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> |
a66979ab | 20 | #include "internal.h" |
1da177e4 | 21 | |
250df6ed | 22 | /* |
4b4563dc | 23 | * Inode locking rules: |
250df6ed DC |
24 | * |
25 | * inode->i_lock protects: | |
26 | * inode->i_state, inode->i_hash, __iget() | |
09cc9fc7 | 27 | * inode->i_sb->s_inode_lru_lock protects: |
98b745c6 | 28 | * inode->i_sb->s_inode_lru, inode->i_lru |
55fa6091 DC |
29 | * inode_sb_list_lock protects: |
30 | * sb->s_inodes, inode->i_sb_list | |
f758eeab | 31 | * bdi->wb.list_lock protects: |
a66979ab | 32 | * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list |
67a23c49 DC |
33 | * inode_hash_lock protects: |
34 | * inode_hashtable, inode->i_hash | |
250df6ed DC |
35 | * |
36 | * Lock ordering: | |
55fa6091 DC |
37 | * |
38 | * inode_sb_list_lock | |
39 | * inode->i_lock | |
09cc9fc7 | 40 | * inode->i_sb->s_inode_lru_lock |
a66979ab | 41 | * |
f758eeab | 42 | * bdi->wb.list_lock |
a66979ab | 43 | * inode->i_lock |
67a23c49 DC |
44 | * |
45 | * inode_hash_lock | |
46 | * inode_sb_list_lock | |
47 | * inode->i_lock | |
48 | * | |
49 | * iunique_lock | |
50 | * inode_hash_lock | |
250df6ed DC |
51 | */ |
52 | ||
fa3536cc ED |
53 | static unsigned int i_hash_mask __read_mostly; |
54 | static unsigned int i_hash_shift __read_mostly; | |
67a23c49 DC |
55 | static struct hlist_head *inode_hashtable __read_mostly; |
56 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock); | |
1da177e4 | 57 | |
55fa6091 DC |
58 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock); |
59 | ||
7dcda1c9 JA |
60 | /* |
61 | * Empty aops. Can be used for the cases where the user does not | |
62 | * define any of the address_space operations. | |
63 | */ | |
64 | const struct address_space_operations empty_aops = { | |
65 | }; | |
66 | EXPORT_SYMBOL(empty_aops); | |
67 | ||
1da177e4 LT |
68 | /* |
69 | * Statistics gathering.. | |
70 | */ | |
71 | struct inodes_stat_t inodes_stat; | |
72 | ||
3e880fb5 | 73 | static DEFINE_PER_CPU(unsigned int, nr_inodes); |
fcb94f72 | 74 | static DEFINE_PER_CPU(unsigned int, nr_unused); |
cffbc8aa | 75 | |
6b3304b5 | 76 | static struct kmem_cache *inode_cachep __read_mostly; |
1da177e4 | 77 | |
3e880fb5 | 78 | static int get_nr_inodes(void) |
cffbc8aa | 79 | { |
3e880fb5 NP |
80 | int i; |
81 | int sum = 0; | |
82 | for_each_possible_cpu(i) | |
83 | sum += per_cpu(nr_inodes, i); | |
84 | return sum < 0 ? 0 : sum; | |
cffbc8aa DC |
85 | } |
86 | ||
87 | static inline int get_nr_inodes_unused(void) | |
88 | { | |
fcb94f72 DC |
89 | int i; |
90 | int sum = 0; | |
91 | for_each_possible_cpu(i) | |
92 | sum += per_cpu(nr_unused, i); | |
93 | return sum < 0 ? 0 : sum; | |
cffbc8aa DC |
94 | } |
95 | ||
96 | int get_nr_dirty_inodes(void) | |
97 | { | |
3e880fb5 | 98 | /* not actually dirty inodes, but a wild approximation */ |
cffbc8aa DC |
99 | int nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); |
100 | return nr_dirty > 0 ? nr_dirty : 0; | |
cffbc8aa DC |
101 | } |
102 | ||
103 | /* | |
104 | * Handle nr_inode sysctl | |
105 | */ | |
106 | #ifdef CONFIG_SYSCTL | |
107 | int proc_nr_inodes(ctl_table *table, int write, | |
108 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
109 | { | |
110 | inodes_stat.nr_inodes = get_nr_inodes(); | |
fcb94f72 | 111 | inodes_stat.nr_unused = get_nr_inodes_unused(); |
cffbc8aa DC |
112 | return proc_dointvec(table, write, buffer, lenp, ppos); |
113 | } | |
114 | #endif | |
115 | ||
2cb1599f DC |
116 | /** |
117 | * inode_init_always - perform inode structure intialisation | |
0bc02f3f RD |
118 | * @sb: superblock inode belongs to |
119 | * @inode: inode to initialise | |
2cb1599f DC |
120 | * |
121 | * These are initializations that need to be done on every inode | |
122 | * allocation as the fields are not initialised by slab allocation. | |
123 | */ | |
54e34621 | 124 | int inode_init_always(struct super_block *sb, struct inode *inode) |
1da177e4 | 125 | { |
6e1d5dcc | 126 | static const struct inode_operations empty_iops; |
99ac48f5 | 127 | static const struct file_operations empty_fops; |
6b3304b5 | 128 | struct address_space *const mapping = &inode->i_data; |
2cb1599f DC |
129 | |
130 | inode->i_sb = sb; | |
131 | inode->i_blkbits = sb->s_blocksize_bits; | |
132 | inode->i_flags = 0; | |
133 | atomic_set(&inode->i_count, 1); | |
134 | inode->i_op = &empty_iops; | |
135 | inode->i_fop = &empty_fops; | |
a78ef704 | 136 | inode->__i_nlink = 1; |
3ddcd056 | 137 | inode->i_opflags = 0; |
92361636 EB |
138 | i_uid_write(inode, 0); |
139 | i_gid_write(inode, 0); | |
2cb1599f DC |
140 | atomic_set(&inode->i_writecount, 0); |
141 | inode->i_size = 0; | |
142 | inode->i_blocks = 0; | |
143 | inode->i_bytes = 0; | |
144 | inode->i_generation = 0; | |
1da177e4 | 145 | #ifdef CONFIG_QUOTA |
2cb1599f | 146 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); |
1da177e4 | 147 | #endif |
2cb1599f DC |
148 | inode->i_pipe = NULL; |
149 | inode->i_bdev = NULL; | |
150 | inode->i_cdev = NULL; | |
151 | inode->i_rdev = 0; | |
152 | inode->dirtied_when = 0; | |
6146f0d5 MZ |
153 | |
154 | if (security_inode_alloc(inode)) | |
54e34621 | 155 | goto out; |
2cb1599f DC |
156 | spin_lock_init(&inode->i_lock); |
157 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
158 | ||
159 | mutex_init(&inode->i_mutex); | |
160 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); | |
161 | ||
bd5fe6c5 | 162 | atomic_set(&inode->i_dio_count, 0); |
2cb1599f DC |
163 | |
164 | mapping->a_ops = &empty_aops; | |
165 | mapping->host = inode; | |
166 | mapping->flags = 0; | |
6fa3eb70 S |
167 | |
168 | mapping_set_gfp_mask(mapping, (GFP_HIGHUSER_MOVABLE & ~__GFP_HIGHMEM) | __GFP_SLOWHIGHMEM); | |
169 | //mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); | |
170 | ||
252aa6f5 | 171 | mapping->private_data = NULL; |
2cb1599f DC |
172 | mapping->backing_dev_info = &default_backing_dev_info; |
173 | mapping->writeback_index = 0; | |
174 | ||
175 | /* | |
176 | * If the block_device provides a backing_dev_info for client | |
177 | * inodes then use that. Otherwise the inode share the bdev's | |
178 | * backing_dev_info. | |
179 | */ | |
180 | if (sb->s_bdev) { | |
181 | struct backing_dev_info *bdi; | |
182 | ||
2c96ce9f | 183 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; |
2cb1599f DC |
184 | mapping->backing_dev_info = bdi; |
185 | } | |
186 | inode->i_private = NULL; | |
187 | inode->i_mapping = mapping; | |
b3d9b7a3 | 188 | INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */ |
f19d4a8f AV |
189 | #ifdef CONFIG_FS_POSIX_ACL |
190 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; | |
191 | #endif | |
2cb1599f | 192 | |
3be25f49 EP |
193 | #ifdef CONFIG_FSNOTIFY |
194 | inode->i_fsnotify_mask = 0; | |
195 | #endif | |
196 | ||
3e880fb5 | 197 | this_cpu_inc(nr_inodes); |
cffbc8aa | 198 | |
54e34621 | 199 | return 0; |
54e34621 CH |
200 | out: |
201 | return -ENOMEM; | |
1da177e4 | 202 | } |
2cb1599f DC |
203 | EXPORT_SYMBOL(inode_init_always); |
204 | ||
205 | static struct inode *alloc_inode(struct super_block *sb) | |
206 | { | |
207 | struct inode *inode; | |
208 | ||
209 | if (sb->s_op->alloc_inode) | |
210 | inode = sb->s_op->alloc_inode(sb); | |
211 | else | |
212 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
213 | ||
54e34621 CH |
214 | if (!inode) |
215 | return NULL; | |
216 | ||
217 | if (unlikely(inode_init_always(sb, inode))) { | |
218 | if (inode->i_sb->s_op->destroy_inode) | |
219 | inode->i_sb->s_op->destroy_inode(inode); | |
220 | else | |
221 | kmem_cache_free(inode_cachep, inode); | |
222 | return NULL; | |
223 | } | |
224 | ||
225 | return inode; | |
2cb1599f | 226 | } |
1da177e4 | 227 | |
ff0c7d15 NP |
228 | void free_inode_nonrcu(struct inode *inode) |
229 | { | |
230 | kmem_cache_free(inode_cachep, inode); | |
231 | } | |
232 | EXPORT_SYMBOL(free_inode_nonrcu); | |
233 | ||
2e00c97e | 234 | void __destroy_inode(struct inode *inode) |
1da177e4 | 235 | { |
b7542f8c | 236 | BUG_ON(inode_has_buffers(inode)); |
1da177e4 | 237 | security_inode_free(inode); |
3be25f49 | 238 | fsnotify_inode_delete(inode); |
7ada4db8 MS |
239 | if (!inode->i_nlink) { |
240 | WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0); | |
241 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
242 | } | |
243 | ||
f19d4a8f AV |
244 | #ifdef CONFIG_FS_POSIX_ACL |
245 | if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) | |
246 | posix_acl_release(inode->i_acl); | |
247 | if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) | |
248 | posix_acl_release(inode->i_default_acl); | |
249 | #endif | |
3e880fb5 | 250 | this_cpu_dec(nr_inodes); |
2e00c97e CH |
251 | } |
252 | EXPORT_SYMBOL(__destroy_inode); | |
253 | ||
fa0d7e3d NP |
254 | static void i_callback(struct rcu_head *head) |
255 | { | |
256 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
257 | kmem_cache_free(inode_cachep, inode); |
258 | } | |
259 | ||
56b0dacf | 260 | static void destroy_inode(struct inode *inode) |
2e00c97e | 261 | { |
7ccf19a8 | 262 | BUG_ON(!list_empty(&inode->i_lru)); |
2e00c97e | 263 | __destroy_inode(inode); |
1da177e4 LT |
264 | if (inode->i_sb->s_op->destroy_inode) |
265 | inode->i_sb->s_op->destroy_inode(inode); | |
266 | else | |
fa0d7e3d | 267 | call_rcu(&inode->i_rcu, i_callback); |
1da177e4 | 268 | } |
1da177e4 | 269 | |
7ada4db8 MS |
270 | /** |
271 | * drop_nlink - directly drop an inode's link count | |
272 | * @inode: inode | |
273 | * | |
274 | * This is a low-level filesystem helper to replace any | |
275 | * direct filesystem manipulation of i_nlink. In cases | |
276 | * where we are attempting to track writes to the | |
277 | * filesystem, a decrement to zero means an imminent | |
278 | * write when the file is truncated and actually unlinked | |
279 | * on the filesystem. | |
280 | */ | |
281 | void drop_nlink(struct inode *inode) | |
282 | { | |
283 | WARN_ON(inode->i_nlink == 0); | |
284 | inode->__i_nlink--; | |
285 | if (!inode->i_nlink) | |
286 | atomic_long_inc(&inode->i_sb->s_remove_count); | |
287 | } | |
288 | EXPORT_SYMBOL(drop_nlink); | |
289 | ||
290 | /** | |
291 | * clear_nlink - directly zero an inode's link count | |
292 | * @inode: inode | |
293 | * | |
294 | * This is a low-level filesystem helper to replace any | |
295 | * direct filesystem manipulation of i_nlink. See | |
296 | * drop_nlink() for why we care about i_nlink hitting zero. | |
297 | */ | |
298 | void clear_nlink(struct inode *inode) | |
299 | { | |
300 | if (inode->i_nlink) { | |
301 | inode->__i_nlink = 0; | |
302 | atomic_long_inc(&inode->i_sb->s_remove_count); | |
303 | } | |
304 | } | |
305 | EXPORT_SYMBOL(clear_nlink); | |
306 | ||
307 | /** | |
308 | * set_nlink - directly set an inode's link count | |
309 | * @inode: inode | |
310 | * @nlink: new nlink (should be non-zero) | |
311 | * | |
312 | * This is a low-level filesystem helper to replace any | |
313 | * direct filesystem manipulation of i_nlink. | |
314 | */ | |
315 | void set_nlink(struct inode *inode, unsigned int nlink) | |
316 | { | |
317 | if (!nlink) { | |
7ada4db8 MS |
318 | clear_nlink(inode); |
319 | } else { | |
320 | /* Yes, some filesystems do change nlink from zero to one */ | |
321 | if (inode->i_nlink == 0) | |
322 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
323 | ||
324 | inode->__i_nlink = nlink; | |
325 | } | |
326 | } | |
327 | EXPORT_SYMBOL(set_nlink); | |
328 | ||
329 | /** | |
330 | * inc_nlink - directly increment an inode's link count | |
331 | * @inode: inode | |
332 | * | |
333 | * This is a low-level filesystem helper to replace any | |
334 | * direct filesystem manipulation of i_nlink. Currently, | |
335 | * it is only here for parity with dec_nlink(). | |
336 | */ | |
337 | void inc_nlink(struct inode *inode) | |
338 | { | |
339 | if (WARN_ON(inode->i_nlink == 0)) | |
340 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
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 | { | |
09cc9fc7 | 405 | spin_lock(&inode->i_sb->s_inode_lru_lock); |
7ccf19a8 | 406 | if (list_empty(&inode->i_lru)) { |
98b745c6 DC |
407 | list_add(&inode->i_lru, &inode->i_sb->s_inode_lru); |
408 | inode->i_sb->s_nr_inodes_unused++; | |
fcb94f72 | 409 | this_cpu_inc(nr_unused); |
9e38d86f | 410 | } |
09cc9fc7 | 411 | spin_unlock(&inode->i_sb->s_inode_lru_lock); |
9e38d86f | 412 | } |
2e147f1e | 413 | |
4eff96dd JK |
414 | /* |
415 | * Add inode to LRU if needed (inode is unused and clean). | |
416 | * | |
417 | * Needs inode->i_lock held. | |
418 | */ | |
419 | void inode_add_lru(struct inode *inode) | |
420 | { | |
421 | if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) && | |
422 | !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE) | |
423 | inode_lru_list_add(inode); | |
424 | } | |
425 | ||
426 | ||
9e38d86f NP |
427 | static void inode_lru_list_del(struct inode *inode) |
428 | { | |
09cc9fc7 | 429 | spin_lock(&inode->i_sb->s_inode_lru_lock); |
7ccf19a8 NP |
430 | if (!list_empty(&inode->i_lru)) { |
431 | list_del_init(&inode->i_lru); | |
98b745c6 | 432 | inode->i_sb->s_nr_inodes_unused--; |
fcb94f72 | 433 | this_cpu_dec(nr_unused); |
9e38d86f | 434 | } |
09cc9fc7 | 435 | spin_unlock(&inode->i_sb->s_inode_lru_lock); |
1da177e4 LT |
436 | } |
437 | ||
646ec461 CH |
438 | /** |
439 | * inode_sb_list_add - add inode to the superblock list of inodes | |
440 | * @inode: inode to add | |
441 | */ | |
442 | void inode_sb_list_add(struct inode *inode) | |
443 | { | |
55fa6091 DC |
444 | spin_lock(&inode_sb_list_lock); |
445 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); | |
446 | spin_unlock(&inode_sb_list_lock); | |
646ec461 CH |
447 | } |
448 | EXPORT_SYMBOL_GPL(inode_sb_list_add); | |
449 | ||
55fa6091 | 450 | static inline void inode_sb_list_del(struct inode *inode) |
646ec461 | 451 | { |
a209dfc7 ED |
452 | if (!list_empty(&inode->i_sb_list)) { |
453 | spin_lock(&inode_sb_list_lock); | |
454 | list_del_init(&inode->i_sb_list); | |
455 | spin_unlock(&inode_sb_list_lock); | |
456 | } | |
646ec461 CH |
457 | } |
458 | ||
4c51acbc DC |
459 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
460 | { | |
461 | unsigned long tmp; | |
462 | ||
463 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
464 | L1_CACHE_BYTES; | |
4b4563dc CH |
465 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift); |
466 | return tmp & i_hash_mask; | |
4c51acbc DC |
467 | } |
468 | ||
469 | /** | |
470 | * __insert_inode_hash - hash an inode | |
471 | * @inode: unhashed inode | |
472 | * @hashval: unsigned long value used to locate this object in the | |
473 | * inode_hashtable. | |
474 | * | |
475 | * Add an inode to the inode hash for this superblock. | |
476 | */ | |
477 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
478 | { | |
646ec461 CH |
479 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); |
480 | ||
67a23c49 | 481 | spin_lock(&inode_hash_lock); |
250df6ed | 482 | spin_lock(&inode->i_lock); |
646ec461 | 483 | hlist_add_head(&inode->i_hash, b); |
250df6ed | 484 | spin_unlock(&inode->i_lock); |
67a23c49 | 485 | spin_unlock(&inode_hash_lock); |
4c51acbc DC |
486 | } |
487 | EXPORT_SYMBOL(__insert_inode_hash); | |
488 | ||
4c51acbc | 489 | /** |
f2ee7abf | 490 | * __remove_inode_hash - remove an inode from the hash |
4c51acbc DC |
491 | * @inode: inode to unhash |
492 | * | |
493 | * Remove an inode from the superblock. | |
494 | */ | |
f2ee7abf | 495 | void __remove_inode_hash(struct inode *inode) |
4c51acbc | 496 | { |
67a23c49 | 497 | spin_lock(&inode_hash_lock); |
250df6ed | 498 | spin_lock(&inode->i_lock); |
4c51acbc | 499 | hlist_del_init(&inode->i_hash); |
250df6ed | 500 | spin_unlock(&inode->i_lock); |
67a23c49 | 501 | spin_unlock(&inode_hash_lock); |
4c51acbc | 502 | } |
f2ee7abf | 503 | EXPORT_SYMBOL(__remove_inode_hash); |
4c51acbc | 504 | |
dbd5768f | 505 | void clear_inode(struct inode *inode) |
b0683aa6 AV |
506 | { |
507 | might_sleep(); | |
08142579 JK |
508 | /* |
509 | * We have to cycle tree_lock here because reclaim can be still in the | |
510 | * process of removing the last page (in __delete_from_page_cache()) | |
511 | * and we must not free mapping under it. | |
512 | */ | |
513 | spin_lock_irq(&inode->i_data.tree_lock); | |
b0683aa6 | 514 | BUG_ON(inode->i_data.nrpages); |
08142579 | 515 | spin_unlock_irq(&inode->i_data.tree_lock); |
b0683aa6 AV |
516 | BUG_ON(!list_empty(&inode->i_data.private_list)); |
517 | BUG_ON(!(inode->i_state & I_FREEING)); | |
518 | BUG_ON(inode->i_state & I_CLEAR); | |
fa0d7e3d | 519 | /* don't need i_lock here, no concurrent mods to i_state */ |
b0683aa6 AV |
520 | inode->i_state = I_FREEING | I_CLEAR; |
521 | } | |
dbd5768f | 522 | EXPORT_SYMBOL(clear_inode); |
b0683aa6 | 523 | |
b2b2af8e DC |
524 | /* |
525 | * Free the inode passed in, removing it from the lists it is still connected | |
526 | * to. We remove any pages still attached to the inode and wait for any IO that | |
527 | * is still in progress before finally destroying the inode. | |
528 | * | |
529 | * An inode must already be marked I_FREEING so that we avoid the inode being | |
530 | * moved back onto lists if we race with other code that manipulates the lists | |
531 | * (e.g. writeback_single_inode). The caller is responsible for setting this. | |
532 | * | |
533 | * An inode must already be removed from the LRU list before being evicted from | |
534 | * the cache. This should occur atomically with setting the I_FREEING state | |
535 | * flag, so no inodes here should ever be on the LRU when being evicted. | |
536 | */ | |
644da596 | 537 | static void evict(struct inode *inode) |
b4272d4c AV |
538 | { |
539 | const struct super_operations *op = inode->i_sb->s_op; | |
540 | ||
b2b2af8e DC |
541 | BUG_ON(!(inode->i_state & I_FREEING)); |
542 | BUG_ON(!list_empty(&inode->i_lru)); | |
543 | ||
b12362bd ED |
544 | if (!list_empty(&inode->i_wb_list)) |
545 | inode_wb_list_del(inode); | |
546 | ||
55fa6091 DC |
547 | inode_sb_list_del(inode); |
548 | ||
169ebd90 JK |
549 | /* |
550 | * Wait for flusher thread to be done with the inode so that filesystem | |
551 | * does not start destroying it while writeback is still running. Since | |
552 | * the inode has I_FREEING set, flusher thread won't start new work on | |
553 | * the inode. We just have to wait for running writeback to finish. | |
554 | */ | |
555 | inode_wait_for_writeback(inode); | |
7994e6f7 | 556 | |
be7ce416 AV |
557 | if (op->evict_inode) { |
558 | op->evict_inode(inode); | |
b4272d4c AV |
559 | } else { |
560 | if (inode->i_data.nrpages) | |
561 | truncate_inode_pages(&inode->i_data, 0); | |
dbd5768f | 562 | clear_inode(inode); |
b4272d4c | 563 | } |
661074e9 AV |
564 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
565 | bd_forget(inode); | |
566 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) | |
567 | cd_forget(inode); | |
b2b2af8e DC |
568 | |
569 | remove_inode_hash(inode); | |
570 | ||
571 | spin_lock(&inode->i_lock); | |
572 | wake_up_bit(&inode->i_state, __I_NEW); | |
573 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); | |
574 | spin_unlock(&inode->i_lock); | |
575 | ||
576 | destroy_inode(inode); | |
b4272d4c AV |
577 | } |
578 | ||
1da177e4 LT |
579 | /* |
580 | * dispose_list - dispose of the contents of a local list | |
581 | * @head: the head of the list to free | |
582 | * | |
583 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
584 | * need to worry about list corruption and SMP locks. | |
585 | */ | |
586 | static void dispose_list(struct list_head *head) | |
587 | { | |
1da177e4 LT |
588 | while (!list_empty(head)) { |
589 | struct inode *inode; | |
590 | ||
7ccf19a8 NP |
591 | inode = list_first_entry(head, struct inode, i_lru); |
592 | list_del_init(&inode->i_lru); | |
1da177e4 | 593 | |
644da596 | 594 | evict(inode); |
1da177e4 | 595 | } |
1da177e4 LT |
596 | } |
597 | ||
63997e98 AV |
598 | /** |
599 | * evict_inodes - evict all evictable inodes for a superblock | |
600 | * @sb: superblock to operate on | |
601 | * | |
602 | * Make sure that no inodes with zero refcount are retained. This is | |
603 | * called by superblock shutdown after having MS_ACTIVE flag removed, | |
604 | * so any inode reaching zero refcount during or after that call will | |
605 | * be immediately evicted. | |
1da177e4 | 606 | */ |
63997e98 | 607 | void evict_inodes(struct super_block *sb) |
1da177e4 | 608 | { |
63997e98 AV |
609 | struct inode *inode, *next; |
610 | LIST_HEAD(dispose); | |
1da177e4 | 611 | |
55fa6091 | 612 | spin_lock(&inode_sb_list_lock); |
63997e98 AV |
613 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
614 | if (atomic_read(&inode->i_count)) | |
aabb8fdb | 615 | continue; |
250df6ed DC |
616 | |
617 | spin_lock(&inode->i_lock); | |
618 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
619 | spin_unlock(&inode->i_lock); | |
1da177e4 | 620 | continue; |
250df6ed | 621 | } |
63997e98 AV |
622 | |
623 | inode->i_state |= I_FREEING; | |
02afc410 | 624 | inode_lru_list_del(inode); |
250df6ed | 625 | spin_unlock(&inode->i_lock); |
02afc410 | 626 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 627 | } |
55fa6091 | 628 | spin_unlock(&inode_sb_list_lock); |
63997e98 AV |
629 | |
630 | dispose_list(&dispose); | |
1da177e4 LT |
631 | } |
632 | ||
1da177e4 | 633 | /** |
a0318786 CH |
634 | * invalidate_inodes - attempt to free all inodes on a superblock |
635 | * @sb: superblock to operate on | |
93b270f7 | 636 | * @kill_dirty: flag to guide handling of dirty inodes |
1da177e4 | 637 | * |
a0318786 CH |
638 | * Attempts to free all inodes for a given superblock. If there were any |
639 | * busy inodes return a non-zero value, else zero. | |
93b270f7 N |
640 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat |
641 | * them as busy. | |
1da177e4 | 642 | */ |
93b270f7 | 643 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) |
1da177e4 | 644 | { |
cffbc8aa | 645 | int busy = 0; |
a0318786 CH |
646 | struct inode *inode, *next; |
647 | LIST_HEAD(dispose); | |
1da177e4 | 648 | |
55fa6091 | 649 | spin_lock(&inode_sb_list_lock); |
a0318786 | 650 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
250df6ed DC |
651 | spin_lock(&inode->i_lock); |
652 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
653 | spin_unlock(&inode->i_lock); | |
aabb8fdb | 654 | continue; |
250df6ed | 655 | } |
93b270f7 | 656 | if (inode->i_state & I_DIRTY && !kill_dirty) { |
250df6ed | 657 | spin_unlock(&inode->i_lock); |
93b270f7 N |
658 | busy = 1; |
659 | continue; | |
660 | } | |
99a38919 | 661 | if (atomic_read(&inode->i_count)) { |
250df6ed | 662 | spin_unlock(&inode->i_lock); |
99a38919 | 663 | busy = 1; |
1da177e4 LT |
664 | continue; |
665 | } | |
99a38919 | 666 | |
99a38919 | 667 | inode->i_state |= I_FREEING; |
02afc410 | 668 | inode_lru_list_del(inode); |
250df6ed | 669 | spin_unlock(&inode->i_lock); |
02afc410 | 670 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 671 | } |
55fa6091 | 672 | spin_unlock(&inode_sb_list_lock); |
1da177e4 | 673 | |
a0318786 | 674 | dispose_list(&dispose); |
1da177e4 LT |
675 | |
676 | return busy; | |
677 | } | |
1da177e4 LT |
678 | |
679 | static int can_unuse(struct inode *inode) | |
680 | { | |
9e38d86f | 681 | if (inode->i_state & ~I_REFERENCED) |
1da177e4 LT |
682 | return 0; |
683 | if (inode_has_buffers(inode)) | |
684 | return 0; | |
685 | if (atomic_read(&inode->i_count)) | |
686 | return 0; | |
687 | if (inode->i_data.nrpages) | |
688 | return 0; | |
689 | return 1; | |
690 | } | |
691 | ||
692 | /* | |
b0d40c92 DC |
693 | * Walk the superblock inode LRU for freeable inodes and attempt to free them. |
694 | * This is called from the superblock shrinker function with a number of inodes | |
695 | * to trim from the LRU. Inodes to be freed are moved to a temporary list and | |
696 | * then are freed outside inode_lock by dispose_list(). | |
1da177e4 LT |
697 | * |
698 | * Any inodes which are pinned purely because of attached pagecache have their | |
9e38d86f NP |
699 | * pagecache removed. If the inode has metadata buffers attached to |
700 | * mapping->private_list then try to remove them. | |
1da177e4 | 701 | * |
9e38d86f NP |
702 | * If the inode has the I_REFERENCED flag set, then it means that it has been |
703 | * used recently - the flag is set in iput_final(). When we encounter such an | |
704 | * inode, clear the flag and move it to the back of the LRU so it gets another | |
705 | * pass through the LRU before it gets reclaimed. This is necessary because of | |
706 | * the fact we are doing lazy LRU updates to minimise lock contention so the | |
707 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes | |
708 | * with this flag set because they are the inodes that are out of order. | |
1da177e4 | 709 | */ |
b0d40c92 | 710 | void prune_icache_sb(struct super_block *sb, int nr_to_scan) |
1da177e4 LT |
711 | { |
712 | LIST_HEAD(freeable); | |
1da177e4 LT |
713 | int nr_scanned; |
714 | unsigned long reap = 0; | |
715 | ||
09cc9fc7 | 716 | spin_lock(&sb->s_inode_lru_lock); |
b0d40c92 | 717 | for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) { |
1da177e4 LT |
718 | struct inode *inode; |
719 | ||
98b745c6 | 720 | if (list_empty(&sb->s_inode_lru)) |
1da177e4 LT |
721 | break; |
722 | ||
98b745c6 | 723 | inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru); |
1da177e4 | 724 | |
02afc410 | 725 | /* |
09cc9fc7 | 726 | * we are inverting the sb->s_inode_lru_lock/inode->i_lock here, |
02afc410 DC |
727 | * so use a trylock. If we fail to get the lock, just move the |
728 | * inode to the back of the list so we don't spin on it. | |
729 | */ | |
730 | if (!spin_trylock(&inode->i_lock)) { | |
5b55d708 | 731 | list_move(&inode->i_lru, &sb->s_inode_lru); |
02afc410 DC |
732 | continue; |
733 | } | |
734 | ||
9e38d86f NP |
735 | /* |
736 | * Referenced or dirty inodes are still in use. Give them | |
737 | * another pass through the LRU as we canot reclaim them now. | |
738 | */ | |
739 | if (atomic_read(&inode->i_count) || | |
740 | (inode->i_state & ~I_REFERENCED)) { | |
7ccf19a8 | 741 | list_del_init(&inode->i_lru); |
f283c86a | 742 | spin_unlock(&inode->i_lock); |
98b745c6 | 743 | sb->s_nr_inodes_unused--; |
fcb94f72 | 744 | this_cpu_dec(nr_unused); |
9e38d86f NP |
745 | continue; |
746 | } | |
747 | ||
748 | /* recently referenced inodes get one more pass */ | |
749 | if (inode->i_state & I_REFERENCED) { | |
9e38d86f | 750 | inode->i_state &= ~I_REFERENCED; |
98b745c6 | 751 | list_move(&inode->i_lru, &sb->s_inode_lru); |
f283c86a | 752 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
753 | continue; |
754 | } | |
755 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { | |
756 | __iget(inode); | |
250df6ed | 757 | spin_unlock(&inode->i_lock); |
09cc9fc7 | 758 | spin_unlock(&sb->s_inode_lru_lock); |
1da177e4 | 759 | if (remove_inode_buffers(inode)) |
fc0ecff6 AM |
760 | reap += invalidate_mapping_pages(&inode->i_data, |
761 | 0, -1); | |
1da177e4 | 762 | iput(inode); |
09cc9fc7 | 763 | spin_lock(&sb->s_inode_lru_lock); |
1da177e4 | 764 | |
98b745c6 | 765 | if (inode != list_entry(sb->s_inode_lru.next, |
7ccf19a8 | 766 | struct inode, i_lru)) |
1da177e4 | 767 | continue; /* wrong inode or list_empty */ |
02afc410 DC |
768 | /* avoid lock inversions with trylock */ |
769 | if (!spin_trylock(&inode->i_lock)) | |
770 | continue; | |
250df6ed DC |
771 | if (!can_unuse(inode)) { |
772 | spin_unlock(&inode->i_lock); | |
1da177e4 | 773 | continue; |
250df6ed | 774 | } |
1da177e4 | 775 | } |
7ef0d737 | 776 | WARN_ON(inode->i_state & I_NEW); |
1da177e4 | 777 | inode->i_state |= I_FREEING; |
250df6ed | 778 | spin_unlock(&inode->i_lock); |
7ccf19a8 | 779 | |
7ccf19a8 | 780 | list_move(&inode->i_lru, &freeable); |
98b745c6 | 781 | sb->s_nr_inodes_unused--; |
fcb94f72 | 782 | this_cpu_dec(nr_unused); |
1da177e4 | 783 | } |
f8891e5e CL |
784 | if (current_is_kswapd()) |
785 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
786 | else | |
787 | __count_vm_events(PGINODESTEAL, reap); | |
09cc9fc7 | 788 | spin_unlock(&sb->s_inode_lru_lock); |
5f8aefd4 KK |
789 | if (current->reclaim_state) |
790 | current->reclaim_state->reclaimed_slab += reap; | |
1da177e4 LT |
791 | |
792 | dispose_list(&freeable); | |
1da177e4 LT |
793 | } |
794 | ||
1da177e4 LT |
795 | static void __wait_on_freeing_inode(struct inode *inode); |
796 | /* | |
797 | * Called with the inode lock held. | |
1da177e4 | 798 | */ |
6b3304b5 MK |
799 | static struct inode *find_inode(struct super_block *sb, |
800 | struct hlist_head *head, | |
801 | int (*test)(struct inode *, void *), | |
802 | void *data) | |
1da177e4 | 803 | { |
6b3304b5 | 804 | struct inode *inode = NULL; |
1da177e4 LT |
805 | |
806 | repeat: | |
b67bfe0d | 807 | hlist_for_each_entry(inode, head, i_hash) { |
67a23c49 DC |
808 | spin_lock(&inode->i_lock); |
809 | if (inode->i_sb != sb) { | |
810 | spin_unlock(&inode->i_lock); | |
1da177e4 | 811 | continue; |
67a23c49 DC |
812 | } |
813 | if (!test(inode, data)) { | |
814 | spin_unlock(&inode->i_lock); | |
1da177e4 | 815 | continue; |
67a23c49 | 816 | } |
a4ffdde6 | 817 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
818 | __wait_on_freeing_inode(inode); |
819 | goto repeat; | |
820 | } | |
f7899bd5 | 821 | __iget(inode); |
250df6ed | 822 | spin_unlock(&inode->i_lock); |
f7899bd5 | 823 | return inode; |
1da177e4 | 824 | } |
f7899bd5 | 825 | return NULL; |
1da177e4 LT |
826 | } |
827 | ||
828 | /* | |
829 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
830 | * iget_locked for details. | |
831 | */ | |
6b3304b5 MK |
832 | static struct inode *find_inode_fast(struct super_block *sb, |
833 | struct hlist_head *head, unsigned long ino) | |
1da177e4 | 834 | { |
6b3304b5 | 835 | struct inode *inode = NULL; |
1da177e4 LT |
836 | |
837 | repeat: | |
b67bfe0d | 838 | hlist_for_each_entry(inode, head, i_hash) { |
67a23c49 DC |
839 | spin_lock(&inode->i_lock); |
840 | if (inode->i_ino != ino) { | |
841 | spin_unlock(&inode->i_lock); | |
1da177e4 | 842 | continue; |
67a23c49 DC |
843 | } |
844 | if (inode->i_sb != sb) { | |
845 | spin_unlock(&inode->i_lock); | |
1da177e4 | 846 | continue; |
67a23c49 | 847 | } |
a4ffdde6 | 848 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
849 | __wait_on_freeing_inode(inode); |
850 | goto repeat; | |
851 | } | |
f7899bd5 | 852 | __iget(inode); |
250df6ed | 853 | spin_unlock(&inode->i_lock); |
f7899bd5 | 854 | return inode; |
1da177e4 | 855 | } |
f7899bd5 | 856 | return NULL; |
8290c35f DC |
857 | } |
858 | ||
f991bd2e ED |
859 | /* |
860 | * Each cpu owns a range of LAST_INO_BATCH numbers. | |
861 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, | |
862 | * to renew the exhausted range. | |
8290c35f | 863 | * |
f991bd2e ED |
864 | * This does not significantly increase overflow rate because every CPU can |
865 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is | |
866 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the | |
867 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase | |
868 | * overflow rate by 2x, which does not seem too significant. | |
869 | * | |
870 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
871 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
872 | * here to attempt to avoid that. | |
8290c35f | 873 | */ |
f991bd2e ED |
874 | #define LAST_INO_BATCH 1024 |
875 | static DEFINE_PER_CPU(unsigned int, last_ino); | |
876 | ||
85fe4025 | 877 | unsigned int get_next_ino(void) |
8290c35f | 878 | { |
f991bd2e ED |
879 | unsigned int *p = &get_cpu_var(last_ino); |
880 | unsigned int res = *p; | |
8290c35f | 881 | |
f991bd2e ED |
882 | #ifdef CONFIG_SMP |
883 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { | |
884 | static atomic_t shared_last_ino; | |
885 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); | |
886 | ||
887 | res = next - LAST_INO_BATCH; | |
888 | } | |
889 | #endif | |
890 | ||
891 | *p = ++res; | |
892 | put_cpu_var(last_ino); | |
893 | return res; | |
8290c35f | 894 | } |
85fe4025 | 895 | EXPORT_SYMBOL(get_next_ino); |
8290c35f | 896 | |
a209dfc7 ED |
897 | /** |
898 | * new_inode_pseudo - obtain an inode | |
899 | * @sb: superblock | |
900 | * | |
901 | * Allocates a new inode for given superblock. | |
902 | * Inode wont be chained in superblock s_inodes list | |
903 | * This means : | |
904 | * - fs can't be unmount | |
905 | * - quotas, fsnotify, writeback can't work | |
906 | */ | |
907 | struct inode *new_inode_pseudo(struct super_block *sb) | |
908 | { | |
909 | struct inode *inode = alloc_inode(sb); | |
910 | ||
911 | if (inode) { | |
912 | spin_lock(&inode->i_lock); | |
913 | inode->i_state = 0; | |
914 | spin_unlock(&inode->i_lock); | |
915 | INIT_LIST_HEAD(&inode->i_sb_list); | |
916 | } | |
917 | return inode; | |
918 | } | |
919 | ||
1da177e4 LT |
920 | /** |
921 | * new_inode - obtain an inode | |
922 | * @sb: superblock | |
923 | * | |
769848c0 | 924 | * Allocates a new inode for given superblock. The default gfp_mask |
3c1d4378 | 925 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
769848c0 MG |
926 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
927 | * for the page cache are not reclaimable or migratable, | |
928 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
929 | * newly created inode's mapping | |
930 | * | |
1da177e4 LT |
931 | */ |
932 | struct inode *new_inode(struct super_block *sb) | |
933 | { | |
6b3304b5 | 934 | struct inode *inode; |
1da177e4 | 935 | |
55fa6091 | 936 | spin_lock_prefetch(&inode_sb_list_lock); |
6b3304b5 | 937 | |
a209dfc7 ED |
938 | inode = new_inode_pseudo(sb); |
939 | if (inode) | |
55fa6091 | 940 | inode_sb_list_add(inode); |
1da177e4 LT |
941 | return inode; |
942 | } | |
1da177e4 LT |
943 | EXPORT_SYMBOL(new_inode); |
944 | ||
14358e6d | 945 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
e096d0c7 JB |
946 | void lockdep_annotate_inode_mutex_key(struct inode *inode) |
947 | { | |
a3314a0e | 948 | if (S_ISDIR(inode->i_mode)) { |
1e89a5e1 PZ |
949 | struct file_system_type *type = inode->i_sb->s_type; |
950 | ||
9a7aa12f | 951 | /* Set new key only if filesystem hasn't already changed it */ |
978d6d8c | 952 | if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) { |
9a7aa12f JK |
953 | /* |
954 | * ensure nobody is actually holding i_mutex | |
955 | */ | |
956 | mutex_destroy(&inode->i_mutex); | |
957 | mutex_init(&inode->i_mutex); | |
958 | lockdep_set_class(&inode->i_mutex, | |
959 | &type->i_mutex_dir_key); | |
960 | } | |
1e89a5e1 | 961 | } |
e096d0c7 JB |
962 | } |
963 | EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key); | |
14358e6d | 964 | #endif |
e096d0c7 JB |
965 | |
966 | /** | |
967 | * unlock_new_inode - clear the I_NEW state and wake up any waiters | |
968 | * @inode: new inode to unlock | |
969 | * | |
970 | * Called when the inode is fully initialised to clear the new state of the | |
971 | * inode and wake up anyone waiting for the inode to finish initialisation. | |
972 | */ | |
973 | void unlock_new_inode(struct inode *inode) | |
974 | { | |
975 | lockdep_annotate_inode_mutex_key(inode); | |
250df6ed | 976 | spin_lock(&inode->i_lock); |
eaff8079 CH |
977 | WARN_ON(!(inode->i_state & I_NEW)); |
978 | inode->i_state &= ~I_NEW; | |
310fa7a3 | 979 | smp_mb(); |
250df6ed DC |
980 | wake_up_bit(&inode->i_state, __I_NEW); |
981 | spin_unlock(&inode->i_lock); | |
1da177e4 | 982 | } |
1da177e4 LT |
983 | EXPORT_SYMBOL(unlock_new_inode); |
984 | ||
0b2d0724 CH |
985 | /** |
986 | * iget5_locked - obtain an inode from a mounted file system | |
987 | * @sb: super block of file system | |
988 | * @hashval: hash value (usually inode number) to get | |
989 | * @test: callback used for comparisons between inodes | |
990 | * @set: callback used to initialize a new struct inode | |
991 | * @data: opaque data pointer to pass to @test and @set | |
992 | * | |
993 | * Search for the inode specified by @hashval and @data in the inode cache, | |
994 | * and if present it is return it with an increased reference count. This is | |
995 | * a generalized version of iget_locked() for file systems where the inode | |
996 | * number is not sufficient for unique identification of an inode. | |
997 | * | |
998 | * If the inode is not in cache, allocate a new inode and return it locked, | |
999 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
1000 | * before unlocking it via unlock_new_inode(). | |
1da177e4 | 1001 | * |
0b2d0724 CH |
1002 | * Note both @test and @set are called with the inode_hash_lock held, so can't |
1003 | * sleep. | |
1da177e4 | 1004 | */ |
0b2d0724 CH |
1005 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, |
1006 | int (*test)(struct inode *, void *), | |
1007 | int (*set)(struct inode *, void *), void *data) | |
1da177e4 | 1008 | { |
0b2d0724 | 1009 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
6b3304b5 | 1010 | struct inode *inode; |
1da177e4 | 1011 | |
0b2d0724 CH |
1012 | spin_lock(&inode_hash_lock); |
1013 | inode = find_inode(sb, head, test, data); | |
1014 | spin_unlock(&inode_hash_lock); | |
1015 | ||
1016 | if (inode) { | |
1017 | wait_on_inode(inode); | |
1018 | return inode; | |
1019 | } | |
1020 | ||
1da177e4 LT |
1021 | inode = alloc_inode(sb); |
1022 | if (inode) { | |
6b3304b5 | 1023 | struct inode *old; |
1da177e4 | 1024 | |
67a23c49 | 1025 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1026 | /* We released the lock, so.. */ |
1027 | old = find_inode(sb, head, test, data); | |
1028 | if (!old) { | |
1029 | if (set(inode, data)) | |
1030 | goto set_failed; | |
1031 | ||
250df6ed DC |
1032 | spin_lock(&inode->i_lock); |
1033 | inode->i_state = I_NEW; | |
646ec461 | 1034 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1035 | spin_unlock(&inode->i_lock); |
55fa6091 | 1036 | inode_sb_list_add(inode); |
67a23c49 | 1037 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1038 | |
1039 | /* Return the locked inode with I_NEW set, the | |
1040 | * caller is responsible for filling in the contents | |
1041 | */ | |
1042 | return inode; | |
1043 | } | |
1044 | ||
1045 | /* | |
1046 | * Uhhuh, somebody else created the same inode under | |
1047 | * us. Use the old inode instead of the one we just | |
1048 | * allocated. | |
1049 | */ | |
67a23c49 | 1050 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1051 | destroy_inode(inode); |
1052 | inode = old; | |
1053 | wait_on_inode(inode); | |
1054 | } | |
1055 | return inode; | |
1056 | ||
1057 | set_failed: | |
67a23c49 | 1058 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1059 | destroy_inode(inode); |
1060 | return NULL; | |
1061 | } | |
0b2d0724 | 1062 | EXPORT_SYMBOL(iget5_locked); |
1da177e4 | 1063 | |
0b2d0724 CH |
1064 | /** |
1065 | * iget_locked - obtain an inode from a mounted file system | |
1066 | * @sb: super block of file system | |
1067 | * @ino: inode number to get | |
1068 | * | |
1069 | * Search for the inode specified by @ino in the inode cache and if present | |
1070 | * return it with an increased reference count. This is for file systems | |
1071 | * where the inode number is sufficient for unique identification of an inode. | |
1072 | * | |
1073 | * If the inode is not in cache, allocate a new inode and return it locked, | |
1074 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
1075 | * before unlocking it via unlock_new_inode(). | |
1da177e4 | 1076 | */ |
0b2d0724 | 1077 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) |
1da177e4 | 1078 | { |
0b2d0724 | 1079 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
6b3304b5 | 1080 | struct inode *inode; |
1da177e4 | 1081 | |
0b2d0724 CH |
1082 | spin_lock(&inode_hash_lock); |
1083 | inode = find_inode_fast(sb, head, ino); | |
1084 | spin_unlock(&inode_hash_lock); | |
1085 | if (inode) { | |
1086 | wait_on_inode(inode); | |
1087 | return inode; | |
1088 | } | |
1089 | ||
1da177e4 LT |
1090 | inode = alloc_inode(sb); |
1091 | if (inode) { | |
6b3304b5 | 1092 | struct inode *old; |
1da177e4 | 1093 | |
67a23c49 | 1094 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1095 | /* We released the lock, so.. */ |
1096 | old = find_inode_fast(sb, head, ino); | |
1097 | if (!old) { | |
1098 | inode->i_ino = ino; | |
250df6ed DC |
1099 | spin_lock(&inode->i_lock); |
1100 | inode->i_state = I_NEW; | |
646ec461 | 1101 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1102 | spin_unlock(&inode->i_lock); |
55fa6091 | 1103 | inode_sb_list_add(inode); |
67a23c49 | 1104 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1105 | |
1106 | /* Return the locked inode with I_NEW set, the | |
1107 | * caller is responsible for filling in the contents | |
1108 | */ | |
1109 | return inode; | |
1110 | } | |
1111 | ||
1112 | /* | |
1113 | * Uhhuh, somebody else created the same inode under | |
1114 | * us. Use the old inode instead of the one we just | |
1115 | * allocated. | |
1116 | */ | |
67a23c49 | 1117 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1118 | destroy_inode(inode); |
1119 | inode = old; | |
1120 | wait_on_inode(inode); | |
1121 | } | |
1122 | return inode; | |
1123 | } | |
0b2d0724 | 1124 | EXPORT_SYMBOL(iget_locked); |
1da177e4 | 1125 | |
ad5e195a CH |
1126 | /* |
1127 | * search the inode cache for a matching inode number. | |
1128 | * If we find one, then the inode number we are trying to | |
1129 | * allocate is not unique and so we should not use it. | |
1130 | * | |
1131 | * Returns 1 if the inode number is unique, 0 if it is not. | |
1132 | */ | |
1133 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) | |
1134 | { | |
1135 | struct hlist_head *b = inode_hashtable + hash(sb, ino); | |
ad5e195a CH |
1136 | struct inode *inode; |
1137 | ||
67a23c49 | 1138 | spin_lock(&inode_hash_lock); |
b67bfe0d | 1139 | hlist_for_each_entry(inode, b, i_hash) { |
67a23c49 DC |
1140 | if (inode->i_ino == ino && inode->i_sb == sb) { |
1141 | spin_unlock(&inode_hash_lock); | |
ad5e195a | 1142 | return 0; |
67a23c49 | 1143 | } |
ad5e195a | 1144 | } |
67a23c49 | 1145 | spin_unlock(&inode_hash_lock); |
ad5e195a CH |
1146 | |
1147 | return 1; | |
1148 | } | |
1149 | ||
1da177e4 LT |
1150 | /** |
1151 | * iunique - get a unique inode number | |
1152 | * @sb: superblock | |
1153 | * @max_reserved: highest reserved inode number | |
1154 | * | |
1155 | * Obtain an inode number that is unique on the system for a given | |
1156 | * superblock. This is used by file systems that have no natural | |
1157 | * permanent inode numbering system. An inode number is returned that | |
1158 | * is higher than the reserved limit but unique. | |
1159 | * | |
1160 | * BUGS: | |
1161 | * With a large number of inodes live on the file system this function | |
1162 | * currently becomes quite slow. | |
1163 | */ | |
1164 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
1165 | { | |
866b04fc JL |
1166 | /* |
1167 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
1168 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
1169 | * here to attempt to avoid that. | |
1170 | */ | |
ad5e195a | 1171 | static DEFINE_SPINLOCK(iunique_lock); |
866b04fc | 1172 | static unsigned int counter; |
1da177e4 | 1173 | ino_t res; |
3361c7be | 1174 | |
ad5e195a | 1175 | spin_lock(&iunique_lock); |
3361c7be JL |
1176 | do { |
1177 | if (counter <= max_reserved) | |
1178 | counter = max_reserved + 1; | |
1da177e4 | 1179 | res = counter++; |
ad5e195a CH |
1180 | } while (!test_inode_iunique(sb, res)); |
1181 | spin_unlock(&iunique_lock); | |
1da177e4 | 1182 | |
3361c7be JL |
1183 | return res; |
1184 | } | |
1da177e4 LT |
1185 | EXPORT_SYMBOL(iunique); |
1186 | ||
1187 | struct inode *igrab(struct inode *inode) | |
1188 | { | |
250df6ed DC |
1189 | spin_lock(&inode->i_lock); |
1190 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { | |
1da177e4 | 1191 | __iget(inode); |
250df6ed DC |
1192 | spin_unlock(&inode->i_lock); |
1193 | } else { | |
1194 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1195 | /* |
1196 | * Handle the case where s_op->clear_inode is not been | |
1197 | * called yet, and somebody is calling igrab | |
1198 | * while the inode is getting freed. | |
1199 | */ | |
1200 | inode = NULL; | |
250df6ed | 1201 | } |
1da177e4 LT |
1202 | return inode; |
1203 | } | |
1da177e4 LT |
1204 | EXPORT_SYMBOL(igrab); |
1205 | ||
1206 | /** | |
0b2d0724 | 1207 | * ilookup5_nowait - search for an inode in the inode cache |
1da177e4 | 1208 | * @sb: super block of file system to search |
0b2d0724 | 1209 | * @hashval: hash value (usually inode number) to search for |
1da177e4 LT |
1210 | * @test: callback used for comparisons between inodes |
1211 | * @data: opaque data pointer to pass to @test | |
1da177e4 | 1212 | * |
0b2d0724 | 1213 | * Search for the inode specified by @hashval and @data in the inode cache. |
1da177e4 LT |
1214 | * If the inode is in the cache, the inode is returned with an incremented |
1215 | * reference count. | |
1216 | * | |
0b2d0724 CH |
1217 | * Note: I_NEW is not waited upon so you have to be very careful what you do |
1218 | * with the returned inode. You probably should be using ilookup5() instead. | |
1da177e4 | 1219 | * |
b6d0ad68 | 1220 | * Note2: @test is called with the inode_hash_lock held, so can't sleep. |
1da177e4 | 1221 | */ |
0b2d0724 CH |
1222 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, |
1223 | int (*test)(struct inode *, void *), void *data) | |
1da177e4 | 1224 | { |
0b2d0724 | 1225 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
1da177e4 LT |
1226 | struct inode *inode; |
1227 | ||
67a23c49 | 1228 | spin_lock(&inode_hash_lock); |
1da177e4 | 1229 | inode = find_inode(sb, head, test, data); |
67a23c49 | 1230 | spin_unlock(&inode_hash_lock); |
88bd5121 | 1231 | |
0b2d0724 | 1232 | return inode; |
88bd5121 | 1233 | } |
88bd5121 AA |
1234 | EXPORT_SYMBOL(ilookup5_nowait); |
1235 | ||
1236 | /** | |
1237 | * ilookup5 - search for an inode in the inode cache | |
1238 | * @sb: super block of file system to search | |
1239 | * @hashval: hash value (usually inode number) to search for | |
1240 | * @test: callback used for comparisons between inodes | |
1241 | * @data: opaque data pointer to pass to @test | |
1242 | * | |
0b2d0724 CH |
1243 | * Search for the inode specified by @hashval and @data in the inode cache, |
1244 | * and if the inode is in the cache, return the inode with an incremented | |
1245 | * reference count. Waits on I_NEW before returning the inode. | |
88bd5121 | 1246 | * returned with an incremented reference count. |
1da177e4 | 1247 | * |
0b2d0724 CH |
1248 | * This is a generalized version of ilookup() for file systems where the |
1249 | * inode number is not sufficient for unique identification of an inode. | |
1da177e4 | 1250 | * |
0b2d0724 | 1251 | * Note: @test is called with the inode_hash_lock held, so can't sleep. |
1da177e4 LT |
1252 | */ |
1253 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
1254 | int (*test)(struct inode *, void *), void *data) | |
1255 | { | |
0b2d0724 | 1256 | struct inode *inode = ilookup5_nowait(sb, hashval, test, data); |
1da177e4 | 1257 | |
0b2d0724 CH |
1258 | if (inode) |
1259 | wait_on_inode(inode); | |
1260 | return inode; | |
1da177e4 | 1261 | } |
1da177e4 LT |
1262 | EXPORT_SYMBOL(ilookup5); |
1263 | ||
1264 | /** | |
1265 | * ilookup - search for an inode in the inode cache | |
1266 | * @sb: super block of file system to search | |
1267 | * @ino: inode number to search for | |
1268 | * | |
0b2d0724 CH |
1269 | * Search for the inode @ino in the inode cache, and if the inode is in the |
1270 | * cache, the inode is returned with an incremented reference count. | |
1da177e4 LT |
1271 | */ |
1272 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1273 | { | |
1274 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1da177e4 LT |
1275 | struct inode *inode; |
1276 | ||
0b2d0724 CH |
1277 | spin_lock(&inode_hash_lock); |
1278 | inode = find_inode_fast(sb, head, ino); | |
1279 | spin_unlock(&inode_hash_lock); | |
1da177e4 | 1280 | |
1da177e4 | 1281 | if (inode) |
0b2d0724 CH |
1282 | wait_on_inode(inode); |
1283 | return inode; | |
1da177e4 | 1284 | } |
0b2d0724 | 1285 | EXPORT_SYMBOL(ilookup); |
1da177e4 | 1286 | |
261bca86 AV |
1287 | int insert_inode_locked(struct inode *inode) |
1288 | { | |
1289 | struct super_block *sb = inode->i_sb; | |
1290 | ino_t ino = inode->i_ino; | |
1291 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
261bca86 | 1292 | |
261bca86 | 1293 | while (1) { |
72a43d63 | 1294 | struct inode *old = NULL; |
67a23c49 | 1295 | spin_lock(&inode_hash_lock); |
b67bfe0d | 1296 | hlist_for_each_entry(old, head, i_hash) { |
72a43d63 AV |
1297 | if (old->i_ino != ino) |
1298 | continue; | |
1299 | if (old->i_sb != sb) | |
1300 | continue; | |
250df6ed DC |
1301 | spin_lock(&old->i_lock); |
1302 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1303 | spin_unlock(&old->i_lock); | |
72a43d63 | 1304 | continue; |
250df6ed | 1305 | } |
72a43d63 AV |
1306 | break; |
1307 | } | |
b67bfe0d | 1308 | if (likely(!old)) { |
250df6ed DC |
1309 | spin_lock(&inode->i_lock); |
1310 | inode->i_state |= I_NEW; | |
261bca86 | 1311 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1312 | spin_unlock(&inode->i_lock); |
67a23c49 | 1313 | spin_unlock(&inode_hash_lock); |
261bca86 AV |
1314 | return 0; |
1315 | } | |
1316 | __iget(old); | |
250df6ed | 1317 | spin_unlock(&old->i_lock); |
67a23c49 | 1318 | spin_unlock(&inode_hash_lock); |
261bca86 | 1319 | wait_on_inode(old); |
1d3382cb | 1320 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1321 | iput(old); |
1322 | return -EBUSY; | |
1323 | } | |
1324 | iput(old); | |
1325 | } | |
1326 | } | |
261bca86 AV |
1327 | EXPORT_SYMBOL(insert_inode_locked); |
1328 | ||
1329 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1330 | int (*test)(struct inode *, void *), void *data) | |
1331 | { | |
1332 | struct super_block *sb = inode->i_sb; | |
1333 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
261bca86 | 1334 | |
261bca86 | 1335 | while (1) { |
72a43d63 AV |
1336 | struct inode *old = NULL; |
1337 | ||
67a23c49 | 1338 | spin_lock(&inode_hash_lock); |
b67bfe0d | 1339 | hlist_for_each_entry(old, head, i_hash) { |
72a43d63 AV |
1340 | if (old->i_sb != sb) |
1341 | continue; | |
1342 | if (!test(old, data)) | |
1343 | continue; | |
250df6ed DC |
1344 | spin_lock(&old->i_lock); |
1345 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1346 | spin_unlock(&old->i_lock); | |
72a43d63 | 1347 | continue; |
250df6ed | 1348 | } |
72a43d63 AV |
1349 | break; |
1350 | } | |
b67bfe0d | 1351 | if (likely(!old)) { |
250df6ed DC |
1352 | spin_lock(&inode->i_lock); |
1353 | inode->i_state |= I_NEW; | |
261bca86 | 1354 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1355 | spin_unlock(&inode->i_lock); |
67a23c49 | 1356 | spin_unlock(&inode_hash_lock); |
261bca86 AV |
1357 | return 0; |
1358 | } | |
1359 | __iget(old); | |
250df6ed | 1360 | spin_unlock(&old->i_lock); |
67a23c49 | 1361 | spin_unlock(&inode_hash_lock); |
261bca86 | 1362 | wait_on_inode(old); |
1d3382cb | 1363 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1364 | iput(old); |
1365 | return -EBUSY; | |
1366 | } | |
1367 | iput(old); | |
1368 | } | |
1369 | } | |
261bca86 AV |
1370 | EXPORT_SYMBOL(insert_inode_locked4); |
1371 | ||
1da177e4 | 1372 | |
45321ac5 AV |
1373 | int generic_delete_inode(struct inode *inode) |
1374 | { | |
1375 | return 1; | |
1376 | } | |
1377 | EXPORT_SYMBOL(generic_delete_inode); | |
1378 | ||
45321ac5 AV |
1379 | /* |
1380 | * Called when we're dropping the last reference | |
1381 | * to an inode. | |
22fe4042 | 1382 | * |
45321ac5 AV |
1383 | * Call the FS "drop_inode()" function, defaulting to |
1384 | * the legacy UNIX filesystem behaviour. If it tells | |
1385 | * us to evict inode, do so. Otherwise, retain inode | |
1386 | * in cache if fs is alive, sync and evict if fs is | |
1387 | * shutting down. | |
22fe4042 | 1388 | */ |
45321ac5 | 1389 | static void iput_final(struct inode *inode) |
1da177e4 LT |
1390 | { |
1391 | struct super_block *sb = inode->i_sb; | |
45321ac5 AV |
1392 | const struct super_operations *op = inode->i_sb->s_op; |
1393 | int drop; | |
1394 | ||
250df6ed DC |
1395 | WARN_ON(inode->i_state & I_NEW); |
1396 | ||
e7f59097 | 1397 | if (op->drop_inode) |
45321ac5 AV |
1398 | drop = op->drop_inode(inode); |
1399 | else | |
1400 | drop = generic_drop_inode(inode); | |
1da177e4 | 1401 | |
b2b2af8e DC |
1402 | if (!drop && (sb->s_flags & MS_ACTIVE)) { |
1403 | inode->i_state |= I_REFERENCED; | |
4eff96dd | 1404 | inode_add_lru(inode); |
b2b2af8e | 1405 | spin_unlock(&inode->i_lock); |
b2b2af8e DC |
1406 | return; |
1407 | } | |
1408 | ||
45321ac5 | 1409 | if (!drop) { |
991114c6 | 1410 | inode->i_state |= I_WILL_FREE; |
250df6ed | 1411 | spin_unlock(&inode->i_lock); |
1da177e4 | 1412 | write_inode_now(inode, 1); |
250df6ed | 1413 | spin_lock(&inode->i_lock); |
7ef0d737 | 1414 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1415 | inode->i_state &= ~I_WILL_FREE; |
1da177e4 | 1416 | } |
7ccf19a8 | 1417 | |
991114c6 | 1418 | inode->i_state |= I_FREEING; |
c4ae0c65 ED |
1419 | if (!list_empty(&inode->i_lru)) |
1420 | inode_lru_list_del(inode); | |
b2b2af8e | 1421 | spin_unlock(&inode->i_lock); |
b2b2af8e | 1422 | |
644da596 | 1423 | evict(inode); |
1da177e4 LT |
1424 | } |
1425 | ||
1da177e4 | 1426 | /** |
6b3304b5 | 1427 | * iput - put an inode |
1da177e4 LT |
1428 | * @inode: inode to put |
1429 | * | |
1430 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1431 | * zero, the inode is then freed and may also be destroyed. | |
1432 | * | |
1433 | * Consequently, iput() can sleep. | |
1434 | */ | |
1435 | void iput(struct inode *inode) | |
1436 | { | |
1437 | if (inode) { | |
a4ffdde6 | 1438 | BUG_ON(inode->i_state & I_CLEAR); |
1da177e4 | 1439 | |
f283c86a | 1440 | if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) |
1da177e4 LT |
1441 | iput_final(inode); |
1442 | } | |
1443 | } | |
1da177e4 LT |
1444 | EXPORT_SYMBOL(iput); |
1445 | ||
1446 | /** | |
1447 | * bmap - find a block number in a file | |
1448 | * @inode: inode of file | |
1449 | * @block: block to find | |
1450 | * | |
1451 | * Returns the block number on the device holding the inode that | |
1452 | * is the disk block number for the block of the file requested. | |
1453 | * That is, asked for block 4 of inode 1 the function will return the | |
6b3304b5 | 1454 | * disk block relative to the disk start that holds that block of the |
1da177e4 LT |
1455 | * file. |
1456 | */ | |
6b3304b5 | 1457 | sector_t bmap(struct inode *inode, sector_t block) |
1da177e4 LT |
1458 | { |
1459 | sector_t res = 0; | |
1460 | if (inode->i_mapping->a_ops->bmap) | |
1461 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1462 | return res; | |
1463 | } | |
1da177e4 LT |
1464 | EXPORT_SYMBOL(bmap); |
1465 | ||
11ff6f05 MG |
1466 | /* |
1467 | * With relative atime, only update atime if the previous atime is | |
1468 | * earlier than either the ctime or mtime or if at least a day has | |
1469 | * passed since the last atime update. | |
1470 | */ | |
1471 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1472 | struct timespec now) | |
1473 | { | |
1474 | ||
1475 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1476 | return 1; | |
1477 | /* | |
1478 | * Is mtime younger than atime? If yes, update atime: | |
1479 | */ | |
1480 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1481 | return 1; | |
1482 | /* | |
1483 | * Is ctime younger than atime? If yes, update atime: | |
1484 | */ | |
1485 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1486 | return 1; | |
1487 | ||
1488 | /* | |
1489 | * Is the previous atime value older than a day? If yes, | |
1490 | * update atime: | |
1491 | */ | |
1492 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1493 | return 1; | |
1494 | /* | |
1495 | * Good, we can skip the atime update: | |
1496 | */ | |
1497 | return 0; | |
1498 | } | |
1499 | ||
c3b2da31 JB |
1500 | /* |
1501 | * This does the actual work of updating an inodes time or version. Must have | |
1502 | * had called mnt_want_write() before calling this. | |
1503 | */ | |
1504 | static int update_time(struct inode *inode, struct timespec *time, int flags) | |
1505 | { | |
1506 | if (inode->i_op->update_time) | |
1507 | return inode->i_op->update_time(inode, time, flags); | |
1508 | ||
1509 | if (flags & S_ATIME) | |
1510 | inode->i_atime = *time; | |
1511 | if (flags & S_VERSION) | |
1512 | inode_inc_iversion(inode); | |
1513 | if (flags & S_CTIME) | |
1514 | inode->i_ctime = *time; | |
1515 | if (flags & S_MTIME) | |
1516 | inode->i_mtime = *time; | |
1517 | mark_inode_dirty_sync(inode); | |
1518 | return 0; | |
1519 | } | |
1520 | ||
1da177e4 | 1521 | /** |
869243a0 | 1522 | * touch_atime - update the access time |
185553b2 | 1523 | * @path: the &struct path to update |
1da177e4 LT |
1524 | * |
1525 | * Update the accessed time on an inode and mark it for writeback. | |
1526 | * This function automatically handles read only file systems and media, | |
1527 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1528 | */ | |
68ac1234 | 1529 | void touch_atime(struct path *path) |
1da177e4 | 1530 | { |
68ac1234 AV |
1531 | struct vfsmount *mnt = path->mnt; |
1532 | struct inode *inode = path->dentry->d_inode; | |
1da177e4 LT |
1533 | struct timespec now; |
1534 | ||
cdb70f3f | 1535 | if (inode->i_flags & S_NOATIME) |
b12536c2 | 1536 | return; |
37756ced | 1537 | if (IS_NOATIME(inode)) |
b12536c2 | 1538 | return; |
b2276138 | 1539 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1540 | return; |
47ae32d6 | 1541 | |
cdb70f3f | 1542 | if (mnt->mnt_flags & MNT_NOATIME) |
b12536c2 | 1543 | return; |
cdb70f3f | 1544 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1545 | return; |
1da177e4 LT |
1546 | |
1547 | now = current_fs_time(inode->i_sb); | |
11ff6f05 MG |
1548 | |
1549 | if (!relatime_need_update(mnt, inode, now)) | |
b12536c2 | 1550 | return; |
11ff6f05 | 1551 | |
47ae32d6 | 1552 | if (timespec_equal(&inode->i_atime, &now)) |
b12536c2 AK |
1553 | return; |
1554 | ||
5d37e9e6 | 1555 | if (!sb_start_write_trylock(inode->i_sb)) |
b12536c2 | 1556 | return; |
47ae32d6 | 1557 | |
5d37e9e6 JK |
1558 | if (__mnt_want_write(mnt)) |
1559 | goto skip_update; | |
c3b2da31 JB |
1560 | /* |
1561 | * File systems can error out when updating inodes if they need to | |
1562 | * allocate new space to modify an inode (such is the case for | |
1563 | * Btrfs), but since we touch atime while walking down the path we | |
1564 | * really don't care if we failed to update the atime of the file, | |
1565 | * so just ignore the return value. | |
2bc55652 AB |
1566 | * We may also fail on filesystems that have the ability to make parts |
1567 | * of the fs read only, e.g. subvolumes in Btrfs. | |
c3b2da31 JB |
1568 | */ |
1569 | update_time(inode, &now, S_ATIME); | |
5d37e9e6 JK |
1570 | __mnt_drop_write(mnt); |
1571 | skip_update: | |
1572 | sb_end_write(inode->i_sb); | |
1da177e4 | 1573 | } |
869243a0 | 1574 | EXPORT_SYMBOL(touch_atime); |
1da177e4 | 1575 | |
3ed37648 CW |
1576 | /* |
1577 | * The logic we want is | |
1578 | * | |
1579 | * if suid or (sgid and xgrp) | |
1580 | * remove privs | |
1581 | */ | |
1582 | int should_remove_suid(struct dentry *dentry) | |
1583 | { | |
1584 | umode_t mode = dentry->d_inode->i_mode; | |
1585 | int kill = 0; | |
1586 | ||
1587 | /* suid always must be killed */ | |
1588 | if (unlikely(mode & S_ISUID)) | |
1589 | kill = ATTR_KILL_SUID; | |
1590 | ||
1591 | /* | |
1592 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1593 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1594 | */ | |
1595 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1596 | kill |= ATTR_KILL_SGID; | |
1597 | ||
1598 | if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) | |
1599 | return kill; | |
1600 | ||
1601 | return 0; | |
1602 | } | |
1603 | EXPORT_SYMBOL(should_remove_suid); | |
1604 | ||
1605 | static int __remove_suid(struct dentry *dentry, int kill) | |
1606 | { | |
1607 | struct iattr newattrs; | |
1608 | ||
1609 | newattrs.ia_valid = ATTR_FORCE | kill; | |
1610 | return notify_change(dentry, &newattrs); | |
1611 | } | |
1612 | ||
1613 | int file_remove_suid(struct file *file) | |
1614 | { | |
1615 | struct dentry *dentry = file->f_path.dentry; | |
1616 | struct inode *inode = dentry->d_inode; | |
1617 | int killsuid; | |
1618 | int killpriv; | |
1619 | int error = 0; | |
1620 | ||
1621 | /* Fast path for nothing security related */ | |
1622 | if (IS_NOSEC(inode)) | |
1623 | return 0; | |
1624 | ||
1625 | killsuid = should_remove_suid(dentry); | |
1626 | killpriv = security_inode_need_killpriv(dentry); | |
1627 | ||
1628 | if (killpriv < 0) | |
1629 | return killpriv; | |
1630 | if (killpriv) | |
1631 | error = security_inode_killpriv(dentry); | |
1632 | if (!error && killsuid) | |
1633 | error = __remove_suid(dentry, killsuid); | |
9b9724dc JK |
1634 | if (!error) |
1635 | inode_has_no_xattr(inode); | |
3ed37648 CW |
1636 | |
1637 | return error; | |
1638 | } | |
1639 | EXPORT_SYMBOL(file_remove_suid); | |
1640 | ||
1da177e4 | 1641 | /** |
870f4817 CH |
1642 | * file_update_time - update mtime and ctime time |
1643 | * @file: file accessed | |
1da177e4 | 1644 | * |
870f4817 CH |
1645 | * Update the mtime and ctime members of an inode and mark the inode |
1646 | * for writeback. Note that this function is meant exclusively for | |
1647 | * usage in the file write path of filesystems, and filesystems may | |
1648 | * choose to explicitly ignore update via this function with the | |
2eadfc0e | 1649 | * S_NOCMTIME inode flag, e.g. for network filesystem where these |
c3b2da31 JB |
1650 | * timestamps are handled by the server. This can return an error for |
1651 | * file systems who need to allocate space in order to update an inode. | |
1da177e4 LT |
1652 | */ |
1653 | ||
c3b2da31 | 1654 | int file_update_time(struct file *file) |
1da177e4 | 1655 | { |
496ad9aa | 1656 | struct inode *inode = file_inode(file); |
1da177e4 | 1657 | struct timespec now; |
c3b2da31 JB |
1658 | int sync_it = 0; |
1659 | int ret; | |
1da177e4 | 1660 | |
ce06e0b2 | 1661 | /* First try to exhaust all avenues to not sync */ |
1da177e4 | 1662 | if (IS_NOCMTIME(inode)) |
c3b2da31 | 1663 | return 0; |
20ddee2c | 1664 | |
1da177e4 | 1665 | now = current_fs_time(inode->i_sb); |
ce06e0b2 AK |
1666 | if (!timespec_equal(&inode->i_mtime, &now)) |
1667 | sync_it = S_MTIME; | |
1da177e4 | 1668 | |
ce06e0b2 AK |
1669 | if (!timespec_equal(&inode->i_ctime, &now)) |
1670 | sync_it |= S_CTIME; | |
870f4817 | 1671 | |
ce06e0b2 AK |
1672 | if (IS_I_VERSION(inode)) |
1673 | sync_it |= S_VERSION; | |
7a224228 | 1674 | |
ce06e0b2 | 1675 | if (!sync_it) |
c3b2da31 | 1676 | return 0; |
ce06e0b2 AK |
1677 | |
1678 | /* Finally allowed to write? Takes lock. */ | |
eb04c282 | 1679 | if (__mnt_want_write_file(file)) |
c3b2da31 | 1680 | return 0; |
ce06e0b2 | 1681 | |
c3b2da31 | 1682 | ret = update_time(inode, &now, sync_it); |
eb04c282 | 1683 | __mnt_drop_write_file(file); |
c3b2da31 JB |
1684 | |
1685 | return ret; | |
1da177e4 | 1686 | } |
870f4817 | 1687 | EXPORT_SYMBOL(file_update_time); |
1da177e4 LT |
1688 | |
1689 | int inode_needs_sync(struct inode *inode) | |
1690 | { | |
1691 | if (IS_SYNC(inode)) | |
1692 | return 1; | |
1693 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1694 | return 1; | |
1695 | return 0; | |
1696 | } | |
1da177e4 LT |
1697 | EXPORT_SYMBOL(inode_needs_sync); |
1698 | ||
1da177e4 LT |
1699 | int inode_wait(void *word) |
1700 | { | |
1701 | schedule(); | |
1702 | return 0; | |
1703 | } | |
d44dab8d | 1704 | EXPORT_SYMBOL(inode_wait); |
1da177e4 LT |
1705 | |
1706 | /* | |
168a9fd6 MS |
1707 | * If we try to find an inode in the inode hash while it is being |
1708 | * deleted, we have to wait until the filesystem completes its | |
1709 | * deletion before reporting that it isn't found. This function waits | |
1710 | * until the deletion _might_ have completed. Callers are responsible | |
1711 | * to recheck inode state. | |
1712 | * | |
eaff8079 | 1713 | * It doesn't matter if I_NEW is not set initially, a call to |
250df6ed DC |
1714 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list |
1715 | * will DTRT. | |
1da177e4 LT |
1716 | */ |
1717 | static void __wait_on_freeing_inode(struct inode *inode) | |
1718 | { | |
1719 | wait_queue_head_t *wq; | |
eaff8079 CH |
1720 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); |
1721 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1da177e4 | 1722 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
250df6ed | 1723 | spin_unlock(&inode->i_lock); |
67a23c49 | 1724 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1725 | schedule(); |
1726 | finish_wait(wq, &wait.wait); | |
67a23c49 | 1727 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1728 | } |
1729 | ||
1da177e4 LT |
1730 | static __initdata unsigned long ihash_entries; |
1731 | static int __init set_ihash_entries(char *str) | |
1732 | { | |
1733 | if (!str) | |
1734 | return 0; | |
1735 | ihash_entries = simple_strtoul(str, &str, 0); | |
1736 | return 1; | |
1737 | } | |
1738 | __setup("ihash_entries=", set_ihash_entries); | |
1739 | ||
1740 | /* | |
1741 | * Initialize the waitqueues and inode hash table. | |
1742 | */ | |
1743 | void __init inode_init_early(void) | |
1744 | { | |
074b8517 | 1745 | unsigned int loop; |
1da177e4 LT |
1746 | |
1747 | /* If hashes are distributed across NUMA nodes, defer | |
1748 | * hash allocation until vmalloc space is available. | |
1749 | */ | |
1750 | if (hashdist) | |
1751 | return; | |
1752 | ||
1753 | inode_hashtable = | |
1754 | alloc_large_system_hash("Inode-cache", | |
1755 | sizeof(struct hlist_head), | |
1756 | ihash_entries, | |
1757 | 14, | |
1758 | HASH_EARLY, | |
1759 | &i_hash_shift, | |
1760 | &i_hash_mask, | |
31fe62b9 | 1761 | 0, |
1da177e4 LT |
1762 | 0); |
1763 | ||
074b8517 | 1764 | for (loop = 0; loop < (1U << i_hash_shift); loop++) |
1da177e4 LT |
1765 | INIT_HLIST_HEAD(&inode_hashtable[loop]); |
1766 | } | |
1767 | ||
74bf17cf | 1768 | void __init inode_init(void) |
1da177e4 | 1769 | { |
074b8517 | 1770 | unsigned int loop; |
1da177e4 LT |
1771 | |
1772 | /* inode slab cache */ | |
b0196009 PJ |
1773 | inode_cachep = kmem_cache_create("inode_cache", |
1774 | sizeof(struct inode), | |
1775 | 0, | |
1776 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1777 | SLAB_MEM_SPREAD), | |
20c2df83 | 1778 | init_once); |
1da177e4 LT |
1779 | |
1780 | /* Hash may have been set up in inode_init_early */ | |
1781 | if (!hashdist) | |
1782 | return; | |
1783 | ||
1784 | inode_hashtable = | |
1785 | alloc_large_system_hash("Inode-cache", | |
1786 | sizeof(struct hlist_head), | |
1787 | ihash_entries, | |
1788 | 14, | |
1789 | 0, | |
1790 | &i_hash_shift, | |
1791 | &i_hash_mask, | |
31fe62b9 | 1792 | 0, |
1da177e4 LT |
1793 | 0); |
1794 | ||
074b8517 | 1795 | for (loop = 0; loop < (1U << i_hash_shift); loop++) |
1da177e4 LT |
1796 | INIT_HLIST_HEAD(&inode_hashtable[loop]); |
1797 | } | |
1798 | ||
1799 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1800 | { | |
1801 | inode->i_mode = mode; | |
1802 | if (S_ISCHR(mode)) { | |
1803 | inode->i_fop = &def_chr_fops; | |
1804 | inode->i_rdev = rdev; | |
1805 | } else if (S_ISBLK(mode)) { | |
1806 | inode->i_fop = &def_blk_fops; | |
1807 | inode->i_rdev = rdev; | |
1808 | } else if (S_ISFIFO(mode)) | |
599a0ac1 | 1809 | inode->i_fop = &pipefifo_fops; |
1da177e4 LT |
1810 | else if (S_ISSOCK(mode)) |
1811 | inode->i_fop = &bad_sock_fops; | |
1812 | else | |
af0d9ae8 MK |
1813 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" |
1814 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1815 | inode->i_ino); | |
1da177e4 LT |
1816 | } |
1817 | EXPORT_SYMBOL(init_special_inode); | |
a1bd120d DM |
1818 | |
1819 | /** | |
eaae668d | 1820 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards |
a1bd120d DM |
1821 | * @inode: New inode |
1822 | * @dir: Directory inode | |
1823 | * @mode: mode of the new inode | |
1824 | */ | |
1825 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
62bb1091 | 1826 | umode_t mode) |
a1bd120d DM |
1827 | { |
1828 | inode->i_uid = current_fsuid(); | |
1829 | if (dir && dir->i_mode & S_ISGID) { | |
1830 | inode->i_gid = dir->i_gid; | |
1831 | if (S_ISDIR(mode)) | |
1832 | mode |= S_ISGID; | |
1833 | } else | |
1834 | inode->i_gid = current_fsgid(); | |
1835 | inode->i_mode = mode; | |
1836 | } | |
1837 | EXPORT_SYMBOL(inode_init_owner); | |
e795b717 | 1838 | |
2e149670 SH |
1839 | /** |
1840 | * inode_owner_or_capable - check current task permissions to inode | |
1841 | * @inode: inode being checked | |
1842 | * | |
4f80c6c1 AL |
1843 | * Return true if current either has CAP_FOWNER in a namespace with the |
1844 | * inode owner uid mapped, or owns the file. | |
e795b717 | 1845 | */ |
2e149670 | 1846 | bool inode_owner_or_capable(const struct inode *inode) |
e795b717 | 1847 | { |
4f80c6c1 AL |
1848 | struct user_namespace *ns; |
1849 | ||
92361636 | 1850 | if (uid_eq(current_fsuid(), inode->i_uid)) |
e795b717 | 1851 | return true; |
4f80c6c1 AL |
1852 | |
1853 | ns = current_user_ns(); | |
1854 | if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid)) | |
e795b717 SH |
1855 | return true; |
1856 | return false; | |
1857 | } | |
2e149670 | 1858 | EXPORT_SYMBOL(inode_owner_or_capable); |
1d59d61f TM |
1859 | |
1860 | /* | |
1861 | * Direct i/o helper functions | |
1862 | */ | |
1863 | static void __inode_dio_wait(struct inode *inode) | |
1864 | { | |
1865 | wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP); | |
1866 | DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP); | |
1867 | ||
1868 | do { | |
1869 | prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE); | |
1870 | if (atomic_read(&inode->i_dio_count)) | |
1871 | schedule(); | |
1872 | } while (atomic_read(&inode->i_dio_count)); | |
1873 | finish_wait(wq, &q.wait); | |
1874 | } | |
1875 | ||
1876 | /** | |
1877 | * inode_dio_wait - wait for outstanding DIO requests to finish | |
1878 | * @inode: inode to wait for | |
1879 | * | |
1880 | * Waits for all pending direct I/O requests to finish so that we can | |
1881 | * proceed with a truncate or equivalent operation. | |
1882 | * | |
1883 | * Must be called under a lock that serializes taking new references | |
1884 | * to i_dio_count, usually by inode->i_mutex. | |
1885 | */ | |
1886 | void inode_dio_wait(struct inode *inode) | |
1887 | { | |
1888 | if (atomic_read(&inode->i_dio_count)) | |
1889 | __inode_dio_wait(inode); | |
1890 | } | |
1891 | EXPORT_SYMBOL(inode_dio_wait); | |
1892 | ||
1893 | /* | |
1894 | * inode_dio_done - signal finish of a direct I/O requests | |
1895 | * @inode: inode the direct I/O happens on | |
1896 | * | |
1897 | * This is called once we've finished processing a direct I/O request, | |
1898 | * and is used to wake up callers waiting for direct I/O to be quiesced. | |
1899 | */ | |
1900 | void inode_dio_done(struct inode *inode) | |
1901 | { | |
1902 | if (atomic_dec_and_test(&inode->i_dio_count)) | |
1903 | wake_up_bit(&inode->i_state, __I_DIO_WAKEUP); | |
1904 | } | |
1905 | EXPORT_SYMBOL(inode_dio_done); |