]>
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 | */ |
e59cc473 | 5 | #include <linux/export.h> |
1da177e4 LT |
6 | #include <linux/fs.h> |
7 | #include <linux/mm.h> | |
1da177e4 | 8 | #include <linux/backing-dev.h> |
1da177e4 LT |
9 | #include <linux/hash.h> |
10 | #include <linux/swap.h> | |
11 | #include <linux/security.h> | |
1da177e4 LT |
12 | #include <linux/cdev.h> |
13 | #include <linux/bootmem.h> | |
3be25f49 | 14 | #include <linux/fsnotify.h> |
fc33a7bb | 15 | #include <linux/mount.h> |
f19d4a8f | 16 | #include <linux/posix_acl.h> |
9ce6e0be | 17 | #include <linux/prefetch.h> |
4b4563dc | 18 | #include <linux/buffer_head.h> /* for inode_has_buffers */ |
7ada4db8 | 19 | #include <linux/ratelimit.h> |
bc3b14cb | 20 | #include <linux/list_lru.h> |
a66979ab | 21 | #include "internal.h" |
1da177e4 | 22 | |
250df6ed | 23 | /* |
4b4563dc | 24 | * Inode locking rules: |
250df6ed DC |
25 | * |
26 | * inode->i_lock protects: | |
27 | * inode->i_state, inode->i_hash, __iget() | |
bc3b14cb | 28 | * Inode LRU list locks protect: |
98b745c6 | 29 | * inode->i_sb->s_inode_lru, inode->i_lru |
55fa6091 DC |
30 | * inode_sb_list_lock protects: |
31 | * sb->s_inodes, inode->i_sb_list | |
f758eeab | 32 | * bdi->wb.list_lock protects: |
a66979ab | 33 | * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list |
67a23c49 DC |
34 | * inode_hash_lock protects: |
35 | * inode_hashtable, inode->i_hash | |
250df6ed DC |
36 | * |
37 | * Lock ordering: | |
55fa6091 DC |
38 | * |
39 | * inode_sb_list_lock | |
40 | * inode->i_lock | |
bc3b14cb | 41 | * Inode LRU list locks |
a66979ab | 42 | * |
f758eeab | 43 | * bdi->wb.list_lock |
a66979ab | 44 | * inode->i_lock |
67a23c49 DC |
45 | * |
46 | * inode_hash_lock | |
47 | * inode_sb_list_lock | |
48 | * inode->i_lock | |
49 | * | |
50 | * iunique_lock | |
51 | * inode_hash_lock | |
250df6ed DC |
52 | */ |
53 | ||
fa3536cc ED |
54 | static unsigned int i_hash_mask __read_mostly; |
55 | static unsigned int i_hash_shift __read_mostly; | |
67a23c49 DC |
56 | static struct hlist_head *inode_hashtable __read_mostly; |
57 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock); | |
1da177e4 | 58 | |
55fa6091 DC |
59 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock); |
60 | ||
7dcda1c9 JA |
61 | /* |
62 | * Empty aops. Can be used for the cases where the user does not | |
63 | * define any of the address_space operations. | |
64 | */ | |
65 | const struct address_space_operations empty_aops = { | |
66 | }; | |
67 | EXPORT_SYMBOL(empty_aops); | |
68 | ||
1da177e4 LT |
69 | /* |
70 | * Statistics gathering.. | |
71 | */ | |
72 | struct inodes_stat_t inodes_stat; | |
73 | ||
3942c07c GC |
74 | static DEFINE_PER_CPU(unsigned long, nr_inodes); |
75 | static DEFINE_PER_CPU(unsigned long, nr_unused); | |
cffbc8aa | 76 | |
6b3304b5 | 77 | static struct kmem_cache *inode_cachep __read_mostly; |
1da177e4 | 78 | |
3942c07c | 79 | static long get_nr_inodes(void) |
cffbc8aa | 80 | { |
3e880fb5 | 81 | int i; |
3942c07c | 82 | long sum = 0; |
3e880fb5 NP |
83 | for_each_possible_cpu(i) |
84 | sum += per_cpu(nr_inodes, i); | |
85 | return sum < 0 ? 0 : sum; | |
cffbc8aa DC |
86 | } |
87 | ||
3942c07c | 88 | static inline long get_nr_inodes_unused(void) |
cffbc8aa | 89 | { |
fcb94f72 | 90 | int i; |
3942c07c | 91 | long sum = 0; |
fcb94f72 DC |
92 | for_each_possible_cpu(i) |
93 | sum += per_cpu(nr_unused, i); | |
94 | return sum < 0 ? 0 : sum; | |
cffbc8aa DC |
95 | } |
96 | ||
3942c07c | 97 | long get_nr_dirty_inodes(void) |
cffbc8aa | 98 | { |
3e880fb5 | 99 | /* not actually dirty inodes, but a wild approximation */ |
3942c07c | 100 | long nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); |
cffbc8aa | 101 | return nr_dirty > 0 ? nr_dirty : 0; |
cffbc8aa DC |
102 | } |
103 | ||
104 | /* | |
105 | * Handle nr_inode sysctl | |
106 | */ | |
107 | #ifdef CONFIG_SYSCTL | |
108 | int proc_nr_inodes(ctl_table *table, int write, | |
109 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
110 | { | |
111 | inodes_stat.nr_inodes = get_nr_inodes(); | |
fcb94f72 | 112 | inodes_stat.nr_unused = get_nr_inodes_unused(); |
3942c07c | 113 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
cffbc8aa DC |
114 | } |
115 | #endif | |
116 | ||
2cb1599f DC |
117 | /** |
118 | * inode_init_always - perform inode structure intialisation | |
0bc02f3f RD |
119 | * @sb: superblock inode belongs to |
120 | * @inode: inode to initialise | |
2cb1599f DC |
121 | * |
122 | * These are initializations that need to be done on every inode | |
123 | * allocation as the fields are not initialised by slab allocation. | |
124 | */ | |
54e34621 | 125 | int inode_init_always(struct super_block *sb, struct inode *inode) |
1da177e4 | 126 | { |
6e1d5dcc | 127 | static const struct inode_operations empty_iops; |
99ac48f5 | 128 | static const struct file_operations empty_fops; |
6b3304b5 | 129 | struct address_space *const mapping = &inode->i_data; |
2cb1599f DC |
130 | |
131 | inode->i_sb = sb; | |
132 | inode->i_blkbits = sb->s_blocksize_bits; | |
133 | inode->i_flags = 0; | |
134 | atomic_set(&inode->i_count, 1); | |
135 | inode->i_op = &empty_iops; | |
136 | inode->i_fop = &empty_fops; | |
a78ef704 | 137 | inode->__i_nlink = 1; |
3ddcd056 | 138 | inode->i_opflags = 0; |
92361636 EB |
139 | i_uid_write(inode, 0); |
140 | i_gid_write(inode, 0); | |
2cb1599f DC |
141 | atomic_set(&inode->i_writecount, 0); |
142 | inode->i_size = 0; | |
143 | inode->i_blocks = 0; | |
144 | inode->i_bytes = 0; | |
145 | inode->i_generation = 0; | |
1da177e4 | 146 | #ifdef CONFIG_QUOTA |
2cb1599f | 147 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); |
1da177e4 | 148 | #endif |
2cb1599f DC |
149 | inode->i_pipe = NULL; |
150 | inode->i_bdev = NULL; | |
151 | inode->i_cdev = NULL; | |
152 | inode->i_rdev = 0; | |
153 | inode->dirtied_when = 0; | |
6146f0d5 MZ |
154 | |
155 | if (security_inode_alloc(inode)) | |
54e34621 | 156 | goto out; |
2cb1599f DC |
157 | spin_lock_init(&inode->i_lock); |
158 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
159 | ||
160 | mutex_init(&inode->i_mutex); | |
161 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); | |
162 | ||
bd5fe6c5 | 163 | atomic_set(&inode->i_dio_count, 0); |
2cb1599f DC |
164 | |
165 | mapping->a_ops = &empty_aops; | |
166 | mapping->host = inode; | |
167 | mapping->flags = 0; | |
3c1d4378 | 168 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); |
252aa6f5 | 169 | mapping->private_data = NULL; |
2cb1599f DC |
170 | mapping->backing_dev_info = &default_backing_dev_info; |
171 | mapping->writeback_index = 0; | |
172 | ||
173 | /* | |
174 | * If the block_device provides a backing_dev_info for client | |
175 | * inodes then use that. Otherwise the inode share the bdev's | |
176 | * backing_dev_info. | |
177 | */ | |
178 | if (sb->s_bdev) { | |
179 | struct backing_dev_info *bdi; | |
180 | ||
2c96ce9f | 181 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; |
2cb1599f DC |
182 | mapping->backing_dev_info = bdi; |
183 | } | |
184 | inode->i_private = NULL; | |
185 | inode->i_mapping = mapping; | |
b3d9b7a3 | 186 | INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */ |
f19d4a8f AV |
187 | #ifdef CONFIG_FS_POSIX_ACL |
188 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; | |
189 | #endif | |
2cb1599f | 190 | |
3be25f49 EP |
191 | #ifdef CONFIG_FSNOTIFY |
192 | inode->i_fsnotify_mask = 0; | |
193 | #endif | |
194 | ||
3e880fb5 | 195 | this_cpu_inc(nr_inodes); |
cffbc8aa | 196 | |
54e34621 | 197 | return 0; |
54e34621 CH |
198 | out: |
199 | return -ENOMEM; | |
1da177e4 | 200 | } |
2cb1599f DC |
201 | EXPORT_SYMBOL(inode_init_always); |
202 | ||
203 | static struct inode *alloc_inode(struct super_block *sb) | |
204 | { | |
205 | struct inode *inode; | |
206 | ||
207 | if (sb->s_op->alloc_inode) | |
208 | inode = sb->s_op->alloc_inode(sb); | |
209 | else | |
210 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
211 | ||
54e34621 CH |
212 | if (!inode) |
213 | return NULL; | |
214 | ||
215 | if (unlikely(inode_init_always(sb, inode))) { | |
216 | if (inode->i_sb->s_op->destroy_inode) | |
217 | inode->i_sb->s_op->destroy_inode(inode); | |
218 | else | |
219 | kmem_cache_free(inode_cachep, inode); | |
220 | return NULL; | |
221 | } | |
222 | ||
223 | return inode; | |
2cb1599f | 224 | } |
1da177e4 | 225 | |
ff0c7d15 NP |
226 | void free_inode_nonrcu(struct inode *inode) |
227 | { | |
228 | kmem_cache_free(inode_cachep, inode); | |
229 | } | |
230 | EXPORT_SYMBOL(free_inode_nonrcu); | |
231 | ||
2e00c97e | 232 | void __destroy_inode(struct inode *inode) |
1da177e4 | 233 | { |
b7542f8c | 234 | BUG_ON(inode_has_buffers(inode)); |
1da177e4 | 235 | security_inode_free(inode); |
3be25f49 | 236 | fsnotify_inode_delete(inode); |
7ada4db8 MS |
237 | if (!inode->i_nlink) { |
238 | WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0); | |
239 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
240 | } | |
241 | ||
f19d4a8f AV |
242 | #ifdef CONFIG_FS_POSIX_ACL |
243 | if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) | |
244 | posix_acl_release(inode->i_acl); | |
245 | if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) | |
246 | posix_acl_release(inode->i_default_acl); | |
247 | #endif | |
3e880fb5 | 248 | this_cpu_dec(nr_inodes); |
2e00c97e CH |
249 | } |
250 | EXPORT_SYMBOL(__destroy_inode); | |
251 | ||
fa0d7e3d NP |
252 | static void i_callback(struct rcu_head *head) |
253 | { | |
254 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
255 | kmem_cache_free(inode_cachep, inode); |
256 | } | |
257 | ||
56b0dacf | 258 | static void destroy_inode(struct inode *inode) |
2e00c97e | 259 | { |
7ccf19a8 | 260 | BUG_ON(!list_empty(&inode->i_lru)); |
2e00c97e | 261 | __destroy_inode(inode); |
1da177e4 LT |
262 | if (inode->i_sb->s_op->destroy_inode) |
263 | inode->i_sb->s_op->destroy_inode(inode); | |
264 | else | |
fa0d7e3d | 265 | call_rcu(&inode->i_rcu, i_callback); |
1da177e4 | 266 | } |
1da177e4 | 267 | |
7ada4db8 MS |
268 | /** |
269 | * drop_nlink - directly drop an inode's link count | |
270 | * @inode: inode | |
271 | * | |
272 | * This is a low-level filesystem helper to replace any | |
273 | * direct filesystem manipulation of i_nlink. In cases | |
274 | * where we are attempting to track writes to the | |
275 | * filesystem, a decrement to zero means an imminent | |
276 | * write when the file is truncated and actually unlinked | |
277 | * on the filesystem. | |
278 | */ | |
279 | void drop_nlink(struct inode *inode) | |
280 | { | |
281 | WARN_ON(inode->i_nlink == 0); | |
282 | inode->__i_nlink--; | |
283 | if (!inode->i_nlink) | |
284 | atomic_long_inc(&inode->i_sb->s_remove_count); | |
285 | } | |
286 | EXPORT_SYMBOL(drop_nlink); | |
287 | ||
288 | /** | |
289 | * clear_nlink - directly zero an inode's link count | |
290 | * @inode: inode | |
291 | * | |
292 | * This is a low-level filesystem helper to replace any | |
293 | * direct filesystem manipulation of i_nlink. See | |
294 | * drop_nlink() for why we care about i_nlink hitting zero. | |
295 | */ | |
296 | void clear_nlink(struct inode *inode) | |
297 | { | |
298 | if (inode->i_nlink) { | |
299 | inode->__i_nlink = 0; | |
300 | atomic_long_inc(&inode->i_sb->s_remove_count); | |
301 | } | |
302 | } | |
303 | EXPORT_SYMBOL(clear_nlink); | |
304 | ||
305 | /** | |
306 | * set_nlink - directly set an inode's link count | |
307 | * @inode: inode | |
308 | * @nlink: new nlink (should be non-zero) | |
309 | * | |
310 | * This is a low-level filesystem helper to replace any | |
311 | * direct filesystem manipulation of i_nlink. | |
312 | */ | |
313 | void set_nlink(struct inode *inode, unsigned int nlink) | |
314 | { | |
315 | if (!nlink) { | |
7ada4db8 MS |
316 | clear_nlink(inode); |
317 | } else { | |
318 | /* Yes, some filesystems do change nlink from zero to one */ | |
319 | if (inode->i_nlink == 0) | |
320 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
321 | ||
322 | inode->__i_nlink = nlink; | |
323 | } | |
324 | } | |
325 | EXPORT_SYMBOL(set_nlink); | |
326 | ||
327 | /** | |
328 | * inc_nlink - directly increment an inode's link count | |
329 | * @inode: inode | |
330 | * | |
331 | * This is a low-level filesystem helper to replace any | |
332 | * direct filesystem manipulation of i_nlink. Currently, | |
333 | * it is only here for parity with dec_nlink(). | |
334 | */ | |
335 | void inc_nlink(struct inode *inode) | |
336 | { | |
f4e0c30c AV |
337 | if (unlikely(inode->i_nlink == 0)) { |
338 | WARN_ON(!(inode->i_state & I_LINKABLE)); | |
7ada4db8 | 339 | atomic_long_dec(&inode->i_sb->s_remove_count); |
f4e0c30c | 340 | } |
7ada4db8 MS |
341 | |
342 | inode->__i_nlink++; | |
343 | } | |
344 | EXPORT_SYMBOL(inc_nlink); | |
345 | ||
2aa15890 MS |
346 | void address_space_init_once(struct address_space *mapping) |
347 | { | |
348 | memset(mapping, 0, sizeof(*mapping)); | |
349 | INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC); | |
350 | spin_lock_init(&mapping->tree_lock); | |
3d48ae45 | 351 | mutex_init(&mapping->i_mmap_mutex); |
2aa15890 MS |
352 | INIT_LIST_HEAD(&mapping->private_list); |
353 | spin_lock_init(&mapping->private_lock); | |
6b2dbba8 | 354 | mapping->i_mmap = RB_ROOT; |
2aa15890 | 355 | INIT_LIST_HEAD(&mapping->i_mmap_nonlinear); |
2aa15890 MS |
356 | } |
357 | EXPORT_SYMBOL(address_space_init_once); | |
358 | ||
1da177e4 LT |
359 | /* |
360 | * These are initializations that only need to be done | |
361 | * once, because the fields are idempotent across use | |
362 | * of the inode, so let the slab aware of that. | |
363 | */ | |
364 | void inode_init_once(struct inode *inode) | |
365 | { | |
366 | memset(inode, 0, sizeof(*inode)); | |
367 | INIT_HLIST_NODE(&inode->i_hash); | |
1da177e4 | 368 | INIT_LIST_HEAD(&inode->i_devices); |
7ccf19a8 NP |
369 | INIT_LIST_HEAD(&inode->i_wb_list); |
370 | INIT_LIST_HEAD(&inode->i_lru); | |
2aa15890 | 371 | address_space_init_once(&inode->i_data); |
1da177e4 | 372 | i_size_ordered_init(inode); |
3be25f49 | 373 | #ifdef CONFIG_FSNOTIFY |
e61ce867 | 374 | INIT_HLIST_HEAD(&inode->i_fsnotify_marks); |
3be25f49 | 375 | #endif |
1da177e4 | 376 | } |
1da177e4 LT |
377 | EXPORT_SYMBOL(inode_init_once); |
378 | ||
51cc5068 | 379 | static void init_once(void *foo) |
1da177e4 | 380 | { |
6b3304b5 | 381 | struct inode *inode = (struct inode *) foo; |
1da177e4 | 382 | |
a35afb83 | 383 | inode_init_once(inode); |
1da177e4 LT |
384 | } |
385 | ||
386 | /* | |
250df6ed | 387 | * inode->i_lock must be held |
1da177e4 | 388 | */ |
6b3304b5 | 389 | void __iget(struct inode *inode) |
1da177e4 | 390 | { |
9e38d86f NP |
391 | atomic_inc(&inode->i_count); |
392 | } | |
2e147f1e | 393 | |
7de9c6ee AV |
394 | /* |
395 | * get additional reference to inode; caller must already hold one. | |
396 | */ | |
397 | void ihold(struct inode *inode) | |
398 | { | |
399 | WARN_ON(atomic_inc_return(&inode->i_count) < 2); | |
400 | } | |
401 | EXPORT_SYMBOL(ihold); | |
402 | ||
9e38d86f NP |
403 | static void inode_lru_list_add(struct inode *inode) |
404 | { | |
bc3b14cb | 405 | if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru)) |
fcb94f72 | 406 | this_cpu_inc(nr_unused); |
9e38d86f | 407 | } |
2e147f1e | 408 | |
4eff96dd JK |
409 | /* |
410 | * Add inode to LRU if needed (inode is unused and clean). | |
411 | * | |
412 | * Needs inode->i_lock held. | |
413 | */ | |
414 | void inode_add_lru(struct inode *inode) | |
415 | { | |
416 | if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) && | |
417 | !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE) | |
418 | inode_lru_list_add(inode); | |
419 | } | |
420 | ||
421 | ||
9e38d86f NP |
422 | static void inode_lru_list_del(struct inode *inode) |
423 | { | |
bc3b14cb DC |
424 | |
425 | if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru)) | |
fcb94f72 | 426 | this_cpu_dec(nr_unused); |
1da177e4 LT |
427 | } |
428 | ||
646ec461 CH |
429 | /** |
430 | * inode_sb_list_add - add inode to the superblock list of inodes | |
431 | * @inode: inode to add | |
432 | */ | |
433 | void inode_sb_list_add(struct inode *inode) | |
434 | { | |
55fa6091 DC |
435 | spin_lock(&inode_sb_list_lock); |
436 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); | |
437 | spin_unlock(&inode_sb_list_lock); | |
646ec461 CH |
438 | } |
439 | EXPORT_SYMBOL_GPL(inode_sb_list_add); | |
440 | ||
55fa6091 | 441 | static inline void inode_sb_list_del(struct inode *inode) |
646ec461 | 442 | { |
a209dfc7 ED |
443 | if (!list_empty(&inode->i_sb_list)) { |
444 | spin_lock(&inode_sb_list_lock); | |
445 | list_del_init(&inode->i_sb_list); | |
446 | spin_unlock(&inode_sb_list_lock); | |
447 | } | |
646ec461 CH |
448 | } |
449 | ||
4c51acbc DC |
450 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
451 | { | |
452 | unsigned long tmp; | |
453 | ||
454 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
455 | L1_CACHE_BYTES; | |
4b4563dc CH |
456 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift); |
457 | return tmp & i_hash_mask; | |
4c51acbc DC |
458 | } |
459 | ||
460 | /** | |
461 | * __insert_inode_hash - hash an inode | |
462 | * @inode: unhashed inode | |
463 | * @hashval: unsigned long value used to locate this object in the | |
464 | * inode_hashtable. | |
465 | * | |
466 | * Add an inode to the inode hash for this superblock. | |
467 | */ | |
468 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
469 | { | |
646ec461 CH |
470 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); |
471 | ||
67a23c49 | 472 | spin_lock(&inode_hash_lock); |
250df6ed | 473 | spin_lock(&inode->i_lock); |
646ec461 | 474 | hlist_add_head(&inode->i_hash, b); |
250df6ed | 475 | spin_unlock(&inode->i_lock); |
67a23c49 | 476 | spin_unlock(&inode_hash_lock); |
4c51acbc DC |
477 | } |
478 | EXPORT_SYMBOL(__insert_inode_hash); | |
479 | ||
4c51acbc | 480 | /** |
f2ee7abf | 481 | * __remove_inode_hash - remove an inode from the hash |
4c51acbc DC |
482 | * @inode: inode to unhash |
483 | * | |
484 | * Remove an inode from the superblock. | |
485 | */ | |
f2ee7abf | 486 | void __remove_inode_hash(struct inode *inode) |
4c51acbc | 487 | { |
67a23c49 | 488 | spin_lock(&inode_hash_lock); |
250df6ed | 489 | spin_lock(&inode->i_lock); |
4c51acbc | 490 | hlist_del_init(&inode->i_hash); |
250df6ed | 491 | spin_unlock(&inode->i_lock); |
67a23c49 | 492 | spin_unlock(&inode_hash_lock); |
4c51acbc | 493 | } |
f2ee7abf | 494 | EXPORT_SYMBOL(__remove_inode_hash); |
4c51acbc | 495 | |
dbd5768f | 496 | void clear_inode(struct inode *inode) |
b0683aa6 AV |
497 | { |
498 | might_sleep(); | |
08142579 JK |
499 | /* |
500 | * We have to cycle tree_lock here because reclaim can be still in the | |
501 | * process of removing the last page (in __delete_from_page_cache()) | |
502 | * and we must not free mapping under it. | |
503 | */ | |
504 | spin_lock_irq(&inode->i_data.tree_lock); | |
b0683aa6 | 505 | BUG_ON(inode->i_data.nrpages); |
08142579 | 506 | spin_unlock_irq(&inode->i_data.tree_lock); |
b0683aa6 AV |
507 | BUG_ON(!list_empty(&inode->i_data.private_list)); |
508 | BUG_ON(!(inode->i_state & I_FREEING)); | |
509 | BUG_ON(inode->i_state & I_CLEAR); | |
fa0d7e3d | 510 | /* don't need i_lock here, no concurrent mods to i_state */ |
b0683aa6 AV |
511 | inode->i_state = I_FREEING | I_CLEAR; |
512 | } | |
dbd5768f | 513 | EXPORT_SYMBOL(clear_inode); |
b0683aa6 | 514 | |
b2b2af8e DC |
515 | /* |
516 | * Free the inode passed in, removing it from the lists it is still connected | |
517 | * to. We remove any pages still attached to the inode and wait for any IO that | |
518 | * is still in progress before finally destroying the inode. | |
519 | * | |
520 | * An inode must already be marked I_FREEING so that we avoid the inode being | |
521 | * moved back onto lists if we race with other code that manipulates the lists | |
522 | * (e.g. writeback_single_inode). The caller is responsible for setting this. | |
523 | * | |
524 | * An inode must already be removed from the LRU list before being evicted from | |
525 | * the cache. This should occur atomically with setting the I_FREEING state | |
526 | * flag, so no inodes here should ever be on the LRU when being evicted. | |
527 | */ | |
644da596 | 528 | static void evict(struct inode *inode) |
b4272d4c AV |
529 | { |
530 | const struct super_operations *op = inode->i_sb->s_op; | |
531 | ||
b2b2af8e DC |
532 | BUG_ON(!(inode->i_state & I_FREEING)); |
533 | BUG_ON(!list_empty(&inode->i_lru)); | |
534 | ||
b12362bd ED |
535 | if (!list_empty(&inode->i_wb_list)) |
536 | inode_wb_list_del(inode); | |
537 | ||
55fa6091 DC |
538 | inode_sb_list_del(inode); |
539 | ||
169ebd90 JK |
540 | /* |
541 | * Wait for flusher thread to be done with the inode so that filesystem | |
542 | * does not start destroying it while writeback is still running. Since | |
543 | * the inode has I_FREEING set, flusher thread won't start new work on | |
544 | * the inode. We just have to wait for running writeback to finish. | |
545 | */ | |
546 | inode_wait_for_writeback(inode); | |
7994e6f7 | 547 | |
be7ce416 AV |
548 | if (op->evict_inode) { |
549 | op->evict_inode(inode); | |
b4272d4c AV |
550 | } else { |
551 | if (inode->i_data.nrpages) | |
552 | truncate_inode_pages(&inode->i_data, 0); | |
dbd5768f | 553 | clear_inode(inode); |
b4272d4c | 554 | } |
661074e9 AV |
555 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
556 | bd_forget(inode); | |
557 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) | |
558 | cd_forget(inode); | |
b2b2af8e DC |
559 | |
560 | remove_inode_hash(inode); | |
561 | ||
562 | spin_lock(&inode->i_lock); | |
563 | wake_up_bit(&inode->i_state, __I_NEW); | |
564 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); | |
565 | spin_unlock(&inode->i_lock); | |
566 | ||
567 | destroy_inode(inode); | |
b4272d4c AV |
568 | } |
569 | ||
1da177e4 LT |
570 | /* |
571 | * dispose_list - dispose of the contents of a local list | |
572 | * @head: the head of the list to free | |
573 | * | |
574 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
575 | * need to worry about list corruption and SMP locks. | |
576 | */ | |
577 | static void dispose_list(struct list_head *head) | |
578 | { | |
1da177e4 LT |
579 | while (!list_empty(head)) { |
580 | struct inode *inode; | |
581 | ||
7ccf19a8 NP |
582 | inode = list_first_entry(head, struct inode, i_lru); |
583 | list_del_init(&inode->i_lru); | |
1da177e4 | 584 | |
644da596 | 585 | evict(inode); |
1da177e4 | 586 | } |
1da177e4 LT |
587 | } |
588 | ||
63997e98 AV |
589 | /** |
590 | * evict_inodes - evict all evictable inodes for a superblock | |
591 | * @sb: superblock to operate on | |
592 | * | |
593 | * Make sure that no inodes with zero refcount are retained. This is | |
594 | * called by superblock shutdown after having MS_ACTIVE flag removed, | |
595 | * so any inode reaching zero refcount during or after that call will | |
596 | * be immediately evicted. | |
1da177e4 | 597 | */ |
63997e98 | 598 | void evict_inodes(struct super_block *sb) |
1da177e4 | 599 | { |
63997e98 AV |
600 | struct inode *inode, *next; |
601 | LIST_HEAD(dispose); | |
1da177e4 | 602 | |
55fa6091 | 603 | spin_lock(&inode_sb_list_lock); |
63997e98 AV |
604 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
605 | if (atomic_read(&inode->i_count)) | |
aabb8fdb | 606 | continue; |
250df6ed DC |
607 | |
608 | spin_lock(&inode->i_lock); | |
609 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
610 | spin_unlock(&inode->i_lock); | |
1da177e4 | 611 | continue; |
250df6ed | 612 | } |
63997e98 AV |
613 | |
614 | inode->i_state |= I_FREEING; | |
02afc410 | 615 | inode_lru_list_del(inode); |
250df6ed | 616 | spin_unlock(&inode->i_lock); |
02afc410 | 617 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 618 | } |
55fa6091 | 619 | spin_unlock(&inode_sb_list_lock); |
63997e98 AV |
620 | |
621 | dispose_list(&dispose); | |
1da177e4 LT |
622 | } |
623 | ||
1da177e4 | 624 | /** |
a0318786 CH |
625 | * invalidate_inodes - attempt to free all inodes on a superblock |
626 | * @sb: superblock to operate on | |
93b270f7 | 627 | * @kill_dirty: flag to guide handling of dirty inodes |
1da177e4 | 628 | * |
a0318786 CH |
629 | * Attempts to free all inodes for a given superblock. If there were any |
630 | * busy inodes return a non-zero value, else zero. | |
93b270f7 N |
631 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat |
632 | * them as busy. | |
1da177e4 | 633 | */ |
93b270f7 | 634 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) |
1da177e4 | 635 | { |
cffbc8aa | 636 | int busy = 0; |
a0318786 CH |
637 | struct inode *inode, *next; |
638 | LIST_HEAD(dispose); | |
1da177e4 | 639 | |
55fa6091 | 640 | spin_lock(&inode_sb_list_lock); |
a0318786 | 641 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
250df6ed DC |
642 | spin_lock(&inode->i_lock); |
643 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
644 | spin_unlock(&inode->i_lock); | |
aabb8fdb | 645 | continue; |
250df6ed | 646 | } |
93b270f7 | 647 | if (inode->i_state & I_DIRTY && !kill_dirty) { |
250df6ed | 648 | spin_unlock(&inode->i_lock); |
93b270f7 N |
649 | busy = 1; |
650 | continue; | |
651 | } | |
99a38919 | 652 | if (atomic_read(&inode->i_count)) { |
250df6ed | 653 | spin_unlock(&inode->i_lock); |
99a38919 | 654 | busy = 1; |
1da177e4 LT |
655 | continue; |
656 | } | |
99a38919 | 657 | |
99a38919 | 658 | inode->i_state |= I_FREEING; |
02afc410 | 659 | inode_lru_list_del(inode); |
250df6ed | 660 | spin_unlock(&inode->i_lock); |
02afc410 | 661 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 662 | } |
55fa6091 | 663 | spin_unlock(&inode_sb_list_lock); |
1da177e4 | 664 | |
a0318786 | 665 | dispose_list(&dispose); |
1da177e4 LT |
666 | |
667 | return busy; | |
668 | } | |
1da177e4 | 669 | |
1da177e4 | 670 | /* |
bc3b14cb | 671 | * Isolate the inode from the LRU in preparation for freeing it. |
1da177e4 LT |
672 | * |
673 | * Any inodes which are pinned purely because of attached pagecache have their | |
9e38d86f NP |
674 | * pagecache removed. If the inode has metadata buffers attached to |
675 | * mapping->private_list then try to remove them. | |
1da177e4 | 676 | * |
9e38d86f NP |
677 | * If the inode has the I_REFERENCED flag set, then it means that it has been |
678 | * used recently - the flag is set in iput_final(). When we encounter such an | |
679 | * inode, clear the flag and move it to the back of the LRU so it gets another | |
680 | * pass through the LRU before it gets reclaimed. This is necessary because of | |
681 | * the fact we are doing lazy LRU updates to minimise lock contention so the | |
682 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes | |
683 | * with this flag set because they are the inodes that are out of order. | |
1da177e4 | 684 | */ |
bc3b14cb DC |
685 | static enum lru_status |
686 | inode_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg) | |
1da177e4 | 687 | { |
bc3b14cb DC |
688 | struct list_head *freeable = arg; |
689 | struct inode *inode = container_of(item, struct inode, i_lru); | |
1da177e4 | 690 | |
bc3b14cb DC |
691 | /* |
692 | * we are inverting the lru lock/inode->i_lock here, so use a trylock. | |
693 | * If we fail to get the lock, just skip it. | |
694 | */ | |
695 | if (!spin_trylock(&inode->i_lock)) | |
696 | return LRU_SKIP; | |
1da177e4 | 697 | |
bc3b14cb DC |
698 | /* |
699 | * Referenced or dirty inodes are still in use. Give them another pass | |
700 | * through the LRU as we canot reclaim them now. | |
701 | */ | |
702 | if (atomic_read(&inode->i_count) || | |
703 | (inode->i_state & ~I_REFERENCED)) { | |
704 | list_del_init(&inode->i_lru); | |
705 | spin_unlock(&inode->i_lock); | |
706 | this_cpu_dec(nr_unused); | |
707 | return LRU_REMOVED; | |
708 | } | |
1da177e4 | 709 | |
bc3b14cb DC |
710 | /* recently referenced inodes get one more pass */ |
711 | if (inode->i_state & I_REFERENCED) { | |
712 | inode->i_state &= ~I_REFERENCED; | |
713 | spin_unlock(&inode->i_lock); | |
714 | return LRU_ROTATE; | |
715 | } | |
1da177e4 | 716 | |
bc3b14cb DC |
717 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { |
718 | __iget(inode); | |
719 | spin_unlock(&inode->i_lock); | |
720 | spin_unlock(lru_lock); | |
721 | if (remove_inode_buffers(inode)) { | |
722 | unsigned long reap; | |
723 | reap = invalidate_mapping_pages(&inode->i_data, 0, -1); | |
724 | if (current_is_kswapd()) | |
725 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
726 | else | |
727 | __count_vm_events(PGINODESTEAL, reap); | |
728 | if (current->reclaim_state) | |
729 | current->reclaim_state->reclaimed_slab += reap; | |
02afc410 | 730 | } |
bc3b14cb DC |
731 | iput(inode); |
732 | spin_lock(lru_lock); | |
733 | return LRU_RETRY; | |
734 | } | |
02afc410 | 735 | |
bc3b14cb DC |
736 | WARN_ON(inode->i_state & I_NEW); |
737 | inode->i_state |= I_FREEING; | |
738 | spin_unlock(&inode->i_lock); | |
9e38d86f | 739 | |
bc3b14cb DC |
740 | list_move(&inode->i_lru, freeable); |
741 | this_cpu_dec(nr_unused); | |
742 | return LRU_REMOVED; | |
743 | } | |
7ccf19a8 | 744 | |
bc3b14cb DC |
745 | /* |
746 | * Walk the superblock inode LRU for freeable inodes and attempt to free them. | |
747 | * This is called from the superblock shrinker function with a number of inodes | |
748 | * to trim from the LRU. Inodes to be freed are moved to a temporary list and | |
749 | * then are freed outside inode_lock by dispose_list(). | |
750 | */ | |
751 | long prune_icache_sb(struct super_block *sb, unsigned long nr_to_scan) | |
752 | { | |
753 | LIST_HEAD(freeable); | |
754 | long freed; | |
1da177e4 | 755 | |
bc3b14cb DC |
756 | freed = list_lru_walk(&sb->s_inode_lru, inode_lru_isolate, |
757 | &freeable, nr_to_scan); | |
1da177e4 | 758 | dispose_list(&freeable); |
0a234c6d | 759 | return freed; |
1da177e4 LT |
760 | } |
761 | ||
1da177e4 LT |
762 | static void __wait_on_freeing_inode(struct inode *inode); |
763 | /* | |
764 | * Called with the inode lock held. | |
1da177e4 | 765 | */ |
6b3304b5 MK |
766 | static struct inode *find_inode(struct super_block *sb, |
767 | struct hlist_head *head, | |
768 | int (*test)(struct inode *, void *), | |
769 | void *data) | |
1da177e4 | 770 | { |
6b3304b5 | 771 | struct inode *inode = NULL; |
1da177e4 LT |
772 | |
773 | repeat: | |
b67bfe0d | 774 | hlist_for_each_entry(inode, head, i_hash) { |
67a23c49 DC |
775 | spin_lock(&inode->i_lock); |
776 | if (inode->i_sb != sb) { | |
777 | spin_unlock(&inode->i_lock); | |
1da177e4 | 778 | continue; |
67a23c49 DC |
779 | } |
780 | if (!test(inode, data)) { | |
781 | spin_unlock(&inode->i_lock); | |
1da177e4 | 782 | continue; |
67a23c49 | 783 | } |
a4ffdde6 | 784 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
785 | __wait_on_freeing_inode(inode); |
786 | goto repeat; | |
787 | } | |
f7899bd5 | 788 | __iget(inode); |
250df6ed | 789 | spin_unlock(&inode->i_lock); |
f7899bd5 | 790 | return inode; |
1da177e4 | 791 | } |
f7899bd5 | 792 | return NULL; |
1da177e4 LT |
793 | } |
794 | ||
795 | /* | |
796 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
797 | * iget_locked for details. | |
798 | */ | |
6b3304b5 MK |
799 | static struct inode *find_inode_fast(struct super_block *sb, |
800 | struct hlist_head *head, unsigned long ino) | |
1da177e4 | 801 | { |
6b3304b5 | 802 | struct inode *inode = NULL; |
1da177e4 LT |
803 | |
804 | repeat: | |
b67bfe0d | 805 | hlist_for_each_entry(inode, head, i_hash) { |
67a23c49 DC |
806 | spin_lock(&inode->i_lock); |
807 | if (inode->i_ino != ino) { | |
808 | spin_unlock(&inode->i_lock); | |
1da177e4 | 809 | continue; |
67a23c49 DC |
810 | } |
811 | if (inode->i_sb != sb) { | |
812 | spin_unlock(&inode->i_lock); | |
1da177e4 | 813 | continue; |
67a23c49 | 814 | } |
a4ffdde6 | 815 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
816 | __wait_on_freeing_inode(inode); |
817 | goto repeat; | |
818 | } | |
f7899bd5 | 819 | __iget(inode); |
250df6ed | 820 | spin_unlock(&inode->i_lock); |
f7899bd5 | 821 | return inode; |
1da177e4 | 822 | } |
f7899bd5 | 823 | return NULL; |
8290c35f DC |
824 | } |
825 | ||
f991bd2e ED |
826 | /* |
827 | * Each cpu owns a range of LAST_INO_BATCH numbers. | |
828 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, | |
829 | * to renew the exhausted range. | |
8290c35f | 830 | * |
f991bd2e ED |
831 | * This does not significantly increase overflow rate because every CPU can |
832 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is | |
833 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the | |
834 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase | |
835 | * overflow rate by 2x, which does not seem too significant. | |
836 | * | |
837 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
838 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
839 | * here to attempt to avoid that. | |
8290c35f | 840 | */ |
f991bd2e ED |
841 | #define LAST_INO_BATCH 1024 |
842 | static DEFINE_PER_CPU(unsigned int, last_ino); | |
843 | ||
85fe4025 | 844 | unsigned int get_next_ino(void) |
8290c35f | 845 | { |
f991bd2e ED |
846 | unsigned int *p = &get_cpu_var(last_ino); |
847 | unsigned int res = *p; | |
8290c35f | 848 | |
f991bd2e ED |
849 | #ifdef CONFIG_SMP |
850 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { | |
851 | static atomic_t shared_last_ino; | |
852 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); | |
853 | ||
854 | res = next - LAST_INO_BATCH; | |
855 | } | |
856 | #endif | |
857 | ||
858 | *p = ++res; | |
859 | put_cpu_var(last_ino); | |
860 | return res; | |
8290c35f | 861 | } |
85fe4025 | 862 | EXPORT_SYMBOL(get_next_ino); |
8290c35f | 863 | |
a209dfc7 ED |
864 | /** |
865 | * new_inode_pseudo - obtain an inode | |
866 | * @sb: superblock | |
867 | * | |
868 | * Allocates a new inode for given superblock. | |
869 | * Inode wont be chained in superblock s_inodes list | |
870 | * This means : | |
871 | * - fs can't be unmount | |
872 | * - quotas, fsnotify, writeback can't work | |
873 | */ | |
874 | struct inode *new_inode_pseudo(struct super_block *sb) | |
875 | { | |
876 | struct inode *inode = alloc_inode(sb); | |
877 | ||
878 | if (inode) { | |
879 | spin_lock(&inode->i_lock); | |
880 | inode->i_state = 0; | |
881 | spin_unlock(&inode->i_lock); | |
882 | INIT_LIST_HEAD(&inode->i_sb_list); | |
883 | } | |
884 | return inode; | |
885 | } | |
886 | ||
1da177e4 LT |
887 | /** |
888 | * new_inode - obtain an inode | |
889 | * @sb: superblock | |
890 | * | |
769848c0 | 891 | * Allocates a new inode for given superblock. The default gfp_mask |
3c1d4378 | 892 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
769848c0 MG |
893 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
894 | * for the page cache are not reclaimable or migratable, | |
895 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
896 | * newly created inode's mapping | |
897 | * | |
1da177e4 LT |
898 | */ |
899 | struct inode *new_inode(struct super_block *sb) | |
900 | { | |
6b3304b5 | 901 | struct inode *inode; |
1da177e4 | 902 | |
55fa6091 | 903 | spin_lock_prefetch(&inode_sb_list_lock); |
6b3304b5 | 904 | |
a209dfc7 ED |
905 | inode = new_inode_pseudo(sb); |
906 | if (inode) | |
55fa6091 | 907 | inode_sb_list_add(inode); |
1da177e4 LT |
908 | return inode; |
909 | } | |
1da177e4 LT |
910 | EXPORT_SYMBOL(new_inode); |
911 | ||
14358e6d | 912 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
e096d0c7 JB |
913 | void lockdep_annotate_inode_mutex_key(struct inode *inode) |
914 | { | |
a3314a0e | 915 | if (S_ISDIR(inode->i_mode)) { |
1e89a5e1 PZ |
916 | struct file_system_type *type = inode->i_sb->s_type; |
917 | ||
9a7aa12f | 918 | /* Set new key only if filesystem hasn't already changed it */ |
978d6d8c | 919 | if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) { |
9a7aa12f JK |
920 | /* |
921 | * ensure nobody is actually holding i_mutex | |
922 | */ | |
923 | mutex_destroy(&inode->i_mutex); | |
924 | mutex_init(&inode->i_mutex); | |
925 | lockdep_set_class(&inode->i_mutex, | |
926 | &type->i_mutex_dir_key); | |
927 | } | |
1e89a5e1 | 928 | } |
e096d0c7 JB |
929 | } |
930 | EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key); | |
14358e6d | 931 | #endif |
e096d0c7 JB |
932 | |
933 | /** | |
934 | * unlock_new_inode - clear the I_NEW state and wake up any waiters | |
935 | * @inode: new inode to unlock | |
936 | * | |
937 | * Called when the inode is fully initialised to clear the new state of the | |
938 | * inode and wake up anyone waiting for the inode to finish initialisation. | |
939 | */ | |
940 | void unlock_new_inode(struct inode *inode) | |
941 | { | |
942 | lockdep_annotate_inode_mutex_key(inode); | |
250df6ed | 943 | spin_lock(&inode->i_lock); |
eaff8079 CH |
944 | WARN_ON(!(inode->i_state & I_NEW)); |
945 | inode->i_state &= ~I_NEW; | |
310fa7a3 | 946 | smp_mb(); |
250df6ed DC |
947 | wake_up_bit(&inode->i_state, __I_NEW); |
948 | spin_unlock(&inode->i_lock); | |
1da177e4 | 949 | } |
1da177e4 LT |
950 | EXPORT_SYMBOL(unlock_new_inode); |
951 | ||
0b2d0724 CH |
952 | /** |
953 | * iget5_locked - obtain an inode from a mounted file system | |
954 | * @sb: super block of file system | |
955 | * @hashval: hash value (usually inode number) to get | |
956 | * @test: callback used for comparisons between inodes | |
957 | * @set: callback used to initialize a new struct inode | |
958 | * @data: opaque data pointer to pass to @test and @set | |
959 | * | |
960 | * Search for the inode specified by @hashval and @data in the inode cache, | |
961 | * and if present it is return it with an increased reference count. This is | |
962 | * a generalized version of iget_locked() for file systems where the inode | |
963 | * number is not sufficient for unique identification of an inode. | |
964 | * | |
965 | * If the inode is not in cache, allocate a new inode and return it locked, | |
966 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
967 | * before unlocking it via unlock_new_inode(). | |
1da177e4 | 968 | * |
0b2d0724 CH |
969 | * Note both @test and @set are called with the inode_hash_lock held, so can't |
970 | * sleep. | |
1da177e4 | 971 | */ |
0b2d0724 CH |
972 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, |
973 | int (*test)(struct inode *, void *), | |
974 | int (*set)(struct inode *, void *), void *data) | |
1da177e4 | 975 | { |
0b2d0724 | 976 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
6b3304b5 | 977 | struct inode *inode; |
1da177e4 | 978 | |
0b2d0724 CH |
979 | spin_lock(&inode_hash_lock); |
980 | inode = find_inode(sb, head, test, data); | |
981 | spin_unlock(&inode_hash_lock); | |
982 | ||
983 | if (inode) { | |
984 | wait_on_inode(inode); | |
985 | return inode; | |
986 | } | |
987 | ||
1da177e4 LT |
988 | inode = alloc_inode(sb); |
989 | if (inode) { | |
6b3304b5 | 990 | struct inode *old; |
1da177e4 | 991 | |
67a23c49 | 992 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
993 | /* We released the lock, so.. */ |
994 | old = find_inode(sb, head, test, data); | |
995 | if (!old) { | |
996 | if (set(inode, data)) | |
997 | goto set_failed; | |
998 | ||
250df6ed DC |
999 | spin_lock(&inode->i_lock); |
1000 | inode->i_state = I_NEW; | |
646ec461 | 1001 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1002 | spin_unlock(&inode->i_lock); |
55fa6091 | 1003 | inode_sb_list_add(inode); |
67a23c49 | 1004 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1005 | |
1006 | /* Return the locked inode with I_NEW set, the | |
1007 | * caller is responsible for filling in the contents | |
1008 | */ | |
1009 | return inode; | |
1010 | } | |
1011 | ||
1012 | /* | |
1013 | * Uhhuh, somebody else created the same inode under | |
1014 | * us. Use the old inode instead of the one we just | |
1015 | * allocated. | |
1016 | */ | |
67a23c49 | 1017 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1018 | destroy_inode(inode); |
1019 | inode = old; | |
1020 | wait_on_inode(inode); | |
1021 | } | |
1022 | return inode; | |
1023 | ||
1024 | set_failed: | |
67a23c49 | 1025 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1026 | destroy_inode(inode); |
1027 | return NULL; | |
1028 | } | |
0b2d0724 | 1029 | EXPORT_SYMBOL(iget5_locked); |
1da177e4 | 1030 | |
0b2d0724 CH |
1031 | /** |
1032 | * iget_locked - obtain an inode from a mounted file system | |
1033 | * @sb: super block of file system | |
1034 | * @ino: inode number to get | |
1035 | * | |
1036 | * Search for the inode specified by @ino in the inode cache and if present | |
1037 | * return it with an increased reference count. This is for file systems | |
1038 | * where the inode number is sufficient for unique identification of an inode. | |
1039 | * | |
1040 | * If the inode is not in cache, allocate a new inode and return it locked, | |
1041 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
1042 | * before unlocking it via unlock_new_inode(). | |
1da177e4 | 1043 | */ |
0b2d0724 | 1044 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) |
1da177e4 | 1045 | { |
0b2d0724 | 1046 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
6b3304b5 | 1047 | struct inode *inode; |
1da177e4 | 1048 | |
0b2d0724 CH |
1049 | spin_lock(&inode_hash_lock); |
1050 | inode = find_inode_fast(sb, head, ino); | |
1051 | spin_unlock(&inode_hash_lock); | |
1052 | if (inode) { | |
1053 | wait_on_inode(inode); | |
1054 | return inode; | |
1055 | } | |
1056 | ||
1da177e4 LT |
1057 | inode = alloc_inode(sb); |
1058 | if (inode) { | |
6b3304b5 | 1059 | struct inode *old; |
1da177e4 | 1060 | |
67a23c49 | 1061 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1062 | /* We released the lock, so.. */ |
1063 | old = find_inode_fast(sb, head, ino); | |
1064 | if (!old) { | |
1065 | inode->i_ino = ino; | |
250df6ed DC |
1066 | spin_lock(&inode->i_lock); |
1067 | inode->i_state = I_NEW; | |
646ec461 | 1068 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1069 | spin_unlock(&inode->i_lock); |
55fa6091 | 1070 | inode_sb_list_add(inode); |
67a23c49 | 1071 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1072 | |
1073 | /* Return the locked inode with I_NEW set, the | |
1074 | * caller is responsible for filling in the contents | |
1075 | */ | |
1076 | return inode; | |
1077 | } | |
1078 | ||
1079 | /* | |
1080 | * Uhhuh, somebody else created the same inode under | |
1081 | * us. Use the old inode instead of the one we just | |
1082 | * allocated. | |
1083 | */ | |
67a23c49 | 1084 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1085 | destroy_inode(inode); |
1086 | inode = old; | |
1087 | wait_on_inode(inode); | |
1088 | } | |
1089 | return inode; | |
1090 | } | |
0b2d0724 | 1091 | EXPORT_SYMBOL(iget_locked); |
1da177e4 | 1092 | |
ad5e195a CH |
1093 | /* |
1094 | * search the inode cache for a matching inode number. | |
1095 | * If we find one, then the inode number we are trying to | |
1096 | * allocate is not unique and so we should not use it. | |
1097 | * | |
1098 | * Returns 1 if the inode number is unique, 0 if it is not. | |
1099 | */ | |
1100 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) | |
1101 | { | |
1102 | struct hlist_head *b = inode_hashtable + hash(sb, ino); | |
ad5e195a CH |
1103 | struct inode *inode; |
1104 | ||
67a23c49 | 1105 | spin_lock(&inode_hash_lock); |
b67bfe0d | 1106 | hlist_for_each_entry(inode, b, i_hash) { |
67a23c49 DC |
1107 | if (inode->i_ino == ino && inode->i_sb == sb) { |
1108 | spin_unlock(&inode_hash_lock); | |
ad5e195a | 1109 | return 0; |
67a23c49 | 1110 | } |
ad5e195a | 1111 | } |
67a23c49 | 1112 | spin_unlock(&inode_hash_lock); |
ad5e195a CH |
1113 | |
1114 | return 1; | |
1115 | } | |
1116 | ||
1da177e4 LT |
1117 | /** |
1118 | * iunique - get a unique inode number | |
1119 | * @sb: superblock | |
1120 | * @max_reserved: highest reserved inode number | |
1121 | * | |
1122 | * Obtain an inode number that is unique on the system for a given | |
1123 | * superblock. This is used by file systems that have no natural | |
1124 | * permanent inode numbering system. An inode number is returned that | |
1125 | * is higher than the reserved limit but unique. | |
1126 | * | |
1127 | * BUGS: | |
1128 | * With a large number of inodes live on the file system this function | |
1129 | * currently becomes quite slow. | |
1130 | */ | |
1131 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
1132 | { | |
866b04fc JL |
1133 | /* |
1134 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
1135 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
1136 | * here to attempt to avoid that. | |
1137 | */ | |
ad5e195a | 1138 | static DEFINE_SPINLOCK(iunique_lock); |
866b04fc | 1139 | static unsigned int counter; |
1da177e4 | 1140 | ino_t res; |
3361c7be | 1141 | |
ad5e195a | 1142 | spin_lock(&iunique_lock); |
3361c7be JL |
1143 | do { |
1144 | if (counter <= max_reserved) | |
1145 | counter = max_reserved + 1; | |
1da177e4 | 1146 | res = counter++; |
ad5e195a CH |
1147 | } while (!test_inode_iunique(sb, res)); |
1148 | spin_unlock(&iunique_lock); | |
1da177e4 | 1149 | |
3361c7be JL |
1150 | return res; |
1151 | } | |
1da177e4 LT |
1152 | EXPORT_SYMBOL(iunique); |
1153 | ||
1154 | struct inode *igrab(struct inode *inode) | |
1155 | { | |
250df6ed DC |
1156 | spin_lock(&inode->i_lock); |
1157 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { | |
1da177e4 | 1158 | __iget(inode); |
250df6ed DC |
1159 | spin_unlock(&inode->i_lock); |
1160 | } else { | |
1161 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1162 | /* |
1163 | * Handle the case where s_op->clear_inode is not been | |
1164 | * called yet, and somebody is calling igrab | |
1165 | * while the inode is getting freed. | |
1166 | */ | |
1167 | inode = NULL; | |
250df6ed | 1168 | } |
1da177e4 LT |
1169 | return inode; |
1170 | } | |
1da177e4 LT |
1171 | EXPORT_SYMBOL(igrab); |
1172 | ||
1173 | /** | |
0b2d0724 | 1174 | * ilookup5_nowait - search for an inode in the inode cache |
1da177e4 | 1175 | * @sb: super block of file system to search |
0b2d0724 | 1176 | * @hashval: hash value (usually inode number) to search for |
1da177e4 LT |
1177 | * @test: callback used for comparisons between inodes |
1178 | * @data: opaque data pointer to pass to @test | |
1da177e4 | 1179 | * |
0b2d0724 | 1180 | * Search for the inode specified by @hashval and @data in the inode cache. |
1da177e4 LT |
1181 | * If the inode is in the cache, the inode is returned with an incremented |
1182 | * reference count. | |
1183 | * | |
0b2d0724 CH |
1184 | * Note: I_NEW is not waited upon so you have to be very careful what you do |
1185 | * with the returned inode. You probably should be using ilookup5() instead. | |
1da177e4 | 1186 | * |
b6d0ad68 | 1187 | * Note2: @test is called with the inode_hash_lock held, so can't sleep. |
1da177e4 | 1188 | */ |
0b2d0724 CH |
1189 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, |
1190 | int (*test)(struct inode *, void *), void *data) | |
1da177e4 | 1191 | { |
0b2d0724 | 1192 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
1da177e4 LT |
1193 | struct inode *inode; |
1194 | ||
67a23c49 | 1195 | spin_lock(&inode_hash_lock); |
1da177e4 | 1196 | inode = find_inode(sb, head, test, data); |
67a23c49 | 1197 | spin_unlock(&inode_hash_lock); |
88bd5121 | 1198 | |
0b2d0724 | 1199 | return inode; |
88bd5121 | 1200 | } |
88bd5121 AA |
1201 | EXPORT_SYMBOL(ilookup5_nowait); |
1202 | ||
1203 | /** | |
1204 | * ilookup5 - search for an inode in the inode cache | |
1205 | * @sb: super block of file system to search | |
1206 | * @hashval: hash value (usually inode number) to search for | |
1207 | * @test: callback used for comparisons between inodes | |
1208 | * @data: opaque data pointer to pass to @test | |
1209 | * | |
0b2d0724 CH |
1210 | * Search for the inode specified by @hashval and @data in the inode cache, |
1211 | * and if the inode is in the cache, return the inode with an incremented | |
1212 | * reference count. Waits on I_NEW before returning the inode. | |
88bd5121 | 1213 | * returned with an incremented reference count. |
1da177e4 | 1214 | * |
0b2d0724 CH |
1215 | * This is a generalized version of ilookup() for file systems where the |
1216 | * inode number is not sufficient for unique identification of an inode. | |
1da177e4 | 1217 | * |
0b2d0724 | 1218 | * Note: @test is called with the inode_hash_lock held, so can't sleep. |
1da177e4 LT |
1219 | */ |
1220 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
1221 | int (*test)(struct inode *, void *), void *data) | |
1222 | { | |
0b2d0724 | 1223 | struct inode *inode = ilookup5_nowait(sb, hashval, test, data); |
1da177e4 | 1224 | |
0b2d0724 CH |
1225 | if (inode) |
1226 | wait_on_inode(inode); | |
1227 | return inode; | |
1da177e4 | 1228 | } |
1da177e4 LT |
1229 | EXPORT_SYMBOL(ilookup5); |
1230 | ||
1231 | /** | |
1232 | * ilookup - search for an inode in the inode cache | |
1233 | * @sb: super block of file system to search | |
1234 | * @ino: inode number to search for | |
1235 | * | |
0b2d0724 CH |
1236 | * Search for the inode @ino in the inode cache, and if the inode is in the |
1237 | * cache, the inode is returned with an incremented reference count. | |
1da177e4 LT |
1238 | */ |
1239 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1240 | { | |
1241 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1da177e4 LT |
1242 | struct inode *inode; |
1243 | ||
0b2d0724 CH |
1244 | spin_lock(&inode_hash_lock); |
1245 | inode = find_inode_fast(sb, head, ino); | |
1246 | spin_unlock(&inode_hash_lock); | |
1da177e4 | 1247 | |
1da177e4 | 1248 | if (inode) |
0b2d0724 CH |
1249 | wait_on_inode(inode); |
1250 | return inode; | |
1da177e4 | 1251 | } |
0b2d0724 | 1252 | EXPORT_SYMBOL(ilookup); |
1da177e4 | 1253 | |
261bca86 AV |
1254 | int insert_inode_locked(struct inode *inode) |
1255 | { | |
1256 | struct super_block *sb = inode->i_sb; | |
1257 | ino_t ino = inode->i_ino; | |
1258 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
261bca86 | 1259 | |
261bca86 | 1260 | while (1) { |
72a43d63 | 1261 | struct inode *old = NULL; |
67a23c49 | 1262 | spin_lock(&inode_hash_lock); |
b67bfe0d | 1263 | hlist_for_each_entry(old, head, i_hash) { |
72a43d63 AV |
1264 | if (old->i_ino != ino) |
1265 | continue; | |
1266 | if (old->i_sb != sb) | |
1267 | continue; | |
250df6ed DC |
1268 | spin_lock(&old->i_lock); |
1269 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1270 | spin_unlock(&old->i_lock); | |
72a43d63 | 1271 | continue; |
250df6ed | 1272 | } |
72a43d63 AV |
1273 | break; |
1274 | } | |
b67bfe0d | 1275 | if (likely(!old)) { |
250df6ed DC |
1276 | spin_lock(&inode->i_lock); |
1277 | inode->i_state |= I_NEW; | |
261bca86 | 1278 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1279 | spin_unlock(&inode->i_lock); |
67a23c49 | 1280 | spin_unlock(&inode_hash_lock); |
261bca86 AV |
1281 | return 0; |
1282 | } | |
1283 | __iget(old); | |
250df6ed | 1284 | spin_unlock(&old->i_lock); |
67a23c49 | 1285 | spin_unlock(&inode_hash_lock); |
261bca86 | 1286 | wait_on_inode(old); |
1d3382cb | 1287 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1288 | iput(old); |
1289 | return -EBUSY; | |
1290 | } | |
1291 | iput(old); | |
1292 | } | |
1293 | } | |
261bca86 AV |
1294 | EXPORT_SYMBOL(insert_inode_locked); |
1295 | ||
1296 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1297 | int (*test)(struct inode *, void *), void *data) | |
1298 | { | |
1299 | struct super_block *sb = inode->i_sb; | |
1300 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
261bca86 | 1301 | |
261bca86 | 1302 | while (1) { |
72a43d63 AV |
1303 | struct inode *old = NULL; |
1304 | ||
67a23c49 | 1305 | spin_lock(&inode_hash_lock); |
b67bfe0d | 1306 | hlist_for_each_entry(old, head, i_hash) { |
72a43d63 AV |
1307 | if (old->i_sb != sb) |
1308 | continue; | |
1309 | if (!test(old, data)) | |
1310 | continue; | |
250df6ed DC |
1311 | spin_lock(&old->i_lock); |
1312 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1313 | spin_unlock(&old->i_lock); | |
72a43d63 | 1314 | continue; |
250df6ed | 1315 | } |
72a43d63 AV |
1316 | break; |
1317 | } | |
b67bfe0d | 1318 | if (likely(!old)) { |
250df6ed DC |
1319 | spin_lock(&inode->i_lock); |
1320 | inode->i_state |= I_NEW; | |
261bca86 | 1321 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1322 | spin_unlock(&inode->i_lock); |
67a23c49 | 1323 | spin_unlock(&inode_hash_lock); |
261bca86 AV |
1324 | return 0; |
1325 | } | |
1326 | __iget(old); | |
250df6ed | 1327 | spin_unlock(&old->i_lock); |
67a23c49 | 1328 | spin_unlock(&inode_hash_lock); |
261bca86 | 1329 | wait_on_inode(old); |
1d3382cb | 1330 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1331 | iput(old); |
1332 | return -EBUSY; | |
1333 | } | |
1334 | iput(old); | |
1335 | } | |
1336 | } | |
261bca86 AV |
1337 | EXPORT_SYMBOL(insert_inode_locked4); |
1338 | ||
1da177e4 | 1339 | |
45321ac5 AV |
1340 | int generic_delete_inode(struct inode *inode) |
1341 | { | |
1342 | return 1; | |
1343 | } | |
1344 | EXPORT_SYMBOL(generic_delete_inode); | |
1345 | ||
45321ac5 AV |
1346 | /* |
1347 | * Called when we're dropping the last reference | |
1348 | * to an inode. | |
22fe4042 | 1349 | * |
45321ac5 AV |
1350 | * Call the FS "drop_inode()" function, defaulting to |
1351 | * the legacy UNIX filesystem behaviour. If it tells | |
1352 | * us to evict inode, do so. Otherwise, retain inode | |
1353 | * in cache if fs is alive, sync and evict if fs is | |
1354 | * shutting down. | |
22fe4042 | 1355 | */ |
45321ac5 | 1356 | static void iput_final(struct inode *inode) |
1da177e4 LT |
1357 | { |
1358 | struct super_block *sb = inode->i_sb; | |
45321ac5 AV |
1359 | const struct super_operations *op = inode->i_sb->s_op; |
1360 | int drop; | |
1361 | ||
250df6ed DC |
1362 | WARN_ON(inode->i_state & I_NEW); |
1363 | ||
e7f59097 | 1364 | if (op->drop_inode) |
45321ac5 AV |
1365 | drop = op->drop_inode(inode); |
1366 | else | |
1367 | drop = generic_drop_inode(inode); | |
1da177e4 | 1368 | |
b2b2af8e DC |
1369 | if (!drop && (sb->s_flags & MS_ACTIVE)) { |
1370 | inode->i_state |= I_REFERENCED; | |
4eff96dd | 1371 | inode_add_lru(inode); |
b2b2af8e | 1372 | spin_unlock(&inode->i_lock); |
b2b2af8e DC |
1373 | return; |
1374 | } | |
1375 | ||
45321ac5 | 1376 | if (!drop) { |
991114c6 | 1377 | inode->i_state |= I_WILL_FREE; |
250df6ed | 1378 | spin_unlock(&inode->i_lock); |
1da177e4 | 1379 | write_inode_now(inode, 1); |
250df6ed | 1380 | spin_lock(&inode->i_lock); |
7ef0d737 | 1381 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1382 | inode->i_state &= ~I_WILL_FREE; |
1da177e4 | 1383 | } |
7ccf19a8 | 1384 | |
991114c6 | 1385 | inode->i_state |= I_FREEING; |
c4ae0c65 ED |
1386 | if (!list_empty(&inode->i_lru)) |
1387 | inode_lru_list_del(inode); | |
b2b2af8e | 1388 | spin_unlock(&inode->i_lock); |
b2b2af8e | 1389 | |
644da596 | 1390 | evict(inode); |
1da177e4 LT |
1391 | } |
1392 | ||
1da177e4 | 1393 | /** |
6b3304b5 | 1394 | * iput - put an inode |
1da177e4 LT |
1395 | * @inode: inode to put |
1396 | * | |
1397 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1398 | * zero, the inode is then freed and may also be destroyed. | |
1399 | * | |
1400 | * Consequently, iput() can sleep. | |
1401 | */ | |
1402 | void iput(struct inode *inode) | |
1403 | { | |
1404 | if (inode) { | |
a4ffdde6 | 1405 | BUG_ON(inode->i_state & I_CLEAR); |
1da177e4 | 1406 | |
f283c86a | 1407 | if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) |
1da177e4 LT |
1408 | iput_final(inode); |
1409 | } | |
1410 | } | |
1da177e4 LT |
1411 | EXPORT_SYMBOL(iput); |
1412 | ||
1413 | /** | |
1414 | * bmap - find a block number in a file | |
1415 | * @inode: inode of file | |
1416 | * @block: block to find | |
1417 | * | |
1418 | * Returns the block number on the device holding the inode that | |
1419 | * is the disk block number for the block of the file requested. | |
1420 | * That is, asked for block 4 of inode 1 the function will return the | |
6b3304b5 | 1421 | * disk block relative to the disk start that holds that block of the |
1da177e4 LT |
1422 | * file. |
1423 | */ | |
6b3304b5 | 1424 | sector_t bmap(struct inode *inode, sector_t block) |
1da177e4 LT |
1425 | { |
1426 | sector_t res = 0; | |
1427 | if (inode->i_mapping->a_ops->bmap) | |
1428 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1429 | return res; | |
1430 | } | |
1da177e4 LT |
1431 | EXPORT_SYMBOL(bmap); |
1432 | ||
11ff6f05 MG |
1433 | /* |
1434 | * With relative atime, only update atime if the previous atime is | |
1435 | * earlier than either the ctime or mtime or if at least a day has | |
1436 | * passed since the last atime update. | |
1437 | */ | |
1438 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1439 | struct timespec now) | |
1440 | { | |
1441 | ||
1442 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1443 | return 1; | |
1444 | /* | |
1445 | * Is mtime younger than atime? If yes, update atime: | |
1446 | */ | |
1447 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1448 | return 1; | |
1449 | /* | |
1450 | * Is ctime younger than atime? If yes, update atime: | |
1451 | */ | |
1452 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1453 | return 1; | |
1454 | ||
1455 | /* | |
1456 | * Is the previous atime value older than a day? If yes, | |
1457 | * update atime: | |
1458 | */ | |
1459 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1460 | return 1; | |
1461 | /* | |
1462 | * Good, we can skip the atime update: | |
1463 | */ | |
1464 | return 0; | |
1465 | } | |
1466 | ||
c3b2da31 JB |
1467 | /* |
1468 | * This does the actual work of updating an inodes time or version. Must have | |
1469 | * had called mnt_want_write() before calling this. | |
1470 | */ | |
1471 | static int update_time(struct inode *inode, struct timespec *time, int flags) | |
1472 | { | |
1473 | if (inode->i_op->update_time) | |
1474 | return inode->i_op->update_time(inode, time, flags); | |
1475 | ||
1476 | if (flags & S_ATIME) | |
1477 | inode->i_atime = *time; | |
1478 | if (flags & S_VERSION) | |
1479 | inode_inc_iversion(inode); | |
1480 | if (flags & S_CTIME) | |
1481 | inode->i_ctime = *time; | |
1482 | if (flags & S_MTIME) | |
1483 | inode->i_mtime = *time; | |
1484 | mark_inode_dirty_sync(inode); | |
1485 | return 0; | |
1486 | } | |
1487 | ||
1da177e4 | 1488 | /** |
869243a0 | 1489 | * touch_atime - update the access time |
185553b2 | 1490 | * @path: the &struct path to update |
1da177e4 LT |
1491 | * |
1492 | * Update the accessed time on an inode and mark it for writeback. | |
1493 | * This function automatically handles read only file systems and media, | |
1494 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1495 | */ | |
badcf2b7 | 1496 | void touch_atime(const struct path *path) |
1da177e4 | 1497 | { |
68ac1234 AV |
1498 | struct vfsmount *mnt = path->mnt; |
1499 | struct inode *inode = path->dentry->d_inode; | |
1da177e4 LT |
1500 | struct timespec now; |
1501 | ||
cdb70f3f | 1502 | if (inode->i_flags & S_NOATIME) |
b12536c2 | 1503 | return; |
37756ced | 1504 | if (IS_NOATIME(inode)) |
b12536c2 | 1505 | return; |
b2276138 | 1506 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1507 | return; |
47ae32d6 | 1508 | |
cdb70f3f | 1509 | if (mnt->mnt_flags & MNT_NOATIME) |
b12536c2 | 1510 | return; |
cdb70f3f | 1511 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1512 | return; |
1da177e4 LT |
1513 | |
1514 | now = current_fs_time(inode->i_sb); | |
11ff6f05 MG |
1515 | |
1516 | if (!relatime_need_update(mnt, inode, now)) | |
b12536c2 | 1517 | return; |
11ff6f05 | 1518 | |
47ae32d6 | 1519 | if (timespec_equal(&inode->i_atime, &now)) |
b12536c2 AK |
1520 | return; |
1521 | ||
5d37e9e6 | 1522 | if (!sb_start_write_trylock(inode->i_sb)) |
b12536c2 | 1523 | return; |
47ae32d6 | 1524 | |
5d37e9e6 JK |
1525 | if (__mnt_want_write(mnt)) |
1526 | goto skip_update; | |
c3b2da31 JB |
1527 | /* |
1528 | * File systems can error out when updating inodes if they need to | |
1529 | * allocate new space to modify an inode (such is the case for | |
1530 | * Btrfs), but since we touch atime while walking down the path we | |
1531 | * really don't care if we failed to update the atime of the file, | |
1532 | * so just ignore the return value. | |
2bc55652 AB |
1533 | * We may also fail on filesystems that have the ability to make parts |
1534 | * of the fs read only, e.g. subvolumes in Btrfs. | |
c3b2da31 JB |
1535 | */ |
1536 | update_time(inode, &now, S_ATIME); | |
5d37e9e6 JK |
1537 | __mnt_drop_write(mnt); |
1538 | skip_update: | |
1539 | sb_end_write(inode->i_sb); | |
1da177e4 | 1540 | } |
869243a0 | 1541 | EXPORT_SYMBOL(touch_atime); |
1da177e4 | 1542 | |
3ed37648 CW |
1543 | /* |
1544 | * The logic we want is | |
1545 | * | |
1546 | * if suid or (sgid and xgrp) | |
1547 | * remove privs | |
1548 | */ | |
1549 | int should_remove_suid(struct dentry *dentry) | |
1550 | { | |
1551 | umode_t mode = dentry->d_inode->i_mode; | |
1552 | int kill = 0; | |
1553 | ||
1554 | /* suid always must be killed */ | |
1555 | if (unlikely(mode & S_ISUID)) | |
1556 | kill = ATTR_KILL_SUID; | |
1557 | ||
1558 | /* | |
1559 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1560 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1561 | */ | |
1562 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1563 | kill |= ATTR_KILL_SGID; | |
1564 | ||
1565 | if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) | |
1566 | return kill; | |
1567 | ||
1568 | return 0; | |
1569 | } | |
1570 | EXPORT_SYMBOL(should_remove_suid); | |
1571 | ||
1572 | static int __remove_suid(struct dentry *dentry, int kill) | |
1573 | { | |
1574 | struct iattr newattrs; | |
1575 | ||
1576 | newattrs.ia_valid = ATTR_FORCE | kill; | |
1577 | return notify_change(dentry, &newattrs); | |
1578 | } | |
1579 | ||
1580 | int file_remove_suid(struct file *file) | |
1581 | { | |
1582 | struct dentry *dentry = file->f_path.dentry; | |
1583 | struct inode *inode = dentry->d_inode; | |
1584 | int killsuid; | |
1585 | int killpriv; | |
1586 | int error = 0; | |
1587 | ||
1588 | /* Fast path for nothing security related */ | |
1589 | if (IS_NOSEC(inode)) | |
1590 | return 0; | |
1591 | ||
1592 | killsuid = should_remove_suid(dentry); | |
1593 | killpriv = security_inode_need_killpriv(dentry); | |
1594 | ||
1595 | if (killpriv < 0) | |
1596 | return killpriv; | |
1597 | if (killpriv) | |
1598 | error = security_inode_killpriv(dentry); | |
1599 | if (!error && killsuid) | |
1600 | error = __remove_suid(dentry, killsuid); | |
1601 | if (!error && (inode->i_sb->s_flags & MS_NOSEC)) | |
1602 | inode->i_flags |= S_NOSEC; | |
1603 | ||
1604 | return error; | |
1605 | } | |
1606 | EXPORT_SYMBOL(file_remove_suid); | |
1607 | ||
1da177e4 | 1608 | /** |
870f4817 CH |
1609 | * file_update_time - update mtime and ctime time |
1610 | * @file: file accessed | |
1da177e4 | 1611 | * |
870f4817 CH |
1612 | * Update the mtime and ctime members of an inode and mark the inode |
1613 | * for writeback. Note that this function is meant exclusively for | |
1614 | * usage in the file write path of filesystems, and filesystems may | |
1615 | * choose to explicitly ignore update via this function with the | |
2eadfc0e | 1616 | * S_NOCMTIME inode flag, e.g. for network filesystem where these |
c3b2da31 JB |
1617 | * timestamps are handled by the server. This can return an error for |
1618 | * file systems who need to allocate space in order to update an inode. | |
1da177e4 LT |
1619 | */ |
1620 | ||
c3b2da31 | 1621 | int file_update_time(struct file *file) |
1da177e4 | 1622 | { |
496ad9aa | 1623 | struct inode *inode = file_inode(file); |
1da177e4 | 1624 | struct timespec now; |
c3b2da31 JB |
1625 | int sync_it = 0; |
1626 | int ret; | |
1da177e4 | 1627 | |
ce06e0b2 | 1628 | /* First try to exhaust all avenues to not sync */ |
1da177e4 | 1629 | if (IS_NOCMTIME(inode)) |
c3b2da31 | 1630 | return 0; |
20ddee2c | 1631 | |
1da177e4 | 1632 | now = current_fs_time(inode->i_sb); |
ce06e0b2 AK |
1633 | if (!timespec_equal(&inode->i_mtime, &now)) |
1634 | sync_it = S_MTIME; | |
1da177e4 | 1635 | |
ce06e0b2 AK |
1636 | if (!timespec_equal(&inode->i_ctime, &now)) |
1637 | sync_it |= S_CTIME; | |
870f4817 | 1638 | |
ce06e0b2 AK |
1639 | if (IS_I_VERSION(inode)) |
1640 | sync_it |= S_VERSION; | |
7a224228 | 1641 | |
ce06e0b2 | 1642 | if (!sync_it) |
c3b2da31 | 1643 | return 0; |
ce06e0b2 AK |
1644 | |
1645 | /* Finally allowed to write? Takes lock. */ | |
eb04c282 | 1646 | if (__mnt_want_write_file(file)) |
c3b2da31 | 1647 | return 0; |
ce06e0b2 | 1648 | |
c3b2da31 | 1649 | ret = update_time(inode, &now, sync_it); |
eb04c282 | 1650 | __mnt_drop_write_file(file); |
c3b2da31 JB |
1651 | |
1652 | return ret; | |
1da177e4 | 1653 | } |
870f4817 | 1654 | EXPORT_SYMBOL(file_update_time); |
1da177e4 LT |
1655 | |
1656 | int inode_needs_sync(struct inode *inode) | |
1657 | { | |
1658 | if (IS_SYNC(inode)) | |
1659 | return 1; | |
1660 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1661 | return 1; | |
1662 | return 0; | |
1663 | } | |
1da177e4 LT |
1664 | EXPORT_SYMBOL(inode_needs_sync); |
1665 | ||
1da177e4 LT |
1666 | int inode_wait(void *word) |
1667 | { | |
1668 | schedule(); | |
1669 | return 0; | |
1670 | } | |
d44dab8d | 1671 | EXPORT_SYMBOL(inode_wait); |
1da177e4 LT |
1672 | |
1673 | /* | |
168a9fd6 MS |
1674 | * If we try to find an inode in the inode hash while it is being |
1675 | * deleted, we have to wait until the filesystem completes its | |
1676 | * deletion before reporting that it isn't found. This function waits | |
1677 | * until the deletion _might_ have completed. Callers are responsible | |
1678 | * to recheck inode state. | |
1679 | * | |
eaff8079 | 1680 | * It doesn't matter if I_NEW is not set initially, a call to |
250df6ed DC |
1681 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list |
1682 | * will DTRT. | |
1da177e4 LT |
1683 | */ |
1684 | static void __wait_on_freeing_inode(struct inode *inode) | |
1685 | { | |
1686 | wait_queue_head_t *wq; | |
eaff8079 CH |
1687 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); |
1688 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1da177e4 | 1689 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
250df6ed | 1690 | spin_unlock(&inode->i_lock); |
67a23c49 | 1691 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1692 | schedule(); |
1693 | finish_wait(wq, &wait.wait); | |
67a23c49 | 1694 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1695 | } |
1696 | ||
1da177e4 LT |
1697 | static __initdata unsigned long ihash_entries; |
1698 | static int __init set_ihash_entries(char *str) | |
1699 | { | |
1700 | if (!str) | |
1701 | return 0; | |
1702 | ihash_entries = simple_strtoul(str, &str, 0); | |
1703 | return 1; | |
1704 | } | |
1705 | __setup("ihash_entries=", set_ihash_entries); | |
1706 | ||
1707 | /* | |
1708 | * Initialize the waitqueues and inode hash table. | |
1709 | */ | |
1710 | void __init inode_init_early(void) | |
1711 | { | |
074b8517 | 1712 | unsigned int loop; |
1da177e4 LT |
1713 | |
1714 | /* If hashes are distributed across NUMA nodes, defer | |
1715 | * hash allocation until vmalloc space is available. | |
1716 | */ | |
1717 | if (hashdist) | |
1718 | return; | |
1719 | ||
1720 | inode_hashtable = | |
1721 | alloc_large_system_hash("Inode-cache", | |
1722 | sizeof(struct hlist_head), | |
1723 | ihash_entries, | |
1724 | 14, | |
1725 | HASH_EARLY, | |
1726 | &i_hash_shift, | |
1727 | &i_hash_mask, | |
31fe62b9 | 1728 | 0, |
1da177e4 LT |
1729 | 0); |
1730 | ||
074b8517 | 1731 | for (loop = 0; loop < (1U << i_hash_shift); loop++) |
1da177e4 LT |
1732 | INIT_HLIST_HEAD(&inode_hashtable[loop]); |
1733 | } | |
1734 | ||
74bf17cf | 1735 | void __init inode_init(void) |
1da177e4 | 1736 | { |
074b8517 | 1737 | unsigned int loop; |
1da177e4 LT |
1738 | |
1739 | /* inode slab cache */ | |
b0196009 PJ |
1740 | inode_cachep = kmem_cache_create("inode_cache", |
1741 | sizeof(struct inode), | |
1742 | 0, | |
1743 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1744 | SLAB_MEM_SPREAD), | |
20c2df83 | 1745 | init_once); |
1da177e4 LT |
1746 | |
1747 | /* Hash may have been set up in inode_init_early */ | |
1748 | if (!hashdist) | |
1749 | return; | |
1750 | ||
1751 | inode_hashtable = | |
1752 | alloc_large_system_hash("Inode-cache", | |
1753 | sizeof(struct hlist_head), | |
1754 | ihash_entries, | |
1755 | 14, | |
1756 | 0, | |
1757 | &i_hash_shift, | |
1758 | &i_hash_mask, | |
31fe62b9 | 1759 | 0, |
1da177e4 LT |
1760 | 0); |
1761 | ||
074b8517 | 1762 | for (loop = 0; loop < (1U << i_hash_shift); loop++) |
1da177e4 LT |
1763 | INIT_HLIST_HEAD(&inode_hashtable[loop]); |
1764 | } | |
1765 | ||
1766 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1767 | { | |
1768 | inode->i_mode = mode; | |
1769 | if (S_ISCHR(mode)) { | |
1770 | inode->i_fop = &def_chr_fops; | |
1771 | inode->i_rdev = rdev; | |
1772 | } else if (S_ISBLK(mode)) { | |
1773 | inode->i_fop = &def_blk_fops; | |
1774 | inode->i_rdev = rdev; | |
1775 | } else if (S_ISFIFO(mode)) | |
599a0ac1 | 1776 | inode->i_fop = &pipefifo_fops; |
1da177e4 LT |
1777 | else if (S_ISSOCK(mode)) |
1778 | inode->i_fop = &bad_sock_fops; | |
1779 | else | |
af0d9ae8 MK |
1780 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" |
1781 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1782 | inode->i_ino); | |
1da177e4 LT |
1783 | } |
1784 | EXPORT_SYMBOL(init_special_inode); | |
a1bd120d DM |
1785 | |
1786 | /** | |
eaae668d | 1787 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards |
a1bd120d DM |
1788 | * @inode: New inode |
1789 | * @dir: Directory inode | |
1790 | * @mode: mode of the new inode | |
1791 | */ | |
1792 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
62bb1091 | 1793 | umode_t mode) |
a1bd120d DM |
1794 | { |
1795 | inode->i_uid = current_fsuid(); | |
1796 | if (dir && dir->i_mode & S_ISGID) { | |
1797 | inode->i_gid = dir->i_gid; | |
1798 | if (S_ISDIR(mode)) | |
1799 | mode |= S_ISGID; | |
1800 | } else | |
1801 | inode->i_gid = current_fsgid(); | |
1802 | inode->i_mode = mode; | |
1803 | } | |
1804 | EXPORT_SYMBOL(inode_init_owner); | |
e795b717 | 1805 | |
2e149670 SH |
1806 | /** |
1807 | * inode_owner_or_capable - check current task permissions to inode | |
1808 | * @inode: inode being checked | |
1809 | * | |
1810 | * Return true if current either has CAP_FOWNER to the inode, or | |
1811 | * owns the file. | |
e795b717 | 1812 | */ |
2e149670 | 1813 | bool inode_owner_or_capable(const struct inode *inode) |
e795b717 | 1814 | { |
92361636 | 1815 | if (uid_eq(current_fsuid(), inode->i_uid)) |
e795b717 | 1816 | return true; |
1a48e2ac | 1817 | if (inode_capable(inode, CAP_FOWNER)) |
e795b717 SH |
1818 | return true; |
1819 | return false; | |
1820 | } | |
2e149670 | 1821 | EXPORT_SYMBOL(inode_owner_or_capable); |
1d59d61f TM |
1822 | |
1823 | /* | |
1824 | * Direct i/o helper functions | |
1825 | */ | |
1826 | static void __inode_dio_wait(struct inode *inode) | |
1827 | { | |
1828 | wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP); | |
1829 | DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP); | |
1830 | ||
1831 | do { | |
1832 | prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE); | |
1833 | if (atomic_read(&inode->i_dio_count)) | |
1834 | schedule(); | |
1835 | } while (atomic_read(&inode->i_dio_count)); | |
1836 | finish_wait(wq, &q.wait); | |
1837 | } | |
1838 | ||
1839 | /** | |
1840 | * inode_dio_wait - wait for outstanding DIO requests to finish | |
1841 | * @inode: inode to wait for | |
1842 | * | |
1843 | * Waits for all pending direct I/O requests to finish so that we can | |
1844 | * proceed with a truncate or equivalent operation. | |
1845 | * | |
1846 | * Must be called under a lock that serializes taking new references | |
1847 | * to i_dio_count, usually by inode->i_mutex. | |
1848 | */ | |
1849 | void inode_dio_wait(struct inode *inode) | |
1850 | { | |
1851 | if (atomic_read(&inode->i_dio_count)) | |
1852 | __inode_dio_wait(inode); | |
1853 | } | |
1854 | EXPORT_SYMBOL(inode_dio_wait); | |
1855 | ||
1856 | /* | |
1857 | * inode_dio_done - signal finish of a direct I/O requests | |
1858 | * @inode: inode the direct I/O happens on | |
1859 | * | |
1860 | * This is called once we've finished processing a direct I/O request, | |
1861 | * and is used to wake up callers waiting for direct I/O to be quiesced. | |
1862 | */ | |
1863 | void inode_dio_done(struct inode *inode) | |
1864 | { | |
1865 | if (atomic_dec_and_test(&inode->i_dio_count)) | |
1866 | wake_up_bit(&inode->i_state, __I_DIO_WAKEUP); | |
1867 | } | |
1868 | EXPORT_SYMBOL(inode_dio_done); |