode *inode) | |
420 | { | |
67a23c49 | 421 | spin_lock(&inode_hash_lock); |
250df6ed | 422 | spin_lock(&inode->i_lock); |
4c51acbc | 423 | hlist_del_init(&inode->i_hash); |
250df6ed | 424 | spin_unlock(&inode->i_lock); |
67a23c49 | 425 | spin_unlock(&inode_hash_lock); |
4c51acbc DC |
426 | } |
427 | EXPORT_SYMBOL(remove_inode_hash); | |
428 | ||
b0683aa6 AV |
429 | void end_writeback(struct inode *inode) |
430 | { | |
431 | might_sleep(); | |
08142579 JK |
432 | /* |
433 | * We have to cycle tree_lock here because reclaim can be still in the | |
434 | * process of removing the last page (in __delete_from_page_cache()) | |
435 | * and we must not free mapping under it. | |
436 | */ | |
437 | spin_lock_irq(&inode->i_data.tree_lock); | |
b0683aa6 | 438 | BUG_ON(inode->i_data.nrpages); |
08142579 | 439 | spin_unlock_irq(&inode->i_data.tree_lock); |
b0683aa6 AV |
440 | BUG_ON(!list_empty(&inode->i_data.private_list)); |
441 | BUG_ON(!(inode->i_state & I_FREEING)); | |
442 | BUG_ON(inode->i_state & I_CLEAR); | |
443 | inode_sync_wait(inode); | |
fa0d7e3d | 444 | /* don't need i_lock here, no concurrent mods to i_state */ |
b0683aa6 AV |
445 | inode->i_state = I_FREEING | I_CLEAR; |
446 | } | |
447 | EXPORT_SYMBOL(end_writeback); | |
448 | ||
b2b2af8e DC |
449 | /* |
450 | * Free the inode passed in, removing it from the lists it is still connected | |
451 | * to. We remove any pages still attached to the inode and wait for any IO that | |
452 | * is still in progress before finally destroying the inode. | |
453 | * | |
454 | * An inode must already be marked I_FREEING so that we avoid the inode being | |
455 | * moved back onto lists if we race with other code that manipulates the lists | |
456 | * (e.g. writeback_single_inode). The caller is responsible for setting this. | |
457 | * | |
458 | * An inode must already be removed from the LRU list before being evicted from | |
459 | * the cache. This should occur atomically with setting the I_FREEING state | |
460 | * flag, so no inodes here should ever be on the LRU when being evicted. | |
461 | */ | |
644da596 | 462 | static void evict(struct inode *inode) |
b4272d4c AV |
463 | { |
464 | const struct super_operations *op = inode->i_sb->s_op; | |
465 | ||
b2b2af8e DC |
466 | BUG_ON(!(inode->i_state & I_FREEING)); |
467 | BUG_ON(!list_empty(&inode->i_lru)); | |
468 | ||
a66979ab | 469 | inode_wb_list_del(inode); |
55fa6091 DC |
470 | inode_sb_list_del(inode); |
471 | ||
be7ce416 AV |
472 | if (op->evict_inode) { |
473 | op->evict_inode(inode); | |
b4272d4c AV |
474 | } else { |
475 | if (inode->i_data.nrpages) | |
476 | truncate_inode_pages(&inode->i_data, 0); | |
30140837 | 477 | end_writeback(inode); |
b4272d4c | 478 | } |
661074e9 AV |
479 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
480 | bd_forget(inode); | |
481 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) | |
482 | cd_forget(inode); | |
b2b2af8e DC |
483 | |
484 | remove_inode_hash(inode); | |
485 | ||
486 | spin_lock(&inode->i_lock); | |
487 | wake_up_bit(&inode->i_state, __I_NEW); | |
488 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); | |
489 | spin_unlock(&inode->i_lock); | |
490 | ||
491 | destroy_inode(inode); | |
b4272d4c AV |
492 | } |
493 | ||
1da177e4 LT |
494 | /* |
495 | * dispose_list - dispose of the contents of a local list | |
496 | * @head: the head of the list to free | |
497 | * | |
498 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
499 | * need to worry about list corruption and SMP locks. | |
500 | */ | |
501 | static void dispose_list(struct list_head *head) | |
502 | { | |
1da177e4 LT |
503 | while (!list_empty(head)) { |
504 | struct inode *inode; | |
505 | ||
7ccf19a8 NP |
506 | inode = list_first_entry(head, struct inode, i_lru); |
507 | list_del_init(&inode->i_lru); | |
1da177e4 | 508 | |
644da596 | 509 | evict(inode); |
1da177e4 | 510 | } |
1da177e4 LT |
511 | } |
512 | ||
63997e98 AV |
513 | /** |
514 | * evict_inodes - evict all evictable inodes for a superblock | |
515 | * @sb: superblock to operate on | |
516 | * | |
517 | * Make sure that no inodes with zero refcount are retained. This is | |
518 | * called by superblock shutdown after having MS_ACTIVE flag removed, | |
519 | * so any inode reaching zero refcount during or after that call will | |
520 | * be immediately evicted. | |
1da177e4 | 521 | */ |
63997e98 | 522 | void evict_inodes(struct super_block *sb) |
1da177e4 | 523 | { |
63997e98 AV |
524 | struct inode *inode, *next; |
525 | LIST_HEAD(dispose); | |
1da177e4 | 526 | |
55fa6091 | 527 | spin_lock(&inode_sb_list_lock); |
63997e98 AV |
528 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
529 | if (atomic_read(&inode->i_count)) | |
aabb8fdb | 530 | continue; |
250df6ed DC |
531 | |
532 | spin_lock(&inode->i_lock); | |
533 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
534 | spin_unlock(&inode->i_lock); | |
1da177e4 | 535 | continue; |
250df6ed | 536 | } |
63997e98 AV |
537 | |
538 | inode->i_state |= I_FREEING; | |
02afc410 | 539 | inode_lru_list_del(inode); |
250df6ed | 540 | spin_unlock(&inode->i_lock); |
02afc410 | 541 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 542 | } |
55fa6091 | 543 | spin_unlock(&inode_sb_list_lock); |
63997e98 AV |
544 | |
545 | dispose_list(&dispose); | |
bab1d944 CH |
546 | |
547 | /* | |
548 | * Cycle through iprune_sem to make sure any inode that prune_icache | |
549 | * moved off the list before we took the lock has been fully torn | |
550 | * down. | |
551 | */ | |
552 | down_write(&iprune_sem); | |
63997e98 | 553 | up_write(&iprune_sem); |
1da177e4 LT |
554 | } |
555 | ||
1da177e4 | 556 | /** |
a0318786 CH |
557 | * invalidate_inodes - attempt to free all inodes on a superblock |
558 | * @sb: superblock to operate on | |
93b270f7 | 559 | * @kill_dirty: flag to guide handling of dirty inodes |
1da177e4 | 560 | * |
a0318786 CH |
561 | * Attempts to free all inodes for a given superblock. If there were any |
562 | * busy inodes return a non-zero value, else zero. | |
93b270f7 N |
563 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat |
564 | * them as busy. | |
1da177e4 | 565 | */ |
93b270f7 | 566 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) |
1da177e4 | 567 | { |
cffbc8aa | 568 | int busy = 0; |
a0318786 CH |
569 | struct inode *inode, *next; |
570 | LIST_HEAD(dispose); | |
1da177e4 | 571 | |
55fa6091 | 572 | spin_lock(&inode_sb_list_lock); |
a0318786 | 573 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
250df6ed DC |
574 | spin_lock(&inode->i_lock); |
575 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
576 | spin_unlock(&inode->i_lock); | |
aabb8fdb | 577 | continue; |
250df6ed | 578 | } |
93b270f7 | 579 | if (inode->i_state & I_DIRTY && !kill_dirty) { |
250df6ed | 580 | spin_unlock(&inode->i_lock); |
93b270f7 N |
581 | busy = 1; |
582 | continue; | |
583 | } | |
99a38919 | 584 | if (atomic_read(&inode->i_count)) { |
250df6ed | 585 | spin_unlock(&inode->i_lock); |
99a38919 | 586 | busy = 1; |
1da177e4 LT |
587 | continue; |
588 | } | |
99a38919 | 589 | |
99a38919 | 590 | inode->i_state |= I_FREEING; |
02afc410 | 591 | inode_lru_list_del(inode); |
250df6ed | 592 | spin_unlock(&inode->i_lock); |
02afc410 | 593 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 594 | } |
55fa6091 | 595 | spin_unlock(&inode_sb_list_lock); |
1da177e4 | 596 | |
a0318786 | 597 | dispose_list(&dispose); |
1da177e4 LT |
598 | |
599 | return busy; | |
600 | } | |
1da177e4 LT |
601 | |
602 | static int can_unuse(struct inode *inode) | |
603 | { | |
9e38d86f | 604 | if (inode->i_state & ~I_REFERENCED) |
1da177e4 LT |
605 | return 0; |
606 | if (inode_has_buffers(inode)) | |
607 | return 0; | |
608 | if (atomic_read(&inode->i_count)) | |
609 | return 0; | |
610 | if (inode->i_data.nrpages) | |
611 | return 0; | |
612 | return 1; | |
613 | } | |
614 | ||
615 | /* | |
9e38d86f | 616 | * Scan `goal' inodes on the unused list for freeable ones. They are moved to a |
02afc410 | 617 | * temporary list and then are freed outside inode_lru_lock by 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 LT |
630 | */ |
631 | static void prune_icache(int nr_to_scan) | |
632 | { | |
633 | LIST_HEAD(freeable); | |
1da177e4 LT |
634 | int nr_scanned; |
635 | unsigned long reap = 0; | |
636 | ||
88e0fbc4 | 637 | down_read(&iprune_sem); |
02afc410 | 638 | spin_lock(&inode_lru_lock); |
1da177e4 LT |
639 | for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) { |
640 | struct inode *inode; | |
641 | ||
7ccf19a8 | 642 | if (list_empty(&inode_lru)) |
1da177e4 LT |
643 | break; |
644 | ||
7ccf19a8 | 645 | inode = list_entry(inode_lru.prev, struct inode, i_lru); |
1da177e4 | 646 | |
02afc410 DC |
647 | /* |
648 | * we are inverting the inode_lru_lock/inode->i_lock here, | |
649 | * so use a trylock. If we fail to get the lock, just move the | |
650 | * inode to the back of the list so we don't spin on it. | |
651 | */ | |
652 | if (!spin_trylock(&inode->i_lock)) { | |
653 | list_move(&inode->i_lru, &inode_lru); | |
654 | continue; | |
655 | } | |
656 | ||
9e38d86f NP |
657 | /* |
658 | * Referenced or dirty inodes are still in use. Give them | |
659 | * another pass through the LRU as we canot reclaim them now. | |
660 | */ | |
661 | if (atomic_read(&inode->i_count) || | |
662 | (inode->i_state & ~I_REFERENCED)) { | |
7ccf19a8 | 663 | list_del_init(&inode->i_lru); |
f283c86a | 664 | spin_unlock(&inode->i_lock); |
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; |
250df6ed | 672 | list_move(&inode->i_lru, &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); |
02afc410 | 679 | spin_unlock(&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); |
02afc410 | 684 | spin_lock(&inode_lru_lock); |
1da177e4 | 685 | |
7ccf19a8 NP |
686 | if (inode != list_entry(inode_lru.next, |
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); |
fcb94f72 | 702 | this_cpu_dec(nr_unused); |
1da177e4 | 703 | } |
f8891e5e CL |
704 | if (current_is_kswapd()) |
705 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
706 | else | |
707 | __count_vm_events(PGINODESTEAL, reap); | |
02afc410 | 708 | spin_unlock(&inode_lru_lock); |
1da177e4 LT |
709 | |
710 | dispose_list(&freeable); | |
88e0fbc4 | 711 | up_read(&iprune_sem); |
1da177e4 LT |
712 | } |
713 | ||
714 | /* | |
715 | * shrink_icache_memory() will attempt to reclaim some unused inodes. Here, | |
716 | * "unused" means that no dentries are referring to the inodes: the files are | |
717 | * not open and the dcache references to those inodes have already been | |
718 | * reclaimed. | |
719 | * | |
720 | * This function is passed the number of inodes to scan, and it returns the | |
721 | * total number of remaining possibly-reclaimable inodes. | |
722 | */ | |
1495f230 YH |
723 | static int shrink_icache_memory(struct shrinker *shrink, |
724 | struct shrink_control *sc) | |
1da177e4 | 725 | { |
1495f230 YH |
726 | int nr = sc->nr_to_scan; |
727 | gfp_t gfp_mask = sc->gfp_mask; | |
728 | ||
1da177e4 LT |
729 | if (nr) { |
730 | /* | |
731 | * Nasty deadlock avoidance. We may hold various FS locks, | |
732 | * and we don't want to recurse into the FS that called us | |
733 | * in clear_inode() and friends.. | |
6b3304b5 | 734 | */ |
1da177e4 LT |
735 | if (!(gfp_mask & __GFP_FS)) |
736 | return -1; | |
737 | prune_icache(nr); | |
738 | } | |
cffbc8aa | 739 | return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure; |
1da177e4 LT |
740 | } |
741 | ||
8e1f936b RR |
742 | static struct shrinker icache_shrinker = { |
743 | .shrink = shrink_icache_memory, | |
744 | .seeks = DEFAULT_SEEKS, | |
745 | }; | |
746 | ||
1da177e4 LT |
747 | static void __wait_on_freeing_inode(struct inode *inode); |
748 | /* | |
749 | * Called with the inode lock held. | |
1da177e4 | 750 | */ |
6b3304b5 MK |
751 | static struct inode *find_inode(struct super_block *sb, |
752 | struct hlist_head *head, | |
753 | int (*test)(struct inode *, void *), | |
754 | void *data) | |
1da177e4 LT |
755 | { |
756 | struct hlist_node *node; | |
6b3304b5 | 757 | struct inode *inode = NULL; |
1da177e4 LT |
758 | |
759 | repeat: | |
c5c8be3c | 760 | hlist_for_each_entry(inode, node, head, i_hash) { |
67a23c49 DC |
761 | spin_lock(&inode->i_lock); |
762 | if (inode->i_sb != sb) { | |
763 | spin_unlock(&inode->i_lock); | |
1da177e4 | 764 | continue; |
67a23c49 DC |
765 | } |
766 | if (!test(inode, data)) { | |
767 | spin_unlock(&inode->i_lock); | |
1da177e4 | 768 | continue; |
67a23c49 | 769 | } |
a4ffdde6 | 770 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
771 | __wait_on_freeing_inode(inode); |
772 | goto repeat; | |
773 | } | |
f7899bd5 | 774 | __iget(inode); |
250df6ed | 775 | spin_unlock(&inode->i_lock); |
f7899bd5 | 776 | return inode; |
1da177e4 | 777 | } |
f7899bd5 | 778 | return NULL; |
1da177e4 LT |
779 | } |
780 | ||
781 | /* | |
782 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
783 | * iget_locked for details. | |
784 | */ | |
6b3304b5 MK |
785 | static struct inode *find_inode_fast(struct super_block *sb, |
786 | struct hlist_head *head, unsigned long ino) | |
1da177e4 LT |
787 | { |
788 | struct hlist_node *node; | |
6b3304b5 | 789 | struct inode *inode = NULL; |
1da177e4 LT |
790 | |
791 | repeat: | |
c5c8be3c | 792 | hlist_for_each_entry(inode, node, head, i_hash) { |
67a23c49 DC |
793 | spin_lock(&inode->i_lock); |
794 | if (inode->i_ino != ino) { | |
795 | spin_unlock(&inode->i_lock); | |
1da177e4 | 796 | continue; |
67a23c49 DC |
797 | } |
798 | if (inode->i_sb != sb) { | |
799 | spin_unlock(&inode->i_lock); | |
1da177e4 | 800 | continue; |
67a23c49 | 801 | } |
a4ffdde6 | 802 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
803 | __wait_on_freeing_inode(inode); |
804 | goto repeat; | |
805 | } | |
f7899bd5 | 806 | __iget(inode); |
250df6ed | 807 | spin_unlock(&inode->i_lock); |
f7899bd5 | 808 | return inode; |
1da177e4 | 809 | } |
f7899bd5 | 810 | return NULL; |
8290c35f DC |
811 | } |
812 | ||
f991bd2e ED |
813 | /* |
814 | * Each cpu owns a range of LAST_INO_BATCH numbers. | |
815 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, | |
816 | * to renew the exhausted range. | |
8290c35f | 817 | * |
f991bd2e ED |
818 | * This does not significantly increase overflow rate because every CPU can |
819 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is | |
820 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the | |
821 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase | |
822 | * overflow rate by 2x, which does not seem too significant. | |
823 | * | |
824 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
825 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
826 | * here to attempt to avoid that. | |
8290c35f | 827 | */ |
f991bd2e ED |
828 | #define LAST_INO_BATCH 1024 |
829 | static DEFINE_PER_CPU(unsigned int, last_ino); | |
830 | ||
85fe4025 | 831 | unsigned int get_next_ino(void) |
8290c35f | 832 | { |
f991bd2e ED |
833 | unsigned int *p = &get_cpu_var(last_ino); |
834 | unsigned int res = *p; | |
8290c35f | 835 | |
f991bd2e ED |
836 | #ifdef CONFIG_SMP |
837 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { | |
838 | static atomic_t shared_last_ino; | |
839 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); | |
840 | ||
841 | res = next - LAST_INO_BATCH; | |
842 | } | |
843 | #endif | |
844 | ||
845 | *p = ++res; | |
846 | put_cpu_var(last_ino); | |
847 | return res; | |
8290c35f | 848 | } |
85fe4025 | 849 | EXPORT_SYMBOL(get_next_ino); |
8290c35f | 850 | |
1da177e4 LT |
851 | /** |
852 | * new_inode - obtain an inode | |
853 | * @sb: superblock | |
854 | * | |
769848c0 | 855 | * Allocates a new inode for given superblock. The default gfp_mask |
3c1d4378 | 856 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
769848c0 MG |
857 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
858 | * for the page cache are not reclaimable or migratable, | |
859 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
860 | * newly created inode's mapping | |
861 | * | |
1da177e4 LT |
862 | */ |
863 | struct inode *new_inode(struct super_block *sb) | |
864 | { | |
6b3304b5 | 865 | struct inode *inode; |
1da177e4 | 866 | |
55fa6091 | 867 | spin_lock_prefetch(&inode_sb_list_lock); |
6b3304b5 | 868 | |
1da177e4 LT |
869 | inode = alloc_inode(sb); |
870 | if (inode) { | |
250df6ed | 871 | spin_lock(&inode->i_lock); |
1da177e4 | 872 | inode->i_state = 0; |
250df6ed | 873 | spin_unlock(&inode->i_lock); |
55fa6091 | 874 | inode_sb_list_add(inode); |
1da177e4 LT |
875 | } |
876 | return inode; | |
877 | } | |
1da177e4 LT |
878 | EXPORT_SYMBOL(new_inode); |
879 | ||
250df6ed DC |
880 | /** |
881 | * unlock_new_inode - clear the I_NEW state and wake up any waiters | |
882 | * @inode: new inode to unlock | |
883 | * | |
884 | * Called when the inode is fully initialised to clear the new state of the | |
885 | * inode and wake up anyone waiting for the inode to finish initialisation. | |
886 | */ | |
1da177e4 LT |
887 | void unlock_new_inode(struct inode *inode) |
888 | { | |
14358e6d | 889 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
a3314a0e | 890 | if (S_ISDIR(inode->i_mode)) { |
1e89a5e1 PZ |
891 | struct file_system_type *type = inode->i_sb->s_type; |
892 | ||
9a7aa12f JK |
893 | /* Set new key only if filesystem hasn't already changed it */ |
894 | if (!lockdep_match_class(&inode->i_mutex, | |
895 | &type->i_mutex_key)) { | |
896 | /* | |
897 | * ensure nobody is actually holding i_mutex | |
898 | */ | |
899 | mutex_destroy(&inode->i_mutex); | |
900 | mutex_init(&inode->i_mutex); | |
901 | lockdep_set_class(&inode->i_mutex, | |
902 | &type->i_mutex_dir_key); | |
903 | } | |
1e89a5e1 | 904 | } |
14358e6d | 905 | #endif |
250df6ed | 906 | spin_lock(&inode->i_lock); |
eaff8079 CH |
907 | WARN_ON(!(inode->i_state & I_NEW)); |
908 | inode->i_state &= ~I_NEW; | |
250df6ed DC |
909 | wake_up_bit(&inode->i_state, __I_NEW); |
910 | spin_unlock(&inode->i_lock); | |
1da177e4 | 911 | } |
1da177e4 LT |
912 | EXPORT_SYMBOL(unlock_new_inode); |
913 | ||
0b2d0724 CH |
914 | /** |
915 | * iget5_locked - obtain an inode from a mounted file system | |
916 | * @sb: super block of file system | |
917 | * @hashval: hash value (usually inode number) to get | |
918 | * @test: callback used for comparisons between inodes | |
919 | * @set: callback used to initialize a new struct inode | |
920 | * @data: opaque data pointer to pass to @test and @set | |
921 | * | |
922 | * Search for the inode specified by @hashval and @data in the inode cache, | |
923 | * and if present it is return it with an increased reference count. This is | |
924 | * a generalized version of iget_locked() for file systems where the inode | |
925 | * number is not sufficient for unique identification of an inode. | |
926 | * | |
927 | * If the inode is not in cache, allocate a new inode and return it locked, | |
928 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
929 | * before unlocking it via unlock_new_inode(). | |
1da177e4 | 930 | * |
0b2d0724 CH |
931 | * Note both @test and @set are called with the inode_hash_lock held, so can't |
932 | * sleep. | |
1da177e4 | 933 | */ |
0b2d0724 CH |
934 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, |
935 | int (*test)(struct inode *, void *), | |
936 | int (*set)(struct inode *, void *), void *data) | |
1da177e4 | 937 | { |
0b2d0724 | 938 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
6b3304b5 | 939 | struct inode *inode; |
1da177e4 | 940 | |
0b2d0724 CH |
941 | spin_lock(&inode_hash_lock); |
942 | inode = find_inode(sb, head, test, data); | |
943 | spin_unlock(&inode_hash_lock); | |
944 | ||
945 | if (inode) { | |
946 | wait_on_inode(inode); | |
947 | return inode; | |
948 | } | |
949 | ||
1da177e4 LT |
950 | inode = alloc_inode(sb); |
951 | if (inode) { | |
6b3304b5 | 952 | struct inode *old; |
1da177e4 | 953 | |
67a23c49 | 954 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
955 | /* We released the lock, so.. */ |
956 | old = find_inode(sb, head, test, data); | |
957 | if (!old) { | |
958 | if (set(inode, data)) | |
959 | goto set_failed; | |
960 | ||
250df6ed DC |
961 | spin_lock(&inode->i_lock); |
962 | inode->i_state = I_NEW; | |
646ec461 | 963 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 964 | spin_unlock(&inode->i_lock); |
55fa6091 | 965 | inode_sb_list_add(inode); |
67a23c49 | 966 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
967 | |
968 | /* Return the locked inode with I_NEW set, the | |
969 | * caller is responsible for filling in the contents | |
970 | */ | |
971 | return inode; | |
972 | } | |
973 | ||
974 | /* | |
975 | * Uhhuh, somebody else created the same inode under | |
976 | * us. Use the old inode instead of the one we just | |
977 | * allocated. | |
978 | */ | |
67a23c49 | 979 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
980 | destroy_inode(inode); |
981 | inode = old; | |
982 | wait_on_inode(inode); | |
983 | } | |
984 | return inode; | |
985 | ||
986 | set_failed: | |
67a23c49 | 987 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
988 | destroy_inode(inode); |
989 | return NULL; | |
990 | } | |
0b2d0724 | 991 | EXPORT_SYMBOL(iget5_locked); |
1da177e4 | 992 | |
0b2d0724 CH |
993 | /** |
994 | * iget_locked - obtain an inode from a mounted file system | |
995 | * @sb: super block of file system | |
996 | * @ino: inode number to get | |
997 | * | |
998 | * Search for the inode specified by @ino in the inode cache and if present | |
999 | * return it with an increased reference count. This is for file systems | |
1000 | * where the inode number is sufficient for unique identification of an inode. | |
1001 | * | |
1002 | * If the inode is not in cache, allocate a new inode and return it locked, | |
1003 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
1004 | * before unlocking it via unlock_new_inode(). | |
1da177e4 | 1005 | */ |
0b2d0724 | 1006 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) |
1da177e4 | 1007 | { |
0b2d0724 | 1008 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
6b3304b5 | 1009 | struct inode *inode; |
1da177e4 | 1010 | |
0b2d0724 CH |
1011 | spin_lock(&inode_hash_lock); |
1012 | inode = find_inode_fast(sb, head, ino); | |
1013 | spin_unlock(&inode_hash_lock); | |
1014 | if (inode) { | |
1015 | wait_on_inode(inode); | |
1016 | return inode; | |
1017 | } | |
1018 | ||
1da177e4 LT |
1019 | inode = alloc_inode(sb); |
1020 | if (inode) { | |
6b3304b5 | 1021 | struct inode *old; |
1da177e4 | 1022 | |
67a23c49 | 1023 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1024 | /* We released the lock, so.. */ |
1025 | old = find_inode_fast(sb, head, ino); | |
1026 | if (!old) { | |
1027 | inode->i_ino = ino; | |
250df6ed DC |
1028 | spin_lock(&inode->i_lock); |
1029 | inode->i_state = I_NEW; | |
646ec461 | 1030 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1031 | spin_unlock(&inode->i_lock); |
55fa6091 | 1032 | inode_sb_list_add(inode); |
67a23c49 | 1033 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1034 | |
1035 | /* Return the locked inode with I_NEW set, the | |
1036 | * caller is responsible for filling in the contents | |
1037 | */ | |
1038 | return inode; | |
1039 | } | |
1040 | ||
1041 | /* | |
1042 | * Uhhuh, somebody else created the same inode under | |
1043 | * us. Use the old inode instead of the one we just | |
1044 | * allocated. | |
1045 | */ | |
67a23c49 | 1046 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1047 | destroy_inode(inode); |
1048 | inode = old; | |
1049 | wait_on_inode(inode); | |
1050 | } | |
1051 | return inode; | |
1052 | } | |
0b2d0724 | 1053 | EXPORT_SYMBOL(iget_locked); |
1da177e4 | 1054 | |
ad5e195a CH |
1055 | /* |
1056 | * search the inode cache for a matching inode number. | |
1057 | * If we find one, then the inode number we are trying to | |
1058 | * allocate is not unique and so we should not use it. | |
1059 | * | |
1060 | * Returns 1 if the inode number is unique, 0 if it is not. | |
1061 | */ | |
1062 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) | |
1063 | { | |
1064 | struct hlist_head *b = inode_hashtable + hash(sb, ino); | |
1065 | struct hlist_node *node; | |
1066 | struct inode *inode; | |
1067 | ||
67a23c49 | 1068 | spin_lock(&inode_hash_lock); |
ad5e195a | 1069 | hlist_for_each_entry(inode, node, b, i_hash) { |
67a23c49 DC |
1070 | if (inode->i_ino == ino && inode->i_sb == sb) { |
1071 | spin_unlock(&inode_hash_lock); | |
ad5e195a | 1072 | return 0; |
67a23c49 | 1073 | } |
ad5e195a | 1074 | } |
67a23c49 | 1075 | spin_unlock(&inode_hash_lock); |
ad5e195a CH |
1076 | |
1077 | return 1; | |
1078 | } | |
1079 | ||
1da177e4 LT |
1080 | /** |
1081 | * iunique - get a unique inode number | |
1082 | * @sb: superblock | |
1083 | * @max_reserved: highest reserved inode number | |
1084 | * | |
1085 | * Obtain an inode number that is unique on the system for a given | |
1086 | * superblock. This is used by file systems that have no natural | |
1087 | * permanent inode numbering system. An inode number is returned that | |
1088 | * is higher than the reserved limit but unique. | |
1089 | * | |
1090 | * BUGS: | |
1091 | * With a large number of inodes live on the file system this function | |
1092 | * currently becomes quite slow. | |
1093 | */ | |
1094 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
1095 | { | |
866b04fc JL |
1096 | /* |
1097 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
1098 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
1099 | * here to attempt to avoid that. | |
1100 | */ | |
ad5e195a | 1101 | static DEFINE_SPINLOCK(iunique_lock); |
866b04fc | 1102 | static unsigned int counter; |
1da177e4 | 1103 | ino_t res; |
3361c7be | 1104 | |
ad5e195a | 1105 | spin_lock(&iunique_lock); |
3361c7be JL |
1106 | do { |
1107 | if (counter <= max_reserved) | |
1108 | counter = max_reserved + 1; | |
1da177e4 | 1109 | res = counter++; |
ad5e195a CH |
1110 | } while (!test_inode_iunique(sb, res)); |
1111 | spin_unlock(&iunique_lock); | |
1da177e4 | 1112 | |
3361c7be JL |
1113 | return res; |
1114 | } | |
1da177e4 LT |
1115 | EXPORT_SYMBOL(iunique); |
1116 | ||
1117 | struct inode *igrab(struct inode *inode) | |
1118 | { | |
250df6ed DC |
1119 | spin_lock(&inode->i_lock); |
1120 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { | |
1da177e4 | 1121 | __iget(inode); |
250df6ed DC |
1122 | spin_unlock(&inode->i_lock); |
1123 | } else { | |
1124 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1125 | /* |
1126 | * Handle the case where s_op->clear_inode is not been | |
1127 | * called yet, and somebody is calling igrab | |
1128 | * while the inode is getting freed. | |
1129 | */ | |
1130 | inode = NULL; | |
250df6ed | 1131 | } |
1da177e4 LT |
1132 | return inode; |
1133 | } | |
1da177e4 LT |
1134 | EXPORT_SYMBOL(igrab); |
1135 | ||
1136 | /** | |
0b2d0724 | 1137 | * ilookup5_nowait - search for an inode in the inode cache |
1da177e4 | 1138 | * @sb: super block of file system to search |
0b2d0724 | 1139 | * @hashval: hash value (usually inode number) to search for |
1da177e4 LT |
1140 | * @test: callback used for comparisons between inodes |
1141 | * @data: opaque data pointer to pass to @test | |
1da177e4 | 1142 | * |
0b2d0724 | 1143 | * Search for the inode specified by @hashval and @data in the inode cache. |
1da177e4 LT |
1144 | * If the inode is in the cache, the inode is returned with an incremented |
1145 | * reference count. | |
1146 | * | |
0b2d0724 CH |
1147 | * Note: I_NEW is not waited upon so you have to be very careful what you do |
1148 | * with the returned inode. You probably should be using ilookup5() instead. | |
1da177e4 | 1149 | * |
b6d0ad68 | 1150 | * Note2: @test is called with the inode_hash_lock held, so can't sleep. |
1da177e4 | 1151 | */ |
0b2d0724 CH |
1152 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, |
1153 | int (*test)(struct inode *, void *), void *data) | |
1da177e4 | 1154 | { |
0b2d0724 | 1155 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
1da177e4 LT |
1156 | struct inode *inode; |
1157 | ||
67a23c49 | 1158 | spin_lock(&inode_hash_lock); |
1da177e4 | 1159 | inode = find_inode(sb, head, test, data); |
67a23c49 | 1160 | spin_unlock(&inode_hash_lock); |
88bd5121 | 1161 | |
0b2d0724 | 1162 | return inode; |
88bd5121 | 1163 | } |
88bd5121 AA |
1164 | EXPORT_SYMBOL(ilookup5_nowait); |
1165 | ||
1166 | /** | |
1167 | * ilookup5 - search for an inode in the inode cache | |
1168 | * @sb: super block of file system to search | |
1169 | * @hashval: hash value (usually inode number) to search for | |
1170 | * @test: callback used for comparisons between inodes | |
1171 | * @data: opaque data pointer to pass to @test | |
1172 | * | |
0b2d0724 CH |
1173 | * Search for the inode specified by @hashval and @data in the inode cache, |
1174 | * and if the inode is in the cache, return the inode with an incremented | |
1175 | * reference count. Waits on I_NEW before returning the inode. | |
88bd5121 | 1176 | * returned with an incremented reference count. |
1da177e4 | 1177 | * |
0b2d0724 CH |
1178 | * This is a generalized version of ilookup() for file systems where the |
1179 | * inode number is not sufficient for unique identification of an inode. | |
1da177e4 | 1180 | * |
0b2d0724 | 1181 | * Note: @test is called with the inode_hash_lock held, so can't sleep. |
1da177e4 LT |
1182 | */ |
1183 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
1184 | int (*test)(struct inode *, void *), void *data) | |
1185 | { | |
0b2d0724 | 1186 | struct inode *inode = ilookup5_nowait(sb, hashval, test, data); |
1da177e4 | 1187 | |
0b2d0724 CH |
1188 | if (inode) |
1189 | wait_on_inode(inode); | |
1190 | return inode; | |
1da177e4 | 1191 | } |
1da177e4 LT |
1192 | EXPORT_SYMBOL(ilookup5); |
1193 | ||
1194 | /** | |
1195 | * ilookup - search for an inode in the inode cache | |
1196 | * @sb: super block of file system to search | |
1197 | * @ino: inode number to search for | |
1198 | * | |
0b2d0724 CH |
1199 | * Search for the inode @ino in the inode cache, and if the inode is in the |
1200 | * cache, the inode is returned with an incremented reference count. | |
1da177e4 LT |
1201 | */ |
1202 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1203 | { | |
1204 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1da177e4 LT |
1205 | struct inode *inode; |
1206 | ||
0b2d0724 CH |
1207 | spin_lock(&inode_hash_lock); |
1208 | inode = find_inode_fast(sb, head, ino); | |
1209 | spin_unlock(&inode_hash_lock); | |
1da177e4 | 1210 | |
1da177e4 | 1211 | if (inode) |
0b2d0724 CH |
1212 | wait_on_inode(inode); |
1213 | return inode; | |
1da177e4 | 1214 | } |
0b2d0724 | 1215 | EXPORT_SYMBOL(ilookup); |
1da177e4 | 1216 | |
261bca86 AV |
1217 | int insert_inode_locked(struct inode *inode) |
1218 | { | |
1219 | struct super_block *sb = inode->i_sb; | |
1220 | ino_t ino = inode->i_ino; | |
1221 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
261bca86 | 1222 | |
261bca86 | 1223 | while (1) { |
72a43d63 AV |
1224 | struct hlist_node *node; |
1225 | struct inode *old = NULL; | |
67a23c49 | 1226 | spin_lock(&inode_hash_lock); |
72a43d63 AV |
1227 | hlist_for_each_entry(old, node, head, i_hash) { |
1228 | if (old->i_ino != ino) | |
1229 | continue; | |
1230 | if (old->i_sb != sb) | |
1231 | continue; | |
250df6ed DC |
1232 | spin_lock(&old->i_lock); |
1233 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1234 | spin_unlock(&old->i_lock); | |
72a43d63 | 1235 | continue; |
250df6ed | 1236 | } |
72a43d63 AV |
1237 | break; |
1238 | } | |
1239 | if (likely(!node)) { | |
250df6ed DC |
1240 | spin_lock(&inode->i_lock); |
1241 | inode->i_state |= I_NEW; | |
261bca86 | 1242 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1243 | spin_unlock(&inode->i_lock); |
67a23c49 | 1244 | spin_unlock(&inode_hash_lock); |
261bca86 AV |
1245 | return 0; |
1246 | } | |
1247 | __iget(old); | |
250df6ed | 1248 | spin_unlock(&old->i_lock); |
67a23c49 | 1249 | spin_unlock(&inode_hash_lock); |
261bca86 | 1250 | wait_on_inode(old); |
1d3382cb | 1251 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1252 | iput(old); |
1253 | return -EBUSY; | |
1254 | } | |
1255 | iput(old); | |
1256 | } | |
1257 | } | |
261bca86 AV |
1258 | EXPORT_SYMBOL(insert_inode_locked); |
1259 | ||
1260 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1261 | int (*test)(struct inode *, void *), void *data) | |
1262 | { | |
1263 | struct super_block *sb = inode->i_sb; | |
1264 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
261bca86 | 1265 | |
261bca86 | 1266 | while (1) { |
72a43d63 AV |
1267 | struct hlist_node *node; |
1268 | struct inode *old = NULL; | |
1269 | ||
67a23c49 | 1270 | spin_lock(&inode_hash_lock); |
72a43d63 AV |
1271 | hlist_for_each_entry(old, node, head, i_hash) { |
1272 | if (old->i_sb != sb) | |
1273 | continue; | |
1274 | if (!test(old, data)) | |
1275 | continue; | |
250df6ed DC |
1276 | spin_lock(&old->i_lock); |
1277 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1278 | spin_unlock(&old->i_lock); | |
72a43d63 | 1279 | continue; |
250df6ed | 1280 | } |
72a43d63 AV |
1281 | break; |
1282 | } | |
1283 | if (likely(!node)) { | |
250df6ed DC |
1284 | spin_lock(&inode->i_lock); |
1285 | inode->i_state |= I_NEW; | |
261bca86 | 1286 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1287 | spin_unlock(&inode->i_lock); |
67a23c49 | 1288 | spin_unlock(&inode_hash_lock); |
261bca86 AV |
1289 | return 0; |
1290 | } | |
1291 | __iget(old); | |
250df6ed | 1292 | spin_unlock(&old->i_lock); |
67a23c49 | 1293 | spin_unlock(&inode_hash_lock); |
261bca86 | 1294 | wait_on_inode(old); |
1d3382cb | 1295 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1296 | iput(old); |
1297 | return -EBUSY; | |
1298 | } | |
1299 | iput(old); | |
1300 | } | |
1301 | } | |
261bca86 AV |
1302 | EXPORT_SYMBOL(insert_inode_locked4); |
1303 | ||
1da177e4 | 1304 | |
45321ac5 AV |
1305 | int generic_delete_inode(struct inode *inode) |
1306 | { | |
1307 | return 1; | |
1308 | } | |
1309 | EXPORT_SYMBOL(generic_delete_inode); | |
1310 | ||
1da177e4 | 1311 | /* |
45321ac5 AV |
1312 | * Normal UNIX filesystem behaviour: delete the |
1313 | * inode when the usage count drops to zero, and | |
1314 | * i_nlink is zero. | |
1da177e4 | 1315 | */ |
45321ac5 | 1316 | int generic_drop_inode(struct inode *inode) |
1da177e4 | 1317 | { |
1d3382cb | 1318 | return !inode->i_nlink || inode_unhashed(inode); |
1da177e4 | 1319 | } |
45321ac5 | 1320 | EXPORT_SYMBOL_GPL(generic_drop_inode); |
1da177e4 | 1321 | |
45321ac5 AV |
1322 | /* |
1323 | * Called when we're dropping the last reference | |
1324 | * to an inode. | |
22fe4042 | 1325 | * |
45321ac5 AV |
1326 | * Call the FS "drop_inode()" function, defaulting to |
1327 | * the legacy UNIX filesystem behaviour. If it tells | |
1328 | * us to evict inode, do so. Otherwise, retain inode | |
1329 | * in cache if fs is alive, sync and evict if fs is | |
1330 | * shutting down. | |
22fe4042 | 1331 | */ |
45321ac5 | 1332 | static void iput_final(struct inode *inode) |
1da177e4 LT |
1333 | { |
1334 | struct super_block *sb = inode->i_sb; | |
45321ac5 AV |
1335 | const struct super_operations *op = inode->i_sb->s_op; |
1336 | int drop; | |
1337 | ||
250df6ed DC |
1338 | WARN_ON(inode->i_state & I_NEW); |
1339 | ||
e7f59097 | 1340 | if (op->drop_inode) |
45321ac5 AV |
1341 | drop = op->drop_inode(inode); |
1342 | else | |
1343 | drop = generic_drop_inode(inode); | |
1da177e4 | 1344 | |
b2b2af8e DC |
1345 | if (!drop && (sb->s_flags & MS_ACTIVE)) { |
1346 | inode->i_state |= I_REFERENCED; | |
1347 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) | |
1348 | inode_lru_list_add(inode); | |
1349 | spin_unlock(&inode->i_lock); | |
b2b2af8e DC |
1350 | return; |
1351 | } | |
1352 | ||
45321ac5 | 1353 | if (!drop) { |
991114c6 | 1354 | inode->i_state |= I_WILL_FREE; |
250df6ed | 1355 | spin_unlock(&inode->i_lock); |
1da177e4 | 1356 | write_inode_now(inode, 1); |
250df6ed | 1357 | spin_lock(&inode->i_lock); |
7ef0d737 | 1358 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1359 | inode->i_state &= ~I_WILL_FREE; |
1da177e4 | 1360 | } |
7ccf19a8 | 1361 | |
991114c6 | 1362 | inode->i_state |= I_FREEING; |
9e38d86f | 1363 | inode_lru_list_del(inode); |
b2b2af8e | 1364 | spin_unlock(&inode->i_lock); |
b2b2af8e | 1365 | |
644da596 | 1366 | evict(inode); |
1da177e4 LT |
1367 | } |
1368 | ||
1da177e4 | 1369 | /** |
6b3304b5 | 1370 | * iput - put an inode |
1da177e4 LT |
1371 | * @inode: inode to put |
1372 | * | |
1373 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1374 | * zero, the inode is then freed and may also be destroyed. | |
1375 | * | |
1376 | * Consequently, iput() can sleep. | |
1377 | */ | |
1378 | void iput(struct inode *inode) | |
1379 | { | |
1380 | if (inode) { | |
a4ffdde6 | 1381 | BUG_ON(inode->i_state & I_CLEAR); |
1da177e4 | 1382 | |
f283c86a | 1383 | if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) |
1da177e4 LT |
1384 | iput_final(inode); |
1385 | } | |
1386 | } | |
1da177e4 LT |
1387 | EXPORT_SYMBOL(iput); |
1388 | ||
1389 | /** | |
1390 | * bmap - find a block number in a file | |
1391 | * @inode: inode of file | |
1392 | * @block: block to find | |
1393 | * | |
1394 | * Returns the block number on the device holding the inode that | |
1395 | * is the disk block number for the block of the file requested. | |
1396 | * That is, asked for block 4 of inode 1 the function will return the | |
6b3304b5 | 1397 | * disk block relative to the disk start that holds that block of the |
1da177e4 LT |
1398 | * file. |
1399 | */ | |
6b3304b5 | 1400 | sector_t bmap(struct inode *inode, sector_t block) |
1da177e4 LT |
1401 | { |
1402 | sector_t res = 0; | |
1403 | if (inode->i_mapping->a_ops->bmap) | |
1404 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1405 | return res; | |
1406 | } | |
1da177e4 LT |
1407 | EXPORT_SYMBOL(bmap); |
1408 | ||
11ff6f05 MG |
1409 | /* |
1410 | * With relative atime, only update atime if the previous atime is | |
1411 | * earlier than either the ctime or mtime or if at least a day has | |
1412 | * passed since the last atime update. | |
1413 | */ | |
1414 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1415 | struct timespec now) | |
1416 | { | |
1417 | ||
1418 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1419 | return 1; | |
1420 | /* | |
1421 | * Is mtime younger than atime? If yes, update atime: | |
1422 | */ | |
1423 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1424 | return 1; | |
1425 | /* | |
1426 | * Is ctime younger than atime? If yes, update atime: | |
1427 | */ | |
1428 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1429 | return 1; | |
1430 | ||
1431 | /* | |
1432 | * Is the previous atime value older than a day? If yes, | |
1433 | * update atime: | |
1434 | */ | |
1435 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1436 | return 1; | |
1437 | /* | |
1438 | * Good, we can skip the atime update: | |
1439 | */ | |
1440 | return 0; | |
1441 | } | |
1442 | ||
1da177e4 | 1443 | /** |
869243a0 CH |
1444 | * touch_atime - update the access time |
1445 | * @mnt: mount the inode is accessed on | |
7045f37b | 1446 | * @dentry: dentry accessed |
1da177e4 LT |
1447 | * |
1448 | * Update the accessed time on an inode and mark it for writeback. | |
1449 | * This function automatically handles read only file systems and media, | |
1450 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1451 | */ | |
869243a0 | 1452 | void touch_atime(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 1453 | { |
869243a0 | 1454 | struct inode *inode = dentry->d_inode; |
1da177e4 LT |
1455 | struct timespec now; |
1456 | ||
cdb70f3f | 1457 | if (inode->i_flags & S_NOATIME) |
b12536c2 | 1458 | return; |
37756ced | 1459 | if (IS_NOATIME(inode)) |
b12536c2 | 1460 | return; |
b2276138 | 1461 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1462 | return; |
47ae32d6 | 1463 | |
cdb70f3f | 1464 | if (mnt->mnt_flags & MNT_NOATIME) |
b12536c2 | 1465 | return; |
cdb70f3f | 1466 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1467 | return; |
1da177e4 LT |
1468 | |
1469 | now = current_fs_time(inode->i_sb); | |
11ff6f05 MG |
1470 | |
1471 | if (!relatime_need_update(mnt, inode, now)) | |
b12536c2 | 1472 | return; |
11ff6f05 | 1473 | |
47ae32d6 | 1474 | if (timespec_equal(&inode->i_atime, &now)) |
b12536c2 AK |
1475 | return; |
1476 | ||
1477 | if (mnt_want_write(mnt)) | |
1478 | return; | |
47ae32d6 VH |
1479 | |
1480 | inode->i_atime = now; | |
1481 | mark_inode_dirty_sync(inode); | |
cdb70f3f | 1482 | mnt_drop_write(mnt); |
1da177e4 | 1483 | } |
869243a0 | 1484 | EXPORT_SYMBOL(touch_atime); |
1da177e4 LT |
1485 | |
1486 | /** | |
870f4817 CH |
1487 | * file_update_time - update mtime and ctime time |
1488 | * @file: file accessed | |
1da177e4 | 1489 | * |
870f4817 CH |
1490 | * Update the mtime and ctime members of an inode and mark the inode |
1491 | * for writeback. Note that this function is meant exclusively for | |
1492 | * usage in the file write path of filesystems, and filesystems may | |
1493 | * choose to explicitly ignore update via this function with the | |
2eadfc0e | 1494 | * S_NOCMTIME inode flag, e.g. for network filesystem where these |
870f4817 | 1495 | * timestamps are handled by the server. |
1da177e4 LT |
1496 | */ |
1497 | ||
870f4817 | 1498 | void file_update_time(struct file *file) |
1da177e4 | 1499 | { |
0f7fc9e4 | 1500 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 | 1501 | struct timespec now; |
ce06e0b2 | 1502 | enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0; |
1da177e4 | 1503 | |
ce06e0b2 | 1504 | /* First try to exhaust all avenues to not sync */ |
1da177e4 LT |
1505 | if (IS_NOCMTIME(inode)) |
1506 | return; | |
20ddee2c | 1507 | |
1da177e4 | 1508 | now = current_fs_time(inode->i_sb); |
ce06e0b2 AK |
1509 | if (!timespec_equal(&inode->i_mtime, &now)) |
1510 | sync_it = S_MTIME; | |
1da177e4 | 1511 | |
ce06e0b2 AK |
1512 | if (!timespec_equal(&inode->i_ctime, &now)) |
1513 | sync_it |= S_CTIME; | |
870f4817 | 1514 | |
ce06e0b2 AK |
1515 | if (IS_I_VERSION(inode)) |
1516 | sync_it |= S_VERSION; | |
7a224228 | 1517 | |
ce06e0b2 AK |
1518 | if (!sync_it) |
1519 | return; | |
1520 | ||
1521 | /* Finally allowed to write? Takes lock. */ | |
1522 | if (mnt_want_write_file(file)) | |
1523 | return; | |
1524 | ||
1525 | /* Only change inode inside the lock region */ | |
1526 | if (sync_it & S_VERSION) | |
1527 | inode_inc_iversion(inode); | |
1528 | if (sync_it & S_CTIME) | |
1529 | inode->i_ctime = now; | |
1530 | if (sync_it & S_MTIME) | |
1531 | inode->i_mtime = now; | |
1532 | mark_inode_dirty_sync(inode); | |
20ddee2c | 1533 | mnt_drop_write(file->f_path.mnt); |
1da177e4 | 1534 | } |
870f4817 | 1535 | EXPORT_SYMBOL(file_update_time); |
1da177e4 LT |
1536 | |
1537 | int inode_needs_sync(struct inode *inode) | |
1538 | { | |
1539 | if (IS_SYNC(inode)) | |
1540 | return 1; | |
1541 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1542 | return 1; | |
1543 | return 0; | |
1544 | } | |
1da177e4 LT |
1545 | EXPORT_SYMBOL(inode_needs_sync); |
1546 | ||
1da177e4 LT |
1547 | int inode_wait(void *word) |
1548 | { | |
1549 | schedule(); | |
1550 | return 0; | |
1551 | } | |
d44dab8d | 1552 | EXPORT_SYMBOL(inode_wait); |
1da177e4 LT |
1553 | |
1554 | /* | |
168a9fd6 MS |
1555 | * If we try to find an inode in the inode hash while it is being |
1556 | * deleted, we have to wait until the filesystem completes its | |
1557 | * deletion before reporting that it isn't found. This function waits | |
1558 | * until the deletion _might_ have completed. Callers are responsible | |
1559 | * to recheck inode state. | |
1560 | * | |
eaff8079 | 1561 | * It doesn't matter if I_NEW is not set initially, a call to |
250df6ed DC |
1562 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list |
1563 | * will DTRT. | |
1da177e4 LT |
1564 | */ |
1565 | static void __wait_on_freeing_inode(struct inode *inode) | |
1566 | { | |
1567 | wait_queue_head_t *wq; | |
eaff8079 CH |
1568 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); |
1569 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1da177e4 | 1570 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
250df6ed | 1571 | spin_unlock(&inode->i_lock); |
67a23c49 | 1572 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1573 | schedule(); |
1574 | finish_wait(wq, &wait.wait); | |
67a23c49 | 1575 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1576 | } |
1577 | ||
1da177e4 LT |
1578 | static __initdata unsigned long ihash_entries; |
1579 | static int __init set_ihash_entries(char *str) | |
1580 | { | |
1581 | if (!str) | |
1582 | return 0; | |
1583 | ihash_entries = simple_strtoul(str, &str, 0); | |
1584 | return 1; | |
1585 | } | |
1586 | __setup("ihash_entries=", set_ihash_entries); | |
1587 | ||
1588 | /* | |
1589 | * Initialize the waitqueues and inode hash table. | |
1590 | */ | |
1591 | void __init inode_init_early(void) | |
1592 | { | |
1593 | int loop; | |
1594 | ||
1595 | /* If hashes are distributed across NUMA nodes, defer | |
1596 | * hash allocation until vmalloc space is available. | |
1597 | */ | |
1598 | if (hashdist) | |
1599 | return; | |
1600 | ||
1601 | inode_hashtable = | |
1602 | alloc_large_system_hash("Inode-cache", | |
1603 | sizeof(struct hlist_head), | |
1604 | ihash_entries, | |
1605 | 14, | |
1606 | HASH_EARLY, | |
1607 | &i_hash_shift, | |
1608 | &i_hash_mask, | |
1609 | 0); | |
1610 | ||
1611 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1612 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1613 | } | |
1614 | ||
74bf17cf | 1615 | void __init inode_init(void) |
1da177e4 LT |
1616 | { |
1617 | int loop; | |
1618 | ||
1619 | /* inode slab cache */ | |
b0196009 PJ |
1620 | inode_cachep = kmem_cache_create("inode_cache", |
1621 | sizeof(struct inode), | |
1622 | 0, | |
1623 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1624 | SLAB_MEM_SPREAD), | |
20c2df83 | 1625 | init_once); |
8e1f936b | 1626 | register_shrinker(&icache_shrinker); |
1da177e4 LT |
1627 | |
1628 | /* Hash may have been set up in inode_init_early */ | |
1629 | if (!hashdist) | |
1630 | return; | |
1631 | ||
1632 | inode_hashtable = | |
1633 | alloc_large_system_hash("Inode-cache", | |
1634 | sizeof(struct hlist_head), | |
1635 | ihash_entries, | |
1636 | 14, | |
1637 | 0, | |
1638 | &i_hash_shift, | |
1639 | &i_hash_mask, | |
1640 | 0); | |
1641 | ||
1642 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1643 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1644 | } | |
1645 | ||
1646 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1647 | { | |
1648 | inode->i_mode = mode; | |
1649 | if (S_ISCHR(mode)) { | |
1650 | inode->i_fop = &def_chr_fops; | |
1651 | inode->i_rdev = rdev; | |
1652 | } else if (S_ISBLK(mode)) { | |
1653 | inode->i_fop = &def_blk_fops; | |
1654 | inode->i_rdev = rdev; | |
1655 | } else if (S_ISFIFO(mode)) | |
1656 | inode->i_fop = &def_fifo_fops; | |
1657 | else if (S_ISSOCK(mode)) | |
1658 | inode->i_fop = &bad_sock_fops; | |
1659 | else | |
af0d9ae8 MK |
1660 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" |
1661 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1662 | inode->i_ino); | |
1da177e4 LT |
1663 | } |
1664 | EXPORT_SYMBOL(init_special_inode); | |
a1bd120d DM |
1665 | |
1666 | /** | |
eaae668d | 1667 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards |
a1bd120d DM |
1668 | * @inode: New inode |
1669 | * @dir: Directory inode | |
1670 | * @mode: mode of the new inode | |
1671 | */ | |
1672 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
1673 | mode_t mode) | |
1674 | { | |
1675 | inode->i_uid = current_fsuid(); | |
1676 | if (dir && dir->i_mode & S_ISGID) { | |
1677 | inode->i_gid = dir->i_gid; | |
1678 | if (S_ISDIR(mode)) | |
1679 | mode |= S_ISGID; | |
1680 | } else | |
1681 | inode->i_gid = current_fsgid(); | |
1682 | inode->i_mode = mode; | |
1683 | } | |
1684 | EXPORT_SYMBOL(inode_init_owner); | |
e795b717 | 1685 | |
2e149670 SH |
1686 | /** |
1687 | * inode_owner_or_capable - check current task permissions to inode | |
1688 | * @inode: inode being checked | |
1689 | * | |
1690 | * Return true if current either has CAP_FOWNER to the inode, or | |
1691 | * owns the file. | |
e795b717 | 1692 | */ |
2e149670 | 1693 | bool inode_owner_or_capable(const struct inode *inode) |
e795b717 SH |
1694 | { |
1695 | struct user_namespace *ns = inode_userns(inode); | |
1696 | ||
1697 | if (current_user_ns() == ns && current_fsuid() == inode->i_uid) | |
1698 | return true; | |
1699 | if (ns_capable(ns, CAP_FOWNER)) | |
1700 | return true; | |
1701 | return false; | |
1702 | } | |
2e149670 | 1703 | EXPORT_SYMBOL(inode_owner_or_capable); |